The recent changes to error on mixing -march=i386 and -fcf-protection broke
bootstrap. This patch changes lib{atomic,gomp,itm} configury, so that it
only adds -march=i486 to flags if really needed (i.e. when 486 or later isn't
on by default already). Similarly, it will not use ifuncs if -mcx16
(or -march=i686 for 32-bit) is on by default.
2021-01-15 Jakub Jelinek <jakub@redhat.com>
PR target/70454
libatomic/
* configure.tgt: For i?86 and x86_64 determine if -march=i486 needs to
be added through preprocessor check on
__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4. Determine if try_ifunc is needed
based on preprocessor check on __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16
or __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8.
libgomp/
* configure.tgt: For i?86 and x86_64 determine if -march=i486 needs to
be added through preprocessor check on
__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4.
libitm/
* configure.tgt: For i?86 and x86_64 determine if -march=i486 needs to
be added through preprocessor check on
__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4.
This patch enables MVE vshr instructions for auto-vectorization. New
MVE patterns are introduced that take a vector of constants as second
operand, all constants being equal.
The existing mve_vshrq_n_<supf><mode> is kept, as it takes a single
immediate as second operand, and is used by arm_mve.h.
The vashr<mode>3 and vlshr<mode>3 expanders are moved fron neon.md to
vec-common.md, updated to rely on the normal expansion scheme to
generate shifts by immediate.
2020-12-03 Christophe Lyon <christophe.lyon@linaro.org>
gcc/
* config/arm/mve.md (mve_vshrq_n_s<mode>_imm): New entry.
(mve_vshrq_n_u<mode>_imm): Likewise.
* config/arm/neon.md (vashr<mode>3, vlshr<mode>3): Move to ...
* config/arm/vec-common.md: ... here.
gcc/testsuite/
* gcc.target/arm/simd/mve-vshr.c: Add tests for vshr.
This patch enables MVE vshlq instructions for auto-vectorization.
The existing mve_vshlq_n_<supf><mode> is kept, as it takes a single
immediate as second operand, and is used by arm_mve.h.
We move the vashl<mode>3 insn from neon.md to an expander in
vec-common.md, and the mve_vshlq_<supf><mode> insn from mve.md to
vec-common.md, adding the second alternative fron neon.md.
mve_vshlq_<supf><mode> will be used by a later patch enabling
vectorization for vshr, as a unified version of
ashl3<mode3>_[signed|unsigned] from neon.md. Keeping the use of unspec
VSHLQ enables to generate both 's' and 'u' variants.
It is not clear whether the neon_shift_[reg|imm]<q> attribute is still
suitable, since this insn is also used for MVE.
I kept the mve_vshlq_<supf><mode> naming instead of renaming it to
ashl3_<supf>_<mode> as discussed because the reference in
arm_mve_builtins.def automatically inserts the "mve_" prefix and I
didn't want to make a special case for this.
I haven't yet found why the v16qi and v8hi tests are not vectorized.
With dest[i] = a[i] << b[i] and:
{
int i;
unsigned int i.24_1;
unsigned int _2;
int16_t * _3;
short int _4;
int _5;
int16_t * _6;
short int _7;
int _8;
int _9;
int16_t * _10;
short int _11;
unsigned int ivtmp_42;
unsigned int ivtmp_43;
<bb 2> [local count: 119292720]:
<bb 3> [local count: 954449105]:
i.24_1 = (unsigned int) i_23;
_2 = i.24_1 * 2;
_3 = a_15(D) + _2;
_4 = *_3;
_5 = (int) _4;
_6 = b_16(D) + _2;
_7 = *_6;
_8 = (int) _7;
_9 = _5 << _8;
_10 = dest_17(D) + _2;
_11 = (short int) _9;
*_10 = _11;
i_19 = i_23 + 1;
ivtmp_42 = ivtmp_43 - 1;
if (ivtmp_42 != 0)
goto <bb 5>; [87.50%]
else
goto <bb 4>; [12.50%]
<bb 5> [local count: 835156386]:
goto <bb 3>; [100.00%]
<bb 4> [local count: 119292720]:
return;
}
the vectorizer says:
mve-vshl.c:37:96: note: ==> examining statement: _5 = (int) _4;
mve-vshl.c:37:96: note: vect_is_simple_use: operand *_3, type of def: internal
mve-vshl.c:37:96: note: vect_is_simple_use: vectype vector(8) short int
mve-vshl.c:37:96: missed: conversion not supported by target.
mve-vshl.c:37:96: note: vect_is_simple_use: operand *_3, type of def: internal
mve-vshl.c:37:96: note: vect_is_simple_use: vectype vector(8) short int
mve-vshl.c:37:96: note: vect_is_simple_use: operand *_3, type of def: internal
mve-vshl.c:37:96: note: vect_is_simple_use: vectype vector(8) short int
mve-vshl.c:37:117: missed: not vectorized: relevant stmt not supported: _5 = (int) _4;
mve-vshl.c:37:96: missed: bad operation or unsupported loop bound.
mve-vshl.c:37:96: note: ***** Analysis failed with vector mode V8HI
2020-12-03 Christophe Lyon <christophe.lyon@linaro.org>
gcc/
* config/arm/mve.md (mve_vshlq_<supf><mode>): Move to
vec-commond.md.
* config/arm/neon.md (vashl<mode>3): Delete.
* config/arm/vec-common.md (mve_vshlq_<supf><mode>): New.
(vasl<mode>3): New expander.
gcc/testsuite/
* gcc.target/arm/simd/mve-vshl.c: Add tests for vshl.
Avoid advancing to the next stmt when inserting at region boundary
and deal with a vector def being not the only child.
2021-01-15 Richard Biener <rguenther@suse.de>
PR tree-optimization/98685
* tree-vect-slp.c (vect_schedule_slp_node): Refactor handling
of vector extern defs.
* gcc.dg/vect/bb-slp-pr98685.c: New testcase.
I ran sed script late over the tests which accidentally
introduced a syntax error in the tests.
This fixes it.
Committed under the obvious rule.
gcc/testsuite/ChangeLog:
* gcc.dg/vect/complex/complex-mla-template.c: Fix sed.
* gcc.dg/vect/complex/complex-mls-template.c: Likewise.
This is the code that parses an embedcfg file, which is a JSON file
created by the go command when it sees go:embed directives. This code
is not yet called, and does not yet do anything. It's being sent as a
separate CL to isolate just the JSON parsing code.
* Make-lang.in (GO_OBJS): Add go/embed.o.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/281532
I removed the "Alpha" warning from the JIT wiki page on
2020-05-18:
https://gcc.gnu.org/wiki/JIT?action=diff&rev1=47&rev2=48
but forgot to remove it from the documentation, which this
patch does.
gcc/jit/ChangeLog:
* docs/cp/index.rst: Remove "Alpha" warning.
* docs/index.rst: Likewise.
* docs/_build/texinfo/libgccjit.texi: Regenerate
This function had two different local variables for TREE_TYPE (field), one
of which shadowed a parameter, and wasn't using them consistently.
gcc/cp/ChangeLog:
* typeck2.c (process_init_constructor_record): Use fldtype
variable consistently.
If fancy_abort is called before the diagnostic subsystem is initialized,
internal_error will crash internally in a way that prevents a useful
message reaching the user.
This can happen with libgccjit in the case of gcc_assert failures
that occur outside of the libgccjit mutex that guards the rest of
gcc's state, including global_dc (when global_dc may not be
initialized yet, or might be in use by another thread).
I tried a few approaches to fixing this as noted in PR jit/98586
e.g. using a temporary diagnostic_context and initializing it for
the call to internal_error, however the more code that runs, the
more chance there is for other errors to occur.
The best fix appears to be to simply fall back to a minimal abort
implementation that only relies on i18n, as implemented by this
patch.
gcc/ChangeLog:
PR jit/98586
* diagnostic.c (diagnostic_kind_text): Break out this array
from...
(diagnostic_build_prefix): ...here.
(fancy_abort): Detect when diagnostic_initialize has not yet been
called and fall back to a minimal implementation of printing the
ICE, rather than segfaulting in internal_error.
GCC has had the ability to emit fix-it hints in machine-readable form
since GCC 7 via -fdiagnostics-parseable-fixits and
-fdiagnostics-generate-patch.
The former emits additional specially-formatted lines to stderr; the
option and its format were directly taken from a pre-existing option
in clang.
Ideally this could be used by IDEs so that the user can select specific
fix-it hints and have the IDE apply them to the user's source code
(perhaps turning them into clickable elements, perhaps with an
"Apply All" option, etc). Eclipse CDT has supported this option in
this way for a few years:
https://bugs.eclipse.org/bugs/show_bug.cgi?id=497670
As a user of Emacs I would like Emacs to support such a feature.
https://debbugs.gnu.org/cgi/bugreport.cgi?bug=25987 tracks supporting
GCC fix-it output in Emacs. The discussion there identifies two issues
with the existing option:
(a) columns in the output are specified as byte-offsets within the
line (for exact compatibility with the option in clang), whereas emacs
would prefer to consume them as what GCC 11 calls "display columns".
https://gcc.gnu.org/onlinedocs/gcc/Diagnostic-Message-Formatting-Options.html#index-fdiagnostics-column-unit
(b) injecting a command-line option into the build is a fiddly manual
step, varying between build systems. It's far easier for the
user if Emacs simply sets an environment variable when compiling,
GCC uses this to enable the option if it recognizes the value, and
the emacs compilation buffer decodes the additional lines of output
and adds appropriate widgets. In some ways it is a workaround for
not having a language server. Doing it this way means that for the
various combinations of older and newer GCC and older and newer Emacs
that a sufficiently modern combination of both can automatically
support the rich fix-it UI, whereas other combinations will either
not provide the envvar, or silently ignore it, gracefully doing
nothing extra.
Hence this patch adds a new GCC_EXTRA_DIAGNOSTIC_OUTPUT environment
variable to GCC which enables output of machine-parseable fix-it hints.
GCC_EXTRA_DIAGNOSTIC_OUTPUT=fixits-v1 is equivalent to the existing
-fdiagnostics-parseable-fixits option.
GCC_EXTRA_DIAGNOSTIC_OUTPUT=fixits-v2 is the same, but changes the
column output mode to "display columns" rather than bytes, as
required by Emacs.
The discussion in that Emacs bug has some concerns about the encoding
of these lines, and, indeed, the encoding of GCC's stderr in general:
currently we emit a mixture of bytes and UTF-8; I believe we emit
filenames as bytes, diagnostic messages as UTF-8, and quote source code
in the original encoding (PR other/93067 covers converting it to UTF-8 on
output). This patch prints octal-escaped bytes for bytes within
filenames and replacement text that aren't printable (as per
-fdiagnostics-parseable-fixits).
gcc/ChangeLog:
* diagnostic.c (diagnostic_initialize): Eliminate
parseable_fixits_p in favor of initializing extra_output_kind from
GCC_EXTRA_DIAGNOSTIC_OUTPUT.
(convert_column_unit): New function, split out from...
(diagnostic_converted_column): ...this.
(print_parseable_fixits): Add "column_unit" and "tabstop" params.
Use them to call convert_column_unit on the column values.
(diagnostic_report_diagnostic): Eliminate conditional on
parseable_fixits_p in favor of a switch statement on
extra_output_kind, passing the appropriate values to the new
params of print_parseable_fixits.
(selftest::test_print_parseable_fixits_none): Update for new
params of print_parseable_fixits.
(selftest::test_print_parseable_fixits_insert): Likewise.
(selftest::test_print_parseable_fixits_remove): Likewise.
(selftest::test_print_parseable_fixits_replace): Likewise.
(selftest::test_print_parseable_fixits_bytes_vs_display_columns):
New.
(selftest::diagnostic_c_tests): Call it.
* diagnostic.h (enum diagnostics_extra_output_kind): New.
(diagnostic_context::parseable_fixits_p): Delete field in favor
of...
(diagnostic_context::extra_output_kind): ...this new field.
* doc/invoke.texi (Environment Variables): Add
GCC_EXTRA_DIAGNOSTIC_OUTPUT.
* opts.c (common_handle_option): Update handling of
OPT_fdiagnostics_parseable_fixits for change to diagnostic_context
fields.
gcc/testsuite/ChangeLog:
* gcc.dg/plugin/diagnostic-test-show-locus-GCC_EXTRA_DIAGNOSTIC_OUTPUT-fixits-v1.c:
New file.
* gcc.dg/plugin/diagnostic-test-show-locus-GCC_EXTRA_DIAGNOSTIC_OUTPUT-fixits-v2.c:
New file.
* gcc.dg/plugin/plugin.exp (plugin_test_list): Add them.
This adds the initial tests for the complex mul, mls and mla.
These will be enabled in the commits that add the optabs.
Committed as obvious variations of existing tests.
gcc/testsuite/ChangeLog:
* gcc.dg/vect/complex/complex-mla-template.c: New test.
* gcc.dg/vect/complex/complex-mls-template.c: New test.
* gcc.dg/vect/complex/complex-mul-template.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-mla-double.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-mla-float.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-mla-half-float.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-mls-double.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-mls-float.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-mls-half-float.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-mul-double.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-mul-float.c: New test.
* gcc.dg/vect/complex/fast-math-bb-slp-complex-mul-half-float.c: New test.
* gcc.dg/vect/complex/fast-math-complex-mla-double.c: New test.
* gcc.dg/vect/complex/fast-math-complex-mla-float.c: New test.
* gcc.dg/vect/complex/fast-math-complex-mla-half-float.c: New test.
* gcc.dg/vect/complex/fast-math-complex-mls-double.c: New test.
* gcc.dg/vect/complex/fast-math-complex-mls-float.c: New test.
* gcc.dg/vect/complex/fast-math-complex-mls-half-float.c: New test.
* gcc.dg/vect/complex/fast-math-complex-mul-double.c: New test.
* gcc.dg/vect/complex/fast-math-complex-mul-float.c: New test.
* gcc.dg/vect/complex/fast-math-complex-mul-half-float.c: New test.
This introduces a common class complex_operations_pattern which encapsulates
the complex add, mul, fma and fms pattern in such a way so that the first match
is shared.
gcc/ChangeLog:
* tree-vect-slp-patterns.c (class complex_operations_pattern,
complex_operations_pattern::matches,
complex_operations_pattern::recognize,
complex_operations_pattern::build): New.
(slp_patterns): Use it.
This introduces a post processing step for the pattern matcher to flatten
permutes introduced by the complex multiplications patterns.
This performs a blend early such that SLP is not cancelled by the LOAD_LANES
permute. This is a temporary workaround to the fact that loads are not CSEd
during building and is required to produce efficient code.
gcc/ChangeLog:
* tree-vect-slp.c (optimize_load_redistribution_1): New.
(optimize_load_redistribution, vect_is_slp_load_node): New.
(vect_match_slp_patterns): Use it.
This applies the same feedback received for MUL and the rest to
ADD which was already committed. In short it elides the intermediate
nodes vec and avoids the use of truncate on the SLP child.
gcc/ChangeLog:
* tree-vect-slp-patterns.c (complex_add_pattern::build):
Elide nodes.
I've been implementing a PyGTK viewer for the output of
-fdump-analyzer-json, to help me debug analyzer issues:
https://github.com/davidmalcolm/gcc-analyzer-viewer
The viewer is very much just a work in progress.
This patch adds some fields that were missing from the dump, and
fixes some mistakes I spotted whilst working on the viewer.
gcc/analyzer/ChangeLog:
* engine.cc (strongly_connected_components::to_json): New.
(worklist::to_json): New.
(exploded_graph::to_json): JSON-ify the worklist.
* exploded-graph.h (strongly_connected_components::to_json): New
decl.
(worklist::to_json): New decl.
* store.cc (store::to_json): Fix comment.
* supergraph.cc (supernode::to_json): Fix reference to
"returning_call" in comment. Add optional "fun" to JSON.
(edge_kind_to_string): New.
(superedge::to_json): Add "kind" to JSON.
This test was failing in C++11 because variable templates are only
available in C++14.
gcc/testsuite/ChangeLog:
* g++.dg/template/pr98372.C: Only run in C++14 and up.
Substrings were not reduced early enough for use in initializations,
such as DATA statements. Add an early simplification for substrings
with constant starting and ending points.
gcc/fortran/ChangeLog:
* gfortran.h (gfc_resolve_substring): Add prototype.
* primary.c (match_string_constant): Simplify substrings with
constant starting and ending points.
* resolve.c: Rename resolve_substring to gfc_resolve_substring.
(gfc_resolve_ref): Use renamed function gfc_resolve_substring.
gcc/testsuite/ChangeLog:
* substr_10.f90: New test.
* substr_9.f90: New test.
We get occasional failures of 30_threads/future/members/poll.cc
on some platforms whose high resolution clock doesn't have such a high
resolution; wait_for_0 ends up as 0, and then some asserts fail as
intervals measured as longer than zero are tested for less than
several times zero.
This patch adds some calibration in the iteration count to set a
measurable base time interval with some additional margin.
for libstdc++-v3/ChangeLog
* testsuite/30_threads/future/members/poll.cc: Calibrate
iteration count.
The sigsetjmp analyzer tests use jmp_buf in sigsetjmp and siglongjmp
calls. Not every system that supports sigsetjmp uses the same data
structure for setjmp and sigsetjmp, which results in type mismatches.
This patch changes the tests to use sigjmp_buf, that is the
POSIX-specific type for use with sigsetjmp and siglongjmp.
for gcc/testsuite/ChnageLog
* gcc.dg/analyzer/sigsetjmp-5.c: Use sigjmp_buf.
* gcc.dg/analyzer/sigsetjmp-6.c: Likewise.
The getpass function is not available on all systems; and not
necessarily declared in unistd.h, as expected by the sensitive-1
analyzer test.
Since this is a compile-only test, it doesn't really matter if the
function is defined in the system libraries. All we need is a
declaration, to avoid warnings from calling an undeclared function.
This patch adds the declaration, in a way that is most unlikely to
conflict with any existing declaration.
for gcc/testsuite/ChangeLog
* gcc.dg/analyzer/sensitive-1.c: Declare getpass.
Similar to nvptx offloading, see PR65099 "nvptx offloading: hard-coded 64-bit
assumptions".
gcc/
* config/gcn/mkoffload.c (main): Create an offload image only in
64-bit configurations.
During error recovery after an invalid derived type specification it was
possible to try to resolve an invalid array specification. We now skip
this if the component has the ALLOCATABLE or POINTER attribute and the
shape is not deferred.
gcc/fortran/ChangeLog:
PR fortran/98661
* resolve.c (resolve_component): Derived type components with
ALLOCATABLE or POINTER attribute shall have a deferred shape.
gcc/testsuite/ChangeLog:
PR fortran/98661
* gfortran.dg/pr98661.f90: New test.
During error recovery after an invalid derived type specification it was
possible to try to resolve an invalid array specification. We now skip
this if the component has the ALLOCATABLE or POINTER attribute and the
shape is not deferred.
gcc/fortran/ChangeLog:
PR fortran/98661
* resolve.c (resolve_component): Derived type components with
ALLOCATABLE or POINTER attribute shall have a deferred shape.
gcc/testsuite/ChangeLog:
PR fortran/98661
* gfortran.dg/pr98661.f90: New test.
As recently again discussed in <https://gcc.gnu.org/PR97436> "[nvptx] -m32
support", nvptx offloading other than for 64-bit host has never been
implemented, tested, supported. So we simply should buildn't the nvptx libgomp
plugin in this case.
This avoids build problems if, for example, in a (standard) bi-arch
x86_64-pc-linux-gnu '-m64'/'-m32' build, libcuda is available only in a 64-bit
variant but not in a 32-bit one, which, for example, is the case if you build
GCC against the CUDA toolkit's 'stubs/libcuda.so' (see
<https://stackoverflow.com/a/52784819>).
This amends PR65099 commit a92defdab7 (r225560)
"[nvptx offloading] Only 64-bit configurations are currently supported" to
match the way we're doing this for the HSA/GCN plugins.
libgomp/
PR libgomp/65099
* plugin/configfrag.ac (PLUGIN_NVPTX): Restrict to supported
configurations.
* configure: Regenerate.
* plugin/plugin-nvptx.c (nvptx_get_num_devices): Remove 64-bit
check.
Fix ordering problem on Windows targets where filesystem_error was used
before being defined.
libstdc++-v3/ChangeLog:
PR libstdc++/98471
* include/bits/fs_path.h (__throw_conversion_error): New
function to throw or abort on character conversion errors.
(__wstr_from_utf8): Move definition after filesystem_error has
been defined. Use __throw_conversion_error.
(path::_S_convert<_EcharT>): Use __throw_conversion_error.
(path::_S_str_convert<_CharT, _Traits, _Allocator>): Likewise.
(path::u8string): Likewise.
This fixes an infinite loop one could see for:
git show b87ec922c4 | ./contrib/mklog.py
contrib/ChangeLog:
* mklog.py: Fix infinite loop for unsupported files.
-fcf-protection with CF_BRANCH inserts ENDBR32 at function entries.
ENDBR32 is NOP only on 64-bit processors and 32-bit TARGET_CMOV
processors. Issue an error for -fcf-protection with CF_BRANCH when
compiling for 32-bit non-TARGET_CMOV targets.
gcc/
PR target/98667
* config/i386/i386-options.c (ix86_option_override_internal):
Issue an error for -fcf-protection with CF_BRANCH when compiling
for 32-bit non-TARGET_CMOV targets.
gcc/testsuite/
PR target/98667
* gcc.target/i386/pr98667-1.c: New file.
* gcc.target/i386/pr98667-2.c: Likewise.
* gcc.target/i386/pr98667-3.c: Likewise.
This improves dependence analysis on refs that access the same
array but with different typed but same sized accesses. That's
obviously safe for the case of types that cannot have any
access function based off them. For the testcase this is
signed short vs. unsigned short.
2021-01-14 Richard Biener <rguenther@suse.de>
PR tree-optimization/98674
* tree-data-ref.c (base_supports_access_fn_components_p): New.
(initialize_data_dependence_relation): For two bases without
possible access fns resort to type size equality when determining
shape compatibility.
* gcc.dg/vect/pr98674.c: New testcase.
Also pass -mpreferred-stack-boundary=4 -mno-stackrealign to avoid
disabling STV by:
/* Disable STV if -mpreferred-stack-boundary={2,3} or
-mincoming-stack-boundary={2,3} or -mstackrealign - the needed
stack realignment will be extra cost the pass doesn't take into
account and the pass can't realign the stack. */
if (ix86_preferred_stack_boundary < 128
|| ix86_incoming_stack_boundary < 128
|| opts->x_ix86_force_align_arg_pointer)
opts->x_target_flags &= ~MASK_STV;
PR target/98676
* gcc.target/i386/pr95021-1.c: Add -mpreferred-stack-boundary=4
-mno-stackrealign.
* gcc.target/i386/pr95021-3.c: Likewise.
The patch adding these files was approved in 2020 but it wasn't
committed until 2021, so the copyright years were not updated along with
the years in all the existing files.
libstdc++-v3/ChangeLog:
* include/std/barrier: Update copyright years. Fix whitespace.
* include/std/version: Fix whitespace.
* testsuite/30_threads/barrier/1.cc: Update copyright years.
* testsuite/30_threads/barrier/2.cc: Likewise.
* testsuite/30_threads/barrier/arrive.cc: Likewise.
* testsuite/30_threads/barrier/arrive_and_drop.cc: Likewise.
* testsuite/30_threads/barrier/arrive_and_wait.cc: Likewise.
* testsuite/30_threads/barrier/completion.cc: Likewise.
I flubbed an application of De Morgan's law. Let's just express the
logic directly and let the compiler figure it out. This bug made it
look like pr52830 was fixed, but it is not.
PR c++/98372
gcc/cp/
* tree.c (cp_tree_equal): Correct map_context logic.
gcc/testsuite/
* g++.dg/cpp0x/constexpr-52830.C: Restore dg-ice
* g++.dg/template/pr98372.C: New.
I've made two mistakes in the *sse4_1_zero_extend* define_insn_and_split
patterns. One is that when it uses vector_operand, it should use Bm rather
than m constraint, and the other one is that because it is a post-reload
splitter it needs isa attribute to select which alternatives are valid for
which ISAs. Sorry for messing this up.
2021-01-14 Jakub Jelinek <jakub@redhat.com>
PR target/98670
* config/i386/sse.md (*sse4_1_zero_extendv8qiv8hi2_3,
*sse4_1_zero_extendv4hiv4si2_3, *sse4_1_zero_extendv2siv2di2_3):
Use Bm instead of m for non-avx. Add isa attribute.
* gcc.target/i386/pr98670.c: New test.
This patch optimizes two GIMPLE operations into just one.
As mentioned in the PR, there is some risk this might create more expensive
constants, but sometimes it will make them on the other side less expensive,
it really depends on the exact value.
And if it is an important issue, we should do it in md or during expansion.
2021-01-14 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/96688
* match.pd (~(X >> Y) -> ~X >> Y): New simplification if
~X can be simplified.
* gcc.dg/tree-ssa/pr96688.c: New test.
* gcc.dg/tree-ssa/reassoc-37.c: Adjust scan-tree-dump regex.
* gcc.target/i386/pr66821.c: Likewise.
At the moment, if we use only one vector of an LD4 result,
we'll treat the LD4 as having the cost of a single load.
But all 4 loads and any associated permutes take place
regardless of which results are actually used.
This patch therefore counts the cost of unused LOAD_LANES
results against the first statement in a group. An alternative
would be to multiply the ncopies of the first stmt by the group
size and treat other stmts in the group as having zero cost,
but I thought that might be more surprising when reading dumps.
gcc/
* tree-vect-stmts.c (vect_model_load_cost): Account for unused
IFN_LOAD_LANES results.
gcc/testsuite/
* gcc.target/aarch64/sve/cost_model_11.c: New test.
* gcc.target/aarch64/sve/mask_struct_load_5.c: Use
-fno-vect-cost-model.
Turns out __builtin_convertvector is not as good a fit for the widening
and narrowing intrinsics as I had hoped.
During the veclower phase we lower most of it to bitfield operations and
hope DCE cleans it back up into
vector pack/unpack and extend operations. I received reports that in
more complex cases GCC fails to do that
and we're left with many vector extract operations that clutter the
output.
I think veclower can be improved on that front, but for GCC 10 I'd like
to just implement these builtins
with a good old RTL builtin rather than inline asm.
gcc/
* config/aarch64/aarch64-simd.md (aarch64_<su>xtl<mode>):
Define.
(aarch64_xtn<mode>): Likewise.
* config/aarch64/aarch64-simd-builtins.def (sxtl, uxtl, xtn):
Define
builtins.
* config/aarch64/arm_neon.h (vmovl_s8): Reimplement using
builtin.
(vmovl_s16): Likewise.
(vmovl_s32): Likewise.
(vmovl_u8): Likewise.
(vmovl_u16): Likewise.
(vmovl_u32): Likewise.
(vmovn_s16): Likewise.
(vmovn_s32): Likewise.
(vmovn_s64): Likewise.
(vmovn_u16): Likewise.
(vmovn_u32): Likewise.
(vmovn_u64): Likewise.
