Successfully identified regression in *gcc* in CI configuration tcwg_gnu/gnu-master-aarch64-check_gcc. So far, this commit has regressed CI configurations: - tcwg_gnu/gnu-master-aarch64-check_gcc

Culprit: <cut> commit 13586172d0b70c9d7ca464fc5a2a46a8532b06d7 Author: Martin Jambor mjambor@suse.cz Date: Tue Jul 27 10:02:38 2021 +0200

ipa: Adjust references to identify read-only globals

this patch has been motivated by SPEC 2017's 544.nab_r in which there is a static variable which is never written to and so zero throughout the run-time of the benchmark. However, it is passed by reference to a function in which it is read and (after some multiplications) passed into __builtin_exp which in turn unnecessarily consumes almost 10% of the total benchmark run-time. The situation is illustrated by the added testcase remref-3.c.

The patch adds a flag to ipa-prop descriptor of each parameter to mark such parameters. IPA-CP and inling then take the effort to remove IPA_REF_ADDR references in the caller and only add IPA_REF_LOAD reference to the clone/overall inlined function. This is sufficient for subsequent symbol table analysis code to identify the read-only variable as such and optimize the code.

There are two changes from the RFC version posted to the list earlier. First, three missing calls to get_base_address were added (there was another one in an assert). Second, references are not stripped off the callers if the cloned function cannot change the signature. The second change reveals a real shortcoming stemming from the fact we cannot adjust function prototypes with fnspecs. But that is a more general problem.

gcc/ChangeLog:

2021-07-20 Martin Jambor mjambor@suse.cz

* cgraph.h (ipa_replace_map): New field force_load_ref. * ipa-prop.h (ipa_param_descriptor): Reduce precision of move_cost, aded new flag load_dereferenced, adjusted comments. (ipa_get_param_dereferenced): New function. (ipa_set_param_dereferenced): Likewise. * cgraphclones.c (cgraph_node::create_virtual_clone): Follow it. * ipa-cp.c: Include gimple.h. (ipcp_discover_new_direct_edges): Take into account dereferenced flag. (get_replacement_map): New parameter force_load_ref, set the appropriate flag in ipa_replace_map if set. (struct symbol_and_index_together): New type. (adjust_refs_in_act_callers): New function. (adjust_references_in_caller): Likewise. (create_specialized_node): When appropriate, call adjust_references_in_caller and force only load references. * ipa-prop.c (load_from_dereferenced_name): New function. (ipa_analyze_controlled_uses): Also detect loads from a dereference, harden testing of call statements. (ipa_write_node_info): Stream the dereferenced flag. (ipa_read_node_info): Likewise. (ipa_set_jf_constant): Also create refdesc when jump function references a variable. (cgraph_node_for_jfunc): Rename to symtab_node_for_jfunc, work also on references of variables and return a symtab_node. Adjust all callers. (propagate_controlled_uses): Also remove references to VAR_DECLs.

gcc/testsuite/ChangeLog:

2021-06-29 Martin Jambor mjambor@suse.cz

* gcc.dg/ipa/remref-3.c: New test. * gcc.dg/ipa/remref-4.c: Likewise. * gcc.dg/ipa/remref-5.c: Likewise. * gcc.dg/ipa/remref-6.c: Likewise. </cut>

Results regressed to (for first_bad == 13586172d0b70c9d7ca464fc5a2a46a8532b06d7) # reset_artifacts: -10 # build_abe binutils: -2 # build_abe gcc: -1 # build_abe dejagnu: 0 # build_abe check_gcc -- --set runtestflags=g++.dg/dg.exp --set runtestflags=g++.dg/vect/vect.exp --set runtestflags=gcc.dg/ipa/ipa.exp: 1 # Getting actual results from build directory /home/tcwg-buildslave/workspace/tcwg_gnu_3/artifacts/build-13586172d0b70c9d7ca464fc5a2a46a8532b06d7/sumfiles # /home/tcwg-buildslave/workspace/tcwg_gnu_3/artifacts/build-13586172d0b70c9d7ca464fc5a2a46a8532b06d7/sumfiles/libstdc++.sum # /home/tcwg-buildslave/workspace/tcwg_gnu_3/artifacts/build-13586172d0b70c9d7ca464fc5a2a46a8532b06d7/sumfiles/gfortran.sum # /home/tcwg-buildslave/workspace/tcwg_gnu_3/artifacts/build-13586172d0b70c9d7ca464fc5a2a46a8532b06d7/sumfiles/libitm.sum # /home/tcwg-buildslave/workspace/tcwg_gnu_3/artifacts/build-13586172d0b70c9d7ca464fc5a2a46a8532b06d7/sumfiles/libgomp.sum # /home/tcwg-buildslave/workspace/tcwg_gnu_3/artifacts/build-13586172d0b70c9d7ca464fc5a2a46a8532b06d7/sumfiles/libatomic.sum # /home/tcwg-buildslave/workspace/tcwg_gnu_3/artifacts/build-13586172d0b70c9d7ca464fc5a2a46a8532b06d7/sumfiles/g++.sum # /home/tcwg-buildslave/workspace/tcwg_gnu_3/artifacts/build-13586172d0b70c9d7ca464fc5a2a46a8532b06d7/sumfiles/gcc.sum # Manifest: gcc-compare-results/contrib/testsuite-management/flaky/gnu-master-aarch64-check_gcc.xfail # Getting actual results from build directory base-artifacts/sumfiles # base-artifacts/sumfiles/libstdc++.sum # base-artifacts/sumfiles/gfortran.sum # base-artifacts/sumfiles/libitm.sum # base-artifacts/sumfiles/libgomp.sum # base-artifacts/sumfiles/libatomic.sum # base-artifacts/sumfiles/g++.sum # base-artifacts/sumfiles/gcc.sum # # # Unexpected results in this build (new failures) # === g++ tests === # # Running g++.dg/vect/vect.exp ... # FAIL: g++.dg/vect/pr99149.cc -std=c++98 scan-tree-dump-times slp2 "stmt.*COMPLEX_MUL" 1 # FAIL: g++.dg/vect/pr99149.cc -std=c++98 scan-tree-dump-times slp2 "stmt.*COMPLEX_FMA" 1 # FAIL: g++.dg/vect/pr99149.cc -std=c++14 scan-tree-dump-times slp2 "stmt.*COMPLEX_MUL" 1 # FAIL: g++.dg/vect/pr99149.cc -std=c++14 scan-tree-dump-times slp2 "stmt.*COMPLEX_FMA" 1 # FAIL: g++.dg/vect/pr99149.cc -std=c++17 scan-tree-dump-times slp2 "stmt.*COMPLEX_MUL" 1 # FAIL: g++.dg/vect/pr99149.cc -std=c++17 scan-tree-dump-times slp2 "stmt.*COMPLEX_FMA" 1 # FAIL: g++.dg/vect/pr99149.cc -std=c++2a scan-tree-dump-times slp2 "stmt.*COMPLEX_MUL" 1 # FAIL: g++.dg/vect/pr99149.cc -std=c++2a scan-tree-dump-times slp2 "stmt.*COMPLEX_FMA" 1 # === gcc tests === # # Running g++.dg/dg.exp ... # FAIL: g++.dg/pr77550.C (test for excess errors) # UNRESOLVED: g++.dg/pr77550.C compilation failed to produce executable # # Running gcc.dg/ipa/ipa.exp ... # UNRESOLVED: gcc.dg/ipa/remref-6.c scan-ipa-dump inline "Removed a reference" # UNRESOLVED: gcc.dg/ipa/remref-6.c scan-ipa-dump inline "replaced it with LOAD" # # === Results Summary ===

from (for last_good == a21bd3cebd6f54af70a37c18b8fbeae933fb6515) # reset_artifacts: -10 # build_abe binutils: -2 # build_abe gcc: -1 # build_abe dejagnu: 0 # build_abe check_gcc -- --set runtestflags=g++.dg/dg.exp --set runtestflags=g++.dg/vect/vect.exp --set runtestflags=gcc.dg/ipa/ipa.exp: 1

Artifacts of last_good build: https://ci.linaro.org/job/tcwg_gcc-bisect-gnu-master-aarch64-check_gcc/1/art... Artifacts of first_bad build: https://ci.linaro.org/job/tcwg_gcc-bisect-gnu-master-aarch64-check_gcc/1/art... Build top page/logs: https://ci.linaro.org/job/tcwg_gcc-bisect-gnu-master-aarch64-check_gcc/1/

Configuration details:

Reproduce builds: <cut> mkdir investigate-gcc-13586172d0b70c9d7ca464fc5a2a46a8532b06d7 cd investigate-gcc-13586172d0b70c9d7ca464fc5a2a46a8532b06d7

git clone https://git.linaro.org/toolchain/jenkins-scripts

mkdir -p artifacts/manifests curl -o artifacts/manifests/build-baseline.sh https://ci.linaro.org/job/tcwg_gcc-bisect-gnu-master-aarch64-check_gcc/1/art... --fail curl -o artifacts/manifests/build-parameters.sh https://ci.linaro.org/job/tcwg_gcc-bisect-gnu-master-aarch64-check_gcc/1/art... --fail curl -o artifacts/test.sh https://ci.linaro.org/job/tcwg_gcc-bisect-gnu-master-aarch64-check_gcc/1/art... --fail chmod +x artifacts/test.sh

# Reproduce the baseline build (build all pre-requisites) ./jenkins-scripts/tcwg_gnu-build.sh @@ artifacts/manifests/build-baseline.sh

# Save baseline build state (which is then restored in artifacts/test.sh) mkdir -p ./bisect rsync -a --del --delete-excluded --exclude /bisect/ --exclude /artifacts/ --exclude /gcc/ ./ ./bisect/baseline/

cd gcc

# Reproduce first_bad build git checkout --detach 13586172d0b70c9d7ca464fc5a2a46a8532b06d7 ../artifacts/test.sh

# Reproduce last_good build git checkout --detach a21bd3cebd6f54af70a37c18b8fbeae933fb6515 ../artifacts/test.sh

cd .. </cut>

History of pending regressions and results: https://git.linaro.org/toolchain/ci/base-artifacts.git/log/?h=linaro-local/c...

Artifacts: https://ci.linaro.org/job/tcwg_gcc-bisect-gnu-master-aarch64-check_gcc/1/art... Build log: https://ci.linaro.org/job/tcwg_gcc-bisect-gnu-master-aarch64-check_gcc/1/con...

Full commit (up to 1000 lines): <cut> commit 13586172d0b70c9d7ca464fc5a2a46a8532b06d7 Author: Martin Jambor mjambor@suse.cz Date: Tue Jul 27 10:02:38 2021 +0200

ipa: Adjust references to identify read-only globals

this patch has been motivated by SPEC 2017's 544.nab_r in which there is a static variable which is never written to and so zero throughout the run-time of the benchmark. However, it is passed by reference to a function in which it is read and (after some multiplications) passed into __builtin_exp which in turn unnecessarily consumes almost 10% of the total benchmark run-time. The situation is illustrated by the added testcase remref-3.c.

The patch adds a flag to ipa-prop descriptor of each parameter to mark such parameters. IPA-CP and inling then take the effort to remove IPA_REF_ADDR references in the caller and only add IPA_REF_LOAD reference to the clone/overall inlined function. This is sufficient for subsequent symbol table analysis code to identify the read-only variable as such and optimize the code.

There are two changes from the RFC version posted to the list earlier. First, three missing calls to get_base_address were added (there was another one in an assert). Second, references are not stripped off the callers if the cloned function cannot change the signature. The second change reveals a real shortcoming stemming from the fact we cannot adjust function prototypes with fnspecs. But that is a more general problem.

gcc/ChangeLog:

2021-07-20 Martin Jambor mjambor@suse.cz

* cgraph.h (ipa_replace_map): New field force_load_ref. * ipa-prop.h (ipa_param_descriptor): Reduce precision of move_cost, aded new flag load_dereferenced, adjusted comments. (ipa_get_param_dereferenced): New function. (ipa_set_param_dereferenced): Likewise. * cgraphclones.c (cgraph_node::create_virtual_clone): Follow it. * ipa-cp.c: Include gimple.h. (ipcp_discover_new_direct_edges): Take into account dereferenced flag. (get_replacement_map): New parameter force_load_ref, set the appropriate flag in ipa_replace_map if set. (struct symbol_and_index_together): New type. (adjust_refs_in_act_callers): New function. (adjust_references_in_caller): Likewise. (create_specialized_node): When appropriate, call adjust_references_in_caller and force only load references. * ipa-prop.c (load_from_dereferenced_name): New function. (ipa_analyze_controlled_uses): Also detect loads from a dereference, harden testing of call statements. (ipa_write_node_info): Stream the dereferenced flag. (ipa_read_node_info): Likewise. (ipa_set_jf_constant): Also create refdesc when jump function references a variable. (cgraph_node_for_jfunc): Rename to symtab_node_for_jfunc, work also on references of variables and return a symtab_node. Adjust all callers. (propagate_controlled_uses): Also remove references to VAR_DECLs.

gcc/testsuite/ChangeLog:

2021-06-29 Martin Jambor mjambor@suse.cz

* gcc.dg/ipa/remref-3.c: New test. * gcc.dg/ipa/remref-4.c: Likewise. * gcc.dg/ipa/remref-5.c: Likewise. * gcc.dg/ipa/remref-6.c: Likewise. --- gcc/cgraph.h | 4 + gcc/cgraphclones.c | 10 ++- gcc/ipa-cp.c | 156 ++++++++++++++++++++++++++++++--- gcc/ipa-prop.c | 166 ++++++++++++++++++++++++++++-------- gcc/ipa-prop.h | 29 ++++++- gcc/testsuite/gcc.dg/ipa/remref-3.c | 23 +++++ gcc/testsuite/gcc.dg/ipa/remref-4.c | 31 +++++++ gcc/testsuite/gcc.dg/ipa/remref-5.c | 38 +++++++++ gcc/testsuite/gcc.dg/ipa/remref-6.c | 24 ++++++ 9 files changed, 431 insertions(+), 50 deletions(-)

diff --git a/gcc/cgraph.h b/gcc/cgraph.h index 8c776d6f3a8..d54a258c2ee 100644 --- a/gcc/cgraph.h +++ b/gcc/cgraph.h @@ -700,6 +700,10 @@ struct GTY(()) ipa_replace_map tree new_tree; /* Parameter number to replace, when old_tree is NULL. */ int parm_num; + /* Set if the newly added reference should not be an address one, but a load + one from the operand of the ADDR_EXPR in NEW_TREE. This is for cases when + the corresponding parameter p is used only as *p. */ + unsigned force_load_ref : 1; };

enum cgraph_simd_clone_arg_type diff --git a/gcc/cgraphclones.c b/gcc/cgraphclones.c index fef4a231f85..b16e68194e1 100644 --- a/gcc/cgraphclones.c +++ b/gcc/cgraphclones.c @@ -633,7 +633,15 @@ cgraph_node::create_virtual_clone (const vec<cgraph_edge *> &redirect_callers, || in_lto_p) new_node->unique_name = true; FOR_EACH_VEC_SAFE_ELT (tree_map, i, map) - new_node->maybe_create_reference (map->new_tree, NULL); + { + tree repl = map->new_tree; + if (map->force_load_ref) + { + gcc_assert (TREE_CODE (repl) == ADDR_EXPR); + repl = get_base_address (TREE_OPERAND (repl, 0)); + } + new_node->maybe_create_reference (repl, NULL); + }

if (ipa_transforms_to_apply.exists ()) new_node->ipa_transforms_to_apply diff --git a/gcc/ipa-cp.c b/gcc/ipa-cp.c index ce28ada19fe..55b9216337f 100644 --- a/gcc/ipa-cp.c +++ b/gcc/ipa-cp.c @@ -106,6 +106,7 @@ along with GCC; see the file COPYING3. If not see #include "backend.h" #include "tree.h" #include "gimple-expr.h" +#include "gimple.h" #include "predict.h" #include "alloc-pool.h" #include "tree-pass.h" @@ -4008,7 +4009,8 @@ ipcp_discover_new_direct_edges (struct cgraph_node *node, { ipa_node_params *info = ipa_node_params_sum->get (node); int c = ipa_get_controlled_uses (info, param_index); - if (c != IPA_UNDESCRIBED_USE) + if (c != IPA_UNDESCRIBED_USE + && !ipa_get_param_load_dereferenced (info, param_index)) { struct ipa_ref *to_del;

@@ -4317,10 +4319,14 @@ gather_edges_for_value (ipcp_value<valtype> *val, cgraph_node *dest, }

/* Construct a replacement map for a know VALUE for a formal parameter PARAM. - Return it or NULL if for some reason it cannot be created. */ + Return it or NULL if for some reason it cannot be created. FORCE_LOAD_REF + should be set to true when the reference created for the constant should be + a load one and not an address one because the corresponding parameter p is + only used as *p. */

static struct ipa_replace_map * -get_replacement_map (class ipa_node_params *info, tree value, int parm_num) +get_replacement_map (class ipa_node_params *info, tree value, int parm_num, + bool force_load_ref) { struct ipa_replace_map *replace_map;

@@ -4332,10 +4338,15 @@ get_replacement_map (class ipa_node_params *info, tree value, int parm_num)

fprintf (dump_file, " with const "); print_generic_expr (dump_file, value); - fprintf (dump_file, "\n"); + + if (force_load_ref) + fprintf (dump_file, " - forcing load reference\n"); + else + fprintf (dump_file, "\n"); } replace_map->parm_num = parm_num; replace_map->new_tree = value; + replace_map->force_load_ref = force_load_ref; return replace_map; }

@@ -4488,6 +4499,113 @@ update_specialized_profile (struct cgraph_node *new_node, dump_profile_updates (orig_node, new_node); }

+static void adjust_references_in_caller (cgraph_edge *cs, + symtab_node *symbol, int index); + +/* Simple structure to pass a symbol and index (with same meaning as parameters + of adjust_references_in_caller) through a void* parameter of a + call_for_symbol_thunks_and_aliases callback. */ +struct symbol_and_index_together +{ + symtab_node *symbol; + int index; +}; + +/* Worker callback of call_for_symbol_thunks_and_aliases to recursively call + adjust_references_in_caller on edges up in the call-graph, if necessary. */ +static bool +adjust_refs_in_act_callers (struct cgraph_node *node, void *data) +{ + symbol_and_index_together *pack = (symbol_and_index_together *) data; + for (cgraph_edge *cs = node->callers; cs; cs = cs->next_caller) + if (!cs->caller->thunk) + adjust_references_in_caller (cs, pack->symbol, pack->index); + return false; +} + +/* At INDEX of a function being called by CS there is an ADDR_EXPR of a + variable which is only dereferenced and which is represented by SYMBOL. See + if we can remove ADDR reference in callers assosiated witht the call. */ + +static void +adjust_references_in_caller (cgraph_edge *cs, symtab_node *symbol, int index) +{ + ipa_edge_args *args = ipa_edge_args_sum->get (cs); + ipa_jump_func *jfunc = ipa_get_ith_jump_func (args, index); + if (jfunc->type == IPA_JF_CONST) + { + ipa_ref *to_del = cs->caller->find_reference (symbol, cs->call_stmt, + cs->lto_stmt_uid); + if (!to_del) + return; + to_del->remove_reference (); + if (dump_file) + fprintf (dump_file, " Removed a reference from %s to %s.\n", + cs->caller->dump_name (), symbol->dump_name ()); + return; + } + + if (jfunc->type != IPA_JF_PASS_THROUGH + || ipa_get_jf_pass_through_operation (jfunc) != NOP_EXPR) + return; + + int fidx = ipa_get_jf_pass_through_formal_id (jfunc); + cgraph_node *caller = cs->caller; + ipa_node_params *caller_info = ipa_node_params_sum->get (caller); + /* TODO: This consistency check may be too big and not really + that useful. Consider removing it. */ + tree cst; + if (caller_info->ipcp_orig_node) + cst = caller_info->known_csts[fidx]; + else + { + ipcp_lattice<tree> *lat = ipa_get_scalar_lat (caller_info, fidx); + gcc_assert (lat->is_single_const ()); + cst = lat->values->value; + } + gcc_assert (TREE_CODE (cst) == ADDR_EXPR + && (symtab_node::get (get_base_address (TREE_OPERAND (cst, 0))) + == symbol)); + + int cuses = ipa_get_controlled_uses (caller_info, fidx); + if (cuses == IPA_UNDESCRIBED_USE) + return; + gcc_assert (cuses > 0); + cuses--; + ipa_set_controlled_uses (caller_info, fidx, cuses); + if (cuses) + return; + + if (caller_info->ipcp_orig_node) + { + /* Cloning machinery has created a reference here, we need to either + remove it or change it to a read one. */ + ipa_ref *to_del = caller->find_reference (symbol, NULL, 0); + if (to_del && to_del->use == IPA_REF_ADDR) + { + to_del->remove_reference (); + if (dump_file) + fprintf (dump_file, " Removed a reference from %s to %s.\n", + cs->caller->dump_name (), symbol->dump_name ()); + if (ipa_get_param_load_dereferenced (caller_info, fidx)) + { + caller->create_reference (symbol, IPA_REF_LOAD, NULL); + if (dump_file) + fprintf (dump_file, + " ...and replaced it with LOAD one.\n"); + } + } + } + + symbol_and_index_together pack; + pack.symbol = symbol; + pack.index = fidx; + if (caller->can_change_signature) + caller->call_for_symbol_thunks_and_aliases (adjust_refs_in_act_callers, + &pack, true); +} + + /* Return true if we would like to remove a parameter from NODE when cloning it with KNOWN_CSTS scalar constants. */

@@ -4595,15 +4713,31 @@ create_specialized_node (struct cgraph_node *node, for (i = 0; i < count; i++) { tree t = known_csts[i]; - if (t) - { - struct ipa_replace_map *replace_map; + if (!t) + continue;

- gcc_checking_assert (TREE_CODE (t) != TREE_BINFO); - replace_map = get_replacement_map (info, t, i); - if (replace_map) - vec_safe_push (replace_trees, replace_map); + gcc_checking_assert (TREE_CODE (t) != TREE_BINFO); + + bool load_ref = false; + symtab_node *ref_symbol; + if (TREE_CODE (t) == ADDR_EXPR) + { + tree base = get_base_address (TREE_OPERAND (t, 0)); + if (TREE_CODE (base) == VAR_DECL + && ipa_get_controlled_uses (info, i) == 0 + && ipa_get_param_load_dereferenced (info, i) + && (ref_symbol = symtab_node::get (base))) + { + load_ref = true; + if (node->can_change_signature) + for (cgraph_edge *caller : callers) + adjust_references_in_caller (caller, ref_symbol, i); + } } + + ipa_replace_map *replace_map = get_replacement_map (info, t, i, load_ref); + if (replace_map) + vec_safe_push (replace_trees, replace_map); } auto_vec<cgraph_edge *, 2> self_recursive_calls; for (i = callers.length () - 1; i >= 0; i--) diff --git a/gcc/ipa-prop.c b/gcc/ipa-prop.c index 43f46a578c6..0afd05eb95c 100644 --- a/gcc/ipa-prop.c +++ b/gcc/ipa-prop.c @@ -544,7 +544,9 @@ ipa_set_jf_constant (struct ipa_jump_func *jfunc, tree constant, jfunc->value.constant.value = unshare_expr_without_location (constant);

if (TREE_CODE (constant) == ADDR_EXPR - && TREE_CODE (TREE_OPERAND (constant, 0)) == FUNCTION_DECL) + && (TREE_CODE (TREE_OPERAND (constant, 0)) == FUNCTION_DECL + || (TREE_CODE (TREE_OPERAND (constant, 0)) == VAR_DECL + && TREE_STATIC (TREE_OPERAND (constant, 0))))) { struct ipa_cst_ref_desc *rdesc;

@@ -2876,6 +2878,16 @@ ipa_analyze_params_uses_in_bb (struct ipa_func_body_info *fbi, basic_block bb) visit_ref_for_mod_analysis); }

+/* Return true EXPR is a load from a dereference of SSA_NAME NAME. */ + +static bool +load_from_dereferenced_name (tree expr, tree name) +{ + tree base = get_base_address (expr); + return (TREE_CODE (base) == MEM_REF + && TREE_OPERAND (base, 0) == name); +} + /* Calculate controlled uses of parameters of NODE. */

static void @@ -2886,7 +2898,8 @@ ipa_analyze_controlled_uses (struct cgraph_node *node) for (int i = 0; i < ipa_get_param_count (info); i++) { tree parm = ipa_get_param (info, i); - int controlled_uses = 0; + int call_uses = 0; + bool load_dereferenced = false;

/* For SSA regs see if parameter is used. For non-SSA we compute the flag during modification analysis. */ @@ -2897,27 +2910,77 @@ ipa_analyze_controlled_uses (struct cgraph_node *node) if (ddef && !has_zero_uses (ddef)) { imm_use_iterator imm_iter; - use_operand_p use_p; + gimple *stmt;

ipa_set_param_used (info, i, true); - FOR_EACH_IMM_USE_FAST (use_p, imm_iter, ddef) - if (!is_gimple_call (USE_STMT (use_p))) - { - if (!is_gimple_debug (USE_STMT (use_p))) - { - controlled_uses = IPA_UNDESCRIBED_USE; - break; - } - } - else - controlled_uses++; + FOR_EACH_IMM_USE_STMT (stmt, imm_iter, ddef) + { + if (is_gimple_debug (stmt)) + continue; + + int all_stmt_uses = 0; + use_operand_p use_p; + FOR_EACH_IMM_USE_ON_STMT (use_p, imm_iter) + all_stmt_uses++; + + if (is_gimple_call (stmt)) + { + if (gimple_call_internal_p (stmt)) + { + call_uses = IPA_UNDESCRIBED_USE; + break; + } + int recognized_stmt_uses; + if (gimple_call_fn (stmt) == ddef) + recognized_stmt_uses = 1; + else + recognized_stmt_uses = 0; + unsigned arg_count = gimple_call_num_args (stmt); + for (unsigned i = 0; i < arg_count; i++) + { + tree arg = gimple_call_arg (stmt, i); + if (arg == ddef) + recognized_stmt_uses++; + else if (load_from_dereferenced_name (arg, ddef)) + { + load_dereferenced = true; + recognized_stmt_uses++; + } + } + + if (recognized_stmt_uses != all_stmt_uses) + { + call_uses = IPA_UNDESCRIBED_USE; + break; + } + if (call_uses >= 0) + call_uses += all_stmt_uses; + } + else if (gimple_assign_single_p (stmt)) + { + tree rhs = gimple_assign_rhs1 (stmt); + if (all_stmt_uses != 1 + || !load_from_dereferenced_name (rhs, ddef)) + { + call_uses = IPA_UNDESCRIBED_USE; + break; + } + load_dereferenced = true; + } + else + { + call_uses = IPA_UNDESCRIBED_USE; + break; + } + } } else - controlled_uses = 0; + call_uses = 0; } else - controlled_uses = IPA_UNDESCRIBED_USE; - ipa_set_controlled_uses (info, i, controlled_uses); + call_uses = IPA_UNDESCRIBED_USE; + ipa_set_controlled_uses (info, i, call_uses); + ipa_set_param_load_dereferenced (info, i, load_dereferenced); } }

@@ -3640,16 +3703,17 @@ jfunc_rdesc_usable (struct ipa_jump_func *jfunc) declaration, return the associated call graph node. Otherwise return NULL. */

-static cgraph_node * -cgraph_node_for_jfunc (struct ipa_jump_func *jfunc) +static symtab_node * +symtab_node_for_jfunc (struct ipa_jump_func *jfunc) { gcc_checking_assert (jfunc->type == IPA_JF_CONST); tree cst = ipa_get_jf_constant (jfunc); if (TREE_CODE (cst) != ADDR_EXPR - || TREE_CODE (TREE_OPERAND (cst, 0)) != FUNCTION_DECL) + || (TREE_CODE (TREE_OPERAND (cst, 0)) != FUNCTION_DECL + && TREE_CODE (TREE_OPERAND (cst, 0)) != VAR_DECL)) return NULL;

- return cgraph_node::get (TREE_OPERAND (cst, 0)); + return symtab_node::get (TREE_OPERAND (cst, 0)); }

@@ -3666,7 +3730,7 @@ try_decrement_rdesc_refcount (struct ipa_jump_func *jfunc) && (rdesc = jfunc_rdesc_usable (jfunc)) && --rdesc->refcount == 0) { - symtab_node *symbol = cgraph_node_for_jfunc (jfunc); + symtab_node *symbol = symtab_node_for_jfunc (jfunc); if (!symbol) return false;

@@ -3721,8 +3785,8 @@ try_make_edge_direct_simple_call (struct cgraph_edge *ie, gcc_checking_assert (cs->callee && (cs != ie || jfunc->type != IPA_JF_CONST - || !cgraph_node_for_jfunc (jfunc) - || cs->callee == cgraph_node_for_jfunc (jfunc))); + || !symtab_node_for_jfunc (jfunc) + || cs->callee == symtab_node_for_jfunc (jfunc))); ok = try_decrement_rdesc_refcount (jfunc); gcc_checking_assert (ok); } @@ -4085,7 +4149,15 @@ propagate_controlled_uses (struct cgraph_edge *cs) == NOP_EXPR || c == IPA_UNDESCRIBED_USE); c = combine_controlled_uses_counters (c, d); ipa_set_controlled_uses (new_root_info, src_idx, c); - if (c == 0 && new_root_info->ipcp_orig_node) + bool lderef = true; + if (c != IPA_UNDESCRIBED_USE) + { + lderef = (ipa_get_param_load_dereferenced (new_root_info, src_idx) + || ipa_get_param_load_dereferenced (old_root_info, i)); + ipa_set_param_load_dereferenced (new_root_info, src_idx, lderef); + } + + if (c == 0 && !lderef && new_root_info->ipcp_orig_node) { struct cgraph_node *n; struct ipa_ref *ref; @@ -4114,17 +4186,27 @@ propagate_controlled_uses (struct cgraph_edge *cs) if (rdesc->refcount == 0) { tree cst = ipa_get_jf_constant (jf); - struct cgraph_node *n; gcc_checking_assert (TREE_CODE (cst) == ADDR_EXPR - && TREE_CODE (TREE_OPERAND (cst, 0)) - == FUNCTION_DECL); - n = cgraph_node::get (TREE_OPERAND (cst, 0)); + && ((TREE_CODE (TREE_OPERAND (cst, 0)) + == FUNCTION_DECL) + || (TREE_CODE (TREE_OPERAND (cst, 0)) + == VAR_DECL))); + + symtab_node *n = symtab_node::get (TREE_OPERAND (cst, 0)); if (n) { struct cgraph_node *clone; - bool ok; - ok = remove_described_reference (n, rdesc); - gcc_checking_assert (ok); + bool removed = remove_described_reference (n, rdesc); + /* The reference might have been removed by IPA-CP. */ + if (removed + && ipa_get_param_load_dereferenced (old_root_info, i)) + { + new_root->create_reference (n, IPA_REF_LOAD, NULL); + if (dump_file) + fprintf (dump_file, "ipa-prop: ...replaced it with " + " LOAD one from %s to %s.\n", + new_root->dump_name (), n->dump_name ()); + }

clone = cs->caller; while (clone->inlined_to @@ -4347,7 +4429,7 @@ ipa_edge_args_sum_t::duplicate (cgraph_edge *src, cgraph_edge *dst, else if (src->caller == dst->caller) { struct ipa_ref *ref; - symtab_node *n = cgraph_node_for_jfunc (src_jf); + symtab_node *n = symtab_node_for_jfunc (src_jf); gcc_checking_assert (n); ref = src->caller->find_reference (n, src->call_stmt, src->lto_stmt_uid); @@ -4574,7 +4656,9 @@ ipa_print_node_params (FILE *f, struct cgraph_node *node) if (c == IPA_UNDESCRIBED_USE) fprintf (f, " undescribed_use"); else - fprintf (f, " controlled_uses=%i", c); + fprintf (f, " controlled_uses=%i %s", c, + ipa_get_param_load_dereferenced (info, i) + ? "(load_dereferenced)" : ""); fprintf (f, "\n"); } } @@ -4969,7 +5053,13 @@ ipa_write_node_info (struct output_block *ob, struct cgraph_node *node) gcc_assert (!info->node_enqueued); gcc_assert (!info->ipcp_orig_node); for (j = 0; j < ipa_get_param_count (info); j++) - bp_pack_value (&bp, ipa_is_param_used (info, j), 1); + { + /* TODO: We could just not stream the bit in the undescribed case. */ + bool d = (ipa_get_controlled_uses (info, j) != IPA_UNDESCRIBED_USE) + ? ipa_get_param_load_dereferenced (info, j) : true; + bp_pack_value (&bp, d, 1); + bp_pack_value (&bp, ipa_is_param_used (info, j), 1); + } streamer_write_bitpack (&bp); for (j = 0; j < ipa_get_param_count (info); j++) { @@ -5095,10 +5185,14 @@ ipa_read_node_info (class lto_input_block *ib, struct cgraph_node *node, bp = streamer_read_bitpack (ib); for (k = 0; k < param_count; k++) { + bool load_dereferenced = bp_unpack_value (&bp, 1); bool used = bp_unpack_value (&bp, 1);

if (prevails) - ipa_set_param_used (info, k, used); + { + ipa_set_param_load_dereferenced (info, k, load_dereferenced); + ipa_set_param_used (info, k, used); + } } for (k = 0; k < param_count; k++) { diff --git a/gcc/ipa-prop.h b/gcc/ipa-prop.h index 2fe220acf84..19751f10974 100644 --- a/gcc/ipa-prop.h +++ b/gcc/ipa-prop.h @@ -550,14 +550,21 @@ struct GTY(()) ipa_param_descriptor tree decl_or_type; /* If all uses of the parameter are described by ipa-prop structures, this says how many there are. If any use could not be described by means of - ipa-prop structures, this is IPA_UNDESCRIBED_USE. */ + ipa-prop structures (which include flag dereferenced below), this is + IPA_UNDESCRIBED_USE. */ int controlled_uses; - unsigned int move_cost : 28; + unsigned int move_cost : 27; /* The parameter is used. */ unsigned used : 1; unsigned used_by_ipa_predicates : 1; unsigned used_by_indirect_call : 1; unsigned used_by_polymorphic_call : 1; + /* Set to true when in addition to being used in call statements, the + parameter has also been used for loads (but not for writes, does not + escape, etc.). This allows us to identify parameters p which are only + used as *p, and so when we propagate a constant to them, we can generate a + LOAD and not ADDR reference to them. */ + unsigned load_dereferenced : 1; };

/* ipa_node_params stores information related to formal parameters of functions @@ -795,6 +802,24 @@ ipa_set_controlled_uses (class ipa_node_params *info, int i, int val) (*info->descriptors)[i].controlled_uses = val; }

+/* Assuming a parameter does not have IPA_UNDESCRIBED_USE controlled uses, + return flag which indicates it has been dereferenced but only in a load. */ +static inline int +ipa_get_param_load_dereferenced (class ipa_node_params *info, int i) +{ + gcc_assert (ipa_get_controlled_uses (info, i) != IPA_UNDESCRIBED_USE); + return (*info->descriptors)[i].load_dereferenced; +} + +/* Set the load_dereferenced flag of a given parameter. */ + +static inline void +ipa_set_param_load_dereferenced (class ipa_node_params *info, int i, bool val) +{ + gcc_checking_assert (info->descriptors); + (*info->descriptors)[i].load_dereferenced = val; +} + /* Return the used flag corresponding to the Ith formal parameter of the function associated with INFO. */

diff --git a/gcc/testsuite/gcc.dg/ipa/remref-3.c b/gcc/testsuite/gcc.dg/ipa/remref-3.c new file mode 100644 index 00000000000..8a82ca65c68 --- /dev/null +++ b/gcc/testsuite/gcc.dg/ipa/remref-3.c @@ -0,0 +1,23 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fno-early-inlining -fdump-ipa-cp-details -fdump-tree-optimized" } */ + +static double global = 0.0; + +double foo_temp5; + +static void foo(double *ptr) { + static double abcd; + double v, exp_res; + v = *ptr; + exp_res = __builtin_exp(v); + foo_temp5 = exp_res * abcd; + abcd += foo_temp5; +} + +void entry() +{ + foo(&global); +} + +/* { dg-final { scan-ipa-dump "Removed a reference" "cp" } } */ +/* { dg-final { scan-tree-dump-not "builtin_exp" "optimized" } } */ diff --git a/gcc/testsuite/gcc.dg/ipa/remref-4.c b/gcc/testsuite/gcc.dg/ipa/remref-4.c new file mode 100644 index 00000000000..5a55437e971 --- /dev/null +++ b/gcc/testsuite/gcc.dg/ipa/remref-4.c @@ -0,0 +1,31 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fno-early-inlining -fdump-ipa-cp-details -fdump-tree-optimized" } */ + +static double global = 0.0; +double foo_temp5; + +static void foo(double *ptr) { + static double abcd; + double v, exp_res; + v = *ptr; + exp_res = __builtin_exp(v); + foo_temp5 = exp_res * abcd; + abcd += foo_temp5; +} + +double last_value; + +static void bar(double *ptr) +{ + last_value = *ptr; + foo (ptr); +} + +void entry() +{ + bar (&global); +} + +/* { dg-final { scan-ipa-dump "Removed a reference" "cp" } } */ +/* { dg-final { scan-ipa-dump "replaced it with LOAD" "cp" } } */ +/* { dg-final { scan-tree-dump-not "builtin_exp" "optimized" } } */ diff --git a/gcc/testsuite/gcc.dg/ipa/remref-5.c b/gcc/testsuite/gcc.dg/ipa/remref-5.c new file mode 100644 index 00000000000..c520e34d809 --- /dev/null +++ b/gcc/testsuite/gcc.dg/ipa/remref-5.c @@ -0,0 +1,38 @@ +/* { dg-do compile } */ +/* { dg-options "-O3 -fno-early-inlining -fdump-ipa-cp-details -fdump-tree-optimized" } */ + +static double global = 0.0; +double foo_temp5; + +static void foo(double *ptr) { + static double abcd; + double v, exp_res; + v = *ptr; + exp_res = __builtin_exp(v); + foo_temp5 = exp_res * abcd; + abcd += foo_temp5; +} + +double last_value; + +static void bar(double *ptr) +{ + last_value = *ptr; + for (unsigned i = 0; i < 200; i++) + foo (ptr); +} + +void entry() +{ + bar (&global); +} + +void decoy(double *ptr) +{ + bar (ptr); +} + + +/* { dg-final { scan-ipa-dump "Removed a reference" "cp" } } */ +/* { dg-final { scan-ipa-dump "replaced it with LOAD" "cp" } } */ +/* { dg-final { scan-tree-dump-times "builtin_exp" 1 "optimized" } } */ diff --git a/gcc/testsuite/gcc.dg/ipa/remref-6.c b/gcc/testsuite/gcc.dg/ipa/remref-6.c new file mode 100644 index 00000000000..de3649315c2 --- /dev/null +++ b/gcc/testsuite/gcc.dg/ipa/remref-6.c @@ -0,0 +1,24 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fno-early-inlining -fno-ipa-cp -fdump-tree-optimized" } */ + +static double global = 0.0; + +double foo_temp5; + +static void foo(double *ptr) { + static double abcd; + double v, exp_res; + v = *ptr; + exp_res = __builtin_exp(v); + foo_temp5 = exp_res * abcd; + abcd += foo_temp5; +} + +void entry() +{ + foo(&global); +} + +/* { dg-final { scan-ipa-dump "Removed a reference" "inline" } } */ +/* { dg-final { scan-ipa-dump "replaced it with LOAD" "inline" } } */ +/* { dg-final { scan-tree-dump-not "builtin_exp" "optimized" } } */ </cut>