diff --git a/gcc/cgraph.c b/gcc/cgraph.c index e505b10e211..eab8216100e 100644 --- a/gcc/cgraph.c +++ b/gcc/cgraph.c @@ -2492,6 +2492,61 @@ cgraph_node::set_nothrow_flag (bool nothrow) return changed; } +/* Worker to set malloc flag. */ +static void +set_malloc_flag_1 (cgraph_node *node, bool malloc_p, bool non_call, + bool *changed) +{ + cgraph_edge *e; + + if (malloc_p && !DECL_IS_MALLOC (node->decl)) + { + /* With non-call exceptions we can't say for sure if other function + body was not possibly optimized to still throw. */ + if (!non_call || node->binds_to_current_def_p ()) + { + DECL_IS_MALLOC (node->decl) = true; + *changed = true; + } + } + else if (!malloc_p && DECL_IS_MALLOC (node->decl)) + { + DECL_IS_MALLOC (node->decl) = false; + *changed = true; + } + + ipa_ref *ref; + FOR_EACH_ALIAS (node, ref) + { + cgraph_node *alias = dyn_cast (ref->referring); + if (!malloc_p || alias->get_availability () > AVAIL_INTERPOSABLE) + set_malloc_flag_1 (alias, malloc_p, non_call, changed); + } +} +/* Set DECL_IS_MALLOC on NODE's decl and on NODE's aliases if any. */ + +bool +cgraph_node::set_malloc_flag (bool malloc_p) +{ + bool changed = false; + bool non_call = opt_for_fn (decl, flag_non_call_exceptions); + + if (!malloc_p || get_availability () > AVAIL_INTERPOSABLE) + set_malloc_flag_1 (this, malloc_p, non_call, &changed); + else + { + ipa_ref *ref; + + FOR_EACH_ALIAS (this, ref) + { + cgraph_node *alias = dyn_cast (ref->referring); + if (!malloc_p || alias->get_availability () > AVAIL_INTERPOSABLE) + set_malloc_flag_1 (alias, malloc_p, non_call, &changed); + } + } + return changed; +} + /* Worker to set_const_flag. */ static void diff --git a/gcc/cgraph.h b/gcc/cgraph.h index be4eaee71e2..6739bb3094d 100644 --- a/gcc/cgraph.h +++ b/gcc/cgraph.h @@ -1116,6 +1116,10 @@ public: if any to NOTHROW. */ bool set_nothrow_flag (bool nothrow); + /* SET DECL_IS_MALLOC on cgraph_node's decl and on aliases of the node + if any. */ + bool set_malloc_flag (bool malloc_p); + /* If SET_CONST is true, mark function, aliases and thunks to be ECF_CONST. If SET_CONST if false, clear the flag. diff --git a/gcc/ipa-pure-const.c b/gcc/ipa-pure-const.c index e457166ea39..724c26e03f6 100644 --- a/gcc/ipa-pure-const.c +++ b/gcc/ipa-pure-const.c @@ -56,6 +56,7 @@ along with GCC; see the file COPYING3. If not see #include "tree-scalar-evolution.h" #include "intl.h" #include "opts.h" +#include "ssa.h" /* Lattice values for const and pure functions. Everything starts out being const, then may drop to pure and then neither depending on @@ -69,6 +70,15 @@ enum pure_const_state_e const char *pure_const_names[3] = {"const", "pure", "neither"}; +enum malloc_state_e +{ + PURE_CONST_MALLOC_TOP, + PURE_CONST_MALLOC, + PURE_CONST_MALLOC_BOTTOM +}; + +const char *malloc_state_names[] = {"malloc_top", "malloc", "malloc_bottom"}; + /* Holder for the const_state. There is one of these per function decl. */ struct funct_state_d @@ -92,11 +102,13 @@ struct funct_state_d /* If function can call free, munmap or otherwise make previously non-trapping memory accesses trapping. */ bool can_free; + + enum malloc_state_e malloc_state; }; /* State used when we know nothing about function. */ static struct funct_state_d varying_state - = { IPA_NEITHER, IPA_NEITHER, true, true, true, true }; + = { IPA_NEITHER, IPA_NEITHER, true, true, true, true, PURE_CONST_MALLOC_BOTTOM }; typedef struct funct_state_d * funct_state; @@ -109,6 +121,10 @@ typedef struct funct_state_d * funct_state; static vec funct_state_vec; +/* A map from node to subset of callees. The subset contains those callees + * whose return-value is returned by the node. */ +static hash_map< cgraph_node *, vec* > *return_callees_map; + static bool gate_pure_const (void); namespace { @@ -812,6 +828,124 @@ check_stmt (gimple_stmt_iterator *gsip, funct_state local, bool ipa) } } +static bool +check_retval_uses (tree retval, gimple *stmt) +{ + imm_use_iterator use_iter; + gimple *use_stmt; + + FOR_EACH_IMM_USE_STMT (use_stmt, use_iter, retval) + if (gcond *cond = dyn_cast (use_stmt)) + { + enum tree_code code = gimple_cond_code (cond); + if (TREE_CODE_CLASS (code) != tcc_comparison) + RETURN_FROM_IMM_USE_STMT (use_iter, false); + } + else if (gassign *ga = dyn_cast (use_stmt)) + { + enum tree_code code = gimple_assign_rhs_code (ga); + if (TREE_CODE_CLASS (code) != tcc_comparison) + RETURN_FROM_IMM_USE_STMT (use_iter, false); + } + else if (use_stmt != stmt) + RETURN_FROM_IMM_USE_STMT (use_iter, false); + + return true; +} + +/* + * Currently this function does a very conservative analysis to check if + * function could be a malloc candidate. + * + * The function is considered to be a candidate if + * 1) The function returns a value of pointer type. + * 2) SSA_NAME_DEF_STMT (return_value) is either a function call or + * a phi, and SSA_NAME_DEF_STMT(each element of phi) is function call. + * 3) The return-value has immediate uses only within comparisons (gcond or gassign) + * and return_stmt (and likewise a phi arg has immediate use only within comparison + * or the phi stmt). + */ + +static bool +malloc_candidate_p (function *fun, vec& callees) +{ + basic_block bb; + FOR_EACH_BB_FN (bb, fun) + { + for (gimple_stmt_iterator gsi = gsi_start_bb (bb); + !gsi_end_p (gsi); + gsi_next (&gsi)) + { + greturn *ret_stmt = dyn_cast (gsi_stmt (gsi)); + if (ret_stmt) + { + tree retval = gimple_return_retval (ret_stmt); + if (!retval) + return false; + + if (TREE_CODE (retval) != SSA_NAME + || TREE_CODE (TREE_TYPE (retval)) != POINTER_TYPE) + return false; + + if (!check_retval_uses (retval, ret_stmt)) + return false; + + gimple *def = SSA_NAME_DEF_STMT (retval); + if (gcall *call_stmt = dyn_cast (def)) + { + tree callee_decl = gimple_call_fndecl (call_stmt); + /* FIXME: In case of indirect call, callee_decl might be NULL + Not sure what to do in that case, punting for now. */ + if (!callee_decl) + return false; + cgraph_node *callee = cgraph_node::get_create (callee_decl); + callees.safe_push (callee); + } + else if (gphi *phi = dyn_cast (def)) + for (unsigned i = 0; i < gimple_phi_num_args (phi); ++i) + { + tree arg = gimple_phi_arg_def (phi, i); + if (TREE_CODE (arg) != SSA_NAME) + return false; + if (!check_retval_uses (arg, phi)) + return false; + + gimple *arg_def = SSA_NAME_DEF_STMT (arg); + gcall *call_stmt = dyn_cast (arg_def); + if (!call_stmt) + return false; + tree callee_decl = gimple_call_fndecl (call_stmt); + if (!callee_decl) + return false; + cgraph_node *callee = cgraph_node::get_create (callee_decl); + callees.safe_push (callee); + } + else + return false; + } + } + } + + if (dump_file && (dump_flags & TDF_DETAILS)) + fprintf (dump_file, "\nFound %s to be candidate for malloc attribute\n", + IDENTIFIER_POINTER (DECL_NAME (fun->decl))); + return true; +} + + +static void +free_return_callees_map (void) +{ + for (hash_map*>::iterator it = return_callees_map->begin (); + it != return_callees_map->end (); + ++it) + { + std::pair*> p = *it; + p.second->release (); + delete p.second; + } + delete return_callees_map; +} /* This is the main routine for finding the reference patterns for global variables within a function FN. */ @@ -921,6 +1055,23 @@ end: if (TREE_NOTHROW (decl)) l->can_throw = false; + if (ipa) + { + vec v = vNULL; + l->malloc_state = PURE_CONST_MALLOC_BOTTOM; + if (DECL_IS_MALLOC (decl)) + l->malloc_state = PURE_CONST_MALLOC; + else if (malloc_candidate_p (DECL_STRUCT_FUNCTION (decl), v)) + { + l->malloc_state = PURE_CONST_MALLOC_TOP; + vec *callees_p = new vec (vNULL); + for (unsigned i = 0; i < v.length (); ++i) + callees_p->safe_push (v[i]); + return_callees_map->put (fn, callees_p); + } + v.release (); + } + pop_cfun (); if (dump_file) { @@ -1003,6 +1154,7 @@ pure_const_generate_summary (void) pass_ipa_pure_const *pass = static_cast (current_pass); pass->register_hooks (); + return_callees_map = new hash_map*>; /* Process all of the functions. @@ -1067,11 +1219,30 @@ pure_const_write_summary (void) bp_pack_value (&bp, fs->looping, 1); bp_pack_value (&bp, fs->can_throw, 1); bp_pack_value (&bp, fs->can_free, 1); + bp_pack_value (&bp, fs->malloc_state, 2); streamer_write_bitpack (&bp); + + vec **callees_pp = return_callees_map->get (node); + if (callees_pp) + { + vec callees = **callees_pp; + streamer_write_uhwi_stream (ob->main_stream, callees.length ()); + for (unsigned i = 0; i < callees.length (); ++i) + { + int callee_ref = lto_symtab_encoder_encode (encoder, callees[i]); + streamer_write_uhwi_stream (ob->main_stream, callee_ref); + } + } + else + streamer_write_uhwi_stream (ob->main_stream, 0); } } lto_destroy_simple_output_block (ob); + + // FIXME: Do not free if -ffat-lto-objects is enabled. + if (!global_options.x_flag_fat_lto_objects) + free_return_callees_map (); } @@ -1087,6 +1258,8 @@ pure_const_read_summary (void) pass_ipa_pure_const *pass = static_cast (current_pass); pass->register_hooks (); + return_callees_map = new hash_map *>; + while ((file_data = file_data_vec[j++])) { const char *data; @@ -1127,6 +1300,22 @@ pure_const_read_summary (void) fs->looping = bp_unpack_value (&bp, 1); fs->can_throw = bp_unpack_value (&bp, 1); fs->can_free = bp_unpack_value (&bp, 1); + fs->malloc_state + = (enum malloc_state_e) bp_unpack_value (&bp, 2); + + unsigned len = streamer_read_uhwi (ib); + if (len) + { + vec *callees_p = new vec (vNULL); + for (unsigned i = 0; i < len; i++) + { + int callee_index = streamer_read_uhwi (ib); + cgraph_node *callee = dyn_cast (lto_symtab_encoder_deref (encoder, callee_index)); + callees_p->safe_push (callee); + } + return_callees_map->put (node, callees_p); + } + if (dump_file) { int flags = flags_from_decl_or_type (node->decl); @@ -1151,6 +1340,8 @@ pure_const_read_summary (void) fprintf (dump_file," function is locally throwing\n"); if (fs->can_free) fprintf (dump_file," function can locally free\n"); + fprintf (dump_file, "\n malloc state: %s\n", + malloc_state_names[fs->malloc_state]); } } @@ -1664,6 +1855,121 @@ propagate_nothrow (void) free (order); } +DEBUG_FUNCTION +static void +dump_malloc_lattice (FILE *dump_file, const char *s) +{ + if (!dump_file) + return; + + fprintf (dump_file, "\n\nMALLOC LATTICE %s:\n", s); + cgraph_node *node; + FOR_EACH_FUNCTION (node) + { + funct_state fs = get_function_state (node); + malloc_state_e state = fs->malloc_state; + fprintf (dump_file, "%s: %s\n", node->name (), malloc_state_names[state]); + } +} + +static void +propagate_malloc (void) +{ + cgraph_node *node; + FOR_EACH_FUNCTION (node) + { + if (DECL_IS_MALLOC (node->decl)) + if (!has_function_state (node)) + { + funct_state l = XCNEW (struct funct_state_d); + *l = varying_state; + l->malloc_state = PURE_CONST_MALLOC; + set_function_state (node, l); + } + } + + dump_malloc_lattice (dump_file, "Initial"); + struct cgraph_node **order + = XNEWVEC (struct cgraph_node *, symtab->cgraph_count); + int order_pos = ipa_reverse_postorder (order); + bool changed = true; + + while (changed) + { + changed = false; + /* Walk in postorder. */ + for (int i = order_pos - 1; i >= 0; --i) + { + cgraph_node *node = order[i]; + if (node->alias + || !node->definition + || !has_function_state (node)) + continue; + + funct_state l = get_function_state (node); + + /* FIXME: add support for indirect-calls. */ + if (node->indirect_calls) + { + l->malloc_state = PURE_CONST_MALLOC_BOTTOM; + continue; + } + + if (node->get_availability () <= AVAIL_INTERPOSABLE) + { + l->malloc_state = PURE_CONST_MALLOC_BOTTOM; + continue; + } + + if (l->malloc_state == PURE_CONST_MALLOC_BOTTOM) + continue; + + vec **callees_pp = return_callees_map->get (node); + if (!callees_pp) + continue; + + vec callees = **callees_pp; + malloc_state_e new_state = l->malloc_state; + for (unsigned j = 0; j < callees.length (); j++) + { + cgraph_node *callee = callees[j]; + if (!has_function_state (callee)) + { + new_state = PURE_CONST_MALLOC_BOTTOM; + break; + } + malloc_state_e callee_state = get_function_state (callee)->malloc_state; + if (new_state < callee_state) + new_state = callee_state; + } + if (new_state != l->malloc_state) + { + changed = true; + l->malloc_state = new_state; + } + } + } + + FOR_EACH_DEFINED_FUNCTION (node) + if (has_function_state (node)) + { + funct_state l = get_function_state (node); + if (!node->alias + && l->malloc_state == PURE_CONST_MALLOC + && !node->global.inlined_to) + { + if (dump_file && (dump_flags & TDF_DETAILS)) + fprintf (dump_file, "Function %s found to be malloc\n", + node->name ()); + node->set_malloc_flag (true); + } + } + + dump_malloc_lattice (dump_file, "after propagation"); + ipa_free_postorder_info (); + free (order); + free_return_callees_map (); +} /* Produce the global information by preforming a transitive closure on the local information that was produced by generate_summary. */ @@ -1682,6 +1988,7 @@ execute (function *) /* Nothrow makes more function to not lead to return and improve later analysis. */ propagate_nothrow (); + propagate_malloc (); remove_p = propagate_pure_const (); /* Cleanup. */ diff --git a/gcc/ssa-iterators.h b/gcc/ssa-iterators.h index c8aa77bd4f3..740cbf13cb2 100644 --- a/gcc/ssa-iterators.h +++ b/gcc/ssa-iterators.h @@ -93,6 +93,12 @@ struct imm_use_iterator break; \ } +/* Similarly for return. */ +#define RETURN_FROM_IMM_USE_STMT(ITER, VAL) \ + { \ + end_imm_use_stmt_traverse (&(ITER)); \ + return (VAL); \ + } /* Use this iterator in combination with FOR_EACH_IMM_USE_STMT to get access to each occurrence of ssavar on the stmt returned by diff --git a/gcc/testsuite/gcc.dg/ipa/propmalloc-1.c b/gcc/testsuite/gcc.dg/ipa/propmalloc-1.c new file mode 100644 index 00000000000..9a95f817079 --- /dev/null +++ b/gcc/testsuite/gcc.dg/ipa/propmalloc-1.c @@ -0,0 +1,21 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-ipa-pure-const-details" } */ + +__attribute__((noinline, no_icf, used)) +static void *f(__SIZE_TYPE__ n) +{ + void *p = __builtin_malloc (n); + if (p == 0) + __builtin_abort (); + return p; +} + +__attribute__((noinline, no_icf, used)) +static void *bar(__SIZE_TYPE__ n) +{ + void *p = f (n); + return p; +} + +/* { dg-final { scan-ipa-dump "Function f found to be malloc" "pure-const" } } */ +/* { dg-final { scan-ipa-dump "Function bar found to be malloc" "pure-const" } } */ diff --git a/gcc/testsuite/gcc.dg/ipa/propmalloc-2.c b/gcc/testsuite/gcc.dg/ipa/propmalloc-2.c new file mode 100644 index 00000000000..95b2fd74a7a --- /dev/null +++ b/gcc/testsuite/gcc.dg/ipa/propmalloc-2.c @@ -0,0 +1,23 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-ipa-pure-const-details" } */ + +__attribute__((noinline, used, no_icf)) +static void *foo (__SIZE_TYPE__ n) +{ + return __builtin_malloc (n * 10); +} + +__attribute__((noinline, used, no_icf)) +static void *bar(__SIZE_TYPE__ n, int cond) +{ + void *p; + if (cond) + p = foo (n); + else + p = __builtin_malloc (n); + + return p; +} + +/* { dg-final { scan-ipa-dump "Function foo found to be malloc" "pure-const" } } */ +/* { dg-final { scan-ipa-dump "Function bar found to be malloc" "pure-const" } } */ diff --git a/gcc/testsuite/gcc.dg/ipa/propmalloc-3.c b/gcc/testsuite/gcc.dg/ipa/propmalloc-3.c new file mode 100644 index 00000000000..13558ddd07d --- /dev/null +++ b/gcc/testsuite/gcc.dg/ipa/propmalloc-3.c @@ -0,0 +1,24 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-ipa-pure-const-details" } */ + +static void *foo(__SIZE_TYPE__, int) __attribute__((noinline, no_icf, used)); + +__attribute__((noinline, used, no_icf)) +static void *bar(__SIZE_TYPE__ n, int m) +{ + return foo (n, m); +} + +static void *foo(__SIZE_TYPE__ n, int m) +{ + void *p; + if (m > 0) + p = bar (n, --m); + else + p = __builtin_malloc (n); + + return p; +} + +/* { dg-final { scan-ipa-dump "Function foo found to be malloc" "pure-const" } } */ +/* { dg-final { scan-ipa-dump "Function bar found to be malloc" "pure-const" } } */