Michael Meissner [Wed, 30 Nov 2016 00:05:46 +0000 (00:05 +0000)]
re PR target/78594 (Bug in November 11th, 2016 change to rs6000.md)
2016-11-29 Michael Meissner <meissner@linux.vnet.ibm.com>
PR target/78594
* config/rs6000/rs6000.md (mov<mode>_internal, QHI iterator): Add
'x' to stxsi<wd>x print pattern, so that QImode and HImode values
residing in traditional altivec registers can be stored
correctly.
Max Filippov [Tue, 29 Nov 2016 22:22:13 +0000 (22:22 +0000)]
xtensa: Fix PR target/78603
2016-11-29 Max Filippov <jcmvbkbc@gmail.com>
gcc/
* config/xtensa/xtensa.c (hwloop_optimize): Don't emit zero
overhead loop start between a call and its CALL_ARG_LOCATION
note.
Chen Gang [Tue, 29 Nov 2016 18:33:20 +0000 (18:33 +0000)]
re PR target/71331 (target-tilegx: nested-function-4.c: r10 is conflict which is both in function frame and in parameter.)
PR target/71331
* config/tilegx/tilegx.c (tilegx_function_profiler): Save r10
to stack before call mcount.
(tilegx_can_use_return_insn_p): Clean up code.
David Malcolm [Tue, 29 Nov 2016 16:25:01 +0000 (16:25 +0000)]
spellcheck bugfixes: don't offer the goal string as a suggestion
gcc/cp/ChangeLog:
PR c++/77922
* name-lookup.c (lookup_name_fuzzy): Filter out reserved words
that were filtered out by init_reswords.
gcc/ChangeLog:
PR c++/72774
PR c++/72786
PR c++/77922
PR c++/78313
* spellcheck.c (selftest::test_find_closest_string): Verify that
we don't offer the goal string as a suggestion.
* spellcheck.h (best_match::get_best_meaningful_candidate): Don't
offer the goal string as a suggestion.
gcc/testsuite/ChangeLog:
PR c++/72774
PR c++/72786
PR c++/77922
PR c++/78313
* g++.dg/spellcheck-c++-11-keyword.C: New test case.
* g++.dg/spellcheck-macro-ordering.C: New test case.
* g++.dg/spellcheck-pr78313.C: New test case.
Richard Biener [Tue, 29 Nov 2016 14:01:32 +0000 (14:01 +0000)]
tree-cfg.c (lower_phi_internal_fn): Do not look for further PHIs after a regular stmt.
2016-11-29 Richard Biener <rguenther@suse.de>
* tree-cfg.c (lower_phi_internal_fn): Do not look for further
PHIs after a regular stmt.
(stmt_starts_bb_p): PHIs not preceeded by a PHI or a label
start a new BB.
Martin Liska [Tue, 29 Nov 2016 13:20:00 +0000 (14:20 +0100)]
Make one extra BB to prevent PHI argument clash (PR
PR gcov-profile/78582
* gcc.dg/pr78582.c: New test.
PR gcov-profile/78582
* tree-profile.c (gimple_gen_time_profiler): Make one extra BB
to prevent PHI argument clash.
These testcases test that we generate the expected code for all of the
rl*i* instructions, that is, rotate-and-mask and rotate-and-mask-insert
for immediate rotation counts. All the testcases do rotate, shift left,
as well as shift right; if that results in an instruction that does not
exist the testcases generate a multiplication instead, so that we can
detect if this is handled properly.
Many 32-bit instructions zero-extend their result properly in 64-bit
mode, but the rs6000 port does not yet know. These testcases test the
status quo, so they will need updating when ever we handle this.
gcc/testsuite/
* gcc.target/powerpc/rldic-0.c: New testcase.
* gcc.target/powerpc/rldic-1.c: New testcase.
* gcc.target/powerpc/rldic-2.c: New testcase.
* gcc.target/powerpc/rldicl-0.c: New testcase.
* gcc.target/powerpc/rldicl-1.c: New testcase.
* gcc.target/powerpc/rldicl-2.c: New testcase.
* gcc.target/powerpc/rldicr-0.c: New testcase.
* gcc.target/powerpc/rldicr-1.c: New testcase.
* gcc.target/powerpc/rldicr-2.c: New testcase.
* gcc.target/powerpc/rldicx.h: New file.
* gcc.target/powerpc/rldimi-0.c: New testcase.
* gcc.target/powerpc/rldimi-1.c: New testcase.
* gcc.target/powerpc/rldimi-2.c: New testcase.
* gcc.target/powerpc/rldimi.h: New file.
* gcc.target/powerpc/rlwimi-0.c: New testcase.
* gcc.target/powerpc/rlwimi-1.c: New testcase.
* gcc.target/powerpc/rlwimi-2.c: New testcase.
* gcc.target/powerpc/rlwimi.h: New file.
* gcc.target/powerpc/rlwinm-0.c: New testcase.
* gcc.target/powerpc/rlwinm-1.c: New testcase.
* gcc.target/powerpc/rlwinm-2.c: New testcase.
* gcc.target/powerpc/rlwinm.h: New file.
change_zero_ext handles (zero_extend:M1 (subreg:M2 (reg:M1) ...))
already; this patch extends it to also deal with any
(zero_extend:M1 (subreg:M2 (reg:M3) ...)) where the subreg is not
paradoxical.
* combine.c (change_zero_ext): Also handle extends from a subreg
to a mode bigger than that of the operand of the subreg.
rs6000: Make deallocation of a large frame work (PR77687)
If we use ABI_V4 and we have a big stack frame, we end the epilogue
with a "mr 1,11" (or similar) instruction. This instruction however
has no dependencies on the earlier restores from stack (done via r11),
so sched2 can end up reordering the insns, which is bad because we
have no red zone so that you then restore from stack that is already
deallocated.
This fixes it by making that restore depend on the memory accesses.
PR target/77687
* config/rs6000/rs6000.c (rs6000_emit_stack_reset): Emit the
stack_restore_tie insn instead of stack_tie, for the SVR4 and
SPE ABIs.
* config/rs6000/rs6000.md (stack_restore_tie): New define_insn.
This patch changes spread_components to use a simpler algorithm that
puts prologue components as early as possible, and epilogue components
as late as possible. This allows better scheduling, and also saves a
bit of code size. The blocks that run with some specific component
enabled after this patch is a strict superset of those that had it
before the patch.
It does this by finding for every component the basic blocks where that
component is not needed on some path from the entry block (it reuses
head_components to store this), and similarly the blocks where the
component is not needed on some path to the exit block (or the exit can
not be reached from that block) (stored in tail_components). Blocks
that then are not in both of those two sets get the component active.
* shrink-wrap.c (init_separate_shrink_wrap): Do not clear
head_components and tail_components.
(spread_components): New algorithm.
(emit_common_tails_for_components): Clear head_components and
tail_components.
(insert_prologue_epilogue_for_components): Write extra output to the
dump file for sibcalls and abnormal exits.
combine: Make code after a new trap unreachable (PR78342)
Combine can turn a conditional trap into an unconditional trap. If it
does that it should make the code after it unreachable (an unconditional
trap should be the last insn in its bb, and that bb has no successors).
This patch seems to work. It is hard to be sure, this is very hard to
trigger. Quite a few other passes look like they need something similar
as well, but I don't see anything else handling it yet either.
PR rtl-optimization/78342
* combine.c: Include "cfghooks.h".
(try_combine): If we create an unconditional trap, break the basic
block in two just after it, and remove the edge between; also, set
the *new_direct_jump_p flag so that cleanup_cfg is run.
Jakub Jelinek [Mon, 28 Nov 2016 22:51:29 +0000 (23:51 +0100)]
re PR middle-end/78540 (ICE: in df_refs_verify, at df-scan.c:4062 with -O -march=core2)
PR middle-end/78540
* rtl.h (remove_reg_equal_equiv_notes): Return bool instead of void.
* rtlanal.c (remove_reg_equal_equiv_notes): Return true if any
note has been removed.
* postreload.c (reload_combine_recognize_pattern): If
remove_reg_equal_equiv_notes returns true, call df_notes_rescan.
Martin Sebor [Mon, 28 Nov 2016 21:41:41 +0000 (21:41 +0000)]
PR middle-end/78521 - [7 Regression] incorrect byte count in -Wformat-length...
PR middle-end/78521 - [7 Regression] incorrect byte count in -Wformat-length warning with non-constant width or precision
PR middle-end/78520 - missing warning for snprintf with size greater than INT_MAX
gcc/ChangeLog:
PR middle-end/78520
* gimple-ssa-sprintf.c (target_max_value): Remove.
(target_int_max, target_size_max): Use TYPE_MAX_VALUE.
(get_width_and_precision): New function.
(format_integer, format_floating, get_string_length, format_string):
Correct handling of width and precision with unknown value.
(format_directive): Add warning.
(pass_sprintf_length::compute_format_length): Allow for precision
to consist of a sole period with no asterisk or digits after it.
gcc/testsuite/ChangeLog:
PR middle-end/78520
* gcc.dg/tree-ssa/builtin-sprintf-5.c: Add test cases.
* gcc.dg/tree-ssa/builtin-sprintf-6.c: New test.
* gcc.dg/tree-ssa/builtin-sprintf-warn-1.c: Add test cases.
* gcc.dg/tree-ssa/builtin-sprintf-warn-3.c: Add test cases.
Jakub Jelinek [Mon, 28 Nov 2016 19:15:51 +0000 (20:15 +0100)]
re PR rtl-optimization/78546 (wrong code at -O2 and above)
PR rtl-optimization/78546
* simplify-rtx.c (neg_const_int): When negating most negative
number in mode wider than HOST_BITS_PER_WIDE_INT, use
simplify_const_unary_operation to produce CONST_DOUBLE or
CONST_WIDE_INT.
(simplify_plus_minus): Hanlde the case where neg_const_int
doesn't return a CONST_INT.
* gcc.dg/torture/pr78546-1.c: New test.
* gcc.dg/torture/pr78546-2.c: New test.
Jakub Jelinek [Mon, 28 Nov 2016 17:31:37 +0000 (18:31 +0100)]
re PR fortran/78298 (ICE in lookup_decl_in_outer_ctx, bei omp-low.c:4115)
PR fortran/78298
* tree-nested.c (convert_local_reference_stmt): After adding
shared (FRAME.NN) clause to omp parallel, task or target,
add it also to all outer omp parallel, task or target constructs.
Uros Bizjak [Mon, 28 Nov 2016 16:53:17 +0000 (17:53 +0100)]
i386.md (UNSPEC_KMASKOP): New.
* config/i386/i386.md (UNSPEC_KMASKOP): New.
(UNSPEC_KMOV): Remove.
(kmovw): Expand to plain HImode move.
(k<any_logic:code><mode>): Rename from *k<logic><mode>. Use
register_operand predicates. Tag pattern with UNSPEC_KMASKOP.
Remove corresponding clobber-removing splitter.
(*anddi_1): Remove mask register alternatives.
(*andsi_1): Ditto.
(*andhi_1): Ditto.
(*andqi_1): Ditto.
(*<any_or:code><mode>_1): Ditto.
(*<any_or:code>qi_1): Ditto.
(kandn<mode>): Use SWI1248_AVX512BW mode iterator. Remove
general register alternatives. Tag pattern with UNSPEC_KMASKOP.
Remove corresponding splitter to operation with general registers.
(*andn<SWI38:mode>): Rename from *bmi_andn_<mode>.
(*andn<SWI12:mode>): New pattern.
(*kxnor<mode>): Remove general register alternatives. Tag pattern
with UNSPEC_KMASKOP. Remove corresponding splitter to operation
with general registers.
(knot<mode>): New insn pattern.
(*one_cmpl<mode>2_1): Remove mask register alternatives.
(one_cmplqi2_1): Ditto.
(*k<any_lshift:code><mode>): Rename from *k<mshift><mode>3.
Tag pattern with UNSPEC_KMASKOP. Add mode attribute.
* config/i386/predicates.md (mask_reg_operand): Remove predicate.
* config/i386/sse.md (vec_unpacks_hi_hi): Update pattern
to generate kmaskop shift.
(vec_unpacks_hi_<mode>): Ditto.
* config/i386/i386-builtin.def (__builtin_ia32_kandhi):
Use CODE_FOR_kandhi.
(__builtin_ia32_knothi): Use CODE_FOR_knothi.
(__builtin_ia32_korhi): Use CODE_FOR_kiorhi.
(__builtin_ia32_kxorhi): Use CODE_FOR_kxorhi.
Jonas Hahnfeld [Mon, 28 Nov 2016 13:42:50 +0000 (13:42 +0000)]
libgomp: fix OpenMP offloading to NVPTX and correct location of generic affinity.c
2016-11-28 Jonas Hahnfeld <Hahnfeld@itc.rwth-aachen.de>
* config/linux/affinity.c [!HAVE_PTHREAD_AFFINITY_NP]: Include
../../affinity.c as fallback.
* config/nvptx/affinity.c: Delete to use fallback implementation.
Jakub Jelinek [Mon, 28 Nov 2016 10:01:30 +0000 (11:01 +0100)]
gimple-ssa-sprintf.c (build_intmax_type_nodes): Look at UINTMAX_TYPE rather than SIZE_TYPE.
* gimple-ssa-sprintf.c (build_intmax_type_nodes): Look at
UINTMAX_TYPE rather than SIZE_TYPE. Add gcc_unreachable if
intmax_t couldn't be determined.
(format_integer): Make {,u}intmax_type_node no longer static,
initialize them only when needed. For z and t use
signed_or_unsigned_type_for instead of assuming size_t and
ptrdiff_t have the same precision.
Paul Thomas [Sun, 27 Nov 2016 19:00:00 +0000 (19:00 +0000)]
re PR fortran/78474 ([F08] gfortran accepts invalid submodule syntax)
2016-11-27 Paul Thomas <pault@gcc.gnu.org>
PR fortran/78474
* module.c (gfc_match_submodule): If there is more than one
colon, it is a syntax error.
PR fortran/78331
* module.c (gfc_use_module): If an smod file does not exist it
is either because the module does not have a module procedure
interface or there is an error in the module.
2016-11-27 Paul Thomas <pault@gcc.gnu.org>
PR fortran/78474
* gfortran.dg/submodule_22.f08: New test.
PR fortran/78331
* gfortran.dg/submodule_21.f08: New test.
Iain Sandoe [Sun, 27 Nov 2016 15:07:22 +0000 (15:07 +0000)]
[Darwin] fix PR67710 by updating 'as' specs to handle newer assembler versions.
A/ Newer versions of ld64 check the min_version command, and newer versions of
the system assembler inserts this in response to "-mmacosx-version-min=" on
the assembler line. Unless one makes sensible versions, some object is bound
to conflict.
B/ Additionally, there's a difference in behaviour between "as" and "ld" when
presented with xx.yy.zz (ld truncates to xx.yy, as doesn't); net result is
that one needs to pass a truncated version to "as".
So (if the assembler supports minversion commands)
(a) provide a truncated minversion (as asm_macosx_version_min, which is a
driver-only var).
(b) pass this to "as"
(c) Update tests to determine 'HAVE_AS_MMACOSX_VERSION_MIN_OPTION'
(Rainer's patch)
(d) For some reason the testcases are "run" (it's not obvious they need to be,
they are checking compile-time issues)
- anyway, to preserve the status quo, I've left them as exec. However, the
minimum version that can be code-gened for is target-dependent (there are no
released x86 versions before 10.4, for example). To avoid conflicts where
the "as" is assuming some minimum, I've set the testversion to 10.5 (which
is supported by all the archs we have)
(e) We need to ensure that libgcc and crts are generated with a sufficiently
old minversion not to conflict.
Iain Sandoe [Sun, 27 Nov 2016 14:50:58 +0000 (14:50 +0000)]
[Darwin] Fix PR57438 by avoiding empty function bodies and trailing labels.
A.
Empty function bodies causes two problems for Darwin's linker (i) zero-length
FDEs and (ii) coincident label addresses that might point to items of
differing weakness.
B.
Trailing local labels can be problematic when they end a function because
similarly they might apparently point to a following weak function, leading
to the linker concluding that there's a pointer-diff to a weak symbol
(which is not allowed).
Both conditions arise from __builtin_unreachable() lowering to a barrier.
The solution for both is to emit some finite amount of code; in the case of A
a trap is emitted, in the case of B a nop.
gcc/
2016-11-27 Iain Sandoe <iain@codesourcery.com>
PR target/57438
* config/i386/i386.c (ix86_code_end): Note that we emitted code
where the function might otherwise appear empty for picbase thunks.
(ix86_output_function_epilogue): If we find a zero-sized function
assume that reaching it is UB and trap. If we find a trailing label
append a nop.
* config/rs6000/rs6000.c (rs6000_output_function_epilogue): If we
find a zero-sized function assume that reaching it is UB and trap.
If we find a trailing label, append a nop.
gcc/testsuite/
2016-11-27 Iain Sandoe <iain@codesourcery.com>
PR target/57438
* gcc.dg/pr57438-1.c: New Test.
* gcc.dg/pr57438-2.c: New Test.
Iain Sandoe [Sun, 27 Nov 2016 14:34:54 +0000 (14:34 +0000)]
[Darwin] Fix PR71767 - adjust the sections used where necessary.
(much) Older Darwin linkers needed separate sections marked "coalesce" to
allow for weak symbol coalescing. This has not been needed for some time
and is now deprecated, newer assemblers warn if the old coalesced sections
are used.
gcc/
2016-11-27 Iain Sandoe <iain@codesourcery.com>
PR target/71767
* config/darwin-sections.def (picbase_thunk_section): New.
* config/darwin.c (darwin_init_sections): Set up picbase thunk
section. (darwin_rodata_section, darwin_objc2_section,
machopic_select_section, darwin_asm_declare_constant_name,
darwin_emit_weak_or_comdat, darwin_function_section): Don’t use
coalesced with newer linkers.
(darwin_override_options): Decide on usage of coalesed sections
on the basis of the target linker version.
* config/darwin.h (MIN_LD64_NO_COAL_SECTS): New.
* config/darwin.opt (mtarget-linker): New.
* config/i386/i386.c (ix86_code_end): Do not force the thunks into
a coalesced section, instead use a thunks section.
Iain Sandoe [Sun, 27 Nov 2016 14:29:36 +0000 (14:29 +0000)]
[Darwin, config] Arrange for ld64 to be detected as Darwin's linker.
This is an initial patch in a series that converts Darwin's configury
to detect ld64 features, rather than the current process of hard-coding
them on target system version.
A ld64-compatible linker is currently required and assumed by Darwin.
If a DEFAULT_LINKER is set via --with-ld= then this will be tested to
see if it is ld64.
The ld64 version is determined for the chosen ld and this is exported for
use in setting a default value for -mtarget-linker (needed for run-time
code-gen changes to section choices).
The support for -rdynamic is converted to be detected at config time, or
by the ld64 version if that is found.
gcc/
2016-11-27 Iain Sandoe <iain@codesourcery.com>
PR target/71767
* configure.ac (with-ld64): New var, set for Darwin, set on
detection of ld64, gcc_cv_ld64_export_dynamic: New, New test.
* config/darwin.h: Use LD64_HAS_DYNAMIC export. DEF_LD64: New,
define.
* config/darwin10.h(DEF_LD64): Update for this target version.
* config/darwin12.h(LINK_GCC_C_SEQUENCE_SPEC): Remove rdynamic
test. (DEF_LD64): Update for this target version.
* configure: Regenerated.
* config.in: Regenerated.
Iain Sandoe [Sun, 27 Nov 2016 14:21:51 +0000 (14:21 +0000)]
[Darwin, ld64] Make PIC indirections and constant labels linker-visible.
Indirections:
If we have a situation like:
global_weak_symbol:
....
Lnon_weak_local:
....
ld64 will be unable to split this into two atoms (because the "L" makes
the second symbol 'invisible'). This means that legitimate direct accesses
to the second symbol will appear to be non-allowed direct accesses to an
atom of type weak, global which are not allowed.
To avoid this, we make the indirections have a leading 'l' (lower-case L)
which has a special meaning: linker can see this and use it to determine
atoms, but it is not placed into the final symbol table.
The implementation here is somewhat heavy-handed in that it will also mark
indirections to the __IMPORT,__pointers section the same way which is
really unnecessary, since ld64 _can_ split those into atoms as they are
fixed size. FIXME: determine if this is a penalty worth extra code to
fix.
Similarly, with:
.const
weak_global_constant:
....
LCxx:
...
ld64 can't split the second, causing a warning when it's directly
accessed.
gcc/
2016-11-27 Iain Sandoe <iain@codesourcery.com>
PR target/71767
* config/darwin.c (imachopic_indirection_name): Make data
section indirections linker-visible.
* config/darwin.h (ASM_GENERATE_INTERNAL_LABEL): Make local
constant labels linker-visible.
Tim Shen [Sun, 27 Nov 2016 00:32:04 +0000 (00:32 +0000)]
re PR libstdc++/78441 ([variant] variant_alternative doesn't allow cv qualifiers)
PR libstdc++/78441
* include/std/variant: Propagate cv qualifications to types returned
by variant_alternative.
* testsuite/20_util/variant/compile.cc: Tests.
Jakub Jelinek [Fri, 25 Nov 2016 17:12:29 +0000 (18:12 +0100)]
re PR rtl-optimization/78527 (ice on valid C code at -O3 in both 32-bit and 64-bit modes on x86_64-linux-gnu (internal compiler error: in smallest_mode_for_size, at stor-layout.c:364))
PR rtl-optimization/78527
* combine.c (make_compound_operation_int): Ignore LSHIFTRT with
out of bounds shift count.
Martin Liska [Fri, 25 Nov 2016 14:23:25 +0000 (15:23 +0100)]
Don't use priority {cd}tors if not supported by a target (PR
PR gcov-profile/78086
* g++.dg/gcov/pr16855.C: Clean up the test case.
* g++.dg/gcov/pr16855-priority.C: New test.
* coverage.c (build_init_ctor): Don't use priority {cd}tors if
not supported by a target. Set priority to 100 if possible.
(build_gcov_exit_decl): Likewise.
Paul Thomas [Fri, 25 Nov 2016 12:23:43 +0000 (12:23 +0000)]
[multiple changes]
2016-11-25 Andre Vehreschild <vehre@gcc.gnu.org>
Paul Thomas <pault@gcc.gnu.org>
PR fortran/78293
* trans-expr.c (gfc_conv_procedure_call): Prepend deallocation
of alloctable components to post, rather than adding to
se->post.
* trans-stmt.c (gfc_trans_allocate): Move deallocation of expr3
allocatable components so that all expr3s are visited.
2016-11-25 Paul Thomas <pault@gcc.gnu.org>
PR fortran/78293
* gfortran.dg/allocatable_function_10.f90: New test.
* gfortran.dg/class_array_15.f03: Increase builtin_free count
from 11 to 12.
Bin Cheng [Fri, 25 Nov 2016 11:45:43 +0000 (11:45 +0000)]
re PR middle-end/78507 (ICE: tree check: expected integer_type or enumeral_type or boolean_type or real_type or fixed_point_type, have pointer_type in int_fits_type_p, at tree.c:9153)
PR middle-end/78507
PR middle-end/78510
PR middle-end/78517
* match.pd ((cond (cmp (convert1? @1) @3) (convert2? @1) @2)): Use
cmp directly, rather than cmp_code. Initialize code to ERROR_MARK
and set it to result code if transformation is valid. Use code EQ
directly in last simplification case.
gcc/testsuite
PR middle-end/78507
PR middle-end/78510
PR middle-end/78517
* g++.dg/torture/pr78507.C: New test.
* gcc.dg/torture/pr78510.c: New test.
* gcc.dg/torture/pr78517.c: New test.
2016-11-25 Thomas Preud'homme <thomas.preudhomme@arm.com>
gcc/
PR tree-optimization/77673
* tree-ssa-math-opts.c (struct symbolic_number): Add new src field.
(init_symbolic_number): Initialize src field from src parameter.
(perform_symbolic_merge): Select most dominated statement as the
source statement. Set src field of resulting n structure from the
input src with the lowest address.
(find_bswap_or_nop): Rename source_stmt into ins_stmt.
(bswap_replace): Rename src_stmt into ins_stmt. Initially get source
of load from src field rather than insertion statement. Cancel
optimization if statement analyzed is not dominated by the insertion
statement.
(pass_optimize_bswap::execute): Rename src_stmt to ins_stmt. Compute
dominance information.
gcc/testsuite/
PR tree-optimization/77673
* gcc.dg/pr77673.c: New test.
James Greenhalgh [Fri, 25 Nov 2016 09:25:31 +0000 (09:25 +0000)]
[Patch i386] PR78509 - TARGET_C_EXCESS_PRECISION should not return
"unpredictable" for EXCESS_PRECISION_TYPE_STANDARD
gcc/
PR target/78509
* config/i386/i386.c (i386_excess_precision): Do not return
FLT_EVAL_METHOD_UNPREDICTABLE when "type" is
EXCESS_PRECISION_TYPE_STANDARD.
* target.def (excess_precision): Document that targets should
not return FLT_EVAL_METHOD_UNPREDICTABLE when "type" is
EXCESS_PRECISION_TYPE_STANDARD or EXCESS_PRECISION_TYPE_FAST.
Fix typo in first sentence.
* doc/tm.texi: Regenerate.
Richard Biener [Fri, 25 Nov 2016 08:59:28 +0000 (08:59 +0000)]
re PR tree-optimization/78396 (gcc.dg/vect/bb-slp-cond-1.c FAILs after fix for PR77848)
2016-11-25 Richard Biener <rguenther@suse.de>
PR tree-optimization/78396
* tree-vectorizer.c (vectorize_loops): When the if-converted
body contains masked loads or stores do not attempt to
basic-block-vectorize it.
The previous code processed the users of a stack slot in order of
decreasing size and allocated the slot based on the first user.
This seems a bit dangerous, since the ordering is based on the
mode of the biggest reference while the allocation is based also
on the size of the register itself (which I think could be larger).
That scheme doesn't scale well to polynomial sizes, since there's
no guarantee that the order of the sizes is known at compile time.
This patch instead records an upper bound on the size required
by all users of a slot. It also records the maximum alignment
requirement.
gcc/
2016-11-15 Richard Sandiford <richard.sandiford@arm.com>
Alan Hayward <alan.hayward@arm.com>
David Sherwood <david.sherwood@arm.com>
* function.h (spill_slot_alignment): Declare.
* function.c (spill_slot_alignment): New function.
* lra-spills.c (slot): Add align and size fields.
(assign_mem_slot): Use them in the call to assign_stack_local.
(add_pseudo_to_slot): Update the fields.
(assign_stack_slot_num_and_sort_pseudos): Initialise the fields.
Co-Authored-By: Alan Hayward <alan.hayward@arm.com> Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r242863
Set mode of decimal floats before calling layout_type
Previously decimal floating-point types were created and laid
out as binary floating-point types, then the caller changed
the mode to a decimal mode later. The problem with that
approach is that not all targets support an equivalent binary
floating-point mode. When they didn't, we would give the
type BLKmode and lay it out as a zero-sized type.
This probably had no effect in practice. If a target doesn't
support a binary mode then it's unlikely to support the decimal
equivalent either. However, with the stricter mode checking
added by later patches, we would assert if a scalar floating-
point type didn't have a scalar floating-point mode.
gcc/
2016-11-16 Richard Sandiford <richard.sandiford@arm.com>
Alan Hayward <alan.hayward@arm.com>
David Sherwood <david.sherwood@arm.com>
* stor-layout.c (layout_type): Allow the caller to set the mode of
a float type. Only choose one here if the mode is still VOIDmode.
* tree.c (build_common_tree_nodes): Set the type mode of decimal
floats before calling layout_type.
* config/rs6000/rs6000.c (rs6000_init_builtins): Likewise.
Co-Authored-By: Alan Hayward <alan.hayward@arm.com> Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r242862