H.J. Lu [Thu, 18 Dec 2014 17:35:45 +0000 (17:35 +0000)]
X86-64: Add -mskip-rax-setup
The Linux kernel never passes floating point arguments around, vararg
functions or not. Hence no vector registers are ever used when calling a
vararg function. But gcc still dutifully emits an "xor %eax,%eax" before
each and every call of a vararg function. Since no callee use that for
anything, these instructions are redundant.
This patch adds the -mskip-rax-setup option to skip setting up RAX
register when SSE is disabled and there are no variable arguments passed
in vector registers. Since RAX register is used to avoid unnecessarily
saving vector registers on stack when passing variable arguments, the
impacts of this option are callees may waste some stack space, misbehave
or jump to a random location. GCC 4.4 or newer don't those issues,
regardless the RAX register value since they don't check the RAX register
value when SSE is disabled.
gcc/
* config/i386/i386.c (ix86_expand_call): Skip setting up RAX
register for -mskip-rax-setup when there are no parameters
passed in vector registers.
* config/i386/i386.opt (mskip-rax-setup): New option.
* doc/invoke.texi: Document -mskip-rax-setup.
gcc/testsuite/
* gcc.target/i386/amd64-abi-7.c: New tests.
* gcc.target/i386/amd64-abi-8.c: Likwise.
* gcc.target/i386/amd64-abi-9.c: Likwise.
Martin Liska [Thu, 18 Dec 2014 13:32:18 +0000 (14:32 +0100)]
re PR tree-optimization/64330 (IPA-ICF merges const exported vars that could be addressable in other TUs)
Fix for PR64330.
PR tree-optimization/64330
* ipa-icf.c (sem_variable::parse): Add checking
for externally visible symbols and do not introduce
an alias for an external declaration.
Martin Liska [Thu, 18 Dec 2014 12:02:22 +0000 (13:02 +0100)]
re PR ipa/64146 (ipa-icf breaks gold dynamic_test_2 test)
Fix for PR ipa/64146
PR ipa/64146
* ipa-icf.c (sem_function::merge): Check for
decl_binds_to_current_def_p is newly added to merge operation.
* g++.dg/ipa/pr64146.C: New test.
Bin Cheng [Thu, 18 Dec 2014 02:53:42 +0000 (02:53 +0000)]
re PR middle-end/62178 ([AArch64] Performance regression on matrix matrix multiply due to r211211)
PR tree-optimization/62178
* tree-ssa-loop-ivopts.c (cheaper_cost_with_cand): New function.
(iv_ca_replace): New function.
(try_improve_iv_set): New parameter try_replace_p.
Break local optimal fixed-point by calling iv_ca_replace.
(find_optimal_iv_set_1): Pass new argument to try_improve_iv_set.
gcc/testsuite/ChangeLog
PR tree-optimization/62178
* gcc.target/aarch64/pr62178.c: New test.
Oleg Endo [Wed, 17 Dec 2014 23:08:14 +0000 (23:08 +0000)]
re PR target/51244 ([SH] Inefficient conditional branch and code around T bit)
gcc/
PR target/51244
* config/sh/sh_treg_combine.cc (is_conditional_insn): New function.
(cbranch_trace): Add member rtx* condition_rtx_in_insn, initialize it
accordingly in constructor.
(cbranch_trace::branch_condition_rtx_ref): New function.
(cbranch_trace::branch_condition_rtx): Use branch_condition_rtx_ref.
(sh_treg_combine::try_invert_branch_condition): Invert condition rtx
in insn using reversed_comparison_code and validate_change instead of
invert_jump_1.
(sh_treg_combine::execute): Look for conditional insns in basic blocks
in addition to conditional branches.
* config/sh/sh.md (*movsicc_div0s): Remove combine patterns.
Make the Ada front-end use the array descr language hook
* gcc-interface/misc.c (gnat_get_array_descr_info): New. Use it for
the get_array_descr_info lang-hook. Use it to tune the DWARF output
for array types.
Enable the array descr language hook for all DWARF versions
* dwarf2out.c (gen_type_die_with_usage): Enable the array lang-hook
even when (dwarf_version < 3 && dwarf_strict).
(gen_descr_array_die): Do not output DW_AT_data_locationn,
DW_AT_associated, DW_AT_allocated and DW_AT_byte_stride DWARF
attributes when (dwarf_version < 3 && dwarf_strict).
Complete information generated through the array descriptor language hook
gcc/
* dwarf2out.h (enum array_descr_ordering): New.
(array_descr_dimen): Add a bounds_type structure field.
(struct array_descr_info): Add a field to hold index type information
and another one to hold ordering information.
* dwarf2out.c (gen_type_die_with_usage): Get the main variant before
invoking the array descriptor language hook. Initialize the
array_descr_info structure before calling the lang-hook.
(gen_descr_array_type_die): Use gen_type_die if not processing the main
type variant. Replace Fortran-specific code with generic one using
this new field. Add a GNAT descriptive type, if any. Output type
information for the array bound subrange, if any.
gcc/fortran
* trans-types.c (gfc_get_array_descr_info): Describe all Fortran arrays
with column major ordering.
H.J. Lu [Wed, 17 Dec 2014 14:22:57 +0000 (14:22 +0000)]
Add -malign-data={abi|compat|cachineline}
Add -malign-data={abi|compat,cachineline} to control how GCC aligns
variables. "compat" uses increased alignment value compatible with
GCC 4.8 and earlier, "abi" uses alignment value as specified by the
psABI, and "cacheline" uses increased alignment value to match the
cache line size. "compat" is the default.
gcc/
PR target/61296
* config/i386/i386-opts.h (ix86_align_data): New enum.
* config/i386/i386.c (ix86_data_alignment): Return the ABI
alignment value for -malign-data=abi, the cachine line size
for -malign-data=cachineline and the older GCC compatible
alignment value for for -malign-data=compat.
* config/i386/i386.opt (malign-data=): New.
* doc/invoke.texi: Document -malign-data=.
Jakub Jelinek [Wed, 17 Dec 2014 09:29:12 +0000 (10:29 +0100)]
re PR tree-optimization/64322 (More optimize opportunity for constant folding)
PR tree-optimization/64322
* tree-vrp.c (extract_range_from_binary_expr_1): Attempt to derive
range for RSHIFT_EXPR even if vr0 range is not VR_RANGE or is symbolic.
Jakub Jelinek [Wed, 17 Dec 2014 09:26:49 +0000 (10:26 +0100)]
re PR sanitizer/64289 (ICE with -fsanitize=float-cast-overflow)
PR sanitizer/64289
* c-convert.c: Include ubsan.h.
(convert): For real -> integral casts and
-fsanitize=float-cast-overflow don't call convert_to_integer, but
instead instrument the float cast directly.
Ian Lance Taylor [Tue, 16 Dec 2014 22:53:38 +0000 (22:53 +0000)]
compiler: Don't built hash/equality functions for thunk structs.
They are never necessary, and they can cause problems when a
thunk is used to pass an unexported type from a different
package to a function defined in that package. The resulting
struct type may need to call the comparison routine from the
other package, which will fail because the type is not
exported. This will be bug492 in the master testsuite.
Jan Hubicka [Tue, 16 Dec 2014 22:16:17 +0000 (23:16 +0100)]
ipa-inline-analysis.c (will_be_nonconstant_predicate): Consider return values of const calls as constants.
* ipa-inline-analysis.c (will_be_nonconstant_predicate): Consider
return values of const calls as constants.
(estimate_function_body_sizes): Expect calls to have false predicates.
Oleg Endo [Tue, 16 Dec 2014 21:28:59 +0000 (21:28 +0000)]
re PR target/53513 ([SH] Add support for fpchg insn and improve fenv support)
gcc/testsuite/
PR target/53513
* gcc.target/sh/fpchg.c: Rename to ...
* gcc.target/sh/pr53513-1.c: ... this. Adjust test case to work for
-m4a and -m4a-single.
* genmatch.c (parser::parser): Initialize capture_ids.
(parser::parse_pattern): Properly allocate capture_ids before
using them. Set capture_ids to zero when its lifetime is
supposed to finish.
(parser::parse_simplify): Allocate capture_ids only if
required.
Janus Weil [Tue, 16 Dec 2014 08:15:38 +0000 (09:15 +0100)]
re PR fortran/64244 (ICE at class.c:236 when using non_overridable)
2014-12-16 Janus Weil <janus@gcc.gnu.org>
PR fortran/64244
* resolve.c (resolve_typebound_call): New argument to pass out the
non-overridable attribute of the specific procedure.
(resolve_typebound_subroutine): Get overridable flag from
resolve_typebound_call.
Uros Bizjak [Tue, 16 Dec 2014 07:49:53 +0000 (08:49 +0100)]
gnu-user.h (TARGET_CAN_SPLIT_STACK): Move from here ...
* config/i386/gnu-user.h (TARGET_CAN_SPLIT_STACK): Move from here ...
* config/i386/gnu-user64.h (TARGET_CAN_SPLIT_STACK): ... and here ...
* config/i386/gnu-user-common.h (TARGET_CAN_SPLIT_STACK): ... to here.
Jakub Jelinek [Mon, 15 Dec 2014 20:10:45 +0000 (21:10 +0100)]
re PR rtl-optimization/64316 (ICE in simplify_const_unary_operation after r218503)
PR rtl-optimization/64316
* simplify-rtx.c (simplify_relational_operation_1): For
(eq/ne (and x y) x) and (eq/ne (and x y) y) optimizations use
CONST0_RTX instead of const0_rtx.
* ira-int.h (ira_prohibited_class_mode_regs): Remove.
(struct target_ira_int): Move x_ira_prohibited_class_mode_regs to
...
* ira.h (struct target_ira_int): ... here.
(ira_prohibited_class_mode_regs): Define.
* lra-constraints.c (process_alt_operands): Add one more condition
to refuse alternative when reload pseudo of given class can not
hold value of given mode.
Steven Bosscher [Mon, 15 Dec 2014 18:37:59 +0000 (18:37 +0000)]
re PR fortran/61669 (Error recovery ICE)
PR fortran/61669
* gfortran.h (struct gfc_namespace): Add OLD_DATA field.
* decl.c (gfc_reject_data): New function.
* parse.c *use_modules): Record roll-back point.
(next_statement): Likewise.
(reject_statement): Roll back to last accepted DATA.
* gfortran.dg/pr61669.f90: New test.
Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r218757
Richard Biener [Mon, 15 Dec 2014 18:33:16 +0000 (18:33 +0000)]
re PR tree-optimization/64312 (ICE: Segmentation fault)
2014-12-15 Richard Biener <rguenther@suse.de>
PR tree-optimization/64312
* tree-ssa-sccvn.c (vn_reference_lookup_pieces): Use
vuse_ssa_val as callback to walk_non_aliased_vuses.
(vn_reference_lookup): Likewise.
Jason Merrill [Mon, 15 Dec 2014 18:31:40 +0000 (13:31 -0500)]
N3778: Sized Deallocation
N3778: Sized Deallocation
gcc/c-family/
* c.opt (-fsized-deallocation, -Wc++14-compat): New.
(-Wsized-deallocation): New.
* c-opts.c (c_common_post_options): -fsized-deallocation defaults
to on in C++14 and up.
gcc/cp/
* call.c (non_placement_deallocation_fn_p): A global sized
operator delete is not a usual deallocation function until C++14.
(build_op_delete_call): Choose the global sized op delete if we
know the size.
* cp-tree.h: Declare non_placement_deallocation_fn_p.
(enum cp_tree_index): Remove CPTI_GLOBAL_DELETE_FNDECL.
(global_delete_fndecl): Remove.
* decl.c (cxx_init_decl_processing): Also declare sized op deletes.
(grok_op_properties): Warn about sized dealloc without the flag.
* init.c (build_builtin_delete_call): Remove.
(build_vec_delete_1, build_delete): Don't call it.
* decl2.c (maybe_warn_sized_delete): New.
(cp_write_global_declarations): Call it.
libstdc++-v3/
* libsupc++/del_ops.cc: New.
* libsupc++/del_opvs.cc: New.
* libsupc++/Makefile.am: Add them.
* libsupc++/Makefile.in: Regenerate.
* config/abi/pre/gnu.ver: Export _ZdlPvm and _ZdaPvm.
Jakub Jelinek [Mon, 15 Dec 2014 10:50:52 +0000 (11:50 +0100)]
re PR target/64210 (FAIL: gcc.target/i386/avx512vl-(vmovdqa64|vpbroadcastd)-1.c ... with -fpic)
PR target/64210
* gcc.target/i386/avx512f-broadcast-gpr-1.c: Use %(?:e|r\[0-9\]+d)
instead of %e in regexps trying to match 32-bit GPR.
* gcc.target/i386/avx512f-vpbroadcastd-1.c: Likewise.
* gcc.target/i386/avx512vl-vpbroadcastd-1.c: Likewise.
* gcc.target/i386/avx512vl-vmovdqa64-1.c: Restrict some scan-assembler-times
lines to nonpic targets only. Fix up \[^\n^x^y\] to \[^\nxy\].
projects / thirdparty / gcc.git / log
