* [PATCH] Fix code emission for FAIL_ALLOC predictor
@ 2016-06-13 13:57 Martin Liška
2016-06-15 11:36 ` Martin Liška
0 siblings, 1 reply; 3+ messages in thread
From: Martin Liška @ 2016-06-13 13:57 UTC (permalink / raw)
Cc: GCC Patches, Jan Hubicka
Hello.
Following patch fixes Fortran FAIL_ALLOC predictor in a way that it introduces a new one (PRED_FORTRAN_REALLOC)
and it properly marks returned values as described in the following 2 examples:
A) allocate_allocatable
original annotation:
if ((logical(kind=4)) __builtin_expect ((integer(kind=8)) (overflow.343 != 0), 0, 33)) // overflow
{
stat.341 = 5014;
}
else
{
if ((logical(kind=4)) __builtin_expect ((integer(kind=8)) (bx_ilow.data != 0B), 0, 34)) // fail alloc
{
stat.341 = 5014;
}
else
{
stat.341 = 0;
bx_ilow.data = (void * restrict) __builtin_malloc (MAX_EXPR <size.342, 1>);
if (bx_ilow.data == 0B)
{
stat.341 = 5014;
}
}
}
if ((logical(kind=4)) __builtin_expect ((integer(kind=8)) (stat.341 == 0), 1, 34)) // fail alloc
{
bx_ilow.dtype = 539;
bx_ilow.dim[0].lbound = (integer(kind=8)) xstart;
bx_ilow.dim[0].ubound = 1;
bx_ilow.dim[0].stride = 1;
bx_ilow.dim[1].lbound = (integer(kind=8)) ystart;
bx_ilow.dim[1].ubound = D.5342;
bx_ilow.dim[1].stride = D.5341;
bx_ilow.dim[2].lbound = (integer(kind=8)) zstart;
bx_ilow.dim[2].ubound = D.5346;
bx_ilow.dim[2].stride = D.5345;
bx_ilow.offset = D.5352;
}
I changed it to:
if ((logical(kind=4)) __builtin_expect ((integer(kind=8)) (overflow.343 != 0), 0, 33)) // overflow
{
stat.341 = 5014;
}
else
{
if ((logical(kind=4)) __builtin_expect ((integer(kind=8)) (bx_ilow.data != 0B), 0, 35)) // repeated allocation/deallocation
{
stat.341 = 5014;
}
else
{
stat.341 = 0;
bx_ilow.data = (void * restrict) __builtin_malloc (MAX_EXPR <size.342, 1>);
if ((logical(kind=4)) __builtin_expect ((integer(kind=8)) (bx_ilow.data == 0B), 0, 34)) // fail alloc
{
stat.341 = 5014;
}
}
}
if (stat.341 == 0) // no expectation
{
bx_ilow.dtype = 539;
bx_ilow.dim[0].lbound = (integer(kind=8)) xstart;
bx_ilow.dim[0].ubound = 1;
bx_ilow.dim[0].stride = 1;
bx_ilow.dim[1].lbound = (integer(kind=8)) ystart;
bx_ilow.dim[1].ubound = D.5342;
bx_ilow.dim[1].stride = D.5341;
bx_ilow.dim[2].lbound = (integer(kind=8)) zstart;
bx_ilow.dim[2].ubound = D.5346;
bx_ilow.dim[2].stride = D.5345;
bx_ilow.offset = D.5352;
}
B) array allocation
<bb 9>:
# size.1478_3210 = PHI <0(7), size.1478_3743(8)>
_21 = _3740 != 0;
_22 = (integer(kind=8)) _21;
_23 = BUILTIN_EXPECT (_22, 0, 33); // overflow
_24 = (logical(kind=4)) _23;
if (_24 != 0)
goto <bb 13>;
else
goto <bb 10>;
<bb 10>:
_25 = hrval.data;
_26 = _25 != 0B;
_27 = (integer(kind=8)) _26;
_28 = BUILTIN_EXPECT (_27, 0, 34); // fail malloc
_29 = (logical(kind=4)) _28;
if (_29 != 0)
goto <bb 13>;
else
goto <bb 11>;
<bb 11>:
_30 = MAX_EXPR <size.1478_3210, 1>;
_31 = __builtin_malloc (_30);
hrval.data = _31;
if (_31 == 0B)
goto <bb 12>;
else
goto <bb 13>;
<bb 12>:
<bb 13>:
# stat.1477_3202 = PHI <5014(9), 5014(10), 0(11), 5014(12)>
_33 = stat.1477_3202 == 0;
_34 = (integer(kind=8)) _33;
_35 = BUILTIN_EXPECT (_34, 1, 34); // fail malloc
_36 = (logical(kind=4)) _35;
if (_36 != 0)
goto <bb 14>;
else
goto <bb 15>;
currently looks as follows:
<bb 9>:
# size.1478_3210 = PHI <0(7), size.1478_3743(8)>
_21 = _3740 != 0;
_22 = (integer(kind=8)) _21;
_23 = BUILTIN_EXPECT (_22, 0, 33); // overflow
_24 = (logical(kind=4)) _23;
if (_24 != 0)
goto <bb 13>;
else
goto <bb 10>;
<bb 10>:
_25 = hrval.data;
_26 = _25 != 0B;
_27 = (integer(kind=8)) _26;
_28 = BUILTIN_EXPECT (_27, 0, 35); // repeated allocation/deallocation
_29 = (logical(kind=4)) _28;
if (_29 != 0)
goto <bb 13>;
else
goto <bb 11>;
<bb 11>:
_30 = MAX_EXPR <size.1478_3210, 1>;
_31 = __builtin_malloc (_30);
hrval.data = _31;
_33 = _31 == 0B;
_34 = (integer(kind=8)) _33;
_35 = BUILTIN_EXPECT (_34, 0, 34); // fail alloc
_36 = (logical(kind=4)) _35;
if (_36 != 0)
goto <bb 12>;
else
goto <bb 13>;
<bb 12>:
<bb 13>:
# stat.1477_3202 = PHI <5014(9), 5014(10), 0(11), 5014(12)>
if (stat.1477_3202 == 0) // no prediction
goto <bb 14>;
else
goto <bb 15>;
I get following numbers with the patch applied:
1) polyhedron benchmark (aermod.f90.061i.profile):
HEURISTICS BRANCHES (REL) HITRATE COVERAGE COVERAGE (REL)
repeated allocation/deallocation 194 4.1% 100.00% / 100.00% 194 194.00 0.0%
fail alloc 377 7.9% 100.00% / 100.00% 377 377.00 0.0%
b) 459.GemsFDTD SPEC2006 benchmark:
HEURISTICS BRANCHES (REL) HITRATE COVERAGE COVERAGE (REL)
repeated allocation/deallocation 203 4.3% 100.00% / 100.00% 203 203.00 0.0%
fail alloc 378 8.0% 100.00% / 100.00% 378 378.00 0.0%
Even though all numbers are 100%, I would suggest to set FAIL_ALLOC to PROB_VERY_LIKELY and
PRED_FORTRAN_REALLOC to PROB_LIKELY.
Ready to install the patch if it survives regression tests?
Thanks,
Martin
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] Fix code emission for FAIL_ALLOC predictor
2016-06-13 13:57 [PATCH] Fix code emission for FAIL_ALLOC predictor Martin Liška
@ 2016-06-15 11:36 ` Martin Liška
2016-06-15 11:52 ` Jan Hubicka
0 siblings, 1 reply; 3+ messages in thread
From: Martin Liška @ 2016-06-15 11:36 UTC (permalink / raw)
To: gcc-patches; +Cc: Jan Hubicka
[-- Attachment #1: Type: text/plain, Size: 30 bytes --]
Adding missing patch.
Martin
[-- Attachment #2: 0001-Fix-code-emission-for-FAIL_ALLOC-predictor.patch --]
[-- Type: text/x-patch, Size: 4680 bytes --]
From 35ba97e0139d955c04e67ca157f8899bbb468bf1 Mon Sep 17 00:00:00 2001
From: marxin <mliska@suse.cz>
Date: Thu, 9 Jun 2016 17:51:38 +0200
Subject: [PATCH] Fix code emission for FAIL_ALLOC predictor
gcc/ChangeLog:
2016-06-13 Martin Liska <mliska@suse.cz>
* predict.def: Define a new predictor.
gcc/fortran/ChangeLog:
2016-06-13 Martin Liska <mliska@suse.cz>
* trans-array.c (gfc_array_allocate): Do not generate expect
stmt.
* trans.c (gfc_allocate_using_malloc): Properly set FAIL_ALLOC
predictor for malloc return value.
(gfc_allocate_allocatable): Use REALLOC predictor instead of
FAIL_ALLOC.
(gfc_deallocate_with_status): Likewise.
---
gcc/fortran/trans-array.c | 2 +-
gcc/fortran/trans.c | 10 ++++------
gcc/predict.def | 15 +++++++++------
3 files changed, 14 insertions(+), 13 deletions(-)
diff --git a/gcc/fortran/trans-array.c b/gcc/fortran/trans-array.c
index 403ce3a..e95c8dd 100644
--- a/gcc/fortran/trans-array.c
+++ b/gcc/fortran/trans-array.c
@@ -5553,7 +5553,7 @@ gfc_array_allocate (gfc_se * se, gfc_expr * expr, tree status, tree errmsg,
build_int_cst (TREE_TYPE (status), 0));
gfc_add_expr_to_block (&se->pre,
fold_build3_loc (input_location, COND_EXPR, void_type_node,
- gfc_likely (cond, PRED_FORTRAN_FAIL_ALLOC),
+ cond,
set_descriptor,
build_empty_stmt (input_location)));
}
diff --git a/gcc/fortran/trans.c b/gcc/fortran/trans.c
index c6688d3..d6b4a56 100644
--- a/gcc/fortran/trans.c
+++ b/gcc/fortran/trans.c
@@ -672,9 +672,6 @@ gfc_allocate_using_malloc (stmtblock_t * block, tree pointer,
gfc_start_block (&on_error);
if (status != NULL_TREE)
{
- gfc_add_expr_to_block (&on_error,
- build_predict_expr (PRED_FORTRAN_FAIL_ALLOC,
- NOT_TAKEN));
tmp = fold_build2_loc (input_location, MODIFY_EXPR, status_type, status,
build_int_cst (status_type, LIBERROR_ALLOCATION));
gfc_add_expr_to_block (&on_error, tmp);
@@ -693,7 +690,8 @@ gfc_allocate_using_malloc (stmtblock_t * block, tree pointer,
boolean_type_node, pointer,
build_int_cst (prvoid_type_node, 0));
tmp = fold_build3_loc (input_location, COND_EXPR, void_type_node,
- error_cond, gfc_finish_block (&on_error),
+ gfc_unlikely (error_cond, PRED_FORTRAN_FAIL_ALLOC),
+ gfc_finish_block (&on_error),
build_empty_stmt (input_location));
gfc_add_expr_to_block (block, tmp);
@@ -796,7 +794,7 @@ gfc_allocate_allocatable (stmtblock_t * block, tree mem, tree size, tree token,
null_mem = gfc_unlikely (fold_build2_loc (input_location, NE_EXPR,
boolean_type_node, mem,
build_int_cst (type, 0)),
- PRED_FORTRAN_FAIL_ALLOC);
+ PRED_FORTRAN_REALLOC);
/* If mem is NULL, we call gfc_allocate_using_malloc or
gfc_allocate_using_lib. */
@@ -1385,7 +1383,7 @@ gfc_deallocate_with_status (tree pointer, tree status, tree errmsg,
cond2 = fold_build2_loc (input_location, NE_EXPR, boolean_type_node,
stat, build_zero_cst (TREE_TYPE (stat)));
tmp = fold_build3_loc (input_location, COND_EXPR, void_type_node,
- gfc_unlikely (cond2, PRED_FORTRAN_FAIL_ALLOC),
+ gfc_unlikely (cond2, PRED_FORTRAN_REALLOC),
tmp, build_empty_stmt (input_location));
gfc_add_expr_to_block (&non_null, tmp);
}
diff --git a/gcc/predict.def b/gcc/predict.def
index c0a3f36..da4f9ab 100644
--- a/gcc/predict.def
+++ b/gcc/predict.def
@@ -163,12 +163,15 @@ DEF_PREDICTOR (PRED_FORTRAN_OVERFLOW, "overflow", PROB_ALWAYS,
PRED_FLAG_FIRST_MATCH)
/* Branch leading to a failure status are unlikely. This can occur for out
- of memory or when trying to allocate an already allocated allocated or
- deallocating an already deallocated allocatable. This predictor only
- occurs when the user explicitly asked for a return status. By default,
- the code aborts, which is handled via PRED_NORETURN.
- FIXME: the hitrate really ought to be close to 100%. */
-DEF_PREDICTOR (PRED_FORTRAN_FAIL_ALLOC, "fail alloc", HITRATE (62), 0)
+ of memory. This predictor only occurs when the user explicitly asked
+ for a return status. By default, the code aborts,
+ which is handled via PRED_NORETURN. */
+DEF_PREDICTOR (PRED_FORTRAN_FAIL_ALLOC, "fail alloc", PROB_VERY_LIKELY, 0)
+
+/* Predictor is used for an allocation of an already allocated memory or
+ deallocating an already deallocated allocatable. */
+DEF_PREDICTOR (PRED_FORTRAN_REALLOC, "repeated allocation/deallocation", \
+ PROB_LIKELY, 0)
/* Branch leading to an I/O failure status are unlikely. This predictor is
used for I/O failures such as for invalid unit numbers. This predictor
--
2.8.3
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] Fix code emission for FAIL_ALLOC predictor
2016-06-15 11:36 ` Martin Liška
@ 2016-06-15 11:52 ` Jan Hubicka
0 siblings, 0 replies; 3+ messages in thread
From: Jan Hubicka @ 2016-06-15 11:52 UTC (permalink / raw)
To: Martin Liška; +Cc: gcc-patches, Jan Hubicka
> Adding missing patch.
>
> Martin
> >From 35ba97e0139d955c04e67ca157f8899bbb468bf1 Mon Sep 17 00:00:00 2001
> From: marxin <mliska@suse.cz>
> Date: Thu, 9 Jun 2016 17:51:38 +0200
> Subject: [PATCH] Fix code emission for FAIL_ALLOC predictor
>
> gcc/ChangeLog:
>
> 2016-06-13 Martin Liska <mliska@suse.cz>
>
> * predict.def: Define a new predictor.
>
> gcc/fortran/ChangeLog:
>
> 2016-06-13 Martin Liska <mliska@suse.cz>
>
> * trans-array.c (gfc_array_allocate): Do not generate expect
> stmt.
> * trans.c (gfc_allocate_using_malloc): Properly set FAIL_ALLOC
> predictor for malloc return value.
> (gfc_allocate_allocatable): Use REALLOC predictor instead of
> FAIL_ALLOC.
> (gfc_deallocate_with_status): Likewise.
OK, thanks!
Honza
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2016-06-15 11:52 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-06-13 13:57 [PATCH] Fix code emission for FAIL_ALLOC predictor Martin Liška
2016-06-15 11:36 ` Martin Liška
2016-06-15 11:52 ` Jan Hubicka
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).