* [RFC] Native Coarrays (finally!)
@ 2020-09-22 12:14 Nicolas König
2020-09-23 6:13 ` Damian Rouson
` (3 more replies)
0 siblings, 4 replies; 17+ messages in thread
From: Nicolas König @ 2020-09-22 12:14 UTC (permalink / raw)
To: fortran
[-- Attachment #1: Type: text/plain, Size: 3220 bytes --]
Hello everyone,
Contrary to rumors, I'm not dead, but I have been working silently on
the Native Coarray patch. Now, a few rewrites later, I think the patch
is now in an acceptable state to get a few comments on it and to push
it on a development branch. My hope is that after fixing a few of the
more obvious bugs and potential portability problems this might still
make it as an experimental feature into GCC 11 (this patch doesn't
disturb the compiler much, almost everything that is done is behind if
(flag_coarray == FCOARRAY_NATIVE) guards)
Supported Features:
- coarrays (duh), both with basic types and user-defined types
- Dynamic allocation and deallocation
- Arrays of Locks
- sync images/all
- Passing coarrays as arguments
- collective subroutines (co_broadcast/co_reduce/co_sum/...)
- critical sections
- this_image/num_images/lcobound/...
Missing Features
- coarrays of classes
- events
- stat & errmsg
- atomics
- proper handling of random_init
- coshape
- coarrays in data statements
- sync memory
- team
A few words on how these native coarrays work:
Each image is its own process, that is forked from the master process at
the start of the program. The number of images is determined by the
environment variable GFORTRAN_NUM_IMAGES or, alternatively, the number
of processors.
Each coarray is identified by its address. Since coarrays always behave
as if they had the SAVE attribute, this works even for allocatable
coarrays. ASLR is not an issue, since the addresses are assigned at
startup and remain valid over forks. If, on two different images, the
allocation function is called with the same descriptor address, the same
piece of memory is allocated.
Internally, the allocator (alloc.c) uses a shared hashmap (hashmap.c) to
remember with which ids pieces of memory allocated. If a new piece of
memory is needed, a simple relatively allocator (allocator.c) is used.
If the allocator doesn't hold any previously free()d memory, it requests
it from the shared memory object (shared_memory.c), which also handles
the translation of shared_mem_ptr's to pointers in the address space of
the image. At the moment shared_memory relies on double-mapping pages
for this (which might restrict the architectures on which this will
work, I have tested this on x86 and POWER), but since any piece of
memory should only be written to through one address within one
alloc/free pair, it shouldn't matter that much performance-wise.
The entry points in the library with the exception of master are defined
in wrapper.c, master(), the function handling launching the images, is
defined in coarraynative.c, and the other files shouldn't require much
explanation.
To compile a program to run with native coarrays, compile with
-fcoarray=native -lgfor_nca -lrt (I've not yet figured out how to
automagically link against the library).
It should be pointed out that Thomas greatly helped me with this
project, both with advice and actual code.
With the realization that this might have been a bit large of project
for a beginner remains
Nicolas
P.S.: Because I don't trust git as far as I can throw it, I've also
attached the patch and a copy of a few of my tests.
[-- Attachment #2: 2020-09-20-2.diff --]
[-- Type: text/x-patch, Size: 461548 bytes --]
diff --git a/gcc/flag-types.h b/gcc/flag-types.h
index b092c563f3d..02a084a037f 100644
--- a/gcc/flag-types.h
+++ b/gcc/flag-types.h
@@ -346,7 +346,8 @@ enum gfc_fcoarray
{
GFC_FCOARRAY_NONE = 0,
GFC_FCOARRAY_SINGLE,
- GFC_FCOARRAY_LIB
+ GFC_FCOARRAY_LIB,
+ GFC_FCOARRAY_NATIVE
};
diff --git a/gcc/fortran/dump-parse-tree.c b/gcc/fortran/dump-parse-tree.c
index f44648879f5..5fe834b4948 100644
--- a/gcc/fortran/dump-parse-tree.c
+++ b/gcc/fortran/dump-parse-tree.c
@@ -1060,7 +1060,7 @@ show_symbol (gfc_symbol *sym)
if (sym == NULL)
return;
- fprintf (dumpfile, "|| symbol: '%s' ", sym->name);
+ fprintf (dumpfile, "|| symbol: '%s' %p ", sym->name, (void *) &(sym->backend_decl));
len = strlen (sym->name);
for (i=len; i<12; i++)
fputc(' ', dumpfile);
diff --git a/gcc/fortran/frontend-passes.c b/gcc/fortran/frontend-passes.c
index 7768fdc25ca..871c354dda3 100644
--- a/gcc/fortran/frontend-passes.c
+++ b/gcc/fortran/frontend-passes.c
@@ -57,6 +57,7 @@ static int call_external_blas (gfc_code **, int *, void *);
static int matmul_temp_args (gfc_code **, int *,void *data);
static int index_interchange (gfc_code **, int*, void *);
static bool is_fe_temp (gfc_expr *e);
+static void rewrite_co_reduce (gfc_namespace *);
#ifdef CHECKING_P
static void check_locus (gfc_namespace *);
@@ -179,6 +180,9 @@ gfc_run_passes (gfc_namespace *ns)
if (flag_realloc_lhs)
realloc_strings (ns);
+
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ rewrite_co_reduce (ns);
}
#ifdef CHECKING_P
@@ -5566,3 +5570,121 @@ void gfc_check_externals (gfc_namespace *ns)
gfc_errors_to_warnings (false);
}
+
+/* Callback function. Create a wrapper around VALUE functions. */
+
+static int
+co_reduce_code (gfc_code **c, int *walk_subtrees ATTRIBUTE_UNUSED, void *data)
+{
+ gfc_code *co = *c;
+ gfc_expr *oper;
+ gfc_symbol *op_sym;
+ gfc_symbol *arg1, *arg2;
+ gfc_namespace *parent_ns;
+ gfc_namespace *proc_ns;
+ gfc_symbol *proc_sym;
+ gfc_symtree *f1t, *f2t;
+ gfc_symbol *f1, *f2;
+ gfc_code *assign;
+ gfc_expr *e1, *e2;
+ char name[GFC_MAX_SYMBOL_LEN + 1];
+ static int num;
+
+ if (co->op != EXEC_CALL || co->resolved_isym == NULL
+ || co->resolved_isym->id != GFC_ISYM_CO_REDUCE)
+ return 0;
+
+ oper = co->ext.actual->next->expr;
+ op_sym = oper->symtree->n.sym;
+ arg1 = op_sym->formal->sym;
+ arg2 = op_sym->formal->next->sym;
+
+ parent_ns = (gfc_namespace *) data;
+
+ /* Generate the wrapper around the function. */
+ proc_ns = gfc_get_namespace (parent_ns, 0);
+ snprintf (name, GFC_MAX_SYMBOL_LEN, "__coreduce_%d_%s", num++, op_sym->name);
+ gfc_get_symbol (name, proc_ns, &proc_sym);
+ proc_sym->attr.flavor = FL_PROCEDURE;
+ proc_sym->attr.subroutine = 1;
+ proc_sym->attr.referenced = 1;
+ proc_sym->attr.access = ACCESS_PRIVATE;
+ gfc_commit_symbol (proc_sym);
+ proc_ns->proc_name = proc_sym;
+
+ /* Make up the formal arguments. */
+ gfc_get_sym_tree (arg1->name, proc_ns, &f1t, false);
+ f1 = f1t->n.sym;
+ f1->ts = arg1->ts;
+ f1->attr.flavor = FL_VARIABLE;
+ f1->attr.dummy = 1;
+ f1->attr.intent = INTENT_INOUT;
+ f1->attr.fe_temp = 1;
+ f1->declared_at = arg1->declared_at;
+ f1->attr.referenced = 1;
+ proc_sym->formal = gfc_get_formal_arglist ();
+ proc_sym->formal->sym = f1;
+ gfc_commit_symbol (f1);
+
+ gfc_get_sym_tree (arg2->name, proc_ns, &f2t, false);
+ f2 = f2t->n.sym;
+ f2->ts = arg2->ts;
+ f2->attr.flavor = FL_VARIABLE;
+ f2->attr.dummy = 1;
+ f2->attr.intent = INTENT_IN;
+ f2->attr.fe_temp = 1;
+ f2->declared_at = arg2->declared_at;
+ f2->attr.referenced = 1;
+ proc_sym->formal->next = gfc_get_formal_arglist ();
+ proc_sym->formal->next->sym = f2;
+ gfc_commit_symbol (f2);
+
+ /* Generate the assignment statement. */
+ assign = gfc_get_code (EXEC_ASSIGN);
+
+ e1 = gfc_lval_expr_from_sym (f1);
+ e2 = gfc_get_expr ();
+ e2->where = proc_sym->declared_at;
+ e2->expr_type = EXPR_FUNCTION;
+ e2->symtree = f2t;
+ e2->ts = arg1->ts;
+ e2->value.function.esym = op_sym;
+ e2->value.function.actual = gfc_get_actual_arglist ();
+ e2->value.function.actual->expr = gfc_lval_expr_from_sym (f1);
+ e2->value.function.actual->next = gfc_get_actual_arglist ();
+ e2->value.function.actual->next->expr = gfc_lval_expr_from_sym (f2);
+ assign->expr1 = e1;
+ assign->expr2 = e2;
+ assign->loc = proc_sym->declared_at;
+
+ proc_ns->code = assign;
+
+ /* And hang it into the sibling list. */
+ proc_ns->sibling = parent_ns->contained;
+ parent_ns->contained = proc_ns;
+
+ /* ... and finally replace the call in the statement. */
+
+ oper->symtree->n.sym = proc_sym;
+ proc_sym->refs ++;
+ return 0;
+}
+
+/* Rewrite functions for co_reduce for a consistent calling
+ signature. This is only necessary if any of the functions
+ has a VALUE argument. */
+
+static void
+rewrite_co_reduce (gfc_namespace *global_ns)
+{
+ gfc_namespace *ns;
+
+ gfc_code_walker (&global_ns->code, co_reduce_code, dummy_expr_callback,
+ (void *) global_ns);
+
+ for (ns = global_ns->contained; ns; ns = ns->sibling)
+ gfc_code_walker (&ns->code, co_reduce_code, dummy_expr_callback,
+ (void *) global_ns);
+
+ return;
+}
diff --git a/gcc/fortran/gfortran.h b/gcc/fortran/gfortran.h
index 24c5101c4cb..b3ae45d11f7 100644
--- a/gcc/fortran/gfortran.h
+++ b/gcc/fortran/gfortran.h
@@ -1974,6 +1974,7 @@ typedef struct gfc_array_ref
int dimen; /* # of components in the reference */
int codimen;
bool in_allocate; /* For coarray checks. */
+ bool native_coarray_argument;
gfc_expr *team;
gfc_expr *stat;
locus where;
diff --git a/gcc/fortran/intrinsic.c b/gcc/fortran/intrinsic.c
index 3518a4e2c87..8a966417732 100644
--- a/gcc/fortran/intrinsic.c
+++ b/gcc/fortran/intrinsic.c
@@ -3734,7 +3734,7 @@ add_subroutines (void)
/* Coarray collectives. */
add_sym_4s ("co_broadcast", GFC_ISYM_CO_BROADCAST, CLASS_IMPURE,
BT_UNKNOWN, 0, GFC_STD_F2018,
- gfc_check_co_broadcast, NULL, NULL,
+ gfc_check_co_broadcast, NULL, gfc_resolve_co_broadcast,
a, BT_REAL, dr, REQUIRED, INTENT_INOUT,
"source_image", BT_INTEGER, di, REQUIRED, INTENT_IN,
stat, BT_INTEGER, di, OPTIONAL, INTENT_OUT,
@@ -3742,7 +3742,7 @@ add_subroutines (void)
add_sym_4s ("co_max", GFC_ISYM_CO_MAX, CLASS_IMPURE,
BT_UNKNOWN, 0, GFC_STD_F2018,
- gfc_check_co_minmax, NULL, NULL,
+ gfc_check_co_minmax, NULL, gfc_resolve_co_max,
a, BT_REAL, dr, REQUIRED, INTENT_INOUT,
result_image, BT_INTEGER, di, OPTIONAL, INTENT_IN,
stat, BT_INTEGER, di, OPTIONAL, INTENT_OUT,
@@ -3750,7 +3750,7 @@ add_subroutines (void)
add_sym_4s ("co_min", GFC_ISYM_CO_MIN, CLASS_IMPURE,
BT_UNKNOWN, 0, GFC_STD_F2018,
- gfc_check_co_minmax, NULL, NULL,
+ gfc_check_co_minmax, NULL, gfc_resolve_co_min,
a, BT_REAL, dr, REQUIRED, INTENT_INOUT,
result_image, BT_INTEGER, di, OPTIONAL, INTENT_IN,
stat, BT_INTEGER, di, OPTIONAL, INTENT_OUT,
@@ -3758,7 +3758,7 @@ add_subroutines (void)
add_sym_4s ("co_sum", GFC_ISYM_CO_SUM, CLASS_IMPURE,
BT_UNKNOWN, 0, GFC_STD_F2018,
- gfc_check_co_sum, NULL, NULL,
+ gfc_check_co_sum, NULL, gfc_resolve_co_sum,
a, BT_REAL, dr, REQUIRED, INTENT_INOUT,
result_image, BT_INTEGER, di, OPTIONAL, INTENT_IN,
stat, BT_INTEGER, di, OPTIONAL, INTENT_OUT,
@@ -3766,7 +3766,7 @@ add_subroutines (void)
add_sym_5s ("co_reduce", GFC_ISYM_CO_REDUCE, CLASS_IMPURE,
BT_UNKNOWN, 0, GFC_STD_F2018,
- gfc_check_co_reduce, NULL, NULL,
+ gfc_check_co_reduce, NULL, gfc_resolve_co_reduce,
a, BT_REAL, dr, REQUIRED, INTENT_INOUT,
"operator", BT_INTEGER, di, REQUIRED, INTENT_IN,
result_image, BT_INTEGER, di, OPTIONAL, INTENT_IN,
diff --git a/gcc/fortran/intrinsic.h b/gcc/fortran/intrinsic.h
index 166ae792939..2ca566ce3c4 100644
--- a/gcc/fortran/intrinsic.h
+++ b/gcc/fortran/intrinsic.h
@@ -677,7 +677,11 @@ void gfc_resolve_system_sub (gfc_code *);
void gfc_resolve_ttynam_sub (gfc_code *);
void gfc_resolve_umask_sub (gfc_code *);
void gfc_resolve_unlink_sub (gfc_code *);
-
+void gfc_resolve_co_sum (gfc_code *);
+void gfc_resolve_co_min (gfc_code *);
+void gfc_resolve_co_max (gfc_code *);
+void gfc_resolve_co_reduce (gfc_code *);
+void gfc_resolve_co_broadcast (gfc_code *);
/* The findloc() subroutine requires the most arguments: six. */
diff --git a/gcc/fortran/iresolve.c b/gcc/fortran/iresolve.c
index 73769615c20..844891e34ab 100644
--- a/gcc/fortran/iresolve.c
+++ b/gcc/fortran/iresolve.c
@@ -36,6 +36,7 @@ along with GCC; see the file COPYING3. If not see
#include "constructor.h"
#include "arith.h"
#include "trans.h"
+#include "options.h"
/* Given printf-like arguments, return a stable version of the result string.
@@ -4030,3 +4031,100 @@ gfc_resolve_unlink_sub (gfc_code *c)
name = gfc_get_string (PREFIX ("unlink_i%d_sub"), kind);
c->resolved_sym = gfc_get_intrinsic_sub_symbol (name);
}
+
+/* Resolve the CO_SUM et al. intrinsic subroutines. */
+
+static void
+gfc_resolve_co_collective (gfc_code *c, const char *oper)
+{
+ int kind;
+ gfc_expr *e;
+ const char *name;
+
+ if (flag_coarray != GFC_FCOARRAY_NATIVE)
+ name = gfc_get_string (PREFIX ("caf_co_sum"));
+ else
+ {
+ e = c->ext.actual->expr;
+ kind = e->ts.kind;
+
+ name = gfc_get_string (PREFIX ("nca_collsub_%s_%s_%c%d"), oper,
+ e->rank ? "array" : "scalar",
+ gfc_type_letter (e->ts.type), kind);
+ }
+
+ c->resolved_sym = gfc_get_intrinsic_sub_symbol (name);
+}
+
+/* Resolve CO_SUM. */
+
+void
+gfc_resolve_co_sum (gfc_code *c)
+{
+ gfc_resolve_co_collective (c, "sum");
+}
+
+/* Resolve CO_MIN. */
+
+void
+gfc_resolve_co_min (gfc_code *c)
+{
+ gfc_resolve_co_collective (c, "min");
+}
+
+/* Resolve CO_MAX. */
+
+void
+gfc_resolve_co_max (gfc_code *c)
+{
+ gfc_resolve_co_collective (c, "max");
+}
+
+/* Resolve CO_REDUCE. */
+
+void
+gfc_resolve_co_reduce (gfc_code *c)
+{
+ gfc_expr *e;
+ const char *name;
+
+ if (flag_coarray != GFC_FCOARRAY_NATIVE)
+ name = gfc_get_string (PREFIX ("caf_co_reduce"));
+
+ else
+ {
+ e = c->ext.actual->expr;
+ if (e->ts.type == BT_CHARACTER)
+ name = gfc_get_string (PREFIX ("nca_collsub_reduce_%s%c%d"),
+ e->rank ? "array" : "scalar",
+ gfc_type_letter (e->ts.type), e->ts.kind);
+ else
+ name = gfc_get_string (PREFIX ("nca_collsub_reduce_%s"),
+ e->rank ? "array" : "scalar" );
+ }
+
+ c->resolved_sym = gfc_get_intrinsic_sub_symbol (name);
+}
+
+void
+gfc_resolve_co_broadcast (gfc_code * c)
+{
+ gfc_expr *e;
+ const char *name;
+
+ if (flag_coarray != GFC_FCOARRAY_NATIVE)
+ name = gfc_get_string (PREFIX ("caf_co_broadcast"));
+ else
+ {
+ e = c->ext.actual->expr;
+ if (e->ts.type == BT_CHARACTER)
+ name = gfc_get_string (PREFIX ("nca_collsub_broadcast_%s%c%d"),
+ e->rank ? "array" : "scalar",
+ gfc_type_letter (e->ts.type), e->ts.kind);
+ else
+ name = gfc_get_string (PREFIX ("nca_collsub_broadcast_%s"),
+ e->rank ? "array" : "scalar" );
+ }
+
+ c->resolved_sym = gfc_get_intrinsic_sub_symbol (name);
+}
diff --git a/gcc/fortran/lang.opt b/gcc/fortran/lang.opt
index da4b1aa879a..e267fb0aef8 100644
--- a/gcc/fortran/lang.opt
+++ b/gcc/fortran/lang.opt
@@ -761,7 +761,7 @@ Copy array sections into a contiguous block on procedure entry.
fcoarray=
Fortran RejectNegative Joined Enum(gfc_fcoarray) Var(flag_coarray) Init(GFC_FCOARRAY_NONE)
--fcoarray=<none|single|lib> Specify which coarray parallelization should be used.
+-fcoarray=<none|single|lib|native> Specify which coarray parallelization should be used.
Enum
Name(gfc_fcoarray) Type(enum gfc_fcoarray) UnknownError(Unrecognized option: %qs)
@@ -775,6 +775,9 @@ Enum(gfc_fcoarray) String(single) Value(GFC_FCOARRAY_SINGLE)
EnumValue
Enum(gfc_fcoarray) String(lib) Value(GFC_FCOARRAY_LIB)
+EnumValue
+Enum(gfc_fcoarray) String(native) Value(GFC_FCOARRAY_NATIVE)
+
fcheck=
Fortran RejectNegative JoinedOrMissing
-fcheck=[...] Specify which runtime checks are to be performed.
diff --git a/gcc/fortran/resolve.c b/gcc/fortran/resolve.c
index 2751c0ccf62..4b986e40587 100644
--- a/gcc/fortran/resolve.c
+++ b/gcc/fortran/resolve.c
@@ -3585,6 +3585,53 @@ resolve_specific_s (gfc_code *c)
return false;
}
+/* Fix up references to native coarrays in call - element references
+ have to be converted to full references if the coarray has to be
+ passed fully. */
+
+static void
+fixup_coarray_args (gfc_symbol *sym, gfc_actual_arglist *actual)
+{
+ gfc_formal_arglist *formal, *f;
+ gfc_actual_arglist *a;
+
+ formal = gfc_sym_get_dummy_args (sym);
+
+ if (formal == NULL)
+ return;
+
+ for (a = actual, f = formal; a && f; a = a->next, f = f->next)
+ {
+ if (a->expr == NULL || f->sym == NULL)
+ continue;
+ if (a->expr->expr_type == EXPR_VARIABLE
+ && a->expr->symtree->n.sym->attr.codimension
+ && f->sym->attr.codimension)
+ {
+ gfc_ref *r;
+ for (r = a->expr->ref; r; r = r->next)
+ {
+ if (r->type == REF_ARRAY && r->u.ar.codimen)
+ {
+ gfc_array_ref *ar = &r->u.ar;
+ int i, eff_dimen = ar->dimen + ar->codimen;
+
+ for (i = ar->dimen; i < eff_dimen; i++)
+ {
+ ar->dimen_type[i] = DIMEN_RANGE;
+ gcc_assert (ar->start[i] == NULL);
+ gcc_assert (ar->end[i] == NULL);
+ }
+
+ if (ar->type == AR_ELEMENT)
+ ar->type = !ar->dimen ? AR_FULL : AR_SECTION;
+
+ ar->native_coarray_argument = true;
+ }
+ }
+ }
+ }
+}
/* Resolve a subroutine call not known to be generic nor specific. */
@@ -3615,7 +3662,7 @@ resolve_unknown_s (gfc_code *c)
found:
gfc_procedure_use (sym, &c->ext.actual, &c->loc);
-
+
c->resolved_sym = sym;
return pure_subroutine (sym, sym->name, &c->loc);
@@ -3740,6 +3787,9 @@ resolve_call (gfc_code *c)
/* Typebound procedure: Assume the worst. */
gfc_current_ns->proc_name->attr.array_outer_dependency = 1;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ fixup_coarray_args (csym, c->ext.actual);
+
return t;
}
@@ -10099,7 +10149,7 @@ resolve_critical (gfc_code *code)
char name[GFC_MAX_SYMBOL_LEN];
static int serial = 0;
- if (flag_coarray != GFC_FCOARRAY_LIB)
+ if (flag_coarray != GFC_FCOARRAY_LIB && flag_coarray != GFC_FCOARRAY_NATIVE)
return;
symtree = gfc_find_symtree (gfc_current_ns->sym_root,
@@ -10136,6 +10186,19 @@ resolve_critical (gfc_code *code)
symtree->n.sym->as->lower[0] = gfc_get_int_expr (gfc_default_integer_kind,
NULL, 1);
gfc_commit_symbols();
+
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ gfc_ref *r = gfc_get_ref ();
+ r->type = REF_ARRAY;
+ r->u.ar.type = AR_ELEMENT;
+ r->u.ar.as = code->resolved_sym->as;
+ for (int i = 0; i < code->resolved_sym->as->corank; i++)
+ r->u.ar.dimen_type [i] = DIMEN_THIS_IMAGE;
+
+ code->expr1 = gfc_lval_expr_from_sym (code->resolved_sym);
+ code->expr1->ref = r;
+ }
}
diff --git a/gcc/fortran/trans-array.c b/gcc/fortran/trans-array.c
index 54e1107c711..340a8dd06d3 100644
--- a/gcc/fortran/trans-array.c
+++ b/gcc/fortran/trans-array.c
@@ -2940,6 +2940,60 @@ gfc_add_loop_ss_code (gfc_loopinfo * loop, gfc_ss * ss, bool subscript,
gfc_add_loop_ss_code (nested_loop, nested_loop->ss, subscript, where);
}
+static tree
+gfc_add_strides (tree expr, tree desc, int beg, int end)
+{
+ int i;
+ tree tmp, stride;
+ tmp = gfc_index_zero_node;
+ for (i = beg; i < end; i++)
+ {
+ stride = gfc_conv_array_stride (desc, i);
+ tmp = fold_build2_loc (input_location, PLUS_EXPR, TREE_TYPE(tmp),
+ tmp, stride);
+ }
+ return fold_build2_loc (input_location, PLUS_EXPR, TREE_TYPE(expr),
+ expr, tmp);
+}
+
+/* This function calculates the new offset via
+ new_offset = offset + this_image ()
+ * arrray.stride[first_codimension]
+ + sum (remaining codimension offsets)
+ If offset is a pointer, we also need to multiply it by the size.*/
+static tree
+gfc_native_coarray_add_this_image_offset (tree offset, tree desc,
+ gfc_array_ref *ar, int is_pointer,
+ int subtract)
+{
+ tree tmp, off;
+ /* Calculate the actual offset. */
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_nca_this_image,
+ 1, integer_zero_node);
+ tmp = convert (TREE_TYPE(gfc_index_zero_node), tmp);
+ tmp = fold_build2_loc (input_location, MINUS_EXPR, TREE_TYPE(tmp), tmp,
+ build_int_cst (TREE_TYPE(tmp), subtract));
+ tmp = fold_build2_loc (input_location, MULT_EXPR, TREE_TYPE(tmp),
+ gfc_conv_array_stride (desc, ar->dimen), tmp);
+ /* We also need to add the missing strides once to compensate for the
+ offset, that is to large now. The loop starts at sym->as.rank+1
+ because we need to skip the first corank stride */
+ off = gfc_add_strides (tmp, desc, ar->as->rank + 1,
+ ar->as->rank + ar->as->corank);
+ if (is_pointer)
+ {
+ /* Remove pointer and array from type in order to get the raw base type. */
+ tmp = TREE_TYPE(TREE_TYPE(TREE_TYPE(offset)));
+ /* And get the size of that base type. */
+ tmp = convert (TREE_TYPE(off), size_in_bytes_loc (input_location, tmp));
+ tmp = fold_build2_loc (input_location, MULT_EXPR, TREE_TYPE(off),
+ off, tmp);
+ return fold_build_pointer_plus_loc (input_location, offset, tmp);
+ }
+ else
+ return fold_build2_loc (input_location, PLUS_EXPR, TREE_TYPE(offset),
+ offset, off);
+}
/* Translate expressions for the descriptor and data pointer of a SS. */
/*GCC ARRAYS*/
@@ -2951,6 +3005,7 @@ gfc_conv_ss_descriptor (stmtblock_t * block, gfc_ss * ss, int base)
gfc_ss_info *ss_info;
gfc_array_info *info;
tree tmp;
+ gfc_ref *ref;
ss_info = ss->info;
info = &ss_info->data.array;
@@ -2982,10 +3037,18 @@ gfc_conv_ss_descriptor (stmtblock_t * block, gfc_ss * ss, int base)
}
/* Also the data pointer. */
tmp = gfc_conv_array_data (se.expr);
+ /* If we have a native coarray with implied this_image (), add the
+ appropriate offset to the data pointer. */
+ ref = ss_info->expr->ref;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && ref
+ && ref->u.ar.dimen_type[ref->u.ar.dimen + ref->u.ar.codimen - 1]
+ == DIMEN_THIS_IMAGE)
+ tmp = gfc_native_coarray_add_this_image_offset (tmp, se.expr, &ref->u.ar, 1, 1);
/* If this is a variable or address of a variable we use it directly.
Otherwise we must evaluate it now to avoid breaking dependency
analysis by pulling the expressions for elemental array indices
inside the loop. */
+
if (!(DECL_P (tmp)
|| (TREE_CODE (tmp) == ADDR_EXPR
&& DECL_P (TREE_OPERAND (tmp, 0)))))
@@ -2993,6 +3056,15 @@ gfc_conv_ss_descriptor (stmtblock_t * block, gfc_ss * ss, int base)
info->data = tmp;
tmp = gfc_conv_array_offset (se.expr);
+ /* If we have a native coarray, adjust the offset to remove the
+ offset for the codimensions. */
+ // TODO: check whether the recipient is a coarray, if it is, disable
+ // all of this
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && ref
+ && ref->u.ar.dimen_type[ref->u.ar.dimen + ref->u.ar.codimen - 1]
+ == DIMEN_THIS_IMAGE)
+ tmp = gfc_add_strides (tmp, se.expr, ref->u.ar.as->rank,
+ ref->u.ar.as->rank + ref->u.ar.as->corank);
info->offset = gfc_evaluate_now (tmp, block);
/* Make absolutely sure that the saved_offset is indeed saved
@@ -3593,6 +3665,7 @@ build_array_ref (tree desc, tree offset, tree decl, tree vptr)
}
+
/* Build an array reference. se->expr already holds the array descriptor.
This should be either a variable, indirect variable reference or component
reference. For arrays which do not have a descriptor, se->expr will be
@@ -3612,8 +3685,20 @@ gfc_conv_array_ref (gfc_se * se, gfc_array_ref * ar, gfc_expr *expr,
gfc_se tmpse;
gfc_symbol * sym = expr->symtree->n.sym;
char *var_name = NULL;
+ bool need_impl_this_image;
+ int eff_dimen;
+
+ need_impl_this_image =
+ ar->dimen_type[ar->dimen + ar->codimen - 1] == DIMEN_THIS_IMAGE;
+
+ if (flag_coarray == GFC_FCOARRAY_NATIVE
+ && !need_impl_this_image)
+ eff_dimen = ar->dimen + ar->codimen - 1;
+ else
+ eff_dimen = ar->dimen - 1;
- if (ar->dimen == 0)
+
+ if (flag_coarray != GFC_FCOARRAY_NATIVE && ar->dimen == 0)
{
gcc_assert (ar->codimen || sym->attr.select_rank_temporary
|| (ar->as && ar->as->corank));
@@ -3681,7 +3766,7 @@ gfc_conv_array_ref (gfc_se * se, gfc_array_ref * ar, gfc_expr *expr,
/* Calculate the offsets from all the dimensions. Make sure to associate
the final offset so that we form a chain of loop invariant summands. */
- for (n = ar->dimen - 1; n >= 0; n--)
+ for (n = eff_dimen; n >= 0; n--)
{
/* Calculate the index for this dimension. */
gfc_init_se (&indexse, se);
@@ -3753,6 +3838,9 @@ gfc_conv_array_ref (gfc_se * se, gfc_array_ref * ar, gfc_expr *expr,
add_to_offset (&cst_offset, &offset, tmp);
}
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && need_impl_this_image)
+ offset = gfc_native_coarray_add_this_image_offset (offset, se->expr, ar, 0, 0);
+
if (!integer_zerop (cst_offset))
offset = fold_build2_loc (input_location, PLUS_EXPR,
gfc_array_index_type, offset, cst_offset);
@@ -5423,7 +5511,7 @@ gfc_conv_descriptor_cosize (tree desc, int rank, int corank)
} */
/*GCC ARRAYS*/
-static tree
+tree
gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
gfc_expr ** lower, gfc_expr ** upper, stmtblock_t * pblock,
stmtblock_t * descriptor_block, tree * overflow,
@@ -5441,6 +5529,8 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
tree elsecase;
tree cond;
tree var;
+ tree conv_lbound;
+ tree conv_ubound;
stmtblock_t thenblock;
stmtblock_t elseblock;
gfc_expr *ubound;
@@ -5454,7 +5544,7 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
/* Set the dtype before the alloc, because registration of coarrays needs
it initialized. */
- if (expr->ts.type == BT_CHARACTER
+ if (expr && expr->ts.type == BT_CHARACTER
&& expr->ts.deferred
&& VAR_P (expr->ts.u.cl->backend_decl))
{
@@ -5462,7 +5552,7 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
tmp = gfc_conv_descriptor_dtype (descriptor);
gfc_add_modify (pblock, tmp, gfc_get_dtype_rank_type (rank, type));
}
- else if (expr->ts.type == BT_CHARACTER
+ else if (expr && expr->ts.type == BT_CHARACTER
&& expr->ts.deferred
&& TREE_CODE (descriptor) == COMPONENT_REF)
{
@@ -5494,9 +5584,6 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
for (n = 0; n < rank; n++)
{
- tree conv_lbound;
- tree conv_ubound;
-
/* We have 3 possibilities for determining the size of the array:
lower == NULL => lbound = 1, ubound = upper[n]
upper[n] = NULL => lbound = 1, ubound = lower[n]
@@ -5646,6 +5733,15 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
}
gfc_conv_descriptor_lbound_set (descriptor_block, descriptor,
gfc_rank_cst[n], se.expr);
+ conv_lbound = se.expr;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+
+ tmp = fold_build2_loc (input_location, MULT_EXPR, gfc_array_index_type,
+ se.expr, stride);
+ offset = fold_build2_loc (input_location, MINUS_EXPR,
+ gfc_array_index_type, offset, tmp);
+ }
if (n < rank + corank - 1)
{
@@ -5655,6 +5751,18 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
gfc_add_block_to_block (pblock, &se.pre);
gfc_conv_descriptor_ubound_set (descriptor_block, descriptor,
gfc_rank_cst[n], se.expr);
+ gfc_conv_descriptor_stride_set (descriptor_block, descriptor,
+ gfc_rank_cst[n], stride);
+ conv_ubound = se.expr;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ size = gfc_conv_array_extent_dim (conv_lbound, conv_ubound,
+ &or_expr);
+ size = gfc_evaluate_now (size, descriptor_block);
+ stride = fold_build2_loc (input_location, MULT_EXPR,
+ gfc_array_index_type, stride, size);
+ stride = gfc_evaluate_now (stride, descriptor_block);
+ }
}
}
@@ -5688,7 +5796,7 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
/* Convert to size_t. */
*element_size = fold_convert (size_type_node, tmp);
- if (rank == 0)
+ if (rank == 0 && !(flag_coarray == GFC_FCOARRAY_NATIVE && corank))
return *element_size;
*nelems = gfc_evaluate_now (stride, pblock);
@@ -5773,6 +5881,38 @@ retrieve_last_ref (gfc_ref **ref_in, gfc_ref **prev_ref_in)
return true;
}
+int
+gfc_native_coarray_get_allocation_type (gfc_symbol * sym)
+{
+ bool is_lock_type, is_event_type;
+ is_lock_type = sym->ts.type == BT_DERIVED
+ && sym->ts.u.derived->from_intmod == INTMOD_ISO_FORTRAN_ENV
+ && sym->ts.u.derived->intmod_sym_id == ISOFORTRAN_LOCK_TYPE;
+
+ is_event_type = sym->ts.type == BT_DERIVED
+ && sym->ts.u.derived->from_intmod == INTMOD_ISO_FORTRAN_ENV
+ && sym->ts.u.derived->intmod_sym_id == ISOFORTRAN_EVENT_TYPE;
+
+ if (is_lock_type)
+ return GFC_NCA_LOCK_COARRAY;
+ else if (is_event_type)
+ return GFC_NCA_EVENT_COARRAY;
+ else
+ return GFC_NCA_NORMAL_COARRAY;
+}
+
+void
+gfc_allocate_native_coarray (stmtblock_t *b, tree decl, tree size, int corank,
+ int alloc_type)
+{
+ gfc_add_expr_to_block (b,
+ build_call_expr_loc (input_location, gfor_fndecl_nca_coarray_allocate,
+ 4, gfc_build_addr_expr (pvoid_type_node, decl),
+ size, build_int_cst (integer_type_node, corank),
+ build_int_cst (integer_type_node, alloc_type)));
+
+}
+
/* Initializes the descriptor and generates a call to _gfor_allocate. Does
the work for an ALLOCATE statement. */
/*GCC ARRAYS*/
@@ -5784,6 +5924,7 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
bool e3_has_nodescriptor)
{
tree tmp;
+ tree allocation;
tree pointer;
tree offset = NULL_TREE;
tree token = NULL_TREE;
@@ -5914,7 +6055,7 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
expr3_elem_size, nelems, expr3, e3_arr_desc,
e3_has_nodescriptor, expr, &element_size);
- if (dimension)
+ if (dimension || (flag_coarray == GFC_FCOARRAY_NATIVE && coarray))
{
var_overflow = gfc_create_var (integer_type_node, "overflow");
gfc_add_modify (&se->pre, var_overflow, overflow);
@@ -5956,7 +6097,7 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
pointer = gfc_conv_descriptor_data_get (se->expr);
STRIP_NOPS (pointer);
- if (allocatable)
+ if (allocatable && !(flag_coarray == GFC_FCOARRAY_NATIVE && coarray))
{
not_prev_allocated = gfc_create_var (logical_type_node,
"not_prev_allocated");
@@ -5969,8 +6110,17 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
gfc_start_block (&elseblock);
+ if (coarray && flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ tree elem_size
+ = size_in_bytes (gfc_get_element_type (TREE_TYPE(se->expr)));
+ int alloc_type
+ = gfc_native_coarray_get_allocation_type (expr->symtree->n.sym);
+ gfc_allocate_native_coarray (&elseblock, se->expr, elem_size,
+ ref->u.ar.as->corank, alloc_type);
+ }
/* The allocatable variant takes the old pointer as first argument. */
- if (allocatable)
+ else if (allocatable)
gfc_allocate_allocatable (&elseblock, pointer, size, token,
status, errmsg, errlen, label_finish, expr,
coref != NULL ? coref->u.ar.as->corank : 0);
@@ -5987,13 +6137,12 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
cond = gfc_unlikely (fold_build2_loc (input_location, NE_EXPR,
logical_type_node, var_overflow, integer_zero_node),
PRED_FORTRAN_OVERFLOW);
- tmp = fold_build3_loc (input_location, COND_EXPR, void_type_node, cond,
+ allocation = fold_build3_loc (input_location, COND_EXPR, void_type_node, cond,
error, gfc_finish_block (&elseblock));
}
else
- tmp = gfc_finish_block (&elseblock);
+ allocation = gfc_finish_block (&elseblock);
- gfc_add_expr_to_block (&se->pre, tmp);
/* Update the array descriptor with the offset and the span. */
if (dimension)
@@ -6004,6 +6153,7 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
}
set_descriptor = gfc_finish_block (&set_descriptor_block);
+
if (status != NULL_TREE)
{
cond = fold_build2_loc (input_location, EQ_EXPR,
@@ -6014,14 +6164,25 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
cond = fold_build2_loc (input_location, TRUTH_OR_EXPR,
logical_type_node, cond, not_prev_allocated);
- gfc_add_expr_to_block (&se->pre,
- fold_build3_loc (input_location, COND_EXPR, void_type_node,
+ set_descriptor = fold_build3_loc (input_location, COND_EXPR, void_type_node,
cond,
set_descriptor,
- build_empty_stmt (input_location)));
+ build_empty_stmt (input_location));
+ }
+
+ // For native coarrays, the size must be set before the allocation routine
+ // can be called.
+ if (coarray && flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ gfc_add_expr_to_block (&se->pre, set_descriptor);
+ gfc_add_expr_to_block (&se->pre, allocation);
}
else
+ {
+ gfc_add_expr_to_block (&se->pre, allocation);
gfc_add_expr_to_block (&se->pre, set_descriptor);
+ }
+
return true;
}
@@ -6518,6 +6679,7 @@ gfc_trans_dummy_array_bias (gfc_symbol * sym, tree tmpdesc,
bool optional_arg;
gfc_array_spec *as;
bool is_classarray = IS_CLASS_ARRAY (sym);
+ int eff_dimen;
/* Do nothing for pointer and allocatable arrays. */
if ((sym->ts.type != BT_CLASS && sym->attr.pointer)
@@ -6632,8 +6794,13 @@ gfc_trans_dummy_array_bias (gfc_symbol * sym, tree tmpdesc,
offset = gfc_index_zero_node;
size = gfc_index_one_node;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ eff_dimen = as->rank + as->corank;
+ else
+ eff_dimen = as->rank;
+
/* Evaluate the bounds of the array. */
- for (n = 0; n < as->rank; n++)
+ for (n = 0; n < eff_dimen; n++)
{
if (checkparm || !as->upper[n])
{
@@ -6718,7 +6885,7 @@ gfc_trans_dummy_array_bias (gfc_symbol * sym, tree tmpdesc,
gfc_array_index_type, offset, tmp);
/* The size of this dimension, and the stride of the next. */
- if (n + 1 < as->rank)
+ if (n + 1 < eff_dimen)
{
stride = GFC_TYPE_ARRAY_STRIDE (type, n + 1);
@@ -6873,20 +7040,35 @@ gfc_get_dataptr_offset (stmtblock_t *block, tree parm, tree desc, tree offset,
return;
}
+ /* if it's a coarray with implicit this_image, add that to the offset. */
+ ref = expr->ref;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && ref && ref->type == REF_ARRAY
+ && ref->u.ar.dimen_type[ref->u.ar.dimen + ref->u.ar.codimen - 1]
+ == DIMEN_THIS_IMAGE
+ && !ref->u.ar.native_coarray_argument)
+ offset = gfc_native_coarray_add_this_image_offset (offset, desc,
+ &ref->u.ar, 0, 1);
+
tmp = build_array_ref (desc, offset, NULL, NULL);
/* Offset the data pointer for pointer assignments from arrays with
subreferences; e.g. my_integer => my_type(:)%integer_component. */
if (subref)
{
- /* Go past the array reference. */
+ /* Go past the array reference. */
for (ref = expr->ref; ref; ref = ref->next)
- if (ref->type == REF_ARRAY &&
- ref->u.ar.type != AR_ELEMENT)
- {
- ref = ref->next;
- break;
- }
+ {
+ if (ref->type == REF_ARRAY &&
+ ref->u.ar.type != AR_ELEMENT)
+ {
+ ref = ref->next;
+ break;
+ }
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE && ref->type == REF_ARRAY &&
+ ref->u.ar.dimen_type[ref->u.ar.dimen +ref->u.ar.codimen -1]
+ == DIMEN_THIS_IMAGE)
+ tmp = gfc_native_coarray_add_this_image_offset (tmp, desc, &ref->u.ar, 0, 1);
+ }
/* Calculate the offset for each subsequent subreference. */
for (; ref; ref = ref->next)
@@ -6949,7 +7131,10 @@ gfc_get_dataptr_offset (stmtblock_t *block, tree parm, tree desc, tree offset,
gfc_array_index_type, stride, itmp);
stride = gfc_evaluate_now (stride, block);
}
-
+ if (flag_coarray == GFC_FCOARRAY_NATIVE &&
+ ref->u.ar.dimen_type[ref->u.ar.dimen +ref->u.ar.codimen -1]
+ == DIMEN_THIS_IMAGE)
+ tmp = gfc_native_coarray_add_this_image_offset (tmp, desc, &ref->u.ar, 0, 1);
/* Apply the index to obtain the array element. */
tmp = gfc_build_array_ref (tmp, index, NULL);
break;
@@ -7283,6 +7468,13 @@ gfc_conv_expr_descriptor (gfc_se *se, gfc_expr *expr)
else
full = gfc_full_array_ref_p (info->ref, NULL);
+ if (flag_coarray == GFC_FCOARRAY_NATIVE &&
+ info->ref->type == REF_ARRAY &&
+ info->ref->u.ar.dimen_type[info->ref->u.ar.dimen
+ + info->ref->u.ar.codimen - 1] ==
+ DIMEN_THIS_IMAGE)
+ full = 0;
+
if (full && !transposed_dims (ss))
{
if (se->direct_byref && !se->byref_noassign)
@@ -7517,9 +7709,19 @@ gfc_conv_expr_descriptor (gfc_se *se, gfc_expr *expr)
tree to;
tree base;
tree offset;
-
+#if 0 /* TK */
ndim = info->ref ? info->ref->u.ar.dimen : ss->dimen;
-
+#else
+ if (info->ref)
+ {
+ if (info->ref->u.ar.native_coarray_argument)
+ ndim = info->ref->u.ar.dimen + info->ref->u.ar.codimen;
+ else
+ ndim = info->ref->u.ar.dimen;
+ }
+ else
+ ndim = ss->dimen;
+#endif
if (se->want_coarray)
{
gfc_array_ref *ar = &info->ref->u.ar;
@@ -7888,7 +8090,15 @@ gfc_conv_array_parameter (gfc_se * se, gfc_expr * expr, bool g77,
expr->ts.u.cl->backend_decl = tmp;
se->string_length = tmp;
}
-
+#if 0
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && fsym && fsym->attr.codimension && sym)
+ {
+ gfc_init_se (se, NULL);
+ tmp = gfc_get_symbol_decl (sym);
+ se->expr = gfc_build_addr_expr (NULL_TREE, tmp);
+ return;
+ }
+#endif
/* Is this the result of the enclosing procedure? */
this_array_result = (full_array_var && sym->attr.flavor == FL_PROCEDURE);
if (this_array_result
@@ -7896,6 +8106,10 @@ gfc_conv_array_parameter (gfc_se * se, gfc_expr * expr, bool g77,
&& (sym->backend_decl != parent))
this_array_result = false;
+#if 1 /* TK */
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && fsym && fsym->attr.codimension)
+ g77 = false;
+#endif
/* Passing address of the array if it is not pointer or assumed-shape. */
if (full_array_var && g77 && !this_array_result
&& sym->ts.type != BT_DERIVED && sym->ts.type != BT_CLASS)
@@ -8030,8 +8244,8 @@ gfc_conv_array_parameter (gfc_se * se, gfc_expr * expr, bool g77,
{
/* Every other type of array. */
se->want_pointer = 1;
- gfc_conv_expr_descriptor (se, expr);
+ gfc_conv_expr_descriptor (se, expr);
if (size)
array_parameter_size (build_fold_indirect_ref_loc (input_location,
se->expr),
@@ -10837,9 +11051,15 @@ gfc_walk_array_ref (gfc_ss * ss, gfc_expr * expr, gfc_ref * ref)
case AR_SECTION:
newss = gfc_get_array_ss (ss, expr, 0, GFC_SS_SECTION);
newss->info->data.array.ref = ref;
-
+#if 1 /* TK */
+ int eff_dimen;
+ if (ar->native_coarray_argument)
+ eff_dimen = ar->dimen + ar->codimen;
+ else
+ eff_dimen = ar->dimen;
+#endif
/* We add SS chains for all the subscripts in the section. */
- for (n = 0; n < ar->dimen; n++)
+ for (n = 0; n < eff_dimen; n++)
{
gfc_ss *indexss;
diff --git a/gcc/fortran/trans-array.h b/gcc/fortran/trans-array.h
index e561605aaed..0bfd1b03022 100644
--- a/gcc/fortran/trans-array.h
+++ b/gcc/fortran/trans-array.h
@@ -23,6 +23,15 @@ along with GCC; see the file COPYING3. If not see
bool gfc_array_allocate (gfc_se *, gfc_expr *, tree, tree, tree, tree,
tree, tree *, gfc_expr *, tree, bool);
+enum gfc_coarray_allocation_type {
+ GFC_NCA_NORMAL_COARRAY = 3,
+ GFC_NCA_LOCK_COARRAY,
+ GFC_NCA_EVENT_COARRAY
+};
+int gfc_native_coarray_get_allocation_type (gfc_symbol *);
+
+void gfc_allocate_native_coarray (stmtblock_t *, tree, tree, int, int);
+
/* Allow the bounds of a loop to be set from a callee's array spec. */
void gfc_set_loop_bounds_from_array_spec (gfc_interface_mapping *,
gfc_se *, gfc_array_spec *);
@@ -57,6 +66,10 @@ tree gfc_bcast_alloc_comp (gfc_symbol *, gfc_expr *, int, tree,
tree gfc_deallocate_alloc_comp_no_caf (gfc_symbol *, tree, int);
tree gfc_reassign_alloc_comp_caf (gfc_symbol *, tree, tree);
+tree gfc_array_init_size (tree, int, int, tree *, gfc_expr **, gfc_expr **,
+ stmtblock_t *, stmtblock_t *, tree *, tree, tree *,
+ gfc_expr *, tree, bool, gfc_expr *, tree *);
+
tree gfc_copy_alloc_comp (gfc_symbol *, tree, tree, int, int);
tree gfc_copy_only_alloc_comp (gfc_symbol *, tree, tree, int);
diff --git a/gcc/fortran/trans-decl.c b/gcc/fortran/trans-decl.c
index 769ab20c82d..cd232544063 100644
--- a/gcc/fortran/trans-decl.c
+++ b/gcc/fortran/trans-decl.c
@@ -170,6 +170,21 @@ tree gfor_fndecl_co_reduce;
tree gfor_fndecl_co_sum;
tree gfor_fndecl_caf_is_present;
+/* Native coarray functions. */
+
+tree gfor_fndecl_nca_master;
+tree gfor_fndecl_nca_coarray_allocate;
+tree gfor_fndecl_nca_coarray_free;
+tree gfor_fndecl_nca_this_image;
+tree gfor_fndecl_nca_num_images;
+tree gfor_fndecl_nca_sync_all;
+tree gfor_fndecl_nca_sync_images;
+tree gfor_fndecl_nca_lock;
+tree gfor_fndecl_nca_unlock;
+tree gfor_fndecl_nca_reduce_scalar;
+tree gfor_fndecl_nca_reduce_array;
+tree gfor_fndecl_nca_broadcast_scalar;
+tree gfor_fndecl_nca_broadcast_array;
/* Math functions. Many other math functions are handled in
trans-intrinsic.c. */
@@ -961,6 +976,7 @@ gfc_build_qualified_array (tree decl, gfc_symbol * sym)
tree type;
int dim;
int nest;
+ int eff_dimen;
gfc_namespace* procns;
symbol_attribute *array_attr;
gfc_array_spec *as;
@@ -1031,8 +1047,12 @@ gfc_build_qualified_array (tree decl, gfc_symbol * sym)
else
gfc_add_decl_to_function (token);
}
+
+ eff_dimen = flag_coarray == GFC_FCOARRAY_NATIVE
+ ? GFC_TYPE_ARRAY_RANK (type) + GFC_TYPE_ARRAY_CORANK (type)
+ : GFC_TYPE_ARRAY_RANK (type);
- for (dim = 0; dim < GFC_TYPE_ARRAY_RANK (type); dim++)
+ for (dim = 0; dim < eff_dimen; dim++)
{
if (GFC_TYPE_ARRAY_LBOUND (type, dim) == NULL_TREE)
{
@@ -1054,22 +1074,30 @@ gfc_build_qualified_array (tree decl, gfc_symbol * sym)
TREE_NO_WARNING (GFC_TYPE_ARRAY_STRIDE (type, dim)) = 1;
}
}
- for (dim = GFC_TYPE_ARRAY_RANK (type);
- dim < GFC_TYPE_ARRAY_RANK (type) + GFC_TYPE_ARRAY_CORANK (type); dim++)
- {
- if (GFC_TYPE_ARRAY_LBOUND (type, dim) == NULL_TREE)
- {
- GFC_TYPE_ARRAY_LBOUND (type, dim) = create_index_var ("lbound", nest);
- TREE_NO_WARNING (GFC_TYPE_ARRAY_LBOUND (type, dim)) = 1;
- }
- /* Don't try to use the unknown ubound for the last coarray dimension. */
- if (GFC_TYPE_ARRAY_UBOUND (type, dim) == NULL_TREE
- && dim < GFC_TYPE_ARRAY_RANK (type) + GFC_TYPE_ARRAY_CORANK (type) - 1)
- {
- GFC_TYPE_ARRAY_UBOUND (type, dim) = create_index_var ("ubound", nest);
- TREE_NO_WARNING (GFC_TYPE_ARRAY_UBOUND (type, dim)) = 1;
- }
- }
+
+ if (flag_coarray != GFC_FCOARRAY_NATIVE)
+ for (dim = GFC_TYPE_ARRAY_RANK (type);
+ dim < GFC_TYPE_ARRAY_RANK (type) + GFC_TYPE_ARRAY_CORANK (type);
+ dim++)
+ {
+ if (GFC_TYPE_ARRAY_LBOUND (type, dim) == NULL_TREE)
+ {
+ GFC_TYPE_ARRAY_LBOUND (type, dim)
+ = create_index_var ("lbound", nest);
+ TREE_NO_WARNING (GFC_TYPE_ARRAY_LBOUND (type, dim)) = 1;
+ }
+ /* Don't try to use the unknown ubound for the last coarray
+ dimension. */
+ if (GFC_TYPE_ARRAY_UBOUND (type, dim) == NULL_TREE
+ && dim < GFC_TYPE_ARRAY_RANK (type)
+ + GFC_TYPE_ARRAY_CORANK (type) - 1)
+ {
+ GFC_TYPE_ARRAY_UBOUND (type, dim)
+ = create_index_var ("ubound", nest);
+ TREE_NO_WARNING (GFC_TYPE_ARRAY_UBOUND (type, dim)) = 1;
+ }
+ }
+
if (GFC_TYPE_ARRAY_OFFSET (type) == NULL_TREE)
{
GFC_TYPE_ARRAY_OFFSET (type) = gfc_create_var_np (gfc_array_index_type,
@@ -1202,6 +1230,10 @@ gfc_build_dummy_array_decl (gfc_symbol * sym, tree dummy)
|| (as && as->type == AS_ASSUMED_RANK))
return dummy;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && sym->attr.codimension
+ && sym->attr.allocatable)
+ return dummy;
+
/* Add to list of variables if not a fake result variable.
These symbols are set on the symbol only, not on the class component. */
if (sym->attr.result || sym->attr.dummy)
@@ -1484,7 +1516,6 @@ add_attributes_to_decl (symbol_attribute sym_attr, tree list)
static void build_function_decl (gfc_symbol * sym, bool global);
-
/* Return the decl for a gfc_symbol, create it if it doesn't already
exist. */
@@ -1800,7 +1831,7 @@ gfc_get_symbol_decl (gfc_symbol * sym)
}
/* Remember this variable for allocation/cleanup. */
- if (sym->attr.dimension || sym->attr.allocatable || sym->attr.codimension
+ if (sym->attr.dimension || sym->attr.codimension || sym->attr.allocatable
|| (sym->ts.type == BT_CLASS &&
(CLASS_DATA (sym)->attr.dimension
|| CLASS_DATA (sym)->attr.allocatable))
@@ -1849,6 +1880,9 @@ gfc_get_symbol_decl (gfc_symbol * sym)
gcc_assert (!sym->value || sym->value->expr_type == EXPR_NULL);
}
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && sym->attr.codimension)
+ TREE_STATIC(decl) = 1;
+
gfc_finish_var_decl (decl, sym);
if (sym->ts.type == BT_CHARACTER)
@@ -3668,6 +3702,7 @@ void
gfc_build_builtin_function_decls (void)
{
tree gfc_int8_type_node = gfc_get_int_type (8);
+ tree pint_type = build_pointer_type (integer_type_node);
gfor_fndecl_stop_numeric = gfc_build_library_function_decl (
get_identifier (PREFIX("stop_numeric")),
@@ -3795,9 +3830,8 @@ gfc_build_builtin_function_decls (void)
/* Coarray library calls. */
if (flag_coarray == GFC_FCOARRAY_LIB)
{
- tree pint_type, pppchar_type;
+ tree pppchar_type;
- pint_type = build_pointer_type (integer_type_node);
pppchar_type
= build_pointer_type (build_pointer_type (pchar_type_node));
@@ -4037,6 +4071,64 @@ gfc_build_builtin_function_decls (void)
integer_type_node, 3, pvoid_type_node, integer_type_node,
pvoid_type_node);
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ gfor_fndecl_nca_master = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_master")), ".r", integer_type_node, 1,
+ build_pointer_type (build_function_type_list (void_type_node, NULL_TREE)));
+ gfor_fndecl_nca_coarray_allocate = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_coarray_alloc")), "..RRR", integer_type_node, 4,
+ pvoid_type_node, integer_type_node, integer_type_node, integer_type_node,
+ NULL_TREE);
+ gfor_fndecl_nca_coarray_free = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_coarray_free")), "..RR", integer_type_node, 3,
+ pvoid_type_node, integer_type_node, integer_type_node, NULL_TREE);
+ gfor_fndecl_nca_this_image = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_coarray_this_image")), ".X", integer_type_node, 1,
+ integer_type_node, NULL_TREE);
+ DECL_PURE_P (gfor_fndecl_nca_this_image) = 1;
+ gfor_fndecl_nca_num_images = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_coarray_num_images")), ".X", integer_type_node, 1,
+ integer_type_node, NULL_TREE);
+ DECL_PURE_P (gfor_fndecl_nca_num_images) = 1;
+ gfor_fndecl_nca_sync_all = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_coarray_sync_all")), ".X", void_type_node, 1,
+ build_pointer_type (integer_type_node), NULL_TREE);
+ gfor_fndecl_nca_sync_images = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_sync_images")), ".RRXXX", void_type_node,
+ 5, integer_type_node, pint_type, pint_type,
+ pchar_type_node, size_type_node, NULL_TREE);
+ gfor_fndecl_nca_lock = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_lock")), ".w", void_type_node, 1,
+ pvoid_type_node, NULL_TREE);
+ gfor_fndecl_nca_unlock = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_unlock")), ".w", void_type_node, 1,
+ pvoid_type_node, NULL_TREE);
+
+ gfor_fndecl_nca_reduce_scalar =
+ gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_collsub_reduce_scalar")), ".wrW",
+ void_type_node, 3, pvoid_type_node,
+ build_pointer_type (build_function_type_list (void_type_node,
+ pvoid_type_node, pvoid_type_node, NULL_TREE)),
+ pint_type, NULL_TREE);
+
+ gfor_fndecl_nca_reduce_array =
+ gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_collsub_reduce_array")), ".wrWR",
+ void_type_node, 4, pvoid_type_node,
+ build_pointer_type (build_function_type_list (void_type_node,
+ pvoid_type_node, pvoid_type_node, NULL_TREE)),
+ pint_type, integer_type_node, NULL_TREE);
+
+ gfor_fndecl_nca_broadcast_scalar = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX ("nca_collsub_broadcast_scalar")), ".w..",
+ void_type_node, 3, pvoid_type_node, size_type_node, integer_type_node);
+ gfor_fndecl_nca_broadcast_array = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX ("nca_collsub_broadcast_array")), ".W.",
+ void_type_node, 2, pvoid_type_node, integer_type_node);
+ }
+
gfc_build_intrinsic_function_decls ();
gfc_build_intrinsic_lib_fndecls ();
@@ -4513,6 +4605,76 @@ get_proc_result (gfc_symbol* sym)
}
+void
+gfc_trans_native_coarray (stmtblock_t * init, stmtblock_t *cleanup, gfc_symbol * sym)
+{
+ tree tmp, decl;
+ tree overflow = build_int_cst (integer_type_node, 0), nelems, element_size; //All unused
+ tree offset;
+ tree elem_size;
+ int alloc_type;
+
+ decl = sym->backend_decl;
+
+ TREE_STATIC(decl) = 1;
+
+ /* Tell the library to handle arrays of locks and event types seperatly. */
+ alloc_type = gfc_native_coarray_get_allocation_type (sym);
+
+ if (init)
+ {
+ gfc_array_init_size (decl, sym->as->rank, sym->as->corank, &offset,
+ sym->as->lower, sym->as->upper, init,
+ init, &overflow,
+ NULL_TREE, &nelems, NULL,
+ NULL_TREE, true, NULL, &element_size);
+ gfc_conv_descriptor_offset_set (init, decl, offset);
+ elem_size = size_in_bytes (gfc_get_element_type (TREE_TYPE(decl)));
+ gfc_allocate_native_coarray (init, decl, elem_size, sym->as->corank,
+ alloc_type);
+ }
+
+ if (cleanup)
+ {
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_nca_coarray_free,
+ 2, gfc_build_addr_expr (pvoid_type_node, decl),
+ build_int_cst (integer_type_node, alloc_type),
+ build_int_cst (integer_type_node,
+ sym->as->corank));
+ gfc_add_expr_to_block (cleanup, tmp);
+ }
+}
+
+static void
+finish_coarray_constructor_function (tree *, tree *);
+
+static void
+generate_coarray_constructor_function (tree *, tree *);
+
+static void
+gfc_trans_native_coarray_static (gfc_symbol * sym)
+{
+ tree save_fn_decl, fndecl;
+ generate_coarray_constructor_function (&save_fn_decl, &fndecl);
+ gfc_trans_native_coarray (&caf_init_block, NULL, sym);
+ finish_coarray_constructor_function (&save_fn_decl, &fndecl);
+}
+
+static void
+gfc_trans_native_coarray_inline (gfc_wrapped_block * block, gfc_symbol * sym)
+{
+ stmtblock_t init, cleanup;
+
+ gfc_init_block (&init);
+ gfc_init_block (&cleanup);
+
+ gfc_trans_native_coarray (&init, &cleanup, sym);
+
+ gfc_add_init_cleanup (block, gfc_finish_block (&init), gfc_finish_block (&cleanup));
+}
+
+
+
/* Generate function entry and exit code, and add it to the function body.
This includes:
Allocation and initialization of array variables.
@@ -4808,7 +4970,8 @@ gfc_trans_deferred_vars (gfc_symbol * proc_sym, gfc_wrapped_block * block)
gfc_trans_deferred_array (sym, block);
}
}
- else if (sym->attr.codimension
+ else if (flag_coarray != GFC_FCOARRAY_NATIVE
+ && sym->attr.codimension
&& TREE_STATIC (sym->backend_decl))
{
gfc_init_block (&tmpblock);
@@ -4818,6 +4981,11 @@ gfc_trans_deferred_vars (gfc_symbol * proc_sym, gfc_wrapped_block * block)
NULL_TREE);
continue;
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE
+ && sym->attr.codimension)
+ {
+ gfc_trans_native_coarray_inline (block, sym);
+ }
else
{
gfc_save_backend_locus (&loc);
@@ -5308,6 +5476,10 @@ gfc_create_module_variable (gfc_symbol * sym)
&& sym->fn_result_spec));
DECL_CONTEXT (decl) = sym->ns->proc_name->backend_decl;
rest_of_decl_compilation (decl, 1, 0);
+
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && sym->attr.codimension)
+ gfc_trans_native_coarray_static (sym);
+
gfc_module_add_decl (cur_module, decl);
/* Also add length of strings. */
@@ -5705,64 +5877,82 @@ generate_coarray_sym_init (gfc_symbol *sym)
}
-/* Generate constructor function to initialize static, nonallocatable
- coarrays. */
static void
-generate_coarray_init (gfc_namespace * ns __attribute((unused)))
+generate_coarray_constructor_function (tree *save_fn_decl, tree *fndecl)
{
- tree fndecl, tmp, decl, save_fn_decl;
+ tree tmp, decl;
- save_fn_decl = current_function_decl;
+ *save_fn_decl = current_function_decl;
push_function_context ();
tmp = build_function_type_list (void_type_node, NULL_TREE);
- fndecl = build_decl (input_location, FUNCTION_DECL,
- create_tmp_var_name ("_caf_init"), tmp);
+ *fndecl = build_decl (input_location, FUNCTION_DECL,
+ create_tmp_var_name (flag_coarray == GFC_FCOARRAY_LIB ? "_caf_init" : "_nca_init"), tmp);
- DECL_STATIC_CONSTRUCTOR (fndecl) = 1;
- SET_DECL_INIT_PRIORITY (fndecl, DEFAULT_INIT_PRIORITY);
+ DECL_STATIC_CONSTRUCTOR (*fndecl) = 1;
+ SET_DECL_INIT_PRIORITY (*fndecl, DEFAULT_INIT_PRIORITY);
decl = build_decl (input_location, RESULT_DECL, NULL_TREE, void_type_node);
DECL_ARTIFICIAL (decl) = 1;
DECL_IGNORED_P (decl) = 1;
- DECL_CONTEXT (decl) = fndecl;
- DECL_RESULT (fndecl) = decl;
+ DECL_CONTEXT (decl) = *fndecl;
+ DECL_RESULT (*fndecl) = decl;
- pushdecl (fndecl);
- current_function_decl = fndecl;
- announce_function (fndecl);
+ pushdecl (*fndecl);
+ current_function_decl = *fndecl;
+ announce_function (*fndecl);
- rest_of_decl_compilation (fndecl, 0, 0);
- make_decl_rtl (fndecl);
- allocate_struct_function (fndecl, false);
+ rest_of_decl_compilation (*fndecl, 0, 0);
+ make_decl_rtl (*fndecl);
+ allocate_struct_function (*fndecl, false);
pushlevel ();
gfc_init_block (&caf_init_block);
+}
- gfc_traverse_ns (ns, generate_coarray_sym_init);
+static void
+finish_coarray_constructor_function (tree *save_fn_decl, tree *fndecl)
+{
+ tree decl;
- DECL_SAVED_TREE (fndecl) = gfc_finish_block (&caf_init_block);
+ DECL_SAVED_TREE (*fndecl) = gfc_finish_block (&caf_init_block);
decl = getdecls ();
poplevel (1, 1);
- BLOCK_SUPERCONTEXT (DECL_INITIAL (fndecl)) = fndecl;
+ BLOCK_SUPERCONTEXT (DECL_INITIAL (*fndecl)) = *fndecl;
- DECL_SAVED_TREE (fndecl)
- = build3_v (BIND_EXPR, decl, DECL_SAVED_TREE (fndecl),
- DECL_INITIAL (fndecl));
- dump_function (TDI_original, fndecl);
+ DECL_SAVED_TREE (*fndecl)
+ = build3_v (BIND_EXPR, decl, DECL_SAVED_TREE (*fndecl),
+ DECL_INITIAL (*fndecl));
+ dump_function (TDI_original, *fndecl);
cfun->function_end_locus = input_location;
set_cfun (NULL);
- if (decl_function_context (fndecl))
- (void) cgraph_node::create (fndecl);
+ if (decl_function_context (*fndecl))
+ (void) cgraph_node::create (*fndecl);
else
- cgraph_node::finalize_function (fndecl, true);
+ cgraph_node::finalize_function (*fndecl, true);
pop_function_context ();
- current_function_decl = save_fn_decl;
+ current_function_decl = *save_fn_decl;
+}
+
+/* Generate constructor function to initialize static, nonallocatable
+ coarrays. */
+
+static void
+generate_coarray_init (gfc_namespace * ns)
+{
+ tree save_fn_decl, fndecl;
+
+ generate_coarray_constructor_function (&save_fn_decl, &fndecl);
+
+ gfc_traverse_ns (ns, generate_coarray_sym_init);
+
+ finish_coarray_constructor_function (&save_fn_decl, &fndecl);
+
}
@@ -6445,7 +6635,11 @@ create_main_function (tree fndecl)
}
/* Call MAIN__(). */
- tmp = build_call_expr_loc (input_location,
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_nca_master, 1,
+ gfc_build_addr_expr (NULL, fndecl));
+ else
+ tmp = build_call_expr_loc (input_location,
fndecl, 0);
gfc_add_expr_to_block (&body, tmp);
diff --git a/gcc/fortran/trans-expr.c b/gcc/fortran/trans-expr.c
index b7c568e90e6..cd776ab325d 100644
--- a/gcc/fortran/trans-expr.c
+++ b/gcc/fortran/trans-expr.c
@@ -2622,8 +2622,14 @@ gfc_maybe_dereference_var (gfc_symbol *sym, tree var, bool descriptor_only_p,
}
else if (!sym->attr.value)
{
+
+ /* Do not derefernce native coarray dummies. */
+ if (false && flag_coarray == GFC_FCOARRAY_NATIVE
+ && sym->attr.codimension && sym->attr.dummy)
+ return var;
+
/* Dereference temporaries for class array dummy arguments. */
- if (sym->attr.dummy && is_classarray
+ else if (sym->attr.dummy && is_classarray
&& GFC_ARRAY_TYPE_P (TREE_TYPE (var)))
{
if (!descriptor_only_p)
@@ -2635,6 +2641,7 @@ gfc_maybe_dereference_var (gfc_symbol *sym, tree var, bool descriptor_only_p,
/* Dereference non-character scalar dummy arguments. */
if (sym->attr.dummy && !sym->attr.dimension
&& !(sym->attr.codimension && sym->attr.allocatable)
+ && !(sym->attr.codimension && flag_coarray == GFC_FCOARRAY_NATIVE)
&& (sym->ts.type != BT_CLASS
|| (!CLASS_DATA (sym)->attr.dimension
&& !(CLASS_DATA (sym)->attr.codimension
@@ -2670,6 +2677,7 @@ gfc_maybe_dereference_var (gfc_symbol *sym, tree var, bool descriptor_only_p,
|| CLASS_DATA (sym)->attr.allocatable
|| CLASS_DATA (sym)->attr.class_pointer))
var = build_fold_indirect_ref_loc (input_location, var);
+
/* And the case where a non-dummy, non-result, non-function,
non-allotable and non-pointer classarray is present. This case was
previously covered by the first if, but with introducing the
@@ -5528,7 +5536,10 @@ gfc_conv_procedure_call (gfc_se * se, gfc_symbol * sym,
nodesc_arg = nodesc_arg || !comp->attr.always_explicit;
else
nodesc_arg = nodesc_arg || !sym->attr.always_explicit;
-
+#if 0
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && fsym->attr.codimension)
+ nodesc_arg = false;
+#endif
/* Class array expressions are sometimes coming completely unadorned
with either arrayspec or _data component. Correct that here.
OOP-TODO: Move this to the frontend. */
@@ -5720,7 +5731,10 @@ gfc_conv_procedure_call (gfc_se * se, gfc_symbol * sym,
parmse.want_coarray = 1;
scalar = false;
}
-
+#if 0
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && fsym->attr.codimension)
+ scalar = false;
+#endif
/* A scalar or transformational function. */
if (scalar)
{
@@ -6233,7 +6247,7 @@ gfc_conv_procedure_call (gfc_se * se, gfc_symbol * sym,
else
gfc_conv_array_parameter (&parmse, e, nodesc_arg, fsym,
sym->name, NULL);
-
+
/* Unallocated allocatable arrays and unassociated pointer arrays
need their dtype setting if they are argument associated with
assumed rank dummies. */
diff --git a/gcc/fortran/trans-intrinsic.c b/gcc/fortran/trans-intrinsic.c
index fd8809902b7..8d3a52dc170 100644
--- a/gcc/fortran/trans-intrinsic.c
+++ b/gcc/fortran/trans-intrinsic.c
@@ -41,6 +41,7 @@ along with GCC; see the file COPYING3. If not see
#include "trans-array.h"
#include "dependency.h" /* For CAF array alias analysis. */
/* Only for gfc_trans_assign and gfc_trans_pointer_assign. */
+#include "trans-stmt.h"
/* This maps Fortran intrinsic math functions to external library or GCC
builtin functions. */
@@ -2350,7 +2351,6 @@ conv_caf_send (gfc_code *code) {
return gfc_finish_block (&block);
}
-
static void
trans_this_image (gfc_se * se, gfc_expr *expr)
{
@@ -2381,14 +2381,18 @@ trans_this_image (gfc_se * se, gfc_expr *expr)
}
else
tmp = integer_zero_node;
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_this_image, 1,
- tmp);
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE ?
+ gfor_fndecl_nca_this_image :
+ gfor_fndecl_caf_this_image,
+ 1, tmp);
se->expr = fold_convert (gfc_get_int_type (gfc_default_integer_kind),
tmp);
return;
}
/* Coarray-argument version: THIS_IMAGE(coarray [, dim]). */
+ /* TODO: NCA handle native coarrays. */
type = gfc_get_int_type (gfc_default_integer_kind);
corank = gfc_get_corank (expr->value.function.actual->expr);
@@ -2477,8 +2481,11 @@ trans_this_image (gfc_se * se, gfc_expr *expr)
*/
/* this_image () - 1. */
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_this_image, 1,
- integer_zero_node);
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE
+ ? gfor_fndecl_nca_this_image
+ : gfor_fndecl_caf_this_image,
+ 1, integer_zero_node);
tmp = fold_build2_loc (input_location, MINUS_EXPR, type,
fold_convert (type, tmp), build_int_cst (type, 1));
if (corank == 1)
@@ -2761,7 +2768,10 @@ trans_image_index (gfc_se * se, gfc_expr *expr)
num_images = build_int_cst (type, 1);
else
{
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_num_images, 2,
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE
+ ? gfor_fndecl_nca_num_images
+ : gfor_fndecl_caf_num_images, 2,
integer_zero_node,
build_int_cst (integer_type_node, -1));
num_images = fold_convert (type, tmp);
@@ -2806,8 +2816,13 @@ trans_num_images (gfc_se * se, gfc_expr *expr)
}
else
failed = build_int_cst (integer_type_node, -1);
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_num_images, 2,
- distance, failed);
+
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_nca_num_images, 1,
+ distance);
+ else
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_num_images, 2,
+ distance, failed);
se->expr = fold_convert (gfc_get_int_type (gfc_default_integer_kind), tmp);
}
@@ -3251,7 +3266,10 @@ conv_intrinsic_cobound (gfc_se * se, gfc_expr * expr)
tree cosize;
cosize = gfc_conv_descriptor_cosize (desc, arg->expr->rank, corank);
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_num_images,
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE
+ ? gfor_fndecl_nca_num_images
+ : gfor_fndecl_caf_num_images,
2, integer_zero_node,
build_int_cst (integer_type_node, -1));
tmp = fold_build2_loc (input_location, MINUS_EXPR,
@@ -3267,7 +3285,9 @@ conv_intrinsic_cobound (gfc_se * se, gfc_expr * expr)
else if (flag_coarray != GFC_FCOARRAY_SINGLE)
{
/* ubound = lbound + num_images() - 1. */
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_num_images,
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE ? gfor_fndecl_nca_num_images :
+ gfor_fndecl_caf_num_images,
2, integer_zero_node,
build_int_cst (integer_type_node, -1));
tmp = fold_build2_loc (input_location, MINUS_EXPR,
@@ -10979,6 +10999,136 @@ gfc_walk_intrinsic_function (gfc_ss * ss, gfc_expr * expr,
}
}
+/* Helper function - advance to the next argument. */
+
+static tree
+trans_argument (gfc_actual_arglist **curr_al, stmtblock_t *blk,
+ stmtblock_t *postblk, gfc_se *argse, tree def)
+{
+ if (!(*curr_al)->expr)
+ return def;
+ if ((*curr_al)->expr->rank > 0)
+ gfc_conv_expr_descriptor (argse, (*curr_al)->expr);
+ else
+ gfc_conv_expr (argse, (*curr_al)->expr);
+ gfc_add_block_to_block (blk, &argse->pre);
+ gfc_add_block_to_block (postblk, &argse->post);
+ *curr_al = (*curr_al)->next;
+ return argse->expr;
+}
+
+/* Convert CO_REDUCE for native coarrays. */
+
+static tree
+conv_nca_reduce (gfc_code *code, stmtblock_t *blk, stmtblock_t *postblk)
+{
+ gfc_actual_arglist *curr_al;
+ tree var, reduce_op, result_image, elem_size;
+ gfc_se argse;
+ int is_array;
+
+ curr_al = code->ext.actual;
+
+ gfc_init_se (&argse, NULL);
+ argse.want_pointer = 1;
+ is_array = curr_al->expr->rank > 0;
+ var = trans_argument (&curr_al, blk, postblk, &argse, NULL_TREE);
+
+ gfc_init_se (&argse, NULL);
+ argse.want_pointer = 1;
+ reduce_op = trans_argument (&curr_al, blk, postblk, &argse, NULL_TREE);
+
+ gfc_init_se (&argse, NULL);
+ argse.want_pointer = 1;
+ result_image = trans_argument (&curr_al, blk, postblk, &argse,
+ null_pointer_node);
+
+ if (is_array)
+ return build_call_expr_loc (input_location, gfor_fndecl_nca_reduce_array,
+ 3, var, reduce_op, result_image);
+
+ elem_size = size_in_bytes(TREE_TYPE(TREE_TYPE(var)));
+ return build_call_expr_loc (input_location, gfor_fndecl_nca_reduce_scalar, 4,
+ var, elem_size, reduce_op, result_image);
+}
+
+static tree
+conv_nca_broadcast (gfc_code *code, stmtblock_t *blk, stmtblock_t *postblk)
+{
+ gfc_actual_arglist *curr_al;
+ tree var, source_image, elem_size;
+ gfc_se argse;
+ int is_array;
+
+ curr_al = code->ext.actual;
+
+ gfc_init_se (&argse, NULL);
+ argse.want_pointer = 1;
+ is_array = curr_al->expr->rank > 0;
+ var = trans_argument (&curr_al, blk, postblk, &argse, NULL_TREE);
+
+ gfc_init_se (&argse, NULL);
+ argse.want_pointer = 0;
+ source_image = trans_argument (&curr_al, blk, postblk, &argse, NULL_TREE);
+
+ if (is_array)
+ return build_call_expr_loc (input_location, gfor_fndecl_nca_broadcast_array,
+ 2, var, source_image);
+
+ elem_size = size_in_bytes(TREE_TYPE(TREE_TYPE(var)));
+ return build_call_expr_loc (input_location, gfor_fndecl_nca_broadcast_scalar,
+ 3, var, elem_size, source_image);
+}
+
+static tree conv_co_collective (gfc_code *);
+
+/* Convert collective subroutines for native coarrays. */
+
+static tree
+conv_nca_collective (gfc_code *code)
+{
+
+ switch (code->resolved_isym->id)
+ {
+ case GFC_ISYM_CO_REDUCE:
+ {
+ stmtblock_t block, postblock;
+ tree fcall;
+
+ gfc_start_block (&block);
+ gfc_init_block (&postblock);
+ fcall = conv_nca_reduce (code, &block, &postblock);
+ gfc_add_expr_to_block (&block, fcall);
+ gfc_add_block_to_block (&block, &postblock);
+ return gfc_finish_block (&block);
+ }
+ case GFC_ISYM_CO_SUM:
+ case GFC_ISYM_CO_MIN:
+ case GFC_ISYM_CO_MAX:
+ return gfc_trans_call (code, false, NULL_TREE, NULL_TREE, false);
+
+ case GFC_ISYM_CO_BROADCAST:
+ {
+ stmtblock_t block, postblock;
+ tree fcall;
+
+ gfc_start_block (&block);
+ gfc_init_block (&postblock);
+ fcall = conv_nca_broadcast (code, &block, &postblock);
+ gfc_add_expr_to_block (&block, fcall);
+ gfc_add_block_to_block (&block, &postblock);
+ return gfc_finish_block (&block);
+ }
+#if 0
+ case GFC_ISYM_CO_BROADCAST:
+ return conv_co_collective (code);
+#endif
+ default:
+ gfc_internal_error ("Invalid or unsupported isym");
+ break;
+ }
+}
+
static tree
conv_co_collective (gfc_code *code)
{
@@ -11086,7 +11236,13 @@ conv_co_collective (gfc_code *code)
errmsg_len = build_zero_cst (size_type_node);
}
+ /* For native coarrays, we only come here for CO_BROADCAST. */
+
+ gcc_assert (code->resolved_isym->id == GFC_ISYM_CO_BROADCAST
+ || flag_coarray != GFC_FCOARRAY_NATIVE);
+
/* Generate the function call. */
+
switch (code->resolved_isym->id)
{
case GFC_ISYM_CO_BROADCAST:
@@ -12079,7 +12235,10 @@ gfc_conv_intrinsic_subroutine (gfc_code *code)
case GFC_ISYM_CO_MAX:
case GFC_ISYM_CO_REDUCE:
case GFC_ISYM_CO_SUM:
- res = conv_co_collective (code);
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ res = conv_nca_collective (code);
+ else
+ res = conv_co_collective (code);
break;
case GFC_ISYM_FREE:
diff --git a/gcc/fortran/trans-stmt.c b/gcc/fortran/trans-stmt.c
index 54b56c4f01d..017757b5f44 100644
--- a/gcc/fortran/trans-stmt.c
+++ b/gcc/fortran/trans-stmt.c
@@ -831,7 +831,9 @@ gfc_trans_lock_unlock (gfc_code *code, gfc_exec_op op)
/* Short cut: For single images without STAT= or LOCK_ACQUIRED
return early. (ERRMSG= is always untouched for -fcoarray=single.) */
- if (!code->expr2 && !code->expr4 && flag_coarray != GFC_FCOARRAY_LIB)
+ if (!code->expr2 && !code->expr4
+ && !(flag_coarray == GFC_FCOARRAY_LIB
+ || flag_coarray == GFC_FCOARRAY_NATIVE))
return NULL_TREE;
if (code->expr2)
@@ -991,6 +993,29 @@ gfc_trans_lock_unlock (gfc_code *code, gfc_exec_op op)
return gfc_finish_block (&se.pre);
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ gfc_se arg;
+ stmtblock_t res;
+ tree call;
+ tree tmp;
+
+ gfc_init_se (&arg, NULL);
+ gfc_start_block (&res);
+ gfc_conv_expr (&arg, code->expr1);
+ gfc_add_block_to_block (&res, &arg.pre);
+ call = build_call_expr_loc (input_location, op == EXEC_LOCK ?
+ gfor_fndecl_nca_lock
+ : gfor_fndecl_nca_unlock,
+ 1, fold_convert (pvoid_type_node,
+ gfc_build_addr_expr (NULL, arg.expr)));
+ gfc_add_expr_to_block (&res, call);
+ gfc_add_block_to_block (&res, &arg.post);
+ tmp = gfc_trans_memory_barrier ();
+ gfc_add_expr_to_block (&res, tmp);
+
+ return gfc_finish_block (&res);
+ }
if (stat != NULL_TREE)
gfc_add_modify (&se.pre, stat, build_int_cst (TREE_TYPE (stat), 0));
@@ -1184,7 +1209,8 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
/* Short cut: For single images without bound checking or without STAT=,
return early. (ERRMSG= is always untouched for -fcoarray=single.) */
if (!code->expr2 && !(gfc_option.rtcheck & GFC_RTCHECK_BOUNDS)
- && flag_coarray != GFC_FCOARRAY_LIB)
+ && flag_coarray != GFC_FCOARRAY_LIB
+ && flag_coarray != GFC_FCOARRAY_NATIVE)
return NULL_TREE;
gfc_init_se (&se, NULL);
@@ -1207,7 +1233,7 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
else
stat = null_pointer_node;
- if (code->expr3 && flag_coarray == GFC_FCOARRAY_LIB)
+ if (code->expr3 && (flag_coarray == GFC_FCOARRAY_LIB || flag_coarray == GFC_FCOARRAY_NATIVE))
{
gcc_assert (code->expr3->expr_type == EXPR_VARIABLE);
gfc_init_se (&argse, NULL);
@@ -1217,7 +1243,7 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
errmsg = gfc_build_addr_expr (NULL, argse.expr);
errmsglen = fold_convert (size_type_node, argse.string_length);
}
- else if (flag_coarray == GFC_FCOARRAY_LIB)
+ else if (flag_coarray == GFC_FCOARRAY_LIB || flag_coarray == GFC_FCOARRAY_NATIVE)
{
errmsg = null_pointer_node;
errmsglen = build_int_cst (size_type_node, 0);
@@ -1230,7 +1256,7 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
{
tree images2 = fold_convert (integer_type_node, images);
tree cond;
- if (flag_coarray != GFC_FCOARRAY_LIB)
+ if (flag_coarray != GFC_FCOARRAY_LIB && flag_coarray != GFC_FCOARRAY_NATIVE)
cond = fold_build2_loc (input_location, NE_EXPR, logical_type_node,
images, build_int_cst (TREE_TYPE (images), 1));
else
@@ -1254,17 +1280,13 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
/* Per F2008, 8.5.1, a SYNC MEMORY is implied by calling the
image control statements SYNC IMAGES and SYNC ALL. */
- if (flag_coarray == GFC_FCOARRAY_LIB)
+ if (flag_coarray == GFC_FCOARRAY_LIB || flag_coarray == GFC_FCOARRAY_NATIVE)
{
- tmp = gfc_build_string_const (strlen ("memory")+1, "memory"),
- tmp = build5_loc (input_location, ASM_EXPR, void_type_node,
- gfc_build_string_const (1, ""), NULL_TREE, NULL_TREE,
- tree_cons (NULL_TREE, tmp, NULL_TREE), NULL_TREE);
- ASM_VOLATILE_P (tmp) = 1;
+ tmp = gfc_trans_memory_barrier ();
gfc_add_expr_to_block (&se.pre, tmp);
}
- if (flag_coarray != GFC_FCOARRAY_LIB)
+ if (flag_coarray != GFC_FCOARRAY_LIB && flag_coarray != GFC_FCOARRAY_NATIVE)
{
/* Set STAT to zero. */
if (code->expr2)
@@ -1286,8 +1308,14 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_memory,
3, stat, errmsg, errmsglen);
else
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_all,
- 3, stat, errmsg, errmsglen);
+ {
+ if (flag_coarray == GFC_FCOARRAY_LIB)
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_all,
+ 3, stat, errmsg, errmsglen);
+ else
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_nca_sync_all,
+ 1, stat);
+ }
gfc_add_expr_to_block (&se.pre, tmp);
}
@@ -1352,7 +1380,10 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
if (TREE_TYPE (stat) == integer_type_node)
stat = gfc_build_addr_expr (NULL, stat);
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_images,
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE
+ ? gfor_fndecl_nca_sync_images
+ : gfor_fndecl_caf_sync_images,
5, fold_convert (integer_type_node, len),
images, stat, errmsg, errmsglen);
gfc_add_expr_to_block (&se.pre, tmp);
@@ -1361,7 +1392,10 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
{
tree tmp_stat = gfc_create_var (integer_type_node, "stat");
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_images,
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE
+ ? gfor_fndecl_nca_sync_images
+ : gfor_fndecl_caf_sync_images,
5, fold_convert (integer_type_node, len),
images, gfc_build_addr_expr (NULL, tmp_stat),
errmsg, errmsglen);
@@ -1597,6 +1631,11 @@ gfc_trans_critical (gfc_code *code)
gfc_add_expr_to_block (&block, tmp);
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ tmp = gfc_trans_lock_unlock (code, EXEC_LOCK);
+ gfc_add_expr_to_block (&block, tmp);
+ }
tmp = gfc_trans_code (code->block->next);
gfc_add_expr_to_block (&block, tmp);
@@ -1621,6 +1660,11 @@ gfc_trans_critical (gfc_code *code)
gfc_add_expr_to_block (&block, tmp);
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ tmp = gfc_trans_lock_unlock (code, EXEC_UNLOCK);
+ gfc_add_expr_to_block (&block, tmp);
+ }
return gfc_finish_block (&block);
}
@@ -7170,6 +7214,7 @@ gfc_trans_deallocate (gfc_code *code)
tree apstat, pstat, stat, errmsg, errlen, tmp;
tree label_finish, label_errmsg;
stmtblock_t block;
+ bool is_native_coarray = false;
pstat = apstat = stat = errmsg = errlen = tmp = NULL_TREE;
label_finish = label_errmsg = NULL_TREE;
@@ -7255,8 +7300,27 @@ gfc_trans_deallocate (gfc_code *code)
? GFC_STRUCTURE_CAF_MODE_DEALLOC_ONLY : 0);
}
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ gfc_ref *ref, *last;
- if (expr->rank || is_coarray_array)
+ for (ref = expr->ref, last = ref; ref; last = ref, ref = ref->next);
+ ref = last;
+ if (ref->type == REF_ARRAY && ref->u.ar.codimen)
+ {
+ gfc_symbol *sym = expr->symtree->n.sym;
+ int alloc_type = gfc_native_coarray_get_allocation_type (sym);
+ tmp = build_call_expr_loc (input_location,
+ gfor_fndecl_nca_coarray_free,
+ 2, gfc_build_addr_expr (pvoid_type_node, se.expr),
+ build_int_cst (integer_type_node,
+ alloc_type));
+ gfc_add_expr_to_block (&block, tmp);
+ is_native_coarray = true;
+ }
+ }
+
+ if ((expr->rank || is_coarray_array) && !is_native_coarray)
{
gfc_ref *ref;
@@ -7345,7 +7409,7 @@ gfc_trans_deallocate (gfc_code *code)
gfc_reset_len (&se.pre, al->expr);
}
}
- else
+ else if (!is_native_coarray)
{
tmp = gfc_deallocate_scalar_with_status (se.expr, pstat, label_finish,
false, al->expr,
diff --git a/gcc/fortran/trans-types.c b/gcc/fortran/trans-types.c
index 99844812505..7ed4209b2e3 100644
--- a/gcc/fortran/trans-types.c
+++ b/gcc/fortran/trans-types.c
@@ -1345,6 +1345,10 @@ gfc_is_nodesc_array (gfc_symbol * sym)
gcc_assert (array_attr->dimension || array_attr->codimension);
+ /* We need a descriptor for native coarrays. */
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && sym->as && sym->as->corank)
+ return 0;
+
/* We only want local arrays. */
if ((sym->ts.type != BT_CLASS && sym->attr.pointer)
|| (sym->ts.type == BT_CLASS && CLASS_DATA (sym)->attr.class_pointer)
@@ -1381,12 +1385,18 @@ gfc_build_array_type (tree type, gfc_array_spec * as,
tree ubound[GFC_MAX_DIMENSIONS];
int n, corank;
- /* Assumed-shape arrays do not have codimension information stored in the
- descriptor. */
- corank = MAX (as->corank, codim);
- if (as->type == AS_ASSUMED_SHAPE ||
- (as->type == AS_ASSUMED_RANK && akind == GFC_ARRAY_ALLOCATABLE))
- corank = codim;
+ /* For -fcoarray=lib, assumed-shape arrays do not have codimension
+ information stored in the descriptor. */
+ if (flag_coarray != GFC_FCOARRAY_NATIVE)
+ {
+ corank = MAX (as->corank, codim);
+
+ if (as->type == AS_ASSUMED_SHAPE ||
+ (as->type == AS_ASSUMED_RANK && akind == GFC_ARRAY_ALLOCATABLE))
+ corank = codim;
+ }
+ else
+ corank = as->corank;
if (as->type == AS_ASSUMED_RANK)
for (n = 0; n < GFC_MAX_DIMENSIONS; n++)
@@ -1427,7 +1437,7 @@ gfc_build_array_type (tree type, gfc_array_spec * as,
corank, lbound, ubound, 0, akind,
restricted);
}
-\f
+
/* Returns the struct descriptor_dimension type. */
static tree
@@ -1598,7 +1608,7 @@ gfc_get_nodesc_array_type (tree etype, gfc_array_spec * as, gfc_packed packed,
/* We don't use build_array_type because this does not include
lang-specific information (i.e. the bounds of the array) when checking
for duplicates. */
- if (as->rank)
+ if (as->rank || (flag_coarray == GFC_FCOARRAY_NATIVE && as->corank))
type = make_node (ARRAY_TYPE);
else
type = build_variant_type_copy (etype);
@@ -1665,6 +1675,7 @@ gfc_get_nodesc_array_type (tree etype, gfc_array_spec * as, gfc_packed packed,
if (packed == PACKED_NO || packed == PACKED_PARTIAL)
known_stride = 0;
}
+
for (n = as->rank; n < as->rank + as->corank; n++)
{
expr = as->lower[n];
@@ -1672,7 +1683,7 @@ gfc_get_nodesc_array_type (tree etype, gfc_array_spec * as, gfc_packed packed,
tmp = gfc_conv_mpz_to_tree (expr->value.integer,
gfc_index_integer_kind);
else
- tmp = NULL_TREE;
+ tmp = NULL_TREE;
GFC_TYPE_ARRAY_LBOUND (type, n) = tmp;
expr = as->upper[n];
@@ -1680,16 +1691,16 @@ gfc_get_nodesc_array_type (tree etype, gfc_array_spec * as, gfc_packed packed,
tmp = gfc_conv_mpz_to_tree (expr->value.integer,
gfc_index_integer_kind);
else
- tmp = NULL_TREE;
+ tmp = NULL_TREE;
if (n < as->rank + as->corank - 1)
- GFC_TYPE_ARRAY_UBOUND (type, n) = tmp;
+ GFC_TYPE_ARRAY_UBOUND (type, n) = tmp;
}
- if (known_offset)
- {
- GFC_TYPE_ARRAY_OFFSET (type) =
- gfc_conv_mpz_to_tree (offset, gfc_index_integer_kind);
- }
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && as->rank == 0 && as->corank != 0)
+ GFC_TYPE_ARRAY_OFFSET (type) = NULL_TREE;
+ else if (known_offset)
+ GFC_TYPE_ARRAY_OFFSET (type) =
+ gfc_conv_mpz_to_tree (offset, gfc_index_integer_kind);
else
GFC_TYPE_ARRAY_OFFSET (type) = NULL_TREE;
@@ -1714,7 +1725,7 @@ gfc_get_nodesc_array_type (tree etype, gfc_array_spec * as, gfc_packed packed,
build_qualified_type (GFC_TYPE_ARRAY_DATAPTR_TYPE (type),
TYPE_QUAL_RESTRICT);
- if (as->rank == 0)
+ if (as->rank == 0 && (flag_coarray != GFC_FCOARRAY_NATIVE || as->corank == 0))
{
if (packed != PACKED_STATIC || flag_coarray == GFC_FCOARRAY_LIB)
{
@@ -1982,7 +1993,7 @@ gfc_get_array_type_bounds (tree etype, int dimen, int codimen, tree * lbound,
/* TODO: known offsets for descriptors. */
GFC_TYPE_ARRAY_OFFSET (fat_type) = NULL_TREE;
- if (dimen == 0)
+ if (flag_coarray != GFC_FCOARRAY_NATIVE && dimen == 0)
{
arraytype = build_pointer_type (etype);
if (restricted)
@@ -2281,6 +2292,10 @@ gfc_sym_type (gfc_symbol * sym)
: GFC_ARRAY_POINTER;
else if (sym->attr.allocatable)
akind = GFC_ARRAY_ALLOCATABLE;
+
+ /* FIXME: For normal coarrays, we pass a bool to an int here.
+ Is this really intended? */
+
type = gfc_build_array_type (type, sym->as, akind, restricted,
sym->attr.contiguous, false);
}
diff --git a/gcc/fortran/trans.c b/gcc/fortran/trans.c
index ed054261452..2b605505445 100644
--- a/gcc/fortran/trans.c
+++ b/gcc/fortran/trans.c
@@ -47,6 +47,21 @@ static gfc_file *gfc_current_backend_file;
const char gfc_msg_fault[] = N_("Array reference out of bounds");
const char gfc_msg_wrong_return[] = N_("Incorrect function return value");
+/* Insert a memory barrier into the code. */
+
+tree
+gfc_trans_memory_barrier (void)
+{
+ tree tmp;
+
+ tmp = gfc_build_string_const (strlen ("memory")+1, "memory"),
+ tmp = build5_loc (input_location, ASM_EXPR, void_type_node,
+ gfc_build_string_const (1, ""), NULL_TREE, NULL_TREE,
+ tree_cons (NULL_TREE, tmp, NULL_TREE), NULL_TREE);
+ ASM_VOLATILE_P (tmp) = 1;
+
+ return tmp;
+}
/* Return a location_t suitable for 'tree' for a gfortran locus. The way the
parser works in gfortran, loc->lb->location contains only the line number
@@ -403,15 +418,16 @@ gfc_build_array_ref (tree base, tree offset, tree decl, tree vptr)
tree tmp;
tree span = NULL_TREE;
- if (GFC_ARRAY_TYPE_P (type) && GFC_TYPE_ARRAY_RANK (type) == 0)
+ if (GFC_ARRAY_TYPE_P (type) && GFC_TYPE_ARRAY_RANK (type) == 0
+ && flag_coarray != GFC_FCOARRAY_NATIVE)
{
gcc_assert (GFC_TYPE_ARRAY_CORANK (type) > 0);
return fold_convert (TYPE_MAIN_VARIANT (type), base);
}
- /* Scalar coarray, there is nothing to do. */
- if (TREE_CODE (type) != ARRAY_TYPE)
+ /* Scalar library coarray, there is nothing to do. */
+ if (TREE_CODE (type) != ARRAY_TYPE && flag_coarray != GFC_FCOARRAY_NATIVE)
{
gcc_assert (decl == NULL_TREE);
gcc_assert (integer_zerop (offset));
diff --git a/gcc/fortran/trans.h b/gcc/fortran/trans.h
index e126fe92782..bb4674166d5 100644
--- a/gcc/fortran/trans.h
+++ b/gcc/fortran/trans.h
@@ -501,6 +501,9 @@ void gfc_conv_expr_reference (gfc_se * se, gfc_expr * expr,
bool add_clobber = false);
void gfc_conv_expr_type (gfc_se * se, gfc_expr *, tree);
+/* Insert a memory barrier into the code. */
+
+tree gfc_trans_memory_barrier (void);
/* trans-expr.c */
tree gfc_conv_scalar_to_descriptor (gfc_se *, tree, symbol_attribute);
@@ -890,6 +893,21 @@ extern GTY(()) tree gfor_fndecl_co_reduce;
extern GTY(()) tree gfor_fndecl_co_sum;
extern GTY(()) tree gfor_fndecl_caf_is_present;
+
+/* Native coarray library function decls. */
+extern GTY(()) tree gfor_fndecl_nca_this_image;
+extern GTY(()) tree gfor_fndecl_nca_num_images;
+extern GTY(()) tree gfor_fndecl_nca_coarray_allocate;
+extern GTY(()) tree gfor_fndecl_nca_coarray_free;
+extern GTY(()) tree gfor_fndecl_nca_sync_images;
+extern GTY(()) tree gfor_fndecl_nca_sync_all;
+extern GTY(()) tree gfor_fndecl_nca_lock;
+extern GTY(()) tree gfor_fndecl_nca_unlock;
+extern GTY(()) tree gfor_fndecl_nca_reduce_scalar;
+extern GTY(()) tree gfor_fndecl_nca_reduce_array;
+extern GTY(()) tree gfor_fndecl_nca_broadcast_scalar;
+extern GTY(()) tree gfor_fndecl_nca_broadcast_array;
+
/* Math functions. Many other math functions are handled in
trans-intrinsic.c. */
diff --git a/libgfortran/Makefile.am b/libgfortran/Makefile.am
index 36b204e1aa3..c404ef8ccb4 100644
--- a/libgfortran/Makefile.am
+++ b/libgfortran/Makefile.am
@@ -36,14 +36,21 @@ gfor_cdir = $(libdir)/gcc/$(target_alias)/$(gcc_version)/include
LTLDFLAGS = $(shell $(SHELL) $(top_srcdir)/../libtool-ldflags $(LDFLAGS)) \
$(lt_host_flags)
+if LIBGFOR_NATIVE_COARRAY
+COARRAY_LIBS=$(PTHREAD_LIBS) $(RT_LIBS)
+else
+COARRAY_LIBS=""
+endif
+
toolexeclib_LTLIBRARIES = libgfortran.la
toolexeclib_DATA = libgfortran.spec
libgfortran_la_LINK = $(LINK) $(libgfortran_la_LDFLAGS)
libgfortran_la_LDFLAGS = -version-info `grep -v '^\#' $(srcdir)/libtool-version` \
$(LTLDFLAGS) $(LIBQUADLIB) ../libbacktrace/libbacktrace.la \
$(HWCAP_LDFLAGS) \
- -lm $(extra_ldflags_libgfortran) \
+ -lm $(COARRAY_LIBS) $(extra_ldflags_libgfortran) \
$(version_arg) -Wc,-shared-libgcc
+
libgfortran_la_DEPENDENCIES = $(version_dep) libgfortran.spec $(LIBQUADLIB_DEP)
cafexeclib_LTLIBRARIES = libcaf_single.la
@@ -53,6 +60,37 @@ libcaf_single_la_LDFLAGS = -static
libcaf_single_la_DEPENDENCIES = caf/libcaf.h
libcaf_single_la_LINK = $(LINK) $(libcaf_single_la_LDFLAGS)
+i_nca_minmax_c = \
+ $(srcdir)/generated/nca_minmax_i1.c \
+ $(srcdir)/generated/nca_minmax_i2.c \
+ $(srcdir)/generated/nca_minmax_i4.c \
+ $(srcdir)/generated/nca_minmax_i8.c \
+ $(srcdir)/generated/nca_minmax_i16.c \
+ $(srcdir)/generated/nca_minmax_r4.c \
+ $(srcdir)/generated/nca_minmax_r8.c \
+ $(srcdir)/generated/nca_minmax_r10.c \
+ $(srcdir)/generated/nca_minmax_r16.c
+
+i_nca_minmax_s_c = \
+ $(srcdir)/generated/nca_minmax_s1.c \
+ $(srcdir)/generated/nca_minmax_s4.c
+
+if LIBGFOR_NATIVE_COARRAY
+
+mylib_LTLIBRARIES = libgfor_nca.la
+mylibdir = $(libdir)/gcc/$(target_alias)/$(gcc_version)$(MULTISUBDIR)
+libgfor_nca_la_SOURCES = nca/alloc.c nca/allocator.c nca/coarraynative.c \
+ nca/hashmap.c \
+ nca/sync.c nca/util.c nca/wrapper.c nca/collective_subroutine.c \
+ nca/shared_memory.c \
+ $(i_nca_minmax_c) $(i_nca_minmax_s_c)
+libgfor_nca_la_DEPENDENCIES = nca/alloc.h nca/allocator.h nca/hashmap.h \
+ nca/libcoarraynative.h nca/sync.h nca/shared_memory.h \
+ nca/util.h nca/lock.h nca/collective_subroutine.h\
+ nca/collective_inline.h
+libgfor_nca_la_LINK = $(LINK) $(libgfor_nca_la_LDFLAGS)
+endif
+
if IEEE_SUPPORT
fincludedir = $(libdir)/gcc/$(target_alias)/$(gcc_version)$(MULTISUBDIR)/finclude
nodist_finclude_HEADERS = ieee_arithmetic.mod ieee_exceptions.mod ieee_features.mod
@@ -83,6 +121,10 @@ if LIBGFOR_MINIMAL
AM_CFLAGS += -DLIBGFOR_MINIMAL
endif
+if LIBGFOR_NATIVE_COARRAY
+AM_CFLAGS += $(PTHREAD_CFLAGS)
+endif
+
gfor_io_src= \
io/size_from_kind.c
@@ -1231,9 +1273,20 @@ $(gfor_built_specific2_src): m4/specific2.m4 m4/head.m4
$(gfor_misc_specifics): m4/misc_specifics.m4 m4/head.m4
$(M4) -Dfile=$@ -I$(srcdir)/m4 misc_specifics.m4 > $@
+
+
+if LIBGFOR_NATIVE_COARRAY
+$(i_nca_minmax_c): m4/nca_minmax.m4 $(I_M4_DEPS)
+ $(M4) -Dfile=$@ -I$(srcdir)/m4 nca_minmax.m4 > $@
+
+$(i_nca_minmax_s_c): m4/nca-minmax-s.m4 $(I_M4_DEPS)
+ $(M4) -Dfile=$@ -I$(srcdir)/m4 nca-minmax-s.m4 > $@
+endif
+
## end of maintainer mode only rules
endif
+
EXTRA_DIST = $(m4_files)
# target overrides
diff --git a/libgfortran/Makefile.in b/libgfortran/Makefile.in
index fe063e7ff91..d577a57c6a2 100644
--- a/libgfortran/Makefile.in
+++ b/libgfortran/Makefile.in
@@ -92,7 +92,8 @@ build_triplet = @build@
host_triplet = @host@
target_triplet = @target@
@LIBGFOR_MINIMAL_TRUE@am__append_1 = -DLIBGFOR_MINIMAL
-@LIBGFOR_MINIMAL_FALSE@am__append_2 = \
+@LIBGFOR_NATIVE_COARRAY_TRUE@am__append_2 = $(PTHREAD_CFLAGS)
+@LIBGFOR_MINIMAL_FALSE@am__append_3 = \
@LIBGFOR_MINIMAL_FALSE@io/close.c \
@LIBGFOR_MINIMAL_FALSE@io/file_pos.c \
@LIBGFOR_MINIMAL_FALSE@io/format.c \
@@ -110,7 +111,7 @@ target_triplet = @target@
@LIBGFOR_MINIMAL_FALSE@io/fbuf.c \
@LIBGFOR_MINIMAL_FALSE@io/async.c
-@LIBGFOR_MINIMAL_FALSE@am__append_3 = \
+@LIBGFOR_MINIMAL_FALSE@am__append_4 = \
@LIBGFOR_MINIMAL_FALSE@intrinsics/access.c \
@LIBGFOR_MINIMAL_FALSE@intrinsics/c99_functions.c \
@LIBGFOR_MINIMAL_FALSE@intrinsics/chdir.c \
@@ -143,9 +144,9 @@ target_triplet = @target@
@LIBGFOR_MINIMAL_FALSE@intrinsics/umask.c \
@LIBGFOR_MINIMAL_FALSE@intrinsics/unlink.c
-@IEEE_SUPPORT_TRUE@am__append_4 = ieee/ieee_helper.c
-@LIBGFOR_MINIMAL_TRUE@am__append_5 = runtime/minimal.c
-@LIBGFOR_MINIMAL_FALSE@am__append_6 = \
+@IEEE_SUPPORT_TRUE@am__append_5 = ieee/ieee_helper.c
+@LIBGFOR_MINIMAL_TRUE@am__append_6 = runtime/minimal.c
+@LIBGFOR_MINIMAL_FALSE@am__append_7 = \
@LIBGFOR_MINIMAL_FALSE@runtime/backtrace.c \
@LIBGFOR_MINIMAL_FALSE@runtime/convert_char.c \
@LIBGFOR_MINIMAL_FALSE@runtime/environ.c \
@@ -157,7 +158,7 @@ target_triplet = @target@
# dummy sources for libtool
-@onestep_TRUE@am__append_7 = libgfortran_c.c libgfortran_f.f90
+@onestep_TRUE@am__append_8 = libgfortran_c.c libgfortran_f.f90
subdir = .
ACLOCAL_M4 = $(top_srcdir)/aclocal.m4
am__aclocal_m4_deps = $(top_srcdir)/../config/depstand.m4 \
@@ -214,25 +215,41 @@ am__uninstall_files_from_dir = { \
|| { echo " ( cd '$$dir' && rm -f" $$files ")"; \
$(am__cd) "$$dir" && rm -f $$files; }; \
}
-am__installdirs = "$(DESTDIR)$(cafexeclibdir)" \
+am__installdirs = "$(DESTDIR)$(cafexeclibdir)" "$(DESTDIR)$(mylibdir)" \
"$(DESTDIR)$(toolexeclibdir)" "$(DESTDIR)$(toolexeclibdir)" \
"$(DESTDIR)$(gfor_cdir)" "$(DESTDIR)$(fincludedir)"
-LTLIBRARIES = $(cafexeclib_LTLIBRARIES) $(toolexeclib_LTLIBRARIES)
+LTLIBRARIES = $(cafexeclib_LTLIBRARIES) $(mylib_LTLIBRARIES) \
+ $(toolexeclib_LTLIBRARIES)
libcaf_single_la_LIBADD =
am_libcaf_single_la_OBJECTS = single.lo
libcaf_single_la_OBJECTS = $(am_libcaf_single_la_OBJECTS)
+libgfor_nca_la_LIBADD =
+am__objects_1 = nca_minmax_i1.lo nca_minmax_i2.lo nca_minmax_i4.lo \
+ nca_minmax_i8.lo nca_minmax_i16.lo nca_minmax_r4.lo \
+ nca_minmax_r8.lo nca_minmax_r10.lo nca_minmax_r16.lo
+am__objects_2 = nca_minmax_s1.lo nca_minmax_s4.lo
+@LIBGFOR_NATIVE_COARRAY_TRUE@am_libgfor_nca_la_OBJECTS = alloc.lo \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ allocator.lo coarraynative.lo \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ hashmap.lo sync.lo util.lo \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ wrapper.lo \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ collective_subroutine.lo \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ shared_memory.lo $(am__objects_1) \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ $(am__objects_2)
+libgfor_nca_la_OBJECTS = $(am_libgfor_nca_la_OBJECTS)
+@LIBGFOR_NATIVE_COARRAY_TRUE@am_libgfor_nca_la_rpath = -rpath \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ $(mylibdir)
libgfortran_la_LIBADD =
-@LIBGFOR_MINIMAL_TRUE@am__objects_1 = minimal.lo
-@LIBGFOR_MINIMAL_FALSE@am__objects_2 = backtrace.lo convert_char.lo \
+@LIBGFOR_MINIMAL_TRUE@am__objects_3 = minimal.lo
+@LIBGFOR_MINIMAL_FALSE@am__objects_4 = backtrace.lo convert_char.lo \
@LIBGFOR_MINIMAL_FALSE@ environ.lo error.lo fpu.lo main.lo \
@LIBGFOR_MINIMAL_FALSE@ pause.lo stop.lo
-am__objects_3 = bounds.lo compile_options.lo memory.lo string.lo \
- select.lo $(am__objects_1) $(am__objects_2)
-am__objects_4 = all_l1.lo all_l2.lo all_l4.lo all_l8.lo all_l16.lo
-am__objects_5 = any_l1.lo any_l2.lo any_l4.lo any_l8.lo any_l16.lo
-am__objects_6 = count_1_l.lo count_2_l.lo count_4_l.lo count_8_l.lo \
+am__objects_5 = bounds.lo compile_options.lo memory.lo string.lo \
+ select.lo $(am__objects_3) $(am__objects_4)
+am__objects_6 = all_l1.lo all_l2.lo all_l4.lo all_l8.lo all_l16.lo
+am__objects_7 = any_l1.lo any_l2.lo any_l4.lo any_l8.lo any_l16.lo
+am__objects_8 = count_1_l.lo count_2_l.lo count_4_l.lo count_8_l.lo \
count_16_l.lo
-am__objects_7 = maxloc0_4_i1.lo maxloc0_8_i1.lo maxloc0_16_i1.lo \
+am__objects_9 = maxloc0_4_i1.lo maxloc0_8_i1.lo maxloc0_16_i1.lo \
maxloc0_4_i2.lo maxloc0_8_i2.lo maxloc0_16_i2.lo \
maxloc0_4_i4.lo maxloc0_8_i4.lo maxloc0_16_i4.lo \
maxloc0_4_i8.lo maxloc0_8_i8.lo maxloc0_16_i8.lo \
@@ -241,7 +258,7 @@ am__objects_7 = maxloc0_4_i1.lo maxloc0_8_i1.lo maxloc0_16_i1.lo \
maxloc0_4_r8.lo maxloc0_8_r8.lo maxloc0_16_r8.lo \
maxloc0_4_r10.lo maxloc0_8_r10.lo maxloc0_16_r10.lo \
maxloc0_4_r16.lo maxloc0_8_r16.lo maxloc0_16_r16.lo
-am__objects_8 = maxloc1_4_i1.lo maxloc1_8_i1.lo maxloc1_16_i1.lo \
+am__objects_10 = maxloc1_4_i1.lo maxloc1_8_i1.lo maxloc1_16_i1.lo \
maxloc1_4_i2.lo maxloc1_8_i2.lo maxloc1_16_i2.lo \
maxloc1_4_i4.lo maxloc1_8_i4.lo maxloc1_16_i4.lo \
maxloc1_4_i8.lo maxloc1_8_i8.lo maxloc1_16_i8.lo \
@@ -250,10 +267,10 @@ am__objects_8 = maxloc1_4_i1.lo maxloc1_8_i1.lo maxloc1_16_i1.lo \
maxloc1_4_r8.lo maxloc1_8_r8.lo maxloc1_16_r8.lo \
maxloc1_4_r10.lo maxloc1_8_r10.lo maxloc1_16_r10.lo \
maxloc1_4_r16.lo maxloc1_8_r16.lo maxloc1_16_r16.lo
-am__objects_9 = maxval_i1.lo maxval_i2.lo maxval_i4.lo maxval_i8.lo \
+am__objects_11 = maxval_i1.lo maxval_i2.lo maxval_i4.lo maxval_i8.lo \
maxval_i16.lo maxval_r4.lo maxval_r8.lo maxval_r10.lo \
maxval_r16.lo
-am__objects_10 = minloc0_4_i1.lo minloc0_8_i1.lo minloc0_16_i1.lo \
+am__objects_12 = minloc0_4_i1.lo minloc0_8_i1.lo minloc0_16_i1.lo \
minloc0_4_i2.lo minloc0_8_i2.lo minloc0_16_i2.lo \
minloc0_4_i4.lo minloc0_8_i4.lo minloc0_16_i4.lo \
minloc0_4_i8.lo minloc0_8_i8.lo minloc0_16_i8.lo \
@@ -262,7 +279,7 @@ am__objects_10 = minloc0_4_i1.lo minloc0_8_i1.lo minloc0_16_i1.lo \
minloc0_4_r8.lo minloc0_8_r8.lo minloc0_16_r8.lo \
minloc0_4_r10.lo minloc0_8_r10.lo minloc0_16_r10.lo \
minloc0_4_r16.lo minloc0_8_r16.lo minloc0_16_r16.lo
-am__objects_11 = minloc1_4_i1.lo minloc1_8_i1.lo minloc1_16_i1.lo \
+am__objects_13 = minloc1_4_i1.lo minloc1_8_i1.lo minloc1_16_i1.lo \
minloc1_4_i2.lo minloc1_8_i2.lo minloc1_16_i2.lo \
minloc1_4_i4.lo minloc1_8_i4.lo minloc1_16_i4.lo \
minloc1_4_i8.lo minloc1_8_i8.lo minloc1_16_i8.lo \
@@ -271,49 +288,49 @@ am__objects_11 = minloc1_4_i1.lo minloc1_8_i1.lo minloc1_16_i1.lo \
minloc1_4_r8.lo minloc1_8_r8.lo minloc1_16_r8.lo \
minloc1_4_r10.lo minloc1_8_r10.lo minloc1_16_r10.lo \
minloc1_4_r16.lo minloc1_8_r16.lo minloc1_16_r16.lo
-am__objects_12 = minval_i1.lo minval_i2.lo minval_i4.lo minval_i8.lo \
+am__objects_14 = minval_i1.lo minval_i2.lo minval_i4.lo minval_i8.lo \
minval_i16.lo minval_r4.lo minval_r8.lo minval_r10.lo \
minval_r16.lo
-am__objects_13 = product_i1.lo product_i2.lo product_i4.lo \
+am__objects_15 = product_i1.lo product_i2.lo product_i4.lo \
product_i8.lo product_i16.lo product_r4.lo product_r8.lo \
product_r10.lo product_r16.lo product_c4.lo product_c8.lo \
product_c10.lo product_c16.lo
-am__objects_14 = sum_i1.lo sum_i2.lo sum_i4.lo sum_i8.lo sum_i16.lo \
+am__objects_16 = sum_i1.lo sum_i2.lo sum_i4.lo sum_i8.lo sum_i16.lo \
sum_r4.lo sum_r8.lo sum_r10.lo sum_r16.lo sum_c4.lo sum_c8.lo \
sum_c10.lo sum_c16.lo
-am__objects_15 = bessel_r4.lo bessel_r8.lo bessel_r10.lo bessel_r16.lo
-am__objects_16 = iall_i1.lo iall_i2.lo iall_i4.lo iall_i8.lo \
+am__objects_17 = bessel_r4.lo bessel_r8.lo bessel_r10.lo bessel_r16.lo
+am__objects_18 = iall_i1.lo iall_i2.lo iall_i4.lo iall_i8.lo \
iall_i16.lo
-am__objects_17 = iany_i1.lo iany_i2.lo iany_i4.lo iany_i8.lo \
+am__objects_19 = iany_i1.lo iany_i2.lo iany_i4.lo iany_i8.lo \
iany_i16.lo
-am__objects_18 = iparity_i1.lo iparity_i2.lo iparity_i4.lo \
+am__objects_20 = iparity_i1.lo iparity_i2.lo iparity_i4.lo \
iparity_i8.lo iparity_i16.lo
-am__objects_19 = norm2_r4.lo norm2_r8.lo norm2_r10.lo norm2_r16.lo
-am__objects_20 = parity_l1.lo parity_l2.lo parity_l4.lo parity_l8.lo \
+am__objects_21 = norm2_r4.lo norm2_r8.lo norm2_r10.lo norm2_r16.lo
+am__objects_22 = parity_l1.lo parity_l2.lo parity_l4.lo parity_l8.lo \
parity_l16.lo
-am__objects_21 = matmul_i1.lo matmul_i2.lo matmul_i4.lo matmul_i8.lo \
+am__objects_23 = matmul_i1.lo matmul_i2.lo matmul_i4.lo matmul_i8.lo \
matmul_i16.lo matmul_r4.lo matmul_r8.lo matmul_r10.lo \
matmul_r16.lo matmul_c4.lo matmul_c8.lo matmul_c10.lo \
matmul_c16.lo
-am__objects_22 = matmul_l4.lo matmul_l8.lo matmul_l16.lo
-am__objects_23 = shape_i1.lo shape_i2.lo shape_i4.lo shape_i8.lo \
+am__objects_24 = matmul_l4.lo matmul_l8.lo matmul_l16.lo
+am__objects_25 = shape_i1.lo shape_i2.lo shape_i4.lo shape_i8.lo \
shape_i16.lo
-am__objects_24 = eoshift1_4.lo eoshift1_8.lo eoshift1_16.lo
-am__objects_25 = eoshift3_4.lo eoshift3_8.lo eoshift3_16.lo
-am__objects_26 = cshift1_4.lo cshift1_8.lo cshift1_16.lo
-am__objects_27 = reshape_i4.lo reshape_i8.lo reshape_i16.lo \
+am__objects_26 = eoshift1_4.lo eoshift1_8.lo eoshift1_16.lo
+am__objects_27 = eoshift3_4.lo eoshift3_8.lo eoshift3_16.lo
+am__objects_28 = cshift1_4.lo cshift1_8.lo cshift1_16.lo
+am__objects_29 = reshape_i4.lo reshape_i8.lo reshape_i16.lo \
reshape_r4.lo reshape_r8.lo reshape_r10.lo reshape_r16.lo \
reshape_c4.lo reshape_c8.lo reshape_c10.lo reshape_c16.lo
-am__objects_28 = in_pack_i1.lo in_pack_i2.lo in_pack_i4.lo \
+am__objects_30 = in_pack_i1.lo in_pack_i2.lo in_pack_i4.lo \
in_pack_i8.lo in_pack_i16.lo in_pack_r4.lo in_pack_r8.lo \
in_pack_r10.lo in_pack_r16.lo in_pack_c4.lo in_pack_c8.lo \
in_pack_c10.lo in_pack_c16.lo
-am__objects_29 = in_unpack_i1.lo in_unpack_i2.lo in_unpack_i4.lo \
+am__objects_31 = in_unpack_i1.lo in_unpack_i2.lo in_unpack_i4.lo \
in_unpack_i8.lo in_unpack_i16.lo in_unpack_r4.lo \
in_unpack_r8.lo in_unpack_r10.lo in_unpack_r16.lo \
in_unpack_c4.lo in_unpack_c8.lo in_unpack_c10.lo \
in_unpack_c16.lo
-am__objects_30 = pow_i4_i4.lo pow_i8_i4.lo pow_i16_i4.lo pow_r16_i4.lo \
+am__objects_32 = pow_i4_i4.lo pow_i8_i4.lo pow_i16_i4.lo pow_r16_i4.lo \
pow_c4_i4.lo pow_c8_i4.lo pow_c10_i4.lo pow_c16_i4.lo \
pow_i4_i8.lo pow_i8_i8.lo pow_i16_i8.lo pow_r4_i8.lo \
pow_r8_i8.lo pow_r10_i8.lo pow_r16_i8.lo pow_c4_i8.lo \
@@ -321,27 +338,27 @@ am__objects_30 = pow_i4_i4.lo pow_i8_i4.lo pow_i16_i4.lo pow_r16_i4.lo \
pow_i8_i16.lo pow_i16_i16.lo pow_r4_i16.lo pow_r8_i16.lo \
pow_r10_i16.lo pow_r16_i16.lo pow_c4_i16.lo pow_c8_i16.lo \
pow_c10_i16.lo pow_c16_i16.lo
-am__objects_31 = pack_i1.lo pack_i2.lo pack_i4.lo pack_i8.lo \
+am__objects_33 = pack_i1.lo pack_i2.lo pack_i4.lo pack_i8.lo \
pack_i16.lo pack_r4.lo pack_r8.lo pack_r10.lo pack_r16.lo \
pack_c4.lo pack_c8.lo pack_c10.lo pack_c16.lo
-am__objects_32 = unpack_i1.lo unpack_i2.lo unpack_i4.lo unpack_i8.lo \
+am__objects_34 = unpack_i1.lo unpack_i2.lo unpack_i4.lo unpack_i8.lo \
unpack_i16.lo unpack_r4.lo unpack_r8.lo unpack_r10.lo \
unpack_r16.lo unpack_c4.lo unpack_c8.lo unpack_c10.lo \
unpack_c16.lo
-am__objects_33 = matmulavx128_i1.lo matmulavx128_i2.lo \
+am__objects_35 = matmulavx128_i1.lo matmulavx128_i2.lo \
matmulavx128_i4.lo matmulavx128_i8.lo matmulavx128_i16.lo \
matmulavx128_r4.lo matmulavx128_r8.lo matmulavx128_r10.lo \
matmulavx128_r16.lo matmulavx128_c4.lo matmulavx128_c8.lo \
matmulavx128_c10.lo matmulavx128_c16.lo
-am__objects_34 = spread_i1.lo spread_i2.lo spread_i4.lo spread_i8.lo \
+am__objects_36 = spread_i1.lo spread_i2.lo spread_i4.lo spread_i8.lo \
spread_i16.lo spread_r4.lo spread_r8.lo spread_r10.lo \
spread_r16.lo spread_c4.lo spread_c8.lo spread_c10.lo \
spread_c16.lo
-am__objects_35 = cshift0_i1.lo cshift0_i2.lo cshift0_i4.lo \
+am__objects_37 = cshift0_i1.lo cshift0_i2.lo cshift0_i4.lo \
cshift0_i8.lo cshift0_i16.lo cshift0_r4.lo cshift0_r8.lo \
cshift0_r10.lo cshift0_r16.lo cshift0_c4.lo cshift0_c8.lo \
cshift0_c10.lo cshift0_c16.lo
-am__objects_36 = cshift1_4_i1.lo cshift1_4_i2.lo cshift1_4_i4.lo \
+am__objects_38 = cshift1_4_i1.lo cshift1_4_i2.lo cshift1_4_i4.lo \
cshift1_4_i8.lo cshift1_4_i16.lo cshift1_4_r4.lo \
cshift1_4_r8.lo cshift1_4_r10.lo cshift1_4_r16.lo \
cshift1_4_c4.lo cshift1_4_c8.lo cshift1_4_c10.lo \
@@ -354,58 +371,58 @@ am__objects_36 = cshift1_4_i1.lo cshift1_4_i2.lo cshift1_4_i4.lo \
cshift1_16_i16.lo cshift1_16_r4.lo cshift1_16_r8.lo \
cshift1_16_r10.lo cshift1_16_r16.lo cshift1_16_c4.lo \
cshift1_16_c8.lo cshift1_16_c10.lo cshift1_16_c16.lo
-am__objects_37 = maxloc0_4_s1.lo maxloc0_4_s4.lo maxloc0_8_s1.lo \
+am__objects_39 = maxloc0_4_s1.lo maxloc0_4_s4.lo maxloc0_8_s1.lo \
maxloc0_8_s4.lo maxloc0_16_s1.lo maxloc0_16_s4.lo
-am__objects_38 = minloc0_4_s1.lo minloc0_4_s4.lo minloc0_8_s1.lo \
+am__objects_40 = minloc0_4_s1.lo minloc0_4_s4.lo minloc0_8_s1.lo \
minloc0_8_s4.lo minloc0_16_s1.lo minloc0_16_s4.lo
-am__objects_39 = maxloc1_4_s1.lo maxloc1_4_s4.lo maxloc1_8_s1.lo \
+am__objects_41 = maxloc1_4_s1.lo maxloc1_4_s4.lo maxloc1_8_s1.lo \
maxloc1_8_s4.lo maxloc1_16_s1.lo maxloc1_16_s4.lo
-am__objects_40 = minloc1_4_s1.lo minloc1_4_s4.lo minloc1_8_s1.lo \
+am__objects_42 = minloc1_4_s1.lo minloc1_4_s4.lo minloc1_8_s1.lo \
minloc1_8_s4.lo minloc1_16_s1.lo minloc1_16_s4.lo
-am__objects_41 = maxloc2_4_s1.lo maxloc2_4_s4.lo maxloc2_8_s1.lo \
+am__objects_43 = maxloc2_4_s1.lo maxloc2_4_s4.lo maxloc2_8_s1.lo \
maxloc2_8_s4.lo maxloc2_16_s1.lo maxloc2_16_s4.lo
-am__objects_42 = minloc2_4_s1.lo minloc2_4_s4.lo minloc2_8_s1.lo \
+am__objects_44 = minloc2_4_s1.lo minloc2_4_s4.lo minloc2_8_s1.lo \
minloc2_8_s4.lo minloc2_16_s1.lo minloc2_16_s4.lo
-am__objects_43 = maxval0_s1.lo maxval0_s4.lo
-am__objects_44 = minval0_s1.lo minval0_s4.lo
-am__objects_45 = maxval1_s1.lo maxval1_s4.lo
-am__objects_46 = minval1_s1.lo minval1_s4.lo
-am__objects_47 = findloc0_i1.lo findloc0_i2.lo findloc0_i4.lo \
+am__objects_45 = maxval0_s1.lo maxval0_s4.lo
+am__objects_46 = minval0_s1.lo minval0_s4.lo
+am__objects_47 = maxval1_s1.lo maxval1_s4.lo
+am__objects_48 = minval1_s1.lo minval1_s4.lo
+am__objects_49 = findloc0_i1.lo findloc0_i2.lo findloc0_i4.lo \
findloc0_i8.lo findloc0_i16.lo findloc0_r4.lo findloc0_r8.lo \
findloc0_r10.lo findloc0_r16.lo findloc0_c4.lo findloc0_c8.lo \
findloc0_c10.lo findloc0_c16.lo
-am__objects_48 = findloc0_s1.lo findloc0_s4.lo
-am__objects_49 = findloc1_i1.lo findloc1_i2.lo findloc1_i4.lo \
+am__objects_50 = findloc0_s1.lo findloc0_s4.lo
+am__objects_51 = findloc1_i1.lo findloc1_i2.lo findloc1_i4.lo \
findloc1_i8.lo findloc1_i16.lo findloc1_r4.lo findloc1_r8.lo \
findloc1_r10.lo findloc1_r16.lo findloc1_c4.lo findloc1_c8.lo \
findloc1_c10.lo findloc1_c16.lo
-am__objects_50 = findloc1_s1.lo findloc1_s4.lo
-am__objects_51 = findloc2_s1.lo findloc2_s4.lo
-am__objects_52 = ISO_Fortran_binding.lo
-am__objects_53 = $(am__objects_4) $(am__objects_5) $(am__objects_6) \
- $(am__objects_7) $(am__objects_8) $(am__objects_9) \
- $(am__objects_10) $(am__objects_11) $(am__objects_12) \
- $(am__objects_13) $(am__objects_14) $(am__objects_15) \
- $(am__objects_16) $(am__objects_17) $(am__objects_18) \
- $(am__objects_19) $(am__objects_20) $(am__objects_21) \
- $(am__objects_22) $(am__objects_23) $(am__objects_24) \
- $(am__objects_25) $(am__objects_26) $(am__objects_27) \
- $(am__objects_28) $(am__objects_29) $(am__objects_30) \
- $(am__objects_31) $(am__objects_32) $(am__objects_33) \
- $(am__objects_34) $(am__objects_35) $(am__objects_36) \
- $(am__objects_37) $(am__objects_38) $(am__objects_39) \
- $(am__objects_40) $(am__objects_41) $(am__objects_42) \
- $(am__objects_43) $(am__objects_44) $(am__objects_45) \
- $(am__objects_46) $(am__objects_47) $(am__objects_48) \
- $(am__objects_49) $(am__objects_50) $(am__objects_51) \
- $(am__objects_52)
-@LIBGFOR_MINIMAL_FALSE@am__objects_54 = close.lo file_pos.lo format.lo \
+am__objects_52 = findloc1_s1.lo findloc1_s4.lo
+am__objects_53 = findloc2_s1.lo findloc2_s4.lo
+am__objects_54 = ISO_Fortran_binding.lo
+am__objects_55 = $(am__objects_6) $(am__objects_7) $(am__objects_8) \
+ $(am__objects_9) $(am__objects_10) $(am__objects_11) \
+ $(am__objects_12) $(am__objects_13) $(am__objects_14) \
+ $(am__objects_15) $(am__objects_16) $(am__objects_17) \
+ $(am__objects_18) $(am__objects_19) $(am__objects_20) \
+ $(am__objects_21) $(am__objects_22) $(am__objects_23) \
+ $(am__objects_24) $(am__objects_25) $(am__objects_26) \
+ $(am__objects_27) $(am__objects_28) $(am__objects_29) \
+ $(am__objects_30) $(am__objects_31) $(am__objects_32) \
+ $(am__objects_33) $(am__objects_34) $(am__objects_35) \
+ $(am__objects_36) $(am__objects_37) $(am__objects_38) \
+ $(am__objects_39) $(am__objects_40) $(am__objects_41) \
+ $(am__objects_42) $(am__objects_43) $(am__objects_44) \
+ $(am__objects_45) $(am__objects_46) $(am__objects_47) \
+ $(am__objects_48) $(am__objects_49) $(am__objects_50) \
+ $(am__objects_51) $(am__objects_52) $(am__objects_53) \
+ $(am__objects_54)
+@LIBGFOR_MINIMAL_FALSE@am__objects_56 = close.lo file_pos.lo format.lo \
@LIBGFOR_MINIMAL_FALSE@ inquire.lo intrinsics.lo list_read.lo \
@LIBGFOR_MINIMAL_FALSE@ lock.lo open.lo read.lo transfer.lo \
@LIBGFOR_MINIMAL_FALSE@ transfer128.lo unit.lo unix.lo write.lo \
@LIBGFOR_MINIMAL_FALSE@ fbuf.lo async.lo
-am__objects_55 = size_from_kind.lo $(am__objects_54)
-@LIBGFOR_MINIMAL_FALSE@am__objects_56 = access.lo c99_functions.lo \
+am__objects_57 = size_from_kind.lo $(am__objects_56)
+@LIBGFOR_MINIMAL_FALSE@am__objects_58 = access.lo c99_functions.lo \
@LIBGFOR_MINIMAL_FALSE@ chdir.lo chmod.lo clock.lo cpu_time.lo \
@LIBGFOR_MINIMAL_FALSE@ ctime.lo date_and_time.lo dtime.lo \
@LIBGFOR_MINIMAL_FALSE@ env.lo etime.lo execute_command_line.lo \
@@ -415,20 +432,20 @@ am__objects_55 = size_from_kind.lo $(am__objects_54)
@LIBGFOR_MINIMAL_FALSE@ rename.lo stat.lo symlnk.lo \
@LIBGFOR_MINIMAL_FALSE@ system_clock.lo time.lo umask.lo \
@LIBGFOR_MINIMAL_FALSE@ unlink.lo
-@IEEE_SUPPORT_TRUE@am__objects_57 = ieee_helper.lo
-am__objects_58 = associated.lo abort.lo args.lo cshift0.lo eoshift0.lo \
+@IEEE_SUPPORT_TRUE@am__objects_59 = ieee_helper.lo
+am__objects_60 = associated.lo abort.lo args.lo cshift0.lo eoshift0.lo \
eoshift2.lo erfc_scaled.lo extends_type_of.lo fnum.lo \
ierrno.lo ishftc.lo is_contiguous.lo mvbits.lo move_alloc.lo \
pack_generic.lo selected_char_kind.lo size.lo \
spread_generic.lo string_intrinsics.lo rand.lo random.lo \
reshape_generic.lo reshape_packed.lo selected_int_kind.lo \
selected_real_kind.lo trigd.lo unpack_generic.lo \
- in_pack_generic.lo in_unpack_generic.lo $(am__objects_56) \
- $(am__objects_57)
-@IEEE_SUPPORT_TRUE@am__objects_59 = ieee_arithmetic.lo \
+ in_pack_generic.lo in_unpack_generic.lo $(am__objects_58) \
+ $(am__objects_59)
+@IEEE_SUPPORT_TRUE@am__objects_61 = ieee_arithmetic.lo \
@IEEE_SUPPORT_TRUE@ ieee_exceptions.lo ieee_features.lo
-am__objects_60 =
-am__objects_61 = _abs_c4.lo _abs_c8.lo _abs_c10.lo _abs_c16.lo \
+am__objects_62 =
+am__objects_63 = _abs_c4.lo _abs_c8.lo _abs_c10.lo _abs_c16.lo \
_abs_i4.lo _abs_i8.lo _abs_i16.lo _abs_r4.lo _abs_r8.lo \
_abs_r10.lo _abs_r16.lo _aimag_c4.lo _aimag_c8.lo \
_aimag_c10.lo _aimag_c16.lo _exp_r4.lo _exp_r8.lo _exp_r10.lo \
@@ -452,19 +469,19 @@ am__objects_61 = _abs_c4.lo _abs_c8.lo _abs_c10.lo _abs_c16.lo \
_conjg_c4.lo _conjg_c8.lo _conjg_c10.lo _conjg_c16.lo \
_aint_r4.lo _aint_r8.lo _aint_r10.lo _aint_r16.lo _anint_r4.lo \
_anint_r8.lo _anint_r10.lo _anint_r16.lo
-am__objects_62 = _sign_i4.lo _sign_i8.lo _sign_i16.lo _sign_r4.lo \
+am__objects_64 = _sign_i4.lo _sign_i8.lo _sign_i16.lo _sign_r4.lo \
_sign_r8.lo _sign_r10.lo _sign_r16.lo _dim_i4.lo _dim_i8.lo \
_dim_i16.lo _dim_r4.lo _dim_r8.lo _dim_r10.lo _dim_r16.lo \
_atan2_r4.lo _atan2_r8.lo _atan2_r10.lo _atan2_r16.lo \
_mod_i4.lo _mod_i8.lo _mod_i16.lo _mod_r4.lo _mod_r8.lo \
_mod_r10.lo _mod_r16.lo
-am__objects_63 = misc_specifics.lo
-am__objects_64 = $(am__objects_61) $(am__objects_62) $(am__objects_63) \
+am__objects_65 = misc_specifics.lo
+am__objects_66 = $(am__objects_63) $(am__objects_64) $(am__objects_65) \
dprod_r8.lo f2c_specifics.lo random_init.lo
-am__objects_65 = $(am__objects_3) $(am__objects_53) $(am__objects_55) \
- $(am__objects_58) $(am__objects_59) $(am__objects_60) \
- $(am__objects_64)
-@onestep_FALSE@am_libgfortran_la_OBJECTS = $(am__objects_65)
+am__objects_67 = $(am__objects_5) $(am__objects_55) $(am__objects_57) \
+ $(am__objects_60) $(am__objects_61) $(am__objects_62) \
+ $(am__objects_66)
+@onestep_FALSE@am_libgfortran_la_OBJECTS = $(am__objects_67)
@onestep_TRUE@am_libgfortran_la_OBJECTS = libgfortran_c.lo
libgfortran_la_OBJECTS = $(am_libgfortran_la_OBJECTS)
AM_V_P = $(am__v_P_@AM_V@)
@@ -530,7 +547,8 @@ AM_V_FC = $(am__v_FC_@AM_V@)
am__v_FC_ = $(am__v_FC_@AM_DEFAULT_V@)
am__v_FC_0 = @echo " FC " $@;
am__v_FC_1 =
-SOURCES = $(libcaf_single_la_SOURCES) $(libgfortran_la_SOURCES)
+SOURCES = $(libcaf_single_la_SOURCES) $(libgfor_nca_la_SOURCES) \
+ $(libgfortran_la_SOURCES)
am__can_run_installinfo = \
case $$AM_UPDATE_INFO_DIR in \
n|no|NO) false;; \
@@ -569,7 +587,7 @@ AMTAR = @AMTAR@
# Some targets require additional compiler options for IEEE compatibility.
AM_CFLAGS = @AM_CFLAGS@ -fcx-fortran-rules $(SECTION_FLAGS) \
- $(IEEE_FLAGS) $(am__append_1)
+ $(IEEE_FLAGS) $(am__append_1) $(am__append_2)
AM_DEFAULT_VERBOSITY = @AM_DEFAULT_VERBOSITY@
AM_FCFLAGS = @AM_FCFLAGS@ $(IEEE_FLAGS)
AR = @AR@
@@ -635,7 +653,10 @@ PACKAGE_TARNAME = @PACKAGE_TARNAME@
PACKAGE_URL = @PACKAGE_URL@
PACKAGE_VERSION = @PACKAGE_VERSION@
PATH_SEPARATOR = @PATH_SEPARATOR@
+PTHREAD_CFLAGS = @PTHREAD_CFLAGS@
+PTHREAD_LIBS = @PTHREAD_LIBS@
RANLIB = @RANLIB@
+RT_LIBS = @RT_LIBS@
SECTION_FLAGS = @SECTION_FLAGS@
SED = @SED@
SET_MAKE = @SET_MAKE@
@@ -696,6 +717,7 @@ pdfdir = @pdfdir@
prefix = @prefix@
program_transform_name = @program_transform_name@
psdir = @psdir@
+runstatedir = @runstatedir@
sbindir = @sbindir@
sharedstatedir = @sharedstatedir@
srcdir = @srcdir@
@@ -726,13 +748,15 @@ gfor_cdir = $(libdir)/gcc/$(target_alias)/$(gcc_version)/include
LTLDFLAGS = $(shell $(SHELL) $(top_srcdir)/../libtool-ldflags $(LDFLAGS)) \
$(lt_host_flags)
+@LIBGFOR_NATIVE_COARRAY_FALSE@COARRAY_LIBS = ""
+@LIBGFOR_NATIVE_COARRAY_TRUE@COARRAY_LIBS = $(PTHREAD_LIBS) $(RT_LIBS)
toolexeclib_LTLIBRARIES = libgfortran.la
toolexeclib_DATA = libgfortran.spec
libgfortran_la_LINK = $(LINK) $(libgfortran_la_LDFLAGS)
libgfortran_la_LDFLAGS = -version-info `grep -v '^\#' $(srcdir)/libtool-version` \
$(LTLDFLAGS) $(LIBQUADLIB) ../libbacktrace/libbacktrace.la \
$(HWCAP_LDFLAGS) \
- -lm $(extra_ldflags_libgfortran) \
+ -lm $(COARRAY_LIBS) $(extra_ldflags_libgfortran) \
$(version_arg) -Wc,-shared-libgcc
libgfortran_la_DEPENDENCIES = $(version_dep) libgfortran.spec $(LIBQUADLIB_DEP)
@@ -742,6 +766,35 @@ libcaf_single_la_SOURCES = caf/single.c
libcaf_single_la_LDFLAGS = -static
libcaf_single_la_DEPENDENCIES = caf/libcaf.h
libcaf_single_la_LINK = $(LINK) $(libcaf_single_la_LDFLAGS)
+i_nca_minmax_c = \
+ $(srcdir)/generated/nca_minmax_i1.c \
+ $(srcdir)/generated/nca_minmax_i2.c \
+ $(srcdir)/generated/nca_minmax_i4.c \
+ $(srcdir)/generated/nca_minmax_i8.c \
+ $(srcdir)/generated/nca_minmax_i16.c \
+ $(srcdir)/generated/nca_minmax_r4.c \
+ $(srcdir)/generated/nca_minmax_r8.c \
+ $(srcdir)/generated/nca_minmax_r10.c \
+ $(srcdir)/generated/nca_minmax_r16.c
+
+i_nca_minmax_s_c = \
+ $(srcdir)/generated/nca_minmax_s1.c \
+ $(srcdir)/generated/nca_minmax_s4.c
+
+@LIBGFOR_NATIVE_COARRAY_TRUE@mylib_LTLIBRARIES = libgfor_nca.la
+@LIBGFOR_NATIVE_COARRAY_TRUE@mylibdir = $(libdir)/gcc/$(target_alias)/$(gcc_version)$(MULTISUBDIR)
+@LIBGFOR_NATIVE_COARRAY_TRUE@libgfor_nca_la_SOURCES = nca/alloc.c nca/allocator.c nca/coarraynative.c \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ nca/hashmap.c \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ nca/sync.c nca/util.c nca/wrapper.c nca/collective_subroutine.c \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ nca/shared_memory.c \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ $(i_nca_minmax_c) $(i_nca_minmax_s_c)
+
+@LIBGFOR_NATIVE_COARRAY_TRUE@libgfor_nca_la_DEPENDENCIES = nca/alloc.h nca/allocator.h nca/hashmap.h \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ nca/libcoarraynative.h nca/sync.h nca/shared_memory.h \
+@LIBGFOR_NATIVE_COARRAY_TRUE@ nca/util.h nca/lock.h nca/collective_subroutine.h\
+@LIBGFOR_NATIVE_COARRAY_TRUE@ nca/collective_inline.h
+
+@LIBGFOR_NATIVE_COARRAY_TRUE@libgfor_nca_la_LINK = $(LINK) $(libgfor_nca_la_LDFLAGS)
@IEEE_SUPPORT_TRUE@fincludedir = $(libdir)/gcc/$(target_alias)/$(gcc_version)$(MULTISUBDIR)/finclude
@IEEE_SUPPORT_TRUE@nodist_finclude_HEADERS = ieee_arithmetic.mod ieee_exceptions.mod ieee_features.mod
AM_CPPFLAGS = -iquote$(srcdir)/io -I$(srcdir)/$(MULTISRCTOP)../gcc \
@@ -753,7 +806,7 @@ AM_CPPFLAGS = -iquote$(srcdir)/io -I$(srcdir)/$(MULTISRCTOP)../gcc \
-I$(MULTIBUILDTOP)../libbacktrace \
-I../libbacktrace
-gfor_io_src = io/size_from_kind.c $(am__append_2)
+gfor_io_src = io/size_from_kind.c $(am__append_3)
gfor_io_headers = \
io/io.h \
io/fbuf.h \
@@ -775,7 +828,7 @@ gfor_helper_src = intrinsics/associated.c intrinsics/abort.c \
intrinsics/selected_int_kind.f90 \
intrinsics/selected_real_kind.f90 intrinsics/trigd.c \
intrinsics/unpack_generic.c runtime/in_pack_generic.c \
- runtime/in_unpack_generic.c $(am__append_3) $(am__append_4)
+ runtime/in_unpack_generic.c $(am__append_4) $(am__append_5)
@IEEE_SUPPORT_FALSE@gfor_ieee_src =
@IEEE_SUPPORT_TRUE@gfor_ieee_src = \
@IEEE_SUPPORT_TRUE@ieee/ieee_arithmetic.F90 \
@@ -783,8 +836,8 @@ gfor_helper_src = intrinsics/associated.c intrinsics/abort.c \
@IEEE_SUPPORT_TRUE@ieee/ieee_features.F90
gfor_src = runtime/bounds.c runtime/compile_options.c runtime/memory.c \
- runtime/string.c runtime/select.c $(am__append_5) \
- $(am__append_6)
+ runtime/string.c runtime/select.c $(am__append_6) \
+ $(am__append_7)
i_all_c = \
$(srcdir)/generated/all_l1.c \
$(srcdir)/generated/all_l2.c \
@@ -1538,7 +1591,7 @@ intrinsics/random_init.f90
BUILT_SOURCES = $(gfor_built_src) $(gfor_built_specific_src) \
$(gfor_built_specific2_src) $(gfor_misc_specifics) \
- $(am__append_7)
+ $(am__append_8)
prereq_SRC = $(gfor_src) $(gfor_built_src) $(gfor_io_src) \
$(gfor_helper_src) $(gfor_ieee_src) $(gfor_io_headers) $(gfor_specific_src)
@@ -1666,6 +1719,41 @@ clean-cafexeclibLTLIBRARIES:
rm -f $${locs}; \
}
+install-mylibLTLIBRARIES: $(mylib_LTLIBRARIES)
+ @$(NORMAL_INSTALL)
+ @list='$(mylib_LTLIBRARIES)'; test -n "$(mylibdir)" || list=; \
+ list2=; for p in $$list; do \
+ if test -f $$p; then \
+ list2="$$list2 $$p"; \
+ else :; fi; \
+ done; \
+ test -z "$$list2" || { \
+ echo " $(MKDIR_P) '$(DESTDIR)$(mylibdir)'"; \
+ $(MKDIR_P) "$(DESTDIR)$(mylibdir)" || exit 1; \
+ echo " $(LIBTOOL) $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=install $(INSTALL) $(INSTALL_STRIP_FLAG) $$list2 '$(DESTDIR)$(mylibdir)'"; \
+ $(LIBTOOL) $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=install $(INSTALL) $(INSTALL_STRIP_FLAG) $$list2 "$(DESTDIR)$(mylibdir)"; \
+ }
+
+uninstall-mylibLTLIBRARIES:
+ @$(NORMAL_UNINSTALL)
+ @list='$(mylib_LTLIBRARIES)'; test -n "$(mylibdir)" || list=; \
+ for p in $$list; do \
+ $(am__strip_dir) \
+ echo " $(LIBTOOL) $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=uninstall rm -f '$(DESTDIR)$(mylibdir)/$$f'"; \
+ $(LIBTOOL) $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=uninstall rm -f "$(DESTDIR)$(mylibdir)/$$f"; \
+ done
+
+clean-mylibLTLIBRARIES:
+ -test -z "$(mylib_LTLIBRARIES)" || rm -f $(mylib_LTLIBRARIES)
+ @list='$(mylib_LTLIBRARIES)'; \
+ locs=`for p in $$list; do echo $$p; done | \
+ sed 's|^[^/]*$$|.|; s|/[^/]*$$||; s|$$|/so_locations|' | \
+ sort -u`; \
+ test -z "$$locs" || { \
+ echo rm -f $${locs}; \
+ rm -f $${locs}; \
+ }
+
install-toolexeclibLTLIBRARIES: $(toolexeclib_LTLIBRARIES)
@$(NORMAL_INSTALL)
@list='$(toolexeclib_LTLIBRARIES)'; test -n "$(toolexeclibdir)" || list=; \
@@ -1704,6 +1792,9 @@ clean-toolexeclibLTLIBRARIES:
libcaf_single.la: $(libcaf_single_la_OBJECTS) $(libcaf_single_la_DEPENDENCIES) $(EXTRA_libcaf_single_la_DEPENDENCIES)
$(AM_V_GEN)$(libcaf_single_la_LINK) -rpath $(cafexeclibdir) $(libcaf_single_la_OBJECTS) $(libcaf_single_la_LIBADD) $(LIBS)
+libgfor_nca.la: $(libgfor_nca_la_OBJECTS) $(libgfor_nca_la_DEPENDENCIES) $(EXTRA_libgfor_nca_la_DEPENDENCIES)
+ $(AM_V_GEN)$(libgfor_nca_la_LINK) $(am_libgfor_nca_la_rpath) $(libgfor_nca_la_OBJECTS) $(libgfor_nca_la_LIBADD) $(LIBS)
+
libgfortran.la: $(libgfortran_la_OBJECTS) $(libgfortran_la_DEPENDENCIES) $(EXTRA_libgfortran_la_DEPENDENCIES)
$(AM_V_GEN)$(libgfortran_la_LINK) -rpath $(toolexeclibdir) $(libgfortran_la_OBJECTS) $(libgfortran_la_LIBADD) $(LIBS)
@@ -1721,6 +1812,8 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/all_l2.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/all_l4.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/all_l8.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/alloc.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/allocator.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/any_l1.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/any_l16.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/any_l2.Plo@am__quote@
@@ -1740,6 +1833,8 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/chmod.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/clock.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/close.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/coarraynative.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/collective_subroutine.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/compile_options.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/convert_char.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/count_16_l.Plo@am__quote@
@@ -1864,6 +1959,7 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/getXid.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/getcwd.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/getlog.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/hashmap.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/hostnm.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/iall_i1.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/iall_i16.Plo@am__quote@
@@ -2123,6 +2219,17 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/minval_r8.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/move_alloc.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/mvbits.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_i1.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_i16.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_i2.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_i4.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_i8.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_r10.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_r16.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_r4.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_r8.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_s1.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/nca_minmax_s4.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/norm2_r10.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/norm2_r16.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/norm2_r4.Plo@am__quote@
@@ -2216,6 +2323,7 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/shape_i2.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/shape_i4.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/shape_i8.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/shared_memory.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/signal.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/single.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/size.Plo@am__quote@
@@ -2253,6 +2361,7 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sum_r4.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sum_r8.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/symlnk.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sync.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/system.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/system_clock.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/time.Plo@am__quote@
@@ -2277,6 +2386,8 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/unpack_r16.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/unpack_r4.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/unpack_r8.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/util.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/wrapper.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/write.Plo@am__quote@
.F90.o:
@@ -2736,6 +2847,146 @@ single.lo: caf/single.c
@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o single.lo `test -f 'caf/single.c' || echo '$(srcdir)/'`caf/single.c
+alloc.lo: nca/alloc.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT alloc.lo -MD -MP -MF $(DEPDIR)/alloc.Tpo -c -o alloc.lo `test -f 'nca/alloc.c' || echo '$(srcdir)/'`nca/alloc.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/alloc.Tpo $(DEPDIR)/alloc.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='nca/alloc.c' object='alloc.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o alloc.lo `test -f 'nca/alloc.c' || echo '$(srcdir)/'`nca/alloc.c
+
+allocator.lo: nca/allocator.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT allocator.lo -MD -MP -MF $(DEPDIR)/allocator.Tpo -c -o allocator.lo `test -f 'nca/allocator.c' || echo '$(srcdir)/'`nca/allocator.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/allocator.Tpo $(DEPDIR)/allocator.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='nca/allocator.c' object='allocator.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o allocator.lo `test -f 'nca/allocator.c' || echo '$(srcdir)/'`nca/allocator.c
+
+coarraynative.lo: nca/coarraynative.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT coarraynative.lo -MD -MP -MF $(DEPDIR)/coarraynative.Tpo -c -o coarraynative.lo `test -f 'nca/coarraynative.c' || echo '$(srcdir)/'`nca/coarraynative.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/coarraynative.Tpo $(DEPDIR)/coarraynative.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='nca/coarraynative.c' object='coarraynative.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o coarraynative.lo `test -f 'nca/coarraynative.c' || echo '$(srcdir)/'`nca/coarraynative.c
+
+hashmap.lo: nca/hashmap.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT hashmap.lo -MD -MP -MF $(DEPDIR)/hashmap.Tpo -c -o hashmap.lo `test -f 'nca/hashmap.c' || echo '$(srcdir)/'`nca/hashmap.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/hashmap.Tpo $(DEPDIR)/hashmap.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='nca/hashmap.c' object='hashmap.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o hashmap.lo `test -f 'nca/hashmap.c' || echo '$(srcdir)/'`nca/hashmap.c
+
+sync.lo: nca/sync.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT sync.lo -MD -MP -MF $(DEPDIR)/sync.Tpo -c -o sync.lo `test -f 'nca/sync.c' || echo '$(srcdir)/'`nca/sync.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/sync.Tpo $(DEPDIR)/sync.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='nca/sync.c' object='sync.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o sync.lo `test -f 'nca/sync.c' || echo '$(srcdir)/'`nca/sync.c
+
+util.lo: nca/util.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT util.lo -MD -MP -MF $(DEPDIR)/util.Tpo -c -o util.lo `test -f 'nca/util.c' || echo '$(srcdir)/'`nca/util.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/util.Tpo $(DEPDIR)/util.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='nca/util.c' object='util.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o util.lo `test -f 'nca/util.c' || echo '$(srcdir)/'`nca/util.c
+
+wrapper.lo: nca/wrapper.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT wrapper.lo -MD -MP -MF $(DEPDIR)/wrapper.Tpo -c -o wrapper.lo `test -f 'nca/wrapper.c' || echo '$(srcdir)/'`nca/wrapper.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/wrapper.Tpo $(DEPDIR)/wrapper.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='nca/wrapper.c' object='wrapper.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o wrapper.lo `test -f 'nca/wrapper.c' || echo '$(srcdir)/'`nca/wrapper.c
+
+collective_subroutine.lo: nca/collective_subroutine.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT collective_subroutine.lo -MD -MP -MF $(DEPDIR)/collective_subroutine.Tpo -c -o collective_subroutine.lo `test -f 'nca/collective_subroutine.c' || echo '$(srcdir)/'`nca/collective_subroutine.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/collective_subroutine.Tpo $(DEPDIR)/collective_subroutine.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='nca/collective_subroutine.c' object='collective_subroutine.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o collective_subroutine.lo `test -f 'nca/collective_subroutine.c' || echo '$(srcdir)/'`nca/collective_subroutine.c
+
+shared_memory.lo: nca/shared_memory.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT shared_memory.lo -MD -MP -MF $(DEPDIR)/shared_memory.Tpo -c -o shared_memory.lo `test -f 'nca/shared_memory.c' || echo '$(srcdir)/'`nca/shared_memory.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/shared_memory.Tpo $(DEPDIR)/shared_memory.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='nca/shared_memory.c' object='shared_memory.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o shared_memory.lo `test -f 'nca/shared_memory.c' || echo '$(srcdir)/'`nca/shared_memory.c
+
+nca_minmax_i1.lo: $(srcdir)/generated/nca_minmax_i1.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_i1.lo -MD -MP -MF $(DEPDIR)/nca_minmax_i1.Tpo -c -o nca_minmax_i1.lo `test -f '$(srcdir)/generated/nca_minmax_i1.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_i1.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_i1.Tpo $(DEPDIR)/nca_minmax_i1.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_i1.c' object='nca_minmax_i1.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_i1.lo `test -f '$(srcdir)/generated/nca_minmax_i1.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_i1.c
+
+nca_minmax_i2.lo: $(srcdir)/generated/nca_minmax_i2.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_i2.lo -MD -MP -MF $(DEPDIR)/nca_minmax_i2.Tpo -c -o nca_minmax_i2.lo `test -f '$(srcdir)/generated/nca_minmax_i2.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_i2.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_i2.Tpo $(DEPDIR)/nca_minmax_i2.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_i2.c' object='nca_minmax_i2.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_i2.lo `test -f '$(srcdir)/generated/nca_minmax_i2.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_i2.c
+
+nca_minmax_i4.lo: $(srcdir)/generated/nca_minmax_i4.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_i4.lo -MD -MP -MF $(DEPDIR)/nca_minmax_i4.Tpo -c -o nca_minmax_i4.lo `test -f '$(srcdir)/generated/nca_minmax_i4.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_i4.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_i4.Tpo $(DEPDIR)/nca_minmax_i4.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_i4.c' object='nca_minmax_i4.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_i4.lo `test -f '$(srcdir)/generated/nca_minmax_i4.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_i4.c
+
+nca_minmax_i8.lo: $(srcdir)/generated/nca_minmax_i8.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_i8.lo -MD -MP -MF $(DEPDIR)/nca_minmax_i8.Tpo -c -o nca_minmax_i8.lo `test -f '$(srcdir)/generated/nca_minmax_i8.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_i8.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_i8.Tpo $(DEPDIR)/nca_minmax_i8.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_i8.c' object='nca_minmax_i8.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_i8.lo `test -f '$(srcdir)/generated/nca_minmax_i8.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_i8.c
+
+nca_minmax_i16.lo: $(srcdir)/generated/nca_minmax_i16.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_i16.lo -MD -MP -MF $(DEPDIR)/nca_minmax_i16.Tpo -c -o nca_minmax_i16.lo `test -f '$(srcdir)/generated/nca_minmax_i16.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_i16.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_i16.Tpo $(DEPDIR)/nca_minmax_i16.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_i16.c' object='nca_minmax_i16.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_i16.lo `test -f '$(srcdir)/generated/nca_minmax_i16.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_i16.c
+
+nca_minmax_r4.lo: $(srcdir)/generated/nca_minmax_r4.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_r4.lo -MD -MP -MF $(DEPDIR)/nca_minmax_r4.Tpo -c -o nca_minmax_r4.lo `test -f '$(srcdir)/generated/nca_minmax_r4.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_r4.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_r4.Tpo $(DEPDIR)/nca_minmax_r4.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_r4.c' object='nca_minmax_r4.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_r4.lo `test -f '$(srcdir)/generated/nca_minmax_r4.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_r4.c
+
+nca_minmax_r8.lo: $(srcdir)/generated/nca_minmax_r8.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_r8.lo -MD -MP -MF $(DEPDIR)/nca_minmax_r8.Tpo -c -o nca_minmax_r8.lo `test -f '$(srcdir)/generated/nca_minmax_r8.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_r8.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_r8.Tpo $(DEPDIR)/nca_minmax_r8.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_r8.c' object='nca_minmax_r8.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_r8.lo `test -f '$(srcdir)/generated/nca_minmax_r8.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_r8.c
+
+nca_minmax_r10.lo: $(srcdir)/generated/nca_minmax_r10.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_r10.lo -MD -MP -MF $(DEPDIR)/nca_minmax_r10.Tpo -c -o nca_minmax_r10.lo `test -f '$(srcdir)/generated/nca_minmax_r10.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_r10.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_r10.Tpo $(DEPDIR)/nca_minmax_r10.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_r10.c' object='nca_minmax_r10.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_r10.lo `test -f '$(srcdir)/generated/nca_minmax_r10.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_r10.c
+
+nca_minmax_r16.lo: $(srcdir)/generated/nca_minmax_r16.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_r16.lo -MD -MP -MF $(DEPDIR)/nca_minmax_r16.Tpo -c -o nca_minmax_r16.lo `test -f '$(srcdir)/generated/nca_minmax_r16.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_r16.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_r16.Tpo $(DEPDIR)/nca_minmax_r16.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_r16.c' object='nca_minmax_r16.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_r16.lo `test -f '$(srcdir)/generated/nca_minmax_r16.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_r16.c
+
+nca_minmax_s1.lo: $(srcdir)/generated/nca_minmax_s1.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_s1.lo -MD -MP -MF $(DEPDIR)/nca_minmax_s1.Tpo -c -o nca_minmax_s1.lo `test -f '$(srcdir)/generated/nca_minmax_s1.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_s1.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_s1.Tpo $(DEPDIR)/nca_minmax_s1.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_s1.c' object='nca_minmax_s1.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_s1.lo `test -f '$(srcdir)/generated/nca_minmax_s1.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_s1.c
+
+nca_minmax_s4.lo: $(srcdir)/generated/nca_minmax_s4.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT nca_minmax_s4.lo -MD -MP -MF $(DEPDIR)/nca_minmax_s4.Tpo -c -o nca_minmax_s4.lo `test -f '$(srcdir)/generated/nca_minmax_s4.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_s4.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/nca_minmax_s4.Tpo $(DEPDIR)/nca_minmax_s4.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='$(srcdir)/generated/nca_minmax_s4.c' object='nca_minmax_s4.lo' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o nca_minmax_s4.lo `test -f '$(srcdir)/generated/nca_minmax_s4.c' || echo '$(srcdir)/'`$(srcdir)/generated/nca_minmax_s4.c
+
bounds.lo: runtime/bounds.c
@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT bounds.lo -MD -MP -MF $(DEPDIR)/bounds.Tpo -c -o bounds.lo `test -f 'runtime/bounds.c' || echo '$(srcdir)/'`runtime/bounds.c
@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/bounds.Tpo $(DEPDIR)/bounds.Plo
@@ -6833,7 +7084,7 @@ check: $(BUILT_SOURCES)
$(MAKE) $(AM_MAKEFLAGS) check-am
all-am: Makefile $(LTLIBRARIES) $(DATA) $(HEADERS) config.h all-local
installdirs:
- for dir in "$(DESTDIR)$(cafexeclibdir)" "$(DESTDIR)$(toolexeclibdir)" "$(DESTDIR)$(toolexeclibdir)" "$(DESTDIR)$(gfor_cdir)" "$(DESTDIR)$(fincludedir)"; do \
+ for dir in "$(DESTDIR)$(cafexeclibdir)" "$(DESTDIR)$(mylibdir)" "$(DESTDIR)$(toolexeclibdir)" "$(DESTDIR)$(toolexeclibdir)" "$(DESTDIR)$(gfor_cdir)" "$(DESTDIR)$(fincludedir)"; do \
test -z "$$dir" || $(MKDIR_P) "$$dir"; \
done
install: $(BUILT_SOURCES)
@@ -6871,7 +7122,8 @@ maintainer-clean-generic:
clean: clean-am
clean-am: clean-cafexeclibLTLIBRARIES clean-generic clean-libtool \
- clean-local clean-toolexeclibLTLIBRARIES mostlyclean-am
+ clean-local clean-mylibLTLIBRARIES \
+ clean-toolexeclibLTLIBRARIES mostlyclean-am
distclean: distclean-am
-rm -f $(am__CONFIG_DISTCLEAN_FILES)
@@ -6892,7 +7144,8 @@ info: info-am
info-am:
-install-data-am: install-gfor_cHEADERS install-nodist_fincludeHEADERS
+install-data-am: install-gfor_cHEADERS install-mylibLTLIBRARIES \
+ install-nodist_fincludeHEADERS
install-dvi: install-dvi-am
@@ -6943,14 +7196,14 @@ ps: ps-am
ps-am:
uninstall-am: uninstall-cafexeclibLTLIBRARIES uninstall-gfor_cHEADERS \
- uninstall-nodist_fincludeHEADERS uninstall-toolexeclibDATA \
- uninstall-toolexeclibLTLIBRARIES
+ uninstall-mylibLTLIBRARIES uninstall-nodist_fincludeHEADERS \
+ uninstall-toolexeclibDATA uninstall-toolexeclibLTLIBRARIES
.MAKE: all check install install-am install-strip
.PHONY: CTAGS GTAGS TAGS all all-am all-local am--refresh check \
check-am clean clean-cafexeclibLTLIBRARIES clean-cscope \
- clean-generic clean-libtool clean-local \
+ clean-generic clean-libtool clean-local clean-mylibLTLIBRARIES \
clean-toolexeclibLTLIBRARIES cscope cscopelist-am ctags \
ctags-am distclean distclean-compile distclean-generic \
distclean-hdr distclean-libtool distclean-local distclean-tags \
@@ -6959,16 +7212,17 @@ uninstall-am: uninstall-cafexeclibLTLIBRARIES uninstall-gfor_cHEADERS \
install-dvi install-dvi-am install-exec install-exec-am \
install-exec-local install-gfor_cHEADERS install-html \
install-html-am install-info install-info-am install-man \
- install-nodist_fincludeHEADERS install-pdf install-pdf-am \
- install-ps install-ps-am install-strip install-toolexeclibDATA \
+ install-mylibLTLIBRARIES install-nodist_fincludeHEADERS \
+ install-pdf install-pdf-am install-ps install-ps-am \
+ install-strip install-toolexeclibDATA \
install-toolexeclibLTLIBRARIES installcheck installcheck-am \
installdirs maintainer-clean maintainer-clean-generic \
maintainer-clean-local mostlyclean mostlyclean-compile \
mostlyclean-generic mostlyclean-libtool mostlyclean-local pdf \
pdf-am ps ps-am tags tags-am uninstall uninstall-am \
uninstall-cafexeclibLTLIBRARIES uninstall-gfor_cHEADERS \
- uninstall-nodist_fincludeHEADERS uninstall-toolexeclibDATA \
- uninstall-toolexeclibLTLIBRARIES
+ uninstall-mylibLTLIBRARIES uninstall-nodist_fincludeHEADERS \
+ uninstall-toolexeclibDATA uninstall-toolexeclibLTLIBRARIES
.PRECIOUS: Makefile
@@ -7191,6 +7445,12 @@ fpu-target.inc: fpu-target.h $(srcdir)/libgfortran.h
@MAINTAINER_MODE_TRUE@$(gfor_misc_specifics): m4/misc_specifics.m4 m4/head.m4
@MAINTAINER_MODE_TRUE@ $(M4) -Dfile=$@ -I$(srcdir)/m4 misc_specifics.m4 > $@
+@LIBGFOR_NATIVE_COARRAY_TRUE@@MAINTAINER_MODE_TRUE@$(i_nca_minmax_c): m4/nca_minmax.m4 $(I_M4_DEPS)
+@LIBGFOR_NATIVE_COARRAY_TRUE@@MAINTAINER_MODE_TRUE@ $(M4) -Dfile=$@ -I$(srcdir)/m4 nca_minmax.m4 > $@
+
+@LIBGFOR_NATIVE_COARRAY_TRUE@@MAINTAINER_MODE_TRUE@$(i_nca_minmax_s_c): m4/nca-minmax-s.m4 $(I_M4_DEPS)
+@LIBGFOR_NATIVE_COARRAY_TRUE@@MAINTAINER_MODE_TRUE@ $(M4) -Dfile=$@ -I$(srcdir)/m4 nca-minmax-s.m4 > $@
+
# target overrides
-include $(tmake_file)
diff --git a/libgfortran/config.h.in b/libgfortran/config.h.in
index 2d58188e50c..795c9fe7b57 100644
--- a/libgfortran/config.h.in
+++ b/libgfortran/config.h.in
@@ -657,6 +657,15 @@
/* Define to 1 if you have the `powf' function. */
#undef HAVE_POWF
+/* Define to 1 if you have the `pthread_barrierattr_setpshared' function. */
+#undef HAVE_PTHREAD_BARRIERATTR_SETPSHARED
+
+/* Define to 1 if you have the `pthread_condattr_setpshared' function. */
+#undef HAVE_PTHREAD_CONDATTR_SETPSHARED
+
+/* Define to 1 if you have the `pthread_mutexattr_setpshared' function. */
+#undef HAVE_PTHREAD_MUTEXATTR_SETPSHARED
+
/* Define to 1 if the system has the type `ptrdiff_t'. */
#undef HAVE_PTRDIFF_T
@@ -687,6 +696,12 @@
/* Define to 1 if you have the `setmode' function. */
#undef HAVE_SETMODE
+/* Define to 1 if you have the `shm_open' function. */
+#undef HAVE_SHM_OPEN
+
+/* Define to 1 if you have the `shm_unlink' function. */
+#undef HAVE_SHM_UNLINK
+
/* Define to 1 if you have the `sigaction' function. */
#undef HAVE_SIGACTION
diff --git a/libgfortran/configure b/libgfortran/configure
index 854656960c4..8b9c9e43343 100755
--- a/libgfortran/configure
+++ b/libgfortran/configure
@@ -637,6 +637,11 @@ am__EXEEXT_TRUE
LTLIBOBJS
LIBOBJS
get_gcc_base_ver
+LIBGFOR_NATIVE_COARRAY_FALSE
+LIBGFOR_NATIVE_COARRAY_TRUE
+RT_LIBS
+PTHREAD_LIBS
+PTHREAD_CFLAGS
HAVE_AVX128_FALSE
HAVE_AVX128_TRUE
tmake_file
@@ -781,6 +786,7 @@ infodir
docdir
oldincludedir
includedir
+runstatedir
localstatedir
sharedstatedir
sysconfdir
@@ -872,6 +878,7 @@ datadir='${datarootdir}'
sysconfdir='${prefix}/etc'
sharedstatedir='${prefix}/com'
localstatedir='${prefix}/var'
+runstatedir='${localstatedir}/run'
includedir='${prefix}/include'
oldincludedir='/usr/include'
docdir='${datarootdir}/doc/${PACKAGE_TARNAME}'
@@ -1124,6 +1131,15 @@ do
| -silent | --silent | --silen | --sile | --sil)
silent=yes ;;
+ -runstatedir | --runstatedir | --runstatedi | --runstated \
+ | --runstate | --runstat | --runsta | --runst | --runs \
+ | --run | --ru | --r)
+ ac_prev=runstatedir ;;
+ -runstatedir=* | --runstatedir=* | --runstatedi=* | --runstated=* \
+ | --runstate=* | --runstat=* | --runsta=* | --runst=* | --runs=* \
+ | --run=* | --ru=* | --r=*)
+ runstatedir=$ac_optarg ;;
+
-sbindir | --sbindir | --sbindi | --sbind | --sbin | --sbi | --sb)
ac_prev=sbindir ;;
-sbindir=* | --sbindir=* | --sbindi=* | --sbind=* | --sbin=* \
@@ -1261,7 +1277,7 @@ fi
for ac_var in exec_prefix prefix bindir sbindir libexecdir datarootdir \
datadir sysconfdir sharedstatedir localstatedir includedir \
oldincludedir docdir infodir htmldir dvidir pdfdir psdir \
- libdir localedir mandir
+ libdir localedir mandir runstatedir
do
eval ac_val=\$$ac_var
# Remove trailing slashes.
@@ -1414,6 +1430,7 @@ Fine tuning of the installation directories:
--sysconfdir=DIR read-only single-machine data [PREFIX/etc]
--sharedstatedir=DIR modifiable architecture-independent data [PREFIX/com]
--localstatedir=DIR modifiable single-machine data [PREFIX/var]
+ --runstatedir=DIR modifiable per-process data [LOCALSTATEDIR/run]
--libdir=DIR object code libraries [EPREFIX/lib]
--includedir=DIR C header files [PREFIX/include]
--oldincludedir=DIR C header files for non-gcc [/usr/include]
@@ -12724,7 +12741,7 @@ else
lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2
lt_status=$lt_dlunknown
cat > conftest.$ac_ext <<_LT_EOF
-#line 12727 "configure"
+#line 12744 "configure"
#include "confdefs.h"
#if HAVE_DLFCN_H
@@ -12830,7 +12847,7 @@ else
lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2
lt_status=$lt_dlunknown
cat > conftest.$ac_ext <<_LT_EOF
-#line 12833 "configure"
+#line 12850 "configure"
#include "confdefs.h"
#if HAVE_DLFCN_H
@@ -16084,7 +16101,7 @@ else
We can't simply define LARGE_OFF_T to be 9223372036854775807,
since some C++ compilers masquerading as C compilers
incorrectly reject 9223372036854775807. */
-#define LARGE_OFF_T (((off_t) 1 << 62) - 1 + ((off_t) 1 << 62))
+#define LARGE_OFF_T ((((off_t) 1 << 31) << 31) - 1 + (((off_t) 1 << 31) << 31))
int off_t_is_large[(LARGE_OFF_T % 2147483629 == 721
&& LARGE_OFF_T % 2147483647 == 1)
? 1 : -1];
@@ -16130,7 +16147,7 @@ else
We can't simply define LARGE_OFF_T to be 9223372036854775807,
since some C++ compilers masquerading as C compilers
incorrectly reject 9223372036854775807. */
-#define LARGE_OFF_T (((off_t) 1 << 62) - 1 + ((off_t) 1 << 62))
+#define LARGE_OFF_T ((((off_t) 1 << 31) << 31) - 1 + (((off_t) 1 << 31) << 31))
int off_t_is_large[(LARGE_OFF_T % 2147483629 == 721
&& LARGE_OFF_T % 2147483647 == 1)
? 1 : -1];
@@ -16154,7 +16171,7 @@ rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
We can't simply define LARGE_OFF_T to be 9223372036854775807,
since some C++ compilers masquerading as C compilers
incorrectly reject 9223372036854775807. */
-#define LARGE_OFF_T (((off_t) 1 << 62) - 1 + ((off_t) 1 << 62))
+#define LARGE_OFF_T ((((off_t) 1 << 31) << 31) - 1 + (((off_t) 1 << 31) << 31))
int off_t_is_large[(LARGE_OFF_T % 2147483629 == 721
&& LARGE_OFF_T % 2147483647 == 1)
? 1 : -1];
@@ -16199,7 +16216,7 @@ else
We can't simply define LARGE_OFF_T to be 9223372036854775807,
since some C++ compilers masquerading as C compilers
incorrectly reject 9223372036854775807. */
-#define LARGE_OFF_T (((off_t) 1 << 62) - 1 + ((off_t) 1 << 62))
+#define LARGE_OFF_T ((((off_t) 1 << 31) << 31) - 1 + (((off_t) 1 << 31) << 31))
int off_t_is_large[(LARGE_OFF_T % 2147483629 == 721
&& LARGE_OFF_T % 2147483647 == 1)
? 1 : -1];
@@ -16223,7 +16240,7 @@ rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
We can't simply define LARGE_OFF_T to be 9223372036854775807,
since some C++ compilers masquerading as C compilers
incorrectly reject 9223372036854775807. */
-#define LARGE_OFF_T (((off_t) 1 << 62) - 1 + ((off_t) 1 << 62))
+#define LARGE_OFF_T ((((off_t) 1 << 31) << 31) - 1 + (((off_t) 1 << 31) << 31))
int off_t_is_large[(LARGE_OFF_T % 2147483629 == 721
&& LARGE_OFF_T % 2147483647 == 1)
? 1 : -1];
@@ -27132,6 +27149,167 @@ rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
CFLAGS="$ac_save_CFLAGS"
+# Tests related to native coarrays
+# Test whether the compiler supports the -pthread option.
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether -pthread is supported" >&5
+$as_echo_n "checking whether -pthread is supported... " >&6; }
+if ${libgfortran_cv_lib_pthread+:} false; then :
+ $as_echo_n "(cached) " >&6
+else
+ CFLAGS_hold=$CFLAGS
+CFLAGS="$CFLAGS -pthread -L../libatomic/.libs"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h. */
+int i;
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+ libgfortran_cv_lib_pthread=yes
+else
+ libgfortran_cv_lib_pthread=no
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+CFLAGS=$CFLAGS_hold
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libgfortran_cv_lib_pthread" >&5
+$as_echo "$libgfortran_cv_lib_pthread" >&6; }
+PTHREAD_CFLAGS=
+if test "$libgfortran_cv_lib_pthread" = yes; then
+ # RISC-V apparently adds -latomic when using -pthread.
+ PTHREAD_CFLAGS="-pthread -L../libatomic/.libs"
+fi
+
+
+# Test for the -lpthread library.
+PTHREAD_LIBS=
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for pthread_create in -lpthread" >&5
+$as_echo_n "checking for pthread_create in -lpthread... " >&6; }
+if ${ac_cv_lib_pthread_pthread_create+:} false; then :
+ $as_echo_n "(cached) " >&6
+else
+ ac_check_lib_save_LIBS=$LIBS
+LIBS="-lpthread $LIBS"
+if test x$gcc_no_link = xyes; then
+ as_fn_error $? "Link tests are not allowed after GCC_NO_EXECUTABLES." "$LINENO" 5
+fi
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h. */
+
+/* Override any GCC internal prototype to avoid an error.
+ Use char because int might match the return type of a GCC
+ builtin and then its argument prototype would still apply. */
+#ifdef __cplusplus
+extern "C"
+#endif
+char pthread_create ();
+int
+main ()
+{
+return pthread_create ();
+ ;
+ return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+ ac_cv_lib_pthread_pthread_create=yes
+else
+ ac_cv_lib_pthread_pthread_create=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+ conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_pthread_pthread_create" >&5
+$as_echo "$ac_cv_lib_pthread_pthread_create" >&6; }
+if test "x$ac_cv_lib_pthread_pthread_create" = xyes; then :
+ PTHREAD_LIBS=-lpthread
+fi
+
+
+
+# Test if -lrt is required
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for shm_open in -lrt" >&5
+$as_echo_n "checking for shm_open in -lrt... " >&6; }
+if ${ac_cv_lib_rt_shm_open+:} false; then :
+ $as_echo_n "(cached) " >&6
+else
+ ac_check_lib_save_LIBS=$LIBS
+LIBS="-lrt $LIBS"
+if test x$gcc_no_link = xyes; then
+ as_fn_error $? "Link tests are not allowed after GCC_NO_EXECUTABLES." "$LINENO" 5
+fi
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h. */
+
+/* Override any GCC internal prototype to avoid an error.
+ Use char because int might match the return type of a GCC
+ builtin and then its argument prototype would still apply. */
+#ifdef __cplusplus
+extern "C"
+#endif
+char shm_open ();
+int
+main ()
+{
+return shm_open ();
+ ;
+ return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+ ac_cv_lib_rt_shm_open=yes
+else
+ ac_cv_lib_rt_shm_open=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+ conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_rt_shm_open" >&5
+$as_echo "$ac_cv_lib_rt_shm_open" >&6; }
+if test "x$ac_cv_lib_rt_shm_open" = xyes; then :
+ RT_LIBS=-lrt
+fi
+
+
+
+CFLAGS_hold="$CFLAGS"
+CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+LIBS_hold="$LIBS"
+LIBS="$LIBS $PTHREAD_LIBS $RT_LIBS"
+
+# Find the functions that we need
+for ac_func in pthread_condattr_setpshared pthread_mutexattr_setpshared pthread_barrierattr_setpshared shm_open shm_unlink
+do :
+ as_ac_var=`$as_echo "ac_cv_func_$ac_func" | $as_tr_sh`
+ac_fn_c_check_func "$LINENO" "$ac_func" "$as_ac_var"
+if eval test \"x\$"$as_ac_var"\" = x"yes"; then :
+ cat >>confdefs.h <<_ACEOF
+#define `$as_echo "HAVE_$ac_func" | $as_tr_cpp` 1
+_ACEOF
+ if true; then
+ LIBGFOR_NATIVE_COARRAY_TRUE=
+ LIBGFOR_NATIVE_COARRAY_FALSE='#'
+else
+ LIBGFOR_NATIVE_COARRAY_TRUE='#'
+ LIBGFOR_NATIVE_COARRAY_FALSE=
+fi
+
+else
+ if false; then
+ LIBGFOR_NATIVE_COARRAY_TRUE=
+ LIBGFOR_NATIVE_COARRAY_FALSE='#'
+else
+ LIBGFOR_NATIVE_COARRAY_TRUE='#'
+ LIBGFOR_NATIVE_COARRAY_FALSE=
+fi
+
+fi
+done
+
+
+CFLAGS="$CFLAGS_hold"
+LIBS="$LIBS_hold"
+
# Determine what GCC version number to use in filesystem paths.
get_gcc_base_ver="cat"
@@ -27423,6 +27601,14 @@ if test -z "${HAVE_AVX128_TRUE}" && test -z "${HAVE_AVX128_FALSE}"; then
as_fn_error $? "conditional \"HAVE_AVX128\" was never defined.
Usually this means the macro was only invoked conditionally." "$LINENO" 5
fi
+if test -z "${LIBGFOR_NATIVE_COARRAY_TRUE}" && test -z "${LIBGFOR_NATIVE_COARRAY_FALSE}"; then
+ as_fn_error $? "conditional \"LIBGFOR_NATIVE_COARRAY\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
+if test -z "${LIBGFOR_NATIVE_COARRAY_TRUE}" && test -z "${LIBGFOR_NATIVE_COARRAY_FALSE}"; then
+ as_fn_error $? "conditional \"LIBGFOR_NATIVE_COARRAY\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
: "${CONFIG_STATUS=./config.status}"
ac_write_fail=0
diff --git a/libgfortran/configure.ac b/libgfortran/configure.ac
index 4109d0fefae..bff8209c151 100644
--- a/libgfortran/configure.ac
+++ b/libgfortran/configure.ac
@@ -670,6 +670,43 @@ LIBGFOR_CHECK_FMA4
# Check if AVX128 works
LIBGFOR_CHECK_AVX128
+# Tests related to native coarrays
+# Test whether the compiler supports the -pthread option.
+AC_CACHE_CHECK([whether -pthread is supported],
+[libgfortran_cv_lib_pthread],
+[CFLAGS_hold=$CFLAGS
+CFLAGS="$CFLAGS -pthread -L../libatomic/.libs"
+AC_COMPILE_IFELSE([AC_LANG_SOURCE([int i;])],
+[libgfortran_cv_lib_pthread=yes],
+[libgfortran_cv_lib_pthread=no])
+CFLAGS=$CFLAGS_hold])
+PTHREAD_CFLAGS=
+if test "$libgfortran_cv_lib_pthread" = yes; then
+ # RISC-V apparently adds -latomic when using -pthread.
+ PTHREAD_CFLAGS="-pthread -L../libatomic/.libs"
+fi
+AC_SUBST(PTHREAD_CFLAGS)
+
+# Test for the -lpthread library.
+PTHREAD_LIBS=
+AC_CHECK_LIB([pthread], [pthread_create], PTHREAD_LIBS=-lpthread)
+AC_SUBST(PTHREAD_LIBS)
+
+# Test if -lrt is required
+AC_CHECK_LIB([rt], [shm_open], RT_LIBS=-lrt)
+AC_SUBST(RT_LIBS)
+
+CFLAGS_hold="$CFLAGS"
+CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+LIBS_hold="$LIBS"
+LIBS="$LIBS $PTHREAD_LIBS $RT_LIBS"
+
+# Find the functions that we need
+AC_CHECK_FUNCS([pthread_condattr_setpshared pthread_mutexattr_setpshared pthread_barrierattr_setpshared shm_open shm_unlink],[AM_CONDITIONAL(LIBGFOR_NATIVE_COARRAY,true)],[AM_CONDITIONAL(LIBGFOR_NATIVE_COARRAY,false)])
+
+CFLAGS="$CFLAGS_hold"
+LIBS="$LIBS_hold"
+
# Determine what GCC version number to use in filesystem paths.
GCC_BASE_VER
diff --git a/libgfortran/generated/nca_minmax_i1.c b/libgfortran/generated/nca_minmax_i1.c
new file mode 100644
index 00000000000..3bc9a2b75ec
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_i1.c
@@ -0,0 +1,653 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_INTEGER_1)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+void nca_collsub_max_scalar_i1 (GFC_INTEGER_1 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_i1);
+
+void
+nca_collsub_max_scalar_i1 (GFC_INTEGER_1 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_1 *a, *b;
+ GFC_INTEGER_1 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_1) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b > *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_i1 (GFC_INTEGER_1 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_i1);
+
+void
+nca_collsub_min_scalar_i1 (GFC_INTEGER_1 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_1 *a, *b;
+ GFC_INTEGER_1 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_1) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b < *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_sum_scalar_i1 (GFC_INTEGER_1 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_sum_scalar_i1);
+
+void
+nca_collsub_sum_scalar_i1 (GFC_INTEGER_1 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_1 *a, *b;
+ GFC_INTEGER_1 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_1) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ *a += *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_i1 (gfc_array_i1 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_max_array_i1);
+
+void
+nca_collsub_max_array_i1 (gfc_array_i1 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_1 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_1 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_1);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_1);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_1);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_1 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_1 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_1 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_1 *a;
+ GFC_INTEGER_1 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b > *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_1 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_1 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_i1 (gfc_array_i1 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_min_array_i1);
+
+void
+nca_collsub_min_array_i1 (gfc_array_i1 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_1 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_1 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_1);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_1);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_1);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_1 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_1 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_1 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_1 *a;
+ GFC_INTEGER_1 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b < *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_1 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_1 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_sum_array_i1 (gfc_array_i1 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_sum_array_i1);
+
+void
+nca_collsub_sum_array_i1 (gfc_array_i1 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_1 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_1 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_1);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_1);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_1);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_1 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_1 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_1 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_1 *a;
+ GFC_INTEGER_1 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ *a += *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_1 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_1 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/generated/nca_minmax_i16.c b/libgfortran/generated/nca_minmax_i16.c
new file mode 100644
index 00000000000..8fbb9481271
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_i16.c
@@ -0,0 +1,653 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_INTEGER_16)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+void nca_collsub_max_scalar_i16 (GFC_INTEGER_16 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_i16);
+
+void
+nca_collsub_max_scalar_i16 (GFC_INTEGER_16 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_16 *a, *b;
+ GFC_INTEGER_16 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_16) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b > *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_i16 (GFC_INTEGER_16 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_i16);
+
+void
+nca_collsub_min_scalar_i16 (GFC_INTEGER_16 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_16 *a, *b;
+ GFC_INTEGER_16 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_16) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b < *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_sum_scalar_i16 (GFC_INTEGER_16 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_sum_scalar_i16);
+
+void
+nca_collsub_sum_scalar_i16 (GFC_INTEGER_16 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_16 *a, *b;
+ GFC_INTEGER_16 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_16) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ *a += *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_i16 (gfc_array_i16 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_max_array_i16);
+
+void
+nca_collsub_max_array_i16 (gfc_array_i16 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_16 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_16 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_16);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_16);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_16);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_16 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_16 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_16 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_16 *a;
+ GFC_INTEGER_16 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b > *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_16 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_16 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_i16 (gfc_array_i16 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_min_array_i16);
+
+void
+nca_collsub_min_array_i16 (gfc_array_i16 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_16 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_16 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_16);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_16);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_16);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_16 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_16 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_16 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_16 *a;
+ GFC_INTEGER_16 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b < *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_16 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_16 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_sum_array_i16 (gfc_array_i16 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_sum_array_i16);
+
+void
+nca_collsub_sum_array_i16 (gfc_array_i16 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_16 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_16 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_16);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_16);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_16);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_16 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_16 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_16 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_16 *a;
+ GFC_INTEGER_16 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ *a += *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_16 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_16 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/generated/nca_minmax_i2.c b/libgfortran/generated/nca_minmax_i2.c
new file mode 100644
index 00000000000..61908d6bf13
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_i2.c
@@ -0,0 +1,653 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_INTEGER_2)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+void nca_collsub_max_scalar_i2 (GFC_INTEGER_2 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_i2);
+
+void
+nca_collsub_max_scalar_i2 (GFC_INTEGER_2 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_2 *a, *b;
+ GFC_INTEGER_2 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_2) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b > *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_i2 (GFC_INTEGER_2 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_i2);
+
+void
+nca_collsub_min_scalar_i2 (GFC_INTEGER_2 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_2 *a, *b;
+ GFC_INTEGER_2 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_2) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b < *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_sum_scalar_i2 (GFC_INTEGER_2 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_sum_scalar_i2);
+
+void
+nca_collsub_sum_scalar_i2 (GFC_INTEGER_2 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_2 *a, *b;
+ GFC_INTEGER_2 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_2) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ *a += *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_i2 (gfc_array_i2 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_max_array_i2);
+
+void
+nca_collsub_max_array_i2 (gfc_array_i2 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_2 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_2 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_2);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_2);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_2);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_2 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_2 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_2 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_2 *a;
+ GFC_INTEGER_2 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b > *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_2 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_2 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_i2 (gfc_array_i2 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_min_array_i2);
+
+void
+nca_collsub_min_array_i2 (gfc_array_i2 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_2 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_2 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_2);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_2);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_2);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_2 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_2 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_2 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_2 *a;
+ GFC_INTEGER_2 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b < *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_2 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_2 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_sum_array_i2 (gfc_array_i2 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_sum_array_i2);
+
+void
+nca_collsub_sum_array_i2 (gfc_array_i2 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_2 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_2 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_2);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_2);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_2);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_2 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_2 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_2 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_2 *a;
+ GFC_INTEGER_2 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ *a += *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_2 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_2 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/generated/nca_minmax_i4.c b/libgfortran/generated/nca_minmax_i4.c
new file mode 100644
index 00000000000..5e37586478d
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_i4.c
@@ -0,0 +1,653 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_INTEGER_4)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+void nca_collsub_max_scalar_i4 (GFC_INTEGER_4 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_i4);
+
+void
+nca_collsub_max_scalar_i4 (GFC_INTEGER_4 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_4 *a, *b;
+ GFC_INTEGER_4 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_4) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b > *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_i4 (GFC_INTEGER_4 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_i4);
+
+void
+nca_collsub_min_scalar_i4 (GFC_INTEGER_4 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_4 *a, *b;
+ GFC_INTEGER_4 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_4) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b < *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_sum_scalar_i4 (GFC_INTEGER_4 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_sum_scalar_i4);
+
+void
+nca_collsub_sum_scalar_i4 (GFC_INTEGER_4 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_4 *a, *b;
+ GFC_INTEGER_4 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_4) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ *a += *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_i4 (gfc_array_i4 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_max_array_i4);
+
+void
+nca_collsub_max_array_i4 (gfc_array_i4 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_4 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_4 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_4);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_4);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_4);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_4 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_4 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_4 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_4 *a;
+ GFC_INTEGER_4 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b > *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_4 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_4 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_i4 (gfc_array_i4 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_min_array_i4);
+
+void
+nca_collsub_min_array_i4 (gfc_array_i4 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_4 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_4 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_4);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_4);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_4);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_4 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_4 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_4 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_4 *a;
+ GFC_INTEGER_4 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b < *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_4 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_4 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_sum_array_i4 (gfc_array_i4 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_sum_array_i4);
+
+void
+nca_collsub_sum_array_i4 (gfc_array_i4 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_4 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_4 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_4);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_4);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_4);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_4 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_4 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_4 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_4 *a;
+ GFC_INTEGER_4 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ *a += *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_4 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_4 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/generated/nca_minmax_i8.c b/libgfortran/generated/nca_minmax_i8.c
new file mode 100644
index 00000000000..b3dc8611981
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_i8.c
@@ -0,0 +1,653 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_INTEGER_8)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+void nca_collsub_max_scalar_i8 (GFC_INTEGER_8 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_i8);
+
+void
+nca_collsub_max_scalar_i8 (GFC_INTEGER_8 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_8 *a, *b;
+ GFC_INTEGER_8 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_8) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b > *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_i8 (GFC_INTEGER_8 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_i8);
+
+void
+nca_collsub_min_scalar_i8 (GFC_INTEGER_8 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_8 *a, *b;
+ GFC_INTEGER_8 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_8) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b < *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_sum_scalar_i8 (GFC_INTEGER_8 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_sum_scalar_i8);
+
+void
+nca_collsub_sum_scalar_i8 (GFC_INTEGER_8 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_INTEGER_8 *a, *b;
+ GFC_INTEGER_8 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_INTEGER_8) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ *a += *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_i8 (gfc_array_i8 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_max_array_i8);
+
+void
+nca_collsub_max_array_i8 (gfc_array_i8 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_8 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_8 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_8);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_8);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_8);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_8 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_8 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_8 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_8 *a;
+ GFC_INTEGER_8 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b > *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_8 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_8 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_i8 (gfc_array_i8 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_min_array_i8);
+
+void
+nca_collsub_min_array_i8 (gfc_array_i8 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_8 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_8 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_8);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_8);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_8);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_8 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_8 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_8 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_8 *a;
+ GFC_INTEGER_8 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b < *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_8 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_8 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_sum_array_i8 (gfc_array_i8 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_sum_array_i8);
+
+void
+nca_collsub_sum_array_i8 (gfc_array_i8 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_INTEGER_8 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_INTEGER_8 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_INTEGER_8);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_INTEGER_8);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_INTEGER_8);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_INTEGER_8 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_INTEGER_8 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_INTEGER_8 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_INTEGER_8 *a;
+ GFC_INTEGER_8 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ *a += *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_INTEGER_8 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_INTEGER_8 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/generated/nca_minmax_r10.c b/libgfortran/generated/nca_minmax_r10.c
new file mode 100644
index 00000000000..10f7324fc92
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_r10.c
@@ -0,0 +1,653 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_REAL_10)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+void nca_collsub_max_scalar_r10 (GFC_REAL_10 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_r10);
+
+void
+nca_collsub_max_scalar_r10 (GFC_REAL_10 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_10 *a, *b;
+ GFC_REAL_10 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_10) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b > *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_r10 (GFC_REAL_10 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_r10);
+
+void
+nca_collsub_min_scalar_r10 (GFC_REAL_10 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_10 *a, *b;
+ GFC_REAL_10 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_10) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b < *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_sum_scalar_r10 (GFC_REAL_10 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_sum_scalar_r10);
+
+void
+nca_collsub_sum_scalar_r10 (GFC_REAL_10 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_10 *a, *b;
+ GFC_REAL_10 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_10) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ *a += *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_r10 (gfc_array_r10 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_max_array_r10);
+
+void
+nca_collsub_max_array_r10 (gfc_array_r10 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_10 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_10 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_10);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_10);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_10);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_10 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_10 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_10 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_10 *a;
+ GFC_REAL_10 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b > *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_10 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_10 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_r10 (gfc_array_r10 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_min_array_r10);
+
+void
+nca_collsub_min_array_r10 (gfc_array_r10 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_10 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_10 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_10);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_10);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_10);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_10 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_10 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_10 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_10 *a;
+ GFC_REAL_10 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b < *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_10 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_10 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_sum_array_r10 (gfc_array_r10 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_sum_array_r10);
+
+void
+nca_collsub_sum_array_r10 (gfc_array_r10 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_10 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_10 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_10);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_10);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_10);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_10 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_10 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_10 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_10 *a;
+ GFC_REAL_10 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ *a += *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_10 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_10 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/generated/nca_minmax_r16.c b/libgfortran/generated/nca_minmax_r16.c
new file mode 100644
index 00000000000..a0a0a5164c0
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_r16.c
@@ -0,0 +1,653 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_REAL_16)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+void nca_collsub_max_scalar_r16 (GFC_REAL_16 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_r16);
+
+void
+nca_collsub_max_scalar_r16 (GFC_REAL_16 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_16 *a, *b;
+ GFC_REAL_16 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_16) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b > *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_r16 (GFC_REAL_16 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_r16);
+
+void
+nca_collsub_min_scalar_r16 (GFC_REAL_16 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_16 *a, *b;
+ GFC_REAL_16 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_16) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b < *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_sum_scalar_r16 (GFC_REAL_16 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_sum_scalar_r16);
+
+void
+nca_collsub_sum_scalar_r16 (GFC_REAL_16 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_16 *a, *b;
+ GFC_REAL_16 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_16) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ *a += *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_r16 (gfc_array_r16 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_max_array_r16);
+
+void
+nca_collsub_max_array_r16 (gfc_array_r16 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_16 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_16 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_16);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_16);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_16);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_16 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_16 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_16 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_16 *a;
+ GFC_REAL_16 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b > *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_16 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_16 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_r16 (gfc_array_r16 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_min_array_r16);
+
+void
+nca_collsub_min_array_r16 (gfc_array_r16 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_16 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_16 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_16);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_16);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_16);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_16 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_16 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_16 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_16 *a;
+ GFC_REAL_16 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b < *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_16 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_16 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_sum_array_r16 (gfc_array_r16 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_sum_array_r16);
+
+void
+nca_collsub_sum_array_r16 (gfc_array_r16 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_16 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_16 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_16);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_16);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_16);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_16 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_16 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_16 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_16 *a;
+ GFC_REAL_16 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ *a += *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_16 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_16 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/generated/nca_minmax_r4.c b/libgfortran/generated/nca_minmax_r4.c
new file mode 100644
index 00000000000..0eb3f1b6340
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_r4.c
@@ -0,0 +1,653 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_REAL_4)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+void nca_collsub_max_scalar_r4 (GFC_REAL_4 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_r4);
+
+void
+nca_collsub_max_scalar_r4 (GFC_REAL_4 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_4 *a, *b;
+ GFC_REAL_4 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_4) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b > *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_r4 (GFC_REAL_4 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_r4);
+
+void
+nca_collsub_min_scalar_r4 (GFC_REAL_4 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_4 *a, *b;
+ GFC_REAL_4 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_4) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b < *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_sum_scalar_r4 (GFC_REAL_4 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_sum_scalar_r4);
+
+void
+nca_collsub_sum_scalar_r4 (GFC_REAL_4 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_4 *a, *b;
+ GFC_REAL_4 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_4) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ *a += *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_r4 (gfc_array_r4 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_max_array_r4);
+
+void
+nca_collsub_max_array_r4 (gfc_array_r4 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_4 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_4 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_4);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_4);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_4);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_4 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_4 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_4 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_4 *a;
+ GFC_REAL_4 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b > *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_4 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_4 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_r4 (gfc_array_r4 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_min_array_r4);
+
+void
+nca_collsub_min_array_r4 (gfc_array_r4 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_4 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_4 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_4);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_4);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_4);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_4 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_4 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_4 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_4 *a;
+ GFC_REAL_4 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b < *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_4 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_4 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_sum_array_r4 (gfc_array_r4 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_sum_array_r4);
+
+void
+nca_collsub_sum_array_r4 (gfc_array_r4 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_4 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_4 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_4);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_4);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_4);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_4 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_4 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_4 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_4 *a;
+ GFC_REAL_4 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ *a += *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_4 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_4 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/generated/nca_minmax_r8.c b/libgfortran/generated/nca_minmax_r8.c
new file mode 100644
index 00000000000..3b3e9623150
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_r8.c
@@ -0,0 +1,653 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_REAL_8)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+void nca_collsub_max_scalar_r8 (GFC_REAL_8 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_r8);
+
+void
+nca_collsub_max_scalar_r8 (GFC_REAL_8 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_8 *a, *b;
+ GFC_REAL_8 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_8) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b > *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_r8 (GFC_REAL_8 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_r8);
+
+void
+nca_collsub_min_scalar_r8 (GFC_REAL_8 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_8 *a, *b;
+ GFC_REAL_8 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_8) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ if (*b < *a)
+ *a = *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_sum_scalar_r8 (GFC_REAL_8 *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_sum_scalar_r8);
+
+void
+nca_collsub_sum_scalar_r8 (GFC_REAL_8 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_REAL_8 *a, *b;
+ GFC_REAL_8 *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof(GFC_REAL_8) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ *a += *b;
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_r8 (gfc_array_r8 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_max_array_r8);
+
+void
+nca_collsub_max_array_r8 (gfc_array_r8 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_8 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_8 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_8);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_8);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_8);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_8 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_8 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_8 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_8 *a;
+ GFC_REAL_8 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b > *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_8 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_8 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_r8 (gfc_array_r8 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_min_array_r8);
+
+void
+nca_collsub_min_array_r8 (gfc_array_r8 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_8 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_8 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_8);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_8);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_8);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_8 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_8 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_8 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_8 *a;
+ GFC_REAL_8 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ if (*b < *a)
+ *a = *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_8 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_8 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_sum_array_r8 (gfc_array_r8 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_sum_array_r8);
+
+void
+nca_collsub_sum_array_r8 (gfc_array_r8 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ GFC_REAL_8 *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ GFC_REAL_8 *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof (GFC_REAL_8);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof (GFC_REAL_8);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof (GFC_REAL_8);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ GFC_REAL_8 *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *((GFC_REAL_8 *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ GFC_REAL_8 * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_REAL_8 *a;
+ GFC_REAL_8 *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ *a += *b;
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ GFC_REAL_8 *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *((GFC_REAL_8 * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/generated/nca_minmax_s1.c b/libgfortran/generated/nca_minmax_s1.c
new file mode 100644
index 00000000000..b081452948d
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_s1.c
@@ -0,0 +1,494 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_UINTEGER_1)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+#if 1 == 4
+
+/* Compare wide character types, which are handled internally as
+ unsigned 4-byte integers. */
+static inline int
+memcmp4 (const void *a, const void *b, size_t len)
+{
+ const GFC_UINTEGER_4 *pa = a;
+ const GFC_UINTEGER_4 *pb = b;
+ while (len-- > 0)
+ {
+ if (*pa != *pb)
+ return *pa < *pb ? -1 : 1;
+ pa ++;
+ pb ++;
+ }
+ return 0;
+}
+
+#endif
+void nca_collsub_max_scalar_s1 (GFC_UINTEGER_1 *obj, int *result_image,
+ int *stat, char *errmsg, index_type char_len, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_s1);
+
+void
+nca_collsub_max_scalar_s1 (GFC_UINTEGER_1 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_UINTEGER_1 *a, *b;
+ GFC_UINTEGER_1 *buffer, *this_image_buf;
+ collsub_iface *ci;
+ index_type type_size;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof (GFC_UINTEGER_1);
+ buffer = get_collsub_buf (ci, type_size * local->num_images);
+ this_image_buf = buffer + this_image.image_num * char_len;
+ memcpy (this_image_buf, obj, type_size);
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset * char_len;
+ if (memcmp (b, a, char_len) > 0)
+ memcpy (a, b, type_size);
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ memcpy (obj, buffer, type_size);
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_s1 (GFC_UINTEGER_1 *obj, int *result_image,
+ int *stat, char *errmsg, index_type char_len, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_s1);
+
+void
+nca_collsub_min_scalar_s1 (GFC_UINTEGER_1 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_UINTEGER_1 *a, *b;
+ GFC_UINTEGER_1 *buffer, *this_image_buf;
+ collsub_iface *ci;
+ index_type type_size;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof (GFC_UINTEGER_1);
+ buffer = get_collsub_buf (ci, type_size * local->num_images);
+ this_image_buf = buffer + this_image.image_num * char_len;
+ memcpy (this_image_buf, obj, type_size);
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset * char_len;
+ if (memcmp (b, a, char_len) < 0)
+ memcpy (a, b, type_size);
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ memcpy (obj, buffer, type_size);
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_s1 (gfc_array_s1 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type char_len,
+ index_type errmsg_len);
+export_proto (nca_collsub_max_array_s1);
+
+void
+nca_collsub_max_array_s1 (gfc_array_s1 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS]; /* stride is byte-based here. */
+ index_type extent[GFC_MAX_DIMENSIONS];
+ char *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ char *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ index_type type_size;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof (GFC_UINTEGER_1);
+ dim = GFC_DESCRIPTOR_RANK (array);
+ num_elems = 1;
+ ssize = type_size;
+ packed = true;
+ span = array->span != 0 ? array->span : type_size;
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (num_elems != GFC_DESCRIPTOR_STRIDE (array,n))
+ packed = false;
+
+ num_elems *= extent[n];
+ }
+
+ ssize = num_elems * type_size;
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * ssize;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ char *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+
+ memcpy (dest, src, type_size);
+ dest += type_size;
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ char *other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_UINTEGER_1 *a;
+ GFC_UINTEGER_1 *b;
+
+ other_shared_ptr = this_shared_ptr + imoffset * ssize;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = (GFC_UINTEGER_1 *) (this_shared_ptr + i * type_size);
+ b = (GFC_UINTEGER_1 *) (other_shared_ptr + i * type_size);
+ if (memcmp (b, a, char_len) > 0)
+ memcpy (a, b, type_size);
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ char *src = buffer;
+ char *restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ memcpy (dest, src, type_size);
+ src += span;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_s1 (gfc_array_s1 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type char_len,
+ index_type errmsg_len);
+export_proto (nca_collsub_min_array_s1);
+
+void
+nca_collsub_min_array_s1 (gfc_array_s1 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS]; /* stride is byte-based here. */
+ index_type extent[GFC_MAX_DIMENSIONS];
+ char *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ char *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ index_type type_size;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof (GFC_UINTEGER_1);
+ dim = GFC_DESCRIPTOR_RANK (array);
+ num_elems = 1;
+ ssize = type_size;
+ packed = true;
+ span = array->span != 0 ? array->span : type_size;
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (num_elems != GFC_DESCRIPTOR_STRIDE (array,n))
+ packed = false;
+
+ num_elems *= extent[n];
+ }
+
+ ssize = num_elems * type_size;
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * ssize;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ char *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+
+ memcpy (dest, src, type_size);
+ dest += type_size;
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ char *other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_UINTEGER_1 *a;
+ GFC_UINTEGER_1 *b;
+
+ other_shared_ptr = this_shared_ptr + imoffset * ssize;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = (GFC_UINTEGER_1 *) (this_shared_ptr + i * type_size);
+ b = (GFC_UINTEGER_1 *) (other_shared_ptr + i * type_size);
+ if (memcmp (b, a, char_len) < 0)
+ memcpy (a, b, type_size);
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ char *src = buffer;
+ char *restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ memcpy (dest, src, type_size);
+ src += span;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/generated/nca_minmax_s4.c b/libgfortran/generated/nca_minmax_s4.c
new file mode 100644
index 00000000000..b202fda0cc4
--- /dev/null
+++ b/libgfortran/generated/nca_minmax_s4.c
@@ -0,0 +1,494 @@
+/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+
+#include "libgfortran.h"
+
+#if defined (HAVE_GFC_UINTEGER_4)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+#if 4 == 4
+
+/* Compare wide character types, which are handled internally as
+ unsigned 4-byte integers. */
+static inline int
+memcmp4 (const void *a, const void *b, size_t len)
+{
+ const GFC_UINTEGER_4 *pa = a;
+ const GFC_UINTEGER_4 *pb = b;
+ while (len-- > 0)
+ {
+ if (*pa != *pb)
+ return *pa < *pb ? -1 : 1;
+ pa ++;
+ pb ++;
+ }
+ return 0;
+}
+
+#endif
+void nca_collsub_max_scalar_s4 (GFC_UINTEGER_4 *obj, int *result_image,
+ int *stat, char *errmsg, index_type char_len, index_type errmsg_len);
+export_proto(nca_collsub_max_scalar_s4);
+
+void
+nca_collsub_max_scalar_s4 (GFC_UINTEGER_4 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_UINTEGER_4 *a, *b;
+ GFC_UINTEGER_4 *buffer, *this_image_buf;
+ collsub_iface *ci;
+ index_type type_size;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof (GFC_UINTEGER_4);
+ buffer = get_collsub_buf (ci, type_size * local->num_images);
+ this_image_buf = buffer + this_image.image_num * char_len;
+ memcpy (this_image_buf, obj, type_size);
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset * char_len;
+ if (memcmp4 (b, a, char_len) > 0)
+ memcpy (a, b, type_size);
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ memcpy (obj, buffer, type_size);
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_min_scalar_s4 (GFC_UINTEGER_4 *obj, int *result_image,
+ int *stat, char *errmsg, index_type char_len, index_type errmsg_len);
+export_proto(nca_collsub_min_scalar_s4);
+
+void
+nca_collsub_min_scalar_s4 (GFC_UINTEGER_4 *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ GFC_UINTEGER_4 *a, *b;
+ GFC_UINTEGER_4 *buffer, *this_image_buf;
+ collsub_iface *ci;
+ index_type type_size;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof (GFC_UINTEGER_4);
+ buffer = get_collsub_buf (ci, type_size * local->num_images);
+ this_image_buf = buffer + this_image.image_num * char_len;
+ memcpy (this_image_buf, obj, type_size);
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset * char_len;
+ if (memcmp4 (b, a, char_len) < 0)
+ memcpy (a, b, type_size);
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ memcpy (obj, buffer, type_size);
+
+ finish_collective_subroutine (ci);
+
+}
+
+void nca_collsub_max_array_s4 (gfc_array_s4 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type char_len,
+ index_type errmsg_len);
+export_proto (nca_collsub_max_array_s4);
+
+void
+nca_collsub_max_array_s4 (gfc_array_s4 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS]; /* stride is byte-based here. */
+ index_type extent[GFC_MAX_DIMENSIONS];
+ char *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ char *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ index_type type_size;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof (GFC_UINTEGER_4);
+ dim = GFC_DESCRIPTOR_RANK (array);
+ num_elems = 1;
+ ssize = type_size;
+ packed = true;
+ span = array->span != 0 ? array->span : type_size;
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (num_elems != GFC_DESCRIPTOR_STRIDE (array,n))
+ packed = false;
+
+ num_elems *= extent[n];
+ }
+
+ ssize = num_elems * type_size;
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * ssize;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ char *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+
+ memcpy (dest, src, type_size);
+ dest += type_size;
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ char *other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_UINTEGER_4 *a;
+ GFC_UINTEGER_4 *b;
+
+ other_shared_ptr = this_shared_ptr + imoffset * ssize;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = (GFC_UINTEGER_4 *) (this_shared_ptr + i * type_size);
+ b = (GFC_UINTEGER_4 *) (other_shared_ptr + i * type_size);
+ if (memcmp4 (b, a, char_len) > 0)
+ memcpy (a, b, type_size);
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ char *src = buffer;
+ char *restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ memcpy (dest, src, type_size);
+ src += span;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+void nca_collsub_min_array_s4 (gfc_array_s4 * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type char_len,
+ index_type errmsg_len);
+export_proto (nca_collsub_min_array_s4);
+
+void
+nca_collsub_min_array_s4 (gfc_array_s4 * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS]; /* stride is byte-based here. */
+ index_type extent[GFC_MAX_DIMENSIONS];
+ char *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ char *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ index_type type_size;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof (GFC_UINTEGER_4);
+ dim = GFC_DESCRIPTOR_RANK (array);
+ num_elems = 1;
+ ssize = type_size;
+ packed = true;
+ span = array->span != 0 ? array->span : type_size;
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (num_elems != GFC_DESCRIPTOR_STRIDE (array,n))
+ packed = false;
+
+ num_elems *= extent[n];
+ }
+
+ ssize = num_elems * type_size;
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * ssize;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ char *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+
+ memcpy (dest, src, type_size);
+ dest += type_size;
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ char *other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ GFC_UINTEGER_4 *a;
+ GFC_UINTEGER_4 *b;
+
+ other_shared_ptr = this_shared_ptr + imoffset * ssize;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = (GFC_UINTEGER_4 *) (this_shared_ptr + i * type_size);
+ b = (GFC_UINTEGER_4 *) (other_shared_ptr + i * type_size);
+ if (memcmp4 (b, a, char_len) < 0)
+ memcpy (a, b, type_size);
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ char *src = buffer;
+ char *restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ memcpy (dest, src, type_size);
+ src += span;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+
+#endif
+
diff --git a/libgfortran/libgfortran.h b/libgfortran/libgfortran.h
index 8c539e0898b..a6b0d5a476d 100644
--- a/libgfortran/libgfortran.h
+++ b/libgfortran/libgfortran.h
@@ -403,6 +403,7 @@ struct {\
}
typedef GFC_FULL_ARRAY_DESCRIPTOR (GFC_MAX_DIMENSIONS, GFC_INTEGER_4) gfc_full_array_i4;
+typedef GFC_FULL_ARRAY_DESCRIPTOR (GFC_MAX_DIMENSIONS, char) gfc_full_array_char;
#define GFC_DESCRIPTOR_RANK(desc) ((desc)->dtype.rank)
#define GFC_DESCRIPTOR_TYPE(desc) ((desc)->dtype.type)
diff --git a/libgfortran/m4/nca-minmax-s.m4 b/libgfortran/m4/nca-minmax-s.m4
new file mode 100644
index 00000000000..2d8891fe673
--- /dev/null
+++ b/libgfortran/m4/nca-minmax-s.m4
@@ -0,0 +1,289 @@
+dnl Support macro file for intrinsic functions.
+dnl Contains the generic sections of gfortran functions.
+dnl This file is part of the GNU Fortran Runtime Library (libgfortran)
+dnl Distributed under the GNU GPL with exception. See COPYING for details.
+dnl
+`/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */'
+
+include(iparm.m4)dnl
+define(`compare_fcn',`ifelse(rtype_kind,1,memcmp,memcmp4)')dnl
+define(SCALAR_FUNCTION,`void nca_collsub_'$1`_scalar_'rtype_code` ('rtype_name` *obj, int *result_image,
+ int *stat, char *errmsg, index_type char_len, index_type errmsg_len);
+export_proto(nca_collsub_'$1`_scalar_'rtype_code`);
+
+void
+nca_collsub_'$1`_scalar_'rtype_code` ('rtype_name` *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ 'rtype_name` *a, *b;
+ 'rtype_name` *buffer, *this_image_buf;
+ collsub_iface *ci;
+ index_type type_size;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof ('rtype_name`);
+ buffer = get_collsub_buf (ci, type_size * local->num_images);
+ this_image_buf = buffer + this_image.image_num * char_len;
+ memcpy (this_image_buf, obj, type_size);
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset * char_len;
+ if ('compare_fcn` (b, a, char_len) '$2` 0)
+ memcpy (a, b, type_size);
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ memcpy (obj, buffer, type_size);
+
+ finish_collective_subroutine (ci);
+
+}
+
+')dnl
+define(ARRAY_FUNCTION,dnl
+`void nca_collsub_'$1`_array_'rtype_code` ('rtype` * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type char_len,
+ index_type errmsg_len);
+export_proto (nca_collsub_'$1`_array_'rtype_code`);
+
+void
+nca_collsub_'$1`_array_'rtype_code` ('rtype` * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS]; /* stride is byte-based here. */
+ index_type extent[GFC_MAX_DIMENSIONS];
+ char *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ char *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ index_type type_size;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof ('rtype_name`);
+ dim = GFC_DESCRIPTOR_RANK (array);
+ num_elems = 1;
+ ssize = type_size;
+ packed = true;
+ span = array->span != 0 ? array->span : type_size;
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (num_elems != GFC_DESCRIPTOR_STRIDE (array,n))
+ packed = false;
+
+ num_elems *= extent[n];
+ }
+
+ ssize = num_elems * type_size;
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * ssize;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ char *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+
+ memcpy (dest, src, type_size);
+ dest += type_size;
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ char *other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ 'rtype_name` *a;
+ 'rtype_name` *b;
+
+ other_shared_ptr = this_shared_ptr + imoffset * ssize;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = ('rtype_name` *) (this_shared_ptr + i * type_size);
+ b = ('rtype_name` *) (other_shared_ptr + i * type_size);
+ if ('compare_fcn` (b, a, char_len) '$2` 0)
+ memcpy (a, b, type_size);
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ char *src = buffer;
+ char *restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ memcpy (dest, src, type_size);
+ src += span;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+')
+`
+#include "libgfortran.h"
+
+#if defined (HAVE_'rtype_name`)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+#if 'rtype_kind` == 4
+
+/* Compare wide character types, which are handled internally as
+ unsigned 4-byte integers. */
+static inline int
+memcmp4 (const void *a, const void *b, size_t len)
+{
+ const GFC_UINTEGER_4 *pa = a;
+ const GFC_UINTEGER_4 *pb = b;
+ while (len-- > 0)
+ {
+ if (*pa != *pb)
+ return *pa < *pb ? -1 : 1;
+ pa ++;
+ pb ++;
+ }
+ return 0;
+}
+
+#endif
+'SCALAR_FUNCTION(`max',`>')dnl
+SCALAR_FUNCTION(`min',`<')dnl
+ARRAY_FUNCTION(`max',`>')dnl
+ARRAY_FUNCTION(`min',`<')dnl
+`
+#endif
+'
diff --git a/libgfortran/m4/nca_minmax.m4 b/libgfortran/m4/nca_minmax.m4
new file mode 100644
index 00000000000..76070c102c0
--- /dev/null
+++ b/libgfortran/m4/nca_minmax.m4
@@ -0,0 +1,259 @@
+dnl Support macro file for intrinsic functions.
+dnl Contains the generic sections of gfortran functions.
+dnl This file is part of the GNU Fortran Runtime Library (libgfortran)
+dnl Distributed under the GNU GPL with exception. See COPYING for details.
+dnl
+`/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */'
+
+include(iparm.m4)dnl
+define(SCALAR_FUNCTION,`void nca_collsub_'$1`_scalar_'rtype_code` ('rtype_name` *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_'$1`_scalar_'rtype_code`);
+
+void
+nca_collsub_'$1`_scalar_'rtype_code` ('rtype_name` *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ 'rtype_name` *a, *b;
+ 'rtype_name` *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof('rtype_name`) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ '$2`
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+')dnl
+define(ARRAY_FUNCTION,dnl
+`void nca_collsub_'$1`_array_'rtype_code` ('rtype` * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_'$1`_array_'rtype_code`);
+
+void
+nca_collsub_'$1`_array_'rtype_code` ('rtype` * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ 'rtype_name` *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ 'rtype_name` *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof ('rtype_name`);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof ('rtype_name`);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof ('rtype_name`);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ 'rtype_name` *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *(('rtype_name` *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ 'rtype_name` * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ 'rtype_name` *a;
+ 'rtype_name` *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ '$2`
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ 'rtype_name` *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *(('rtype_name` * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+')
+`#include "libgfortran.h"
+
+#if defined (HAVE_'rtype_name`)'
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+SCALAR_FUNCTION(`max',`if (*b > *a)
+ *a = *b;')dnl
+SCALAR_FUNCTION(`min',`if (*b < *a)
+ *a = *b;')dnl
+SCALAR_FUNCTION(`sum',`*a += *b;')dnl
+ARRAY_FUNCTION(`max',`if (*b > *a)
+ *a = *b;')dnl
+ARRAY_FUNCTION(`min',`if (*b < *a)
+ *a = *b;')dnl
+ARRAY_FUNCTION(`sum',`*a += *b;')dnl
+`
+#endif
+'
diff --git a/libgfortran/nca/.tags b/libgfortran/nca/.tags
new file mode 100644
index 00000000000..07d260ddb9d
--- /dev/null
+++ b/libgfortran/nca/.tags
@@ -0,0 +1,275 @@
+!_TAG_FILE_FORMAT 2 /extended format; --format=1 will not append ;" to lines/
+!_TAG_FILE_SORTED 1 /0=unsorted, 1=sorted, 2=foldcase/
+!_TAG_PROGRAM_AUTHOR Darren Hiebert /dhiebert@users.sourceforge.net/
+!_TAG_PROGRAM_NAME Exuberant Ctags //
+!_TAG_PROGRAM_URL http://ctags.sourceforge.net /official site/
+!_TAG_PROGRAM_VERSION 5.9~svn20110310 //
+ALIGN_TO serialize.c 47;" d file:
+ALLOC_H alloc.h 26;" d
+COARRAY_LOCK_HDR lock.h 27;" d
+COARRAY_NATIVE_HDR libcoarraynative.h 30;" d
+COLLECTIVE_SUBROUTINE_HDR collective_subroutine.h 3;" d
+CRITICAL_LOOKAHEAD hashmap.c 30;" d file:
+DEBUG_PRINTF libcoarraynative.h 45;" d
+DEBUG_PRINTF libcoarraynative.h 47;" d
+FILL_VALUE serialize.c 57;" d file:
+GFC_NCA_EVENT_COARRAY wrapper.c /^ GFC_NCA_EVENT_COARRAY,$/;" e enum:gfc_coarray_allocation_type file:
+GFC_NCA_LOCK_COARRAY wrapper.c /^ GFC_NCA_LOCK_COARRAY,$/;" e enum:gfc_coarray_allocation_type file:
+GFC_NCA_NORMAL_COARRAY wrapper.c /^ GFC_NCA_NORMAL_COARRAY = 3,$/;" e enum:gfc_coarray_allocation_type file:
+GFORTRAN_ENV_NUM_IMAGES coarraynative.c 37;" d file:
+HASHMAP_H hashmap.h 69;" d
+IMAGE_FAILED libcoarraynative.h /^ IMAGE_FAILED$/;" e enum:__anon7
+IMAGE_FAILED master.c /^ IMAGE_FAILED$/;" e enum:__anon5 file:
+IMAGE_OK libcoarraynative.h /^ IMAGE_OK,$/;" e enum:__anon7
+IMAGE_OK master.c /^ IMAGE_OK,$/;" e enum:__anon5 file:
+IMAGE_UNKNOWN libcoarraynative.h /^ IMAGE_UNKNOWN = 0,$/;" e enum:__anon7
+IMAGE_UNKNOWN master.c /^ IMAGE_UNKNOWN = 0,$/;" e enum:__anon5 file:
+INDENT hashmap.c 403;" d file:
+INDENT hashmap.c 414;" d file:
+INDENT hashmap.c 415;" d file:
+INDENT hashmap.c 429;" d file:
+INDENT hashmap.c 430;" d file:
+INITIAL_BITNUM hashmap.c 28;" d file:
+INITIAL_SIZE hashmap.c 29;" d file:
+IPSYNC_HDR sync.h 26;" d
+ITER malloc_test.c 13;" d file:
+MAX_ALIGN allocator.c 50;" d file:
+MAX_NUM malloc_test.c 10;" d file:
+MEMOBJ_NAME util.c 11;" d file:
+MIN_NUM malloc_test.c 11;" d file:
+NUM_BITS malloc_test.c 8;" d file:
+NUM_SIZES malloc_test.c 9;" d file:
+PE hashmap.c 402;" d file:
+PTR_BITS util.h 32;" d
+SHARED_ALLOCATOR_HDR allocator.h 2;" d
+SHARED_MEMORY_H shared_memory.h 77;" d
+SHARED_MEMORY_RAW_ALLOC shared_memory.h 50;" d
+SHARED_MEMORY_RAW_ALLOC_PTR shared_memory.h 53;" d
+SHMPTR_AS shared_memory.h 46;" d
+SHMPTR_DEREF shared_memory.h 44;" d
+SHMPTR_EQUALS shared_memory.h 48;" d
+SHMPTR_IS_NULL shared_memory.h 42;" d
+SHMPTR_NULL shared_memory.h 41;" d
+SHMPTR_SET shared_memory.h 47;" d
+SZ malloc_test.c 12;" d file:
+UTIL_HDR util.h 26;" d
+a hashmap.h /^ allocator *a;$/;" m struct:hashmap
+a sync.h /^ allocator *a;$/;" m struct:__anon3
+ai libcoarraynative.h /^ alloc_iface ai;$/;" m struct:__anon11
+alignto util.c /^alignto(size_t size, size_t align) {$/;" f
+alloc alloc.h /^ allocator alloc;$/;" m struct:alloc_iface
+alloc_iface alloc.h /^typedef struct alloc_iface$/;" s
+alloc_iface alloc.h /^} alloc_iface;$/;" t typeref:struct:alloc_iface
+alloc_iface_init alloc.c /^alloc_iface_init (alloc_iface *iface, shared_memory *mem)$/;" f
+alloc_iface_init alloc.h /^internal_proto (alloc_iface_init);$/;" v
+alloc_iface_shared alloc.h /^typedef struct alloc_iface_shared$/;" s
+alloc_iface_shared alloc.h /^} alloc_iface_shared;$/;" t typeref:struct:alloc_iface_shared
+allocator allocator.h /^} allocator;$/;" t typeref:struct:__anon17
+allocator_init allocator.c /^allocator_init (allocator *a, allocator_shared *s, shared_memory *sm)$/;" f
+allocator_s alloc.h /^ allocator_shared allocator_s;$/;" m struct:alloc_iface_shared
+allocator_shared allocator.h /^} allocator_shared;$/;" t typeref:struct:__anon16
+allocs shared_memory.c /^ } allocs[];$/;" m struct:shared_memory_act typeref:struct:shared_memory_act::local_alloc file:
+arr lock.h /^ pthread_mutex_t arr[];$/;" m struct:__anon12
+as alloc.h /^ alloc_iface_shared *as;$/;" m struct:alloc_iface
+barrier collective_subroutine.h /^ pthread_barrier_t barrier;$/;" m struct:collsub_iface_shared
+barrier libcoarraynative.h /^ pthread_barrier_t barrier;$/;" m struct:__anon6
+base shared_memory.c /^ void *base;$/;" m struct:shared_memory_act::local_alloc file:
+bitnum hashmap.h /^ int bitnum;$/;" m struct:__anon14
+bucket allocator.c /^} bucket;$/;" t typeref:struct:__anon1 file:
+ci libcoarraynative.h /^ collsub_iface ci;$/;" m struct:__anon11
+cis sync.h /^ sync_iface_shared *cis;$/;" m struct:__anon3
+collsub_broadcast collective_subroutine.h /^internal_proto (collsub_broadcast);$/;" v
+collsub_buf collective_subroutine.c /^void *collsub_buf = NULL;$/;" v
+collsub_buf collective_subroutine.h /^ shared_mem_ptr collsub_buf;$/;" m struct:collsub_iface_shared
+collsub_buf collective_subroutine.h /^ void *collsub_buf; \/* Cached pointer to shared collsub_buf. *\/$/;" m struct:collsub_iface
+collsub_buf collective_subroutine.h /^internal_proto (collsub_buf);$/;" v
+collsub_iface collective_subroutine.h /^typedef struct collsub_iface$/;" s
+collsub_iface collective_subroutine.h /^} collsub_iface;$/;" t typeref:struct:collsub_iface
+collsub_iface_shared collective_subroutine.h /^typedef struct collsub_iface_shared $/;" s
+collsub_iface_shared collective_subroutine.h /^} collsub_iface_shared;$/;" t typeref:struct:collsub_iface_shared
+collsub_reduce collective_subroutine.c /^collsub_reduce (void *obj, size_t nobjs, int *result_image, size_t size, $/;" f
+collsub_reduce collective_subroutine.h /^internal_proto (collsub_reduce);$/;" v
+collsub_sync collective_subroutine.c /^collsub_sync (void) {$/;" f
+collsub_sync collective_subroutine.h /^internal_proto (collsub_sync);$/;" v
+copy_from collective_inline.h /^copy_from (int image) $/;" f
+copy_in collective_inline.h /^copy_in (void *obj) {$/;" f
+copy_out collective_inline.h /^copy_out (void *obj, int image)$/;" f
+copy_to collective_inline.h /^copy_to (void *obj, int image)$/;" f
+curr_size collective_subroutine.c /^size_t curr_size = 0;$/;" v
+curr_size collective_subroutine.h /^ size_t curr_size;$/;" m struct:collsub_iface_shared
+curr_size collective_subroutine.h /^internal_proto (curr_size);$/;" v
+data hashmap.h /^ shared_mem_ptr data;$/;" m struct:__anon14
+deserialize_memory serialize.c /^export_proto (deserialize_memory);$/;" v
+div_ru wrapper.c /^div_ru (int divident, int divisor)$/;" f file:
+dump_hm hashmap.c /^dump_hm(hashmap *hm) {$/;" f
+enlarge_hashmap_mem hashmap.c /^enlarge_hashmap_mem (hashmap *hm, hashmap_entry **data, bool f)$/;" f file:
+ensure_initialization coarraynative.c /^ensure_initialization(void) {$/;" f
+fd shared_memory.c /^ int fd;$/;" m struct:__anon18 file:
+finish_collective_subroutine collective_inline.h /^finish_collective_subroutine (void) $/;" f
+free_bucket_head allocator.h /^ shared_mem_ptr free_bucket_head[PTR_BITS];$/;" m struct:__anon16
+free_memory_with_id alloc.c /^free_memory_with_id (alloc_iface* iface, memid id)$/;" f
+free_memory_with_id alloc.h /^internal_proto (free_memory_with_id);$/;" v
+gen_mask hashmap.c /^gen_mask (hashmap *hm)$/;" f file:
+get_allocator alloc.c /^get_allocator (alloc_iface * iface)$/;" f
+get_allocator alloc.h /^internal_proto (get_allocator);$/;" v
+get_data hashmap.c /^get_data(hashmap *hm)$/;" f file:
+get_environ_image_num coarraynative.c /^get_environ_image_num (void)$/;" f file:
+get_expected_offset hashmap.c /^get_expected_offset (hashmap *hm, memid id)$/;" f file:
+get_locked_table sync.c /^get_locked_table(sync_iface *si) { \/\/ The initialization of the table has to $/;" f file:
+get_master coarraynative.c /^get_master (void) {$/;" f file:
+get_memory_by_id alloc.c /^get_memory_by_id (alloc_iface *iface, size_t size, memid id)$/;" f
+get_memory_by_id alloc.h /^internal_proto (get_memory_by_id);$/;" v
+get_obj_ptr collective_inline.h /^get_obj_ptr (int image) $/;" f
+get_shared_memory_act_size shared_memory.c /^get_shared_memory_act_size (int nallocs)$/;" f file:
+get_shmem_fd util.c /^get_shmem_fd (void)$/;" f
+gfc_coarray_allocation_type wrapper.c /^enum gfc_coarray_allocation_type {$/;" g file:
+global_shared_memory_meta shared_memory.c /^} global_shared_memory_meta;$/;" t typeref:struct:__anon18 file:
+has_failed_image libcoarraynative.h /^ int has_failed_image;$/;" m struct:__anon9
+has_failed_image master.c /^ int has_failed_image = 0;$/;" m struct:__anon4 file:
+hash hashmap.c /^hash (uint64_t key)$/;" f file:
+hashmap hashmap.h /^typedef struct hashmap$/;" s
+hashmap hashmap.h /^} hashmap;$/;" t typeref:struct:hashmap
+hashmap_change_refcnt hashmap.c /^hashmap_change_refcnt (hashmap *hm, memid id, hashmap_search_result *res,$/;" f file:
+hashmap_dec hashmap.c /^hashmap_dec (hashmap *hm, memid id, hashmap_search_result * res)$/;" f
+hashmap_entry hashmap.c /^} hashmap_entry;$/;" t typeref:struct:__anon13 file:
+hashmap_get hashmap.c /^hashmap_get (hashmap *hm, memid id)$/;" f
+hashmap_inc hashmap.c /^hashmap_inc (hashmap *hm, memid id, hashmap_search_result * res)$/;" f
+hashmap_init hashmap.c /^hashmap_init (hashmap *hm, hashmap_shared *hs, allocator *a,$/;" f
+hashmap_search_result hashmap.h /^} hashmap_search_result;$/;" t typeref:struct:__anon15
+hashmap_set hashmap.c /^hashmap_set (hashmap *hm, memid id, hashmap_search_result *hsr,$/;" f
+hashmap_shared hashmap.h /^} hashmap_shared;$/;" t typeref:struct:__anon14
+header shared_memory.c /^ void *header;$/;" m struct:shared_memory_act file:
+hm alloc.h /^ hashmap hm;$/;" m struct:alloc_iface
+hm_search_result_contains hashmap.c /^hm_search_result_contains (hashmap_search_result *res)$/;" f
+hm_search_result_ptr hashmap.c /^hm_search_result_ptr (hashmap_search_result *res)$/;" f
+hm_search_result_size hashmap.c /^hm_search_result_size (hashmap_search_result *res)$/;" f
+hmiadd hashmap.c /^hmiadd (hashmap *hm, size_t s, ssize_t o) {$/;" f file:
+hms alloc.h /^ hashmap_shared hms;$/;" m struct:alloc_iface_shared
+id hashmap.c /^ memid id;$/;" m struct:__anon13 file:
+image libcoarraynative.h /^} image;$/;" t typeref:struct:__anon10
+image_main_wrapper master.c /^image_main_wrapper (void (*image_main) (void), int this_image_num)$/;" f file:
+image_num libcoarraynative.h /^ int image_num;$/;" m struct:__anon10
+image_status libcoarraynative.h /^} image_status;$/;" t typeref:enum:__anon7
+image_tracker libcoarraynative.h /^} image_tracker;$/;" t typeref:struct:__anon8
+images libcoarraynative.h /^ image_tracker images[];$/;" m struct:__anon9
+images master.c /^ struct image_status * images;$/;" m struct:__anon4 typeref:struct:__anon4::image_status file:
+init_collsub collective_subroutine.c /^init_collsub (void) {$/;" f
+init_collsub collective_subroutine.h /^internal_proto (init_collsub);$/;" v
+initialize_shared_mutex util.c /^initialize_shared_mutex (pthread_mutex_t *mutex)$/;" f
+initialized lock.h /^ int initialized;$/;" m struct:__anon12
+ipcollsub libcoarraynative.h /^} ipcollsub;$/;" t typeref:struct:__anon6
+last_base shared_memory.c /^last_base (shared_memory_act *mem)$/;" f file:
+last_seen_size shared_memory.c /^ size_t last_seen_size;$/;" m struct:shared_memory_act file:
+local coarraynative.c /^nca_local_data *local = NULL;$/;" v
+local_alloc shared_memory.c /^ struct local_alloc {$/;" s struct:shared_memory_act file:
+lock alloc.h /^ pthread_mutex_t lock;$/;" m struct:alloc_iface_shared
+lock_array lock.h /^} lock_array;$/;" t typeref:struct:__anon12
+lock_table sync.c /^lock_table (sync_iface *si)$/;" f file:
+m libcoarraynative.h /^ master *m;$/;" m struct:__anon10
+main malloc_test.c /^int main()$/;" f
+map_memory shared_memory.c /^map_memory (int fd, size_t size, off_t offset)$/;" f file:
+master libcoarraynative.h /^} master;$/;" t typeref:struct:__anon9
+master master.c /^} master;$/;" t typeref:struct:__anon4 file:
+max_lookahead hashmap.c /^ int max_lookahead; $/;" m struct:__anon13 file:
+maximg libcoarraynative.h /^ int maximg;$/;" m struct:__anon6
+mem alloc.h /^ shared_memory *mem;$/;" m struct:alloc_iface
+memid hashmap.h /^typedef intptr_t memid;$/;" t
+meta shared_memory.c /^ global_shared_memory_meta *meta;$/;" m struct:shared_memory_act file:
+n_ent hashmap.c /^static ssize_t n_ent;$/;" v file:
+nca_co_broadcast collective_subroutine.c /^export_proto (nca_co_broadcast);$/;" v
+nca_co_broadcast collective_subroutine.c /^nca_co_broadcast (gfc_array_char * restrict a, int source_image,$/;" f
+nca_coarray_alloc wrapper.c /^export_proto (nca_coarray_alloc);$/;" v
+nca_coarray_alloc wrapper.c /^nca_coarray_alloc (gfc_array_void *desc, int elem_size, int corank,$/;" f
+nca_coarray_free wrapper.c /^export_proto (nca_coarray_free);$/;" v
+nca_coarray_free wrapper.c /^nca_coarray_free (gfc_array_void *desc, int alloc_type)$/;" f
+nca_coarray_num_images wrapper.c /^export_proto (nca_coarray_num_images);$/;" v
+nca_coarray_num_images wrapper.c /^nca_coarray_num_images (int distance __attribute__((unused)))$/;" f
+nca_coarray_sync_all wrapper.c /^export_proto (nca_coarray_sync_all);$/;" v
+nca_coarray_sync_all wrapper.c /^nca_coarray_sync_all (int *stat __attribute__((unused)))$/;" f
+nca_coarray_this_image wrapper.c /^export_proto (nca_coarray_this_image);$/;" v
+nca_coarray_this_image wrapper.c /^nca_coarray_this_image (int distance __attribute__((unused)))$/;" f
+nca_collsub_reduce_array wrapper.c /^export_proto (nca_collsub_reduce_array);$/;" v
+nca_collsub_reduce_array wrapper.c /^nca_collsub_reduce_array (gfc_array_void *desc, void (*assign_function) (void *, void *),$/;" f
+nca_collsub_reduce_scalar wrapper.c /^export_proto (nca_collsub_reduce_scalar);$/;" v
+nca_collsub_reduce_scalar wrapper.c /^nca_collsub_reduce_scalar (void *obj, index_type elem_size,$/;" f
+nca_local_data libcoarraynative.h /^} nca_local_data;$/;" t typeref:struct:__anon11
+nca_lock wrapper.c /^export_proto (nca_lock);$/;" v
+nca_lock wrapper.c /^nca_lock (void *lock)$/;" f
+nca_master coarraynative.c /^nca_master (void (*image_main) (void)) {$/;" f
+nca_master master.c /^nca_master (void (*image_main) (void)) {$/;" f
+nca_master master.c /^nca_master (void (*image_main) (void))$/;" f
+nca_sync_images wrapper.c /^export_proto (nca_sync_images);$/;" v
+nca_sync_images wrapper.c /^nca_sync_images (size_t s, int *images,$/;" f
+nca_unlock wrapper.c /^export_proto (nca_unlock);$/;" v
+nca_unlock wrapper.c /^nca_unlock (void *lock)$/;" f
+new_base_mapping shared_memory.c /^new_base_mapping (shared_memory_act *mem)$/;" f file:
+next allocator.c /^ shared_mem_ptr next;$/;" m struct:__anon1 file:
+next_power_of_two util.c /^next_power_of_two(size_t size) {$/;" f
+num_entries hashmap.c /^num_entries (hashmap_entry *data, size_t size)$/;" f file:
+num_images libcoarraynative.h /^ int num_images;$/;" m struct:__anon11
+num_local_allocs shared_memory.c /^ size_t num_local_allocs;$/;" m struct:shared_memory_act file:
+offset shared_memory.h /^ ssize_t offset;$/;" m struct:shared_mem_ptr
+owner lock.h /^ int owner;$/;" m struct:__anon12
+p hashmap.c /^ shared_mem_ptr p; \/* If p == SHMPTR_NULL, the entry is empty. *\/$/;" m struct:__anon13 file:
+p hashmap.h /^ shared_mem_ptr p;$/;" m struct:__anon15
+pagesize util.c /^size_t pagesize = 1<<17;$/;" v
+pid libcoarraynative.h /^ pid_t pid;$/;" m struct:__anon8
+prepare_collective_subroutine collective_subroutine.c /^prepare_collective_subroutine (size_t size)$/;" f
+prepare_collective_subroutine collective_subroutine.h /^internal_proto (prepare_collective_subroutine);$/;" v
+refcnt hashmap.c /^ int refcnt;$/;" m struct:__anon13 file:
+res_offset hashmap.h /^ ssize_t res_offset;$/;" m struct:__anon15
+resize_hm hashmap.c /^resize_hm (hashmap *hm, hashmap_entry **data)$/;" f file:
+round_to_pagesize util.c /^round_to_pagesize(size_t s) {$/;" f
+s allocator.h /^ allocator_shared *s;$/;" m struct:__anon17
+s collective_subroutine.h /^ collsub_iface_shared *s;$/;" m struct:collsub_iface
+s hashmap.c /^ size_t s;$/;" m struct:__anon13 file:
+s hashmap.h /^ hashmap_shared *s;$/;" m struct:hashmap
+scan_empty hashmap.c /^scan_empty (hashmap *hm, ssize_t expected_off, memid id)$/;" f file:
+scan_inside_lookahead hashmap.c /^scan_inside_lookahead (hashmap *hm, ssize_t expected_off, memid id)$/;" f file:
+serialize_memory serialize.c /^export_proto (serialize_memory);$/;" v
+serialize_memory serialize.c /^serialize_memory (gfc_array_char * const restrict source, char *d)$/;" f
+shared_free allocator.c /^shared_free (allocator *a, shared_mem_ptr p, size_t size) {$/;" f
+shared_malloc allocator.c /^shared_malloc (allocator *a, size_t size)$/;" f
+shared_mem_ptr shared_memory.h /^typedef struct shared_mem_ptr$/;" s
+shared_mem_ptr shared_memory.h /^} shared_mem_ptr;$/;" t typeref:struct:shared_mem_ptr
+shared_mem_ptr_to_void_ptr shared_memory.c /^shared_mem_ptr_to_void_ptr(shared_memory_act **pmem, shared_mem_ptr smp)$/;" f
+shared_mem_ptr_to_void_ptr shared_memory.h /^internal_proto (shared_mem_ptr_to_void_ptr);$/;" v
+shared_memory shared_memory.h /^typedef struct shared_memory_act * shared_memory;$/;" t typeref:struct:shared_memory_act
+shared_memory_act shared_memory.c /^typedef struct shared_memory_act$/;" s file:
+shared_memory_act shared_memory.c /^} shared_memory_act;$/;" t typeref:struct:shared_memory_act file:
+shared_memory_get_mem_with_alignment shared_memory.c /^shared_memory_get_mem_with_alignment (shared_memory_act **pmem, size_t size,$/;" f
+shared_memory_get_mem_with_alignment shared_memory.h /^internal_proto (shared_memory_get_mem_with_alignment);$/;" v
+shared_memory_init shared_memory.c /^shared_memory_init (shared_memory_act **pmem)$/;" f
+shared_memory_init shared_memory.h /^internal_proto (shared_memory_init);$/;" v
+shared_memory_prepare shared_memory.c /^shared_memory_prepare (shared_memory_act **pmem)$/;" f
+shared_memory_prepare shared_memory.h /^internal_proto (shared_memory_prepare);$/;" v
+shm allocator.h /^ shared_memory *shm;$/;" m struct:__anon17
+si libcoarraynative.h /^ sync_iface si;$/;" m struct:__anon11
+size hashmap.h /^ size_t size;$/;" m struct:__anon14
+size hashmap.h /^ size_t size;$/;" m struct:__anon15
+size shared_memory.c /^ size_t size;$/;" m struct:shared_memory_act::local_alloc file:
+size shared_memory.c /^ size_t size;$/;" m struct:__anon18 file:
+sm collective_subroutine.h /^ shared_memory *sm;$/;" m struct:collsub_iface
+sm hashmap.h /^ shared_memory *sm;$/;" m struct:hashmap
+sm libcoarraynative.h /^ shared_memory sm;$/;" m struct:__anon11
+sm sync.h /^ shared_memory *sm;$/;" m struct:__anon3
+status libcoarraynative.h /^ image_status status;$/;" m struct:__anon8
+sync_all sync.c /^sync_all (sync_iface *si)$/;" f
+sync_all sync.h /^ pthread_barrier_t sync_all;$/;" m struct:__anon2
+sync_all sync.h /^internal_proto (sync_all);$/;" v
+sync_all_init sync.c /^sync_all_init (pthread_barrier_t *b)$/;" f file:
+sync_iface sync.h /^} sync_iface;$/;" t typeref:struct:__anon3
+sync_iface_init sync.c /^sync_iface_init (sync_iface *si, alloc_iface *ai, shared_memory *sm)$/;" f
+sync_iface_init sync.h /^internal_proto (sync_iface_init);$/;" v
+sync_iface_shared sync.h /^} sync_iface_shared;$/;" t typeref:struct:__anon2
+sync_table sync.c /^sync_table (sync_iface *si, int *images, size_t size)$/;" f
+sync_table sync.h /^internal_proto (sync_table);$/;" v
+table sync.h /^ int *table; \/\/ we can cache the table and the trigger pointers here$/;" m struct:__anon3
+table sync.h /^ shared_mem_ptr table;$/;" m struct:__anon2
+table_lock sync.h /^ pthread_mutex_t table_lock;$/;" m struct:__anon2
+this_image coarraynative.c /^image this_image;$/;" v
+triggers sync.h /^ pthread_cond_t *triggers;$/;" m struct:__anon3
+triggers sync.h /^ shared_mem_ptr triggers;$/;" m struct:__anon2
+unlock_table sync.c /^unlock_table (sync_iface *si)$/;" f file:
+used shared_memory.c /^ size_t used;$/;" m struct:__anon18 file:
+wait_table_cond sync.c /^wait_table_cond (sync_iface *si, pthread_cond_t *cond)$/;" f file:
diff --git a/libgfortran/nca/alloc.c b/libgfortran/nca/alloc.c
new file mode 100644
index 00000000000..174fe330c32
--- /dev/null
+++ b/libgfortran/nca/alloc.c
@@ -0,0 +1,152 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+/* This provides the coarray-specific features (like IDs etc) for
+ allocator.c, in turn calling routines from shared_memory.c.
+*/
+
+#include "libgfortran.h"
+#include "shared_memory.h"
+#include "allocator.h"
+#include "hashmap.h"
+#include "alloc.h"
+
+#include <string.h>
+
+/* Return a local pointer into a shared memory object identified by
+ id. If the object is already found, it has been allocated before,
+ so just increase the reference counter.
+
+ The pointers returned by this function remain valid even if the
+ size of the memory allocation changes (see shared_memory.c). */
+
+static void *
+get_memory_by_id_internal (alloc_iface *iface, size_t size, memid id,
+ bool zero_mem)
+{
+ hashmap_search_result res;
+ shared_mem_ptr shared_ptr;
+ void *ret;
+
+ pthread_mutex_lock (&iface->as->lock);
+ shared_memory_prepare(iface->mem);
+
+ res = hashmap_get (&iface->hm, id);
+
+ if (hm_search_result_contains (&res))
+ {
+ size_t found_size;
+ found_size = hm_search_result_size (&res);
+ if (found_size != size)
+ {
+ dprintf (2, "Size mismatch for coarray allocation id %p: "
+ "found = %lu != size = %lu\n", (void *) id, found_size, size);
+ pthread_mutex_unlock (&iface->as->lock);
+ exit (1);
+ }
+ shared_ptr = hm_search_result_ptr (&res);
+ hashmap_inc (&iface->hm, id, &res);
+ }
+ else
+ {
+ shared_ptr = shared_malloc (&iface->alloc, size);
+ hashmap_set (&iface->hm, id, NULL, shared_ptr, size);
+ }
+
+ ret = SHMPTR_AS (void *, shared_ptr, iface->mem);
+ if (zero_mem)
+ memset (ret, '\0', size);
+
+ pthread_mutex_unlock (&iface->as->lock);
+ return ret;
+}
+
+void *
+get_memory_by_id (alloc_iface *iface, size_t size, memid id)
+{
+ return get_memory_by_id_internal (iface, size, id, 0);
+}
+
+void *
+get_memory_by_id_zero (alloc_iface *iface, size_t size, memid id)
+{
+ return get_memory_by_id_internal (iface, size, id, 1);
+}
+
+/* Free memory with id. Free it if this is the last image which
+ holds that memory segment, decrease the reference count otherwise. */
+
+void
+free_memory_with_id (alloc_iface* iface, memid id)
+{
+ hashmap_search_result res;
+ int entries_left;
+
+ pthread_mutex_lock (&iface->as->lock);
+ shared_memory_prepare(iface->mem);
+
+ res = hashmap_get (&iface->hm, id);
+ if (!hm_search_result_contains (&res))
+ {
+ pthread_mutex_unlock (&iface->as->lock);
+ char buffer[100];
+ snprintf (buffer, sizeof(buffer), "Error in free_memory_with_id: "
+ "%p not found", (void *) id);
+ dprintf (2, buffer);
+ // internal_error (NULL, buffer);
+ exit (1);
+ }
+
+ entries_left = hashmap_dec (&iface->hm, id, &res);
+ assert (entries_left >=0);
+
+ if (entries_left == 0)
+ {
+ shared_free (&iface->alloc, hm_search_result_ptr (&res),
+ hm_search_result_size (&res));
+ }
+
+ pthread_mutex_unlock (&iface->as->lock);
+ return;
+}
+
+/* Allocate the shared memory interface. This is called before we have
+ multiple images. */
+
+void
+alloc_iface_init (alloc_iface *iface, shared_memory *mem)
+{
+
+ iface->as = SHARED_MEMORY_RAW_ALLOC_PTR (mem, alloc_iface_shared);
+ iface->mem = mem;
+ initialize_shared_mutex (&iface->as->lock);
+ allocator_init (&iface->alloc, &iface->as->allocator_s, mem);
+ hashmap_init (&iface->hm, &iface->as->hms, &iface->alloc, mem);
+}
+
+allocator *
+get_allocator (alloc_iface * iface)
+{
+ return &iface->alloc;
+}
diff --git a/libgfortran/nca/alloc.h b/libgfortran/nca/alloc.h
new file mode 100644
index 00000000000..f65121c25cd
--- /dev/null
+++ b/libgfortran/nca/alloc.h
@@ -0,0 +1,67 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#ifndef ALLOC_H
+#define ALLOC_H
+
+#include "allocator.h"
+#include "hashmap.h"
+
+/* High-level interface for shared memory allocation. */
+
+/* This part of the alloc interface goes into shared memory. */
+
+typedef struct alloc_iface_shared
+{
+ allocator_shared allocator_s;
+ hashmap_shared hms;
+ pthread_mutex_t lock;
+} alloc_iface_shared;
+
+/* This is the local part. */
+
+typedef struct alloc_iface
+{
+ alloc_iface_shared *as;
+ shared_memory *mem;
+ allocator alloc;
+ hashmap hm;
+} alloc_iface;
+
+void *get_memory_by_id (alloc_iface *, size_t, memid);
+internal_proto (get_memory_by_id);
+
+void *get_memory_by_id_zero (alloc_iface *, size_t, memid);
+internal_proto (get_memory_by_id_zero);
+
+void free_memory_with_id (alloc_iface *, memid);
+internal_proto (free_memory_with_id);
+
+void alloc_iface_init (alloc_iface *, shared_memory *);
+internal_proto (alloc_iface_init);
+
+allocator *get_allocator (alloc_iface *);
+internal_proto (get_allocator);
+
+#endif
diff --git a/libgfortran/nca/allocator.c b/libgfortran/nca/allocator.c
new file mode 100644
index 00000000000..e7aa9fdefd0
--- /dev/null
+++ b/libgfortran/nca/allocator.c
@@ -0,0 +1,90 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+/* A malloc() - and free() - like interface, but for shared memory
+ pointers, except that we pass the size to free as well. */
+
+#include "libgfortran.h"
+#include "shared_memory.h"
+#include "allocator.h"
+
+typedef struct {
+ shared_mem_ptr next;
+} bucket;
+
+/* Initialize the allocator. */
+
+void
+allocator_init (allocator *a, allocator_shared *s, shared_memory *sm)
+{
+ a->s = s;
+ a->shm = sm;
+ for (int i = 0; i < PTR_BITS; i++)
+ s->free_bucket_head[i] = SHMPTR_NULL;
+}
+
+/* Main allocation routine, works like malloc. Round up allocations
+ to the next power of two and keep free lists in buckets. */
+
+#define MAX_ALIGN 16
+
+shared_mem_ptr
+shared_malloc (allocator *a, size_t size)
+{
+ shared_mem_ptr ret;
+ size_t sz;
+ size_t act_size;
+ int bucket_list_index;
+
+ sz = next_power_of_two (size);
+ act_size = sz > sizeof (bucket) ? sz : sizeof (bucket);
+ bucket_list_index = __builtin_clzl(act_size);
+
+ if (SHMPTR_IS_NULL (a->s->free_bucket_head[bucket_list_index]))
+ return shared_memory_get_mem_with_alignment (a->shm, act_size, MAX_ALIGN);
+
+ ret = a->s->free_bucket_head[bucket_list_index];
+ a->s->free_bucket_head[bucket_list_index]
+ = (SHMPTR_AS (bucket *, ret, a->shm)->next);
+ assert(ret.offset != 0);
+ return ret;
+}
+
+/* Free memory. */
+
+void
+shared_free (allocator *a, shared_mem_ptr p, size_t size) {
+ bucket *b;
+ size_t sz;
+ int bucket_list_index;
+ size_t act_size;
+
+ sz = next_power_of_two (size);
+ act_size = sz > sizeof (bucket) ? sz : sizeof (bucket);
+ bucket_list_index = __builtin_clzl(act_size);
+
+ b = SHMPTR_AS(bucket *, p, a->shm);
+ b->next = a->s->free_bucket_head[bucket_list_index];
+ a->s->free_bucket_head[bucket_list_index] = p;
+}
diff --git a/libgfortran/nca/allocator.h b/libgfortran/nca/allocator.h
new file mode 100644
index 00000000000..306022a5f39
--- /dev/null
+++ b/libgfortran/nca/allocator.h
@@ -0,0 +1,21 @@
+#ifndef SHARED_ALLOCATOR_HDR
+#define SHARED_ALLOCATOR_HDR
+
+#include "util.h"
+#include "shared_memory.h"
+
+typedef struct {
+ shared_mem_ptr free_bucket_head[PTR_BITS];
+} allocator_shared;
+
+typedef struct {
+ allocator_shared *s;
+ shared_memory *shm;
+} allocator;
+
+void allocator_init (allocator *, allocator_shared *, shared_memory *);
+
+shared_mem_ptr shared_malloc (allocator *, size_t size);
+void shared_free (allocator *, shared_mem_ptr, size_t size);
+
+#endif
diff --git a/libgfortran/nca/coarraynative.c b/libgfortran/nca/coarraynative.c
new file mode 100644
index 00000000000..c9d13ee92ac
--- /dev/null
+++ b/libgfortran/nca/coarraynative.c
@@ -0,0 +1,145 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+#include "libcoarraynative.h"
+#include "allocator.h"
+#include "hashmap.h"
+#include "util.h"
+#include "lock.h"
+#include "collective_subroutine.h"
+
+#include <unistd.h>
+#include <sys/mman.h>
+// #include <stdlib.h>
+#include <sys/wait.h>
+
+#define GFORTRAN_ENV_NUM_IMAGES "GFORTRAN_NUM_IMAGES"
+
+nca_local_data *local = NULL;
+
+image this_image;
+
+static int
+get_environ_image_num (void)
+{
+ char *num_images_char;
+ int nimages;
+ num_images_char = getenv (GFORTRAN_ENV_NUM_IMAGES);
+ if (!num_images_char)
+ return sysconf (_SC_NPROCESSORS_ONLN); /* TODO: Make portable. */
+ /* TODO: Error checking. */
+ nimages = atoi (num_images_char);
+ return nimages;
+}
+
+void
+ensure_initialization(void)
+{
+ if (local)
+ return;
+
+ local = malloc(sizeof(nca_local_data)); // Is malloc already init'ed at that
+ // point? Maybe use mmap(MAP_ANON)
+ // instead
+ pagesize = sysconf (_SC_PAGE_SIZE);
+ local->num_images = get_environ_image_num ();
+ shared_memory_init (&local->sm);
+ shared_memory_prepare (&local->sm);
+ alloc_iface_init (&local->ai, &local->sm);
+ collsub_iface_init (&local->ci, &local->ai, &local->sm);
+ sync_iface_init (&local->si, &local->ai, &local->sm);
+}
+
+static void __attribute__((noreturn))
+image_main_wrapper (void (*image_main) (void), image *this)
+{
+ this_image = *this;
+
+ sync_all(&local->si);
+
+ image_main ();
+
+ exit (0);
+}
+
+static master *
+get_master (void) {
+ master *m;
+ m = SHMPTR_AS (master *,
+ shared_memory_get_mem_with_alignment
+ (&local->sm,
+ sizeof (master) + sizeof(image_status) * local->num_images,
+ __alignof__(master)), &local->sm);
+ m->has_failed_image = 0;
+ return m;
+}
+
+/* This is called from main, with a pointer to the user's program as
+ argument. It forks the images and waits for their completion. */
+
+void
+nca_master (void (*image_main) (void)) {
+ master *m;
+ int i, j;
+ pid_t new;
+ image im;
+ int exit_code = 0;
+ int chstatus;
+ ensure_initialization();
+ m = get_master();
+
+ im.m = m;
+
+ for (im.image_num = 0; im.image_num < local->num_images; im.image_num++)
+ {
+ if ((new = fork()))
+ {
+ if (new == -1)
+ {
+ dprintf(2, "error spawning child\n");
+ exit_code = 1;
+ }
+ m->images[im.image_num].pid = new;
+ m->images[im.image_num].status = IMAGE_OK;
+ }
+ else
+ image_main_wrapper(image_main, &im);
+ }
+ for (i = 0; i < local->num_images; i++)
+ {
+ new = wait (&chstatus);
+ if (!WIFEXITED (chstatus) || WEXITSTATUS (chstatus))
+ {
+ j = 0;
+ for (; j < local->num_images && m->images[j].pid != new; j++);
+ m->images[j].status = IMAGE_FAILED;
+ m->has_failed_image++; //FIXME: Needs to be atomic, probably
+ dprintf (2, "ERROR: Image %d(%#x) failed\n", j, new);
+ exit_code = 1;
+ }
+ }
+ exit (exit_code);
+}
diff --git a/libgfortran/nca/collective_inline.h b/libgfortran/nca/collective_inline.h
new file mode 100644
index 00000000000..4e7107b359d
--- /dev/null
+++ b/libgfortran/nca/collective_inline.h
@@ -0,0 +1,42 @@
+#include "collective_subroutine.h"
+
+static inline void
+finish_collective_subroutine (collsub_iface *ci)
+{
+ collsub_sync (ci);
+}
+
+#if 0
+static inline void *
+get_obj_ptr (void *buffer, int image)
+{
+ return (char *) + curr_size * image;
+}
+
+/* If obj is NULL, copy the object from the entry in this image. */
+static inline void
+copy_to (void *buffer, void *obj, int image)
+{
+ if (obj == 0)
+ obj = get_obj_ptr (this_image.image_num);
+ memcpy (get_obj_ptr (image), obj, curr_size);
+}
+
+static inline void
+copy_out (void *buffer, void *obj, int image)
+{
+ memcpy (obj, get_obj_ptr (image), curr_size);
+}
+
+static inline void
+copy_from (void *buffer, int image)
+{
+ copy_out (get_obj_ptr (this_image.image_num), image);
+}
+
+static inline void
+copy_in (void *buffer, void *obj)
+{
+ copy_to (obj, this_image.image_num);
+}
+#endif
diff --git a/libgfortran/nca/collective_subroutine.c b/libgfortran/nca/collective_subroutine.c
new file mode 100644
index 00000000000..8a8a7d659f0
--- /dev/null
+++ b/libgfortran/nca/collective_subroutine.c
@@ -0,0 +1,416 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#include <string.h>
+#include "libgfortran.h"
+#include "libcoarraynative.h"
+#include "collective_subroutine.h"
+#include "collective_inline.h"
+#include "allocator.h"
+
+void *
+get_collsub_buf (collsub_iface *ci, size_t size)
+{
+ void *ret;
+
+ pthread_mutex_lock (&ci->s->mutex);
+ if (size > ci->s->curr_size)
+ {
+ shared_free (ci->a, ci->s->collsub_buf, ci->s->curr_size);
+ ci->s->collsub_buf = shared_malloc (ci->a, size);
+ ci->s->curr_size = size;
+ }
+
+ ret = SHMPTR_AS (void *, ci->s->collsub_buf, ci->sm);
+ pthread_mutex_unlock (&ci->s->mutex);
+ return ret;
+}
+
+/* It appears as if glibc's barrier implementation does not spin (at
+ least that is what I got from a quick glance at the source code),
+ so performance would be improved quite a bit if we spun a few times
+ here so we don't run into the futex syscall. */
+
+void
+collsub_sync (collsub_iface *ci)
+{
+ //dprintf (2, "Calling collsub_sync %d times\n", ++called);
+ pthread_barrier_wait (&ci->s->barrier);
+}
+
+/* assign_function is needed since we only know how to assign the type inside
+ the compiler. It should be implemented as follows:
+
+ void assign_function (void *a, void *b)
+ {
+ *((t *) a) = reduction_operation ((t *) a, (t *) b);
+ }
+
+ */
+
+void
+collsub_reduce_array (collsub_iface *ci, gfc_array_char *desc, int *result_image,
+ void (*assign_function) (void *, void *))
+{
+ void *buffer;
+ pack_info pi;
+ bool packed;
+ int cbit = 0;
+ int imoffset;
+ index_type elem_size;
+ index_type this_image_size_bytes;
+ char *this_image_buf;
+
+ packed = pack_array_prepare (&pi, desc);
+ if (pi.num_elem == 0)
+ return;
+
+ elem_size = GFC_DESCRIPTOR_SIZE (desc);
+ this_image_size_bytes = elem_size * pi.num_elem;
+
+ buffer = get_collsub_buf (ci, this_image_size_bytes * local->num_images);
+ this_image_buf = buffer + this_image_size_bytes * this_image.image_num;
+
+ if (packed)
+ memcpy (this_image_buf, GFC_DESCRIPTOR_DATA (desc), this_image_size_bytes);
+ else
+ pack_array_finish (&pi, desc, this_image_buf);
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ /* Reduce arrays elementwise. */
+ for (size_t i = 0; i < pi.num_elem; i++)
+ assign_function (this_image_buf + elem_size * i,
+ this_image_buf + this_image_size_bytes * imoffset + elem_size * i);
+
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || *result_image == this_image.image_num)
+ {
+ if (packed)
+ memcpy (GFC_DESCRIPTOR_DATA (desc), buffer, this_image_size_bytes);
+ else
+ unpack_array_finish(&pi, desc, buffer);
+ }
+
+ finish_collective_subroutine (ci);
+}
+
+void
+collsub_reduce_scalar (collsub_iface *ci, void *obj, index_type elem_size,
+ int *result_image,
+ void (*assign_function) (void *, void *))
+{
+ void *buffer;
+ int cbit = 0;
+ int imoffset;
+ char *this_image_buf;
+
+ buffer = get_collsub_buf (ci, elem_size * local->num_images);
+ this_image_buf = buffer + elem_size * this_image.image_num;
+
+ memcpy (this_image_buf, obj, elem_size);
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ /* Reduce arrays elementwise. */
+ assign_function (this_image_buf, this_image_buf + elem_size*imoffset);
+
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || *result_image == this_image.image_num)
+ memcpy(obj, buffer, elem_size);
+
+ finish_collective_subroutine (ci);
+}
+
+/* Do not use sync_all(), because the program should deadlock in the case that
+ * some images are on a sync_all barrier while others are in a collective
+ * subroutine. */
+
+void
+collsub_iface_init (collsub_iface *ci, alloc_iface *ai, shared_memory *sm)
+{
+ pthread_barrierattr_t attr;
+ shared_mem_ptr p;
+ ci->s = SHARED_MEMORY_RAW_ALLOC_PTR(sm, collsub_iface_shared);
+
+ ci->s->collsub_buf = shared_malloc(get_allocator(ai), sizeof(double)*local->num_images);
+ ci->s->curr_size = sizeof(double)*local->num_images;
+ ci->sm = sm;
+ ci->a = get_allocator(ai);
+
+ pthread_barrierattr_init (&attr);
+ pthread_barrierattr_setpshared (&attr, PTHREAD_PROCESS_SHARED);
+ pthread_barrier_init (&ci->s->barrier, &attr, local->num_images);
+ pthread_barrierattr_destroy(&attr);
+
+ initialize_shared_mutex (&ci->s->mutex);
+}
+
+void
+collsub_broadcast_scalar (collsub_iface *ci, void *obj, index_type elem_size,
+ int source_image /* Adjusted in the wrapper. */)
+{
+ void *buffer;
+
+ buffer = get_collsub_buf (ci, elem_size);
+
+ dprintf(2, "Source image: %d\n", source_image);
+
+ if (source_image == this_image.image_num)
+ {
+ memcpy (buffer, obj, elem_size);
+ collsub_sync (ci);
+ }
+ else
+ {
+ collsub_sync (ci);
+ memcpy (obj, buffer, elem_size);
+ }
+
+ finish_collective_subroutine (ci);
+}
+
+void
+collsub_broadcast_array (collsub_iface *ci, gfc_array_char *desc,
+ int source_image)
+{
+ void *buffer;
+ pack_info pi;
+ bool packed;
+ index_type elem_size;
+ index_type size_bytes;
+ char *this_image_buf;
+
+ packed = pack_array_prepare (&pi, desc);
+ if (pi.num_elem == 0)
+ return;
+
+ elem_size = GFC_DESCRIPTOR_SIZE (desc);
+ size_bytes = elem_size * pi.num_elem;
+
+ buffer = get_collsub_buf (ci, size_bytes);
+
+ if (source_image == this_image.image_num)
+ {
+ if (packed)
+ memcpy (buffer, GFC_DESCRIPTOR_DATA (desc), size_bytes);
+ else
+ pack_array_finish (&pi, desc, buffer);
+ collsub_sync (ci);
+ }
+ else
+ {
+ collsub_sync (ci);
+ if (packed)
+ memcpy (GFC_DESCRIPTOR_DATA (desc), buffer, size_bytes);
+ else
+ unpack_array_finish(&pi, desc, buffer);
+ }
+
+ finish_collective_subroutine (ci);
+}
+
+#if 0
+
+void nca_co_broadcast (gfc_array_char *, int, int*, char *, size_t);
+export_proto (nca_co_broadcast);
+
+void
+nca_co_broadcast (gfc_array_char * restrict a, int source_image,
+ int *stat, char *errmsg __attribute__ ((unused)),
+ size_t errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ index_type type_size;
+ index_type dim;
+ index_type span;
+ bool packed, empty;
+ index_type num_elems;
+ index_type ssize, ssize_bytes;
+ char *this_shared_ptr, *other_shared_ptr;
+
+ if (stat)
+ *stat = 0;
+
+ dim = GFC_DESCRIPTOR_RANK (a);
+ type_size = GFC_DESCRIPTOR_SIZE (a);
+
+ /* Source image, gather. */
+ if (source_image - 1 == image_num)
+ {
+ num_elems = 1;
+ if (dim > 0)
+ {
+ span = a->span != 0 ? a->span : type_size;
+ packed = true;
+ empty = false;
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (a, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (a, n);
+
+ empty = empty || extent[n] <= 0;
+
+ if (num_elems != GFC_DESCRIPTOR_STRIDE (a, n))
+ packed = false;
+
+ num_elems *= extent[n];
+ }
+ ssize_bytes = num_elems * type_size;
+ }
+ else
+ {
+ ssize_bytes = type_size;
+ packed = true;
+ empty = false;
+ }
+
+ prepare_collective_subroutine (ssize_bytes); // broadcast barrier 1
+ this_shared_ptr = get_obj_ptr (image_num);
+ if (packed)
+ memcpy (this_shared_ptr, a->base_addr, ssize_bytes);
+ else
+ {
+ char *src = (char *) a->base_addr;
+ char * restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+
+ memcpy (dest, src, type_size);
+ dest += type_size;
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it
+ and increment the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+ collsub_sync (ci); /* Broadcast barrier 2. */
+ }
+ else /* Target image, scatter. */
+ {
+ collsub_sync (ci); /* Broadcast barrier 1. */
+ packed = 1;
+ num_elems = 1;
+ span = a->span != 0 ? a->span : type_size;
+
+ for (index_type n = 0; n < dim; n++)
+ {
+ index_type stride_n;
+ count[n] = 0;
+ stride_n = GFC_DESCRIPTOR_STRIDE (a, n);
+ stride[n] = stride_n * type_size;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (a, n);
+ if (extent[n] <= 0)
+ {
+ packed = true;
+ num_elems = 0;
+ break;
+ }
+ if (num_elems != stride_n)
+ packed = false;
+
+ num_elems *= extent[n];
+ }
+ ssize = num_elems * type_size;
+ prepare_collective_subroutine (ssize); /* Broadcaset barrier 2. */
+ other_shared_ptr = get_obj_ptr (source_image - 1);
+ if (packed)
+ memcpy (a->base_addr, other_shared_ptr, ssize);
+ else
+ {
+ char *src = other_shared_ptr;
+ char * restrict dest = (char *) a->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ memcpy (dest, src, type_size);
+ src += span;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci); /* Broadcast barrier 3. */
+}
+
+#endif
diff --git a/libgfortran/nca/collective_subroutine.h b/libgfortran/nca/collective_subroutine.h
new file mode 100644
index 00000000000..6147dd6d793
--- /dev/null
+++ b/libgfortran/nca/collective_subroutine.h
@@ -0,0 +1,44 @@
+
+#ifndef COLLECTIVE_SUBROUTINE_HDR
+#define COLLECTIVE_SUBROUTINE_HDR
+
+#include "shared_memory.h"
+
+typedef struct collsub_iface_shared
+{
+ size_t curr_size;
+ shared_mem_ptr collsub_buf;
+ pthread_barrier_t barrier;
+ pthread_mutex_t mutex;
+} collsub_iface_shared;
+
+typedef struct collsub_iface
+{
+ collsub_iface_shared *s;
+ allocator *a;
+ shared_memory *sm;
+} collsub_iface;
+
+void collsub_broadcast_scalar (collsub_iface *, void *, index_type, int);
+internal_proto (collsub_broadcast_scalar);
+
+void collsub_broadcast_array (collsub_iface *, gfc_array_char *, int);
+internal_proto (collsub_broadcast_array);
+
+void collsub_reduce_array (collsub_iface *, gfc_array_char *, int *,
+ void (*) (void *, void *));
+internal_proto (collsub_reduce_array);
+
+void collsub_reduce_scalar (collsub_iface *, void *, index_type, int *,
+ void (*) (void *, void *));
+internal_proto (collsub_reduce_scalar);
+
+void collsub_sync (collsub_iface *);
+internal_proto (collsub_sync);
+
+void collsub_iface_init (collsub_iface *, alloc_iface *, shared_memory *);
+internal_proto (collsub_iface_init);
+
+void * get_collsub_buf (collsub_iface *ci, size_t size);
+internal_proto (get_collsub_buf);
+#endif
diff --git a/libgfortran/nca/hashmap.c b/libgfortran/nca/hashmap.c
new file mode 100644
index 00000000000..61f5487e63e
--- /dev/null
+++ b/libgfortran/nca/hashmap.c
@@ -0,0 +1,447 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#include "libgfortran.h"
+#include "hashmap.h"
+#include <string.h>
+
+#define INITIAL_BITNUM (5)
+#define INITIAL_SIZE (1<<INITIAL_BITNUM)
+#define CRITICAL_LOOKAHEAD (16)
+
+static ssize_t n_ent;
+
+typedef struct {
+ memid id;
+ shared_mem_ptr p; /* If p == SHMPTR_NULL, the entry is empty. */
+ size_t s;
+ int max_lookahead;
+ int refcnt;
+} hashmap_entry;
+
+static ssize_t
+num_entries (hashmap_entry *data, size_t size)
+{
+ ssize_t i;
+ ssize_t ret = 0;
+ for (i = 0; i < size; i++)
+ {
+ if (!SHMPTR_IS_NULL (data[i].p))
+ ret ++;
+ }
+ return ret;
+}
+
+/* 64 bit to 64 bit hash function. */
+
+/*
+static inline uint64_t
+hash (uint64_t x)
+{
+ return x * 11400714819323198485lu;
+}
+*/
+
+#define ASSERT_HM(hm, cond) assert_hashmap(hm, cond, #cond)
+
+static void
+assert_hashmap(hashmap *hm, bool asserted, const char *cond)
+{
+ if (!asserted)
+ {
+ dprintf(2, cond);
+ dump_hm(hm);
+ }
+ assert(asserted);
+}
+
+static inline uint64_t
+hash (uint64_t key)
+{
+ key ^= (key >> 30);
+ key *= 0xbf58476d1ce4e5b9ul;
+ key ^= (key >> 27);
+ key *= 0x94d049bb133111ebul;
+ key ^= (key >> 31);
+
+ return key;
+}
+
+/* Gets a pointer to the current data in the hashmap. */
+static inline hashmap_entry *
+get_data(hashmap *hm)
+{
+ return SHMPTR_AS (hashmap_entry *, hm->s->data, hm->sm);
+}
+
+/* Generate mask from current number of bits. */
+
+static inline intptr_t
+gen_mask (hashmap *hm)
+{
+ return (1 << hm->s->bitnum) - 1;
+}
+
+/* Add with wrap-around at hashmap size. */
+
+static inline size_t
+hmiadd (hashmap *hm, size_t s, ssize_t o) {
+ return (s + o) & gen_mask (hm);
+}
+
+/* Get the expected offset for entry id. */
+
+static inline ssize_t
+get_expected_offset (hashmap *hm, memid id)
+{
+ return hash(id) >> (PTR_BITS - hm->s->bitnum);
+}
+
+/* Initialize the hashmap. */
+
+void
+hashmap_init (hashmap *hm, hashmap_shared *hs, allocator *a,
+ shared_memory *mem)
+{
+ hashmap_entry *data;
+ hm->s = hs;
+ hm->sm = mem;
+ hm->s->data = shared_malloc (a, INITIAL_SIZE * sizeof(hashmap_entry));
+ data = get_data (hm);
+ memset(data, '\0', INITIAL_SIZE*sizeof(hashmap_entry));
+
+ for (int i = 0; i < INITIAL_SIZE; i++)
+ data[i].p = SHMPTR_NULL;
+
+ hm->s->size = INITIAL_SIZE;
+ hm->s->bitnum = INITIAL_BITNUM;
+ hm->a = a;
+}
+
+/* This checks if the entry id exists in that range the range between
+ the expected position and the maximum lookahead. */
+
+static ssize_t
+scan_inside_lookahead (hashmap *hm, ssize_t expected_off, memid id)
+{
+ ssize_t lookahead;
+ hashmap_entry *data;
+
+ data = get_data (hm);
+ lookahead = data[expected_off].max_lookahead;
+ ASSERT_HM (hm, lookahead < CRITICAL_LOOKAHEAD);
+
+ for (int i = 0; i <= lookahead; i++) /* For performance, this could
+ iterate backwards. */
+ if (data[hmiadd (hm, expected_off, i)].id == id)
+ return hmiadd (hm, expected_off, i);
+
+ return -1;
+}
+
+/* Scan for the next empty slot we can use. Returns offset relative
+ to the expected position. */
+
+static ssize_t
+scan_empty (hashmap *hm, ssize_t expected_off, memid id)
+{
+ hashmap_entry *data;
+
+ data = get_data(hm);
+ for (int i = 0; i < CRITICAL_LOOKAHEAD; i++)
+ if (SHMPTR_IS_NULL (data[hmiadd (hm, expected_off, i)].p))
+ return i;
+
+ return -1;
+}
+
+/* Search the hashmap for id. */
+
+hashmap_search_result
+hashmap_get (hashmap *hm, memid id)
+{
+ hashmap_search_result ret;
+ hashmap_entry *data;
+ size_t expected_offset;
+ ssize_t res;
+
+ data = get_data (hm);
+ expected_offset = get_expected_offset (hm, id);
+ res = scan_inside_lookahead (hm, expected_offset, id);
+
+ if (res != -1)
+ ret = ((hashmap_search_result)
+ { .p = data[res].p, .size=data[res].s, .res_offset = res });
+ else
+ ret.p = SHMPTR_NULL;
+
+ return ret;
+}
+
+/* Return size of a hashmap search result. */
+size_t
+hm_search_result_size (hashmap_search_result *res)
+{
+ return res->size;
+}
+
+/* Return pointer of a hashmap search result. */
+
+shared_mem_ptr
+hm_search_result_ptr (hashmap_search_result *res)
+{
+ return res->p;
+}
+
+/* Return pointer of a hashmap search result. */
+
+bool
+hm_search_result_contains (hashmap_search_result *res)
+{
+ return !SHMPTR_IS_NULL(res->p);
+}
+
+/* Enlarge hashmap memory. */
+
+static void
+enlarge_hashmap_mem (hashmap *hm, hashmap_entry **data, bool f)
+{
+ shared_mem_ptr old_data_p;
+ size_t old_size;
+
+ old_data_p = hm->s->data;
+ old_size = hm->s->size;
+
+ hm->s->data = shared_malloc (hm->a, (hm->s->size *= 2)*sizeof(hashmap_entry));
+ fprintf (stderr,"enlarge_hashmap_mem: %ld\n", hm->s->data.offset);
+ hm->s->bitnum++;
+
+ *data = get_data(hm);
+ for (size_t i = 0; i < hm->s->size; i++)
+ (*data)[i] = ((hashmap_entry) { .id = 0, .p = SHMPTR_NULL, .s=0,
+ .max_lookahead = 0, .refcnt=0 });
+
+ if (f)
+ shared_free(hm->a, old_data_p, old_size);
+}
+
+/* Resize hashmap. */
+
+static void
+resize_hm (hashmap *hm, hashmap_entry **data)
+{
+ shared_mem_ptr old_data_p;
+ hashmap_entry *old_data, *new_data;
+ size_t old_size;
+ ssize_t new_offset, inital_index, new_index;
+ memid id;
+ ssize_t max_lookahead;
+ ssize_t old_count, new_count;
+
+ /* old_data points to the old block containing the hashmap. We
+ redistribute the data from there into the new block. */
+
+ old_data_p = hm->s->data;
+ old_data = *data;
+ old_size = hm->s->size;
+ old_count = num_entries (old_data, old_size);
+
+ fprintf(stderr, "Occupancy at resize: %f\n", ((double) old_count)/old_size);
+
+ //fprintf (stderr,"\n====== Resizing hashmap =========\n\nOld map:\n\n");
+ //dump_hm (hm);
+ enlarge_hashmap_mem (hm, &new_data, false);
+ //fprintf (stderr,"old_data: %p new_data: %p\n", old_data, new_data);
+ retry_resize:
+ for (size_t i = 0; i < old_size; i++)
+ {
+ if (SHMPTR_IS_NULL (old_data[i].p))
+ continue;
+
+ id = old_data[i].id;
+ inital_index = get_expected_offset (hm, id);
+ new_offset = scan_empty (hm, inital_index, id);
+
+ /* If we didn't find a free slot, just resize the hashmap
+ again. */
+ if (new_offset == -1)
+ {
+ enlarge_hashmap_mem (hm, &new_data, true);
+ //fprintf (stderr,"\n====== AGAIN Resizing hashmap =========\n\n");
+ //fprintf (stderr,"old_data: %p new_data %p\n", old_data, new_data);
+ goto retry_resize; /* Sue me. */
+ }
+
+ ASSERT_HM (hm, new_offset < CRITICAL_LOOKAHEAD);
+ new_index = hmiadd (hm, inital_index, new_offset);
+ max_lookahead = new_data[inital_index].max_lookahead;
+ new_data[inital_index].max_lookahead
+ = new_offset > max_lookahead ? new_offset : max_lookahead;
+
+ new_data[new_index] = ((hashmap_entry) {.id = id, .p = old_data[i].p,
+ .s = old_data[i].s,
+ .max_lookahead = new_data[new_index].max_lookahead,
+ .refcnt = old_data[i].refcnt});
+ }
+ new_count = num_entries (new_data, hm->s->size);
+ //fprintf (stderr,"Number of elements: %ld to %ld\n", old_count, new_count);
+ //fprintf (stderr,"============ After resizing: =======\n\n");
+ //dump_hm (hm);
+
+ shared_free (hm->a, old_data_p, old_size);
+ *data = new_data;
+}
+
+/* Set an entry in the hashmap. */
+
+void
+hashmap_set (hashmap *hm, memid id, hashmap_search_result *hsr,
+ shared_mem_ptr p, size_t size)
+{
+ hashmap_entry *data;
+ ssize_t expected_offset, lookahead;
+ ssize_t empty_offset;
+ ssize_t delta;
+
+ // //fprintf (stderr,"hashmap_set: id = %-16p\n", (void *) id);
+ data = get_data(hm);
+
+ if (hsr) {
+ data[hsr->res_offset].s = size;
+ data[hsr->res_offset].p = p;
+ return;
+ }
+
+ expected_offset = get_expected_offset (hm, id);
+ while ((delta = scan_empty (hm, expected_offset, id)) == -1)
+ {
+ resize_hm (hm, &data);
+ expected_offset = get_expected_offset (hm, id);
+ }
+
+ empty_offset = hmiadd (hm, expected_offset, delta);
+ lookahead = data[expected_offset].max_lookahead;
+ data[expected_offset].max_lookahead = delta > lookahead ? delta : lookahead;
+ data[empty_offset] = ((hashmap_entry) {.id = id, .p = p, .s = size,
+ .max_lookahead = data[empty_offset].max_lookahead,
+ .refcnt = 1});
+
+ n_ent ++;
+ fprintf (stderr,"hashmap_set: Setting %p at %p, n_ent = %ld\n", (void *) id, data + empty_offset,
+ n_ent);
+ // dump_hm (hm);
+ // fprintf(stderr, "--------------------------------------------------\n");
+ /* TODO: Shouldn't reset refcnt, but this doesn't matter at the
+ moment because of the way the function is used. */
+}
+
+/* Change the refcount of a hashmap entry. */
+
+static int
+hashmap_change_refcnt (hashmap *hm, memid id, hashmap_search_result *res,
+ int delta)
+{
+ hashmap_entry *data;
+ hashmap_search_result r;
+ hashmap_search_result *pr;
+ int ret;
+ hashmap_entry *entry;
+
+ data = get_data (hm);
+
+ if (res)
+ pr = res;
+ else
+ {
+ r = hashmap_get (hm, id);
+ pr = &r;
+ }
+
+ entry = &data[pr->res_offset];
+ ret = (entry->refcnt += delta);
+ if (ret == 0)
+ {
+ n_ent --;
+ //fprintf (stderr, "hashmap_change_refcnt: removing %p at %p, n_ent = %ld\n",
+ // (void *) id, entry, n_ent);
+ entry->id = 0;
+ entry->p = SHMPTR_NULL;
+ entry->s = 0;
+ }
+
+ return ret;
+}
+
+/* Increase hashmap entry refcount. */
+
+void
+hashmap_inc (hashmap *hm, memid id, hashmap_search_result * res)
+{
+ int ret;
+ ret = hashmap_change_refcnt (hm, id, res, 1);
+ ASSERT_HM (hm, ret > 0);
+}
+
+/* Decrease hashmap entry refcount. */
+
+int
+hashmap_dec (hashmap *hm, memid id, hashmap_search_result * res)
+{
+ int ret;
+ ret = hashmap_change_refcnt (hm, id, res, -1);
+ ASSERT_HM (hm, ret >= 0);
+ return ret;
+}
+
+#define PE(str, ...) fprintf(stderr, INDENT str, ##__VA_ARGS__)
+#define INDENT ""
+
+void
+dump_hm(hashmap *hm) {
+ hashmap_entry *data;
+ size_t exp;
+ size_t occ_num = 0;
+ PE("h %p (size: %lu, bitnum: %d)\n", hm, hm->s->size, hm->s->bitnum);
+ data = get_data (hm);
+ fprintf (stderr,"offset = %lx data = %p\n", (unsigned long) hm->s->data.offset, data);
+
+#undef INDENT
+#define INDENT " "
+ for (size_t i = 0; i < hm->s->size; i++) {
+ exp = get_expected_offset(hm, data[i].id);
+ if (!SHMPTR_IS_NULL(data[i].p)) {
+ PE("%2lu. (exp: %2lu w la %d) id %#-16lx p %#-14lx s %-7lu -- la %u ref %u %-16p\n",
+ i, exp, data[exp].max_lookahead, data[i].id, data[i].p.offset, data[i].s,
+ data[i].max_lookahead, data[i].refcnt, data + i);
+ occ_num++;
+ }
+ else
+ PE("%2lu. empty -- la %u %p\n", i, data[i].max_lookahead,
+ data + i);
+
+ }
+#undef INDENT
+#define INDENT ""
+ PE("occupancy: %lu %f\n", occ_num, ((double) occ_num)/hm->s->size);
+}
diff --git a/libgfortran/nca/hashmap.h b/libgfortran/nca/hashmap.h
new file mode 100644
index 00000000000..4d999e3e3d3
--- /dev/null
+++ b/libgfortran/nca/hashmap.h
@@ -0,0 +1,70 @@
+#ifndef HASHMAP_H
+
+#include "shared_memory.h"
+#include "allocator.h"
+
+#include <stdint.h>
+#include <stddef.h>
+
+
+/* Data structures and variables:
+
+ memid is a unique identifier for the coarray, the address of its
+ descriptor (which is unique in the program). */
+typedef intptr_t memid;
+
+typedef struct {
+ shared_mem_ptr data;
+ size_t size;
+ int bitnum;
+} hashmap_shared;
+
+typedef struct hashmap
+{
+ hashmap_shared *s;
+ shared_memory *sm;
+ allocator *a;
+} hashmap;
+
+typedef struct {
+ shared_mem_ptr p;
+ size_t size;
+ ssize_t res_offset;
+} hashmap_search_result;
+
+void hashmap_init (hashmap *, hashmap_shared *, allocator *a, shared_memory *);
+
+/* Look up memid in the hashmap. The result can be inspected via the
+ hm_search_result_* functions. */
+
+hashmap_search_result hashmap_get (hashmap *, memid);
+
+/* Given a search result, returns the size. */
+size_t hm_search_result_size (hashmap_search_result *);
+
+/* Given a search result, returns the pointer. */
+shared_mem_ptr hm_search_result_ptr (hashmap_search_result *);
+
+/* Given a search result, returns whether something was found. */
+bool hm_search_result_contains (hashmap_search_result *);
+
+/* Sets the hashmap entry for memid to shared_mem_ptr and
+ size_t. Optionally, if a hashmap_search_result is supplied, it is
+ used to make the lookup faster. */
+
+void hashmap_set (hashmap *, memid, hashmap_search_result *, shared_mem_ptr p,
+ size_t);
+
+/* Increments the hashmap entry for memid. Optionally, if a
+ hashmap_search_result is supplied, it is used to make the lookup
+ faster. */
+
+void hashmap_inc (hashmap *, memid, hashmap_search_result *);
+
+/* Same, but decrement. */
+int hashmap_dec (hashmap *, memid, hashmap_search_result *);
+
+void dump_hm (hashmap *hm);
+
+#define HASHMAP_H
+#endif
diff --git a/libgfortran/nca/libcoarraynative.h b/libgfortran/nca/libcoarraynative.h
new file mode 100644
index 00000000000..507de0cde8e
--- /dev/null
+++ b/libgfortran/nca/libcoarraynative.h
@@ -0,0 +1,103 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#ifndef LIBGFOR_H
+#error "Include libgfortran.h before libcoarraynative.h"
+#endif
+
+#ifndef COARRAY_NATIVE_HDR
+#define COARRAY_NATIVE_HDR
+
+#include "libgfortran.h"
+
+#include <sys/types.h>
+#include <stdint.h>
+#include <stdio.h>
+
+
+/* This is to create a _nca_gfortrani_ prefix for all variables and
+ function used only by nca. */
+#if 0
+#define NUM_ADDR_BITS (8 * sizeof (int *))
+#endif
+
+#define DEBUG_NATIVE_COARRAY 1
+
+#ifdef DEBUG_NATIVE_COARRAY
+#define DEBUG_PRINTF(...) dprintf (2,__VA_ARGS__)
+#else
+#define DEBUG_PRINTF(...) do {} while(0)
+#endif
+
+#include "allocator.h"
+#include "hashmap.h"
+#include "sync.h"
+#include "lock.h"
+#include "collective_subroutine.h"
+
+typedef struct {
+ pthread_barrier_t barrier;
+ int maximg;
+} ipcollsub;
+
+typedef enum {
+ IMAGE_UNKNOWN = 0,
+ IMAGE_OK,
+ IMAGE_FAILED
+} image_status;
+
+typedef struct {
+ image_status status;
+ pid_t pid;
+} image_tracker;
+
+typedef struct {
+ int has_failed_image;
+ image_tracker images[];
+} master;
+
+typedef struct {
+ int image_num;
+ master *m;
+} image;
+
+extern image this_image;
+
+typedef struct {
+ int num_images;
+ shared_memory sm;
+ alloc_iface ai;
+ collsub_iface ci;
+ sync_iface si;
+} nca_local_data;
+
+extern nca_local_data *local;
+internal_proto (local);
+void ensure_initialization(void);
+internal_proto(ensure_initialization);
+
+void nca_master(void (*)(void));
+export_proto (nca_master);
+
+#endif
diff --git a/libgfortran/nca/lock.h b/libgfortran/nca/lock.h
new file mode 100644
index 00000000000..469739598c5
--- /dev/null
+++ b/libgfortran/nca/lock.h
@@ -0,0 +1,37 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#ifndef COARRAY_LOCK_HDR
+#define COARRAY_LOCK_HDR
+
+#include <pthread.h>
+
+typedef struct {
+ int owner;
+ int initialized;
+ pthread_mutex_t arr[];
+} lock_array;
+
+#endif
diff --git a/libgfortran/nca/shared_memory.c b/libgfortran/nca/shared_memory.c
new file mode 100644
index 00000000000..bc3093d0ef2
--- /dev/null
+++ b/libgfortran/nca/shared_memory.c
@@ -0,0 +1,221 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#include "libgfortran.h"
+#include "libcoarraynative.h"
+#include "util.h"
+#include <sys/mman.h>
+#include <unistd.h>
+#include <string.h>
+
+#include "shared_memory.h"
+
+/* This implements shared memory based on POSIX mmap. We start with
+ memory block of the size of the global shared memory data, rounded
+ up to one pagesize, and enlarge as needed.
+
+ We address the memory via a shared_memory_ptr, which is an offset into
+ the shared memory block. The metadata is situated at offset 0.
+
+ In order to be able to resize the memory and to keep pointers
+ valid, we keep the old mapping around, so the memory is actually
+ visible several times to the process. Thus, pointers returned by
+ shared_memory_get_mem_with_alignment remain valid even when
+ resizing. */
+
+/* Global metadata for shared memory, always kept at offset 0. */
+
+typedef struct
+{
+ size_t size;
+ size_t used;
+ int fd;
+} global_shared_memory_meta;
+
+/* Type realization for opaque type shared_memory. */
+
+typedef struct shared_memory_act
+{
+ global_shared_memory_meta *meta;
+ void *header;
+ size_t last_seen_size;
+
+ /* We don't need to free these. We probably also don't need to keep
+ track of them, but it is much more future proof if we do. */
+
+ size_t num_local_allocs;
+
+ struct local_alloc {
+ void *base;
+ size_t size;
+ } allocs[];
+
+} shared_memory_act;
+
+/* Convenience wrapper for mmap. */
+
+static inline void *
+map_memory (int fd, size_t size, off_t offset)
+{
+ void *ret = mmap (NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, offset);
+ if (ret == MAP_FAILED)
+ {
+ perror("mmap failed");
+ exit(1);
+ }
+ return ret;
+}
+
+/* Returns the size of shared_memory_act. */
+
+static inline size_t
+get_shared_memory_act_size (int nallocs)
+{
+ return sizeof(shared_memory_act) + nallocs*sizeof(struct local_alloc);
+}
+
+/* When the shared memory block is enlarged, we need to map it into
+ virtual memory again. */
+
+static inline shared_memory_act *
+new_base_mapping (shared_memory_act *mem)
+{
+ shared_memory_act *newmem;
+ /* We need another entry in the alloc table. */
+ mem->num_local_allocs++;
+ newmem = realloc (mem, get_shared_memory_act_size (mem->num_local_allocs));
+ newmem->allocs[newmem->num_local_allocs - 1]
+ = ((struct local_alloc)
+ {.base = map_memory (newmem->meta->fd, newmem->meta->size, 0),
+ .size = newmem->meta->size});
+ newmem->last_seen_size = newmem->meta->size;
+ return newmem;
+}
+
+/* Return the most recently allocated base pointer. */
+
+static inline void *
+last_base (shared_memory_act *mem)
+{
+ return mem->allocs[mem->num_local_allocs - 1].base;
+}
+
+/* Get a pointer into the shared memory block with alignemnt
+ (works similar to sbrk). */
+
+shared_mem_ptr
+shared_memory_get_mem_with_alignment (shared_memory_act **pmem, size_t size,
+ size_t align)
+{
+ shared_memory_act *mem = *pmem;
+ size_t new_size;
+ size_t orig_used;
+
+ /* Offset into memory block with alignment. */
+ size_t used_wa = alignto (mem->meta->used, align);
+
+ if (used_wa + size <= mem->meta->size)
+ {
+ memset(last_base(mem) + mem->meta->used, 0xCA, used_wa - mem->meta->used);
+ memset(last_base(mem) + used_wa, 0x42, size);
+ mem->meta->used = used_wa + size;
+
+ DEBUG_PRINTF ("Shared Memory: New memory of size %#lx requested, returned %#lx\n", size, used_wa);
+ return (shared_mem_ptr) {.offset = used_wa};
+ }
+
+ /* We need to enlarge the memory segment. Double the size if that
+ is big enough, otherwise get what's needed. */
+
+ if (mem->meta->size * 2 < used_wa + size)
+ new_size = mem->meta->size * 2;
+ else
+ new_size = round_to_pagesize (used_wa + size);
+
+ orig_used = mem->meta->used;
+ mem->meta->size = new_size;
+ mem->meta->used = used_wa + size;
+ ftruncate (mem->meta->fd, mem->meta->size);
+ /* This also sets the new base pointer where the shared memory
+ can be found in the address space. */
+
+ mem = new_base_mapping (mem);
+
+ *pmem = mem;
+ assert(used_wa != 0);
+
+ dprintf(2, "Shared Memory: New memory of size %#lx requested, returned %#lx\n", size, used_wa);
+ memset(last_base(mem) + orig_used, 0xCA, used_wa - orig_used);
+ memset(last_base(mem) + used_wa, 0x42, size);
+
+ return (shared_mem_ptr) {.offset = used_wa};
+}
+
+/* If another image changed the size, update the size accordingly. */
+
+void
+shared_memory_prepare (shared_memory_act **pmem)
+{
+ shared_memory_act *mem = *pmem;
+ if (mem->meta->size == mem->last_seen_size)
+ return;
+ mem = new_base_mapping(mem);
+ *pmem = mem;
+}
+
+/* Initialize the memory with one page, the shared metadata of the
+ shared memory is stored at the beginning. */
+
+void
+shared_memory_init (shared_memory_act **pmem)
+{
+ shared_memory_act *mem;
+ int fd;
+ size_t initial_size = round_to_pagesize (sizeof (global_shared_memory_meta));
+
+ mem = malloc (get_shared_memory_act_size(1));
+ fd = get_shmem_fd();
+
+ ftruncate(fd, initial_size);
+ mem->meta = map_memory (fd, initial_size, 0);
+ *mem->meta = ((global_shared_memory_meta) {.size = initial_size,
+ .used = sizeof(global_shared_memory_meta),
+ .fd = fd});
+ mem->last_seen_size = initial_size;
+ mem->num_local_allocs = 1;
+ mem->allocs[0] = ((struct local_alloc) {.base = mem->meta,
+ .size = initial_size});
+
+ *pmem = mem;
+}
+
+/* Convert a shared memory pointer (i.e. an offset into the shared
+ memory block) to a pointer. */
+
+void *
+shared_mem_ptr_to_void_ptr(shared_memory_act **pmem, shared_mem_ptr smp)
+{
+ return last_base(*pmem) + smp.offset;
+}
+
diff --git a/libgfortran/nca/shared_memory.h b/libgfortran/nca/shared_memory.h
new file mode 100644
index 00000000000..4adc104801d
--- /dev/null
+++ b/libgfortran/nca/shared_memory.h
@@ -0,0 +1,78 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#ifndef SHARED_MEMORY_H
+#include <stdbool.h>
+#include <stdint.h>
+#include <stddef.h>
+#include <sys/types.h>
+#include <pthread.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <assert.h>
+#include <limits.h>
+
+/* A struct to serve as an opaque shared memory object. */
+
+struct shared_memory_act;
+typedef struct shared_memory_act * shared_memory;
+
+#define SHMPTR_NULL ((shared_mem_ptr) {.offset = -1})
+#define SHMPTR_IS_NULL(x) (x.offset == -1)
+
+#define SHMPTR_DEREF(x, s, sm) \
+ ((x) = *(__typeof(x) *) shared_mem_ptr_to_void_ptr (sm, s);
+#define SHMPTR_AS(t, s, sm) ((t) shared_mem_ptr_to_void_ptr(sm, s))
+#define SHMPTR_SET(v, s, sm) (v = SHMPTR_AS(__typeof(v), s, sm))
+#define SHMPTR_EQUALS(s1, s2) (s1.offset == s2.offset)
+
+#define SHARED_MEMORY_RAW_ALLOC(mem, t, n) \
+ shared_memory_get_mem_with_alignment(mem, sizeof(t)*n, __alignof__(t))
+
+#define SHARED_MEMORY_RAW_ALLOC_PTR(mem, t) \
+ SHMPTR_AS (t *, SHARED_MEMORY_RAW_ALLOC (mem, t, 1), mem)
+
+/* A shared-memory pointer is implemented as an offset into the shared
+ memory region. */
+
+typedef struct shared_mem_ptr
+{
+ ssize_t offset;
+} shared_mem_ptr;
+
+void shared_memory_init (shared_memory *);
+internal_proto (shared_memory_init);
+
+void shared_memory_prepare (shared_memory *);
+internal_proto (shared_memory_prepare);
+
+shared_mem_ptr shared_memory_get_mem_with_alignment (shared_memory *mem,
+ size_t size, size_t align);
+internal_proto (shared_memory_get_mem_with_alignment);
+
+void *shared_mem_ptr_to_void_ptr (shared_memory *, shared_mem_ptr);
+internal_proto (shared_mem_ptr_to_void_ptr);
+
+#define SHARED_MEMORY_H
+#endif
diff --git a/libgfortran/nca/sync.c b/libgfortran/nca/sync.c
new file mode 100644
index 00000000000..6d7f7caee47
--- /dev/null
+++ b/libgfortran/nca/sync.c
@@ -0,0 +1,156 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include <string.h>
+
+#include "libgfortran.h"
+#include "libcoarraynative.h"
+#include "sync.h"
+#include "util.h"
+
+static void
+sync_all_init (pthread_barrier_t *b)
+{
+ pthread_barrierattr_t battr;
+ pthread_barrierattr_init (&battr);
+ pthread_barrierattr_setpshared (&battr, PTHREAD_PROCESS_SHARED);
+ pthread_barrier_init (b, &battr, local->num_images);
+ pthread_barrierattr_destroy (&battr);
+}
+
+static inline void
+lock_table (sync_iface *si)
+{
+ pthread_mutex_lock (&si->cis->table_lock);
+}
+
+static inline void
+unlock_table (sync_iface *si)
+{
+ pthread_mutex_unlock (&si->cis->table_lock);
+}
+
+static inline void
+wait_table_cond (sync_iface *si, pthread_cond_t *cond)
+{
+ pthread_cond_wait (cond,&si->cis->table_lock);
+}
+
+static int *
+get_locked_table(sync_iface *si) { // The initialization of the table has to
+ // be delayed, since we might not know the
+ // number of images when the library is
+ // initialized
+ lock_table(si);
+ return si->table;
+ /*
+ if (si->table)
+ return si->table;
+ else if (!SHMPTR_IS_NULL(si->cis->table))
+ {
+ si->table = SHMPTR_AS(int *, si->cis->table, si->sm);
+ si->triggers = SHMPTR_AS(pthread_cond_t *, si->cis->triggers, si->sm);
+ return si->table;
+ }
+
+ si->cis->table =
+ shared_malloc(si->a, sizeof(int)*local->num_images * local->num_images);
+ si->cis->triggers =
+ shared_malloc(si->a, sizeof(int)*local->num_images);
+
+ si->table = SHMPTR_AS(int *, si->cis->table, si->sm);
+ si->triggers = SHMPTR_AS(pthread_cond_t *, si->cis->triggers, si->sm);
+
+ for (int i = 0; i < local->num_images; i++)
+ initialize_shared_condition (&si->triggers[i]);
+
+ return si->table;
+ */
+}
+
+void
+sync_iface_init (sync_iface *si, alloc_iface *ai, shared_memory *sm)
+{
+ si->cis = SHMPTR_AS (sync_iface_shared *,
+ shared_malloc (get_allocator(ai),
+ sizeof(collsub_iface_shared)),
+ sm);
+ DEBUG_PRINTF ("%s: num_images is %d\n", __PRETTY_FUNCTION__, local->num_images);
+
+ sync_all_init (&si->cis->sync_all);
+ initialize_shared_mutex (&si->cis->table_lock);
+ si->sm = sm;
+ si->a = get_allocator(ai);
+
+ si->cis->table =
+ shared_malloc(si->a, sizeof(int)*local->num_images * local->num_images);
+ si->cis->triggers =
+ shared_malloc(si->a, sizeof(pthread_cond_t)*local->num_images);
+
+ si->table = SHMPTR_AS(int *, si->cis->table, si->sm);
+ si->triggers = SHMPTR_AS(pthread_cond_t *, si->cis->triggers, si->sm);
+
+ for (int i = 0; i < local->num_images; i++)
+ initialize_shared_condition (&si->triggers[i]);
+}
+
+void
+sync_table (sync_iface *si, int *images, size_t size)
+{
+#ifdef DEBUG_NATIVE_COARRAY
+ dprintf (2, "Image %d waiting for these %ld images: ", this_image.image_num + 1, size);
+ for (int d_i = 0; d_i < size; d_i++)
+ dprintf (2, "%d ", images[d_i]);
+ dprintf (2, "\n");
+#endif
+ size_t i;
+ int done;
+ int *table = get_locked_table(si);
+ for (i = 0; i < size; i++)
+ {
+ table[images[i] - 1 + local->num_images*this_image.image_num]++;
+ pthread_cond_signal (&si->triggers[images[i] - 1]);
+ }
+ for (;;)
+ {
+ done = 1;
+ for (i = 0; i < size; i++)
+ done &= si->table[images[i] - 1 + this_image.image_num*local->num_images]
+ == si->table[this_image.image_num + (images[i] - 1)*local->num_images];
+ if (done)
+ break;
+ wait_table_cond (si, &si->triggers[this_image.image_num]);
+ }
+ unlock_table (si);
+}
+
+void
+sync_all (sync_iface *si)
+{
+
+ DEBUG_PRINTF("Syncing all\n");
+
+ pthread_barrier_wait (&si->cis->sync_all);
+}
diff --git a/libgfortran/nca/sync.h b/libgfortran/nca/sync.h
new file mode 100644
index 00000000000..4b494416d6a
--- /dev/null
+++ b/libgfortran/nca/sync.h
@@ -0,0 +1,56 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#ifndef IPSYNC_HDR
+#define IPSYNC_HDR
+
+#include "shared_memory.h"
+#include "alloc.h"
+#include<pthread.h>
+
+typedef struct {
+ pthread_barrier_t sync_all;
+ pthread_mutex_t table_lock;
+ shared_mem_ptr table;
+ shared_mem_ptr triggers;
+} sync_iface_shared;
+
+typedef struct {
+ sync_iface_shared *cis;
+ shared_memory *sm;
+ allocator *a;
+ int *table; // we can cache the table and the trigger pointers here
+ pthread_cond_t *triggers;
+} sync_iface;
+
+void sync_iface_init (sync_iface *, alloc_iface *, shared_memory *);
+internal_proto (sync_iface_init);
+
+void sync_all (sync_iface *);
+internal_proto (sync_all);
+
+void sync_table (sync_iface *, int *, size_t);
+internal_proto (sync_table);
+
+#endif
diff --git a/libgfortran/nca/util.c b/libgfortran/nca/util.c
new file mode 100644
index 00000000000..5805218f18c
--- /dev/null
+++ b/libgfortran/nca/util.c
@@ -0,0 +1,197 @@
+#include "libgfortran.h"
+#include "util.h"
+#include <string.h>
+#include <stddef.h>
+#include <stdlib.h>
+#include <limits.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+
+#define MEMOBJ_NAME "/gfortran_coarray_memfd"
+
+size_t
+alignto(size_t size, size_t align) {
+ return align*((size + align - 1)/align);
+}
+
+size_t pagesize;
+
+size_t
+round_to_pagesize(size_t s) {
+ return alignto(s, pagesize);
+}
+
+size_t
+next_power_of_two(size_t size) {
+ return 1 << (PTR_BITS - __builtin_clzl(size-1)); //FIXME: There's an off-by-one error, I can feel it
+}
+
+void
+initialize_shared_mutex (pthread_mutex_t *mutex)
+{
+ pthread_mutexattr_t mattr;
+ pthread_mutexattr_init (&mattr);
+ pthread_mutexattr_setpshared (&mattr, PTHREAD_PROCESS_SHARED);
+ pthread_mutex_init (mutex, &mattr);
+ pthread_mutexattr_destroy (&mattr);
+}
+
+void
+initialize_shared_condition (pthread_cond_t *cond)
+{
+ pthread_condattr_t cattr;
+ pthread_condattr_init (&cattr);
+ pthread_condattr_setpshared (&cattr, PTHREAD_PROCESS_SHARED);
+ pthread_cond_init (cond, &cattr);
+ pthread_condattr_destroy (&cattr);
+}
+
+int
+get_shmem_fd (void)
+{
+ char buffer[1<<10];
+ int fd, id;
+ id = random ();
+ do
+ {
+ snprintf (buffer, sizeof (buffer),
+ MEMOBJ_NAME "_%u_%d", (unsigned int) getpid (), id++);
+ fd = shm_open (buffer, O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR);
+ }
+ while (fd == -1);
+ shm_unlink (buffer);
+ return fd;
+}
+
+bool
+pack_array_prepare (pack_info * restrict pi, const gfc_array_char * restrict source)
+{
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type type_size;
+ index_type ssize;
+
+ dim = GFC_DESCRIPTOR_RANK (source);
+ type_size = GFC_DESCRIPTOR_SIZE (source);
+ ssize = type_size;
+
+ pi->num_elem = 1;
+ packed = true;
+ span = source->span != 0 ? source->span : type_size;
+ for (index_type n = 0; n < dim; n++)
+ {
+ pi->stride[n] = GFC_DESCRIPTOR_STRIDE (source,n) * span;
+ pi->extent[n] = GFC_DESCRIPTOR_EXTENT (source,n);
+ if (pi->extent[n] <= 0)
+ {
+ /* Do nothing. */
+ packed = 1;
+ pi->num_elem = 0;
+ break;
+ }
+
+ if (ssize != pi->stride[n])
+ packed = 0;
+
+ pi->num_elem *= pi->extent[n];
+ ssize *= pi->extent[n];
+ }
+
+ return packed;
+}
+
+void
+pack_array_finish (pack_info * const restrict pi, const gfc_array_char * const restrict source,
+ char * restrict dest)
+{
+ index_type dim;
+ const char *restrict src;
+
+ index_type size;
+ index_type stride0;
+ index_type count[GFC_MAX_DIMENSIONS];
+
+ dim = GFC_DESCRIPTOR_RANK (source);
+ src = source->base_addr;
+ stride0 = pi->stride[0];
+ size = GFC_DESCRIPTOR_SIZE (source);
+
+ memset (count, 0, sizeof(count));
+ while (src)
+ {
+ /* Copy the data. */
+ memcpy(dest, src, size);
+ /* Advance to the next element. */
+ dest += size;
+ src += stride0;
+ count[0]++;
+ /* Advance to the next source element. */
+ index_type n = 0;
+ while (count[n] == pi->extent[n])
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ /* We could precalculate these products, but this is a less
+ frequently used path so probably not worth it. */
+ src -= pi->stride[n] * pi->extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += pi->stride[n];
+ }
+ }
+ }
+}
+
+void
+unpack_array_finish (pack_info * const restrict pi,
+ const gfc_array_char * restrict d,
+ const char *restrict src)
+{
+ index_type stride0;
+ char * restrict dest;
+ index_type size;
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type dim;
+
+ size = GFC_DESCRIPTOR_SIZE (d);
+ stride0 = pi->stride[0];
+ dest = d->base_addr;
+ dim = GFC_DESCRIPTOR_RANK (d);
+
+ while (dest)
+ {
+ memcpy (dest, src, size);
+ src += size;
+ dest += stride0;
+ count[0]++;
+ index_type n = 0;
+ while (count[n] == pi->extent[n])
+ {
+ count[n] = 0;
+ dest -= pi->stride[n] * pi->extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n] ++;
+ dest += pi->stride[n];
+ }
+ }
+ }
+}
diff --git a/libgfortran/nca/util.h b/libgfortran/nca/util.h
new file mode 100644
index 00000000000..9abd7adf708
--- /dev/null
+++ b/libgfortran/nca/util.h
@@ -0,0 +1,86 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#ifndef UTIL_HDR
+#define UTIL_HDR
+
+#include <stdint.h>
+#include <stddef.h>
+#include <pthread.h>
+
+#define PTR_BITS (CHAR_BIT*sizeof(void *))
+
+size_t alignto (size_t, size_t);
+internal_proto (alignto);
+
+size_t round_to_pagesize (size_t);
+internal_proto (round_to_pagesize);
+
+size_t next_power_of_two (size_t);
+internal_proto (next_power_of_two);
+
+int get_shmem_fd (void);
+internal_proto (get_shmem_fd);
+
+void initialize_shared_mutex (pthread_mutex_t *);
+internal_proto (initialize_shared_mutex);
+
+void initialize_shared_condition (pthread_cond_t *);
+internal_proto (initialize_shared_condition);
+
+extern size_t pagesize;
+internal_proto (pagesize);
+
+/* Usage:
+ pack_info pi;
+ packed = pack_array_prepare (&pi, source);
+
+ // Awesome allocation of destptr using pi.num_elem
+ if (packed)
+ memcpy (...);
+ else
+ pack_array_finish (&pi, source, destptr);
+
+ This could also be used in in_pack_generic.c. Additionally, since
+ pack_array_prepare is the same for all type sizes, we would only have to
+ specialize pack_array_finish, saving on code size. */
+
+typedef struct
+{
+ index_type num_elem;
+ index_type extent[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS]; /* Stride is byte-based. */
+} pack_info;
+
+bool pack_array_prepare (pack_info *restrict, const gfc_array_char * restrict);
+internal_proto (pack_array_prepare);
+
+void pack_array_finish (pack_info * const restrict, const gfc_array_char * const restrict,
+ char * restrict);
+
+internal_proto (pack_array_finish);
+
+void unpack_array_finish (pack_info * const restrict, const gfc_array_char * const,
+ const char * restrict);
+#endif
diff --git a/libgfortran/nca/wrapper.c b/libgfortran/nca/wrapper.c
new file mode 100644
index 00000000000..eeb64d3aac9
--- /dev/null
+++ b/libgfortran/nca/wrapper.c
@@ -0,0 +1,258 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#include <string.h>
+#include "libgfortran.h"
+#include "libcoarraynative.h"
+#include "sync.h"
+#include "lock.h"
+#include "util.h"
+#include "collective_subroutine.h"
+
+static inline int
+div_ru (int divident, int divisor)
+{
+ return (divident + divisor - 1)/divisor;
+}
+
+enum gfc_coarray_allocation_type {
+ GFC_NCA_NORMAL_COARRAY = 3,
+ GFC_NCA_LOCK_COARRAY,
+ GFC_NCA_EVENT_COARRAY,
+};
+
+void nca_coarray_alloc (gfc_array_void *, int, int, int);
+export_proto (nca_coarray_alloc);
+
+void
+nca_coarray_free (gfc_array_void *, int);
+export_proto (nca_coarray_free);
+
+int nca_coarray_this_image (int);
+export_proto (nca_coarray_this_image);
+
+int nca_coarray_num_images (int);
+export_proto (nca_coarray_num_images);
+
+void nca_coarray_sync_all (int *);
+export_proto (nca_coarray_sync_all);
+
+void nca_sync_images (size_t, int *, int*, char *, size_t);
+export_proto (nca_sync_images);
+
+void nca_lock (void *);
+export_proto (nca_lock);
+
+void nca_unlock (void *);
+export_proto (nca_unlock);
+
+void nca_collsub_reduce_array (gfc_array_char *, void (*) (void *, void *),
+ int *);
+export_proto (nca_collsub_reduce_array);
+
+void nca_collsub_reduce_scalar (void *, index_type, void (*) (void *, void *),
+ int *);
+export_proto (nca_collsub_reduce_scalar);
+
+void nca_collsub_broadcast_array (gfc_array_char * restrict, int/*, int *, char *,
+ size_t*/);
+export_proto (nca_collsub_broadcast_array);
+
+void nca_collsub_broadcast_scalar (void * restrict, size_t, int/*, int *, char *,
+ size_t*/);
+export_proto(nca_collsub_broadcast_scalar);
+
+void
+nca_coarray_alloc (gfc_array_void *desc, int elem_size, int corank,
+ int alloc_type)
+{
+ int i, last_rank_index;
+ int num_coarray_elems, num_elems; /* Excludes the last dimension, because it
+ will have to be determined later. */
+ int extent_last_codimen;
+ size_t last_lbound;
+ size_t size_in_bytes;
+
+ ensure_initialization(); /* This function might be the first one to be
+ called, if it is called in a constructor. */
+
+ if (alloc_type == GFC_NCA_LOCK_COARRAY)
+ elem_size = sizeof (pthread_mutex_t);
+ else if (alloc_type == GFC_NCA_EVENT_COARRAY)
+ elem_size = sizeof(char); /* replace with proper type. */
+
+ last_rank_index = GFC_DESCRIPTOR_RANK(desc) + corank -1;
+
+ num_elems = 1;
+ num_coarray_elems = 1;
+ for (i = 0; i < GFC_DESCRIPTOR_RANK(desc); i++)
+ num_elems *= GFC_DESCRIPTOR_EXTENT(desc, i);
+ for (i = GFC_DESCRIPTOR_RANK(desc); i < last_rank_index; i++)
+ {
+ num_elems *= GFC_DESCRIPTOR_EXTENT(desc, i);
+ num_coarray_elems *= GFC_DESCRIPTOR_EXTENT(desc, i);
+ }
+
+ extent_last_codimen = div_ru (local->num_images, num_coarray_elems);
+
+ last_lbound = GFC_DIMENSION_LBOUND(desc->dim[last_rank_index]);
+ GFC_DIMENSION_SET(desc->dim[last_rank_index], last_lbound,
+ last_lbound + extent_last_codimen - 1,
+ num_elems);
+
+ size_in_bytes = elem_size * num_elems * extent_last_codimen;
+ if (alloc_type == GFC_NCA_LOCK_COARRAY)
+ {
+ lock_array *addr;
+ int expected = 0;
+ /* Allocate enough space for the metadata infront of the lock
+ array. */
+ addr = get_memory_by_id_zero (&local->ai, size_in_bytes
+ + sizeof (lock_array),
+ (intptr_t) desc);
+
+ /* Use of a traditional spin lock to avoid race conditions with
+ the initization of the mutex. We could alternatively put a
+ global lock around allocate, but that would probably be
+ slower. */
+ while (!__atomic_compare_exchange_n (&addr->owner, &expected,
+ this_image.image_num + 1,
+ false, __ATOMIC_SEQ_CST,
+ __ATOMIC_SEQ_CST));
+ if (!addr->initialized++)
+ {
+ for (i = 0; i < local->num_images; i++)
+ initialize_shared_mutex (&addr->arr[i]);
+ }
+ __atomic_store_n (&addr->owner, 0, __ATOMIC_SEQ_CST);
+ desc->base_addr = &addr->arr;
+ }
+ else if (alloc_type == GFC_NCA_EVENT_COARRAY)
+ (void) 0; // TODO
+ else
+ desc->base_addr = get_memory_by_id (&local->ai, size_in_bytes,
+ (intptr_t) desc);
+ dprintf(2, "Base address of desc for image %d: %p\n", this_image.image_num + 1, desc->base_addr);
+}
+
+void
+nca_coarray_free (gfc_array_void *desc, int alloc_type)
+{
+ int i;
+ if (alloc_type == GFC_NCA_LOCK_COARRAY)
+ {
+ lock_array *la;
+ int expected = 0;
+ la = desc->base_addr - offsetof (lock_array, arr);
+ while (!__atomic_compare_exchange_n (&la->owner, &expected,
+ this_image.image_num+1,
+ false, __ATOMIC_SEQ_CST,
+ __ATOMIC_SEQ_CST));
+ if (!--la->initialized)
+ {
+ /* Coarray locks can be removed and just normal
+ pthread_mutex can be used. */
+ for (i = 0; i < local->num_images; i++)
+ pthread_mutex_destroy (&la->arr[i]);
+ }
+ __atomic_store_n (&la->owner, 0, __ATOMIC_SEQ_CST);
+ }
+ else if (alloc_type == GFC_NCA_EVENT_COARRAY)
+ (void) 0; //TODO
+
+ free_memory_with_id (&local->ai, (intptr_t) desc);
+ desc->base_addr = NULL;
+}
+
+int
+nca_coarray_this_image (int distance __attribute__((unused)))
+{
+ return this_image.image_num + 1;
+}
+
+int
+nca_coarray_num_images (int distance __attribute__((unused)))
+{
+ return local->num_images;
+}
+
+void
+nca_coarray_sync_all (int *stat __attribute__((unused)))
+{
+ sync_all (&local->si);
+}
+
+void
+nca_sync_images (size_t s, int *images,
+ int *stat __attribute__((unused)),
+ char *error __attribute__((unused)),
+ size_t err_size __attribute__((unused)))
+{
+ sync_table (&local->si, images, s);
+}
+
+void
+nca_lock (void *lock)
+{
+ pthread_mutex_lock (lock);
+}
+
+void
+nca_unlock (void *lock)
+{
+ pthread_mutex_unlock (lock);
+}
+
+void
+nca_collsub_reduce_array (gfc_array_char *desc, void (*assign_function) (void *, void *),
+ int *result_image)
+{
+ collsub_reduce_array (&local->ci, desc, result_image, assign_function);
+}
+
+void
+nca_collsub_reduce_scalar (void *obj, index_type elem_size,
+ void (*assign_function) (void *, void *),
+ int *result_image)
+{
+ collsub_reduce_scalar (&local->ci, obj, elem_size, result_image, assign_function);
+}
+
+void
+nca_collsub_broadcast_array (gfc_array_char * restrict a, int source_image
+ /* , int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ size_t errmsg_len __attribute__ ((unused))*/)
+{
+ collsub_broadcast_array (&local->ci, a, source_image - 1);
+}
+
+void
+nca_collsub_broadcast_scalar (void * restrict obj, size_t size, int source_image/*,
+ int *stat __attribute__((unused)),
+ char *errmsg __attribute__ ((unused)),
+ size_t errmsg_len __attribute__ ((unused))*/)
+{
+ collsub_broadcast_scalar (&local->ci, obj, size, source_image - 1);
+}
[-- Attachment #3: test.tar.gz --]
[-- Type: application/gzip, Size: 7558 bytes --]
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!)
2020-09-22 12:14 [RFC] Native Coarrays (finally!) Nicolas König
@ 2020-09-23 6:13 ` Damian Rouson
2020-09-30 12:10 ` Andre Vehreschild
` (2 subsequent siblings)
3 siblings, 0 replies; 17+ messages in thread
From: Damian Rouson @ 2020-09-23 6:13 UTC (permalink / raw)
To: Nicolas König; +Cc: gfortran
I very much look forward to trying this out -- preferably after it's
on git a branch. I've often had difficulty applying patches and would
much prefer checking out a git branch.
As I mentioned on Slack, it would be great if the flag could be
something like -fcoarray=shared instead of -fcoarray=native. The
-fcoarray=shared approach is closer to what the Intel compiler does
for shared-memory compilation and it's in keeping with the spirit of
what the NAG compiler does.
Also, I think the following statement is incorrect: "Since coarrays
always behave as if they had the SAVE attribute, this works even for
allocatable coarrays." Constraint C826 in the Fortran 2018 standard
states, "A coarray or an object with a coarray ultimate component
shall be an associate name, a dummy argument, or have the ALLOCATABLE
or SAVE attribute." This reads to me as SAVE being just one of
several options. If coarrays automatically had the SAVE attribute,
then there would be less need for the automatic synchronization that
happens when an allocatable coarrays gets allocated or deallocated.
Damian
On Tue, Sep 22, 2020 at 4:17 AM Nicolas König <koenigni@student.ethz.ch> wrote:
>
> Hello everyone,
>
> Contrary to rumors, I'm not dead, but I have been working silently on
> the Native Coarray patch. Now, a few rewrites later, I think the patch
> is now in an acceptable state to get a few comments on it and to push
> it on a development branch. My hope is that after fixing a few of the
> more obvious bugs and potential portability problems this might still
> make it as an experimental feature into GCC 11 (this patch doesn't
> disturb the compiler much, almost everything that is done is behind if
> (flag_coarray == FCOARRAY_NATIVE) guards)
>
>
> Supported Features:
> - coarrays (duh), both with basic types and user-defined types
> - Dynamic allocation and deallocation
> - Arrays of Locks
> - sync images/all
> - Passing coarrays as arguments
> - collective subroutines (co_broadcast/co_reduce/co_sum/...)
> - critical sections
> - this_image/num_images/lcobound/...
>
> Missing Features
> - coarrays of classes
> - events
> - stat & errmsg
> - atomics
> - proper handling of random_init
> - coshape
> - coarrays in data statements
> - sync memory
> - team
>
>
> A few words on how these native coarrays work:
>
> Each image is its own process, that is forked from the master process at
> the start of the program. The number of images is determined by the
> environment variable GFORTRAN_NUM_IMAGES or, alternatively, the number
> of processors.
>
> Each coarray is identified by its address. Since coarrays always behave
> as if they had the SAVE attribute, this works even for allocatable
> coarrays. ASLR is not an issue, since the addresses are assigned at
> startup and remain valid over forks. If, on two different images, the
> allocation function is called with the same descriptor address, the same
> piece of memory is allocated.
>
> Internally, the allocator (alloc.c) uses a shared hashmap (hashmap.c) to
> remember with which ids pieces of memory allocated. If a new piece of
> memory is needed, a simple relatively allocator (allocator.c) is used.
> If the allocator doesn't hold any previously free()d memory, it requests
> it from the shared memory object (shared_memory.c), which also handles
> the translation of shared_mem_ptr's to pointers in the address space of
> the image. At the moment shared_memory relies on double-mapping pages
> for this (which might restrict the architectures on which this will
> work, I have tested this on x86 and POWER), but since any piece of
> memory should only be written to through one address within one
> alloc/free pair, it shouldn't matter that much performance-wise.
>
> The entry points in the library with the exception of master are defined
> in wrapper.c, master(), the function handling launching the images, is
> defined in coarraynative.c, and the other files shouldn't require much
> explanation.
>
>
> To compile a program to run with native coarrays, compile with
> -fcoarray=native -lgfor_nca -lrt (I've not yet figured out how to
> automagically link against the library).
>
>
> It should be pointed out that Thomas greatly helped me with this
> project, both with advice and actual code.
>
> With the realization that this might have been a bit large of project
> for a beginner remains
>
> Nicolas
>
> P.S.: Because I don't trust git as far as I can throw it, I've also
> attached the patch and a copy of a few of my tests.
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!)
2020-09-22 12:14 [RFC] Native Coarrays (finally!) Nicolas König
2020-09-23 6:13 ` Damian Rouson
@ 2020-09-30 12:10 ` Andre Vehreschild
2020-09-30 14:09 ` Nicolas König
2020-09-30 13:46 ` Paul Richard Thomas
2020-10-05 9:54 ` Tobias Burnus
3 siblings, 1 reply; 17+ messages in thread
From: Andre Vehreschild @ 2020-09-30 12:10 UTC (permalink / raw)
To: Nicolas König; +Cc: fortran, Damian Rouson
Hi Nicolas,
first of all thank you for the work you spent on the patch. It's a huge amount
of code and will take good time to review. This brings me to the main issue:
The patch is too big to grasp (at least for me). Can you split the huge patch
into logical parts? Say one parts takes care about parameter handling, one
about the setup of the new library, one about handling allocatables in gfortran
itself. This would greatly improve the speed in review, because one/I will be
able to concentrate on one thing at a time. You know best which pieces belong
together.
@Damian: I have added a users branch to gcc.gnu.org's git to test this. To get
it use:
git fetch origin refs/users/vehre/heads/nikkoenig_caf_shared
git checkout users/vehre/nikkoenig_caf_shared
may be you need to strip the "users/" from the last line. And yes, I am aware
that I miss-typed the abbreviation of Nicolas. At the moment correcting this is
hard, because there is some mess up in the commit-hooks on the repo not
allowing the deletion of my own branch.
I had to fix the configure.ac in libgfortran/ where -pthreads is analyzed. I
assume that you only copied it from libgo/configure.ac. On my x86_64 at
configuration time no libatomic/.libs will be present, which does not help
there, but instead bugs out the compiler not being able to access the directory
(which it does not need). So this is the first small remark.
Anyway, what are you using pthread_mutexes for when doing process
parallelization? How do you ensure, that not two locks are initialized
concurrently? Can elaborate how synchronisation on memory allocation is done?
Next I have to second Damian in the naming of the option -fcoarray=native
option. I also prefer to name this something like "shared". Given that you are
using processes for parallelization I propose to use "proc_shared", "pshared"
or the like to distinguish it once someone adds a similar approach using
threads.
Furthermore, there is a naming scheme for coarray runtime libraries. They all
start with "caf_". Therefore "caf_proc_shared" would be natural for the name of
your support library, at least to me.
After adapting the configure.ac I was at least able to compile gfortran with it
on x86_64-linux-gnu/f31. I try to continue to test it and am looking forward to
your answer.
Regards,
Andre
On Tue, 22 Sep 2020 14:14:35 +0200
Nicolas König <koenigni@student.ethz.ch> wrote:
> Hello everyone,
>
> Contrary to rumors, I'm not dead, but I have been working silently on
> the Native Coarray patch. Now, a few rewrites later, I think the patch
> is now in an acceptable state to get a few comments on it and to push
> it on a development branch. My hope is that after fixing a few of the
> more obvious bugs and potential portability problems this might still
> make it as an experimental feature into GCC 11 (this patch doesn't
> disturb the compiler much, almost everything that is done is behind if
> (flag_coarray == FCOARRAY_NATIVE) guards)
>
>
> Supported Features:
> - coarrays (duh), both with basic types and user-defined types
> - Dynamic allocation and deallocation
> - Arrays of Locks
> - sync images/all
> - Passing coarrays as arguments
> - collective subroutines (co_broadcast/co_reduce/co_sum/...)
> - critical sections
> - this_image/num_images/lcobound/...
>
> Missing Features
> - coarrays of classes
> - events
> - stat & errmsg
> - atomics
> - proper handling of random_init
> - coshape
> - coarrays in data statements
> - sync memory
> - team
>
>
> A few words on how these native coarrays work:
>
> Each image is its own process, that is forked from the master process at
> the start of the program. The number of images is determined by the
> environment variable GFORTRAN_NUM_IMAGES or, alternatively, the number
> of processors.
>
> Each coarray is identified by its address. Since coarrays always behave
> as if they had the SAVE attribute, this works even for allocatable
> coarrays. ASLR is not an issue, since the addresses are assigned at
> startup and remain valid over forks. If, on two different images, the
> allocation function is called with the same descriptor address, the same
> piece of memory is allocated.
>
> Internally, the allocator (alloc.c) uses a shared hashmap (hashmap.c) to
> remember with which ids pieces of memory allocated. If a new piece of
> memory is needed, a simple relatively allocator (allocator.c) is used.
> If the allocator doesn't hold any previously free()d memory, it requests
> it from the shared memory object (shared_memory.c), which also handles
> the translation of shared_mem_ptr's to pointers in the address space of
> the image. At the moment shared_memory relies on double-mapping pages
> for this (which might restrict the architectures on which this will
> work, I have tested this on x86 and POWER), but since any piece of
> memory should only be written to through one address within one
> alloc/free pair, it shouldn't matter that much performance-wise.
>
> The entry points in the library with the exception of master are defined
> in wrapper.c, master(), the function handling launching the images, is
> defined in coarraynative.c, and the other files shouldn't require much
> explanation.
>
>
> To compile a program to run with native coarrays, compile with
> -fcoarray=native -lgfor_nca -lrt (I've not yet figured out how to
> automagically link against the library).
>
>
> It should be pointed out that Thomas greatly helped me with this
> project, both with advice and actual code.
>
> With the realization that this might have been a bit large of project
> for a beginner remains
>
> Nicolas
>
> P.S.: Because I don't trust git as far as I can throw it, I've also
> attached the patch and a copy of a few of my tests.
--
Andre Vehreschild * Email: vehre ad gmx dot de
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [RFC] Native Coarrays (finally!)
2020-09-30 12:10 ` Andre Vehreschild
@ 2020-09-30 14:09 ` Nicolas König
2020-09-30 14:11 ` Andre Vehreschild
0 siblings, 1 reply; 17+ messages in thread
From: Nicolas König @ 2020-09-30 14:09 UTC (permalink / raw)
To: Andre Vehreschild; +Cc: fortran, Damian Rouson
Hello Andre,
On 30/09/2020 14:10, Andre Vehreschild wrote:
> Hi Nicolas,
>
> first of all thank you for the work you spent on the patch. It's a huge amount
> of code and will take good time to review. This brings me to the main issue:
> The patch is too big to grasp (at least for me). Can you split the huge patch
> into logical parts? Say one parts takes care about parameter handling, one
> about the setup of the new library, one about handling allocatables in gfortran
> itself. This would greatly improve the speed in review, because one/I will be
> able to concentrate on one thing at a time. You know best which pieces belong
> together.
I will do that, though I first need to fix a few things before formally
submitting it (like the build system problems mentioned below). Then
I'll try my best, but I don't think the library can really be split up
without creating a non-building commit. The compiler part can be easily
separated, though.
>
> @Damian: I have added a users branch to gcc.gnu.org's git to test this. To get
> it use:
>
> git fetch origin refs/users/vehre/heads/nikkoenig_caf_shared
> git checkout users/vehre/nikkoenig_caf_shared
It is now also a development branch: devel/coarray_native. There was
some serious fighting with git involved, but in the end it worked.
>
> may be you need to strip the "users/" from the last line. And yes, I am aware
> that I miss-typed the abbreviation of Nicolas. At the moment correcting this is
> hard, because there is some mess up in the commit-hooks on the repo not
> allowing the deletion of my own branch.
>
> I had to fix the configure.ac in libgfortran/ where -pthreads is analyzed. I
> assume that you only copied it from libgo/configure.ac. On my x86_64 at
> configuration time no libatomic/.libs will be present, which does not help
> there, but instead bugs out the compiler not being able to access the directory
> (which it does not need). So this is the first small remark.
I'll look into that. I have to admit, I find the gcc build system to be
very hard to penetrate, so this may take a few iterations. It would be
great if you could send me your fix,
>
> Anyway, what are you using pthread_mutexes for when doing process
> parallelization?
As it turns out, they have a SET_PROCESS_SHARED attribute that allows
you to use them for synchronization between processes if they lie within
a shared memory region (see util.c:initialize_shared_mutex()). They are
a very fast and convenient --the library is using shared memory anyway--
method to synchronize.
> How do you ensure, that not two locks are initialized concurrently?
Most of the locks are initialized before the images are created.
> Can elaborate how synchronisation on memory allocation is done?
At the moment, the allocator itself is big-locked. Otherwise, we don't
need any synchronization for allocation, because, when a new coarray is
allocated, it checks whether there already exists an allocation with the
same ID, and if it does, then it returns the memory allocated for it and
increments the refcount.
>
> Next I have to second Damian in the naming of the option -fcoarray=native
> option. I also prefer to name this something like "shared". Given that you are
> using processes for parallelization I propose to use "proc_shared", "pshared"
> or the like to distinguish it once someone adds a similar approach using
> threads.
I think I prefer "shared", "pshared" sounds like it involves pthreads
and proc_shared is rather long and unnecessarily detailed. At the end of
the day, the user doesn't really care about how it is implemented. Also,
Damian said that's what it is called by other compilers that use some
kind of shared memory implementation. Mind you, the difference between
this and any potential thread-based implementation aren't large, so I'm
not sure that anyone will ever try that.
>
> Furthermore, there is a naming scheme for coarray runtime libraries. They all
> start with "caf_". Therefore "caf_proc_shared" would be natural for the name of
> your support library, at least to me.
All two of them :P Even though the name might be slightly confusing,
since the library doesn't use caf_* as a prefix for its functions, I
would then probably suggest caf_shared_mem or something similiar. That
it uses shared memory is, after all, its defining quality.
>
> After adapting the configure.ac I was at least able to compile gfortran with it
> on x86_64-linux-gnu/f31. I try to continue to test it and am looking forward to
> your answer.
Good to hear the build and some of the basics work. I have only tested
it on POWER up till now.
Kind regards
Nicolas
PS: One thing I forgot to mention is that the use of processes means the
coarrays don't interfere with OpenMP, so coarray code with that can be
locally tested without install MPI. This (together with the pain of
making all global variables thread-local and its corresponding
performance hit) is the main reason for using processes instead of threads.
>
> Regards,
> Andre
>
> On Tue, 22 Sep 2020 14:14:35 +0200
> Nicolas König <koenigni@student.ethz.ch> wrote:
>
>> Hello everyone,
>>
>> Contrary to rumors, I'm not dead, but I have been working silently on
>> the Native Coarray patch. Now, a few rewrites later, I think the patch
>> is now in an acceptable state to get a few comments on it and to push
>> it on a development branch. My hope is that after fixing a few of the
>> more obvious bugs and potential portability problems this might still
>> make it as an experimental feature into GCC 11 (this patch doesn't
>> disturb the compiler much, almost everything that is done is behind if
>> (flag_coarray == FCOARRAY_NATIVE) guards)
>>
>>
>> Supported Features:
>> - coarrays (duh), both with basic types and user-defined types
>> - Dynamic allocation and deallocation
>> - Arrays of Locks
>> - sync images/all
>> - Passing coarrays as arguments
>> - collective subroutines (co_broadcast/co_reduce/co_sum/...)
>> - critical sections
>> - this_image/num_images/lcobound/...
>>
>> Missing Features
>> - coarrays of classes
>> - events
>> - stat & errmsg
>> - atomics
>> - proper handling of random_init
>> - coshape
>> - coarrays in data statements
>> - sync memory
>> - team
>>
>>
>> A few words on how these native coarrays work:
>>
>> Each image is its own process, that is forked from the master process at
>> the start of the program. The number of images is determined by the
>> environment variable GFORTRAN_NUM_IMAGES or, alternatively, the number
>> of processors.
>>
>> Each coarray is identified by its address. Since coarrays always behave
>> as if they had the SAVE attribute, this works even for allocatable
>> coarrays. ASLR is not an issue, since the addresses are assigned at
>> startup and remain valid over forks. If, on two different images, the
>> allocation function is called with the same descriptor address, the same
>> piece of memory is allocated.
>>
>> Internally, the allocator (alloc.c) uses a shared hashmap (hashmap.c) to
>> remember with which ids pieces of memory allocated. If a new piece of
>> memory is needed, a simple relatively allocator (allocator.c) is used.
>> If the allocator doesn't hold any previously free()d memory, it requests
>> it from the shared memory object (shared_memory.c), which also handles
>> the translation of shared_mem_ptr's to pointers in the address space of
>> the image. At the moment shared_memory relies on double-mapping pages
>> for this (which might restrict the architectures on which this will
>> work, I have tested this on x86 and POWER), but since any piece of
>> memory should only be written to through one address within one
>> alloc/free pair, it shouldn't matter that much performance-wise.
>>
>> The entry points in the library with the exception of master are defined
>> in wrapper.c, master(), the function handling launching the images, is
>> defined in coarraynative.c, and the other files shouldn't require much
>> explanation.
>>
>>
>> To compile a program to run with native coarrays, compile with
>> -fcoarray=native -lgfor_nca -lrt (I've not yet figured out how to
>> automagically link against the library).
>>
>>
>> It should be pointed out that Thomas greatly helped me with this
>> project, both with advice and actual code.
>>
>> With the realization that this might have been a bit large of project
>> for a beginner remains
>>
>> Nicolas
>>
>> P.S.: Because I don't trust git as far as I can throw it, I've also
>> attached the patch and a copy of a few of my tests.
>
>
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!)
2020-09-30 14:09 ` Nicolas König
@ 2020-09-30 14:11 ` Andre Vehreschild
0 siblings, 0 replies; 17+ messages in thread
From: Andre Vehreschild @ 2020-09-30 14:11 UTC (permalink / raw)
To: Nicolas König; +Cc: fortran, Damian Rouson
[-- Attachment #1: Type: text/plain, Size: 1595 bytes --]
Hi Nicolas,
<snipp>
> > git fetch origin refs/users/vehre/heads/nikkoenig_caf_shared
> > git checkout users/vehre/nikkoenig_caf_shared
>
> It is now also a development branch: devel/coarray_native. There was
> some serious fighting with git involved, but in the end it worked.
Great.
<snipp>
> > I had to fix the configure.ac in libgfortran/ where -pthreads is analyzed. I
> > assume that you only copied it from libgo/configure.ac. On my x86_64 at
> > configuration time no libatomic/.libs will be present, which does not help
> > there, but instead bugs out the compiler not being able to access the
> > directory (which it does not need). So this is the first small remark.
>
> I'll look into that. I have to admit, I find the gcc build system to be
> very hard to penetrate, so this may take a few iterations. It would be
> great if you could send me your fix,
Well, the build system of gcc is grown over the time (not to call it ancient
:-)) and has great flexibility. But with great flexibility comes great
complexity. That's what we have to deal with.
The patch I had to do is attached. Note the patch is also employing all
coarray tests of the testsuite to check your "native" approach. If you don't
want that remove the first chunk.
Anyway from the (few) testcases in the coarray directory about 30 fail with
internal compiler errors and at least one does not terminate. I have had the
time to figure which one.
I will come back to this after my holidays next week.
Regards,
Andre
--
Andre Vehreschild * Email: vehre ad gmx dot de
[-- Attachment #2: build.patch --]
[-- Type: text/x-patch, Size: 2370 bytes --]
diff --git a/gcc/testsuite/gfortran.dg/coarray/caf.exp b/gcc/testsuite/gfortran.dg/coarray/caf.exp
index cd2a7ed2e45..2ca91f7ee15 100644
--- a/gcc/testsuite/gfortran.dg/coarray/caf.exp
+++ b/gcc/testsuite/gfortran.dg/coarray/caf.exp
@@ -107,6 +107,13 @@ foreach test [lsort [glob -nocomplain $srcdir/$subdir/*.\[fF\]{,90,95,03,08} ]]
dg-test $test "-fcoarray=lib $flags -lcaf_single $maybe_atomic_lib" ""
cleanup-modules ""
}
+
+ foreach flags $option_list {
+ verbose "Testing $nshort (native), $flags" 1
+ set gfortran_aux_module_flags "-fcoarray=native $flags -lgfor_nca -lrt"
+ dg-test $test "-fcoarray=native $flags -lgfor_nca -lrt $maybe_atomic_lib" ""
+ cleanup-modules ""
+ }
}
torture-finish
dg-finish
diff --git a/libgfortran/configure b/libgfortran/configure
index b2faede3aa6..2c00593c567 100755
--- a/libgfortran/configure
+++ b/libgfortran/configure
@@ -27160,7 +27160,7 @@ if ${libgfortran_cv_lib_pthread+:} false; then :
$as_echo_n "(cached) " >&6
else
CFLAGS_hold=$CFLAGS
-CFLAGS="$CFLAGS -pthread -L../libatomic/.libs"
+CFLAGS="$CFLAGS -pthread"
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
/* end confdefs.h. */
int i;
@@ -27176,7 +27176,7 @@ fi
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libgfortran_cv_lib_pthread" >&5
$as_echo "$libgfortran_cv_lib_pthread" >&6; }
PTHREAD_CFLAGS=
-if test "$libgfortran_cv_lib_pthread" = yes; then
+if test "$libgfortran_cv_lib_pthread" = "yes" -a "$cpu_type" = "riscv"; then
# RISC-V apparently adds -latomic when using -pthread.
PTHREAD_CFLAGS="-pthread -L../libatomic/.libs"
fi
diff --git a/libgfortran/configure.ac b/libgfortran/configure.ac
index 1aac140c905..6339e5028ff 100644
--- a/libgfortran/configure.ac
+++ b/libgfortran/configure.ac
@@ -688,13 +688,13 @@ LIBGFOR_CHECK_AVX128
AC_CACHE_CHECK([whether -pthread is supported],
[libgfortran_cv_lib_pthread],
[CFLAGS_hold=$CFLAGS
-CFLAGS="$CFLAGS -pthread -L../libatomic/.libs"
+CFLAGS="$CFLAGS -pthread"
AC_COMPILE_IFELSE([AC_LANG_SOURCE([int i;])],
[libgfortran_cv_lib_pthread=yes],
[libgfortran_cv_lib_pthread=no])
CFLAGS=$CFLAGS_hold])
PTHREAD_CFLAGS=
-if test "$libgfortran_cv_lib_pthread" = yes; then
+if test "$libgfortran_cv_lib_pthread" = "yes" -a "$cpu_type" = "riscv"; then
# RISC-V apparently adds -latomic when using -pthread.
PTHREAD_CFLAGS="-pthread -L../libatomic/.libs"
fi
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [RFC] Native Coarrays (finally!)
2020-09-22 12:14 [RFC] Native Coarrays (finally!) Nicolas König
2020-09-23 6:13 ` Damian Rouson
2020-09-30 12:10 ` Andre Vehreschild
@ 2020-09-30 13:46 ` Paul Richard Thomas
2020-10-05 9:54 ` Tobias Burnus
3 siblings, 0 replies; 17+ messages in thread
From: Paul Richard Thomas @ 2020-09-30 13:46 UTC (permalink / raw)
To: Nicolas König; +Cc: fortran
Hi Nicolas,
I just pushed the patch for PR97045 to master so I thought that I would
give native/shared co-arrays a try on my clean tree. The git apply bombed
completely:
[pault@pc30 gcc]$ git apply < ~/prs/nca/2*.diff
<stdin>:403: trailing whitespace.
<stdin>:424: trailing whitespace.
<stdin>:449: trailing whitespace.
<stdin>:916: indent with spaces.
== DIMEN_THIS_IMAGE
<stdin>:995: trailing whitespace.
#endif
error: patch failed: gcc/flag-types.h:346
error: gcc/flag-types.h: patch does not apply
error: patch failed: gcc/fortran/dump-parse-tree.c:1060
error: gcc/fortran/dump-parse-tree.c: patch does not apply
error: patch failed: gcc/fortran/frontend-passes.c:57
error: gcc/fortran/frontend-passes.c: patch does not apply
error: patch failed: gcc/fortran/gfortran.h:1974
error: gcc/fortran/gfortran.h: patch does not apply
etc., etc.
I have done a hard reset and will try again another time.
As Andre said, that's an impressive amount of code :-)
Perhaps a good way to overcome with the problem of review is that the
reviewer check that every change in the frontend is only activated by the
-fcoarray=native (or whatever)? If the patch is hidden away on a
development branch, it will not be exercised as quickly and the bugs shaken
out.
Cheers
Paul
On Tue, 22 Sep 2020 at 12:18, Nicolas König <koenigni@student.ethz.ch>
wrote:
> Hello everyone,
>
> Contrary to rumors, I'm not dead, but I have been working silently on
> the Native Coarray patch. Now, a few rewrites later, I think the patch
> is now in an acceptable state to get a few comments on it and to push
> it on a development branch. My hope is that after fixing a few of the
> more obvious bugs and potential portability problems this might still
> make it as an experimental feature into GCC 11 (this patch doesn't
> disturb the compiler much, almost everything that is done is behind if
> (flag_coarray == FCOARRAY_NATIVE) guards)
>
>
> Supported Features:
> - coarrays (duh), both with basic types and user-defined types
> - Dynamic allocation and deallocation
> - Arrays of Locks
> - sync images/all
> - Passing coarrays as arguments
> - collective subroutines (co_broadcast/co_reduce/co_sum/...)
> - critical sections
> - this_image/num_images/lcobound/...
>
> Missing Features
> - coarrays of classes
> - events
> - stat & errmsg
> - atomics
> - proper handling of random_init
> - coshape
> - coarrays in data statements
> - sync memory
> - team
>
>
> A few words on how these native coarrays work:
>
> Each image is its own process, that is forked from the master process at
> the start of the program. The number of images is determined by the
> environment variable GFORTRAN_NUM_IMAGES or, alternatively, the number
> of processors.
>
> Each coarray is identified by its address. Since coarrays always behave
> as if they had the SAVE attribute, this works even for allocatable
> coarrays. ASLR is not an issue, since the addresses are assigned at
> startup and remain valid over forks. If, on two different images, the
> allocation function is called with the same descriptor address, the same
> piece of memory is allocated.
>
> Internally, the allocator (alloc.c) uses a shared hashmap (hashmap.c) to
> remember with which ids pieces of memory allocated. If a new piece of
> memory is needed, a simple relatively allocator (allocator.c) is used.
> If the allocator doesn't hold any previously free()d memory, it requests
> it from the shared memory object (shared_memory.c), which also handles
> the translation of shared_mem_ptr's to pointers in the address space of
> the image. At the moment shared_memory relies on double-mapping pages
> for this (which might restrict the architectures on which this will
> work, I have tested this on x86 and POWER), but since any piece of
> memory should only be written to through one address within one
> alloc/free pair, it shouldn't matter that much performance-wise.
>
> The entry points in the library with the exception of master are defined
> in wrapper.c, master(), the function handling launching the images, is
> defined in coarraynative.c, and the other files shouldn't require much
> explanation.
>
>
> To compile a program to run with native coarrays, compile with
> -fcoarray=native -lgfor_nca -lrt (I've not yet figured out how to
> automagically link against the library).
>
>
> It should be pointed out that Thomas greatly helped me with this
> project, both with advice and actual code.
>
> With the realization that this might have been a bit large of project
> for a beginner remains
>
> Nicolas
>
> P.S.: Because I don't trust git as far as I can throw it, I've also
> attached the patch and a copy of a few of my tests.
>
--
"If you can't explain it simply, you don't understand it well enough" -
Albert Einstein
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!)
2020-09-22 12:14 [RFC] Native Coarrays (finally!) Nicolas König
` (2 preceding siblings ...)
2020-09-30 13:46 ` Paul Richard Thomas
@ 2020-10-05 9:54 ` Tobias Burnus
2020-10-05 13:23 ` Nicolas König
3 siblings, 1 reply; 17+ messages in thread
From: Tobias Burnus @ 2020-10-05 9:54 UTC (permalink / raw)
To: Nicolas König, fortran
Hi Nicolas,
admittedly, I have not yet looked at your patch. However, I have to
admit that I do not like the name. I understand that "native" refers
to not needing an external library (libcaf.../libopencoarray...),
but I still wonder whether something like "-fcoarray=shared" (i.e.
working on a shared-memory system) would be better name from an end-user
point of view.
Tobias,
who likes that coarray can be used without extra libs and thinks
that this will help with users starting to use coarrays.
-----------------
Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München / Germany
Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung, Alexander Walter
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!)
2020-10-05 9:54 ` Tobias Burnus
@ 2020-10-05 13:23 ` Nicolas König
2020-10-12 12:32 ` [RFC] Native Coarrays (finally!) [Review part 1] Andre Vehreschild
` (2 more replies)
0 siblings, 3 replies; 17+ messages in thread
From: Nicolas König @ 2020-10-05 13:23 UTC (permalink / raw)
To: Tobias Burnus; +Cc: fortran
Hello Tobias,
On 05/10/2020 11:54, Tobias Burnus wrote:
> Hi Nicolas,
>
> admittedly, I have not yet looked at your patch. However, I have to
> admit that I do not like the name. I understand that "native" refers
> to not needing an external library (libcaf.../libopencoarray...),
> but I still wonder whether something like "-fcoarray=shared" (i.e.
> working on a shared-memory system) would be better name from an end-user
> point of view.
I think the name has been the most critized point of the entire patch up
till now. I'm going to change it to -fcoarray=shared, as you (and a few
other people) suggested :)
>
> Tobias,
> who likes that coarray can be used without extra libs and thinks
> that this will help with users starting to use coarrays.
That is the main reason I wrote the patch.
>
> -----------------
> Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München /
> Germany
> Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung,
> Alexander Walter
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!) [Review part 1]
2020-10-05 13:23 ` Nicolas König
@ 2020-10-12 12:32 ` Andre Vehreschild
2020-10-12 13:48 ` [RFC] Native Coarrays (finally!) [Review part 2] Andre Vehreschild
2020-10-13 13:01 ` [RFC] Native Coarrays (finally!) [Review part 3] Andre Vehreschild
2 siblings, 0 replies; 17+ messages in thread
From: Andre Vehreschild @ 2020-10-12 12:32 UTC (permalink / raw)
To: Nicolas König; +Cc: fortran
[-- Attachment #1: Type: text/plain, Size: 1875 bytes --]
Hi Nicolas,
I had the time to start a review. The base I am using is on your
devel/coarray_native branch the commits between these hashes:
git diff 9044db88d634c631920eaa9f66c0275adf18fdf5.. \
b96fdc7b84eb288dea0c3e99a212e6483007a35a
Attached is the first part of my comments. I have prefixed each comment line
with '###AV: ' for you to easily search for them. I would have loved to use
some review tool like gitlab/github or gerrit, but gcc.gnu.org needs to do
everything more 'sophisticated' then common practice. Therefore I had to
recline to do commenting text.
Hopefully more to come.
Regards,
Andre
On Mon, 5 Oct 2020 15:23:27 +0200
Nicolas König <koenigni@student.ethz.ch> wrote:
> Hello Tobias,
>
> On 05/10/2020 11:54, Tobias Burnus wrote:
> > Hi Nicolas,
> >
> > admittedly, I have not yet looked at your patch. However, I have to
> > admit that I do not like the name. I understand that "native" refers
> > to not needing an external library (libcaf.../libopencoarray...),
> > but I still wonder whether something like "-fcoarray=shared" (i.e.
> > working on a shared-memory system) would be better name from an end-user
> > point of view.
>
> I think the name has been the most critized point of the entire patch up
> till now. I'm going to change it to -fcoarray=shared, as you (and a few
> other people) suggested :)
>
> >
> > Tobias,
> > who likes that coarray can be used without extra libs and thinks
> > that this will help with users starting to use coarrays.
>
> That is the main reason I wrote the patch.
>
> >
> > -----------------
> > Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München /
> > Germany
> > Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung,
> > Alexander Walter
--
Andre Vehreschild * Email: vehre ad gmx dot de
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: caf_shared_part_1.patch --]
[-- Type: text/x-patch, Size: 17605 bytes --]
diff --git a/gcc/flag-types.h b/gcc/flag-types.h
index 852ea76eaa2..51e698db139 100644
--- a/gcc/flag-types.h
+++ b/gcc/flag-types.h
@@ -346,7 +346,8 @@ enum gfc_fcoarray
{
GFC_FCOARRAY_NONE = 0,
GFC_FCOARRAY_SINGLE,
- GFC_FCOARRAY_LIB
+ GFC_FCOARRAY_LIB,
+ GFC_FCOARRAY_NATIVE
###AV: I'd prefer this to be GFC_FCOARRAY_SHARED, too.
};
diff --git a/gcc/fortran/dump-parse-tree.c b/gcc/fortran/dump-parse-tree.c
index 6e265f4520d..acff75a901f 100644
--- a/gcc/fortran/dump-parse-tree.c
+++ b/gcc/fortran/dump-parse-tree.c
@@ -1060,7 +1060,7 @@ show_symbol (gfc_symbol *sym)
if (sym == NULL)
return;
- fprintf (dumpfile, "|| symbol: '%s' ", sym->name);
+ fprintf (dumpfile, "|| symbol: '%s' %p ", sym->name, (void *) &(sym->backend_decl));
###AV: Isn't this a left-over from debugging and should be removed?
len = strlen (sym->name);
for (i=len; i<12; i++)
fputc(' ', dumpfile);
diff --git a/gcc/fortran/frontend-passes.c b/gcc/fortran/frontend-passes.c
index 83f6fd804b1..c573731a7ce 100644
--- a/gcc/fortran/frontend-passes.c
+++ b/gcc/fortran/frontend-passes.c
@@ -57,6 +57,7 @@ static int call_external_blas (gfc_code **, int *, void *);
static int matmul_temp_args (gfc_code **, int *,void *data);
static int index_interchange (gfc_code **, int*, void *);
static bool is_fe_temp (gfc_expr *e);
+static void rewrite_co_reduce (gfc_namespace *);
#ifdef CHECKING_P
static void check_locus (gfc_namespace *);
@@ -179,6 +180,9 @@ gfc_run_passes (gfc_namespace *ns)
if (flag_realloc_lhs)
realloc_strings (ns);
+
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ rewrite_co_reduce (ns);
}
#ifdef CHECKING_P
@@ -5895,3 +5899,121 @@ gfc_fix_implicit_pure (gfc_namespace *ns)
return changed;
}
+
+/* Callback function. Create a wrapper around VALUE functions. */
+
+static int
+co_reduce_code (gfc_code **c, int *walk_subtrees ATTRIBUTE_UNUSED, void *data)
+{
+ gfc_code *co = *c;
+ gfc_expr *oper;
+ gfc_symbol *op_sym;
+ gfc_symbol *arg1, *arg2;
+ gfc_namespace *parent_ns;
+ gfc_namespace *proc_ns;
+ gfc_symbol *proc_sym;
+ gfc_symtree *f1t, *f2t;
+ gfc_symbol *f1, *f2;
+ gfc_code *assign;
+ gfc_expr *e1, *e2;
+ char name[GFC_MAX_SYMBOL_LEN + 1];
+ static int num;
###AV: Better intialize num = 0 or 1 just to make sure a reasonable value is set.
+
+ if (co->op != EXEC_CALL || co->resolved_isym == NULL
+ || co->resolved_isym->id != GFC_ISYM_CO_REDUCE)
+ return 0;
+
+ oper = co->ext.actual->next->expr;
+ op_sym = oper->symtree->n.sym;
+ arg1 = op_sym->formal->sym;
+ arg2 = op_sym->formal->next->sym;
+
+ parent_ns = (gfc_namespace *) data;
+
+ /* Generate the wrapper around the function. */
+ proc_ns = gfc_get_namespace (parent_ns, 0);
+ snprintf (name, GFC_MAX_SYMBOL_LEN, "__coreduce_%d_%s", num++, op_sym->name);
###AV: When you init num with -1 above (and comment about it), you can use ++num
###AV: here preventing the use of a unnecessary temporary.
+ gfc_get_symbol (name, proc_ns, &proc_sym);
+ proc_sym->attr.flavor = FL_PROCEDURE;
+ proc_sym->attr.subroutine = 1;
+ proc_sym->attr.referenced = 1;
+ proc_sym->attr.access = ACCESS_PRIVATE;
+ gfc_commit_symbol (proc_sym);
+ proc_ns->proc_name = proc_sym;
+
+ /* Make up the formal arguments. */
+ gfc_get_sym_tree (arg1->name, proc_ns, &f1t, false);
+ f1 = f1t->n.sym;
+ f1->ts = arg1->ts;
+ f1->attr.flavor = FL_VARIABLE;
+ f1->attr.dummy = 1;
+ f1->attr.intent = INTENT_INOUT;
+ f1->attr.fe_temp = 1;
+ f1->declared_at = arg1->declared_at;
+ f1->attr.referenced = 1;
+ proc_sym->formal = gfc_get_formal_arglist ();
+ proc_sym->formal->sym = f1;
+ gfc_commit_symbol (f1);
+
+ gfc_get_sym_tree (arg2->name, proc_ns, &f2t, false);
+ f2 = f2t->n.sym;
+ f2->ts = arg2->ts;
+ f2->attr.flavor = FL_VARIABLE;
+ f2->attr.dummy = 1;
+ f2->attr.intent = INTENT_IN;
+ f2->attr.fe_temp = 1;
+ f2->declared_at = arg2->declared_at;
+ f2->attr.referenced = 1;
+ proc_sym->formal->next = gfc_get_formal_arglist ();
+ proc_sym->formal->next->sym = f2;
+ gfc_commit_symbol (f2);
+
+ /* Generate the assignment statement. */
+ assign = gfc_get_code (EXEC_ASSIGN);
+
+ e1 = gfc_lval_expr_from_sym (f1);
+ e2 = gfc_get_expr ();
+ e2->where = proc_sym->declared_at;
+ e2->expr_type = EXPR_FUNCTION;
+ e2->symtree = f2t;
+ e2->ts = arg1->ts;
+ e2->value.function.esym = op_sym;
+ e2->value.function.actual = gfc_get_actual_arglist ();
+ e2->value.function.actual->expr = gfc_lval_expr_from_sym (f1);
+ e2->value.function.actual->next = gfc_get_actual_arglist ();
+ e2->value.function.actual->next->expr = gfc_lval_expr_from_sym (f2);
+ assign->expr1 = e1;
+ assign->expr2 = e2;
+ assign->loc = proc_sym->declared_at;
+
+ proc_ns->code = assign;
+
+ /* And hang it into the sibling list. */
+ proc_ns->sibling = parent_ns->contained;
+ parent_ns->contained = proc_ns;
+
+ /* ... and finally replace the call in the statement. */
+
+ oper->symtree->n.sym = proc_sym;
+ proc_sym->refs ++;
+ return 0;
+}
+
+/* Rewrite functions for co_reduce for a consistent calling
+ signature. This is only necessary if any of the functions
+ has a VALUE argument. */
+
+static void
+rewrite_co_reduce (gfc_namespace *global_ns)
+{
+ gfc_namespace *ns;
+
+ gfc_code_walker (&global_ns->code, co_reduce_code, dummy_expr_callback,
+ (void *) global_ns);
+
+ for (ns = global_ns->contained; ns; ns = ns->sibling)
+ gfc_code_walker (&ns->code, co_reduce_code, dummy_expr_callback,
+ (void *) global_ns);
+
+ return;
+}
diff --git a/gcc/fortran/gfortran.h b/gcc/fortran/gfortran.h
index d0cea838444..6940c24a6b5 100644
--- a/gcc/fortran/gfortran.h
+++ b/gcc/fortran/gfortran.h
@@ -2010,6 +2010,7 @@ typedef struct gfc_array_ref
int dimen; /* # of components in the reference */
int codimen;
bool in_allocate; /* For coarray checks. */
+ bool native_coarray_argument;
###AV: Rename to 'shared_...'. I propose to use 'caf_shared_coarray_argument'
###AV: to make clear, that is only for the shared coarray implementation
###AV: and not a shared coarray (whatever that would be).
gfc_expr *team;
gfc_expr *stat;
locus where;
diff --git a/gcc/fortran/intrinsic.c b/gcc/fortran/intrinsic.c
index ef33587a774..6a5b3e913a8 100644
--- a/gcc/fortran/intrinsic.c
+++ b/gcc/fortran/intrinsic.c
@@ -3734,7 +3734,7 @@ add_subroutines (void)
/* Coarray collectives. */
add_sym_4s ("co_broadcast", GFC_ISYM_CO_BROADCAST, CLASS_IMPURE,
BT_UNKNOWN, 0, GFC_STD_F2018,
- gfc_check_co_broadcast, NULL, NULL,
+ gfc_check_co_broadcast, NULL, gfc_resolve_co_broadcast,
a, BT_REAL, dr, REQUIRED, INTENT_INOUT,
"source_image", BT_INTEGER, di, REQUIRED, INTENT_IN,
stat, BT_INTEGER, di, OPTIONAL, INTENT_OUT,
@@ -3742,7 +3742,7 @@ add_subroutines (void)
add_sym_4s ("co_max", GFC_ISYM_CO_MAX, CLASS_IMPURE,
BT_UNKNOWN, 0, GFC_STD_F2018,
- gfc_check_co_minmax, NULL, NULL,
+ gfc_check_co_minmax, NULL, gfc_resolve_co_max,
a, BT_REAL, dr, REQUIRED, INTENT_INOUT,
result_image, BT_INTEGER, di, OPTIONAL, INTENT_IN,
stat, BT_INTEGER, di, OPTIONAL, INTENT_OUT,
@@ -3750,7 +3750,7 @@ add_subroutines (void)
add_sym_4s ("co_min", GFC_ISYM_CO_MIN, CLASS_IMPURE,
BT_UNKNOWN, 0, GFC_STD_F2018,
- gfc_check_co_minmax, NULL, NULL,
+ gfc_check_co_minmax, NULL, gfc_resolve_co_min,
a, BT_REAL, dr, REQUIRED, INTENT_INOUT,
result_image, BT_INTEGER, di, OPTIONAL, INTENT_IN,
stat, BT_INTEGER, di, OPTIONAL, INTENT_OUT,
@@ -3758,7 +3758,7 @@ add_subroutines (void)
add_sym_4s ("co_sum", GFC_ISYM_CO_SUM, CLASS_IMPURE,
BT_UNKNOWN, 0, GFC_STD_F2018,
- gfc_check_co_sum, NULL, NULL,
+ gfc_check_co_sum, NULL, gfc_resolve_co_sum,
a, BT_REAL, dr, REQUIRED, INTENT_INOUT,
result_image, BT_INTEGER, di, OPTIONAL, INTENT_IN,
stat, BT_INTEGER, di, OPTIONAL, INTENT_OUT,
@@ -3766,7 +3766,7 @@ add_subroutines (void)
add_sym_5s ("co_reduce", GFC_ISYM_CO_REDUCE, CLASS_IMPURE,
BT_UNKNOWN, 0, GFC_STD_F2018,
- gfc_check_co_reduce, NULL, NULL,
+ gfc_check_co_reduce, NULL, gfc_resolve_co_reduce,
a, BT_REAL, dr, REQUIRED, INTENT_INOUT,
"operator", BT_INTEGER, di, REQUIRED, INTENT_IN,
result_image, BT_INTEGER, di, OPTIONAL, INTENT_IN,
diff --git a/gcc/fortran/intrinsic.h b/gcc/fortran/intrinsic.h
index 166ae792939..2ca566ce3c4 100644
--- a/gcc/fortran/intrinsic.h
+++ b/gcc/fortran/intrinsic.h
@@ -677,7 +677,11 @@ void gfc_resolve_system_sub (gfc_code *);
void gfc_resolve_ttynam_sub (gfc_code *);
void gfc_resolve_umask_sub (gfc_code *);
void gfc_resolve_unlink_sub (gfc_code *);
-
+void gfc_resolve_co_sum (gfc_code *);
+void gfc_resolve_co_min (gfc_code *);
+void gfc_resolve_co_max (gfc_code *);
+void gfc_resolve_co_reduce (gfc_code *);
+void gfc_resolve_co_broadcast (gfc_code *);
/* The findloc() subroutine requires the most arguments: six. */
diff --git a/gcc/fortran/iresolve.c b/gcc/fortran/iresolve.c
index 73769615c20..844891e34ab 100644
--- a/gcc/fortran/iresolve.c
+++ b/gcc/fortran/iresolve.c
@@ -36,6 +36,7 @@ along with GCC; see the file COPYING3. If not see
#include "constructor.h"
#include "arith.h"
#include "trans.h"
+#include "options.h"
/* Given printf-like arguments, return a stable version of the result string.
@@ -4030,3 +4031,100 @@ gfc_resolve_unlink_sub (gfc_code *c)
name = gfc_get_string (PREFIX ("unlink_i%d_sub"), kind);
c->resolved_sym = gfc_get_intrinsic_sub_symbol (name);
}
+
+/* Resolve the CO_SUM et al. intrinsic subroutines. */
+
+static void
+gfc_resolve_co_collective (gfc_code *c, const char *oper)
+{
+ int kind;
+ gfc_expr *e;
+ const char *name;
+
###AV: I prefer to check for a condition and handle exactly that and then
###AV: check for the next ... So this would IMHO better be.
###AV: >>>
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ e = c->ext.actual->expr;
+ kind = e->ts.kind;
+
+ name = gfc_get_string (PREFIX ("nca_collsub_%s_%s_%c%d"), oper,
###AV: The symbol's name should be changed, too. Something like "cas_collsub..."
###AV: to reference the coarray shared approach.
+ e->rank ? "array" : "scalar",
+ gfc_type_letter (e->ts.type), kind);
+ }
+ else if (flag_coarray == GFC_FCOARRAY_LIB)
###AV: Do we need this for _SINGLE, too?
+ name = gfc_get_string (PREFIX ("caf_co_sum"));
+ else
+ gcc_unreachable ();
###AV: ===
###AV: This way, an unexpected jump into the routine is tracked and reported.
###AV: May be some better error message can be produced pointing to the actual
###AV: offending code.
+ if (flag_coarray != GFC_FCOARRAY_NATIVE)
+ name = gfc_get_string (PREFIX ("caf_co_sum"));
+ else
+ {
+ e = c->ext.actual->expr;
+ kind = e->ts.kind;
+
+ name = gfc_get_string (PREFIX ("nca_collsub_%s_%s_%c%d"), oper,
+ e->rank ? "array" : "scalar",
+ gfc_type_letter (e->ts.type), kind);
+ }
###AV: <<<
+
+ c->resolved_sym = gfc_get_intrinsic_sub_symbol (name);
+}
+
+/* Resolve CO_SUM. */
+
+void
+gfc_resolve_co_sum (gfc_code *c)
+{
+ gfc_resolve_co_collective (c, "sum");
+}
+
+/* Resolve CO_MIN. */
+
+void
+gfc_resolve_co_min (gfc_code *c)
+{
+ gfc_resolve_co_collective (c, "min");
+}
+
+/* Resolve CO_MAX. */
+
+void
+gfc_resolve_co_max (gfc_code *c)
+{
+ gfc_resolve_co_collective (c, "max");
+}
###AV: Do we need separate routine for each of the above? Couldn't
###AV: gfc_resolve_co_collecitve (c, c->function.sym.name) <- just guessed
###AV: be more general?
+
+/* Resolve CO_REDUCE. */
+
+void
+gfc_resolve_co_reduce (gfc_code *c)
+{
+ gfc_expr *e;
+ const char *name;
+
+ if (flag_coarray != GFC_FCOARRAY_NATIVE)
+ name = gfc_get_string (PREFIX ("caf_co_reduce"));
+
+ else
+ {
+ e = c->ext.actual->expr;
+ if (e->ts.type == BT_CHARACTER)
+ name = gfc_get_string (PREFIX ("nca_collsub_reduce_%s%c%d"),
+ e->rank ? "array" : "scalar",
+ gfc_type_letter (e->ts.type), e->ts.kind);
+ else
+ name = gfc_get_string (PREFIX ("nca_collsub_reduce_%s"),
+ e->rank ? "array" : "scalar" );
+ }
+
+ c->resolved_sym = gfc_get_intrinsic_sub_symbol (name);
+}
+
+void
+gfc_resolve_co_broadcast (gfc_code * c)
+{
+ gfc_expr *e;
+ const char *name;
+
+ if (flag_coarray != GFC_FCOARRAY_NATIVE)
+ name = gfc_get_string (PREFIX ("caf_co_broadcast"));
+ else
+ {
+ e = c->ext.actual->expr;
+ if (e->ts.type == BT_CHARACTER)
+ name = gfc_get_string (PREFIX ("nca_collsub_broadcast_%s%c%d"),
+ e->rank ? "array" : "scalar",
+ gfc_type_letter (e->ts.type), e->ts.kind);
+ else
+ name = gfc_get_string (PREFIX ("nca_collsub_broadcast_%s"),
+ e->rank ? "array" : "scalar" );
+ }
+
+ c->resolved_sym = gfc_get_intrinsic_sub_symbol (name);
+}
diff --git a/gcc/fortran/lang.opt b/gcc/fortran/lang.opt
index da4b1aa879a..61803554c6c 100644
--- a/gcc/fortran/lang.opt
+++ b/gcc/fortran/lang.opt
@@ -761,7 +761,7 @@ Copy array sections into a contiguous block on procedure entry.
fcoarray=
Fortran RejectNegative Joined Enum(gfc_fcoarray) Var(flag_coarray) Init(GFC_FCOARRAY_NONE)
--fcoarray=<none|single|lib> Specify which coarray parallelization should be used.
+-fcoarray=<none|single|lib|shared> Specify which coarray parallelization should be used.
Enum
Name(gfc_fcoarray) Type(enum gfc_fcoarray) UnknownError(Unrecognized option: %qs)
@@ -775,6 +775,9 @@ Enum(gfc_fcoarray) String(single) Value(GFC_FCOARRAY_SINGLE)
EnumValue
Enum(gfc_fcoarray) String(lib) Value(GFC_FCOARRAY_LIB)
+EnumValue
+Enum(gfc_fcoarray) String(shared) Value(GFC_FCOARRAY_NATIVE)
+
fcheck=
Fortran RejectNegative JoinedOrMissing
-fcheck=[...] Specify which runtime checks are to be performed.
diff --git a/gcc/fortran/resolve.c b/gcc/fortran/resolve.c
index f4ce49f8432..6d6984b022b 100644
--- a/gcc/fortran/resolve.c
+++ b/gcc/fortran/resolve.c
@@ -3585,6 +3585,53 @@ resolve_specific_s (gfc_code *c)
return false;
}
+/* Fix up references to native coarrays in call - element references
###AV: ... to shared coarrays in ...
+ have to be converted to full references if the coarray has to be
+ passed fully. */
+
+static void
+fixup_coarray_args (gfc_symbol *sym, gfc_actual_arglist *actual)
###AV: For now this only for the shared coarray approach, which should be
###AV: reflected in the name. 'fixup_shared_coarray_args ()' may be?
+{
+ gfc_formal_arglist *formal, *f;
###AV: The variable 'formal' is not really needed here. Replace it with 'f'
###AV: and you are done.
+ gfc_actual_arglist *a;
+
+ formal = gfc_sym_get_dummy_args (sym);
+
+ if (formal == NULL)
+ return;
+
+ for (a = actual, f = formal; a && f; a = a->next, f = f->next)
+ {
+ if (a->expr == NULL || f->sym == NULL)
+ continue;
+ if (a->expr->expr_type == EXPR_VARIABLE
+ && a->expr->symtree->n.sym->attr.codimension
+ && f->sym->attr.codimension)
+ {
+ gfc_ref *r;
+ for (r = a->expr->ref; r; r = r->next)
+ {
+ if (r->type == REF_ARRAY && r->u.ar.codimen)
+ {
+ gfc_array_ref *ar = &r->u.ar;
+ int i, eff_dimen = ar->dimen + ar->codimen;
+
+ for (i = ar->dimen; i < eff_dimen; i++)
###AV: Use '++i' which is more (memory) efficient in general.
+ {
+ ar->dimen_type[i] = DIMEN_RANGE;
+ gcc_assert (ar->start[i] == NULL);
+ gcc_assert (ar->end[i] == NULL);
+ }
+
+ if (ar->type == AR_ELEMENT)
+ ar->type = !ar->dimen ? AR_FULL : AR_SECTION;
+
+ ar->native_coarray_argument = true;
+ }
+ }
+ }
+ }
+}
/* Resolve a subroutine call not known to be generic nor specific. */
@@ -3615,7 +3662,7 @@ resolve_unknown_s (gfc_code *c)
found:
gfc_procedure_use (sym, &c->ext.actual, &c->loc);
-
+
###AV: ???
c->resolved_sym = sym;
return pure_subroutine (sym, sym->name, &c->loc);
@@ -3740,6 +3787,9 @@ resolve_call (gfc_code *c)
/* Typebound procedure: Assume the worst. */
gfc_current_ns->proc_name->attr.array_outer_dependency = 1;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ fixup_coarray_args (csym, c->ext.actual);
+
return t;
}
@@ -10117,7 +10167,7 @@ resolve_critical (gfc_code *code)
char name[GFC_MAX_SYMBOL_LEN];
static int serial = 0;
- if (flag_coarray != GFC_FCOARRAY_LIB)
+ if (flag_coarray != GFC_FCOARRAY_LIB && flag_coarray != GFC_FCOARRAY_NATIVE)
return;
symtree = gfc_find_symtree (gfc_current_ns->sym_root,
@@ -10154,6 +10204,19 @@ resolve_critical (gfc_code *code)
symtree->n.sym->as->lower[0] = gfc_get_int_expr (gfc_default_integer_kind,
NULL, 1);
gfc_commit_symbols();
+
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ gfc_ref *r = gfc_get_ref ();
+ r->type = REF_ARRAY;
+ r->u.ar.type = AR_ELEMENT;
+ r->u.ar.as = code->resolved_sym->as;
+ for (int i = 0; i < code->resolved_sym->as->corank; i++)
+ r->u.ar.dimen_type [i] = DIMEN_THIS_IMAGE;
+
+ code->expr1 = gfc_lval_expr_from_sym (code->resolved_sym);
+ code->expr1->ref = r;
+ }
}
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!) [Review part 2]
2020-10-05 13:23 ` Nicolas König
2020-10-12 12:32 ` [RFC] Native Coarrays (finally!) [Review part 1] Andre Vehreschild
@ 2020-10-12 13:48 ` Andre Vehreschild
2020-10-13 12:42 ` Nicolas König
2020-10-13 13:01 ` [RFC] Native Coarrays (finally!) [Review part 3] Andre Vehreschild
2 siblings, 1 reply; 17+ messages in thread
From: Andre Vehreschild @ 2020-10-12 13:48 UTC (permalink / raw)
To: Nicolas König; +Cc: fortran
[-- Attachment #1: Type: text/plain, Size: 1482 bytes --]
Hi Nicolas,
here is part two of the review of the compiler components. I will do the
testsuite and library parts another day, because now I am already completely
bonkers. (Yes, I know, that's normal for me :-)
Regards,
Andre
On Mon, 5 Oct 2020 15:23:27 +0200
Nicolas König <koenigni@student.ethz.ch> wrote:
> Hello Tobias,
>
> On 05/10/2020 11:54, Tobias Burnus wrote:
> > Hi Nicolas,
> >
> > admittedly, I have not yet looked at your patch. However, I have to
> > admit that I do not like the name. I understand that "native" refers
> > to not needing an external library (libcaf.../libopencoarray...),
> > but I still wonder whether something like "-fcoarray=shared" (i.e.
> > working on a shared-memory system) would be better name from an end-user
> > point of view.
>
> I think the name has been the most critized point of the entire patch up
> till now. I'm going to change it to -fcoarray=shared, as you (and a few
> other people) suggested :)
>
> >
> > Tobias,
> > who likes that coarray can be used without extra libs and thinks
> > that this will help with users starting to use coarrays.
>
> That is the main reason I wrote the patch.
>
> >
> > -----------------
> > Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München /
> > Germany
> > Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung,
> > Alexander Walter
--
Andre Vehreschild * Email: vehre ad gmx dot de
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: caf_shared_part_2.patch --]
[-- Type: text/x-patch, Size: 73465 bytes --]
diff --git a/gcc/fortran/trans-array.c b/gcc/fortran/trans-array.c
index 6566c47d4ae..9013f1984af 100644
--- a/gcc/fortran/trans-array.c
+++ b/gcc/fortran/trans-array.c
@@ -2940,6 +2940,60 @@ gfc_add_loop_ss_code (gfc_loopinfo * loop, gfc_ss * ss, bool subscript,
gfc_add_loop_ss_code (nested_loop, nested_loop->ss, subscript, where);
}
+static tree
+gfc_add_strides (tree expr, tree desc, int beg, int end)
+{
+ int i;
+ tree tmp, stride;
+ tmp = gfc_index_zero_node;
+ for (i = beg; i < end; i++)
+ {
+ stride = gfc_conv_array_stride (desc, i);
+ tmp = fold_build2_loc (input_location, PLUS_EXPR, TREE_TYPE(tmp),
+ tmp, stride);
+ }
+ return fold_build2_loc (input_location, PLUS_EXPR, TREE_TYPE(expr),
+ expr, tmp);
+}
+
+/* This function calculates the new offset via
+ new_offset = offset + this_image ()
+ * arrray.stride[first_codimension]
###AV: Fix typo: arrray -> array
+ + sum (remaining codimension offsets)
+ If offset is a pointer, we also need to multiply it by the size.*/
###AV: Gcc style enforces '. */' for the end of a comment. Or did that change?
+static tree
+gfc_native_coarray_add_this_image_offset (tree offset, tree desc,
+ gfc_array_ref *ar, int is_pointer,
+ int subtract)
+{
+ tree tmp, off;
+ /* Calculate the actual offset. */
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_nca_this_image,
+ 1, integer_zero_node);
+ tmp = convert (TREE_TYPE(gfc_index_zero_node), tmp);
###AV: Style! It should be 'TREE_TYPE (...'.
+ tmp = fold_build2_loc (input_location, MINUS_EXPR, TREE_TYPE(tmp), tmp,
+ build_int_cst (TREE_TYPE(tmp), subtract));
+ tmp = fold_build2_loc (input_location, MULT_EXPR, TREE_TYPE(tmp),
+ gfc_conv_array_stride (desc, ar->dimen), tmp);
+ /* We also need to add the missing strides once to compensate for the
+ offset, that is to large now. The loop starts at sym->as.rank+1
+ because we need to skip the first corank stride */
+ off = gfc_add_strides (tmp, desc, ar->as->rank + 1,
+ ar->as->rank + ar->as->corank);
+ if (is_pointer)
+ {
+ /* Remove pointer and array from type in order to get the raw base type. */
+ tmp = TREE_TYPE(TREE_TYPE(TREE_TYPE(offset)));
+ /* And get the size of that base type. */
+ tmp = convert (TREE_TYPE(off), size_in_bytes_loc (input_location, tmp));
+ tmp = fold_build2_loc (input_location, MULT_EXPR, TREE_TYPE(off),
+ off, tmp);
+ return fold_build_pointer_plus_loc (input_location, offset, tmp);
+ }
+ else
+ return fold_build2_loc (input_location, PLUS_EXPR, TREE_TYPE(offset),
+ offset, off);
+}
/* Translate expressions for the descriptor and data pointer of a SS. */
/*GCC ARRAYS*/
@@ -2951,6 +3005,7 @@ gfc_conv_ss_descriptor (stmtblock_t * block, gfc_ss * ss, int base)
gfc_ss_info *ss_info;
gfc_array_info *info;
tree tmp;
+ gfc_ref *ref;
ss_info = ss->info;
info = &ss_info->data.array;
@@ -2982,10 +3037,18 @@ gfc_conv_ss_descriptor (stmtblock_t * block, gfc_ss * ss, int base)
}
/* Also the data pointer. */
tmp = gfc_conv_array_data (se.expr);
+ /* If we have a native coarray with implied this_image (), add the
+ appropriate offset to the data pointer. */
+ ref = ss_info->expr->ref;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && ref
+ && ref->u.ar.dimen_type[ref->u.ar.dimen + ref->u.ar.codimen - 1]
+ == DIMEN_THIS_IMAGE)
+ tmp = gfc_native_coarray_add_this_image_offset (tmp, se.expr, &ref->u.ar, 1, 1);
/* If this is a variable or address of a variable we use it directly.
Otherwise we must evaluate it now to avoid breaking dependency
analysis by pulling the expressions for elemental array indices
inside the loop. */
+
if (!(DECL_P (tmp)
|| (TREE_CODE (tmp) == ADDR_EXPR
&& DECL_P (TREE_OPERAND (tmp, 0)))))
@@ -2993,6 +3056,15 @@ gfc_conv_ss_descriptor (stmtblock_t * block, gfc_ss * ss, int base)
info->data = tmp;
tmp = gfc_conv_array_offset (se.expr);
+ /* If we have a native coarray, adjust the offset to remove the
+ offset for the codimensions. */
+ // TODO: check whether the recipient is a coarray, if it is, disable
+ // all of this
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && ref
+ && ref->u.ar.dimen_type[ref->u.ar.dimen + ref->u.ar.codimen - 1]
+ == DIMEN_THIS_IMAGE)
+ tmp = gfc_add_strides (tmp, se.expr, ref->u.ar.as->rank,
+ ref->u.ar.as->rank + ref->u.ar.as->corank);
info->offset = gfc_evaluate_now (tmp, block);
/* Make absolutely sure that the saved_offset is indeed saved
@@ -3593,6 +3665,7 @@ build_array_ref (tree desc, tree offset, tree decl, tree vptr)
}
+
###AV: ???
/* Build an array reference. se->expr already holds the array descriptor.
This should be either a variable, indirect variable reference or component
reference. For arrays which do not have a descriptor, se->expr will be
@@ -3612,8 +3685,20 @@ gfc_conv_array_ref (gfc_se * se, gfc_array_ref * ar, gfc_expr *expr,
gfc_se tmpse;
gfc_symbol * sym = expr->symtree->n.sym;
char *var_name = NULL;
+ bool need_impl_this_image;
###AV: Why name this 'need_...' when you're checking that it is an impl_this_image?
###AV: Better 'is_impl_this_image'?
+ int eff_dimen;
+
+ need_impl_this_image =
+ ar->dimen_type[ar->dimen + ar->codimen - 1] == DIMEN_THIS_IMAGE;
+
+ if (flag_coarray == GFC_FCOARRAY_NATIVE
+ && !need_impl_this_image)
+ eff_dimen = ar->dimen + ar->codimen - 1;
+ else
+ eff_dimen = ar->dimen - 1;
- if (ar->dimen == 0)
+
+ if (flag_coarray != GFC_FCOARRAY_NATIVE && ar->dimen == 0)
{
gcc_assert (ar->codimen || sym->attr.select_rank_temporary
|| (ar->as && ar->as->corank));
@@ -3681,7 +3766,7 @@ gfc_conv_array_ref (gfc_se * se, gfc_array_ref * ar, gfc_expr *expr,
/* Calculate the offsets from all the dimensions. Make sure to associate
the final offset so that we form a chain of loop invariant summands. */
- for (n = ar->dimen - 1; n >= 0; n--)
+ for (n = eff_dimen; n >= 0; n--)
{
/* Calculate the index for this dimension. */
gfc_init_se (&indexse, se);
@@ -3753,6 +3838,9 @@ gfc_conv_array_ref (gfc_se * se, gfc_array_ref * ar, gfc_expr *expr,
add_to_offset (&cst_offset, &offset, tmp);
}
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && need_impl_this_image)
+ offset = gfc_native_coarray_add_this_image_offset (offset, se->expr, ar, 0, 0);
+
if (!integer_zerop (cst_offset))
offset = fold_build2_loc (input_location, PLUS_EXPR,
gfc_array_index_type, offset, cst_offset);
@@ -5423,7 +5511,7 @@ gfc_conv_descriptor_cosize (tree desc, int rank, int corank)
} */
/*GCC ARRAYS*/
-static tree
+tree
gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
gfc_expr ** lower, gfc_expr ** upper, stmtblock_t * pblock,
stmtblock_t * descriptor_block, tree * overflow,
@@ -5441,6 +5529,8 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
tree elsecase;
tree cond;
tree var;
+ tree conv_lbound;
+ tree conv_ubound;
stmtblock_t thenblock;
stmtblock_t elseblock;
gfc_expr *ubound;
@@ -5454,7 +5544,7 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
/* Set the dtype before the alloc, because registration of coarrays needs
it initialized. */
- if (expr->ts.type == BT_CHARACTER
+ if (expr && expr->ts.type == BT_CHARACTER
&& expr->ts.deferred
&& VAR_P (expr->ts.u.cl->backend_decl))
{
@@ -5462,7 +5552,7 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
tmp = gfc_conv_descriptor_dtype (descriptor);
gfc_add_modify (pblock, tmp, gfc_get_dtype_rank_type (rank, type));
}
- else if (expr->ts.type == BT_CHARACTER
+ else if (expr && expr->ts.type == BT_CHARACTER
&& expr->ts.deferred
&& TREE_CODE (descriptor) == COMPONENT_REF)
{
@@ -5494,9 +5584,6 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
for (n = 0; n < rank; n++)
{
- tree conv_lbound;
- tree conv_ubound;
-
###AV: This changes semantics when not in shared coarray mode. I can't see the
###AV: adjacent code here, but you better should do something like
###AV: >>>
if (flag_coarray != GFC_FCOARRAY_NATIVE)
conv_lbound = conv_ubound = NULL_TREE;
###AV: <<<
###AV: here to keep semantics the same when not in shared coarray mode. Or you
###AV: have to carefully examine each line of code in the loop to make sure
###AV: that keeping a former iteration's l|ubound does mean no harm.
/* We have 3 possibilities for determining the size of the array:
lower == NULL => lbound = 1, ubound = upper[n]
upper[n] = NULL => lbound = 1, ubound = lower[n]
@@ -5646,6 +5733,15 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
}
gfc_conv_descriptor_lbound_set (descriptor_block, descriptor,
gfc_rank_cst[n], se.expr);
+ conv_lbound = se.expr;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+
+ tmp = fold_build2_loc (input_location, MULT_EXPR, gfc_array_index_type,
+ se.expr, stride);
+ offset = fold_build2_loc (input_location, MINUS_EXPR,
+ gfc_array_index_type, offset, tmp);
+ }
if (n < rank + corank - 1)
{
@@ -5655,6 +5751,18 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
gfc_add_block_to_block (pblock, &se.pre);
gfc_conv_descriptor_ubound_set (descriptor_block, descriptor,
gfc_rank_cst[n], se.expr);
+ gfc_conv_descriptor_stride_set (descriptor_block, descriptor,
+ gfc_rank_cst[n], stride);
###AV: This line looks dubious to me. Why is always needed and not only in
###AV: shared coarray mode?
+ conv_ubound = se.expr;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ size = gfc_conv_array_extent_dim (conv_lbound, conv_ubound,
+ &or_expr);
###AV: Something went wrong with the indentation here. But that also be just my
###AV: editor.
+ size = gfc_evaluate_now (size, descriptor_block);
+ stride = fold_build2_loc (input_location, MULT_EXPR,
+ gfc_array_index_type, stride, size);
+ stride = gfc_evaluate_now (stride, descriptor_block);
+ }
}
}
@@ -5688,7 +5796,7 @@ gfc_array_init_size (tree descriptor, int rank, int corank, tree * poffset,
/* Convert to size_t. */
*element_size = fold_convert (size_type_node, tmp);
- if (rank == 0)
+ if (rank == 0 && !(flag_coarray == GFC_FCOARRAY_NATIVE && corank))
return *element_size;
*nelems = gfc_evaluate_now (stride, pblock);
@@ -5773,6 +5881,38 @@ retrieve_last_ref (gfc_ref **ref_in, gfc_ref **prev_ref_in)
return true;
}
+int
+gfc_native_coarray_get_allocation_type (gfc_symbol * sym)
+{
+ bool is_lock_type, is_event_type;
+ is_lock_type = sym->ts.type == BT_DERIVED
+ && sym->ts.u.derived->from_intmod == INTMOD_ISO_FORTRAN_ENV
+ && sym->ts.u.derived->intmod_sym_id == ISOFORTRAN_LOCK_TYPE;
+
+ is_event_type = sym->ts.type == BT_DERIVED
+ && sym->ts.u.derived->from_intmod == INTMOD_ISO_FORTRAN_ENV
+ && sym->ts.u.derived->intmod_sym_id == ISOFORTRAN_EVENT_TYPE;
+
+ if (is_lock_type)
+ return GFC_NCA_LOCK_COARRAY;
+ else if (is_event_type)
+ return GFC_NCA_EVENT_COARRAY;
+ else
+ return GFC_NCA_NORMAL_COARRAY;
+}
+
+void
+gfc_allocate_native_coarray (stmtblock_t *b, tree decl, tree size, int corank,
+ int alloc_type)
+{
+ gfc_add_expr_to_block (b,
+ build_call_expr_loc (input_location, gfor_fndecl_nca_coarray_allocate,
+ 4, gfc_build_addr_expr (pvoid_type_node, decl),
+ size, build_int_cst (integer_type_node, corank),
+ build_int_cst (integer_type_node, alloc_type)));
+
+}
+
/* Initializes the descriptor and generates a call to _gfor_allocate. Does
the work for an ALLOCATE statement. */
/*GCC ARRAYS*/
@@ -5784,6 +5924,7 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
bool e3_has_nodescriptor)
{
tree tmp;
+ tree allocation;
tree pointer;
tree offset = NULL_TREE;
tree token = NULL_TREE;
@@ -5914,7 +6055,7 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
expr3_elem_size, nelems, expr3, e3_arr_desc,
e3_has_nodescriptor, expr, &element_size);
- if (dimension)
+ if (dimension || (flag_coarray == GFC_FCOARRAY_NATIVE && coarray))
###AV: How about setting 'bool is_shared_coarray = flag_coarray == GFC_FCOARRAY_NATIVE && coarray;'
###AV: above and reusing it here over and over again?
{
var_overflow = gfc_create_var (integer_type_node, "overflow");
gfc_add_modify (&se->pre, var_overflow, overflow);
@@ -5956,7 +6097,7 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
pointer = gfc_conv_descriptor_data_get (se->expr);
STRIP_NOPS (pointer);
- if (allocatable)
+ if (allocatable && !(flag_coarray == GFC_FCOARRAY_NATIVE && coarray))
{
not_prev_allocated = gfc_create_var (logical_type_node,
"not_prev_allocated");
@@ -5969,8 +6110,17 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
gfc_start_block (&elseblock);
+ if (coarray && flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ tree elem_size
+ = size_in_bytes (gfc_get_element_type (TREE_TYPE(se->expr)));
+ int alloc_type
+ = gfc_native_coarray_get_allocation_type (expr->symtree->n.sym);
+ gfc_allocate_native_coarray (&elseblock, se->expr, elem_size,
+ ref->u.ar.as->corank, alloc_type);
+ }
/* The allocatable variant takes the old pointer as first argument. */
- if (allocatable)
+ else if (allocatable)
gfc_allocate_allocatable (&elseblock, pointer, size, token,
status, errmsg, errlen, label_finish, expr,
coref != NULL ? coref->u.ar.as->corank : 0);
@@ -5987,13 +6137,12 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
cond = gfc_unlikely (fold_build2_loc (input_location, NE_EXPR,
logical_type_node, var_overflow, integer_zero_node),
PRED_FORTRAN_OVERFLOW);
- tmp = fold_build3_loc (input_location, COND_EXPR, void_type_node, cond,
+ allocation = fold_build3_loc (input_location, COND_EXPR, void_type_node, cond,
error, gfc_finish_block (&elseblock));
}
else
- tmp = gfc_finish_block (&elseblock);
+ allocation = gfc_finish_block (&elseblock);
- gfc_add_expr_to_block (&se->pre, tmp);
/* Update the array descriptor with the offset and the span. */
if (dimension)
@@ -6004,6 +6153,7 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
}
set_descriptor = gfc_finish_block (&set_descriptor_block);
+
if (status != NULL_TREE)
{
cond = fold_build2_loc (input_location, EQ_EXPR,
@@ -6014,14 +6164,25 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
cond = fold_build2_loc (input_location, TRUTH_OR_EXPR,
logical_type_node, cond, not_prev_allocated);
- gfc_add_expr_to_block (&se->pre,
- fold_build3_loc (input_location, COND_EXPR, void_type_node,
+ set_descriptor = fold_build3_loc (input_location, COND_EXPR, void_type_node,
cond,
set_descriptor,
- build_empty_stmt (input_location)));
+ build_empty_stmt (input_location));
+ }
+
+ // For native coarrays, the size must be set before the allocation routine
+ // can be called.
###AV: GCC coding prefers block comments!
+ if (coarray && flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ gfc_add_expr_to_block (&se->pre, set_descriptor);
+ gfc_add_expr_to_block (&se->pre, allocation);
}
else
+ {
+ gfc_add_expr_to_block (&se->pre, allocation);
gfc_add_expr_to_block (&se->pre, set_descriptor);
+ }
+
return true;
}
@@ -6524,6 +6685,7 @@ gfc_trans_dummy_array_bias (gfc_symbol * sym, tree tmpdesc,
bool optional_arg;
gfc_array_spec *as;
bool is_classarray = IS_CLASS_ARRAY (sym);
+ int eff_dimen;
/* Do nothing for pointer and allocatable arrays. */
if ((sym->ts.type != BT_CLASS && sym->attr.pointer)
@@ -6638,8 +6800,13 @@ gfc_trans_dummy_array_bias (gfc_symbol * sym, tree tmpdesc,
offset = gfc_index_zero_node;
size = gfc_index_one_node;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ eff_dimen = as->rank + as->corank;
+ else
+ eff_dimen = as->rank;
+
/* Evaluate the bounds of the array. */
- for (n = 0; n < as->rank; n++)
+ for (n = 0; n < eff_dimen; n++)
{
if (checkparm || !as->upper[n])
{
@@ -6724,7 +6891,7 @@ gfc_trans_dummy_array_bias (gfc_symbol * sym, tree tmpdesc,
gfc_array_index_type, offset, tmp);
/* The size of this dimension, and the stride of the next. */
- if (n + 1 < as->rank)
+ if (n + 1 < eff_dimen)
{
stride = GFC_TYPE_ARRAY_STRIDE (type, n + 1);
@@ -6879,20 +7046,35 @@ gfc_get_dataptr_offset (stmtblock_t *block, tree parm, tree desc, tree offset,
return;
}
+ /* if it's a coarray with implicit this_image, add that to the offset. */
+ ref = expr->ref;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && ref && ref->type == REF_ARRAY
+ && ref->u.ar.dimen_type[ref->u.ar.dimen + ref->u.ar.codimen - 1]
+ == DIMEN_THIS_IMAGE
+ && !ref->u.ar.native_coarray_argument)
+ offset = gfc_native_coarray_add_this_image_offset (offset, desc,
+ &ref->u.ar, 0, 1);
+
tmp = build_array_ref (desc, offset, NULL, NULL);
/* Offset the data pointer for pointer assignments from arrays with
subreferences; e.g. my_integer => my_type(:)%integer_component. */
if (subref)
{
- /* Go past the array reference. */
+ /* Go past the array reference. */
###AV: Nop, '. */' i.e. dot, two spaces, closing comment marked is according
###AV: to the style guide the way required.
for (ref = expr->ref; ref; ref = ref->next)
- if (ref->type == REF_ARRAY &&
- ref->u.ar.type != AR_ELEMENT)
- {
- ref = ref->next;
- break;
- }
+ {
+ if (ref->type == REF_ARRAY &&
+ ref->u.ar.type != AR_ELEMENT)
+ {
+ ref = ref->next;
+ break;
+ }
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE && ref->type == REF_ARRAY &&
+ ref->u.ar.dimen_type[ref->u.ar.dimen +ref->u.ar.codimen -1]
+ == DIMEN_THIS_IMAGE)
+ tmp = gfc_native_coarray_add_this_image_offset (tmp, desc, &ref->u.ar, 0, 1);
+ }
/* Calculate the offset for each subsequent subreference. */
for (; ref; ref = ref->next)
@@ -6955,7 +7137,10 @@ gfc_get_dataptr_offset (stmtblock_t *block, tree parm, tree desc, tree offset,
gfc_array_index_type, stride, itmp);
stride = gfc_evaluate_now (stride, block);
}
-
+ if (flag_coarray == GFC_FCOARRAY_NATIVE &&
+ ref->u.ar.dimen_type[ref->u.ar.dimen +ref->u.ar.codimen -1]
###AV: A space goes before and after each operator!
+ == DIMEN_THIS_IMAGE)
+ tmp = gfc_native_coarray_add_this_image_offset (tmp, desc, &ref->u.ar, 0, 1);
/* Apply the index to obtain the array element. */
tmp = gfc_build_array_ref (tmp, index, NULL);
break;
@@ -7306,6 +7491,13 @@ gfc_conv_expr_descriptor (gfc_se *se, gfc_expr *expr)
else
full = gfc_full_array_ref_p (info->ref, NULL);
+ if (flag_coarray == GFC_FCOARRAY_NATIVE &&
+ info->ref->type == REF_ARRAY &&
+ info->ref->u.ar.dimen_type[info->ref->u.ar.dimen
+ + info->ref->u.ar.codimen - 1] ==
+ DIMEN_THIS_IMAGE)
+ full = 0;
+
if (full && !transposed_dims (ss))
{
if (se->direct_byref && !se->byref_noassign)
@@ -7540,9 +7732,19 @@ gfc_conv_expr_descriptor (gfc_se *se, gfc_expr *expr)
tree to;
tree base;
tree offset;
-
+#if 0 /* TK */
ndim = info->ref ? info->ref->u.ar.dimen : ss->dimen;
-
+#else
+ if (info->ref)
+ {
+ if (info->ref->u.ar.native_coarray_argument)
+ ndim = info->ref->u.ar.dimen + info->ref->u.ar.codimen;
+ else
+ ndim = info->ref->u.ar.dimen;
+ }
+ else
+ ndim = ss->dimen;
+#endif
###AV: FIX !!!
if (se->want_coarray)
{
gfc_array_ref *ar = &info->ref->u.ar;
@@ -7911,7 +8113,15 @@ gfc_conv_array_parameter (gfc_se * se, gfc_expr * expr, bool g77,
expr->ts.u.cl->backend_decl = tmp;
se->string_length = tmp;
}
-
+#if 0
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && fsym && fsym->attr.codimension && sym)
+ {
+ gfc_init_se (se, NULL);
+ tmp = gfc_get_symbol_decl (sym);
+ se->expr = gfc_build_addr_expr (NULL_TREE, tmp);
+ return;
+ }
+#endif
###AV: FIX !!!
/* Is this the result of the enclosing procedure? */
this_array_result = (full_array_var && sym->attr.flavor == FL_PROCEDURE);
if (this_array_result
@@ -7919,6 +8129,10 @@ gfc_conv_array_parameter (gfc_se * se, gfc_expr * expr, bool g77,
&& (sym->backend_decl != parent))
this_array_result = false;
+#if 1 /* TK */
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && fsym && fsym->attr.codimension)
+ g77 = false;
+#endif
###AV: Fix !!!
/* Passing address of the array if it is not pointer or assumed-shape. */
if (full_array_var && g77 && !this_array_result
&& sym->ts.type != BT_DERIVED && sym->ts.type != BT_CLASS)
@@ -8053,8 +8267,8 @@ gfc_conv_array_parameter (gfc_se * se, gfc_expr * expr, bool g77,
{
/* Every other type of array. */
se->want_pointer = 1;
- gfc_conv_expr_descriptor (se, expr);
+ gfc_conv_expr_descriptor (se, expr);
###AV: Uhh?
if (size)
array_parameter_size (build_fold_indirect_ref_loc (input_location,
se->expr),
@@ -10869,9 +11083,15 @@ gfc_walk_array_ref (gfc_ss * ss, gfc_expr * expr, gfc_ref * ref)
case AR_SECTION:
newss = gfc_get_array_ss (ss, expr, 0, GFC_SS_SECTION);
newss->info->data.array.ref = ref;
-
+#if 1 /* TK */
+ int eff_dimen;
+ if (ar->native_coarray_argument)
+ eff_dimen = ar->dimen + ar->codimen;
+ else
+ eff_dimen = ar->dimen;
+#endif
###AV: Fix !!!
/* We add SS chains for all the subscripts in the section. */
- for (n = 0; n < ar->dimen; n++)
+ for (n = 0; n < eff_dimen; n++)
{
gfc_ss *indexss;
diff --git a/gcc/fortran/trans-array.h b/gcc/fortran/trans-array.h
index e561605aaed..0bfd1b03022 100644
--- a/gcc/fortran/trans-array.h
+++ b/gcc/fortran/trans-array.h
@@ -23,6 +23,15 @@ along with GCC; see the file COPYING3. If not see
bool gfc_array_allocate (gfc_se *, gfc_expr *, tree, tree, tree, tree,
tree, tree *, gfc_expr *, tree, bool);
+enum gfc_coarray_allocation_type {
+ GFC_NCA_NORMAL_COARRAY = 3,
###AV: Why does this start with 3?
+ GFC_NCA_LOCK_COARRAY,
+ GFC_NCA_EVENT_COARRAY
+};
###AV: Add an empty line between the enum the func decl.
+int gfc_native_coarray_get_allocation_type (gfc_symbol *);
+
+void gfc_allocate_native_coarray (stmtblock_t *, tree, tree, int, int);
+
/* Allow the bounds of a loop to be set from a callee's array spec. */
void gfc_set_loop_bounds_from_array_spec (gfc_interface_mapping *,
gfc_se *, gfc_array_spec *);
@@ -57,6 +66,10 @@ tree gfc_bcast_alloc_comp (gfc_symbol *, gfc_expr *, int, tree,
tree gfc_deallocate_alloc_comp_no_caf (gfc_symbol *, tree, int);
tree gfc_reassign_alloc_comp_caf (gfc_symbol *, tree, tree);
+tree gfc_array_init_size (tree, int, int, tree *, gfc_expr **, gfc_expr **,
+ stmtblock_t *, stmtblock_t *, tree *, tree, tree *,
+ gfc_expr *, tree, bool, gfc_expr *, tree *);
+
tree gfc_copy_alloc_comp (gfc_symbol *, tree, tree, int, int);
tree gfc_copy_only_alloc_comp (gfc_symbol *, tree, tree, int);
diff --git a/gcc/fortran/trans-decl.c b/gcc/fortran/trans-decl.c
index 92242771dde..5eadf40e367 100644
--- a/gcc/fortran/trans-decl.c
+++ b/gcc/fortran/trans-decl.c
@@ -170,6 +170,21 @@ tree gfor_fndecl_co_reduce;
tree gfor_fndecl_co_sum;
tree gfor_fndecl_caf_is_present;
+/* Native coarray functions. */
+
+tree gfor_fndecl_nca_master;
+tree gfor_fndecl_nca_coarray_allocate;
+tree gfor_fndecl_nca_coarray_free;
+tree gfor_fndecl_nca_this_image;
+tree gfor_fndecl_nca_num_images;
+tree gfor_fndecl_nca_sync_all;
+tree gfor_fndecl_nca_sync_images;
+tree gfor_fndecl_nca_lock;
+tree gfor_fndecl_nca_unlock;
+tree gfor_fndecl_nca_reduce_scalar;
+tree gfor_fndecl_nca_reduce_array;
+tree gfor_fndecl_nca_broadcast_scalar;
+tree gfor_fndecl_nca_broadcast_array;
/* Math functions. Many other math functions are handled in
trans-intrinsic.c. */
@@ -961,6 +976,7 @@ gfc_build_qualified_array (tree decl, gfc_symbol * sym)
tree type;
int dim;
int nest;
+ int eff_dimen;
gfc_namespace* procns;
symbol_attribute *array_attr;
gfc_array_spec *as;
@@ -1031,8 +1047,12 @@ gfc_build_qualified_array (tree decl, gfc_symbol * sym)
else
gfc_add_decl_to_function (token);
}
+
+ eff_dimen = flag_coarray == GFC_FCOARRAY_NATIVE
+ ? GFC_TYPE_ARRAY_RANK (type) + GFC_TYPE_ARRAY_CORANK (type)
+ : GFC_TYPE_ARRAY_RANK (type);
- for (dim = 0; dim < GFC_TYPE_ARRAY_RANK (type); dim++)
+ for (dim = 0; dim < eff_dimen; dim++)
{
if (GFC_TYPE_ARRAY_LBOUND (type, dim) == NULL_TREE)
{
@@ -1054,22 +1074,30 @@ gfc_build_qualified_array (tree decl, gfc_symbol * sym)
TREE_NO_WARNING (GFC_TYPE_ARRAY_STRIDE (type, dim)) = 1;
}
}
- for (dim = GFC_TYPE_ARRAY_RANK (type);
- dim < GFC_TYPE_ARRAY_RANK (type) + GFC_TYPE_ARRAY_CORANK (type); dim++)
- {
- if (GFC_TYPE_ARRAY_LBOUND (type, dim) == NULL_TREE)
- {
- GFC_TYPE_ARRAY_LBOUND (type, dim) = create_index_var ("lbound", nest);
- TREE_NO_WARNING (GFC_TYPE_ARRAY_LBOUND (type, dim)) = 1;
- }
- /* Don't try to use the unknown ubound for the last coarray dimension. */
- if (GFC_TYPE_ARRAY_UBOUND (type, dim) == NULL_TREE
- && dim < GFC_TYPE_ARRAY_RANK (type) + GFC_TYPE_ARRAY_CORANK (type) - 1)
- {
- GFC_TYPE_ARRAY_UBOUND (type, dim) = create_index_var ("ubound", nest);
- TREE_NO_WARNING (GFC_TYPE_ARRAY_UBOUND (type, dim)) = 1;
- }
- }
+
+ if (flag_coarray != GFC_FCOARRAY_NATIVE)
+ for (dim = GFC_TYPE_ARRAY_RANK (type);
+ dim < GFC_TYPE_ARRAY_RANK (type) + GFC_TYPE_ARRAY_CORANK (type);
+ dim++)
+ {
+ if (GFC_TYPE_ARRAY_LBOUND (type, dim) == NULL_TREE)
+ {
+ GFC_TYPE_ARRAY_LBOUND (type, dim)
+ = create_index_var ("lbound", nest);
+ TREE_NO_WARNING (GFC_TYPE_ARRAY_LBOUND (type, dim)) = 1;
+ }
+ /* Don't try to use the unknown ubound for the last coarray
+ dimension. */
+ if (GFC_TYPE_ARRAY_UBOUND (type, dim) == NULL_TREE
+ && dim < GFC_TYPE_ARRAY_RANK (type)
+ + GFC_TYPE_ARRAY_CORANK (type) - 1)
+ {
+ GFC_TYPE_ARRAY_UBOUND (type, dim)
+ = create_index_var ("ubound", nest);
+ TREE_NO_WARNING (GFC_TYPE_ARRAY_UBOUND (type, dim)) = 1;
+ }
+ }
+
###AV: No need to do something similar for shared coarrays?
if (GFC_TYPE_ARRAY_OFFSET (type) == NULL_TREE)
{
GFC_TYPE_ARRAY_OFFSET (type) = gfc_create_var_np (gfc_array_index_type,
@@ -1202,6 +1230,10 @@ gfc_build_dummy_array_decl (gfc_symbol * sym, tree dummy)
|| (as && as->type == AS_ASSUMED_RANK))
return dummy;
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && sym->attr.codimension
+ && sym->attr.allocatable)
+ return dummy;
+
/* Add to list of variables if not a fake result variable.
These symbols are set on the symbol only, not on the class component. */
if (sym->attr.result || sym->attr.dummy)
@@ -1504,7 +1536,6 @@ add_attributes_to_decl (symbol_attribute sym_attr, tree list)
static void build_function_decl (gfc_symbol * sym, bool global);
-
/* Return the decl for a gfc_symbol, create it if it doesn't already
exist. */
@@ -1820,7 +1851,7 @@ gfc_get_symbol_decl (gfc_symbol * sym)
}
/* Remember this variable for allocation/cleanup. */
- if (sym->attr.dimension || sym->attr.allocatable || sym->attr.codimension
+ if (sym->attr.dimension || sym->attr.codimension || sym->attr.allocatable
###AV: Unnecessary!
|| (sym->ts.type == BT_CLASS &&
(CLASS_DATA (sym)->attr.dimension
|| CLASS_DATA (sym)->attr.allocatable))
@@ -1869,6 +1900,9 @@ gfc_get_symbol_decl (gfc_symbol * sym)
gcc_assert (!sym->value || sym->value->expr_type == EXPR_NULL);
}
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && sym->attr.codimension)
+ TREE_STATIC(decl) = 1;
+
gfc_finish_var_decl (decl, sym);
if (sym->ts.type == BT_CHARACTER)
@@ -3693,6 +3727,7 @@ void
gfc_build_builtin_function_decls (void)
{
tree gfc_int8_type_node = gfc_get_int_type (8);
+ tree pint_type = build_pointer_type (integer_type_node);
gfor_fndecl_stop_numeric = gfc_build_library_function_decl (
get_identifier (PREFIX("stop_numeric")),
@@ -3820,9 +3855,8 @@ gfc_build_builtin_function_decls (void)
/* Coarray library calls. */
if (flag_coarray == GFC_FCOARRAY_LIB)
{
- tree pint_type, pppchar_type;
+ tree pppchar_type;
- pint_type = build_pointer_type (integer_type_node);
pppchar_type
= build_pointer_type (build_pointer_type (pchar_type_node));
@@ -4062,6 +4096,68 @@ gfc_build_builtin_function_decls (void)
integer_type_node, 3, pvoid_type_node, integer_type_node,
pvoid_type_node);
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ gfor_fndecl_nca_master = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_master")), ".r", integer_type_node, 1,
+ build_pointer_type (build_function_type_list (void_type_node, NULL_TREE)));
+ gfor_fndecl_nca_coarray_allocate = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_coarray_alloc")), "..RRR", integer_type_node, 4,
+ pvoid_type_node, integer_type_node, integer_type_node, integer_type_node,
+ NULL_TREE);
+ gfor_fndecl_nca_coarray_free = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_coarray_free")), "..R", integer_type_node, 2,
+ pvoid_type_node, /* Pointer to the descriptor to be deallocated. */
+ integer_type_node, /* Type of allocation (normal, event, lock). */
+ NULL_TREE);
+ gfor_fndecl_nca_this_image = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_coarray_this_image")), ".X", integer_type_node, 1,
+ integer_type_node, /* This is the team number. Currently ignored. */
+ NULL_TREE);
+ DECL_PURE_P (gfor_fndecl_nca_this_image) = 1;
+ gfor_fndecl_nca_num_images = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_coarray_num_images")), ".X", integer_type_node, 1,
+ integer_type_node, /* See above. */
+ NULL_TREE);
+ DECL_PURE_P (gfor_fndecl_nca_num_images) = 1;
+ gfor_fndecl_nca_sync_all = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_coarray_sync_all")), ".X", void_type_node, 1,
+ build_pointer_type (integer_type_node), NULL_TREE);
+ gfor_fndecl_nca_sync_images = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_sync_images")), ".RRXXX", void_type_node,
+ 5, integer_type_node, pint_type, pint_type,
+ pchar_type_node, size_type_node, NULL_TREE);
+ gfor_fndecl_nca_lock = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_lock")), ".w", void_type_node, 1,
+ pvoid_type_node, NULL_TREE);
+ gfor_fndecl_nca_unlock = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_unlock")), ".w", void_type_node, 1,
+ pvoid_type_node, NULL_TREE);
+
+ gfor_fndecl_nca_reduce_scalar =
+ gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_collsub_reduce_scalar")), ".wrW",
+ void_type_node, 3, pvoid_type_node,
+ build_pointer_type (build_function_type_list (void_type_node,
+ pvoid_type_node, pvoid_type_node, NULL_TREE)),
+ pint_type, NULL_TREE);
+
+ gfor_fndecl_nca_reduce_array =
+ gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX("nca_collsub_reduce_array")), ".wrWR",
+ void_type_node, 4, pvoid_type_node,
+ build_pointer_type (build_function_type_list (void_type_node,
+ pvoid_type_node, pvoid_type_node, NULL_TREE)),
+ pint_type, integer_type_node, NULL_TREE);
+
+ gfor_fndecl_nca_broadcast_scalar = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX ("nca_collsub_broadcast_scalar")), ".w..",
+ void_type_node, 3, pvoid_type_node, size_type_node, integer_type_node);
+ gfor_fndecl_nca_broadcast_array = gfc_build_library_function_decl_with_spec (
+ get_identifier (PREFIX ("nca_collsub_broadcast_array")), ".W.",
+ void_type_node, 2, pvoid_type_node, integer_type_node);
+ }
###AV: Note, I just learned, that the spec strings ".W." have changed. There is
###AV: a space after each character. Making them something like ". W . ".
+
gfc_build_intrinsic_function_decls ();
gfc_build_intrinsic_lib_fndecls ();
@@ -4538,6 +4634,74 @@ get_proc_result (gfc_symbol* sym)
}
+void
+gfc_trans_native_coarray (stmtblock_t * init, stmtblock_t *cleanup, gfc_symbol * sym)
+{
+ tree tmp, decl;
+ tree overflow = build_int_cst (integer_type_node, 0), nelems, element_size; //All unused
###AV: "//All unused" means what? Remove what is not used anymore.
+ tree offset;
+ tree elem_size;
+ int alloc_type;
+
+ decl = sym->backend_decl;
+
+ TREE_STATIC(decl) = 1;
+
+ /* Tell the library to handle arrays of locks and event types seperatly. */
+ alloc_type = gfc_native_coarray_get_allocation_type (sym);
+
+ if (init)
+ {
+ gfc_array_init_size (decl, sym->as->rank, sym->as->corank, &offset,
+ sym->as->lower, sym->as->upper, init,
+ init, &overflow,
+ NULL_TREE, &nelems, NULL,
+ NULL_TREE, true, NULL, &element_size);
+ gfc_conv_descriptor_offset_set (init, decl, offset);
+ elem_size = size_in_bytes (gfc_get_element_type (TREE_TYPE(decl)));
+ gfc_allocate_native_coarray (init, decl, elem_size, sym->as->corank,
+ alloc_type);
+ }
+
+ if (cleanup)
+ {
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_nca_coarray_free,
+ 2, gfc_build_addr_expr (pvoid_type_node, decl),
+ build_int_cst (integer_type_node, alloc_type));
+ gfc_add_expr_to_block (cleanup, tmp);
+ }
+}
+
+static void
+finish_coarray_constructor_function (tree *, tree *);
+
+static void
+generate_coarray_constructor_function (tree *, tree *);
+
+static void
+gfc_trans_native_coarray_static (gfc_symbol * sym)
+{
+ tree save_fn_decl, fndecl;
+ generate_coarray_constructor_function (&save_fn_decl, &fndecl);
+ gfc_trans_native_coarray (&caf_init_block, NULL, sym);
+ finish_coarray_constructor_function (&save_fn_decl, &fndecl);
+}
+
+static void
+gfc_trans_native_coarray_inline (gfc_wrapped_block * block, gfc_symbol * sym)
+{
+ stmtblock_t init, cleanup;
+
+ gfc_init_block (&init);
+ gfc_init_block (&cleanup);
+
+ gfc_trans_native_coarray (&init, &cleanup, sym);
+
+ gfc_add_init_cleanup (block, gfc_finish_block (&init), gfc_finish_block (&cleanup));
+}
+
+
+
/* Generate function entry and exit code, and add it to the function body.
This includes:
Allocation and initialization of array variables.
@@ -4833,7 +4997,8 @@ gfc_trans_deferred_vars (gfc_symbol * proc_sym, gfc_wrapped_block * block)
gfc_trans_deferred_array (sym, block);
}
}
- else if (sym->attr.codimension
+ else if (flag_coarray != GFC_FCOARRAY_NATIVE
+ && sym->attr.codimension
&& TREE_STATIC (sym->backend_decl))
{
gfc_init_block (&tmpblock);
@@ -4843,6 +5008,11 @@ gfc_trans_deferred_vars (gfc_symbol * proc_sym, gfc_wrapped_block * block)
NULL_TREE);
continue;
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE
+ && sym->attr.codimension)
+ {
+ gfc_trans_native_coarray_inline (block, sym);
+ }
else
{
gfc_save_backend_locus (&loc);
@@ -5333,6 +5503,11 @@ gfc_create_module_variable (gfc_symbol * sym)
&& sym->fn_result_spec));
DECL_CONTEXT (decl) = sym->ns->proc_name->backend_decl;
rest_of_decl_compilation (decl, 1, 0);
+
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && sym->attr.codimension
+ && !sym->attr.allocatable)
+ gfc_trans_native_coarray_static (sym);
+
gfc_module_add_decl (cur_module, decl);
/* Also add length of strings. */
@@ -5730,64 +5905,82 @@ generate_coarray_sym_init (gfc_symbol *sym)
}
-/* Generate constructor function to initialize static, nonallocatable
- coarrays. */
static void
-generate_coarray_init (gfc_namespace * ns __attribute((unused)))
+generate_coarray_constructor_function (tree *save_fn_decl, tree *fndecl)
{
- tree fndecl, tmp, decl, save_fn_decl;
+ tree tmp, decl;
- save_fn_decl = current_function_decl;
+ *save_fn_decl = current_function_decl;
push_function_context ();
tmp = build_function_type_list (void_type_node, NULL_TREE);
- fndecl = build_decl (input_location, FUNCTION_DECL,
- create_tmp_var_name ("_caf_init"), tmp);
+ *fndecl = build_decl (input_location, FUNCTION_DECL,
+ create_tmp_var_name (flag_coarray == GFC_FCOARRAY_LIB ? "_caf_init" : "_nca_init"), tmp);
###AV: IMHO there is a line length limit of 80 characters!
- DECL_STATIC_CONSTRUCTOR (fndecl) = 1;
- SET_DECL_INIT_PRIORITY (fndecl, DEFAULT_INIT_PRIORITY);
+ DECL_STATIC_CONSTRUCTOR (*fndecl) = 1;
+ SET_DECL_INIT_PRIORITY (*fndecl, DEFAULT_INIT_PRIORITY);
decl = build_decl (input_location, RESULT_DECL, NULL_TREE, void_type_node);
DECL_ARTIFICIAL (decl) = 1;
DECL_IGNORED_P (decl) = 1;
- DECL_CONTEXT (decl) = fndecl;
- DECL_RESULT (fndecl) = decl;
+ DECL_CONTEXT (decl) = *fndecl;
+ DECL_RESULT (*fndecl) = decl;
- pushdecl (fndecl);
- current_function_decl = fndecl;
- announce_function (fndecl);
+ pushdecl (*fndecl);
+ current_function_decl = *fndecl;
+ announce_function (*fndecl);
- rest_of_decl_compilation (fndecl, 0, 0);
- make_decl_rtl (fndecl);
- allocate_struct_function (fndecl, false);
+ rest_of_decl_compilation (*fndecl, 0, 0);
+ make_decl_rtl (*fndecl);
+ allocate_struct_function (*fndecl, false);
pushlevel ();
gfc_init_block (&caf_init_block);
+}
- gfc_traverse_ns (ns, generate_coarray_sym_init);
+static void
+finish_coarray_constructor_function (tree *save_fn_decl, tree *fndecl)
+{
+ tree decl;
- DECL_SAVED_TREE (fndecl) = gfc_finish_block (&caf_init_block);
+ DECL_SAVED_TREE (*fndecl) = gfc_finish_block (&caf_init_block);
decl = getdecls ();
poplevel (1, 1);
- BLOCK_SUPERCONTEXT (DECL_INITIAL (fndecl)) = fndecl;
+ BLOCK_SUPERCONTEXT (DECL_INITIAL (*fndecl)) = *fndecl;
- DECL_SAVED_TREE (fndecl)
- = build3_v (BIND_EXPR, decl, DECL_SAVED_TREE (fndecl),
- DECL_INITIAL (fndecl));
- dump_function (TDI_original, fndecl);
+ DECL_SAVED_TREE (*fndecl)
+ = build3_v (BIND_EXPR, decl, DECL_SAVED_TREE (*fndecl),
+ DECL_INITIAL (*fndecl));
+ dump_function (TDI_original, *fndecl);
cfun->function_end_locus = input_location;
set_cfun (NULL);
- if (decl_function_context (fndecl))
- (void) cgraph_node::create (fndecl);
+ if (decl_function_context (*fndecl))
+ (void) cgraph_node::create (*fndecl);
else
- cgraph_node::finalize_function (fndecl, true);
+ cgraph_node::finalize_function (*fndecl, true);
pop_function_context ();
- current_function_decl = save_fn_decl;
+ current_function_decl = *save_fn_decl;
+}
+
+/* Generate constructor function to initialize static, nonallocatable
+ coarrays. */
+
+static void
+generate_coarray_init (gfc_namespace * ns)
+{
+ tree save_fn_decl, fndecl;
+
+ generate_coarray_constructor_function (&save_fn_decl, &fndecl);
+
+ gfc_traverse_ns (ns, generate_coarray_sym_init);
+
+ finish_coarray_constructor_function (&save_fn_decl, &fndecl);
+
}
@@ -6470,7 +6663,11 @@ create_main_function (tree fndecl)
}
/* Call MAIN__(). */
- tmp = build_call_expr_loc (input_location,
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_nca_master, 1,
+ gfc_build_addr_expr (NULL, fndecl));
+ else
+ tmp = build_call_expr_loc (input_location,
fndecl, 0);
gfc_add_expr_to_block (&body, tmp);
diff --git a/gcc/fortran/trans-expr.c b/gcc/fortran/trans-expr.c
index 36ff9b5cbc6..99799801fcb 100644
--- a/gcc/fortran/trans-expr.c
+++ b/gcc/fortran/trans-expr.c
@@ -2622,8 +2622,14 @@ gfc_maybe_dereference_var (gfc_symbol *sym, tree var, bool descriptor_only_p,
}
else if (!sym->attr.value)
{
+
+ /* Do not derefernce native coarray dummies. */
+ if (false && flag_coarray == GFC_FCOARRAY_NATIVE
+ && sym->attr.codimension && sym->attr.dummy)
+ return var;
+
###AV: Unused code! Remove! 'if (false && ...' can never eval to true.
/* Dereference temporaries for class array dummy arguments. */
- if (sym->attr.dummy && is_classarray
+ else if (sym->attr.dummy && is_classarray
&& GFC_ARRAY_TYPE_P (TREE_TYPE (var)))
{
if (!descriptor_only_p)
@@ -2635,6 +2641,7 @@ gfc_maybe_dereference_var (gfc_symbol *sym, tree var, bool descriptor_only_p,
/* Dereference non-character scalar dummy arguments. */
if (sym->attr.dummy && !sym->attr.dimension
&& !(sym->attr.codimension && sym->attr.allocatable)
+ && !(sym->attr.codimension && flag_coarray == GFC_FCOARRAY_NATIVE)
&& (sym->ts.type != BT_CLASS
|| (!CLASS_DATA (sym)->attr.dimension
&& !(CLASS_DATA (sym)->attr.codimension
@@ -2670,6 +2677,7 @@ gfc_maybe_dereference_var (gfc_symbol *sym, tree var, bool descriptor_only_p,
|| CLASS_DATA (sym)->attr.allocatable
|| CLASS_DATA (sym)->attr.class_pointer))
var = build_fold_indirect_ref_loc (input_location, var);
+
/* And the case where a non-dummy, non-result, non-function,
non-allotable and non-pointer classarray is present. This case was
previously covered by the first if, but with introducing the
@@ -5528,7 +5536,10 @@ gfc_conv_procedure_call (gfc_se * se, gfc_symbol * sym,
nodesc_arg = nodesc_arg || !comp->attr.always_explicit;
else
nodesc_arg = nodesc_arg || !sym->attr.always_explicit;
-
+#if 0
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && fsym->attr.codimension)
+ nodesc_arg = false;
+#endif
###AV: Fix or remove!
/* Class array expressions are sometimes coming completely unadorned
with either arrayspec or _data component. Correct that here.
OOP-TODO: Move this to the frontend. */
@@ -5720,7 +5731,10 @@ gfc_conv_procedure_call (gfc_se * se, gfc_symbol * sym,
parmse.want_coarray = 1;
scalar = false;
}
-
+#if 0
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && fsym->attr.codimension)
+ scalar = false;
+#endif
###AV: Fix or remove!
/* A scalar or transformational function. */
if (scalar)
{
@@ -6233,7 +6247,7 @@ gfc_conv_procedure_call (gfc_se * se, gfc_symbol * sym,
else
gfc_conv_array_parameter (&parmse, e, nodesc_arg, fsym,
sym->name, NULL);
-
+
###AV: hu?
/* Unallocated allocatable arrays and unassociated pointer arrays
need their dtype setting if they are argument associated with
assumed rank dummies. */
diff --git a/gcc/fortran/trans-intrinsic.c b/gcc/fortran/trans-intrinsic.c
index 32fe9886c57..b4183217f49 100644
--- a/gcc/fortran/trans-intrinsic.c
+++ b/gcc/fortran/trans-intrinsic.c
@@ -41,6 +41,7 @@ along with GCC; see the file COPYING3. If not see
#include "trans-array.h"
#include "dependency.h" /* For CAF array alias analysis. */
/* Only for gfc_trans_assign and gfc_trans_pointer_assign. */
+#include "trans-stmt.h"
/* This maps Fortran intrinsic math functions to external library or GCC
builtin functions. */
@@ -2363,7 +2364,6 @@ conv_caf_send (gfc_code *code) {
return gfc_finish_block (&block);
}
-
static void
trans_this_image (gfc_se * se, gfc_expr *expr)
{
@@ -2394,14 +2394,18 @@ trans_this_image (gfc_se * se, gfc_expr *expr)
}
else
tmp = integer_zero_node;
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_this_image, 1,
- tmp);
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE ?
+ gfor_fndecl_nca_this_image :
+ gfor_fndecl_caf_this_image,
+ 1, tmp);
se->expr = fold_convert (gfc_get_int_type (gfc_default_integer_kind),
tmp);
return;
}
/* Coarray-argument version: THIS_IMAGE(coarray [, dim]). */
+ /* TODO: NCA handle native coarrays. */
###AV: What?
type = gfc_get_int_type (gfc_default_integer_kind);
corank = gfc_get_corank (expr->value.function.actual->expr);
@@ -2490,8 +2494,11 @@ trans_this_image (gfc_se * se, gfc_expr *expr)
*/
/* this_image () - 1. */
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_this_image, 1,
- integer_zero_node);
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE
+ ? gfor_fndecl_nca_this_image
+ : gfor_fndecl_caf_this_image,
###AV: Style: Align the colon beneath the '?'.
+ 1, integer_zero_node);
tmp = fold_build2_loc (input_location, MINUS_EXPR, type,
fold_convert (type, tmp), build_int_cst (type, 1));
if (corank == 1)
@@ -2774,7 +2781,10 @@ trans_image_index (gfc_se * se, gfc_expr *expr)
num_images = build_int_cst (type, 1);
else
{
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_num_images, 2,
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE
+ ? gfor_fndecl_nca_num_images
+ : gfor_fndecl_caf_num_images, 2,
integer_zero_node,
build_int_cst (integer_type_node, -1));
num_images = fold_convert (type, tmp);
@@ -2819,8 +2829,13 @@ trans_num_images (gfc_se * se, gfc_expr *expr)
}
else
failed = build_int_cst (integer_type_node, -1);
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_num_images, 2,
- distance, failed);
+
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_nca_num_images, 1,
+ distance);
+ else
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_num_images, 2,
+ distance, failed);
se->expr = fold_convert (gfc_get_int_type (gfc_default_integer_kind), tmp);
}
@@ -3264,7 +3279,10 @@ conv_intrinsic_cobound (gfc_se * se, gfc_expr * expr)
tree cosize;
cosize = gfc_conv_descriptor_cosize (desc, arg->expr->rank, corank);
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_num_images,
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE
+ ? gfor_fndecl_nca_num_images
+ : gfor_fndecl_caf_num_images,
2, integer_zero_node,
build_int_cst (integer_type_node, -1));
tmp = fold_build2_loc (input_location, MINUS_EXPR,
@@ -3280,7 +3298,9 @@ conv_intrinsic_cobound (gfc_se * se, gfc_expr * expr)
else if (flag_coarray != GFC_FCOARRAY_SINGLE)
{
/* ubound = lbound + num_images() - 1. */
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_num_images,
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE ? gfor_fndecl_nca_num_images :
+ gfor_fndecl_caf_num_images,
###AV: Style!
2, integer_zero_node,
build_int_cst (integer_type_node, -1));
tmp = fold_build2_loc (input_location, MINUS_EXPR,
@@ -11004,6 +11024,136 @@ gfc_walk_intrinsic_function (gfc_ss * ss, gfc_expr * expr,
}
}
+/* Helper function - advance to the next argument. */
+
+static tree
+trans_argument (gfc_actual_arglist **curr_al, stmtblock_t *blk,
###AV: I don't like the name. This name implies that only one argument is
###AV: translated, but it is doing so much more
+ stmtblock_t *postblk, gfc_se *argse, tree def)
+{
+ if (!(*curr_al)->expr)
+ return def;
+ if ((*curr_al)->expr->rank > 0)
###AV: Why is this testing for rank only and not also for corank?
+ gfc_conv_expr_descriptor (argse, (*curr_al)->expr);
+ else
+ gfc_conv_expr (argse, (*curr_al)->expr);
+ gfc_add_block_to_block (blk, &argse->pre);
+ gfc_add_block_to_block (postblk, &argse->post);
+ *curr_al = (*curr_al)->next;
+ return argse->expr;
+}
+
+/* Convert CO_REDUCE for native coarrays. */
+
+static tree
+conv_nca_reduce (gfc_code *code, stmtblock_t *blk, stmtblock_t *postblk)
+{
+ gfc_actual_arglist *curr_al;
+ tree var, reduce_op, result_image, elem_size;
+ gfc_se argse;
+ int is_array;
+
+ curr_al = code->ext.actual;
+
+ gfc_init_se (&argse, NULL);
+ argse.want_pointer = 1;
+ is_array = curr_al->expr->rank > 0;
###AV: Move the argse treatment into the trans_argument and stop repeating
###AV: yourself.
+ var = trans_argument (&curr_al, blk, postblk, &argse, NULL_TREE);
+
+ gfc_init_se (&argse, NULL);
+ argse.want_pointer = 1;
+ reduce_op = trans_argument (&curr_al, blk, postblk, &argse, NULL_TREE);
+
+ gfc_init_se (&argse, NULL);
+ argse.want_pointer = 1;
+ result_image = trans_argument (&curr_al, blk, postblk, &argse,
+ null_pointer_node);
+
+ if (is_array)
+ return build_call_expr_loc (input_location, gfor_fndecl_nca_reduce_array,
+ 3, var, reduce_op, result_image);
+
+ elem_size = size_in_bytes(TREE_TYPE(TREE_TYPE(var)));
+ return build_call_expr_loc (input_location, gfor_fndecl_nca_reduce_scalar, 4,
+ var, elem_size, reduce_op, result_image);
+}
+
+static tree
+conv_nca_broadcast (gfc_code *code, stmtblock_t *blk, stmtblock_t *postblk)
+{
+ gfc_actual_arglist *curr_al;
+ tree var, source_image, elem_size;
+ gfc_se argse;
+ int is_array;
+
+ curr_al = code->ext.actual;
+
+ gfc_init_se (&argse, NULL);
+ argse.want_pointer = 1;
+ is_array = curr_al->expr->rank > 0;
+ var = trans_argument (&curr_al, blk, postblk, &argse, NULL_TREE);
+
+ gfc_init_se (&argse, NULL);
+ argse.want_pointer = 0;
+ source_image = trans_argument (&curr_al, blk, postblk, &argse, NULL_TREE);
+
+ if (is_array)
+ return build_call_expr_loc (input_location, gfor_fndecl_nca_broadcast_array,
+ 2, var, source_image);
+
+ elem_size = size_in_bytes(TREE_TYPE(TREE_TYPE(var)));
+ return build_call_expr_loc (input_location, gfor_fndecl_nca_broadcast_scalar,
+ 3, var, elem_size, source_image);
+}
+
+static tree conv_co_collective (gfc_code *);
+
+/* Convert collective subroutines for native coarrays. */
+
+static tree
+conv_nca_collective (gfc_code *code)
+{
+
+ switch (code->resolved_isym->id)
+ {
+ case GFC_ISYM_CO_REDUCE:
+ {
+ stmtblock_t block, postblock;
+ tree fcall;
+
+ gfc_start_block (&block);
+ gfc_init_block (&postblock);
+ fcall = conv_nca_reduce (code, &block, &postblock);
+ gfc_add_expr_to_block (&block, fcall);
+ gfc_add_block_to_block (&block, &postblock);
+ return gfc_finish_block (&block);
+ }
+ case GFC_ISYM_CO_SUM:
+ case GFC_ISYM_CO_MIN:
+ case GFC_ISYM_CO_MAX:
+ return gfc_trans_call (code, false, NULL_TREE, NULL_TREE, false);
+
+ case GFC_ISYM_CO_BROADCAST:
+ {
+ stmtblock_t block, postblock;
+ tree fcall;
+
+ gfc_start_block (&block);
+ gfc_init_block (&postblock);
+ fcall = conv_nca_broadcast (code, &block, &postblock);
+ gfc_add_expr_to_block (&block, fcall);
+ gfc_add_block_to_block (&block, &postblock);
+ return gfc_finish_block (&block);
+ }
+#if 0
+ case GFC_ISYM_CO_BROADCAST:
+ return conv_co_collective (code);
+#endif
###AV: Fix!!!
+ default:
+ gfc_internal_error ("Invalid or unsupported isym");
+ break;
+ }
+}
+
static tree
conv_co_collective (gfc_code *code)
{
@@ -11111,7 +11261,13 @@ conv_co_collective (gfc_code *code)
errmsg_len = build_zero_cst (size_type_node);
}
+ /* For native coarrays, we only come here for CO_BROADCAST. */
+
+ gcc_assert (code->resolved_isym->id == GFC_ISYM_CO_BROADCAST
+ || flag_coarray != GFC_FCOARRAY_NATIVE);
+
/* Generate the function call. */
+
switch (code->resolved_isym->id)
{
case GFC_ISYM_CO_BROADCAST:
@@ -12104,7 +12260,10 @@ gfc_conv_intrinsic_subroutine (gfc_code *code)
case GFC_ISYM_CO_MAX:
case GFC_ISYM_CO_REDUCE:
case GFC_ISYM_CO_SUM:
- res = conv_co_collective (code);
+ if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ res = conv_nca_collective (code);
+ else
+ res = conv_co_collective (code);
break;
case GFC_ISYM_FREE:
diff --git a/gcc/fortran/trans-stmt.c b/gcc/fortran/trans-stmt.c
index 1f183b9dcd0..4897fa10d9d 100644
--- a/gcc/fortran/trans-stmt.c
+++ b/gcc/fortran/trans-stmt.c
@@ -830,7 +830,9 @@ gfc_trans_lock_unlock (gfc_code *code, gfc_exec_op op)
/* Short cut: For single images without STAT= or LOCK_ACQUIRED
return early. (ERRMSG= is always untouched for -fcoarray=single.) */
- if (!code->expr2 && !code->expr4 && flag_coarray != GFC_FCOARRAY_LIB)
+ if (!code->expr2 && !code->expr4
+ && !(flag_coarray == GFC_FCOARRAY_LIB
+ || flag_coarray == GFC_FCOARRAY_NATIVE))
return NULL_TREE;
if (code->expr2)
@@ -990,6 +992,29 @@ gfc_trans_lock_unlock (gfc_code *code, gfc_exec_op op)
return gfc_finish_block (&se.pre);
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ gfc_se arg;
+ stmtblock_t res;
+ tree call;
+ tree tmp;
+
+ gfc_init_se (&arg, NULL);
+ gfc_start_block (&res);
+ gfc_conv_expr (&arg, code->expr1);
+ gfc_add_block_to_block (&res, &arg.pre);
+ call = build_call_expr_loc (input_location, op == EXEC_LOCK ?
+ gfor_fndecl_nca_lock
+ : gfor_fndecl_nca_unlock,
+ 1, fold_convert (pvoid_type_node,
+ gfc_build_addr_expr (NULL, arg.expr)));
+ gfc_add_expr_to_block (&res, call);
+ gfc_add_block_to_block (&res, &arg.post);
+ tmp = gfc_trans_memory_barrier ();
+ gfc_add_expr_to_block (&res, tmp);
+
+ return gfc_finish_block (&res);
+ }
if (stat != NULL_TREE)
gfc_add_modify (&se.pre, stat, build_int_cst (TREE_TYPE (stat), 0));
@@ -1183,7 +1208,8 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
/* Short cut: For single images without bound checking or without STAT=,
return early. (ERRMSG= is always untouched for -fcoarray=single.) */
if (!code->expr2 && !(gfc_option.rtcheck & GFC_RTCHECK_BOUNDS)
- && flag_coarray != GFC_FCOARRAY_LIB)
+ && flag_coarray != GFC_FCOARRAY_LIB
+ && flag_coarray != GFC_FCOARRAY_NATIVE)
###AV: How about adding a helper for this kind of tests, like
###AV: #define IS_NON_SINGLE_CAF (flag_coarray == GFC_FCOARRAY_LIB || flag_coarray == GFC_FCOARRAY_NATIVE)
###AV: and using that everywhere?
return NULL_TREE;
gfc_init_se (&se, NULL);
@@ -1206,7 +1232,7 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
else
stat = null_pointer_node;
- if (code->expr3 && flag_coarray == GFC_FCOARRAY_LIB)
+ if (code->expr3 && (flag_coarray == GFC_FCOARRAY_LIB || flag_coarray == GFC_FCOARRAY_NATIVE))
{
gcc_assert (code->expr3->expr_type == EXPR_VARIABLE);
gfc_init_se (&argse, NULL);
@@ -1216,7 +1242,7 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
errmsg = gfc_build_addr_expr (NULL, argse.expr);
errmsglen = fold_convert (size_type_node, argse.string_length);
}
- else if (flag_coarray == GFC_FCOARRAY_LIB)
+ else if (flag_coarray == GFC_FCOARRAY_LIB || flag_coarray == GFC_FCOARRAY_NATIVE)
{
errmsg = null_pointer_node;
errmsglen = build_int_cst (size_type_node, 0);
@@ -1229,7 +1255,7 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
{
tree images2 = fold_convert (integer_type_node, images);
tree cond;
- if (flag_coarray != GFC_FCOARRAY_LIB)
+ if (flag_coarray != GFC_FCOARRAY_LIB && flag_coarray != GFC_FCOARRAY_NATIVE)
cond = fold_build2_loc (input_location, NE_EXPR, logical_type_node,
images, build_int_cst (TREE_TYPE (images), 1));
else
@@ -1253,17 +1279,13 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
/* Per F2008, 8.5.1, a SYNC MEMORY is implied by calling the
image control statements SYNC IMAGES and SYNC ALL. */
- if (flag_coarray == GFC_FCOARRAY_LIB)
+ if (flag_coarray == GFC_FCOARRAY_LIB || flag_coarray == GFC_FCOARRAY_NATIVE)
{
- tmp = gfc_build_string_const (strlen ("memory")+1, "memory"),
- tmp = build5_loc (input_location, ASM_EXPR, void_type_node,
- gfc_build_string_const (1, ""), NULL_TREE, NULL_TREE,
- tree_cons (NULL_TREE, tmp, NULL_TREE), NULL_TREE);
- ASM_VOLATILE_P (tmp) = 1;
+ tmp = gfc_trans_memory_barrier ();
gfc_add_expr_to_block (&se.pre, tmp);
}
- if (flag_coarray != GFC_FCOARRAY_LIB)
+ if (flag_coarray != GFC_FCOARRAY_LIB && flag_coarray != GFC_FCOARRAY_NATIVE)
{
/* Set STAT to zero. */
if (code->expr2)
@@ -1285,8 +1307,14 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_memory,
3, stat, errmsg, errmsglen);
else
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_all,
- 3, stat, errmsg, errmsglen);
+ {
+ if (flag_coarray == GFC_FCOARRAY_LIB)
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_all,
+ 3, stat, errmsg, errmsglen);
+ else
+ tmp = build_call_expr_loc (input_location, gfor_fndecl_nca_sync_all,
+ 1, stat);
+ }
gfc_add_expr_to_block (&se.pre, tmp);
}
@@ -1351,7 +1379,10 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
if (TREE_TYPE (stat) == integer_type_node)
stat = gfc_build_addr_expr (NULL, stat);
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_images,
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE
+ ? gfor_fndecl_nca_sync_images
+ : gfor_fndecl_caf_sync_images,
5, fold_convert (integer_type_node, len),
images, stat, errmsg, errmsglen);
gfc_add_expr_to_block (&se.pre, tmp);
@@ -1360,7 +1391,10 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
{
tree tmp_stat = gfc_create_var (integer_type_node, "stat");
- tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_images,
+ tmp = build_call_expr_loc (input_location,
+ flag_coarray == GFC_FCOARRAY_NATIVE
+ ? gfor_fndecl_nca_sync_images
+ : gfor_fndecl_caf_sync_images,
5, fold_convert (integer_type_node, len),
images, gfc_build_addr_expr (NULL, tmp_stat),
errmsg, errmsglen);
@@ -1596,6 +1630,11 @@ gfc_trans_critical (gfc_code *code)
gfc_add_expr_to_block (&block, tmp);
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ tmp = gfc_trans_lock_unlock (code, EXEC_LOCK);
+ gfc_add_expr_to_block (&block, tmp);
+ }
tmp = gfc_trans_code (code->block->next);
gfc_add_expr_to_block (&block, tmp);
@@ -1620,6 +1659,11 @@ gfc_trans_critical (gfc_code *code)
gfc_add_expr_to_block (&block, tmp);
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ tmp = gfc_trans_lock_unlock (code, EXEC_UNLOCK);
+ gfc_add_expr_to_block (&block, tmp);
+ }
return gfc_finish_block (&block);
}
@@ -7169,6 +7213,7 @@ gfc_trans_deallocate (gfc_code *code)
tree apstat, pstat, stat, errmsg, errlen, tmp;
tree label_finish, label_errmsg;
stmtblock_t block;
+ bool is_native_coarray = false;
pstat = apstat = stat = errmsg = errlen = tmp = NULL_TREE;
label_finish = label_errmsg = NULL_TREE;
@@ -7254,8 +7299,27 @@ gfc_trans_deallocate (gfc_code *code)
? GFC_STRUCTURE_CAF_MODE_DEALLOC_ONLY : 0);
}
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ gfc_ref *ref, *last;
- if (expr->rank || is_coarray_array)
+ for (ref = expr->ref, last = ref; ref; last = ref, ref = ref->next);
+ ref = last;
+ if (ref->type == REF_ARRAY && ref->u.ar.codimen)
+ {
+ gfc_symbol *sym = expr->symtree->n.sym;
+ int alloc_type = gfc_native_coarray_get_allocation_type (sym);
+ tmp = build_call_expr_loc (input_location,
+ gfor_fndecl_nca_coarray_free,
+ 2, gfc_build_addr_expr (pvoid_type_node, se.expr),
+ build_int_cst (integer_type_node,
+ alloc_type));
+ gfc_add_expr_to_block (&block, tmp);
+ is_native_coarray = true;
+ }
+ }
+
+ if ((expr->rank || is_coarray_array) && !is_native_coarray)
{
gfc_ref *ref;
@@ -7344,7 +7408,7 @@ gfc_trans_deallocate (gfc_code *code)
gfc_reset_len (&se.pre, al->expr);
}
}
- else
+ else if (!is_native_coarray)
{
tmp = gfc_deallocate_scalar_with_status (se.expr, pstat, label_finish,
false, al->expr,
diff --git a/gcc/fortran/trans-types.c b/gcc/fortran/trans-types.c
index 26fdb2803a7..f100d34d65b 100644
--- a/gcc/fortran/trans-types.c
+++ b/gcc/fortran/trans-types.c
@@ -1345,6 +1345,10 @@ gfc_is_nodesc_array (gfc_symbol * sym)
gcc_assert (array_attr->dimension || array_attr->codimension);
+ /* We need a descriptor for native coarrays. */
###AV: Style! Comments end with two spaces after the '.', with a tab and a space.
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && sym->as && sym->as->corank)
+ return 0;
+
/* We only want local arrays. */
if ((sym->ts.type != BT_CLASS && sym->attr.pointer)
|| (sym->ts.type == BT_CLASS && CLASS_DATA (sym)->attr.class_pointer)
@@ -1381,12 +1385,18 @@ gfc_build_array_type (tree type, gfc_array_spec * as,
tree ubound[GFC_MAX_DIMENSIONS];
int n, corank;
- /* Assumed-shape arrays do not have codimension information stored in the
- descriptor. */
- corank = MAX (as->corank, codim);
- if (as->type == AS_ASSUMED_SHAPE ||
- (as->type == AS_ASSUMED_RANK && akind == GFC_ARRAY_ALLOCATABLE))
- corank = codim;
+ /* For -fcoarray=lib, assumed-shape arrays do not have codimension
+ information stored in the descriptor. */
+ if (flag_coarray != GFC_FCOARRAY_NATIVE)
+ {
+ corank = MAX (as->corank, codim);
+
+ if (as->type == AS_ASSUMED_SHAPE ||
+ (as->type == AS_ASSUMED_RANK && akind == GFC_ARRAY_ALLOCATABLE))
+ corank = codim;
+ }
+ else
+ corank = as->corank;
if (as->type == AS_ASSUMED_RANK)
for (n = 0; n < GFC_MAX_DIMENSIONS; n++)
@@ -1427,7 +1437,7 @@ gfc_build_array_type (tree type, gfc_array_spec * as,
corank, lbound, ubound, 0, akind,
restricted);
}
-\f
+
/* Returns the struct descriptor_dimension type. */
static tree
@@ -1598,7 +1608,7 @@ gfc_get_nodesc_array_type (tree etype, gfc_array_spec * as, gfc_packed packed,
/* We don't use build_array_type because this does not include
lang-specific information (i.e. the bounds of the array) when checking
for duplicates. */
- if (as->rank)
+ if (as->rank || (flag_coarray == GFC_FCOARRAY_NATIVE && as->corank))
type = make_node (ARRAY_TYPE);
else
type = build_variant_type_copy (etype);
@@ -1665,6 +1675,7 @@ gfc_get_nodesc_array_type (tree etype, gfc_array_spec * as, gfc_packed packed,
if (packed == PACKED_NO || packed == PACKED_PARTIAL)
known_stride = 0;
}
+
for (n = as->rank; n < as->rank + as->corank; n++)
{
expr = as->lower[n];
@@ -1672,7 +1683,7 @@ gfc_get_nodesc_array_type (tree etype, gfc_array_spec * as, gfc_packed packed,
tmp = gfc_conv_mpz_to_tree (expr->value.integer,
gfc_index_integer_kind);
else
- tmp = NULL_TREE;
+ tmp = NULL_TREE;
GFC_TYPE_ARRAY_LBOUND (type, n) = tmp;
expr = as->upper[n];
@@ -1680,16 +1691,16 @@ gfc_get_nodesc_array_type (tree etype, gfc_array_spec * as, gfc_packed packed,
tmp = gfc_conv_mpz_to_tree (expr->value.integer,
gfc_index_integer_kind);
else
- tmp = NULL_TREE;
+ tmp = NULL_TREE;
if (n < as->rank + as->corank - 1)
- GFC_TYPE_ARRAY_UBOUND (type, n) = tmp;
+ GFC_TYPE_ARRAY_UBOUND (type, n) = tmp;
}
- if (known_offset)
- {
- GFC_TYPE_ARRAY_OFFSET (type) =
- gfc_conv_mpz_to_tree (offset, gfc_index_integer_kind);
- }
+ if (flag_coarray == GFC_FCOARRAY_NATIVE && as->rank == 0 && as->corank != 0)
+ GFC_TYPE_ARRAY_OFFSET (type) = NULL_TREE;
+ else if (known_offset)
+ GFC_TYPE_ARRAY_OFFSET (type) =
+ gfc_conv_mpz_to_tree (offset, gfc_index_integer_kind);
else
GFC_TYPE_ARRAY_OFFSET (type) = NULL_TREE;
@@ -1714,7 +1725,7 @@ gfc_get_nodesc_array_type (tree etype, gfc_array_spec * as, gfc_packed packed,
build_qualified_type (GFC_TYPE_ARRAY_DATAPTR_TYPE (type),
TYPE_QUAL_RESTRICT);
- if (as->rank == 0)
+ if (as->rank == 0 && (flag_coarray != GFC_FCOARRAY_NATIVE || as->corank == 0))
{
if (packed != PACKED_STATIC || flag_coarray == GFC_FCOARRAY_LIB)
{
@@ -1982,7 +1993,7 @@ gfc_get_array_type_bounds (tree etype, int dimen, int codimen, tree * lbound,
/* TODO: known offsets for descriptors. */
GFC_TYPE_ARRAY_OFFSET (fat_type) = NULL_TREE;
- if (dimen == 0)
+ if (flag_coarray != GFC_FCOARRAY_NATIVE && dimen == 0)
{
arraytype = build_pointer_type (etype);
if (restricted)
@@ -2281,6 +2292,10 @@ gfc_sym_type (gfc_symbol * sym)
: GFC_ARRAY_POINTER;
else if (sym->attr.allocatable)
akind = GFC_ARRAY_ALLOCATABLE;
+
+ /* FIXME: For normal coarrays, we pass a bool to an int here.
+ Is this really intended? */
+
type = gfc_build_array_type (type, sym->as, akind, restricted,
sym->attr.contiguous, false);
}
diff --git a/gcc/fortran/trans.c b/gcc/fortran/trans.c
index ed054261452..7d9cd324828 100644
--- a/gcc/fortran/trans.c
+++ b/gcc/fortran/trans.c
@@ -33,6 +33,7 @@ along with GCC; see the file COPYING3. If not see
#include "trans-array.h"
#include "trans-types.h"
#include "trans-const.h"
+#include "diagnostic-core.h"
/* Naming convention for backend interface code:
@@ -47,6 +48,21 @@ static gfc_file *gfc_current_backend_file;
const char gfc_msg_fault[] = N_("Array reference out of bounds");
const char gfc_msg_wrong_return[] = N_("Incorrect function return value");
+/* Insert a memory barrier into the code. */
+
+tree
+gfc_trans_memory_barrier (void)
+{
+ tree tmp;
+
+ tmp = gfc_build_string_const (strlen ("memory")+1, "memory"),
###AV: That line has to end in a ';' and not a comma. I know it's a copy, but
###AV: the comma although legal here, is not what is meant.
+ tmp = build5_loc (input_location, ASM_EXPR, void_type_node,
###AV: The indent needs to be fixed, too.
+ gfc_build_string_const (1, ""), NULL_TREE, NULL_TREE,
+ tree_cons (NULL_TREE, tmp, NULL_TREE), NULL_TREE);
+ ASM_VOLATILE_P (tmp) = 1;
+
+ return tmp;
+}
/* Return a location_t suitable for 'tree' for a gfortran locus. The way the
parser works in gfortran, loc->lb->location contains only the line number
@@ -403,15 +419,16 @@ gfc_build_array_ref (tree base, tree offset, tree decl, tree vptr)
tree tmp;
tree span = NULL_TREE;
- if (GFC_ARRAY_TYPE_P (type) && GFC_TYPE_ARRAY_RANK (type) == 0)
+ if (GFC_ARRAY_TYPE_P (type) && GFC_TYPE_ARRAY_RANK (type) == 0
+ && flag_coarray != GFC_FCOARRAY_NATIVE)
{
gcc_assert (GFC_TYPE_ARRAY_CORANK (type) > 0);
return fold_convert (TYPE_MAIN_VARIANT (type), base);
}
- /* Scalar coarray, there is nothing to do. */
- if (TREE_CODE (type) != ARRAY_TYPE)
+ /* Scalar library coarray, there is nothing to do. */
+ if (TREE_CODE (type) != ARRAY_TYPE && flag_coarray != GFC_FCOARRAY_NATIVE)
{
gcc_assert (decl == NULL_TREE);
gcc_assert (integer_zerop (offset));
@@ -1335,9 +1352,10 @@ gfc_deallocate_with_status (tree pointer, tree status, tree errmsg,
tree cond, tmp, error;
tree status_type = NULL_TREE;
tree token = NULL_TREE;
+ tree orig_desc = NULL_TREE;
gfc_coarray_deregtype caf_dereg_type = GFC_CAF_COARRAY_DEREGISTER;
- if (coarray_dealloc_mode >= GFC_CAF_COARRAY_ANALYZE)
+ if (coarray_dealloc_mode >= GFC_CAF_COARRAY_ANALYZE )
{
if (flag_coarray == GFC_FCOARRAY_LIB)
{
@@ -1358,7 +1376,7 @@ gfc_deallocate_with_status (tree pointer, tree status, tree errmsg,
{
gcc_assert (GFC_ARRAY_TYPE_P (caf_type)
&& GFC_TYPE_ARRAY_CAF_TOKEN (caf_type)
- != NULL_TREE);
+ != NULL_TREE);
###AV: The old indent was correct!
token = GFC_TYPE_ARRAY_CAF_TOKEN (caf_type);
}
}
@@ -1374,6 +1392,11 @@ gfc_deallocate_with_status (tree pointer, tree status, tree errmsg,
else
caf_dereg_type = (enum gfc_coarray_deregtype) coarray_dealloc_mode;
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE)
+ {
+ orig_desc = pointer;
+ pointer = gfc_conv_descriptor_data_get (pointer);
+ }
else if (flag_coarray == GFC_FCOARRAY_SINGLE)
pointer = gfc_conv_descriptor_data_get (pointer);
}
@@ -1425,7 +1448,7 @@ gfc_deallocate_with_status (tree pointer, tree status, tree errmsg,
gfc_add_expr_to_block (&non_null, add_when_allocated);
gfc_add_finalizer_call (&non_null, expr);
if (coarray_dealloc_mode == GFC_CAF_COARRAY_NOCOARRAY
- || flag_coarray != GFC_FCOARRAY_LIB)
+ || (flag_coarray != GFC_FCOARRAY_LIB && flag_coarray != GFC_FCOARRAY_NATIVE))
{
tmp = build_call_expr_loc (input_location,
builtin_decl_explicit (BUILT_IN_FREE), 1,
@@ -1453,6 +1476,19 @@ gfc_deallocate_with_status (tree pointer, tree status, tree errmsg,
gfc_add_expr_to_block (&non_null, tmp);
}
}
+ else if (flag_coarray == GFC_FCOARRAY_NATIVE
+ && coarray_dealloc_mode >= GFC_CAF_COARRAY_ANALYZE)
+ {
+ tmp = build_call_expr_loc(input_location, gfor_fndecl_nca_coarray_free,
+ 2, gfc_build_addr_expr (pvoid_type_node, orig_desc),
+ build_int_cst(integer_type_node, GFC_NCA_NORMAL_COARRAY));
+ gfc_add_expr_to_block (&non_null, tmp);
+ gfc_add_modify (&non_null, pointer, build_int_cst (TREE_TYPE (pointer),
+ 0));
+
+ if (status != NULL_TREE && !integer_zerop(status))
+ sorry("Status not yet implemented");
###AV: Fix!!!
+ }
else
{
tree cond2, pstat = null_pointer_node;
diff --git a/gcc/fortran/trans.h b/gcc/fortran/trans.h
index d257963d5f8..974785f3f10 100644
--- a/gcc/fortran/trans.h
+++ b/gcc/fortran/trans.h
@@ -501,6 +501,9 @@ void gfc_conv_expr_reference (gfc_se * se, gfc_expr * expr,
bool add_clobber = false);
void gfc_conv_expr_type (gfc_se * se, gfc_expr *, tree);
+/* Insert a memory barrier into the code. */
+
###AV: Remove the empty line, because the comment belongs to the routine.
+tree gfc_trans_memory_barrier (void);
/* trans-expr.c */
tree gfc_conv_scalar_to_descriptor (gfc_se *, tree, symbol_attribute);
@@ -890,6 +893,21 @@ extern GTY(()) tree gfor_fndecl_co_reduce;
extern GTY(()) tree gfor_fndecl_co_sum;
extern GTY(()) tree gfor_fndecl_caf_is_present;
+
+/* Native coarray library function decls. */
+extern GTY(()) tree gfor_fndecl_nca_this_image;
+extern GTY(()) tree gfor_fndecl_nca_num_images;
+extern GTY(()) tree gfor_fndecl_nca_coarray_allocate;
+extern GTY(()) tree gfor_fndecl_nca_coarray_free;
+extern GTY(()) tree gfor_fndecl_nca_sync_images;
+extern GTY(()) tree gfor_fndecl_nca_sync_all;
+extern GTY(()) tree gfor_fndecl_nca_lock;
+extern GTY(()) tree gfor_fndecl_nca_unlock;
+extern GTY(()) tree gfor_fndecl_nca_reduce_scalar;
+extern GTY(()) tree gfor_fndecl_nca_reduce_array;
+extern GTY(()) tree gfor_fndecl_nca_broadcast_scalar;
+extern GTY(()) tree gfor_fndecl_nca_broadcast_array;
+
/* Math functions. Many other math functions are handled in
trans-intrinsic.c. */
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!) [Review part 2]
2020-10-12 13:48 ` [RFC] Native Coarrays (finally!) [Review part 2] Andre Vehreschild
@ 2020-10-13 12:42 ` Nicolas König
0 siblings, 0 replies; 17+ messages in thread
From: Nicolas König @ 2020-10-13 12:42 UTC (permalink / raw)
To: Andre Vehreschild; +Cc: fortran
Hi Andre,
Thanks for the reviews! I'll incorporate the comments this weekend (I'll
also implement stat & errmsg then).
Sorry about all of these #if 0, I completely forgot they were still in
there :)
Kind regards
Nicolas König
On 12/10/2020 15:48, Andre Vehreschild wrote:
> Hi Nicolas,
>
> here is part two of the review of the compiler components. I will do the
> testsuite and library parts another day, because now I am already completely
> bonkers. (Yes, I know, that's normal for me :-)
>
> Regards,
> Andre
>
> On Mon, 5 Oct 2020 15:23:27 +0200
> Nicolas König <koenigni@student.ethz.ch> wrote:
>
>> Hello Tobias,
>>
>> On 05/10/2020 11:54, Tobias Burnus wrote:
>>> Hi Nicolas,
>>>
>>> admittedly, I have not yet looked at your patch. However, I have to
>>> admit that I do not like the name. I understand that "native" refers
>>> to not needing an external library (libcaf.../libopencoarray...),
>>> but I still wonder whether something like "-fcoarray=shared" (i.e.
>>> working on a shared-memory system) would be better name from an end-user
>>> point of view.
>>
>> I think the name has been the most critized point of the entire patch up
>> till now. I'm going to change it to -fcoarray=shared, as you (and a few
>> other people) suggested :)
>>
>>>
>>> Tobias,
>>> who likes that coarray can be used without extra libs and thinks
>>> that this will help with users starting to use coarrays.
>>
>> That is the main reason I wrote the patch.
>>
>>>
>>> -----------------
>>> Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München /
>>> Germany
>>> Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung,
>>> Alexander Walter
>
>
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [RFC] Native Coarrays (finally!) [Review part 3]
2020-10-05 13:23 ` Nicolas König
2020-10-12 12:32 ` [RFC] Native Coarrays (finally!) [Review part 1] Andre Vehreschild
2020-10-12 13:48 ` [RFC] Native Coarrays (finally!) [Review part 2] Andre Vehreschild
@ 2020-10-13 13:01 ` Andre Vehreschild
2020-10-14 13:27 ` Thomas Koenig
2 siblings, 1 reply; 17+ messages in thread
From: Andre Vehreschild @ 2020-10-13 13:01 UTC (permalink / raw)
To: Nicolas König; +Cc: fortran
[-- Attachment #1: Type: text/plain, Size: 1605 bytes --]
Hi Nicolas,
here is the third part of the review mostly focusing on the runtime library. I
have deliberately skipped all generated code and the Makefile/configure stuff.
To the end of the review I got at bit bonkers again and may be will take another
look later on. Therefore the comments are thinner there.
Thanks for the work.
Regards,
Andre
On Mon, 5 Oct 2020 15:23:27 +0200
Nicolas König <koenigni@student.ethz.ch> wrote:
> Hello Tobias,
>
> On 05/10/2020 11:54, Tobias Burnus wrote:
> > Hi Nicolas,
> >
> > admittedly, I have not yet looked at your patch. However, I have to
> > admit that I do not like the name. I understand that "native" refers
> > to not needing an external library (libcaf.../libopencoarray...),
> > but I still wonder whether something like "-fcoarray=shared" (i.e.
> > working on a shared-memory system) would be better name from an end-user
> > point of view.
>
> I think the name has been the most critized point of the entire patch up
> till now. I'm going to change it to -fcoarray=shared, as you (and a few
> other people) suggested :)
>
> >
> > Tobias,
> > who likes that coarray can be used without extra libs and thinks
> > that this will help with users starting to use coarrays.
>
> That is the main reason I wrote the patch.
>
> >
> > -----------------
> > Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München /
> > Germany
> > Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung,
> > Alexander Walter
--
Andre Vehreschild * Email: vehre ad gmx dot de
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: caf_shared_part_3.patch --]
[-- Type: text/x-patch, Size: 105613 bytes --]
###AV: I have skipped all the Makefile, configure and generated stuff assuming
###AV: that when it works it's reasonable.
diff --git a/libgfortran/libgfortran.h b/libgfortran/libgfortran.h
index 8c539e0898b..a6b0d5a476d 100644
--- a/libgfortran/libgfortran.h
+++ b/libgfortran/libgfortran.h
@@ -403,6 +403,7 @@ struct {\
}
typedef GFC_FULL_ARRAY_DESCRIPTOR (GFC_MAX_DIMENSIONS, GFC_INTEGER_4) gfc_full_array_i4;
+typedef GFC_FULL_ARRAY_DESCRIPTOR (GFC_MAX_DIMENSIONS, char) gfc_full_array_char;
#define GFC_DESCRIPTOR_RANK(desc) ((desc)->dtype.rank)
#define GFC_DESCRIPTOR_TYPE(desc) ((desc)->dtype.type)
diff --git a/libgfortran/m4/nca-minmax-s.m4 b/libgfortran/m4/nca-minmax-s.m4
new file mode 100644
index 00000000000..2d8891fe673
--- /dev/null
+++ b/libgfortran/m4/nca-minmax-s.m4
@@ -0,0 +1,289 @@
+dnl Support macro file for intrinsic functions.
+dnl Contains the generic sections of gfortran functions.
+dnl This file is part of the GNU Fortran Runtime Library (libgfortran)
+dnl Distributed under the GNU GPL with exception. See COPYING for details.
+dnl
+`/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */'
+
+include(iparm.m4)dnl
+define(`compare_fcn',`ifelse(rtype_kind,1,memcmp,memcmp4)')dnl
+define(SCALAR_FUNCTION,`void nca_collsub_'$1`_scalar_'rtype_code` ('rtype_name` *obj, int *result_image,
+ int *stat, char *errmsg, index_type char_len, index_type errmsg_len);
+export_proto(nca_collsub_'$1`_scalar_'rtype_code`);
+
+void
+nca_collsub_'$1`_scalar_'rtype_code` ('rtype_name` *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ 'rtype_name` *a, *b;
+ 'rtype_name` *buffer, *this_image_buf;
+ collsub_iface *ci;
+ index_type type_size;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof ('rtype_name`);
+ buffer = get_collsub_buf (ci, type_size * local->num_images);
+ this_image_buf = buffer + this_image.image_num * char_len;
+ memcpy (this_image_buf, obj, type_size);
+
+ collsub_sync (ci);
###AV: (1)
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
###AV: At runtime use ++cbit. It's faster and prevents creation of a local
###AV: temporary.
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset * char_len;
+ if ('compare_fcn` (b, a, char_len) '$2` 0)
+ memcpy (a, b, type_size);
+ }
+ collsub_sync (ci);
###AV: (2)
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
###AV: At (1) and (2) you do a sync. At (1) just for the current image.
###AV: At (2) for the current image, but in the loop over all participating
###AV: images. And here you once again sync the current image with all
###AV: participating images. What is the use of this? And why is necessary?
###AV: What are you doing here at all? Please add a comment what the routine
###AV: is supposed to do!
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ memcpy (obj, buffer, type_size);
+
+ finish_collective_subroutine (ci);
###AV: When there is a finish_... why is there no start_...?
+
+}
+
+')dnl
+define(ARRAY_FUNCTION,dnl
+`void nca_collsub_'$1`_array_'rtype_code` ('rtype` * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type char_len,
+ index_type errmsg_len);
+export_proto (nca_collsub_'$1`_array_'rtype_code`);
+
+void
+nca_collsub_'$1`_array_'rtype_code` ('rtype` * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type char_len,
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS]; /* stride is byte-based here. */
###AV: When it's the stride of an array_descriptor it usually is not byte based,
###AV: but type_size-based.
+ index_type extent[GFC_MAX_DIMENSIONS];
+ char *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ char *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ index_type type_size;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ type_size = char_len * sizeof ('rtype_name`);
+ dim = GFC_DESCRIPTOR_RANK (array);
+ num_elems = 1;
+ ssize = type_size;
###AV: Overwritten before use. Remove.
+ packed = true;
+ span = array->span != 0 ? array->span : type_size;
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
###AV: Ahhh, you're storing the byte-strides here. That's not what the comment
###AV: above indicates. Furthermore, what about the strides type size here?
###AV: I.e., if the datatype of stride is to small to hold
###AV: GFC_DESCRIPTOR_STRIDE (array, n) * span then you're running into issues.
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (num_elems != GFC_DESCRIPTOR_STRIDE (array,n))
+ packed = false;
+
+ num_elems *= extent[n];
+ }
+
+ ssize = num_elems * type_size;
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * ssize;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ char *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+
+ memcpy (dest, src, type_size);
+ dest += type_size;
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0
+ && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ char *other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ 'rtype_name` *a;
+ 'rtype_name` *b;
+
+ other_shared_ptr = this_shared_ptr + imoffset * ssize;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = ('rtype_name` *) (this_shared_ptr + i * type_size);
+ b = ('rtype_name` *) (other_shared_ptr + i * type_size);
+ if ('compare_fcn` (b, a, char_len) '$2` 0)
+ memcpy (a, b, type_size);
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ char *src = buffer;
+ char *restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
###AV: memset (count, 0, sizeof (index_type) * dim); ???
+
+ while (dest)
+ {
+ memcpy (dest, src, type_size);
+ src += span;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+')
+`
+#include "libgfortran.h"
+
+#if defined (HAVE_'rtype_name`)
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+#if 'rtype_kind` == 4
+
+/* Compare wide character types, which are handled internally as
+ unsigned 4-byte integers. */
+static inline int
+memcmp4 (const void *a, const void *b, size_t len)
+{
+ const GFC_UINTEGER_4 *pa = a;
+ const GFC_UINTEGER_4 *pb = b;
+ while (len-- > 0)
+ {
+ if (*pa != *pb)
+ return *pa < *pb ? -1 : 1;
+ pa ++;
+ pb ++;
+ }
+ return 0;
+}
+
+#endif
+'SCALAR_FUNCTION(`max',`>')dnl
+SCALAR_FUNCTION(`min',`<')dnl
+ARRAY_FUNCTION(`max',`>')dnl
+ARRAY_FUNCTION(`min',`<')dnl
+`
+#endif
+'
diff --git a/libgfortran/m4/nca_minmax.m4 b/libgfortran/m4/nca_minmax.m4
new file mode 100644
index 00000000000..76070c102c0
--- /dev/null
+++ b/libgfortran/m4/nca_minmax.m4
@@ -0,0 +1,259 @@
+dnl Support macro file for intrinsic functions.
+dnl Contains the generic sections of gfortran functions.
+dnl This file is part of the GNU Fortran Runtime Library (libgfortran)
+dnl Distributed under the GNU GPL with exception. See COPYING for details.
+dnl
+`/* Implementation of collective subroutines minmax.
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig <tkoenig@gcc.gnu.org>.
+
+This file is part of the GNU Fortran runtime library (libgfortran).
+
+Libgfortran is free software; you can redistribute it and/or
+modify it under the terms of the GNU General Public
+License as published by the Free Software Foundation; either
+version 3 of the License, or (at your option) any later version.
+
+Libgfortran is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */'
+
+include(iparm.m4)dnl
+define(SCALAR_FUNCTION,`void nca_collsub_'$1`_scalar_'rtype_code` ('rtype_name` *obj, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto(nca_collsub_'$1`_scalar_'rtype_code`);
+
+void
+nca_collsub_'$1`_scalar_'rtype_code` ('rtype_name` *obj, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ int cbit = 0;
+ int imoffset;
+ 'rtype_name` *a, *b;
+ 'rtype_name` *buffer, *this_image_buf;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ buffer = get_collsub_buf (ci, sizeof('rtype_name`) * local->num_images);
+ this_image_buf = buffer + this_image.image_num;
+ *this_image_buf = *obj;
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ a = this_image_buf;
+ b = this_image_buf + imoffset;
+ '$2`
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ *obj = *buffer;
+
+ finish_collective_subroutine (ci);
+
+}
+
+')dnl
+define(ARRAY_FUNCTION,dnl
+`void nca_collsub_'$1`_array_'rtype_code` ('rtype` * restrict array, int *result_image,
+ int *stat, char *errmsg, index_type errmsg_len);
+export_proto (nca_collsub_'$1`_array_'rtype_code`);
+
+void
+nca_collsub_'$1`_array_'rtype_code` ('rtype` * restrict array, int *result_image,
+ int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ index_type errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ 'rtype_name` *this_shared_ptr; /* Points to the shared memory allocated to this image. */
+ 'rtype_name` *buffer;
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type ssize, num_elems;
+ int cbit = 0;
+ int imoffset;
+ collsub_iface *ci;
+
+ ci = &local->ci;
+
+ dim = GFC_DESCRIPTOR_RANK (array);
+ ssize = sizeof ('rtype_name`);
+ packed = true;
+ span = array->span != 0 ? array->span : (index_type) sizeof ('rtype_name`);
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (array, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (array, n);
+
+ /* No-op for an empty array. */
+ if (extent[n] <= 0)
+ return;
+
+ if (ssize != stride[n])
+ packed = false;
+
+ ssize *= extent[n];
+ }
+
+ num_elems = ssize / sizeof ('rtype_name`);
+
+ buffer = get_collsub_buf (ci, ssize * local->num_images);
+ this_shared_ptr = buffer + this_image.image_num * num_elems;
+
+ if (packed)
+ memcpy (this_shared_ptr, array->base_addr, ssize);
+ else
+ {
+ char *src = (char *) array->base_addr;
+ 'rtype_name` *restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+ *(dest++) = *(('rtype_name` *) src);
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+
+ collsub_sync (ci);
+
+ /* Reduce the array to image zero. Here the general scheme:
+
+ abababababab
+ a_b_a_b_a_b_
+ a___b___a___
+ a_______b___
+ r___________
+ */
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ {
+ 'rtype_name` * other_shared_ptr; /* Points to the shared memory
+ allocated to another image. */
+ 'rtype_name` *a;
+ 'rtype_name` *b;
+
+ other_shared_ptr = this_shared_ptr + num_elems * imoffset;
+ for (index_type i = 0; i < num_elems; i++)
+ {
+ a = this_shared_ptr + i;
+ b = other_shared_ptr + i;
+ '$2`
+ }
+ }
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || (*result_image - 1) == this_image.image_num)
+ {
+ if (packed)
+ memcpy (array->base_addr, buffer, ssize);
+ else
+ {
+ 'rtype_name` *src = buffer;
+ char * restrict dest = (char *) array->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ *(('rtype_name` * ) dest) = *src++;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci);
+}
+')
+`#include "libgfortran.h"
+
+#if defined (HAVE_'rtype_name`)'
+#include <string.h>
+#include "../nca/libcoarraynative.h"
+#include "../nca/collective_subroutine.h"
+#include "../nca/collective_inline.h"
+
+SCALAR_FUNCTION(`max',`if (*b > *a)
+ *a = *b;')dnl
+SCALAR_FUNCTION(`min',`if (*b < *a)
+ *a = *b;')dnl
+SCALAR_FUNCTION(`sum',`*a += *b;')dnl
+ARRAY_FUNCTION(`max',`if (*b > *a)
+ *a = *b;')dnl
+ARRAY_FUNCTION(`min',`if (*b < *a)
+ *a = *b;')dnl
+ARRAY_FUNCTION(`sum',`*a += *b;')dnl
+`
+#endif
+'
diff --git a/libgfortran/nca/.tags b/libgfortran/nca/.tags
new file mode 100644
index 00000000000..07d260ddb9d
--- /dev/null
+++ b/libgfortran/nca/.tags
@@ -0,0 +1,275 @@
###AV: ^^^ This file is from your IDE and should not be checked into git.
###AV: <snip>
diff --git a/libgfortran/nca/README.native_coarrays b/libgfortran/nca/README.native_coarrays
new file mode 100644
index 00000000000..6eea35e9044
--- /dev/null
+++ b/libgfortran/nca/README.native_coarrays
@@ -0,0 +1,35 @@
+Each image is its own process, that is forked from the master process
+at the start of the program. The number of images is determined by the
+environment variable GFORTRAN_NUM_IMAGES or, alternatively, the number
+of processors.
+
+Each coarray is identified by its address. Since coarrays always
+behave as if they had the SAVE attribute, this works even for
+allocatable coarrays. ASLR is not an issue, since the addresses are
+assigned at startup and remain valid over forks. If, on two different
+images, the allocation function is called with the same descriptor
+address, the same piece of memory is allocated.
+
+Internally, the allocator (alloc.c) uses a shared hashmap (hashmap.c)
+to remember with which ids pieces of memory allocated. If a new piece
+of memory is needed, a simple relatively allocator (allocator.c) is
+used. If the allocator doesn't hold any previously free()d memory, it
+requests it from the shared memory object (shared_memory.c), which
+also handles the translation of shared_mem_ptr's to pointers in the
+address space of the image. At the moment shared_memory relies on
+double-mapping pages for this (which might restrict the architectures
+on which this will work, I have tested this on x86 and POWER), but
+since any piece of memory should only be written to through one
+address within one alloc/free pair, it shouldn't matter that much
+performance-wise.
+
+The entry points in the library with the exception of master are
+defined in wrapper.c, master(), the function handling launching the
+images, is defined in coarraynative.c, and the other files shouldn't
+require much explanation.
+
+
+To compile a program to run with native coarrays, compile with
+-fcoarray=shared -lcaf_shared -lrt (I've not yet figured out how to
+automagically link against the library).
###AV: The latter is not possible on Unixoid systems. Only Windows has a way of
###AV: telling the linker to statically link "sub-libraries", when I remember
###AV: correctly.
+
diff --git a/libgfortran/nca/alloc.c b/libgfortran/nca/alloc.c
new file mode 100644
index 00000000000..174fe330c32
--- /dev/null
+++ b/libgfortran/nca/alloc.c
@@ -0,0 +1,152 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Thomas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+/* This provides the coarray-specific features (like IDs etc) for
+ allocator.c, in turn calling routines from shared_memory.c.
+*/
+
+#include "libgfortran.h"
+#include "shared_memory.h"
+#include "allocator.h"
+#include "hashmap.h"
+#include "alloc.h"
+
+#include <string.h>
+
+/* Return a local pointer into a shared memory object identified by
+ id. If the object is already found, it has been allocated before,
+ so just increase the reference counter.
+
+ The pointers returned by this function remain valid even if the
+ size of the memory allocation changes (see shared_memory.c). */
+
+static void *
+get_memory_by_id_internal (alloc_iface *iface, size_t size, memid id,
+ bool zero_mem)
+{
+ hashmap_search_result res;
+ shared_mem_ptr shared_ptr;
+ void *ret;
+
+ pthread_mutex_lock (&iface->as->lock);
+ shared_memory_prepare(iface->mem);
+
+ res = hashmap_get (&iface->hm, id);
+
+ if (hm_search_result_contains (&res))
+ {
+ size_t found_size;
+ found_size = hm_search_result_size (&res);
+ if (found_size != size)
+ {
+ dprintf (2, "Size mismatch for coarray allocation id %p: "
+ "found = %lu != size = %lu\n", (void *) id, found_size, size);
+ pthread_mutex_unlock (&iface->as->lock);
+ exit (1);
+ }
+ shared_ptr = hm_search_result_ptr (&res);
+ hashmap_inc (&iface->hm, id, &res);
+ }
+ else
+ {
+ shared_ptr = shared_malloc (&iface->alloc, size);
+ hashmap_set (&iface->hm, id, NULL, shared_ptr, size);
+ }
+
+ ret = SHMPTR_AS (void *, shared_ptr, iface->mem);
+ if (zero_mem)
+ memset (ret, '\0', size);
+
+ pthread_mutex_unlock (&iface->as->lock);
+ return ret;
+}
+
+void *
+get_memory_by_id (alloc_iface *iface, size_t size, memid id)
+{
+ return get_memory_by_id_internal (iface, size, id, 0);
+}
+
+void *
+get_memory_by_id_zero (alloc_iface *iface, size_t size, memid id)
+{
+ return get_memory_by_id_internal (iface, size, id, 1);
+}
+
+/* Free memory with id. Free it if this is the last image which
+ holds that memory segment, decrease the reference count otherwise. */
+
+void
+free_memory_with_id (alloc_iface* iface, memid id)
+{
+ hashmap_search_result res;
+ int entries_left;
+
+ pthread_mutex_lock (&iface->as->lock);
+ shared_memory_prepare(iface->mem);
+
+ res = hashmap_get (&iface->hm, id);
+ if (!hm_search_result_contains (&res))
+ {
+ pthread_mutex_unlock (&iface->as->lock);
+ char buffer[100];
+ snprintf (buffer, sizeof(buffer), "Error in free_memory_with_id: "
+ "%p not found", (void *) id);
###AV: Why composing the error message in a buffer first and then print?
+ dprintf (2, buffer);
+ // internal_error (NULL, buffer);
+ exit (1);
+ }
+
+ entries_left = hashmap_dec (&iface->hm, id, &res);
+ assert (entries_left >=0);
+
+ if (entries_left == 0)
+ {
+ shared_free (&iface->alloc, hm_search_result_ptr (&res),
+ hm_search_result_size (&res));
+ }
+
+ pthread_mutex_unlock (&iface->as->lock);
+ return;
+}
+
+/* Allocate the shared memory interface. This is called before we have
+ multiple images. */
+
+void
+alloc_iface_init (alloc_iface *iface, shared_memory *mem)
+{
+
+ iface->as = SHARED_MEMORY_RAW_ALLOC_PTR (mem, alloc_iface_shared);
+ iface->mem = mem;
+ initialize_shared_mutex (&iface->as->lock);
+ allocator_init (&iface->alloc, &iface->as->allocator_s, mem);
+ hashmap_init (&iface->hm, &iface->as->hms, &iface->alloc, mem);
+}
+
+allocator *
+get_allocator (alloc_iface * iface)
+{
+ return &iface->alloc;
+}
diff --git a/libgfortran/nca/alloc.h b/libgfortran/nca/alloc.h
new file mode 100644
index 00000000000..f65121c25cd
--- /dev/null
+++ b/libgfortran/nca/alloc.h
@@ -0,0 +1,67 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#ifndef ALLOC_H
+#define ALLOC_H
+
+#include "allocator.h"
+#include "hashmap.h"
+
+/* High-level interface for shared memory allocation. */
+
+/* This part of the alloc interface goes into shared memory. */
+
+typedef struct alloc_iface_shared
+{
+ allocator_shared allocator_s;
+ hashmap_shared hms;
+ pthread_mutex_t lock;
+} alloc_iface_shared;
+
+/* This is the local part. */
+
+typedef struct alloc_iface
+{
+ alloc_iface_shared *as;
+ shared_memory *mem;
+ allocator alloc;
+ hashmap hm;
+} alloc_iface;
+
+void *get_memory_by_id (alloc_iface *, size_t, memid);
+internal_proto (get_memory_by_id);
+
+void *get_memory_by_id_zero (alloc_iface *, size_t, memid);
+internal_proto (get_memory_by_id_zero);
+
+void free_memory_with_id (alloc_iface *, memid);
+internal_proto (free_memory_with_id);
+
+void alloc_iface_init (alloc_iface *, shared_memory *);
+internal_proto (alloc_iface_init);
+
+allocator *get_allocator (alloc_iface *);
+internal_proto (get_allocator);
+
+#endif
diff --git a/libgfortran/nca/allocator.c b/libgfortran/nca/allocator.c
new file mode 100644
index 00000000000..e7aa9fdefd0
--- /dev/null
+++ b/libgfortran/nca/allocator.c
@@ -0,0 +1,90 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+/* A malloc() - and free() - like interface, but for shared memory
+ pointers, except that we pass the size to free as well. */
+
+#include "libgfortran.h"
+#include "shared_memory.h"
+#include "allocator.h"
+
+typedef struct {
+ shared_mem_ptr next;
+} bucket;
+
+/* Initialize the allocator. */
+
+void
+allocator_init (allocator *a, allocator_shared *s, shared_memory *sm)
+{
+ a->s = s;
+ a->shm = sm;
+ for (int i = 0; i < PTR_BITS; i++)
+ s->free_bucket_head[i] = SHMPTR_NULL;
+}
+
+/* Main allocation routine, works like malloc. Round up allocations
+ to the next power of two and keep free lists in buckets. */
+
+#define MAX_ALIGN 16
+
+shared_mem_ptr
+shared_malloc (allocator *a, size_t size)
+{
+ shared_mem_ptr ret;
+ size_t sz;
+ size_t act_size;
+ int bucket_list_index;
+
+ sz = next_power_of_two (size);
+ act_size = sz > sizeof (bucket) ? sz : sizeof (bucket);
+ bucket_list_index = __builtin_clzl(act_size);
+
+ if (SHMPTR_IS_NULL (a->s->free_bucket_head[bucket_list_index]))
+ return shared_memory_get_mem_with_alignment (a->shm, act_size, MAX_ALIGN);
+
+ ret = a->s->free_bucket_head[bucket_list_index];
+ a->s->free_bucket_head[bucket_list_index]
+ = (SHMPTR_AS (bucket *, ret, a->shm)->next);
+ assert(ret.offset != 0);
+ return ret;
+}
+
+/* Free memory. */
+
+void
+shared_free (allocator *a, shared_mem_ptr p, size_t size) {
+ bucket *b;
+ size_t sz;
+ int bucket_list_index;
+ size_t act_size;
+
+ sz = next_power_of_two (size);
+ act_size = sz > sizeof (bucket) ? sz : sizeof (bucket);
+ bucket_list_index = __builtin_clzl(act_size);
+
+ b = SHMPTR_AS(bucket *, p, a->shm);
+ b->next = a->s->free_bucket_head[bucket_list_index];
+ a->s->free_bucket_head[bucket_list_index] = p;
+}
diff --git a/libgfortran/nca/allocator.h b/libgfortran/nca/allocator.h
new file mode 100644
index 00000000000..306022a5f39
--- /dev/null
+++ b/libgfortran/nca/allocator.h
@@ -0,0 +1,21 @@
+#ifndef SHARED_ALLOCATOR_HDR
+#define SHARED_ALLOCATOR_HDR
+
+#include "util.h"
+#include "shared_memory.h"
+
+typedef struct {
+ shared_mem_ptr free_bucket_head[PTR_BITS];
+} allocator_shared;
+
+typedef struct {
+ allocator_shared *s;
+ shared_memory *shm;
+} allocator;
###AV: I am unhappy with the term allocator, because in C++-STL it is a functor
###AV: i.e. something that can act, while here it is just a data structure
###AV: that is dumb. A data structure IMO should name a thing like 'car' or
###AV: 'human'. What you are storing here rather is a 'memory_pool' or
###AV: 'mem_chunks_list'. Maybe some better name would prevent any mislead.
+
+void allocator_init (allocator *, allocator_shared *, shared_memory *);
+
+shared_mem_ptr shared_malloc (allocator *, size_t size);
+void shared_free (allocator *, shared_mem_ptr, size_t size);
+
+#endif
diff --git a/libgfortran/nca/coarraynative.c b/libgfortran/nca/coarraynative.c
new file mode 100644
index 00000000000..c9d13ee92ac
--- /dev/null
+++ b/libgfortran/nca/coarraynative.c
@@ -0,0 +1,145 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include "libgfortran.h"
+#include "libcoarraynative.h"
+#include "allocator.h"
+#include "hashmap.h"
+#include "util.h"
+#include "lock.h"
+#include "collective_subroutine.h"
+
+#include <unistd.h>
+#include <sys/mman.h>
+// #include <stdlib.h>
+#include <sys/wait.h>
+
+#define GFORTRAN_ENV_NUM_IMAGES "GFORTRAN_NUM_IMAGES"
+
+nca_local_data *local = NULL;
+
+image this_image;
+
+static int
+get_environ_image_num (void)
+{
+ char *num_images_char;
+ int nimages;
+ num_images_char = getenv (GFORTRAN_ENV_NUM_IMAGES);
+ if (!num_images_char)
+ return sysconf (_SC_NPROCESSORS_ONLN); /* TODO: Make portable. */
+ /* TODO: Error checking. */
+ nimages = atoi (num_images_char);
+ return nimages;
+}
+
+void
+ensure_initialization(void)
+{
+ if (local)
+ return;
+
+ local = malloc(sizeof(nca_local_data)); // Is malloc already init'ed at that
+ // point? Maybe use mmap(MAP_ANON)
+ // instead
###AV: Well, the libc's init should have run by now. Anyway, I always had the
###AV: assumption in mind, that malloc is bound by the loader.
+ pagesize = sysconf (_SC_PAGE_SIZE);
###AV: Is this avail on all systems?
+ local->num_images = get_environ_image_num ();
+ shared_memory_init (&local->sm);
+ shared_memory_prepare (&local->sm);
+ alloc_iface_init (&local->ai, &local->sm);
+ collsub_iface_init (&local->ci, &local->ai, &local->sm);
+ sync_iface_init (&local->si, &local->ai, &local->sm);
+}
+
+static void __attribute__((noreturn))
+image_main_wrapper (void (*image_main) (void), image *this)
+{
+ this_image = *this;
+
+ sync_all(&local->si);
+
+ image_main ();
+
+ exit (0);
+}
+
+static master *
+get_master (void) {
+ master *m;
+ m = SHMPTR_AS (master *,
+ shared_memory_get_mem_with_alignment
+ (&local->sm,
+ sizeof (master) + sizeof(image_status) * local->num_images,
+ __alignof__(master)), &local->sm);
+ m->has_failed_image = 0;
+ return m;
+}
+
+/* This is called from main, with a pointer to the user's program as
+ argument. It forks the images and waits for their completion. */
+
+void
+nca_master (void (*image_main) (void)) {
+ master *m;
+ int i, j;
+ pid_t new;
+ image im;
+ int exit_code = 0;
+ int chstatus;
+ ensure_initialization();
+ m = get_master();
+
+ im.m = m;
+
+ for (im.image_num = 0; im.image_num < local->num_images; im.image_num++)
+ {
+ if ((new = fork()))
+ {
+ if (new == -1)
+ {
+ dprintf(2, "error spawning child\n");
+ exit_code = 1;
+ }
+ m->images[im.image_num].pid = new;
+ m->images[im.image_num].status = IMAGE_OK;
+ }
+ else
+ image_main_wrapper(image_main, &im);
+ }
+ for (i = 0; i < local->num_images; i++)
+ {
+ new = wait (&chstatus);
+ if (!WIFEXITED (chstatus) || WEXITSTATUS (chstatus))
+ {
+ j = 0;
+ for (; j < local->num_images && m->images[j].pid != new; j++);
+ m->images[j].status = IMAGE_FAILED;
+ m->has_failed_image++; //FIXME: Needs to be atomic, probably
###AV: This should either be a bool, where 'has_failed_images' is correct, but
###AV: then set here to true, or be named 'num_failed_images' which is how it
###AV: is used here currently.
+ dprintf (2, "ERROR: Image %d(%#x) failed\n", j, new);
+ exit_code = 1;
+ }
###AV: If an image terminated successfully, should that not be reflected in the
###AV: image's status? Like IMAGE_ENDED or IMAGE_STOPPED, which when I am not
###AV: mistaken can be checked for to num_images_stopped()?!
+ }
+ exit (exit_code);
+}
diff --git a/libgfortran/nca/collective_inline.h b/libgfortran/nca/collective_inline.h
new file mode 100644
index 00000000000..4e7107b359d
--- /dev/null
+++ b/libgfortran/nca/collective_inline.h
@@ -0,0 +1,42 @@
+#include "collective_subroutine.h"
+
+static inline void
+finish_collective_subroutine (collsub_iface *ci)
+{
+ collsub_sync (ci);
+}
+
+#if 0
+static inline void *
+get_obj_ptr (void *buffer, int image)
+{
+ return (char *) + curr_size * image;
+}
+
+/* If obj is NULL, copy the object from the entry in this image. */
+static inline void
+copy_to (void *buffer, void *obj, int image)
+{
+ if (obj == 0)
+ obj = get_obj_ptr (this_image.image_num);
+ memcpy (get_obj_ptr (image), obj, curr_size);
+}
+
+static inline void
+copy_out (void *buffer, void *obj, int image)
+{
+ memcpy (obj, get_obj_ptr (image), curr_size);
+}
+
+static inline void
+copy_from (void *buffer, int image)
+{
+ copy_out (get_obj_ptr (this_image.image_num), image);
+}
+
+static inline void
+copy_in (void *buffer, void *obj)
+{
+ copy_to (obj, this_image.image_num);
+}
+#endif
###AV: Remove or use.
diff --git a/libgfortran/nca/collective_subroutine.c b/libgfortran/nca/collective_subroutine.c
new file mode 100644
index 00000000000..8a8a7d659f0
--- /dev/null
+++ b/libgfortran/nca/collective_subroutine.c
@@ -0,0 +1,416 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#include <string.h>
+#include "libgfortran.h"
+#include "libcoarraynative.h"
+#include "collective_subroutine.h"
+#include "collective_inline.h"
+#include "allocator.h"
+
+void *
+get_collsub_buf (collsub_iface *ci, size_t size)
+{
+ void *ret;
+
+ pthread_mutex_lock (&ci->s->mutex);
+ if (size > ci->s->curr_size)
###AV: Is a size of zero possible here? Is a guard needed?
+ {
+ shared_free (ci->a, ci->s->collsub_buf, ci->s->curr_size);
+ ci->s->collsub_buf = shared_malloc (ci->a, size);
+ ci->s->curr_size = size;
+ }
+
+ ret = SHMPTR_AS (void *, ci->s->collsub_buf, ci->sm);
+ pthread_mutex_unlock (&ci->s->mutex);
+ return ret;
+}
+
+/* It appears as if glibc's barrier implementation does not spin (at
+ least that is what I got from a quick glance at the source code),
+ so performance would be improved quite a bit if we spun a few times
+ here so we don't run into the futex syscall. */
+
+void
+collsub_sync (collsub_iface *ci)
+{
+ //dprintf (2, "Calling collsub_sync %d times\n", ++called);
###AV: Remove debug.
+ pthread_barrier_wait (&ci->s->barrier);
+}
+
+/* assign_function is needed since we only know how to assign the type inside
+ the compiler. It should be implemented as follows:
+
+ void assign_function (void *a, void *b)
+ {
+ *((t *) a) = reduction_operation ((t *) a, (t *) b);
+ }
+
+ */
+
+void
+collsub_reduce_array (collsub_iface *ci, gfc_array_char *desc, int *result_image,
+ void (*assign_function) (void *, void *))
+{
+ void *buffer;
+ pack_info pi;
+ bool packed;
+ int cbit = 0;
+ int imoffset;
+ index_type elem_size;
+ index_type this_image_size_bytes;
+ char *this_image_buf;
+
+ packed = pack_array_prepare (&pi, desc);
+ if (pi.num_elem == 0)
+ return;
+
+ elem_size = GFC_DESCRIPTOR_SIZE (desc);
+ this_image_size_bytes = elem_size * pi.num_elem;
+
+ buffer = get_collsub_buf (ci, this_image_size_bytes * local->num_images);
+ this_image_buf = buffer + this_image_size_bytes * this_image.image_num;
+
+ if (packed)
+ memcpy (this_image_buf, GFC_DESCRIPTOR_DATA (desc), this_image_size_bytes);
+ else
+ pack_array_finish (&pi, desc, this_image_buf);
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ /* Reduce arrays elementwise. */
+ for (size_t i = 0; i < pi.num_elem; i++)
+ assign_function (this_image_buf + elem_size * i,
+ this_image_buf + this_image_size_bytes * imoffset + elem_size * i);
+
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || *result_image == this_image.image_num)
+ {
+ if (packed)
+ memcpy (GFC_DESCRIPTOR_DATA (desc), buffer, this_image_size_bytes);
+ else
+ unpack_array_finish(&pi, desc, buffer);
+ }
+
+ finish_collective_subroutine (ci);
+}
+
+void
+collsub_reduce_scalar (collsub_iface *ci, void *obj, index_type elem_size,
+ int *result_image,
+ void (*assign_function) (void *, void *))
+{
+ void *buffer;
+ int cbit = 0;
+ int imoffset;
+ char *this_image_buf;
+
+ buffer = get_collsub_buf (ci, elem_size * local->num_images);
+ this_image_buf = buffer + elem_size * this_image.image_num;
+
+ memcpy (this_image_buf, obj, elem_size);
+
+ collsub_sync (ci);
+ for (; ((this_image.image_num >> cbit) & 1) == 0 && (local->num_images >> cbit) != 0; cbit++)
+ {
+ imoffset = 1 << cbit;
+ if (this_image.image_num + imoffset < local->num_images)
+ /* Reduce arrays elementwise. */
+ assign_function (this_image_buf, this_image_buf + elem_size*imoffset);
+
+ collsub_sync (ci);
+ }
+ for ( ; (local->num_images >> cbit) != 0; cbit++)
+ collsub_sync (ci);
+
+ if (!result_image || *result_image == this_image.image_num)
+ memcpy(obj, buffer, elem_size);
+
+ finish_collective_subroutine (ci);
+}
+
+/* Do not use sync_all(), because the program should deadlock in the case that
+ * some images are on a sync_all barrier while others are in a collective
+ * subroutine. */
+
+void
+collsub_iface_init (collsub_iface *ci, alloc_iface *ai, shared_memory *sm)
+{
+ pthread_barrierattr_t attr;
+ shared_mem_ptr p;
+ ci->s = SHARED_MEMORY_RAW_ALLOC_PTR(sm, collsub_iface_shared);
+
+ ci->s->collsub_buf = shared_malloc(get_allocator(ai), sizeof(double)*local->num_images);
+ ci->s->curr_size = sizeof(double)*local->num_images;
+ ci->sm = sm;
+ ci->a = get_allocator(ai);
+
+ pthread_barrierattr_init (&attr);
+ pthread_barrierattr_setpshared (&attr, PTHREAD_PROCESS_SHARED);
+ pthread_barrier_init (&ci->s->barrier, &attr, local->num_images);
+ pthread_barrierattr_destroy(&attr);
+
+ initialize_shared_mutex (&ci->s->mutex);
+}
+
+void
+collsub_broadcast_scalar (collsub_iface *ci, void *obj, index_type elem_size,
+ int source_image /* Adjusted in the wrapper. */)
+{
+ void *buffer;
+
+ buffer = get_collsub_buf (ci, elem_size);
+
+ dprintf(2, "Source image: %d\n", source_image);
+
+ if (source_image == this_image.image_num)
+ {
+ memcpy (buffer, obj, elem_size);
+ collsub_sync (ci);
+ }
+ else
+ {
+ collsub_sync (ci);
+ memcpy (obj, buffer, elem_size);
+ }
+
+ finish_collective_subroutine (ci);
+}
+
+void
+collsub_broadcast_array (collsub_iface *ci, gfc_array_char *desc,
+ int source_image)
+{
+ void *buffer;
+ pack_info pi;
+ bool packed;
+ index_type elem_size;
+ index_type size_bytes;
+ char *this_image_buf;
+
+ packed = pack_array_prepare (&pi, desc);
+ if (pi.num_elem == 0)
+ return;
+
+ elem_size = GFC_DESCRIPTOR_SIZE (desc);
+ size_bytes = elem_size * pi.num_elem;
+
+ buffer = get_collsub_buf (ci, size_bytes);
+
+ if (source_image == this_image.image_num)
+ {
+ if (packed)
+ memcpy (buffer, GFC_DESCRIPTOR_DATA (desc), size_bytes);
+ else
+ pack_array_finish (&pi, desc, buffer);
+ collsub_sync (ci);
+ }
+ else
+ {
+ collsub_sync (ci);
+ if (packed)
+ memcpy (GFC_DESCRIPTOR_DATA (desc), buffer, size_bytes);
+ else
+ unpack_array_finish(&pi, desc, buffer);
+ }
+
+ finish_collective_subroutine (ci);
+}
+
+#if 0
+
+void nca_co_broadcast (gfc_array_char *, int, int*, char *, size_t);
+export_proto (nca_co_broadcast);
+
+void
+nca_co_broadcast (gfc_array_char * restrict a, int source_image,
+ int *stat, char *errmsg __attribute__ ((unused)),
+ size_t errmsg_len __attribute__ ((unused)))
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS];
+ index_type extent[GFC_MAX_DIMENSIONS];
+ index_type type_size;
+ index_type dim;
+ index_type span;
+ bool packed, empty;
+ index_type num_elems;
+ index_type ssize, ssize_bytes;
+ char *this_shared_ptr, *other_shared_ptr;
+
+ if (stat)
+ *stat = 0;
+
+ dim = GFC_DESCRIPTOR_RANK (a);
+ type_size = GFC_DESCRIPTOR_SIZE (a);
+
+ /* Source image, gather. */
+ if (source_image - 1 == image_num)
+ {
+ num_elems = 1;
+ if (dim > 0)
+ {
+ span = a->span != 0 ? a->span : type_size;
+ packed = true;
+ empty = false;
+ for (index_type n = 0; n < dim; n++)
+ {
+ count[n] = 0;
+ stride[n] = GFC_DESCRIPTOR_STRIDE (a, n) * span;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (a, n);
+
+ empty = empty || extent[n] <= 0;
+
+ if (num_elems != GFC_DESCRIPTOR_STRIDE (a, n))
+ packed = false;
+
+ num_elems *= extent[n];
+ }
+ ssize_bytes = num_elems * type_size;
+ }
+ else
+ {
+ ssize_bytes = type_size;
+ packed = true;
+ empty = false;
+ }
+
+ prepare_collective_subroutine (ssize_bytes); // broadcast barrier 1
+ this_shared_ptr = get_obj_ptr (image_num);
+ if (packed)
+ memcpy (this_shared_ptr, a->base_addr, ssize_bytes);
+ else
+ {
+ char *src = (char *) a->base_addr;
+ char * restrict dest = this_shared_ptr;
+ index_type stride0 = stride[0];
+
+ while (src)
+ {
+ /* Copy the data. */
+
+ memcpy (dest, src, type_size);
+ dest += type_size;
+ src += stride0;
+ count[0] ++;
+ /* Advance to the next source element. */
+ for (index_type n = 0; count[n] == extent[n] ; )
+ {
+ /* When we get to the end of a dimension, reset it
+ and increment the next dimension. */
+ count[n] = 0;
+ src -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += stride[n];
+ }
+ }
+ }
+ }
+ collsub_sync (ci); /* Broadcast barrier 2. */
+ }
+ else /* Target image, scatter. */
+ {
+ collsub_sync (ci); /* Broadcast barrier 1. */
+ packed = 1;
+ num_elems = 1;
+ span = a->span != 0 ? a->span : type_size;
+
+ for (index_type n = 0; n < dim; n++)
+ {
+ index_type stride_n;
+ count[n] = 0;
+ stride_n = GFC_DESCRIPTOR_STRIDE (a, n);
+ stride[n] = stride_n * type_size;
+ extent[n] = GFC_DESCRIPTOR_EXTENT (a, n);
+ if (extent[n] <= 0)
+ {
+ packed = true;
+ num_elems = 0;
+ break;
+ }
+ if (num_elems != stride_n)
+ packed = false;
+
+ num_elems *= extent[n];
+ }
+ ssize = num_elems * type_size;
+ prepare_collective_subroutine (ssize); /* Broadcaset barrier 2. */
+ other_shared_ptr = get_obj_ptr (source_image - 1);
+ if (packed)
+ memcpy (a->base_addr, other_shared_ptr, ssize);
+ else
+ {
+ char *src = other_shared_ptr;
+ char * restrict dest = (char *) a->base_addr;
+ index_type stride0 = stride[0];
+
+ for (index_type n = 0; n < dim; n++)
+ count[n] = 0;
+
+ while (dest)
+ {
+ memcpy (dest, src, type_size);
+ src += span;
+ dest += stride0;
+ count[0] ++;
+ for (index_type n = 0; count[n] == extent[n] ;)
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ dest -= stride[n] * extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ dest += stride[n];
+ }
+ }
+ }
+ }
+ }
+ finish_collective_subroutine (ci); /* Broadcast barrier 3. */
+}
+
+#endif
diff --git a/libgfortran/nca/collective_subroutine.h b/libgfortran/nca/collective_subroutine.h
new file mode 100644
index 00000000000..6147dd6d793
--- /dev/null
+++ b/libgfortran/nca/collective_subroutine.h
@@ -0,0 +1,44 @@
+
+#ifndef COLLECTIVE_SUBROUTINE_HDR
+#define COLLECTIVE_SUBROUTINE_HDR
+
+#include "shared_memory.h"
+
+typedef struct collsub_iface_shared
+{
+ size_t curr_size;
+ shared_mem_ptr collsub_buf;
+ pthread_barrier_t barrier;
+ pthread_mutex_t mutex;
+} collsub_iface_shared;
+
+typedef struct collsub_iface
+{
+ collsub_iface_shared *s;
+ allocator *a;
+ shared_memory *sm;
+} collsub_iface;
+
+void collsub_broadcast_scalar (collsub_iface *, void *, index_type, int);
+internal_proto (collsub_broadcast_scalar);
+
+void collsub_broadcast_array (collsub_iface *, gfc_array_char *, int);
+internal_proto (collsub_broadcast_array);
+
+void collsub_reduce_array (collsub_iface *, gfc_array_char *, int *,
+ void (*) (void *, void *));
+internal_proto (collsub_reduce_array);
+
+void collsub_reduce_scalar (collsub_iface *, void *, index_type, int *,
+ void (*) (void *, void *));
+internal_proto (collsub_reduce_scalar);
+
+void collsub_sync (collsub_iface *);
+internal_proto (collsub_sync);
+
+void collsub_iface_init (collsub_iface *, alloc_iface *, shared_memory *);
+internal_proto (collsub_iface_init);
+
+void * get_collsub_buf (collsub_iface *ci, size_t size);
+internal_proto (get_collsub_buf);
+#endif
diff --git a/libgfortran/nca/hashmap.c b/libgfortran/nca/hashmap.c
new file mode 100644
index 00000000000..61f5487e63e
--- /dev/null
+++ b/libgfortran/nca/hashmap.c
@@ -0,0 +1,447 @@
###AV: Where is this hashmap stuff coming from?
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#include "libgfortran.h"
+#include "hashmap.h"
+#include <string.h>
+
+#define INITIAL_BITNUM (5)
+#define INITIAL_SIZE (1<<INITIAL_BITNUM)
+#define CRITICAL_LOOKAHEAD (16)
+
+static ssize_t n_ent;
+
+typedef struct {
+ memid id;
+ shared_mem_ptr p; /* If p == SHMPTR_NULL, the entry is empty. */
+ size_t s;
+ int max_lookahead;
+ int refcnt;
+} hashmap_entry;
+
+static ssize_t
+num_entries (hashmap_entry *data, size_t size)
+{
+ ssize_t i;
+ ssize_t ret = 0;
+ for (i = 0; i < size; i++)
+ {
+ if (!SHMPTR_IS_NULL (data[i].p))
+ ret ++;
+ }
+ return ret;
+}
+
+/* 64 bit to 64 bit hash function. */
+
+/*
+static inline uint64_t
+hash (uint64_t x)
+{
+ return x * 11400714819323198485lu;
+}
+*/
+
+#define ASSERT_HM(hm, cond) assert_hashmap(hm, cond, #cond)
+
+static void
+assert_hashmap(hashmap *hm, bool asserted, const char *cond)
+{
+ if (!asserted)
+ {
+ dprintf(2, cond);
+ dump_hm(hm);
+ }
+ assert(asserted);
+}
+
+static inline uint64_t
+hash (uint64_t key)
+{
+ key ^= (key >> 30);
+ key *= 0xbf58476d1ce4e5b9ul;
+ key ^= (key >> 27);
+ key *= 0x94d049bb133111ebul;
+ key ^= (key >> 31);
###AV: This algo is used in an example of "The Standard C Library" by Plauger
###AV: as a rand()-algorithm...
+
+ return key;
+}
+
+/* Gets a pointer to the current data in the hashmap. */
+static inline hashmap_entry *
+get_data(hashmap *hm)
+{
+ return SHMPTR_AS (hashmap_entry *, hm->s->data, hm->sm);
+}
+
+/* Generate mask from current number of bits. */
+
+static inline intptr_t
+gen_mask (hashmap *hm)
+{
+ return (1 << hm->s->bitnum) - 1;
+}
+
+/* Add with wrap-around at hashmap size. */
+
+static inline size_t
+hmiadd (hashmap *hm, size_t s, ssize_t o) {
+ return (s + o) & gen_mask (hm);
+}
+
+/* Get the expected offset for entry id. */
+
+static inline ssize_t
+get_expected_offset (hashmap *hm, memid id)
+{
+ return hash(id) >> (PTR_BITS - hm->s->bitnum);
+}
+
+/* Initialize the hashmap. */
+
+void
+hashmap_init (hashmap *hm, hashmap_shared *hs, allocator *a,
+ shared_memory *mem)
+{
+ hashmap_entry *data;
+ hm->s = hs;
+ hm->sm = mem;
+ hm->s->data = shared_malloc (a, INITIAL_SIZE * sizeof(hashmap_entry));
+ data = get_data (hm);
+ memset(data, '\0', INITIAL_SIZE*sizeof(hashmap_entry));
+
+ for (int i = 0; i < INITIAL_SIZE; i++)
+ data[i].p = SHMPTR_NULL;
+
+ hm->s->size = INITIAL_SIZE;
+ hm->s->bitnum = INITIAL_BITNUM;
+ hm->a = a;
+}
+
+/* This checks if the entry id exists in that range the range between
+ the expected position and the maximum lookahead. */
+
+static ssize_t
+scan_inside_lookahead (hashmap *hm, ssize_t expected_off, memid id)
+{
+ ssize_t lookahead;
+ hashmap_entry *data;
+
+ data = get_data (hm);
+ lookahead = data[expected_off].max_lookahead;
+ ASSERT_HM (hm, lookahead < CRITICAL_LOOKAHEAD);
+
+ for (int i = 0; i <= lookahead; i++) /* For performance, this could
+ iterate backwards. */
+ if (data[hmiadd (hm, expected_off, i)].id == id)
+ return hmiadd (hm, expected_off, i);
+
+ return -1;
+}
+
+/* Scan for the next empty slot we can use. Returns offset relative
+ to the expected position. */
+
+static ssize_t
+scan_empty (hashmap *hm, ssize_t expected_off, memid id)
+{
+ hashmap_entry *data;
+
+ data = get_data(hm);
+ for (int i = 0; i < CRITICAL_LOOKAHEAD; i++)
+ if (SHMPTR_IS_NULL (data[hmiadd (hm, expected_off, i)].p))
+ return i;
+
+ return -1;
+}
+
+/* Search the hashmap for id. */
+
+hashmap_search_result
+hashmap_get (hashmap *hm, memid id)
+{
+ hashmap_search_result ret;
+ hashmap_entry *data;
+ size_t expected_offset;
+ ssize_t res;
+
+ data = get_data (hm);
+ expected_offset = get_expected_offset (hm, id);
+ res = scan_inside_lookahead (hm, expected_offset, id);
+
+ if (res != -1)
+ ret = ((hashmap_search_result)
+ { .p = data[res].p, .size=data[res].s, .res_offset = res });
+ else
+ ret.p = SHMPTR_NULL;
+
+ return ret;
+}
+
+/* Return size of a hashmap search result. */
+size_t
+hm_search_result_size (hashmap_search_result *res)
+{
+ return res->size;
+}
+
+/* Return pointer of a hashmap search result. */
+
+shared_mem_ptr
+hm_search_result_ptr (hashmap_search_result *res)
+{
+ return res->p;
+}
+
+/* Return pointer of a hashmap search result. */
+
+bool
+hm_search_result_contains (hashmap_search_result *res)
+{
+ return !SHMPTR_IS_NULL(res->p);
+}
+
+/* Enlarge hashmap memory. */
+
+static void
+enlarge_hashmap_mem (hashmap *hm, hashmap_entry **data, bool f)
+{
+ shared_mem_ptr old_data_p;
+ size_t old_size;
+
+ old_data_p = hm->s->data;
+ old_size = hm->s->size;
+
+ hm->s->data = shared_malloc (hm->a, (hm->s->size *= 2)*sizeof(hashmap_entry));
+ fprintf (stderr,"enlarge_hashmap_mem: %ld\n", hm->s->data.offset);
###AV: Remove debug output!
+ hm->s->bitnum++;
+
+ *data = get_data(hm);
+ for (size_t i = 0; i < hm->s->size; i++)
+ (*data)[i] = ((hashmap_entry) { .id = 0, .p = SHMPTR_NULL, .s=0,
+ .max_lookahead = 0, .refcnt=0 });
+
+ if (f)
+ shared_free(hm->a, old_data_p, old_size);
+}
+
+/* Resize hashmap. */
+
+static void
+resize_hm (hashmap *hm, hashmap_entry **data)
+{
+ shared_mem_ptr old_data_p;
+ hashmap_entry *old_data, *new_data;
+ size_t old_size;
+ ssize_t new_offset, inital_index, new_index;
+ memid id;
+ ssize_t max_lookahead;
+ ssize_t old_count, new_count;
+
+ /* old_data points to the old block containing the hashmap. We
+ redistribute the data from there into the new block. */
+
+ old_data_p = hm->s->data;
+ old_data = *data;
+ old_size = hm->s->size;
+ old_count = num_entries (old_data, old_size);
+
+ fprintf(stderr, "Occupancy at resize: %f\n", ((double) old_count)/old_size);
+
+ //fprintf (stderr,"\n====== Resizing hashmap =========\n\nOld map:\n\n");
+ //dump_hm (hm);
+ enlarge_hashmap_mem (hm, &new_data, false);
+ //fprintf (stderr,"old_data: %p new_data: %p\n", old_data, new_data);
+ retry_resize:
+ for (size_t i = 0; i < old_size; i++)
+ {
+ if (SHMPTR_IS_NULL (old_data[i].p))
+ continue;
+
+ id = old_data[i].id;
+ inital_index = get_expected_offset (hm, id);
+ new_offset = scan_empty (hm, inital_index, id);
+
+ /* If we didn't find a free slot, just resize the hashmap
+ again. */
+ if (new_offset == -1)
+ {
+ enlarge_hashmap_mem (hm, &new_data, true);
+ //fprintf (stderr,"\n====== AGAIN Resizing hashmap =========\n\n");
+ //fprintf (stderr,"old_data: %p new_data %p\n", old_data, new_data);
+ goto retry_resize; /* Sue me. */
+ }
+
+ ASSERT_HM (hm, new_offset < CRITICAL_LOOKAHEAD);
+ new_index = hmiadd (hm, inital_index, new_offset);
+ max_lookahead = new_data[inital_index].max_lookahead;
+ new_data[inital_index].max_lookahead
+ = new_offset > max_lookahead ? new_offset : max_lookahead;
+
+ new_data[new_index] = ((hashmap_entry) {.id = id, .p = old_data[i].p,
+ .s = old_data[i].s,
+ .max_lookahead = new_data[new_index].max_lookahead,
+ .refcnt = old_data[i].refcnt});
+ }
+ new_count = num_entries (new_data, hm->s->size);
+ //fprintf (stderr,"Number of elements: %ld to %ld\n", old_count, new_count);
+ //fprintf (stderr,"============ After resizing: =======\n\n");
+ //dump_hm (hm);
+
+ shared_free (hm->a, old_data_p, old_size);
+ *data = new_data;
+}
+
+/* Set an entry in the hashmap. */
+
+void
+hashmap_set (hashmap *hm, memid id, hashmap_search_result *hsr,
+ shared_mem_ptr p, size_t size)
+{
+ hashmap_entry *data;
+ ssize_t expected_offset, lookahead;
+ ssize_t empty_offset;
+ ssize_t delta;
+
+ // //fprintf (stderr,"hashmap_set: id = %-16p\n", (void *) id);
+ data = get_data(hm);
+
+ if (hsr) {
+ data[hsr->res_offset].s = size;
+ data[hsr->res_offset].p = p;
+ return;
+ }
+
+ expected_offset = get_expected_offset (hm, id);
+ while ((delta = scan_empty (hm, expected_offset, id)) == -1)
+ {
+ resize_hm (hm, &data);
+ expected_offset = get_expected_offset (hm, id);
+ }
+
+ empty_offset = hmiadd (hm, expected_offset, delta);
+ lookahead = data[expected_offset].max_lookahead;
+ data[expected_offset].max_lookahead = delta > lookahead ? delta : lookahead;
+ data[empty_offset] = ((hashmap_entry) {.id = id, .p = p, .s = size,
+ .max_lookahead = data[empty_offset].max_lookahead,
+ .refcnt = 1});
+
+ n_ent ++;
+ fprintf (stderr,"hashmap_set: Setting %p at %p, n_ent = %ld\n", (void *) id, data + empty_offset,
+ n_ent);
+ // dump_hm (hm);
+ // fprintf(stderr, "--------------------------------------------------\n");
+ /* TODO: Shouldn't reset refcnt, but this doesn't matter at the
+ moment because of the way the function is used. */
+}
+
+/* Change the refcount of a hashmap entry. */
+
+static int
+hashmap_change_refcnt (hashmap *hm, memid id, hashmap_search_result *res,
+ int delta)
+{
+ hashmap_entry *data;
+ hashmap_search_result r;
+ hashmap_search_result *pr;
+ int ret;
+ hashmap_entry *entry;
+
+ data = get_data (hm);
+
+ if (res)
+ pr = res;
+ else
+ {
+ r = hashmap_get (hm, id);
+ pr = &r;
+ }
+
+ entry = &data[pr->res_offset];
+ ret = (entry->refcnt += delta);
+ if (ret == 0)
+ {
+ n_ent --;
+ //fprintf (stderr, "hashmap_change_refcnt: removing %p at %p, n_ent = %ld\n",
+ // (void *) id, entry, n_ent);
+ entry->id = 0;
+ entry->p = SHMPTR_NULL;
+ entry->s = 0;
+ }
+
+ return ret;
+}
+
+/* Increase hashmap entry refcount. */
+
+void
+hashmap_inc (hashmap *hm, memid id, hashmap_search_result * res)
+{
+ int ret;
+ ret = hashmap_change_refcnt (hm, id, res, 1);
+ ASSERT_HM (hm, ret > 0);
+}
+
+/* Decrease hashmap entry refcount. */
+
+int
+hashmap_dec (hashmap *hm, memid id, hashmap_search_result * res)
+{
+ int ret;
+ ret = hashmap_change_refcnt (hm, id, res, -1);
+ ASSERT_HM (hm, ret >= 0);
+ return ret;
+}
+
+#define PE(str, ...) fprintf(stderr, INDENT str, ##__VA_ARGS__)
+#define INDENT ""
+
+void
+dump_hm(hashmap *hm) {
+ hashmap_entry *data;
+ size_t exp;
+ size_t occ_num = 0;
+ PE("h %p (size: %lu, bitnum: %d)\n", hm, hm->s->size, hm->s->bitnum);
+ data = get_data (hm);
+ fprintf (stderr,"offset = %lx data = %p\n", (unsigned long) hm->s->data.offset, data);
+
+#undef INDENT
+#define INDENT " "
+ for (size_t i = 0; i < hm->s->size; i++) {
+ exp = get_expected_offset(hm, data[i].id);
+ if (!SHMPTR_IS_NULL(data[i].p)) {
+ PE("%2lu. (exp: %2lu w la %d) id %#-16lx p %#-14lx s %-7lu -- la %u ref %u %-16p\n",
+ i, exp, data[exp].max_lookahead, data[i].id, data[i].p.offset, data[i].s,
+ data[i].max_lookahead, data[i].refcnt, data + i);
+ occ_num++;
+ }
+ else
+ PE("%2lu. empty -- la %u %p\n", i, data[i].max_lookahead,
+ data + i);
+
+ }
+#undef INDENT
+#define INDENT ""
+ PE("occupancy: %lu %f\n", occ_num, ((double) occ_num)/hm->s->size);
+}
diff --git a/libgfortran/nca/hashmap.h b/libgfortran/nca/hashmap.h
new file mode 100644
index 00000000000..4d999e3e3d3
--- /dev/null
+++ b/libgfortran/nca/hashmap.h
@@ -0,0 +1,70 @@
###AV: Copyright notice missing!
+#ifndef HASHMAP_H
+
+#include "shared_memory.h"
+#include "allocator.h"
+
+#include <stdint.h>
+#include <stddef.h>
+
+
+/* Data structures and variables:
+
+ memid is a unique identifier for the coarray, the address of its
+ descriptor (which is unique in the program). */
+typedef intptr_t memid;
+
+typedef struct {
+ shared_mem_ptr data;
+ size_t size;
+ int bitnum;
+} hashmap_shared;
+
+typedef struct hashmap
+{
+ hashmap_shared *s;
+ shared_memory *sm;
+ allocator *a;
+} hashmap;
+
+typedef struct {
+ shared_mem_ptr p;
+ size_t size;
+ ssize_t res_offset;
+} hashmap_search_result;
+
+void hashmap_init (hashmap *, hashmap_shared *, allocator *a, shared_memory *);
+
+/* Look up memid in the hashmap. The result can be inspected via the
+ hm_search_result_* functions. */
+
+hashmap_search_result hashmap_get (hashmap *, memid);
+
+/* Given a search result, returns the size. */
+size_t hm_search_result_size (hashmap_search_result *);
+
+/* Given a search result, returns the pointer. */
+shared_mem_ptr hm_search_result_ptr (hashmap_search_result *);
+
+/* Given a search result, returns whether something was found. */
+bool hm_search_result_contains (hashmap_search_result *);
+
+/* Sets the hashmap entry for memid to shared_mem_ptr and
+ size_t. Optionally, if a hashmap_search_result is supplied, it is
+ used to make the lookup faster. */
+
+void hashmap_set (hashmap *, memid, hashmap_search_result *, shared_mem_ptr p,
+ size_t);
+
+/* Increments the hashmap entry for memid. Optionally, if a
+ hashmap_search_result is supplied, it is used to make the lookup
+ faster. */
+
+void hashmap_inc (hashmap *, memid, hashmap_search_result *);
+
+/* Same, but decrement. */
+int hashmap_dec (hashmap *, memid, hashmap_search_result *);
+
+void dump_hm (hashmap *hm);
+
+#define HASHMAP_H
+#endif
diff --git a/libgfortran/nca/libcoarraynative.h b/libgfortran/nca/libcoarraynative.h
new file mode 100644
index 00000000000..507de0cde8e
--- /dev/null
+++ b/libgfortran/nca/libcoarraynative.h
@@ -0,0 +1,103 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#ifndef LIBGFOR_H
+#error "Include libgfortran.h before libcoarraynative.h"
+#endif
###AV: Remove, because in fifth line following you are including libgfortran.h.
+
+#ifndef COARRAY_NATIVE_HDR
+#define COARRAY_NATIVE_HDR
+
+#include "libgfortran.h"
+
+#include <sys/types.h>
+#include <stdint.h>
+#include <stdio.h>
+
+
+/* This is to create a _nca_gfortrani_ prefix for all variables and
+ function used only by nca. */
+#if 0
+#define NUM_ADDR_BITS (8 * sizeof (int *))
+#endif
###AV: But it can't be used. Remove!
+
+#define DEBUG_NATIVE_COARRAY 1
+
+#ifdef DEBUG_NATIVE_COARRAY
+#define DEBUG_PRINTF(...) dprintf (2,__VA_ARGS__)
+#else
+#define DEBUG_PRINTF(...) do {} while(0)
+#endif
+
+#include "allocator.h"
+#include "hashmap.h"
+#include "sync.h"
+#include "lock.h"
+#include "collective_subroutine.h"
+
+typedef struct {
+ pthread_barrier_t barrier;
+ int maximg;
+} ipcollsub;
+
+typedef enum {
+ IMAGE_UNKNOWN = 0,
+ IMAGE_OK,
+ IMAGE_FAILED
+} image_status;
+
+typedef struct {
+ image_status status;
+ pid_t pid;
+} image_tracker;
+
+typedef struct {
+ int has_failed_image;
+ image_tracker images[];
+} master;
+
+typedef struct {
+ int image_num;
+ master *m;
+} image;
+
+extern image this_image;
+
+typedef struct {
+ int num_images;
+ shared_memory sm;
+ alloc_iface ai;
+ collsub_iface ci;
+ sync_iface si;
+} nca_local_data;
+
+extern nca_local_data *local;
+internal_proto (local);
+void ensure_initialization(void);
+internal_proto(ensure_initialization);
+
+void nca_master(void (*)(void));
+export_proto (nca_master);
+
+#endif
diff --git a/libgfortran/nca/lock.h b/libgfortran/nca/lock.h
new file mode 100644
index 00000000000..469739598c5
--- /dev/null
+++ b/libgfortran/nca/lock.h
@@ -0,0 +1,37 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#ifndef COARRAY_LOCK_HDR
+#define COARRAY_LOCK_HDR
+
+#include <pthread.h>
+
+typedef struct {
+ int owner;
+ int initialized;
+ pthread_mutex_t arr[];
+} lock_array;
+
+#endif
diff --git a/libgfortran/nca/shared_memory.c b/libgfortran/nca/shared_memory.c
new file mode 100644
index 00000000000..bc3093d0ef2
--- /dev/null
+++ b/libgfortran/nca/shared_memory.c
@@ -0,0 +1,221 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#include "libgfortran.h"
+#include "libcoarraynative.h"
+#include "util.h"
+#include <sys/mman.h>
+#include <unistd.h>
+#include <string.h>
+
+#include "shared_memory.h"
+
+/* This implements shared memory based on POSIX mmap. We start with
+ memory block of the size of the global shared memory data, rounded
+ up to one pagesize, and enlarge as needed.
+
+ We address the memory via a shared_memory_ptr, which is an offset into
+ the shared memory block. The metadata is situated at offset 0.
+
+ In order to be able to resize the memory and to keep pointers
+ valid, we keep the old mapping around, so the memory is actually
+ visible several times to the process. Thus, pointers returned by
+ shared_memory_get_mem_with_alignment remain valid even when
+ resizing. */
+
+/* Global metadata for shared memory, always kept at offset 0. */
+
+typedef struct
+{
+ size_t size;
+ size_t used;
+ int fd;
+} global_shared_memory_meta;
+
+/* Type realization for opaque type shared_memory. */
+
+typedef struct shared_memory_act
+{
+ global_shared_memory_meta *meta;
+ void *header;
+ size_t last_seen_size;
+
+ /* We don't need to free these. We probably also don't need to keep
+ track of them, but it is much more future proof if we do. */
+
+ size_t num_local_allocs;
+
+ struct local_alloc {
+ void *base;
+ size_t size;
+ } allocs[];
+
+} shared_memory_act;
+
+/* Convenience wrapper for mmap. */
+
+static inline void *
+map_memory (int fd, size_t size, off_t offset)
+{
+ void *ret = mmap (NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, offset);
+ if (ret == MAP_FAILED)
+ {
+ perror("mmap failed");
+ exit(1);
+ }
+ return ret;
+}
+
+/* Returns the size of shared_memory_act. */
+
+static inline size_t
+get_shared_memory_act_size (int nallocs)
+{
+ return sizeof(shared_memory_act) + nallocs*sizeof(struct local_alloc);
+}
+
+/* When the shared memory block is enlarged, we need to map it into
+ virtual memory again. */
+
+static inline shared_memory_act *
+new_base_mapping (shared_memory_act *mem)
+{
+ shared_memory_act *newmem;
+ /* We need another entry in the alloc table. */
+ mem->num_local_allocs++;
+ newmem = realloc (mem, get_shared_memory_act_size (mem->num_local_allocs));
+ newmem->allocs[newmem->num_local_allocs - 1]
+ = ((struct local_alloc)
+ {.base = map_memory (newmem->meta->fd, newmem->meta->size, 0),
+ .size = newmem->meta->size});
+ newmem->last_seen_size = newmem->meta->size;
+ return newmem;
+}
+
+/* Return the most recently allocated base pointer. */
+
+static inline void *
+last_base (shared_memory_act *mem)
+{
+ return mem->allocs[mem->num_local_allocs - 1].base;
+}
+
+/* Get a pointer into the shared memory block with alignemnt
+ (works similar to sbrk). */
+
+shared_mem_ptr
+shared_memory_get_mem_with_alignment (shared_memory_act **pmem, size_t size,
+ size_t align)
+{
+ shared_memory_act *mem = *pmem;
+ size_t new_size;
+ size_t orig_used;
+
+ /* Offset into memory block with alignment. */
+ size_t used_wa = alignto (mem->meta->used, align);
+
+ if (used_wa + size <= mem->meta->size)
+ {
+ memset(last_base(mem) + mem->meta->used, 0xCA, used_wa - mem->meta->used);
+ memset(last_base(mem) + used_wa, 0x42, size);
+ mem->meta->used = used_wa + size;
+
+ DEBUG_PRINTF ("Shared Memory: New memory of size %#lx requested, returned %#lx\n", size, used_wa);
###AV: Consistently use dprintf or DEBUG_PRINTF.
+ return (shared_mem_ptr) {.offset = used_wa};
+ }
+
+ /* We need to enlarge the memory segment. Double the size if that
+ is big enough, otherwise get what's needed. */
+
+ if (mem->meta->size * 2 < used_wa + size)
+ new_size = mem->meta->size * 2;
+ else
+ new_size = round_to_pagesize (used_wa + size);
+
+ orig_used = mem->meta->used;
+ mem->meta->size = new_size;
+ mem->meta->used = used_wa + size;
+ ftruncate (mem->meta->fd, mem->meta->size);
+ /* This also sets the new base pointer where the shared memory
+ can be found in the address space. */
+
+ mem = new_base_mapping (mem);
+
+ *pmem = mem;
+ assert(used_wa != 0);
+
+ dprintf(2, "Shared Memory: New memory of size %#lx requested, returned %#lx\n", size, used_wa);
+ memset(last_base(mem) + orig_used, 0xCA, used_wa - orig_used);
+ memset(last_base(mem) + used_wa, 0x42, size);
+
+ return (shared_mem_ptr) {.offset = used_wa};
+}
+
+/* If another image changed the size, update the size accordingly. */
+
+void
+shared_memory_prepare (shared_memory_act **pmem)
+{
+ shared_memory_act *mem = *pmem;
+ if (mem->meta->size == mem->last_seen_size)
+ return;
+ mem = new_base_mapping(mem);
+ *pmem = mem;
+}
+
+/* Initialize the memory with one page, the shared metadata of the
+ shared memory is stored at the beginning. */
+
+void
+shared_memory_init (shared_memory_act **pmem)
+{
+ shared_memory_act *mem;
+ int fd;
+ size_t initial_size = round_to_pagesize (sizeof (global_shared_memory_meta));
+
+ mem = malloc (get_shared_memory_act_size(1));
+ fd = get_shmem_fd();
+
+ ftruncate(fd, initial_size);
+ mem->meta = map_memory (fd, initial_size, 0);
+ *mem->meta = ((global_shared_memory_meta) {.size = initial_size,
+ .used = sizeof(global_shared_memory_meta),
+ .fd = fd});
+ mem->last_seen_size = initial_size;
+ mem->num_local_allocs = 1;
+ mem->allocs[0] = ((struct local_alloc) {.base = mem->meta,
+ .size = initial_size});
+
+ *pmem = mem;
+}
+
+/* Convert a shared memory pointer (i.e. an offset into the shared
+ memory block) to a pointer. */
+
+void *
+shared_mem_ptr_to_void_ptr(shared_memory_act **pmem, shared_mem_ptr smp)
+{
+ return last_base(*pmem) + smp.offset;
+}
+
diff --git a/libgfortran/nca/shared_memory.h b/libgfortran/nca/shared_memory.h
new file mode 100644
index 00000000000..4adc104801d
--- /dev/null
+++ b/libgfortran/nca/shared_memory.h
@@ -0,0 +1,78 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#ifndef SHARED_MEMORY_H
+#include <stdbool.h>
+#include <stdint.h>
+#include <stddef.h>
+#include <sys/types.h>
+#include <pthread.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <assert.h>
+#include <limits.h>
+
+/* A struct to serve as an opaque shared memory object. */
+
+struct shared_memory_act;
+typedef struct shared_memory_act * shared_memory;
+
+#define SHMPTR_NULL ((shared_mem_ptr) {.offset = -1})
+#define SHMPTR_IS_NULL(x) (x.offset == -1)
+
+#define SHMPTR_DEREF(x, s, sm) \
+ ((x) = *(__typeof(x) *) shared_mem_ptr_to_void_ptr (sm, s);
+#define SHMPTR_AS(t, s, sm) ((t) shared_mem_ptr_to_void_ptr(sm, s))
+#define SHMPTR_SET(v, s, sm) (v = SHMPTR_AS(__typeof(v), s, sm))
+#define SHMPTR_EQUALS(s1, s2) (s1.offset == s2.offset)
+
+#define SHARED_MEMORY_RAW_ALLOC(mem, t, n) \
+ shared_memory_get_mem_with_alignment(mem, sizeof(t)*n, __alignof__(t))
+
+#define SHARED_MEMORY_RAW_ALLOC_PTR(mem, t) \
+ SHMPTR_AS (t *, SHARED_MEMORY_RAW_ALLOC (mem, t, 1), mem)
+
+/* A shared-memory pointer is implemented as an offset into the shared
+ memory region. */
+
+typedef struct shared_mem_ptr
+{
+ ssize_t offset;
+} shared_mem_ptr;
+
+void shared_memory_init (shared_memory *);
+internal_proto (shared_memory_init);
+
+void shared_memory_prepare (shared_memory *);
+internal_proto (shared_memory_prepare);
+
+shared_mem_ptr shared_memory_get_mem_with_alignment (shared_memory *mem,
+ size_t size, size_t align);
+internal_proto (shared_memory_get_mem_with_alignment);
+
+void *shared_mem_ptr_to_void_ptr (shared_memory *, shared_mem_ptr);
+internal_proto (shared_mem_ptr_to_void_ptr);
+
+#define SHARED_MEMORY_H
###AV: By convention this goes directly after the '#ifndef SHARED_MEMORY_H'
###AV: at the top of the file, to prevent multiple includes.
+#endif
diff --git a/libgfortran/nca/sync.c b/libgfortran/nca/sync.c
new file mode 100644
index 00000000000..6d7f7caee47
--- /dev/null
+++ b/libgfortran/nca/sync.c
@@ -0,0 +1,156 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+
+#include <string.h>
+
+#include "libgfortran.h"
+#include "libcoarraynative.h"
+#include "sync.h"
+#include "util.h"
+
+static void
+sync_all_init (pthread_barrier_t *b)
+{
+ pthread_barrierattr_t battr;
+ pthread_barrierattr_init (&battr);
+ pthread_barrierattr_setpshared (&battr, PTHREAD_PROCESS_SHARED);
+ pthread_barrier_init (b, &battr, local->num_images);
+ pthread_barrierattr_destroy (&battr);
+}
+
+static inline void
+lock_table (sync_iface *si)
+{
+ pthread_mutex_lock (&si->cis->table_lock);
+}
+
+static inline void
+unlock_table (sync_iface *si)
+{
+ pthread_mutex_unlock (&si->cis->table_lock);
+}
+
+static inline void
+wait_table_cond (sync_iface *si, pthread_cond_t *cond)
+{
+ pthread_cond_wait (cond,&si->cis->table_lock);
+}
+
+static int *
+get_locked_table(sync_iface *si) { // The initialization of the table has to
+ // be delayed, since we might not know the
+ // number of images when the library is
+ // initialized
+ lock_table(si);
+ return si->table;
+ /*
+ if (si->table)
+ return si->table;
+ else if (!SHMPTR_IS_NULL(si->cis->table))
+ {
+ si->table = SHMPTR_AS(int *, si->cis->table, si->sm);
+ si->triggers = SHMPTR_AS(pthread_cond_t *, si->cis->triggers, si->sm);
+ return si->table;
+ }
+
+ si->cis->table =
+ shared_malloc(si->a, sizeof(int)*local->num_images * local->num_images);
+ si->cis->triggers =
+ shared_malloc(si->a, sizeof(int)*local->num_images);
+
+ si->table = SHMPTR_AS(int *, si->cis->table, si->sm);
+ si->triggers = SHMPTR_AS(pthread_cond_t *, si->cis->triggers, si->sm);
+
+ for (int i = 0; i < local->num_images; i++)
+ initialize_shared_condition (&si->triggers[i]);
+
+ return si->table;
+ */
+}
+
+void
+sync_iface_init (sync_iface *si, alloc_iface *ai, shared_memory *sm)
+{
+ si->cis = SHMPTR_AS (sync_iface_shared *,
+ shared_malloc (get_allocator(ai),
+ sizeof(collsub_iface_shared)),
+ sm);
+ DEBUG_PRINTF ("%s: num_images is %d\n", __PRETTY_FUNCTION__, local->num_images);
+
+ sync_all_init (&si->cis->sync_all);
+ initialize_shared_mutex (&si->cis->table_lock);
+ si->sm = sm;
+ si->a = get_allocator(ai);
+
+ si->cis->table =
+ shared_malloc(si->a, sizeof(int)*local->num_images * local->num_images);
+ si->cis->triggers =
+ shared_malloc(si->a, sizeof(pthread_cond_t)*local->num_images);
+
+ si->table = SHMPTR_AS(int *, si->cis->table, si->sm);
+ si->triggers = SHMPTR_AS(pthread_cond_t *, si->cis->triggers, si->sm);
+
+ for (int i = 0; i < local->num_images; i++)
+ initialize_shared_condition (&si->triggers[i]);
+}
+
+void
+sync_table (sync_iface *si, int *images, size_t size)
+{
+#ifdef DEBUG_NATIVE_COARRAY
+ dprintf (2, "Image %d waiting for these %ld images: ", this_image.image_num + 1, size);
+ for (int d_i = 0; d_i < size; d_i++)
+ dprintf (2, "%d ", images[d_i]);
+ dprintf (2, "\n");
+#endif
+ size_t i;
+ int done;
+ int *table = get_locked_table(si);
+ for (i = 0; i < size; i++)
+ {
+ table[images[i] - 1 + local->num_images*this_image.image_num]++;
+ pthread_cond_signal (&si->triggers[images[i] - 1]);
+ }
+ for (;;)
+ {
+ done = 1;
+ for (i = 0; i < size; i++)
+ done &= si->table[images[i] - 1 + this_image.image_num*local->num_images]
+ == si->table[this_image.image_num + (images[i] - 1)*local->num_images];
+ if (done)
+ break;
+ wait_table_cond (si, &si->triggers[this_image.image_num]);
+ }
+ unlock_table (si);
+}
+
+void
+sync_all (sync_iface *si)
+{
+
+ DEBUG_PRINTF("Syncing all\n");
+
+ pthread_barrier_wait (&si->cis->sync_all);
+}
diff --git a/libgfortran/nca/sync.h b/libgfortran/nca/sync.h
new file mode 100644
index 00000000000..4b494416d6a
--- /dev/null
+++ b/libgfortran/nca/sync.h
@@ -0,0 +1,56 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#ifndef IPSYNC_HDR
+#define IPSYNC_HDR
+
+#include "shared_memory.h"
+#include "alloc.h"
+#include<pthread.h>
###AV: Space between #include and opening '<', please.
+
+typedef struct {
+ pthread_barrier_t sync_all;
+ pthread_mutex_t table_lock;
+ shared_mem_ptr table;
+ shared_mem_ptr triggers;
+} sync_iface_shared;
+
+typedef struct {
+ sync_iface_shared *cis;
+ shared_memory *sm;
+ allocator *a;
+ int *table; // we can cache the table and the trigger pointers here
+ pthread_cond_t *triggers;
+} sync_iface;
+
+void sync_iface_init (sync_iface *, alloc_iface *, shared_memory *);
+internal_proto (sync_iface_init);
+
+void sync_all (sync_iface *);
+internal_proto (sync_all);
+
+void sync_table (sync_iface *, int *, size_t);
+internal_proto (sync_table);
+
+#endif
diff --git a/libgfortran/nca/util.c b/libgfortran/nca/util.c
new file mode 100644
index 00000000000..5805218f18c
--- /dev/null
+++ b/libgfortran/nca/util.c
@@ -0,0 +1,197 @@
+#include "libgfortran.h"
+#include "util.h"
+#include <string.h>
+#include <stddef.h>
+#include <stdlib.h>
+#include <limits.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+
+#define MEMOBJ_NAME "/gfortran_coarray_memfd"
###AV: This is the filename of the shm-region right? On Linux it might not be
###AV: possible to write to the root-directory for the unprivileged user.
###AV: At least on Linux this should be /dev/shm/... Not?
+
+size_t
+alignto(size_t size, size_t align) {
+ return align*((size + align - 1)/align);
+}
+
+size_t pagesize;
+
+size_t
+round_to_pagesize(size_t s) {
+ return alignto(s, pagesize);
+}
+
+size_t
+next_power_of_two(size_t size) {
+ return 1 << (PTR_BITS - __builtin_clzl(size-1)); //FIXME: There's an off-by-one error, I can feel it
###AV: size == 0 ???
+}
+
+void
+initialize_shared_mutex (pthread_mutex_t *mutex)
+{
+ pthread_mutexattr_t mattr;
+ pthread_mutexattr_init (&mattr);
+ pthread_mutexattr_setpshared (&mattr, PTHREAD_PROCESS_SHARED);
+ pthread_mutex_init (mutex, &mattr);
+ pthread_mutexattr_destroy (&mattr);
+}
+
+void
+initialize_shared_condition (pthread_cond_t *cond)
+{
+ pthread_condattr_t cattr;
+ pthread_condattr_init (&cattr);
+ pthread_condattr_setpshared (&cattr, PTHREAD_PROCESS_SHARED);
+ pthread_cond_init (cond, &cattr);
+ pthread_condattr_destroy (&cattr);
+}
+
+int
+get_shmem_fd (void)
+{
+ char buffer[1<<10];
+ int fd, id;
+ id = random ();
+ do
+ {
+ snprintf (buffer, sizeof (buffer),
+ MEMOBJ_NAME "_%u_%d", (unsigned int) getpid (), id++);
+ fd = shm_open (buffer, O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR);
+ }
+ while (fd == -1);
+ shm_unlink (buffer);
+ return fd;
+}
+
+bool
+pack_array_prepare (pack_info * restrict pi, const gfc_array_char * restrict source)
+{
+ index_type dim;
+ bool packed;
+ index_type span;
+ index_type type_size;
+ index_type ssize;
+
+ dim = GFC_DESCRIPTOR_RANK (source);
+ type_size = GFC_DESCRIPTOR_SIZE (source);
+ ssize = type_size;
+
+ pi->num_elem = 1;
+ packed = true;
+ span = source->span != 0 ? source->span : type_size;
+ for (index_type n = 0; n < dim; n++)
+ {
+ pi->stride[n] = GFC_DESCRIPTOR_STRIDE (source,n) * span;
+ pi->extent[n] = GFC_DESCRIPTOR_EXTENT (source,n);
+ if (pi->extent[n] <= 0)
+ {
+ /* Do nothing. */
+ packed = 1;
+ pi->num_elem = 0;
+ break;
+ }
+
+ if (ssize != pi->stride[n])
+ packed = 0;
+
+ pi->num_elem *= pi->extent[n];
+ ssize *= pi->extent[n];
+ }
+
+ return packed;
+}
+
+void
+pack_array_finish (pack_info * const restrict pi, const gfc_array_char * const restrict source,
+ char * restrict dest)
+{
+ index_type dim;
+ const char *restrict src;
+
+ index_type size;
+ index_type stride0;
+ index_type count[GFC_MAX_DIMENSIONS];
+
+ dim = GFC_DESCRIPTOR_RANK (source);
+ src = source->base_addr;
+ stride0 = pi->stride[0];
+ size = GFC_DESCRIPTOR_SIZE (source);
+
+ memset (count, 0, sizeof(count));
+ while (src)
+ {
+ /* Copy the data. */
+ memcpy(dest, src, size);
+ /* Advance to the next element. */
+ dest += size;
+ src += stride0;
+ count[0]++;
+ /* Advance to the next source element. */
+ index_type n = 0;
+ while (count[n] == pi->extent[n])
+ {
+ /* When we get to the end of a dimension, reset it and increment
+ the next dimension. */
+ count[n] = 0;
+ /* We could precalculate these products, but this is a less
+ frequently used path so probably not worth it. */
+ src -= pi->stride[n] * pi->extent[n];
+ n++;
+ if (n == dim)
+ {
+ src = NULL;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ src += pi->stride[n];
+ }
+ }
+ }
+}
+
+void
+unpack_array_finish (pack_info * const restrict pi,
+ const gfc_array_char * restrict d,
+ const char *restrict src)
+{
+ index_type stride0;
+ char * restrict dest;
+ index_type size;
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type dim;
+
+ size = GFC_DESCRIPTOR_SIZE (d);
+ stride0 = pi->stride[0];
+ dest = d->base_addr;
+ dim = GFC_DESCRIPTOR_RANK (d);
+
+ while (dest)
+ {
+ memcpy (dest, src, size);
+ src += size;
+ dest += stride0;
+ count[0]++;
+ index_type n = 0;
+ while (count[n] == pi->extent[n])
+ {
+ count[n] = 0;
+ dest -= pi->stride[n] * pi->extent[n];
+ n++;
+ if (n == dim)
+ {
+ dest = NULL;
+ break;
+ }
+ else
+ {
+ count[n] ++;
+ dest += pi->stride[n];
+ }
+ }
+ }
+}
diff --git a/libgfortran/nca/util.h b/libgfortran/nca/util.h
new file mode 100644
index 00000000000..9abd7adf708
--- /dev/null
+++ b/libgfortran/nca/util.h
@@ -0,0 +1,86 @@
+/* Copyright (C) 2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#ifndef UTIL_HDR
+#define UTIL_HDR
+
+#include <stdint.h>
+#include <stddef.h>
+#include <pthread.h>
+
+#define PTR_BITS (CHAR_BIT*sizeof(void *))
+
+size_t alignto (size_t, size_t);
+internal_proto (alignto);
+
+size_t round_to_pagesize (size_t);
+internal_proto (round_to_pagesize);
+
+size_t next_power_of_two (size_t);
+internal_proto (next_power_of_two);
+
+int get_shmem_fd (void);
+internal_proto (get_shmem_fd);
+
+void initialize_shared_mutex (pthread_mutex_t *);
+internal_proto (initialize_shared_mutex);
+
+void initialize_shared_condition (pthread_cond_t *);
+internal_proto (initialize_shared_condition);
+
+extern size_t pagesize;
+internal_proto (pagesize);
+
+/* Usage:
+ pack_info pi;
+ packed = pack_array_prepare (&pi, source);
+
+ // Awesome allocation of destptr using pi.num_elem
+ if (packed)
+ memcpy (...);
+ else
+ pack_array_finish (&pi, source, destptr);
+
+ This could also be used in in_pack_generic.c. Additionally, since
+ pack_array_prepare is the same for all type sizes, we would only have to
+ specialize pack_array_finish, saving on code size. */
+
+typedef struct
+{
+ index_type num_elem;
+ index_type extent[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS]; /* Stride is byte-based. */
+} pack_info;
+
+bool pack_array_prepare (pack_info *restrict, const gfc_array_char * restrict);
+internal_proto (pack_array_prepare);
+
+void pack_array_finish (pack_info * const restrict, const gfc_array_char * const restrict,
+ char * restrict);
+
+internal_proto (pack_array_finish);
+
+void unpack_array_finish (pack_info * const restrict, const gfc_array_char * const,
+ const char * restrict);
+#endif
diff --git a/libgfortran/nca/wrapper.c b/libgfortran/nca/wrapper.c
new file mode 100644
index 00000000000..eeb64d3aac9
--- /dev/null
+++ b/libgfortran/nca/wrapper.c
@@ -0,0 +1,258 @@
+/* Copyright (C) 2019-2020 Free Software Foundation, Inc.
+ Contributed by Nicolas Koenig
+
+This file is part of the GNU Fortran Native Coarray Library (libnca).
+
+Libnca is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 3, or (at your option)
+any later version.
+
+Libnca is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
+
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
+
+#include <string.h>
+#include "libgfortran.h"
+#include "libcoarraynative.h"
+#include "sync.h"
+#include "lock.h"
+#include "util.h"
+#include "collective_subroutine.h"
+
+static inline int
+div_ru (int divident, int divisor)
+{
+ return (divident + divisor - 1)/divisor;
+}
+
+enum gfc_coarray_allocation_type {
+ GFC_NCA_NORMAL_COARRAY = 3,
+ GFC_NCA_LOCK_COARRAY,
+ GFC_NCA_EVENT_COARRAY,
+};
+
+void nca_coarray_alloc (gfc_array_void *, int, int, int);
+export_proto (nca_coarray_alloc);
+
+void
+nca_coarray_free (gfc_array_void *, int);
+export_proto (nca_coarray_free);
+
+int nca_coarray_this_image (int);
+export_proto (nca_coarray_this_image);
+
+int nca_coarray_num_images (int);
+export_proto (nca_coarray_num_images);
+
+void nca_coarray_sync_all (int *);
+export_proto (nca_coarray_sync_all);
+
+void nca_sync_images (size_t, int *, int*, char *, size_t);
+export_proto (nca_sync_images);
+
+void nca_lock (void *);
+export_proto (nca_lock);
+
+void nca_unlock (void *);
+export_proto (nca_unlock);
+
+void nca_collsub_reduce_array (gfc_array_char *, void (*) (void *, void *),
+ int *);
+export_proto (nca_collsub_reduce_array);
+
+void nca_collsub_reduce_scalar (void *, index_type, void (*) (void *, void *),
+ int *);
+export_proto (nca_collsub_reduce_scalar);
+
+void nca_collsub_broadcast_array (gfc_array_char * restrict, int/*, int *, char *,
+ size_t*/);
+export_proto (nca_collsub_broadcast_array);
+
+void nca_collsub_broadcast_scalar (void * restrict, size_t, int/*, int *, char *,
+ size_t*/);
+export_proto(nca_collsub_broadcast_scalar);
+
+void
+nca_coarray_alloc (gfc_array_void *desc, int elem_size, int corank,
+ int alloc_type)
+{
+ int i, last_rank_index;
+ int num_coarray_elems, num_elems; /* Excludes the last dimension, because it
+ will have to be determined later. */
+ int extent_last_codimen;
+ size_t last_lbound;
+ size_t size_in_bytes;
+
+ ensure_initialization(); /* This function might be the first one to be
+ called, if it is called in a constructor. */
+
+ if (alloc_type == GFC_NCA_LOCK_COARRAY)
+ elem_size = sizeof (pthread_mutex_t);
+ else if (alloc_type == GFC_NCA_EVENT_COARRAY)
+ elem_size = sizeof(char); /* replace with proper type. */
+
+ last_rank_index = GFC_DESCRIPTOR_RANK(desc) + corank -1;
+
+ num_elems = 1;
+ num_coarray_elems = 1;
+ for (i = 0; i < GFC_DESCRIPTOR_RANK(desc); i++)
+ num_elems *= GFC_DESCRIPTOR_EXTENT(desc, i);
+ for (i = GFC_DESCRIPTOR_RANK(desc); i < last_rank_index; i++)
+ {
+ num_elems *= GFC_DESCRIPTOR_EXTENT(desc, i);
+ num_coarray_elems *= GFC_DESCRIPTOR_EXTENT(desc, i);
+ }
+
+ extent_last_codimen = div_ru (local->num_images, num_coarray_elems);
+
+ last_lbound = GFC_DIMENSION_LBOUND(desc->dim[last_rank_index]);
+ GFC_DIMENSION_SET(desc->dim[last_rank_index], last_lbound,
+ last_lbound + extent_last_codimen - 1,
+ num_elems);
+
+ size_in_bytes = elem_size * num_elems * extent_last_codimen;
+ if (alloc_type == GFC_NCA_LOCK_COARRAY)
+ {
+ lock_array *addr;
+ int expected = 0;
+ /* Allocate enough space for the metadata infront of the lock
+ array. */
+ addr = get_memory_by_id_zero (&local->ai, size_in_bytes
+ + sizeof (lock_array),
+ (intptr_t) desc);
+
+ /* Use of a traditional spin lock to avoid race conditions with
+ the initization of the mutex. We could alternatively put a
+ global lock around allocate, but that would probably be
+ slower. */
+ while (!__atomic_compare_exchange_n (&addr->owner, &expected,
+ this_image.image_num + 1,
+ false, __ATOMIC_SEQ_CST,
+ __ATOMIC_SEQ_CST));
+ if (!addr->initialized++)
+ {
+ for (i = 0; i < local->num_images; i++)
+ initialize_shared_mutex (&addr->arr[i]);
+ }
+ __atomic_store_n (&addr->owner, 0, __ATOMIC_SEQ_CST);
+ desc->base_addr = &addr->arr;
+ }
+ else if (alloc_type == GFC_NCA_EVENT_COARRAY)
+ (void) 0; // TODO
+ else
+ desc->base_addr = get_memory_by_id (&local->ai, size_in_bytes,
+ (intptr_t) desc);
+ dprintf(2, "Base address of desc for image %d: %p\n", this_image.image_num + 1, desc->base_addr);
+}
+
+void
+nca_coarray_free (gfc_array_void *desc, int alloc_type)
+{
+ int i;
+ if (alloc_type == GFC_NCA_LOCK_COARRAY)
+ {
+ lock_array *la;
+ int expected = 0;
+ la = desc->base_addr - offsetof (lock_array, arr);
+ while (!__atomic_compare_exchange_n (&la->owner, &expected,
+ this_image.image_num+1,
+ false, __ATOMIC_SEQ_CST,
+ __ATOMIC_SEQ_CST));
+ if (!--la->initialized)
+ {
+ /* Coarray locks can be removed and just normal
+ pthread_mutex can be used. */
+ for (i = 0; i < local->num_images; i++)
+ pthread_mutex_destroy (&la->arr[i]);
+ }
+ __atomic_store_n (&la->owner, 0, __ATOMIC_SEQ_CST);
+ }
+ else if (alloc_type == GFC_NCA_EVENT_COARRAY)
+ (void) 0; //TODO
+
+ free_memory_with_id (&local->ai, (intptr_t) desc);
+ desc->base_addr = NULL;
+}
+
+int
+nca_coarray_this_image (int distance __attribute__((unused)))
+{
+ return this_image.image_num + 1;
+}
+
+int
+nca_coarray_num_images (int distance __attribute__((unused)))
+{
+ return local->num_images;
+}
+
+void
+nca_coarray_sync_all (int *stat __attribute__((unused)))
+{
+ sync_all (&local->si);
+}
+
+void
+nca_sync_images (size_t s, int *images,
+ int *stat __attribute__((unused)),
+ char *error __attribute__((unused)),
+ size_t err_size __attribute__((unused)))
+{
+ sync_table (&local->si, images, s);
+}
+
+void
+nca_lock (void *lock)
+{
+ pthread_mutex_lock (lock);
+}
+
+void
+nca_unlock (void *lock)
+{
+ pthread_mutex_unlock (lock);
+}
+
+void
+nca_collsub_reduce_array (gfc_array_char *desc, void (*assign_function) (void *, void *),
+ int *result_image)
+{
+ collsub_reduce_array (&local->ci, desc, result_image, assign_function);
+}
+
+void
+nca_collsub_reduce_scalar (void *obj, index_type elem_size,
+ void (*assign_function) (void *, void *),
+ int *result_image)
+{
+ collsub_reduce_scalar (&local->ci, obj, elem_size, result_image, assign_function);
+}
+
+void
+nca_collsub_broadcast_array (gfc_array_char * restrict a, int source_image
+ /* , int *stat __attribute__ ((unused)),
+ char *errmsg __attribute__ ((unused)),
+ size_t errmsg_len __attribute__ ((unused))*/)
+{
+ collsub_broadcast_array (&local->ci, a, source_image - 1);
+}
+
+void
+nca_collsub_broadcast_scalar (void * restrict obj, size_t size, int source_image/*,
+ int *stat __attribute__((unused)),
+ char *errmsg __attribute__ ((unused)),
+ size_t errmsg_len __attribute__ ((unused))*/)
+{
+ collsub_broadcast_scalar (&local->ci, obj, size, source_image - 1);
+}
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!) [Review part 3]
2020-10-13 13:01 ` [RFC] Native Coarrays (finally!) [Review part 3] Andre Vehreschild
@ 2020-10-14 13:27 ` Thomas Koenig
2020-10-16 8:45 ` Andre Vehreschild
0 siblings, 1 reply; 17+ messages in thread
From: Thomas Koenig @ 2020-10-14 13:27 UTC (permalink / raw)
To: Andre Vehreschild, Nicolas König; +Cc: fortran
Hi Andre,
just one remark on one of your remarks.
+{
+ index_type count[GFC_MAX_DIMENSIONS];
+ index_type stride[GFC_MAX_DIMENSIONS]; /* stride is byte-based here. */
###AV: When it's the stride of an array_descriptor it usually is not
byte based,
You're correct, it is usually not byte based, but that is a design
mistake that I hope to partially rectify, at least as far as
library code is concerned, for the gcc 11 timeframe. (This is
PR 95101, but that doesn't have a lot of explanation).
Basically, when we do something like
type foo
integer :: i
real :: r
end type foo
type(foo) :: a, b
and then evaluate something like
a%i = cshift(b%i,2)
we create temporary arrays for a%i and b%i. Needless to say,
this is extremely inefficient.
The descriptor has all the necessary information in the span
field (which is byte-based), so it is my aim to use that information
directly in the library function and not to generate that temporary.
The part of that not generating a temporary is easy - it is
--- a/gcc/fortran/trans-expr.c
+++ b/gcc/fortran/trans-expr.c
@@ -9823,13 +9823,9 @@ arrayfunc_assign_needs_temporary (gfc_expr *
expr1, gfc_expr * expr2)
/* If we have reached here with an intrinsic function, we do not
need a temporary except in the particular case that reallocation
- on assignment is active and the lhs is allocatable and a target,
- or a pointer which may be a subref pointer. FIXME: The last
- condition can go away when we use span in the intrinsics
- directly.*/
+ on assignment is active and the lhs is allocatable and a target. */
if (expr2->value.function.isym)
- return (flag_realloc_lhs && sym->attr.allocatable && sym->attr.target)
- || (sym->attr.pointer && sym->attr.subref_array_pointer);
+ return (flag_realloc_lhs && sym->attr.allocatable && sym->attr.target);
/* If the LHS is a dummy, we need a temporary if it is not
INTENT(OUT). */
but all the library functions would have to be adjusted for that.
Clearly, for the collective subroutines, we should avoid one extra
copy on entry and one on exit.
Best regards
Thomas
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!) [Review part 3]
2020-10-14 13:27 ` Thomas Koenig
@ 2020-10-16 8:45 ` Andre Vehreschild
2020-10-16 9:43 ` Thomas Koenig
0 siblings, 1 reply; 17+ messages in thread
From: Andre Vehreschild @ 2020-10-16 8:45 UTC (permalink / raw)
To: Thomas Koenig; +Cc: Nicolas König, fortran, Paul Richard Thomas
Hi Thomas, hi Nicolas,
I agree with you, that doing copies of array's data should be avoided, but
please keep in mind, that the gfc_descriptor_t is so to say "public". I.e.,
there are also libraries using it, that are not in the gcc repo and those
depend on the descriptor being usable as is.
Furthermore, the span-member is not necessarily the size of the data element in
the array, right @Paul? Assume a structure whose data components need an odd
number of bytes. Putting this structure into an array, the span-member would
hold the padded data size, right?
I am not sure about the above, therefore Paul is in copy. This is just like I
have memorized it (may be wrong).
Anyway, my remark was more the comment not being precise or rather misleading.
I would expect something like "/* store byte-based strides here */". Which
makes it more clear, what will been done.
Regards,
Andre
On Wed, 14 Oct 2020 15:27:19 +0200
Thomas Koenig <tkoenig@netcologne.de> wrote:
> Hi Andre,
>
> just one remark on one of your remarks.
>
> +{
> + index_type count[GFC_MAX_DIMENSIONS];
> + index_type stride[GFC_MAX_DIMENSIONS]; /* stride is byte-based here. */
>
> ###AV: When it's the stride of an array_descriptor it usually is not
> byte based,
>
> You're correct, it is usually not byte based, but that is a design
> mistake that I hope to partially rectify, at least as far as
> library code is concerned, for the gcc 11 timeframe. (This is
> PR 95101, but that doesn't have a lot of explanation).
>
> Basically, when we do something like
>
> type foo
> integer :: i
> real :: r
> end type foo
>
> type(foo) :: a, b
>
> and then evaluate something like
>
> a%i = cshift(b%i,2)
>
> we create temporary arrays for a%i and b%i. Needless to say,
> this is extremely inefficient.
>
> The descriptor has all the necessary information in the span
> field (which is byte-based), so it is my aim to use that information
> directly in the library function and not to generate that temporary.
>
> The part of that not generating a temporary is easy - it is
>
> --- a/gcc/fortran/trans-expr.c
> +++ b/gcc/fortran/trans-expr.c
> @@ -9823,13 +9823,9 @@ arrayfunc_assign_needs_temporary (gfc_expr *
> expr1, gfc_expr * expr2)
>
> /* If we have reached here with an intrinsic function, we do not
> need a temporary except in the particular case that reallocation
> - on assignment is active and the lhs is allocatable and a target,
> - or a pointer which may be a subref pointer. FIXME: The last
> - condition can go away when we use span in the intrinsics
> - directly.*/
> + on assignment is active and the lhs is allocatable and a target. */
> if (expr2->value.function.isym)
> - return (flag_realloc_lhs && sym->attr.allocatable && sym->attr.target)
> - || (sym->attr.pointer && sym->attr.subref_array_pointer);
> + return (flag_realloc_lhs && sym->attr.allocatable && sym->attr.target);
>
> /* If the LHS is a dummy, we need a temporary if it is not
> INTENT(OUT). */
>
>
> but all the library functions would have to be adjusted for that.
> Clearly, for the collective subroutines, we should avoid one extra
> copy on entry and one on exit.
>
> Best regards
>
> Thomas
>
--
Andre Vehreschild * Email: vehre ad gmx dot de
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!) [Review part 3]
2020-10-16 8:45 ` Andre Vehreschild
@ 2020-10-16 9:43 ` Thomas Koenig
2020-10-16 14:37 ` Paul Richard Thomas
0 siblings, 1 reply; 17+ messages in thread
From: Thomas Koenig @ 2020-10-16 9:43 UTC (permalink / raw)
To: Andre Vehreschild; +Cc: Nicolas König, fortran, Paul Richard Thomas
Am 16.10.20 um 10:45 schrieb Andre Vehreschild:
> Hi Thomas, hi Nicolas,
>
> I agree with you, that doing copies of array's data should be avoided, but
> please keep in mind, that the gfc_descriptor_t is so to say "public". I.e.,
> there are also libraries using it, that are not in the gcc repo and those
> depend on the descriptor being usable as is.
I am quite aware of that.
However, what can happen?
Since the last ABI revision, we set the span. Old code calling a new
library will work anyway because it will do the packing / unpacking.
New code calling the old library is something we do not, in general,
support (we do add new functions to the library), but it would also work
because we set the span anyway.
The only thing we cannot handle is new user code (without the temporary)
calling old user code (which does not handle the span). We cannot
change that without an ABI upgrade, but upgrading the library will
not be affected.
> Furthermore, the span-member is not necessarily the size of the data element in
> the array, right @Paul?
That's the point of using the span.
> Assume a structure whose data components need an odd
> number of bytes. Putting this structure into an array, the span-member would
> hold the padded data size, right?
Not the size, the distance. Fortunately, alignment requirements mean
that this will work out OK.
> I am not sure about the above, therefore Paul is in copy. This is just like I
> have memorized it (may be wrong).
>
> Anyway, my remark was more the comment not being precise or rather misleading.
> I would expect something like "/* store byte-based strides here */". Which
> makes it more clear, what will been done.
OK :-)
Thanks for looking at this!
Best regards
Thomas
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [RFC] Native Coarrays (finally!) [Review part 3]
2020-10-16 9:43 ` Thomas Koenig
@ 2020-10-16 14:37 ` Paul Richard Thomas
2021-09-03 21:46 ` Damian Rouson
0 siblings, 1 reply; 17+ messages in thread
From: Paul Richard Thomas @ 2020-10-16 14:37 UTC (permalink / raw)
To: Thomas Koenig; +Cc: Andre Vehreschild, Nicolas König, fortran
Hi Andre,
Yes, as Thomas confirms, you are correct about the function of span.
Unfortunately the originators of gfortran decided on strides in words for
array descriptors rather than bytes. As far as I am aware, all the other
vendors use lbound, stride measure and extent, with stride measure in
bytes. Once committed to this line, allowing the F95 feature of pointers to
components of derived type arrays necessitated the introduction of span.
Tobias and I got a long way to do the change over to {lbound, sm, extent}
in fortran-dev but we both ran out of availability with a long way to go.
We did this by calculating stride in words from the new array descriptor
and substituting it everywhere that the stride was needed. This resulted in
ugly and inefficient repetitions of sm/element_length in array element
addressing. When I last looked at it everything worked except coarrays,
which I broke somewhere in libcaf.
Unfortunately, I am still pressed by daytime activities so something on
this scale is well beyond my capacity. I will be concentrating on
regressions in the coming weeks.
Nicolas, I haven't responded to your coarrays patch. Well done for this
impressive bit of work and thanks to Andre for reviewing it. I sincerely
hope to see it in master, sometime in the next weeks. As has been said
elsewhere, this should be an important aid to familiarising fortran users
with coarrays. Does anybody have access to an AMD Ryzen 32 core machine?
Best regards
Paul
On Fri, 16 Oct 2020 at 10:43, Thomas Koenig <tkoenig@netcologne.de> wrote:
> Am 16.10.20 um 10:45 schrieb Andre Vehreschild:
> > Hi Thomas, hi Nicolas,
> >
> > I agree with you, that doing copies of array's data should be avoided,
> but
> > please keep in mind, that the gfc_descriptor_t is so to say "public".
> I.e.,
> > there are also libraries using it, that are not in the gcc repo and those
> > depend on the descriptor being usable as is.
>
> I am quite aware of that.
>
> However, what can happen?
>
> Since the last ABI revision, we set the span. Old code calling a new
> library will work anyway because it will do the packing / unpacking.
> New code calling the old library is something we do not, in general,
> support (we do add new functions to the library), but it would also work
> because we set the span anyway.
>
> The only thing we cannot handle is new user code (without the temporary)
> calling old user code (which does not handle the span). We cannot
> change that without an ABI upgrade, but upgrading the library will
> not be affected.
>
> > Furthermore, the span-member is not necessarily the size of the data
> element in
> > the array, right @Paul?
>
> That's the point of using the span.
>
> > Assume a structure whose data components need an odd
> > number of bytes. Putting this structure into an array, the span-member
> would
> > hold the padded data size, right?
>
> Not the size, the distance. Fortunately, alignment requirements mean
> that this will work out OK.
>
> > I am not sure about the above, therefore Paul is in copy. This is just
> like I
> > have memorized it (may be wrong).
> >
> > Anyway, my remark was more the comment not being precise or rather
> misleading.
> > I would expect something like "/* store byte-based strides here */".
> Which
> > makes it more clear, what will been done.
>
> OK :-)
>
> Thanks for looking at this!
>
> Best regards
>
> Thomas
>
--
"If you can't explain it simply, you don't understand it well enough" -
Albert Einstein
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [RFC] Native Coarrays (finally!) [Review part 3]
2020-10-16 14:37 ` Paul Richard Thomas
@ 2021-09-03 21:46 ` Damian Rouson
0 siblings, 0 replies; 17+ messages in thread
From: Damian Rouson @ 2021-09-03 21:46 UTC (permalink / raw)
To: Paul Richard Thomas; +Cc: Thomas Koenig, fortran
The email below seems to be the most recent email on native coarray
support. Has there been any more recent work or plans for committing a
patch? Could someone tell me whether the planned feature will work on
Windows without requiring the Windows Subsystem for Linux? Asking for a
friend. :)
Damian
On Fri, Oct 16, 2020 at 7:38 AM Paul Richard Thomas via Fortran <
fortran@gcc.gnu.org> wrote:
> Hi Andre,
>
> Yes, as Thomas confirms, you are correct about the function of span.
> Unfortunately the originators of gfortran decided on strides in words for
> array descriptors rather than bytes. As far as I am aware, all the other
> vendors use lbound, stride measure and extent, with stride measure in
> bytes. Once committed to this line, allowing the F95 feature of pointers to
> components of derived type arrays necessitated the introduction of span.
> Tobias and I got a long way to do the change over to {lbound, sm, extent}
> in fortran-dev but we both ran out of availability with a long way to go.
> We did this by calculating stride in words from the new array descriptor
> and substituting it everywhere that the stride was needed. This resulted in
> ugly and inefficient repetitions of sm/element_length in array element
> addressing. When I last looked at it everything worked except coarrays,
> which I broke somewhere in libcaf.
>
> Unfortunately, I am still pressed by daytime activities so something on
> this scale is well beyond my capacity. I will be concentrating on
> regressions in the coming weeks.
>
> Nicolas, I haven't responded to your coarrays patch. Well done for this
> impressive bit of work and thanks to Andre for reviewing it. I sincerely
> hope to see it in master, sometime in the next weeks. As has been said
> elsewhere, this should be an important aid to familiarising fortran users
> with coarrays. Does anybody have access to an AMD Ryzen 32 core machine?
>
> Best regards
>
> Paul
>
>
>
>
> On Fri, 16 Oct 2020 at 10:43, Thomas Koenig <tkoenig@netcologne.de> wrote:
>
> > Am 16.10.20 um 10:45 schrieb Andre Vehreschild:
> > > Hi Thomas, hi Nicolas,
> > >
> > > I agree with you, that doing copies of array's data should be avoided,
> > but
> > > please keep in mind, that the gfc_descriptor_t is so to say "public".
> > I.e.,
> > > there are also libraries using it, that are not in the gcc repo and
> those
> > > depend on the descriptor being usable as is.
> >
> > I am quite aware of that.
> >
> > However, what can happen?
> >
> > Since the last ABI revision, we set the span. Old code calling a new
> > library will work anyway because it will do the packing / unpacking.
> > New code calling the old library is something we do not, in general,
> > support (we do add new functions to the library), but it would also work
> > because we set the span anyway.
> >
> > The only thing we cannot handle is new user code (without the temporary)
> > calling old user code (which does not handle the span). We cannot
> > change that without an ABI upgrade, but upgrading the library will
> > not be affected.
> >
> > > Furthermore, the span-member is not necessarily the size of the data
> > element in
> > > the array, right @Paul?
> >
> > That's the point of using the span.
> >
> > > Assume a structure whose data components need an odd
> > > number of bytes. Putting this structure into an array, the span-member
> > would
> > > hold the padded data size, right?
> >
> > Not the size, the distance. Fortunately, alignment requirements mean
> > that this will work out OK.
> >
> > > I am not sure about the above, therefore Paul is in copy. This is just
> > like I
> > > have memorized it (may be wrong).
> > >
> > > Anyway, my remark was more the comment not being precise or rather
> > misleading.
> > > I would expect something like "/* store byte-based strides here */".
> > Which
> > > makes it more clear, what will been done.
> >
> > OK :-)
> >
> > Thanks for looking at this!
> >
> > Best regards
> >
> > Thomas
> >
>
>
> --
> "If you can't explain it simply, you don't understand it well enough" -
> Albert Einstein
>
^ permalink raw reply [flat|nested] 17+ messages in thread
end of thread, other threads:[~2021-09-03 21:47 UTC | newest]
Thread overview: 17+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-09-22 12:14 [RFC] Native Coarrays (finally!) Nicolas König
2020-09-23 6:13 ` Damian Rouson
2020-09-30 12:10 ` Andre Vehreschild
2020-09-30 14:09 ` Nicolas König
2020-09-30 14:11 ` Andre Vehreschild
2020-09-30 13:46 ` Paul Richard Thomas
2020-10-05 9:54 ` Tobias Burnus
2020-10-05 13:23 ` Nicolas König
2020-10-12 12:32 ` [RFC] Native Coarrays (finally!) [Review part 1] Andre Vehreschild
2020-10-12 13:48 ` [RFC] Native Coarrays (finally!) [Review part 2] Andre Vehreschild
2020-10-13 12:42 ` Nicolas König
2020-10-13 13:01 ` [RFC] Native Coarrays (finally!) [Review part 3] Andre Vehreschild
2020-10-14 13:27 ` Thomas Koenig
2020-10-16 8:45 ` Andre Vehreschild
2020-10-16 9:43 ` Thomas Koenig
2020-10-16 14:37 ` Paul Richard Thomas
2021-09-03 21:46 ` Damian Rouson
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).