liuhongt [Thu, 10 Aug 2023 08:26:13 +0000 (16:26 +0800)]
Support -m[no-]gather -m[no-]scatter to enable/disable vectorization for all gather/scatter instructions
Rename original use_gather to use_gather_8parts, Support
-mtune-ctrl={,^}use_gather to set/clear tune features
use_gather_{2parts, 4parts, 8parts}. Support the new option -mgather
as alias of -mtune-ctrl=, use_gather, ^use_gather.
Similar for use_scatter.
gcc/ChangeLog:
* config/i386/i386-builtins.cc
(ix86_vectorize_builtin_gather): Adjust for use_gather_8parts.
* config/i386/i386-options.cc (parse_mtune_ctrl_str):
Set/Clear tune features use_{gather,scatter}_{2parts, 4parts,
8parts} for -mtune-crtl={,^}{use_gather,use_scatter}.
* config/i386/i386.cc (ix86_vectorize_builtin_scatter): Adjust
for use_scatter_8parts
* config/i386/i386.h (TARGET_USE_GATHER): Rename to ..
(TARGET_USE_GATHER_8PARTS): .. this.
(TARGET_USE_SCATTER): Rename to ..
(TARGET_USE_SCATTER_8PARTS): .. this.
* config/i386/x86-tune.def (X86_TUNE_USE_GATHER): Rename to
(X86_TUNE_USE_GATHER_8PARTS): .. this.
(X86_TUNE_USE_SCATTER): Rename to
(X86_TUNE_USE_SCATTER_8PARTS): .. this.
* config/i386/i386.opt: Add new options mgather, mscatter.
liuhongt [Thu, 10 Aug 2023 03:41:39 +0000 (11:41 +0800)]
Software mitigation: Disable gather generation in vectorization for GDS affected Intel Processors.
For more details of GDS (Gather Data Sampling), refer to
https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/advisory-guidance/gather-data-sampling.html
After microcode update, there's performance regression. To avoid that,
the patch disables gather generation in autovectorization but uses
gather scalar emulation instead.
gcc/ChangeLog:
* config/i386/i386-options.cc (m_GDS): New macro.
* config/i386/x86-tune.def (X86_TUNE_USE_GATHER_2PARTS): Don't
enable for m_GDS.
(X86_TUNE_USE_GATHER_4PARTS): Ditto.
(X86_TUNE_USE_GATHER): Ditto.
Jason Merrill [Wed, 26 Jul 2023 14:39:34 +0000 (10:39 -0400)]
c++: member vs global template [PR106310]
For backward compatibility we still want to allow patterns like
this->A<T>::foo, but the template keyword in a qualified name is
specifically to specify that a dependent name is a template, so don't look
in the enclosing scope at all.
Also fix handling of dependent bases: if member lookup in the current
instantiation fails and we have dependent bases, the lookup is dependent.
We were already handling that for the case where lookup in the enclosing
scope also fails, but we also want it to affect that lookup itself.
PR c++/106310
gcc/cp/ChangeLog:
* parser.cc (cp_parser_template_name): Skip non-member
lookup after the template keyword.
(cp_parser_lookup_name): Pass down template_keyword_p.
Jason Merrill [Mon, 1 May 2023 14:57:20 +0000 (10:57 -0400)]
c++: array DMI and member fn [PR109666]
Here it turns out I also needed to adjust cfun when stepping out of the
member function to instantiate the DMI. But instead of adding that tweak,
let's unify with instantiate_body and just push_to_top_level instead of
trying to do the minimum subset of it. There was no measurable change in
compile time on stdc++.h.
This should also resolve 109506 without yet another tweak.
In the comments for PR108099 Jakub provided some testcases that demonstrated
that even before the regression noted in the patch we were getting the
semantics of this extension wrong: in the unsigned case we weren't producing
the corresponding standard unsigned type but another distinct one of the
same size, and in the signed case we were just dropping it on the floor and
not actually returning a signed type at all.
The former issue is fixed by using c_common_signed_or_unsigned_type instead
of unsigned_type_for, and the latter issue by adding a (signed_p &&
typedef_decl) case.
This patch introduces a failure on std/ranges/iota/max_size_type.cc due to
the latter issue, since the testcase expects 'signed rep_t' to do something
sensible, and previously we didn't. Now that we do, it exposes a bug in the
__max_diff_type::operator>>= handling of sign extension: when we evaluate
-1000 >> 2 in __max_diff_type we keep the MSB set, but leave the
second-most-significant bit cleared.
PR c++/108099
gcc/cp/ChangeLog:
* decl.cc (grokdeclarator): Don't clear typedef_decl after 'unsigned
typedef' pedwarn. Use c_common_signed_or_unsigned_type. Also
handle 'signed typedef'.
gcc/testsuite/ChangeLog:
* g++.dg/ext/int128-7.C: New test.
* g++.dg/ext/int128-8.C: New test.
* g++.dg/ext/unsigned-typedef2.C: New test.
* g++.dg/ext/unsigned-typedef3.C: New test.
liuhongt [Fri, 4 Aug 2023 01:27:39 +0000 (09:27 +0800)]
Workaround possible CPUID bug in Sandy Bridge.
Don't access leaf 7 subleaf 1 unless subleaf 0 says it is
supported via EAX.
Intel documentation says invalid subleaves return 0. We had been
relying on that behavior instead of checking the max sublef number.
It appears that some Sandy Bridge CPUs return at least the subleaf 0
EDX value for subleaf 1. Best guess is that this is a bug in a
microcode patch since all of the bits we're seeing set in EDX were
introduced after Sandy Bridge was originally released.
This is causing avxvnniint16 to be incorrectly enabled with
-march=native on these CPUs.
gcc/ChangeLog:
* common/config/i386/cpuinfo.h (get_available_features): Check
max_subleaf_level for valid subleaf before use CPUID.
Patrick Palka [Tue, 9 May 2023 19:06:34 +0000 (15:06 -0400)]
c++: noexcept-spec from nested class confusion [PR109761]
When late processing a noexcept-spec from a nested class after completion
of the outer class (since it's a complete-class context), we pass the wrong
class context to noexcept_override_late_checks -- the outer class type
instead of the nested class type -- which leads to bogus errors in the
below test.
This patch fixes this by making noexcept_override_late_checks obtain the
class context directly via DECL_CONTEXT instead of via an additional
parameter.
PR c++/109761
gcc/cp/ChangeLog:
* parser.cc (cp_parser_class_specifier): Don't pass a class
context to noexcept_override_late_checks.
(noexcept_override_late_checks): Remove 'type' parameter
and use DECL_CONTEXT of 'fndecl' instead.
Jakub Jelinek [Mon, 19 Dec 2022 10:24:55 +0000 (11:24 +0100)]
testsuite: Fix up pr107397.f90 test [PR107397]
The pr107397.f90 test FAILs for me, one problem was that the
added diagnostics has an indefinite article before BOZ, but
the test dg-error didn't. The other problem was that on the
other dg-error there was no space between the string and closing
}, so it was completely ignored and the error was an excess
error.
2022-12-19 Jakub Jelinek <jakub@redhat.com>
PR fortran/107397
* gfortran.dg/pr107397.f90: Adjust expected diagnostic wording and
add space between dg-error string and closing }.
Kewen Lin [Wed, 26 Jul 2023 08:42:29 +0000 (03:42 -0500)]
rs6000: Correct vsx operands output for xxeval [PR110741]
PR110741 exposes one issue that we didn't use the correct
character for vsx operands in output operand substitution,
consequently it can map to the wrong registers which hold
some unexpected values.
PR target/110741
gcc/ChangeLog:
* config/rs6000/vsx.md (define_insn xxeval): Correct vsx
operands output with "x".
Harald Anlauf [Sun, 16 Jul 2023 20:17:27 +0000 (22:17 +0200)]
Fortran: intrinsics and deferred-length character arguments [PR95947,PR110658]
gcc/fortran/ChangeLog:
PR fortran/95947
PR fortran/110658
* trans-expr.cc (gfc_conv_procedure_call): For intrinsic procedures
whose result characteristics depends on the first argument and which
can be of type character, the character length will not be deferred.
gcc/testsuite/ChangeLog:
PR fortran/95947
PR fortran/110658
* gfortran.dg/deferred_character_37.f90: New test.
testsuite: Require vectors of doubles for pr97428.c
The pr97428.c test assumes support for vectors of doubles, but some
targets only support vectors of floats, causing this test to fail with
such targets. Limit this test to targets that support vectors of
doubles then.
gcc/testsuite/
* gcc.dg/vect/pr97428.c: Limit to `vect_double' targets.
We rely on the "undefined" vals to have a specific value (from the earlier
REG_EQUAL note) but actual code generation doesn't ensure this (it doesn't
need to). That said, the issue isn't the constant folding per-se but that
we do not actually constant fold but register an equality that doesn't hold.
PR target/110206
gcc/ChangeLog:
* fwprop.cc (contains_paradoxical_subreg_p): Move to ...
* rtlanal.cc (contains_paradoxical_subreg_p): ... here.
* rtlanal.h (contains_paradoxical_subreg_p): Add prototype.
* cprop.cc (try_replace_reg): Do not set REG_EQUAL note
when the original source contains a paradoxical subreg.
Harald Anlauf [Tue, 11 Jul 2023 19:21:25 +0000 (21:21 +0200)]
Fortran: formal symbol attributes for intrinsic procedures [PR110288]
gcc/fortran/ChangeLog:
PR fortran/110288
* symbol.cc (gfc_copy_formal_args_intr): When deriving the formal
argument attributes from the actual ones for intrinsic procedure
calls, take special care of CHARACTER arguments that we do not
wrongly treat them formally as deferred-length.
gcc/testsuite/ChangeLog:
PR fortran/110288
* gfortran.dg/findloc_10.f90: New test.
Jonathan Wakely [Wed, 12 Jul 2023 13:40:19 +0000 (14:40 +0100)]
libstdc++: Check conversion from filesystem::path to wide strings [PR95048]
The testcase added for this bug only checks conversion from wide strings
on construction, but the fix also covered conversion to wide strings via
path::wstring(). Add checks for that, and u16string() and u32string().
Patrick Palka [Thu, 29 Jun 2023 20:10:18 +0000 (16:10 -0400)]
c++: NSDMI instantiation during overload resolution [PR110468]
Here we find ourselves instantiating the NSDMI for A<1>::m when
computing argument conversions during overload resolution, and
thus tf_conv is set. The flag causes mark_used for the constructor
used in the NSDMI to exit early and not instantiate its noexcept-spec,
which eventually leads to an ICE from nothrow_spec_p.
This patch fixes this by clearing any special tsubst flags during
instantiation of an NSDMI, since the result should be independent of
the context that requires the instantiation.
PR c++/110468
gcc/cp/ChangeLog:
* init.cc (maybe_instantiate_nsdmi_init): Mask out all
tsubst flags except for tf_warning_or_error.
d: Fix PR 108842: Cannot use enum array with -fno-druntime
Restrict the generating of CONST_DECLs for D manifest constants to just
scalars without pointers. It shouldn't happen that a reference to a
manifest constant has not been expanded within a function body during
codegen, but it has been found to occur in older versions of the D
front-end (PR98277), so if the decl of a non-scalar constant is
requested, just return its initializer as an expression.
PR d/108842
gcc/d/ChangeLog:
* decl.cc (DeclVisitor::visit (VarDeclaration *)): Only emit scalar
manifest constants.
(get_symbol_decl): Don't generate CONST_DECL for non-scalar manifest
constants.
* imports.cc (ImportVisitor::visit (VarDeclaration *)): New method.
gcc/testsuite/ChangeLog:
* gdc.dg/pr98277.d: Add more tests.
* gdc.dg/pr108842.d: New test.
Fix power10 fusion bug with prefixed loads, PR target/105325
This changes fixes PR target/105325. PR target/105325 is a bug where an
invalid lwa instruction is generated due to power10 fusion of a load
instruction to a GPR and an compare immediate instruction with the immediate
being -1, 0, or 1.
In some cases, when the load instruction is done, the GCC compiler would
generate a load instruction with an offset that was too large to fit into the
normal load instruction.
In particular, loads from the stack might originally have a small offset, so
that the load is not a prefixed load. However, after the stack is set up, and
register allocation has been done, the offset now is large enough that we would
have to use a prefixed load instruction.
The support for prefixed loads did not consider that patterns with a fused load
and compare might have a prefixed address. Without this support, the proper
prefixed load won't be generated.
In the original code, when the split2 pass is run after reload has finished the
ds_form_mem_operand predicate that was used for lwa and ld no longer returns
true. When the pattern was created, ds_form_mem_operand recognized the insn as
being valid since the offset was small. But after register allocation,
ds_form_mem_operand did not return true. Because it didn't return true, the
insn could not be split. Since the insn was not split and the prefix support
did not indicate a prefixed instruction was used, the wrong load is generated.
The solution involves:
1) Don't use ds_form_mem_operand for ld and lwa, always use
non_update_memory_operand.
2) Delete ds_form_mem_operand since it is no longer used.
3) Use the "YZ" constraints for ld/lwa instead of "m".
4) If we don't need to sign extend the lwa, convert it to lwz, and use
cmpwi instead of cmpdi. Adjust the insn name to reflect the code
generate.
5) Insure that the insn using lwa will be recognized as having a prefixed
operand (and hence the insn length will be 16 bytes instead of 8
bytes).
5a) Set the prefixed and maybe_prefix attributes to know that
fused_load_cmpi are also load insns;
5b) In the case where we are just setting CC and not using the memory
afterward, set the clobber to use a DI register, and put an
explicit sign_extend operation in the split;
5c) Set the sign_extend attribute to "yes" for lwa.
5d) 5a-5c are the things that prefixed_load_p in rs6000.cc checks to
ensure that lwa is treated as a ds-form instruction and not as
a d-form instruction (i.e. lwz).
6) Add a new test case for this case.
7) Adjust the insn counts in fusion-p10-ldcmpi.c. Because we are no
longer using ds_form_mem_operand, the ld and lwa instructions will fuse
x-form (reg+reg) addresses in addition ds-form (reg+offset or reg).
2023-06-23 Michael Meissner <meissner@linux.ibm.com>
gcc/
PR target/105325
* config/rs6000/genfusion.pl (gen_ld_cmpi_p10_one): Fix problems that
allowed prefixed lwa to be generated.
* config/rs6000/fusion.md: Regenerate.
* config/rs6000/predicates.md (ds_form_mem_operand): Delete.
* config/rs6000/rs6000.md (prefixed attribute): Add support for load
plus compare immediate fused insns.
(maybe_prefixed): Likewise.
This makes the code more readable, more digestible, more maintainable,
more extensible. That kind of thing. It does that by pulling things
apart a bit, but also making what stays together more cohesive lumps.
The original function was a bunch of loops and early-outs, and then
quite a bit of stuff done per iteration, with the iterations essentially
independent of each other. This patch moves the stuff done for one
iteration to a new _one function.
The second big thing is the stuff printed to the .md file is done in
"here documents" now, which is a lot more readable than having to quote
and escape and double-escape pieces of text. Whitespace inside the
here-document is significant (will be printed as-is), which is a bit
awkward sometimes, or might take some getting used to, but it is also
one of the benefits of using them.
Local variables are declared at first use (or close to first use).
There also shouldn't be many at all, often you can write easier to
read and manage code by omitting to name something that is hard to name
in the first place.
Finally some things are done in more typical, more modern, and tighter
Perl style, for example REs in "if"s or "qw" for lists of constants.
d: Fix core.volatile.volatileLoad discarded if result is unused
The first pass of code generation in the D front-end splits up all
compound expressions and discards expressions that have no side effects.
This included calls to the `volatileLoad' intrinsic if its result was
not used, causing such calls to be eliminated from the program.
We already set TREE_THIS_VOLATILE on the expression, however the
tree documentation says if this bit is set in an expression, so is
TREE_SIDE_EFFECTS. So set TREE_SIDE_EFFECTS on the expression too.
This prevents any early discarding from occuring.
PR d/110516
gcc/d/ChangeLog:
* intrinsics.cc (expand_volatile_load): Set TREE_SIDE_EFFECTS on the
expanded expression.
(expand_volatile_store): Likewise.
gcc/testsuite/ChangeLog:
* gdc.dg/torture/pr110516a.d: New test.
* gdc.dg/torture/pr110516b.d: New test.
d: Fix accesses of immutable arrays using constant index still bounds checked
Starts setting TREE_READONLY against specific kinds of VAR_DECLs, so
that the middle-end/optimization passes can more aggressively constant
fold D code that makes use of `immutable' or `const'.
PR d/110514
gcc/d/ChangeLog:
* decl.cc (get_symbol_decl): Set TREE_READONLY on certain kinds of
const and immutable variables.
* expr.cc (ExprVisitor::visit (ArrayLiteralExp *)): Set TREE_READONLY
on immutable dynamic array literals.
gcc/testsuite/ChangeLog:
* gdc.dg/pr110514a.d: New test.
* gdc.dg/pr110514b.d: New test.
* gdc.dg/pr110514c.d: New test.
* gdc.dg/pr110514d.d: New test.
Eric Botcazou [Fri, 30 Jun 2023 16:05:34 +0000 (18:05 +0200)]
Fix couple of endianness issues in fold_ctor_reference
fold_ctor_reference attempts to use a recursive local processing in order
to call native_encode_expr on the leaf nodes of the constructor, before
falling back to calling native_encode_initializer if this fails.
There are a couple of issues related to endianness present in it:
1) it does not specifically handle integral bit-fields; now these are left
justified on big-endian platforms so cannot be treated like ordinary fields.
2) it does not check that the constructor uses the native storage order.
gcc/
* gimple-fold.cc (fold_array_ctor_reference): Fix head comment.
(fold_nonarray_ctor_reference): Likewise. Specifically deal
with integral bit-fields.
(fold_ctor_reference): Make sure that the constructor uses the
native storage order.
If mem_addr points to a memory region with less than whole vector size
bytes of accessible memory and k is a mask that would prevent reading
the inaccessible bytes from mem_addr, add UNSPEC_MASKMOV to prevent
it to be transformed to any other whole memory access instructions.
gcc/ChangeLog:
PR rtl-optimization/110237
* config/i386/sse.md (<avx512>_store<mode>_mask): Refine with
UNSPEC_MASKMOV.
(maskstore<mode><avx512fmaskmodelower): Ditto.
(*<avx512>_store<mode>_mask): New define_insn, it's renamed
from original <avx512>_store<mode>_mask.
liuhongt [Tue, 20 Jun 2023 07:41:00 +0000 (15:41 +0800)]
Refine maskloadmn pattern with UNSPEC_MASKLOAD.
If mem_addr points to a memory region with less than whole vector size
bytes of accessible memory and k is a mask that would prevent reading
the inaccessible bytes from mem_addr, add UNSPEC_MASKLOAD to prevent
it to be transformed to vpblendd.
gcc/ChangeLog:
PR target/110309
* config/i386/sse.md (maskload<mode><avx512fmaskmodelower>):
Refine pattern with UNSPEC_MASKLOAD.
(maskload<mode><avx512fmaskmodelower>): Ditto.
(*<avx512>_load<mode>_mask): Extend mode iterator to
VI12HF_AVX512VL.
(*<avx512>_load<mode>): Ditto.
Hongyu Wang [Sun, 25 Jun 2023 01:50:21 +0000 (09:50 +0800)]
i386: Sync tune_string with arch_string for target attribute arch=*
For function with target attribute arch=*, current logic will set its
tune to -mtune from command line so all target_clones will get same
tuning flags which would affect the performance for each clone. Override
tune with arch if tune was not explicitly specified to get proper tuning
flags for target_clones.
gcc/ChangeLog:
* config/i386/i386-options.cc (ix86_valid_target_attribute_tree):
Override tune_string with arch_string if tune_string is not
explicitly specified.
Thomas Schwinge [Mon, 15 May 2023 18:00:07 +0000 (20:00 +0200)]
Support parallel testing in libgomp: fallback Perl 'flock' [PR66005]
Follow-up to commit 6c3b30ef9e0578509bdaf59c13da4a212fe6c2ba
"Support parallel testing in libgomp, part II [PR66005]"
("..., and enable if 'flock' is available for serializing execution testing"),
where we saw:
> On my Dell Precision 7530 laptop:
>
> $ uname -srvi
> Linux 5.15.0-71-generic #78-Ubuntu SMP Tue Apr 18 09:00:29 UTC 2023 x86_64
> $ grep '^model name' < /proc/cpuinfo | uniq -c
> 12 model name : Intel(R) Core(TM) i7-8850H CPU @ 2.60GHz
> $ nvidia-smi -L
> GPU 0: Quadro P1000 (UUID: GPU-e043973b-b52a-d02b-c066-a8fdbf64e8ea)
>
> ... [...]: case (c) standard configuration, no offloading
> configured, [...]
PR testsuite/66005
gcc/
* doc/install.texi: Document (optional) Perl usage for parallel
testing of libgomp.
libgomp/
* testsuite/lib/libgomp.exp: 'flock' through stdout.
* testsuite/flock: New.
* configure.ac (FLOCK): Point to that if no 'flock' available, but
'perl' is.
* configure: Regenerate.
Thomas Schwinge [Tue, 25 Apr 2023 21:53:12 +0000 (23:53 +0200)]
Support parallel testing in libgomp, part II [PR66005]
..., and enable if 'flock' is available for serializing execution testing.
Regarding the default of 19 parallel slots, this turned out to be a local
minimum for wall time when testing this on:
$ uname -srvi
Linux 4.2.0-42-generic #49~14.04.1-Ubuntu SMP Wed Jun 29 20:22:11 UTC 2016 x86_64
$ grep '^model name' < /proc/cpuinfo | uniq -c
32 model name : Intel(R) Xeon(R) CPU E5-2640 v3 @ 2.60GHz
... in two configurations: case (a) standard configuration, no offloading
configured, case (b) offloading for GCN and nvptx configured but no devices
available. For both cases, default plus '-m32' variant.
$ \time make check-target-libgomp RUNTESTFLAGS="--target_board=unix\{,-m32\}"
$ uname -srvi
Linux 5.15.0-71-generic #78-Ubuntu SMP Tue Apr 18 09:00:29 UTC 2023 x86_64
$ grep '^model name' < /proc/cpuinfo | uniq -c
12 model name : Intel(R) Core(TM) i7-8850H CPU @ 2.60GHz
$ nvidia-smi -L
GPU 0: Quadro P1000 (UUID: GPU-e043973b-b52a-d02b-c066-a8fdbf64e8ea)
... in two configurations: case (c) standard configuration, no offloading
configured, case (d) offloading for nvptx configured and device available.
For both cases, only default variant, no '-m32'.
$ \time make check-target-libgomp
Case (c), baseline; roughly half of case (a) (just one variant):
Worth noting is that with nvptx offloading, there is one execution test case
that times out ('libgomp.fortran/reverse-offload-5.f90'). This effectively
stalls progress for almost 5 min: quickly other executions test cases queue up
on the lock for all parallel slots. That's working as expected; just noting
this as it accordingly does skew the wall time numbers.
That is, if 'c.exp' executes first, it does successfully evaluate
'dg-require-effective-target openacc_cublas' -- and does cache this result (so
it isn't reevaluated for 'c++.exp'). However, for 'c++.exp' alone (that is,
without the 'c.exp' result cached), we run into:
spawn -ignore SIGHUP [xgcc] [...] -x c++ openacc_cublas2311907.c [...]
In file included from /usr/include/cuda_fp16.h:3673,
from /usr/include/cublas_api.h:75,
from /usr/include/cublas_v2.h:65,
from openacc_cublas2311907.c:3:
/usr/include/cuda_fp16.hpp:67:10: fatal error: utility: No such file or directory
We're missing include paths to C++/libstdc++ build-tree headers.
Fix this by using the mechanism introduced for Fortran in
r212268 (commit f707da16f714f7fe5a42391748212c84dfec639b) re
"libgomp.fortran/fortran.exp - add -fintrinsic-modules-path ${blddir}".
libgomp/
* testsuite/libgomp.c++/c++.exp: Use 'lang_include_flags' instead
of 'libstdcxx_includes'.
* testsuite/libgomp.oacc-c++/c++.exp: Likewise.
Paul E. Murphy [Thu, 22 Jun 2023 22:53:46 +0000 (17:53 -0500)]
go: Update usage of TARGET_AIX to TARGET_AIX_OS
TARGET_AIX is defined to a non-zero value on linux and maybe other
powerpc64le targets. This leads to unexpected behavior such as
dropping the .go_export section when linking a shared library
on linux/powerpc64le.
Instead, use TARGET_AIX_OS to toggle AIX specific behavior.
Fixes golang/go#60798.
2023-06-22 Paul E. Murphy <murphyp@linux.ibm.com>
gcc/go/
* go-backend.cc [TARGET_AIX]: Rename and update usage to TARGET_AIX_OS.
* go-lang.cc: Likewise.
But the document of mvzeroupper doesn't mention the insertion
required -O2 and above, it may confuse users when they explicitly
use -Os -mvzeroupper.
------------
mvzeroupper
Target Mask(VZEROUPPER) Save
Generate vzeroupper instruction before a transfer of control flow out of
the function.
------------
The patch moves flag_expensive_optimizations && !optimize_size to
ix86_option_override_internal. It makes -mvzeroupper independent of
optimization level, but still keeps the behavior of architecture
tuning(emit_vzeroupper) unchanged.
gcc/ChangeLog:
* config/i386/i386-features.cc (pass_insert_vzeroupper:gate):
Move flag_expensive_optimizations && !optimize_size to ..
* config/i386/i386-options.cc (ix86_option_override_internal):
.. this, it makes -mvzeroupper independent of optimization
level, but still keeps the behavior of architecture
tuning(emit_vzeroupper) unchanged.
Andrew Pinski [Tue, 27 Jun 2023 00:14:06 +0000 (17:14 -0700)]
Mark asm goto with outputs as volatile
The manual references asm goto as being implicitly volatile already
and that was done when asm goto could not have outputs. When outputs
were added to `asm goto`, only asm goto without outputs were still being
marked as volatile. Now some parts of GCC decide, removing the `asm goto`
is ok if the output is not used, though not updating the CFG (this happens
on both the RTL level and the gimple level). Since the biggest user of `asm goto`
is the Linux kernel and they expect them to be volatile (they use them to
copy to/from userspace), we should just mark the inline-asm as volatile.
Iain Buclaw [Mon, 26 Jun 2023 01:24:27 +0000 (03:24 +0200)]
d: Suboptimal codegen for __builtin_expect(cond, false)
Since PR96435, both boolean objects and expressions have been evaluated
in the following way.
(*(ubyte*)&obj_or_expr) & 1
It has been noted that sometimes this can cause the back-end to optimize
in non-obvious ways - in particular with __builtin_expect.
This @safe feature is now restricted to just when reading the value of a
bool field that comes from a union.
PR d/110359
gcc/d/ChangeLog:
* d-convert.cc (convert_for_rvalue): Only apply the @safe boolean
conversion to boolean fields of a union.
(convert_for_condition): Call convert_for_rvalue in the default case.
Iain Buclaw [Mon, 26 Jun 2023 00:29:46 +0000 (02:29 +0200)]
d: Fix crash in d/dmd/root/aav.d:127 dmd_aaGetRvalue from DsymbolTable::lookup
Backports patch from upstream dmd mainline for fixing PR110113.
The data being Mem.xrealloc'd contains many Array(T) fields, some of
which have self references in their data.ptr field thanks to the
smallarray optimization used by Array.
Naturally then, the memcpy from old GC data to new retains those self
referenced addresses, and the GC marks the old data as "free". Some time
later GC.malloc will return a pointer to said "free" data. So now we
have two GC references to the same memory. One that is treating the data
as an Array(VarDeclaration) in dmd.escape.escapeByStorage, and the other
as an AA in the symtab of a dmd.dsymbol.ScopeDsymbol.
Fix this memory corruption by not storing the data in a global variable
for reuse. If there are no more live references, the GC will free it.
PR d/110113
gcc/d/ChangeLog:
* dmd/escape.d (checkMutableArguments): Always allocate new buffer for
computing escapeBy.
projects / thirdparty / gcc.git / log
