Introduces the nodes used to model connectivity in the escape graph
and related state: a node's escape level and an encoding that will
be added to import and export data.
Uros Bizjak [Fri, 6 May 2016 21:14:20 +0000 (23:14 +0200)]
i386.md (LEAMODE): New mode attribute.
* config/i386/i386.md (LEAMODE): New mode attribute.
(plus to LEA splitter): Rewrite splitter using LEAMODE mode attribute.
(ashift to LEA splitter): Rewrte splitter using SWI mode iterator
and LEAMODE mode attribute. Use VOIDmode const_0_to_3_operand as
operand 2 predicate.
(*lea<mode>_general_2): Use VOIDmode for const248_operand.
(*lea<mode>_general_3): Ditto.
(*lea<mode>_general_4): Use VOIDmode for const_0_to_3_operand.
Chris Manghane [Fri, 6 May 2016 17:37:55 +0000 (17:37 +0000)]
escape: Add skeleton for gc analysis.
Introduces a skeleton replacement escape analysis
which contains four different phases extracted from the escape
analysis implementation in gc/esc.go. Also introduces the
Escape_context each phase uses to make decisions.
Jakub Jelinek [Fri, 6 May 2016 15:23:56 +0000 (17:23 +0200)]
re PR target/70941 (Test miscompiled with -O2.)
PR middle-end/70941
* gcc.dg/torture/pr70941.c (abort): Remove prototype.
(a, b, c, d): Change type from char to signed char.
(main): Compare against (signed char) -1634678893 instead of
hardcoded -109. Use __builtin_abort instead of abort.
David Malcolm [Fri, 6 May 2016 15:18:59 +0000 (15:18 +0000)]
Move name_to_pass_map into class pass_manager
gcc/ChangeLog:
* pass_manager.h (pass_manager::register_pass_name): New method.
(pass_manager::get_pass_by_name): New method.
(pass_manager::create_pass_tab): New method.
(pass_manager::m_name_to_pass_map): New field.
* passes.c (name_to_pass_map): Delete global in favor of field
"m_name_to_pass_map" of pass_manager.
(register_pass_name): Rename from a function to...
(pass_manager::register_pass_name): ...this method, updating
for renaming of global "name_to_pass_map" to field
"m_name_to_pass_map".
(create_pass_tab): Rename from a function to...
(pass_manager::create_pass_tab): ...this method, updating
for renaming of global "name_to_pass_map" to field.
(get_pass_by_name): Rename from a function to...
(pass_manager::get_pass_by_name): ...this method.
(enable_disable_pass): Convert use of get_pass_by_name to
a method call, locating the pass_manager singleton.
Jakub Jelinek [Fri, 6 May 2016 13:13:09 +0000 (15:13 +0200)]
sse.md (*vec_extractv4sf_0, [...]): Use v instead of x in vex or maybe_vex alternatives...
* config/i386/sse.md (*vec_extractv4sf_0, *sse4_1_extractps,
*vec_extractv4sf_mem, vec_extract_lo_v16hi, vec_extract_hi_v16hi,
vec_extract_lo_v32qi, vec_extract_hi_v32qi): Use v instead of x
in vex or maybe_vex alternatives, use maybe_evex instead of vex
in prefix.
Jakub Jelinek [Fri, 6 May 2016 13:12:32 +0000 (15:12 +0200)]
sse.md (*vec_concatv2sf_sse4_1, [...]): Use v instead of x in vex or maybe_vex alternatives...
* config/i386/sse.md (*vec_concatv2sf_sse4_1, *vec_concatv4sf): Use
v instead of x in vex or maybe_vex alternatives, use
maybe_evex instead of vex in prefix.
Jakub Jelinek [Fri, 6 May 2016 13:11:56 +0000 (15:11 +0200)]
sse.md (sse_shufps_<mode>, [...]): Use v instead of x in vex or maybe_vex alternatives...
* config/i386/sse.md (sse_shufps_<mode>, sse_storehps, sse_loadhps,
sse_storelps, sse_movss, avx2_vec_dup<mode>, avx2_vec_dupv8sf_1,
sse2_shufpd_<mode>, sse2_storehpd, sse2_storelpd, sse2_loadhpd,
sse2_loadlpd, sse2_movsd): Use v instead of x in vex or maybe_vex
alternatives, use maybe_evex instead of vex in prefix.
Richard Biener [Fri, 6 May 2016 12:53:26 +0000 (12:53 +0000)]
re PR tree-optimization/70948 (r235622 caused gcc.c-torture/execute/va-arg-pack-1.c execution failure AArch64)
2016-05-06 Richard Biener <rguenther@suse.de>
PR tree-optimization/70948
* tree-ssa-structalias.c (find_func_aliases_for_builtin_call):
Properly clobber all fields of va_list for __builtin_va_start.
Uros Bizjak [Thu, 5 May 2016 22:48:29 +0000 (00:48 +0200)]
re PR target/70873 ([7 Regressio] 20% performance regression at 482.sphinx3 after r235442 with -O2 -m32 on Haswell.)
PR target/70873
* config/i386/i386-protos.h (ix86_standard_x87sse_constant_load_p):
New prototype.
* config/i386/i386.c (ix86_standard_x87sse_constant_load_p): New.
* config/i386/i386.md (push mem splitter): Use find_constant_src in
the splitter condition.
(FP load splitter): Use ix86_standard_x87sse_constant_load_p in
the splitter condition.
(FP float_extend load splitter): Ditto.
Uros Bizjak [Thu, 5 May 2016 20:33:42 +0000 (22:33 +0200)]
i386.md (peehole2 patterns): Change true_regnum to REGNUM in all peephole2 patterns.
* config/i386/i386.md (peehole2 patterns): Change true_regnum
to REGNUM in all peephole2 patterns.
(post-reload splitters): Change true_regnum to REGNUM in
post-reload splitters.
(zero_extend splitters): Use general_reg_operand and
nonimmediate_gr_operand predicates.
* openmp.c (gfc_match_omp_clauses): Restructuralize, so that clause
parsing is done in a big switch based on gfc_peek_ascii_char and
individual clauses under their first letters are sorted too.
Jakub Jelinek [Thu, 5 May 2016 13:26:59 +0000 (15:26 +0200)]
c-parser.c (c_parser_switch_statement): Add IF_P argument, parse it through to c_parser_c99_block_statement.
* c-parser.c (c_parser_switch_statement): Add IF_P argument,
parse it through to c_parser_c99_block_statement.
(c_parser_statement_after_labels): Adjust c_parser_switch_statement
caller.
* parser.c (cp_parser_selection_statement): For RID_SWITCH,
pass if_p instead of NULL to cp_parser_implicitly_scoped_statement.
Alan Modra [Thu, 5 May 2016 00:07:27 +0000 (09:37 +0930)]
[RS6000] TARGET_RELOCATABLE
For ABI_V4, -mrelocatable and -fPIC both generate position independent
code, with some extra "fixup" output for -mrelocatable. The
similarity of these two options has led to the situation where the
sysv4.h SUBTARGET_OVERRIDE_OPTIONS sets flag_pic on seeing
-mrelocatable, and sets TARGET_RELOCATABLE on seeing -fPIC. That
prevents LTO from properly optimizing position dependent executables,
because the mutual dependence of the flags and the fact that LTO
streaming records the state of rs6000_isa_flags, result in flag_pic
being set when it shouldn't be.
So, don't set TARGET_RELOCATABLE when -fPIC. Places that currently
test TARGET_RELOCATABLE can instead test
TARGET_RELOCATABLE || (DEFAULT_ABI == ABI_V4 && flag_pic > 1)
or since TARGET_RELOCATABLE can only be enabled when ABI_V4,
DEFAULT_ABI == ABI_V4 && (TARGET_RELOCATABLE || flag_pic > 1).
Also, since flag_pic is set by -mrelocatable, a number of places that
currently test TARGET_RELOCATABLE can be simplified. I also made
-mrelocatable set TARGET_NO_FP_IN_TOC, allowing TARGET_RELOCATABLE to
be removed from ASM_OUTPUT_SPECIAL_POOL_ENTRY_P. Reducing occurrences
of TARGET_RELOCATABLE is a good thing.
Uros Bizjak [Wed, 4 May 2016 21:13:13 +0000 (23:13 +0200)]
re PR target/70873 ([7 Regressio] 20% performance regression at 482.sphinx3 after r235442 with -O2 -m32 on Haswell.)
PR target/70873
* config/i386/i386.md
(TARGET_SSE_PARTIAL_REG_DEPENDENCY float_extend sf->df peephole2):
Change to post-epilogue_completed late splitter. Use sse_reg_operand
as operand 0 predicate.
(TARGET_SSE_PARTIAL_REG_DEPENDENCY float_truncate df->sf peephole2):
Ditto.
(TARGET_SSE_PARTIAL_REG_DEPENDENCY float {si,di}->{sf,df} peephole2):
Ditto. Emit the pattern using RTX.
(TARGET_USE_VECTOR_FP_CONVERTS float_extend sf->df splitter):
Use sse_reg_opreand as operand 0 predicate. Do not use true_regnum in
the post-reload splitter. Use lowpart_subreg instead of gen_rtx_REG.
(TARGET_USE_VECTOR_FP_CONVERTS float_truncate df->sf splitter):
Ditto.
(TARGET_USE_VECTOR_CONVERTS float si->{sf,df} splitter): Use
sse_reg_operand as operand 0 predicate.
(TARGET_SPLIT_MEM_OPND_FOR_FP_CONVERTS float_extend sf->df peephole2):
Use sse_reg_opreand as operand 0 predicate. Use lowpart_subreg
instead of gen_rtx_REG.
(TARGET_SPLIT_MEM_OPND_FOR_FP_CONVERTS float_truncate sf->df peephole2):
Ditto.
cfgcleanup: Fold jumps and conditional branches with returns
This patch makes cfgcleanup optimize jumps to returns. There are three
cases this handles:
-- A jump to a return; this is simplified to just that return.
-- A conditional branch to a return; simplified to a conditional return.
-- A conditional branch that falls through to a return. This is simplified
to a conditional return (with the condition inverted), falling through
to a jump to the original destination. That jump can then be optimized
further, as usual.
This handles all cases the current function.c does, and a few it misses.
* cfgcleanup.c (bb_is_just_return): New function.
(try_optimize_cfg): Simplify jumps to return, branches to return,
and branches around return.
Jakub Jelinek [Wed, 4 May 2016 20:44:40 +0000 (22:44 +0200)]
re PR c++/70906 (ice in add_expr, at tree.c:7925)
PR c++/70906
PR c++/70933
* tree-core.h (enum operand_equal_flag): Add OEP_HASH_CHECK.
* tree.c (inchash::add_expr): If !IS_EXPR_CODE_CLASS (tclass),
assert flags & OEP_HASH_CHECK, instead of asserting it
never happens. Handle TARGET_EXPR.
* fold-const.c (operand_equal_p): For hash verification,
or in OEP_HASH_CHECK into flags.
* g++.dg/opt/pr70906.C: New test.
* g++.dg/opt/pr70933.C: New test.
Alan Modra [Wed, 4 May 2016 13:23:58 +0000 (22:53 +0930)]
[RS6000] Correct PIC_OFFSET_TABLE_REGNUM
Leaving this as r30 results in pic_offset_table_rtx of (reg 30)
for -m64, which is completely bogus. Various rtl analysis predicate
functions treat pic_offset_table_rtx specially..
Alan Modra [Wed, 4 May 2016 13:21:01 +0000 (22:51 +0930)]
[RS6000] Simplify sysv4.h TARGET_TOC
We can use the TARGET_* defines here. There isn't any reason to use
the underlying variable and masks. Also, TARGET_RELOCATABLE implies
TARGET_MINIMAL_TOC, so there is no need to test TARGET_RELOCATABLE.
Alan Modra [Wed, 4 May 2016 13:17:51 +0000 (22:47 +0930)]
[RS6000] Align .toc section
Lack of any .toc section alignment causes kexec and kdump failure
when linking without the usual linker script. This of course is
really a kexec-tools error, but it is also true that .toc ought to
always be word aligned.
Alan Modra [Wed, 4 May 2016 00:11:32 +0000 (09:41 +0930)]
Combine simplify_set WORD_REGISTER_OPERATIONS
The comment says this test is supposed to prevent "a narrower
operation than requested", but it actually only allows a larger
subreg not one the same size. Fix that.
gcc/
PR target/70866
* config/rs6000/rs6000.c (rs6000_stack_info): Don't set cr_save_p
when cr2,3,4 are all fixed regs.
gcc/testsuite/
* gcc.target/powerpc/pr70866.c: New.
Martin Sebor [Tue, 3 May 2016 21:15:28 +0000 (21:15 +0000)]
PR c++/66561 - __builtin_LINE at al. should yield constant expressions
PR c++/66561 - __builtin_LINE at al. should yield constant expressions
PR c++/66639 - declare __func__, __FUNCTION__ & __PRETTY_FUNCTION__ constexpr
gcc/testsuite/ChangeLog:
2016-05-03 Martin Sebor <msebor@redhat.com>
PR c++/66561
* c-c++-common/builtin_location.c: New test.
* g++.dg/cpp1y/builtin_location.C: New test.
gcc/cp/ChangeLog:
2016-05-03 Martin Sebor <msebor@redhat.com>
PR c++/66561
* tree.c (builtin_valid_in_constant_expr_p): Treat BUILT_IN_FILE,
BUILT_IN_FUNCTION, and BUILT_IN_LINE as constant expressions.
gcc/ChangeLog:
2016-05-03 Martin Sebor <msebor@redhat.com>
PR c++/66561
* builtins.c (fold_builtin_FILE): New function.
(fold_builtin_FUNCTION, fold_builtin_LINE): New functions.
(fold_builtin_0): Call them.
* gimplify.c (gimplify_call_expr): Remove the handling of
BUILT_IN_FILE, BUILT_IN_FUNCTION, and BUILT_IN_LINE.
PR c++/66561
* doc/extend.texi (Other Builtins): Update __builtin_FILE,
__builtin_FUNCTION, and __builtin_LINE to reflect they yield
constants.
PR c++/66639
* doc/extend.texi (Function Names as Strings): Update __func__,
__FUNCTION__, __PRETTY_FUNCTION__ to reflect they evaluate to
constants.
Jakub Jelinek [Tue, 3 May 2016 19:30:22 +0000 (21:30 +0200)]
re PR tree-optimization/70916 (gcc ICE at -O3 on valid code on x86_64-linux-gnu in "tree_operand_check")
PR tree-optimization/70916
* tree-if-conv.c: Include cfganal.h.
(pass_if_conversion::execute): Call connect_infinite_loops_to_exit
and remove_fake_exit_edges around the optimization pass.
Co-Authored-By: Richard Biener <rguenther@suse.de>
From-SVN: r235842
Jan Hubicka [Tue, 3 May 2016 19:05:31 +0000 (21:05 +0200)]
cgraph.c (symbol_table::create_edge): Set inline_failed.
* cgraph.c (symbol_table::create_edge): Set inline_failed.
(cgraph_edge::make_direct): Likewise.
(cgraph_edge::dump_edge_flags): Dump call_stmt_cannot_inline_p.
* cgraphclones.c (duplicate_thunk_for_node): Set inline_failed.
* cif-code.def (CIF_LTO_MISMATCHED_DECLARATIONS): New code
(CIF_THUNK): New code.
* ipa-inline-analysis.c (initialize_inline_failed): Preserve
CIF_FINAL_ERROR codes; do not deal with call_stmt_cannot_inline_p.
(compute_inline_parameters): Set inline_failed for thunks.
(inline_analyze_function): Cleanup.
* ipa-inline.c (can_inline_edge_p): Do not deal with
call_stmt_cannot_inline_p.
(can_early_inline_edge_p): Likewise.
(early_inliner): Initialize inline_failed.
* lto-cgraph.c (lto_output_edge): Sanity check inline_failed.
Uros Bizjak [Tue, 3 May 2016 18:48:18 +0000 (20:48 +0200)]
predicates.md (x87nonimm_ssenomem_operand): Rename from nonimm_ssenomem_operand.
* config/i386/predicates.md (x87nonimm_ssenomem_operand): Rename
from nonimm_ssenomem_operand.
(nonimm_ssenomem_operand): New predicate.
* config/i386/i386.md (extendsfdf2): Use nonimm_ssenomem_operand
as operand 0 predicate.
(*extendsfdf2): Merge from *extendsfdf2_mixed and *extendsfdf2_i387.
Disable unsupported alternatives using "enabled" attribute.
Use register_ssemem_operand as operand 0 predicate.
(*fop_<mode>_1): Use x87nonimm_ssenomem_operand as operand 1 predicate.
Under specific circumstances for Ada programs, such as in the testcase
this change adds, the DWARF back-end currently crashes because of
inconsistent internal state. This is due to a typo: a local variable is
called frame_offset_ but resolve_args_picking_1 wrongly modifies
emit-rtl.h's frame_offset instead.
This change fixes this typo.
gcc/
* dwarf2out.c (resolve_args_picking_1): Replace the frame_offset
occurence with frame_offset_ ones.
gcc/testsuite/
* gnat.dg/debug5.adb: New testcase.
Dominik Vogt [Tue, 3 May 2016 13:30:33 +0000 (13:30 +0000)]
S/390: Add patterns for r<nox>sbg instructions.
gcc/ChangeLog:
2016-05-03 Dominik Vogt <vogt@linux.vnet.ibm.com>
* config/s390/s390.md ("*r<noxa>sbg_<mode>_sll")
("*r<noxa>sbg_<mode>_srl"): New define_insns.
("*r<noxa>sbg_<mode>_srl_bitmask"): Rename by adding "_bitmask".
("*r<noxa>sbg_<mode>_sll_bitmask"): Likewise.
gcc/testsuite/ChangeLog:
2016-05-03 Dominik Vogt <vogt@linux.vnet.ibm.com>
* gcc.target/s390/md/rXsbg_mode_sXl.c: New test.
* gcc.target/s390/s390.exp (check_effective_target_z10_instructions):
Procedure to check for z10 instruction set.
Alan Modra [Tue, 3 May 2016 12:38:32 +0000 (22:08 +0930)]
[RS6000] Fix ICE caused by rs6000_savres_strategy thinko
rev 235672 (git cffc0b35) changed the condition for SAVE_MULTIPLE/
STORE_MULTIPLE, wrongly allowing a single reg.
gcc/
* config/rs6000/rs6000.c (rs6000_savres_strategy): Correct condition
for SAVE_MULTIPLE/STORE_MULTIPLE.
gcc/testsuite/
* gcc.target/powerpc/savres.c: Add func using a single gpr.
* gimplify.h (get_initialized_tmp_var): Add allow_ssa parameter
default true.
(gimplify_arg): Likewise.
* gimplify.c (gimplify_expr): Add overload with allow_ssa parameter,
re-writing the result to a decl if required.
(internal_get_tmp_var): Add allow_ssa parameter
and override into_ssa with it.
(get_formal_tmp_var): Adjust.
(get_initialized_tmp_var): Add allow_ssa parameter.
(gimplify_arg): Add allow_ssa parameter and avoid generating
SSA names for the result false.
(gimplify_call_expr): If the call may return twice do not
gimplify parameters into SSA.
(prepare_gimple_addressable): Do not allow an SSA name as
temporary.
(gimplify_modify_expr): Adjust assert. For noreturn calls
with a SSA name LHS adjust its def.
(gimplify_save_expr): Do not allow an SSA name as save-expr
result.
(gimplify_one_sizepos): Do not allow an SSA name as a sizepos.
(gimplify_body): Init GIMPLE SSA data structures and gimplify
into-SSA.
(gimplify_scan_omp_clauses): Make sure OMP_CLAUSE_SIZE is not
an SSA name. Likewise for OMP_CLAUSE_REDUCTION operands.
(gimplify_omp_for): Likewise for OMP_CLAUSE_DECL. Likewise
for OMP_FOR_COND, OMP_FOR_INCR and OMP_CLAUSE_LINEAR_STEP.
(optimize_target_teams): Do not allow SSA names for clause operands.
(gimplify_expr): Likewise for where we mark the result addressable.
* passes.def (pass_init_datastructures): Remove.
* tree-into-ssa.c (mark_def_sites): Ignore existing SSA names.
(rewrite_stmt): Likewise.
* tree-inline.c (initialize_cfun): Properly transfer SSA state.
(replace_locals_op): Replace SSA names.
(copy_gimple_seq_and_replace_locals): Init src_cfun.
* gimple-low.c (lower_builtin_setjmp): Deal with SSA.
* cgraph.c (release_function_body): Free CFG annotations only
when we have a CFG. Simplify.
* gimple-fold.c (gimplify_and_update_call_from_tree): Use
force_gimple_operand instead of get_initialized_tmp_var.
* tree-pass.h (make_pass_init_datastructures): Remove.
* tree-ssa.c (execute_init_datastructures): Remove.
(pass_data_init_datastructures): Likewise.
(class pass_init_datastructures): Likewise.
(make_pass_init_datastructures): Likewise.
* omp-low.c (create_omp_child_function): Init SSA data structures.
(grid_expand_target_grid_body): Likewise.
* tree-cfg.c (move_block_to_fn): Double-check the DEF is an SSA
name before adding it to names_to_release.
(remove_bb): Always release SSA defs.
* tree-ssa-ccp.c (get_default_value): Check SSA_NAME_VAR
before dereferencing it.
* cgraphunit.c (init_lowered_empty_function): Always
int SSA data structures.
* tree-ssanames.c (release_defs): Remove assert that we are in
SSA form.
* trans-mem.c (diagnose_tm_1): Handle SSA name function.
c-family/
* cilk.c (cilk_gimplify_call_params_in_spawned_fn): Do not
allow call args to gimplify to SSA names.
Jakub Jelinek [Tue, 3 May 2016 11:45:04 +0000 (13:45 +0200)]
re PR rtl-optimization/70467 (Useless "and [esp],-1" emitted on AND with uint64_t variable)
PR rtl-optimization/70467
* config/i386/predicates.md (x86_64_hilo_int_operand,
x86_64_hilo_general_operand): New predicates.
* config/i386/constraints.md (Wd): New constraint.
* config/i386/i386.md (mode attr di): Use Wd instead of e.
(general_hilo_operand): New mode attr.
(add<mode>3, sub<mode>3): Use <general_hilo_operand>
instead of <general_operand>.
(*add<dwi>3_doubleword, *sub<dwi>3_doubleword): Use
x86_64_hilo_general_operand instead of <general_operand>.
* gcc.target/i386/pr70467-3.c: New test.
* gcc.target/i386/pr70467-4.c: New test.
projects / thirdparty / gcc.git / log
