Jeff Law [Wed, 23 Mar 2016 13:20:16 +0000 (07:20 -0600)]
re PR tree-optimization/64058 (Performance degradation after r216304)
PR tree-optimization/64058
* tree-ssa-coalesce.c (struct coalesce_pair): Add new field
CONFLICT_COUNT.
(struct ssa_conflicts): Move up earlier in the file.
(conflicts_, var_map_): New static variables.
(initialize_conflict_count): New function to initialize the
CONFLICT_COUNT field for each conflict pair.
(compare_pairs): Lazily initialize the conflict count and use it
as the first tie-breaker.
(sort_coalesce_list): Add new arguments conflicts, map. Initialize
and wipe conflicts_ and map_ around the call to qsort. Remove
special case for 2 coalesce pairs.
* bitmap.c (bitmap_count_unique_bits): New function.
(bitmap_count_bits_in_word): New function, extracted from
bitmap_count_bits.
(bitmap_count_bits): Use bitmap_count_bits_in_word.
* bitmap.h (bitmap_count_unique_bits): Declare it.
Jakub Jelinek [Wed, 23 Mar 2016 09:52:00 +0000 (10:52 +0100)]
re PR tree-optimization/70354 (Wrong code with -O3 -march=broadwell and -march=skylake-avx512.)
PR tree-optimization/70354
* tree-vect-patterns.c (vect_recog_vector_vector_shift_pattern): If
oprnd0 is wider than oprnd1 and there is a cast from the wider
type to oprnd1, mask it with the mask of the narrower type.
* gcc.dg/vect/pr70354-1.c: New test.
* gcc.dg/vect/pr70354-2.c: New test.
* gcc.target/i386/avx2-pr70354-1.c: New test.
* gcc.target/i386/avx2-pr70354-2.c: New test.
Jakub Jelinek [Wed, 23 Mar 2016 09:49:12 +0000 (10:49 +0100)]
re PR target/70321 (STV generates less optimized code)
PR target/70321
* config/i386/i386.md (*anddi3_doubleword, *<code>di3_doubleword):
Optimize TARGET_STV splitters, if high or low word of last argument
is 0 or -1.
* include/Makefile.am (install-freestanding-headers): Add
concept_check.h and move.h to the installed headers.
* include/Makefile.in: Regenerated.
* include/bits/concept_check.h: Ignore _GLIBCXX_CONCEPT_CHECKS for
freestanding implementations.
* doc/xml/manual/using.xml (_GLIBCXX_CONCEPT_CHECKS): Mention
that this macro has no effect for freestanding implementations.
* doc/html/manual/using_macros.html: Likewise.
Jeff Law [Tue, 22 Mar 2016 21:32:34 +0000 (15:32 -0600)]
re PR target/70232 (excessive stack usage with -O2)
PR target/70232
tree-ssa-threadbackward.c
(fsm_find_control_statement_thread_paths): Correctly distinguish
between old style jump threads vs FSM jump threads.
PR target/70232
* gcc.dg/tree-ssa/pr70232.c: New test.
Michael Meissner [Tue, 22 Mar 2016 21:05:43 +0000 (21:05 +0000)]
re PR libgcc/70363 (PowerPC __float128 to long double doesn't link if built with an assember without ISA 3.0 support)
2016-03-22 Michael Meissner <meissner@linux.vnet.ibm.com>
PR libgcc/70363
* config/rs6000/extendkftf2-sw.c (__extendkftf2_sw): If libgcc was
built with an assembler that does not support ISA 3.0
instructions, rename __extendkftf2_sw to __extendkftf2.
Richard Biener [Tue, 22 Mar 2016 14:38:42 +0000 (14:38 +0000)]
re PR tree-optimization/70251 (Wrong code with -O3 -march=skylake-avx512.)
2016-03-22 Richard Biener <rguenther@suse.de>
PR middle-end/70251
* genmatch.c (gen_transform): Adjust last parameter to a three-state
int...
(capture::gen_transform): ... to change behavior when substituting
a condition into cond or not-cond expr context.
(dt_simplify::gen_1): Adjust.
* gimple-match-head.c: Include gimplify.h for unshare_expr.
* match.pd (A + (B vcmp C ? 1 : 0) -> A - (B vcmp C)): Revert
last change and instead change to
A + (B vcmp C ? 1 : 0) -> A - (B vcmp C ? -1 : 0).
(A - (B vcmp C ? 1 : 0) -> A + (B vcmp C)): Likewise.
David Malcolm [Tue, 22 Mar 2016 14:20:49 +0000 (14:20 +0000)]
PR c/69993: improvements to wording of -Wmisleading-indentation
gcc/c-family/ChangeLog:
PR c/69993
* c-indentation.c (warn_for_misleading_indentation): Rewrite the
diagnostic text, reversing the order of the warning and note so
that they appear in source order.
gcc/testsuite/ChangeLog:
PR c/69993
* c-c++-common/Wmisleading-indentation-3.c: New test, based on
Wmisleading-indentation.c.
* c-c++-common/Wmisleading-indentation.c: Update thoughout to
reflect change to diagnostic text and order of messages.
* gcc.dg/plugin/location-overflow-test-2.c: Likewise.
Kirill Yukhin [Tue, 22 Mar 2016 11:09:03 +0000 (11:09 +0000)]
re PR target/70325 (ICE on __builtin_ia32_storedquqi256_mask)
PR target/70325
gcc/
* config/i386/i386.c (def_builtin): Handle
OPTION_MASK_ISA_AVX512VL to be and-ed with other
bits.
(const struct builtin_description bdesc_special_args[]):
Remove duplicate ISA bits.
gcc/testsuite/
* gcc.target/i386/pr70325.c: New test.
Jakub Jelinek [Tue, 22 Mar 2016 08:28:49 +0000 (09:28 +0100)]
re PR target/70329 (wrong code with -mavx512bw and 64byte vectors)
PR target/70329
* config/i386/i386.c (ix86_expand_vecop_qihi): Don't bother computing
d.perm[i] for i >= d.nelt. If not full_interleave, compute d.perm[i]
in a way that works also for AVX512BW.
* gcc.target/i386/avx512bw-pr70329-1.c: New test.
* gcc.target/i386/avx512bw-pr70329-2.c: New test.
Jakub Jelinek [Tue, 22 Mar 2016 08:14:24 +0000 (09:14 +0100)]
re PR target/70300 (ICE: in extract_constrain_insn, at recog.c:2190 (insn does not satisfy its constraints) with -mtune=amdfam10 -mavx512bw)
PR target/70300
* config/i386/i386.md (cvtsd2ss splitter): Unpack in destination
instead of source if operands[1] is xmm16 and above and
!TARGET_AVX512VL. Use avx512f_vec_dupv16sf_1 instead of
vec_interleave_lowv4sf if we need to unpack xmm16 and above.
Jakub Jelinek [Tue, 22 Mar 2016 08:11:14 +0000 (09:11 +0100)]
re PR c++/70295 (bogus -Wnonnull-compare for "dynamic_cast<const FuSelection *>(this) != nullptr")
PR c++/70295
* gimplify.c (gimplify_modify_expr): Call gimple_set_no_warning
on assign if (*from_p) is a comparison, set it to
TREE_NO_WARNING (*from_p).
* c-c++-common/nonnull-1.c (func): Remove parens around cp4 != 0.
(func2): New function for cond with parens, xfail warning for c++.
* g++.dg/warn/Wnonnull-compare-8.C: New test.
Uros Bizjak [Mon, 21 Mar 2016 19:56:50 +0000 (20:56 +0100)]
re PR target/70327 (ICE: in extract_insn, at recog.c:2287 (unrecognizable insn) with -mavx512ifma and v4ti argument)
PR target/70327
* config/i386/i386.md (movxi): Use ix86_expand_vector_move instead
of ix86_expand_move.
(movoi): Ditto.
(movti): Use general_operand for operand 1 predicate.
testsuite/ChangeLog:
PR target/70327
* gcc.target/i386/pr70327.c: New test.
Jakub Jelinek [Mon, 21 Mar 2016 15:41:13 +0000 (16:41 +0100)]
re PR target/70296 (Incorrect handling of vector X; if X is function-like macro)
PR target/70296
* include/cpplib.h (cpp_fun_like_macro_p): New prototype.
* macro.c (cpp_fun_like_macro_p): New function.
* config/rs6000/rs6000-c.c (rs6000_macro_to_expand): If IDENT is
function-like macro, peek following token(s) if it is followed
by CPP_OPEN_PAREN token with optional padding in between, and
if not, don't treat it like a macro.
Martin Liska [Mon, 21 Mar 2016 10:27:53 +0000 (11:27 +0100)]
Retry to emit global variables in HSA (PR hsa/70234)
PR hsa/70234
* hsa-brig.c (emit_function_directives): Mark unemitted
global variables for emission.
* hsa-gen.c (hsa_symbol::hsa_symbol): Initialize a new flag.
(get_symbol_for_decl): Likewise.
* hsa.h (struct hsa_symbol): New flag.
Richard Biener [Mon, 21 Mar 2016 09:11:37 +0000 (09:11 +0000)]
re PR tree-optimization/70288 (ICE with -O1 -fno-tree-dominator-opts)
2016-03-21 Richard Biener <rguenther@suse.de>
PR tree-optimization/70288
* tree-ssa-loop-ivcanon.c (tree_estimate_loop_size): Make sure
we do not estimate unsimplified all-constant conditionals or
switches as optimized away.
* gcc.dg/torture/pr70288-1.c: New testcase.
* gcc.dg/torture/pr70288-2.c: Likewise.
Jeff Law [Fri, 18 Mar 2016 19:30:20 +0000 (13:30 -0600)]
re PR rtl-optimization/70263 (ICE at -O1 and above in both 32-bit and 64-bit modes on x86_64-linux-gnu (segmentation fault))
PR rtl-optimization/70263
* ira.c (memref_used_between_p): Assert we found END in the insn chain.
(update_equiv_regs): When trying to move a store to after the insn
that sets the source of the store, make sure the store occurs after
the insn that sets the source of the store. When successful note
the REG_EQUIV note created in the dump file.
PR rtl-optimization/70263
* gcc.c-torture/compile/pr70263-1.c: New test.
* gcc.target/i386/pr70263-2.c: New test.
David Malcolm [Fri, 18 Mar 2016 17:45:42 +0000 (17:45 +0000)]
PR c/70281: C FE: fix uninitialized range for __builtin_types_compatible_p
gcc/c/ChangeLog:
PR c/70281
* c-parser.c (c_parser_postfix_expression): Set the source range
for uses of "__builtin_types_compatible_p".
gcc/testsuite/ChangeLog:
PR c/70281
* gcc.dg/plugin/diagnostic-test-expressions-1.c
(test_builtin_types_compatible_p): New test function.
* gcc.dg/pr70281.c: New test case.
Jason Merrill [Fri, 18 Mar 2016 15:31:35 +0000 (11:31 -0400)]
PR c++/70147 - handle primary virtual bases
* class.c (vptr_via_virtual_p): New.
(most_primary_binfo): Factor out of build_rtti_vtbl_entries.
* cp-ubsan.c (cp_ubsan_dfs_initialize_vtbl_ptrs): Don't clear
a vptr from any virtual base in a not-in-charge 'structor.
Uros Bizjak [Fri, 18 Mar 2016 15:28:54 +0000 (16:28 +0100)]
Revert:
2015-07-02 Uros Bizjak <ubizjak@gmail.com>
* libsupc++/guard.cc (__test_and_acquire): Use __p after __atomic_load
to avoid unused variable warning.
(__set_and_release): Use __p after __atomic_store to avoid unused
variable warning.
Ilya Enkovich [Fri, 18 Mar 2016 09:36:32 +0000 (09:36 +0000)]
re PR tree-optimization/70252 (ICE in vect_get_vec_def_for_stmt_copy with -O3 -march=skylake-avx512.)
gcc/
PR tree-optimization/70252
* tree-vect-stmts.c (supportable_widening_operation): Check resulting
boolean vector has a proper number of elements.
(supportable_narrowing_operation): Likewise.
gcc/testsuite/
PR tree-optimization/70252
* gcc.dg/pr70252.c: New test.
Jakub Jelinek [Thu, 17 Mar 2016 23:00:04 +0000 (00:00 +0100)]
re PR middle-end/70280 (-fcompare-debug failure (length) with --param=integer-share-limit=4016 -mavx512bw)
PR c/70280
* c-typeck.c (composite_type): Don't count void_list_node
into len, if the list is terminated by void_list_node, start
with void_list_node instead of NULL for newargs. Stop
at void_list_node.
David Malcolm [Thu, 17 Mar 2016 18:27:47 +0000 (18:27 +0000)]
PR c/70264: fix crash in compatible_locations_p with BUILTINS_LOCATION
In r234088 my fix for PR c++/70105 didn't allow for the possibility that
when comparing a pair of macro expansion histories that one of the
macros in the history might not be located within a line-map, and
PR c/70264 reports a crash due to encountering BUILTINS_LOCATION within
the traversal.
Fixed thusly.
Successfully bootstrapped on x86_64-pc-linux-gnu; adds 4 PASS results
to gcc.sum and 12 to g++.sum.
gcc/ChangeLog:
PR c/70264
* diagnostic-show-locus.c (compatible_locations_p): Handle the case
where one or both locations aren't within a line_map.
gcc/testsuite/ChangeLog:
PR c/70264
* c-c++-common/pr70264.c: New test case.
Jakub Jelinek [Thu, 17 Mar 2016 15:58:22 +0000 (16:58 +0100)]
re PR tree-optimization/70144 (g++ ICE at -O1 and above on valid code on x86_64-linux-gnu in "copy_reference_ops_from_ref")
PR c++/70144
* cp-tree.h (magic_varargs_p): Return int instead of bool.
* call.c (magic_varargs_p): Return int instead of bool, return 2 for
Cilk+ reductions, otherwise 1 for magic varargs and 0 for normal
varargs.
(build_over_call): If magic_varargs_p == 2, call reject_gcc_builtin,
if magic_varargs_p == 1, call decay_conversion
instead of mark_type_use. Don't store error_mark_node arguments to
argarray, instead return error_mark_node.
* c-c++-common/pr70144-1.c: New test.
* c-c++-common/pr70144-2.c: New test.
H.J. Lu [Thu, 17 Mar 2016 15:11:35 +0000 (15:11 +0000)]
Properly set flag_pie and flag_pic
We can't set flag_pie to the default when flag_pic == 0, which may be
set by -fno-pic or -fno-PIC, since the default value of flag_pie is
non-zero when GCC is configured with --enable-default-pie. We need
to initialize flag_pic to -1 so that we can tell if -fpic, -fPIC,
-fno-pic or -fno-PIC is used.
Since Darwin defaults to PIC (__PIC__ == 2) and the PIC setting can't
be changed, skip tests of default __PIC__ and __PIE__ setting for
*-*-darwin* targets.
gcc/
PR driver/70192
* opts.c (finish_options): Don't set flag_pie to the default if
-fpic, -fPIC, -fno-pic or -fno-PIC is used. Set flag_pic to 0
if it is -1.
Jakub Jelinek [Thu, 17 Mar 2016 11:53:53 +0000 (12:53 +0100)]
re PR middle-end/70245 (Miscompilation of ICU on i386 with atom tuning starting with r227382)
PR target/70245
* rtl.h (replace_rtx): Add ALL_REGS argument.
* rtlanal.c (replace_rtx): Likewise. If true, use REGNO
equality and assert mode is the same, instead of just rtx pointer
equality.
* config/i386/i386.md (mov + arithmetics with load peephole): Pass
true as ALL_REGS argument to replace_rtx.
re PR target/70048 ([AArch64] Inefficient local array addressing)
PR target/70048
* config/aarch64/aarch64.c (virt_or_elim_regno_p): New.
(aarch64_classify_address): Use it.
(aarch64_legitimize_address): Force all subexpressions of PLUS
into registers. Simplify as (sfp+const)+reg or (reg+reg)+const.
projects / thirdparty / gcc.git / log
