Using -funwind-tables is necessary to permit Go code to correctly
throw a panic through C code. This hasn't been necessary in the past
as -funwind-tables is the default on x86. However, it is not the
default for PPC AIX.
Reviewed-on: https://go-review.googlesource.com/56650
From-SVN: r251179
2017-08-17 Martin Liska <mliska@suse.cz>
PR bootstrap/81864
* tree-loop-distribution.c (ddrs_table): Change type to pointer
type.
(get_data_dependence): Use it as pointer type.
(distribute_loop): Likewise.
From-SVN: r251165
2017-08-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
* config/rs6000/altivec.md (UNSPEC_VMRGOW_DIRECT): New constant.
(p8_vmrgew_v4sf_direct): Generalize to p8_vmrgew_<mode>_direct.
(p8_vmrgow_<mode>_direct): New define_insn.
* config/rs6000/rs6000.c (altivec_expand_vec_perm_const): Properly
handle endianness for vmrgew and vmrgow permute patterns.
From-SVN: r251161
gcc/ChangeLog:
PR c/81859
* pretty-print.c (pp_format): Use strnlen in %.*s to avoid reading
past the end of an array.
(test_pp_format): Add test cases.
From-SVN: r251157
PR middle-end/81814
* fold-const.c (operand_equal_for_comparison_p): Remove code that used
to mimic what shorten_compare did. Change the return type to bool.
(fold_cond_expr_with_comparison): Update call to
operand_equal_for_comparison_p.
(fold_ternary_loc): Likewise.
* gcc.dg/torture/pr81814.c: New test.
From-SVN: r251152
This patch changes patterns in aarch64-simd.md to replace
movi v0.4s, 0
str q0, [x0, 16]
With:
stp xzr, xzr, [x0, 16]
When we are storing zeros to vectors like this:
void f(uint32x4_t *p) {
uint32x4_t x = { 0, 0, 0, 0};
p[1] = x;
}
gcc/
2017-08-17 Jackson Woodruff <jackson.woodruff@arm.com>
* aarch64-simd.md (mov<mode>): No longer force zero immediate into
register.
(*aarch64_simd_mov<mode>): Add new case for stp using zero immediate.
gcc/testsuite/
2017-08-17 Jackson Woodruff <jackson.woodruff@arm.com>
* gcc.target/aarch64/simd/vect_str_zero.c: New testcase.
From-SVN: r251149
2017-08-17 Richard Biener <rguenther@suse.de>
* tree-ssa-structalias.c (solve_graph): When propagating
to successors update the graphs succ edges and avoid duplicate work.
From-SVN: r251146
2017-08-17 Maxim Ostapenko <m.ostapenko@samsung.com>
PR target/81861
* config/i386/i386.c (ix86_option_override_internal): Save target
specific options after ix86_stack_protector_guard_reg was changed.
From-SVN: r251145
2017-08-17 Richard Biener <rguenther@suse.de>
* tree-vrp.c (vrp_int_const_binop): Do not set *overflow_p
to true when overflow is undefined and we saturated the
result.
* gcc.dg/tree-ssa/vrp117.c: New testcase.
From-SVN: r251141
It is possible when using out-of-line register saves or store multiple
to save some registers unnecessarily, for example one reg in the block
saved might be unused. We don't need to emit frame info for those
registers as that just bloats the info, and also can result in an ICE
when shrink-wrap gives multiple paths through the function saving
different sets of registers. Join points need to have identical frame
register save state regardless of the path taken.
This patch reverts the previous fix for PR80939 "Use SAVE_MULTIPLE
only if we restore what it saves (PR80938)" and instead fixes the PR
by correcting the frame info. The change to rs6000_savres_strategy
is an optimization, but note that it hides the underlying problem in
the PR testcase.
PR target/80938
* config/rs6000/rs6000.c (rs6000_savres_strategy): Revert 2017-08-09.
Don't use store multiple if only one reg needs saving.
(interesting_frame_related_regno): New function.
(rs6000_frame_related): Don't emit frame info for regs that
don't need saving.
(rs6000_emit_epilogue): Likewise.
From-SVN: r251140
* tree-core.h (tree_type_non_common): Rename binfo to lang_1.
* tree.h (TYPE_BINFO): Use type_non_common.maxval.
(TYPE_LANG_SLOT_1): Use type_non_common.lang_1, for any type.
* tree.c (free_lang_data_in_type): Use else-if chain. Always
clear TYPE_LANG_1. Remove obsolete member-function stripping.
(find_decls_types_r): Comment about TYPE_MAX_VALUES_RAW.
(verify_type): Adjust for TYPE_BINFO move.
* lto-streamer-out.c (DFS::DFS_write_tree_body): No need to
process TYPE_BINFO directly.
(hash_tree): Likewise.
* tree-streamer-in.c (lto_input_ts_type_non_common_tree_pointers):
Likewise.
* tree-streamer-out.c (write_ts_type_non_common_tree_pointers):
Likewise.
lto/
* lto.c (mentions_vars_p_type): Use TYPE_LANG_SLOT_1.
(compare_tree_sccs_1): No need to compare TYPE_BINFO directly.
(lto_fixup_prevailing_decls): Use TYPE_LANG_SLOT_1.
From-SVN: r251129
2017-08-16 Thomas Koenig <tkoenig@gcc.gnu.org>
PR fortran/81116
* frontend-passes.c (realloc_string_callback): If expression is a
concatenation, also check for dependency.
(constant_string_length): Check for presence of symtree.
2017-08-16 Thomas Koenig <tkoenig@gcc.gnu.org>
PR fortran/81116
* gfortran.dg/realloc_on_assignment_29.f90: New test.
From-SVN: r251125
We weren't checking whether the phi in a conditional reduction was
used by the condition itself (which isn't a case we handle).
2017-08-11 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
PR tree-optimization/81835
* tree-vect-loop.c (vect_is_simple_reduction): Simply checks for
the phi SSA_NAME. Check that the condition in a COND_EXPR does
not depend on the phi.
gcc/testsuite/
PR tree-optimization/81835
* gcc.dg/vect/pr81815.c: New test.
From-SVN: r251117
rs6000_reg_live_or_pic_offset_p is just save_reg_p with special
handling for the pic register and eh_return. This merge also
simplifies the eh_return handling. The intent of
https://gcc.gnu.org/ml/gcc-patches/2010-09/msg01838.html was to say
the PIC reg needed to be saved for eh_return, not all gprs. Of
course, it doesn't hurt to say all gprs need to be saved for eh_return
as that is what the target-independent code does by setting DF live,
but it's unnecessary in the backend.
* config/rs6000/rs6000.c (rs6000_reg_live_or_pic_offset_p): Merge..
(save_reg_p): ..into this. Update all callers.
(first_reg_to_save): Simplify.
From-SVN: r251115
GCC mainline built for sh4-linux-gnu runs out of memory building a
glibc test, which calls strncmp with very large constant size
argument, resulting in the SH inline strncmp expansion trying to
inline a fully unrolled expansion of strncmp for that size.
This patch limits that fully unrolled expansion to the case of less
than 32 bytes. This is explicitly *not* trying to be optimal in any
way (very likely a lower threshold makes sense), just to limit enough
to avoid the out-of-memory issue in the glibc testsuite.
I have *not* run the GCC testsuite for SH. I have verified that this
allows the glibc testsuite to build OK, with both GCC mainline and GCC
7 branch (and that the included test builds quickly with patched GCC,
runs out of memory with unpatched GCC).
PR target/78460
PR target/67712
gcc:
* config/sh/sh-mem.cc (sh_expand_cmpnstr): Only unroll for
constant count if that count is less than 32.
gcc/testsuite:
* gcc.c-torture/compile/string-large-1.c: New test.
From-SVN: r251108
2017-08-15 Richard Biener <rguenther@suse.de>
PR tree-optimization/81790
* tree-ssa-sccvn.c (vn_lookup_simplify_result): Handle both
CONSTRUCTORs from simplifying and VN.
* gcc.dg/torture/pr81790.c: New testcase.
From-SVN: r251103
Fix longbranch test so it still generates long tbz branches.
gcc/testsuite/
PR target/81643
* gcc.target/aarch64/long_branch_1.c: Improve testcase.
From-SVN: r251094
