Roman Lebedev [Tue, 25 Aug 2020 07:25:34 +0000 (10:25 +0300)]
[InstCombine] PHI-of-insertvalues -> insertvalue-of-PHI's
As per statistic, this happens pretty exceedingly rare,
but i have seen it in exactly the situations the
Phi-aware aggregate reconstruction would have handled,
eventually, and allowed invoke -> call fold later on.
So while this might be something that other fold
will have to learn about, i believe we should be
doing this transform in general.
Here, we are okay with adding two PHI's to get both the base aggregate,
and the inserted value. I'm not sure it makes much sense to restrict
it to a single phi (to just the inserted value?), because originally
we'd be receiving the final aggregate already..
llvm test-suite + RawSpeed:
```
| statistic name | baseline | proposed | Δ | % | \|%\| |
|--------------------------------------------|-----------|-----------|-----:|-------:|------:|
| instcombine.NumPHIsOfInsertValues | 0 | 12 | 12 | 0.00% | 0.00% |
| asm-printer.EmittedInsts | 8926643 | 8926595 | -48 | 0.00% | 0.00% |
| instcombine.NumCombined | 3846614 | 3846640 | 26 | 0.00% | 0.00% |
| instcombine.NumConstProp | 24302 | 24293 | -9 | -0.04% | 0.04% |
| instcombine.NumDeadInst | 1620140 | 1620112 | -28 | 0.00% | 0.00% |
| instcount.NumBrInst | 898466 | 898464 | -2 | 0.00% | 0.00% |
| instcount.NumCallInst | 1760819 | 1760875 | 56 | 0.00% | 0.00% |
| instcount.NumExtractValueInst | 45659 | 45649 | -10 | -0.02% | 0.02% |
| instcount.NumInsertValueInst | 4991 | 4981 | -10 | -0.20% | 0.20% |
| instcount.NumIntToPtrInst | 27084 | 27087 | 3 | 0.01% | 0.01% |
| instcount.NumPHIInst | 371435 | 371429 | -6 | 0.00% | 0.00% |
| instcount.NumStoreInst | 906011 | 906019 | 8 | 0.00% | 0.00% |
| instcount.TotalBlocks | 1105520 | 1105518 | -2 | 0.00% | 0.00% |
| instcount.TotalInsts | 9795737 | 9795776 | 39 | 0.00% | 0.00% |
| simplifycfg.NumInvokes | 2784 | 2786 | 2 | 0.07% | 0.07% |
| simplifycfg.NumSimpl | 1001840 | 1001850 | 10 | 0.00% | 0.00% |
| simplifycfg.NumSinkCommonInstrs | 15174 | 15170 | -4 | -0.03% | 0.03% |
```
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D86306
Sam Parker [Tue, 25 Aug 2020 07:35:07 +0000 (08:35 +0100)]
[NFC][RDA] Add explicit def check
Explicitly check that there is a local def prior to the given
instruction in getReachingLocalMIDef instead of just relying on
a nullptr return from getInstFromId.
Rainer Orth [Tue, 25 Aug 2020 07:36:51 +0000 (09:36 +0200)]
[compiler-rt][asan][test] Set LD_LIBRARY_PATH_{32,64} on Solaris
The dynamically linked ASan tests rely on `LD_LIBRARY_PATH` to find
`libclang_rt.asan-*.so` at runtime.
However, the Solaris runtime linker `ld.so.1` also supports more specific
variables: `LD_LIBRARY_PATH_32` and `LD_LIBRARY_PATH_64` respectively. If
those happen to be set, `LD_LIBRARY_PATH` is ignored. In such a case, all
dynamically linked ASan tests `FAIL`. For i386 alone, this affects about
200 tests.
The following patch fixes that by also setting `LD_LIBRARY_PATH_{32,64}` on
Solaris.
Tested on `amd64-pc-solaris2.11` both with only `LD_LIBRARY_PATH` set and
with `LD_LIBRARY_PATH_{32,64}` set too.
Differential Revision: https://reviews.llvm.org/D86333
Mateusz Mikuła [Tue, 25 Aug 2020 07:16:40 +0000 (10:16 +0300)]
[Compiler-RT] Fix profiler building with MinGW GCC
Differential Revision: https://reviews.llvm.org/D86405
Eduardo Caldas [Mon, 24 Aug 2020 15:54:22 +0000 (15:54 +0000)]
[SyntaxTree] Update `Modifiable` tests to dump `NodeRole` and `unmodifiable` tag
Eduardo Caldas [Fri, 21 Aug 2020 16:51:50 +0000 (16:51 +0000)]
[SyntaxTree] Update `Declaration` tests to dump `NodeRole`
Eduardo Caldas [Fri, 21 Aug 2020 12:57:16 +0000 (12:57 +0000)]
[SyntaxTree] Update `Expression` tests to dump `NodeRole`
Eduardo Caldas [Fri, 21 Aug 2020 09:57:15 +0000 (09:57 +0000)]
[SyntaxTree] Update `Statement` tests to dump `NodeRole`
Eduardo Caldas [Wed, 5 Aug 2020 13:55:17 +0000 (13:55 +0000)]
[SyntaxTree] Extend the syntax tree dump to also cover `NodeRole`
We should see `NodeRole` information in the dump because that exposes how the
accessors will behave.
Functional changes in the dump:
* Surround Leaf tokens with `'`
* Append `Node` dumps with `NodeRole` information, except for unknown roles
* Append marks to `Node` dumps, instead of prepending
Non-functional changes:
* `::dumpTokens(llvm::raw_ostream, ArrayRef<syntax::Token>, const
SourceManager &SM)` always received as parameter a `syntax::Token *`
pointing to `Leaf::token()`. Changed the function to
`dumpLeaf(llvm::raw_ostream, syntax::Leaf *, const SourceManager&)`
* `dumpTree` acted on a Node, rename to `dumpNode`
Differential Revision: https://reviews.llvm.org/D85330
Raphael Isemann [Tue, 25 Aug 2020 06:25:58 +0000 (08:25 +0200)]
[lldb] Don't depend on psutil in TestCompletion.py
psutil isn't reall a dependency of the test suite so this shouldn't be
unconditionally be imported here. Instead just check for the process name
by looking for the "a.out" string to get the bots green again.
Richard Smith [Tue, 25 Aug 2020 05:49:41 +0000 (22:49 -0700)]
PR37556: Don't diagnose conflicts between instantiated unqualified
friend declarations and declarations found in inline namespaces within
the target context.
Freddy Ye [Tue, 25 Aug 2020 04:27:02 +0000 (12:27 +0800)]
[X86] Support -march=sapphirerapids
Support -march=sapphirerapids for x86.
Compare with Icelake Server, it includes 14 more new features. They are
amxtile, amxint8, amxbf16, avx512bf16, avx512vp2intersect, cldemote,
enqcmd, movdir64b, movdiri, ptwrite, serialize, shstk, tsxldtrk, waitpkg.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D86503
Eduardo Caldas [Mon, 24 Aug 2020 16:38:38 +0000 (16:38 +0000)]
[SyntaxTree] Use annotations on ClassTemplate_MemberClassDefinition test
Differential Revision: https://reviews.llvm.org/D86470
Eduardo Caldas [Mon, 24 Aug 2020 16:33:09 +0000 (16:33 +0000)]
[SyntaxTree] Split ConstVolatileQualifiers tests
Differential Revision: https://reviews.llvm.org/D86469
Eduardo Caldas [Mon, 24 Aug 2020 16:13:34 +0000 (16:13 +0000)]
[SyntaxTree] Split `MemberPointer` tests with annotations
Differential Revision: https://reviews.llvm.org/D86467
Petr Hosek [Tue, 25 Aug 2020 06:00:54 +0000 (23:00 -0700)]
[CMake] Fix ncurses/zlib in LLVM_SYSTEM_LIBS for Windows GNU
For the Windows GNU platform, CMAKE_FIND_LIBRARY_PREFIXES is a list
containing an empty string, which ended up in a regex capturing group,
which is invalid in CMake's regex engine. With this change, we get the
following:
set(CMAKE_FIND_LIBRARY_PREFIXES "lib" "")
set(CMAKE_FIND_LIBRARY_SUFFIXES ".dll.a" ".a" ".lib")
get_system_libname(path/to/libz.dll.a zlib)
message("${zlib}")
outputs z, as expected.
Patch By: haampie
Differential Revision: https://reviews.llvm.org/D86434
Eric Christopher [Tue, 25 Aug 2020 04:51:31 +0000 (21:51 -0700)]
Temporarily Revert "[DebugInfo] Move constructor homing case in shouldOmitDefinition."
as it's causing test failures.
This reverts commit
589ce5f7050dd83fd3f7dbc182ea0fb051ece994.
Amy Huang [Tue, 25 Aug 2020 03:17:59 +0000 (20:17 -0700)]
[DebugInfo] Move constructor homing case in shouldOmitDefinition.
For some reason the ctor homing case was before the template
specialization case, and could have returned false too early.
I moved the code out into a separate function to avoid this.
Also added a run line to the template specialization test. I guess
all the -debug-info-kind=limited tests should still pass with =constructor,
but it's probably unnecessary to test for all of those.
Differential Revision: https://reviews.llvm.org/D86491
Alexandre Ganea [Tue, 25 Aug 2020 01:55:34 +0000 (21:55 -0400)]
Disable 'not' test on Windows because 'env' from GnuWin32 cannot be used without arguments.
Tim Keith [Tue, 25 Aug 2020 01:20:22 +0000 (18:20 -0700)]
[flang] Don't attempt to compute element size if no DynamicType
If an error has occurred a symbol may have a DeclTypeSpec but no
valid DynamicType. There is no need to compute the size of erroneous
symbols.
Also, we only need to process object entities and procedure entities.
All other kinds of symbols can be skipped.
This fixes another problem revealed by https://bugs.llvm.org/show_bug.cgi?id=47265
Differential Revision: https://reviews.llvm.org/D86484
Eric Christopher [Tue, 25 Aug 2020 00:58:17 +0000 (17:58 -0700)]
Add NDEBUG checks around debug only loop to avoid unused variable
warnings.
Richard Smith [Mon, 24 Aug 2020 23:28:41 +0000 (16:28 -0700)]
Properly pass modules flags to frontend when using -std=c++20 instead of -std=c++2a.
In passing, also teach the driver to map /std:c++latest to -std=c++20
not -std=c++2a.
Mircea Trofin [Mon, 24 Aug 2020 18:36:22 +0000 (11:36 -0700)]
[MLInliner] Support training that doesn't require partial rewards
If we use training algorithms that don't need partial rewards, we don't
need to worry about an ir2native model. In that case, training logs
won't contain a 'delta_size' feature either (since that's the partial
reward).
Differential Revision: https://reviews.llvm.org/D86481
Fangrui Song [Tue, 25 Aug 2020 00:27:34 +0000 (17:27 -0700)]
[not][test] Fix disable-symbolization.test when 'printenv' is not available
On Windows, 'env' or 'printenv' may not exist.
Also switch back to 'env' which is specified by POSIX.1-2017. 'printenv' is not
standard (I picked it because 'printenv' exists on GnuWin32 but 'env' does not).
Reviewed By: zequanwu
Differential Revision: https://reviews.llvm.org/D86496
Muhammad Omair Javaid [Mon, 24 Aug 2020 19:41:01 +0000 (00:41 +0500)]
[LLDB] Fix SVE offset calculation in NativeRegisterContextLinux_arm64
There was typo left from changes in CalculateSVEOffset where we moved
FPSR/FPCR offset calculation into WriteRegister and ReadRegister.
Differential Revision: https://reviews.llvm.org/D79699
Elliott Hughes [Thu, 20 Aug 2020 22:59:18 +0000 (15:59 -0700)]
Fix test for D77924.
The trailing 'L' was missing in the expectation.
Differential Revision: https://reviews.llvm.org/D86321
shafik [Mon, 24 Aug 2020 21:38:55 +0000 (14:38 -0700)]
[LLDB] Fix how ValueObjectVariable handles DW_AT_const_value when the DWARFExpression holds the data that represents a constant value
In some cases when we have a DW_AT_const_value and the data can be found in the
DWARFExpression then ValueObjectVariable does not handle it properly and we end
up with an extracting data from value failed error.
The test is a very stripped down assembly file since reproducing this relies on the results of compiling with -O1 which may not be stable over time.
Differential Revision: https://reviews.llvm.org/D86311
Venkataramanan Kumar [Mon, 24 Aug 2020 22:12:32 +0000 (18:12 -0400)]
[DAGCombine]: Fold X/Sqrt(X) to Sqrt(X)
With FMF ( "nsz" and " reassoc") fold X/Sqrt(X) to Sqrt(X).
This is done after targets have the chance to produce a
reciprocal sqrt estimate sequence because that expansion
is probably more efficient than an expansion of a
non-reciprocal sqrt. That is also why we deferred doing
this transform in IR (D85709).
Differential Revision: https://reviews.llvm.org/D86403
Sanjay Patel [Mon, 24 Aug 2020 22:05:05 +0000 (18:05 -0400)]
[x86][AArch64] adjust fast-math-flags in tests; NFC
This goes with the proposal in D86403.
Jonas Devlieghere [Mon, 24 Aug 2020 22:07:02 +0000 (15:07 -0700)]
[lldb] Add a SymbolFileProvider to record and replay calls to dsymForUUID
When replaying a reproducer captured from a core file, we always use
dsymForUUID for the kernel binary. When enabled, we also use it to find
kexts. Since these files are already contained in the reproducer,
there's no reason to call out to an external tool. If the tool returns a
different result, e.g. because the dSYM got garbage collected, it will
break reproducer replay. The SymbolFileProvider solves the issue by
mapping UUIDs to module and symbol paths in the reproducer.
Differential revision: https://reviews.llvm.org/D86389
Matt Arsenault [Wed, 12 Aug 2020 15:17:40 +0000 (11:17 -0400)]
AMDGPU/GlobalISel: Handle AGPRs used for SGPR operands.
We would still need to waterfall if the value were somehow an AGPR,
and also need to explicitly copy to a VGPR.
Nemanja Ivanovic [Mon, 24 Aug 2020 21:51:19 +0000 (16:51 -0500)]
[PowerPC] Do not use FISel for calls and TOC-based accesses with PC-Rel
PC-Relative addressing introduces a fair bit of complexity for correctly
eliminating TOC accesses. FastISel does not include any of that handling so we
miscompile code with -mcpu=pwr10 -O0 if it includes an external call that
FastISel does not handle followed by any of the following:
Floating point constant materialization
Materialization of a GlobalValue
Call that FastISel does handle
This patch switches to SDISel for any of the above.
Differential revision: https://reviews.llvm.org/D86343
Craig Topper [Mon, 24 Aug 2020 17:51:56 +0000 (10:51 -0700)]
[X86] Copy the tuning features and scheduler model from pentium4/x86-64 to generic
This is preparation for making clang default to -mtune=generic when no -march is specified. This will allow the default tuning to be "generic" even though our default march is "pentium4" or "x86-64".
To avoid llc lit test regressions, if no mcpu is specified, I've defaulted tune to use i586 to match the old tuning settings of no CPU. Some tests explicitly used -mcpu=generic which I've removed so they instead get this default of architecture features from generic and tune from i586.
I updated one llvm-mca test to check a different CPU since generic has a scheduler model now
Differential Revision: https://reviews.llvm.org/D86312
Matt Arsenault [Thu, 30 Jul 2020 02:01:32 +0000 (22:01 -0400)]
AMDGPU: Have a few selection failure tests check both paths
SelectionDAG and GlobalISel take different failure paths for these and
end up producing different failure errors. Check both so the test
passes when the default is switched.
Julian Lettner [Mon, 24 Aug 2020 21:36:18 +0000 (14:36 -0700)]
[TSan][libdispatch] Add interceptors for dispatch_barrier_async_and_wait(), pt. 2
Also intercept these target functions on Linux.
Nemanja Ivanovic [Mon, 24 Aug 2020 21:22:44 +0000 (16:22 -0500)]
[PowerPC] Handle SUBFIC in reg+reg -> reg+imm transformation
We initially missed the subtract-immediate in this transformation.
This patch just adds that.
Differential revision: https://reviews.llvm.org/D84659
Ryan Prichard [Mon, 24 Aug 2020 21:07:20 +0000 (14:07 -0700)]
[libunwind] Remove static_assert / __has_feature macros
The static_assert macro broke on GCC when a scope had two asserts and a
condition that depended on a template parameter. Remove the macro and
rely on the compiler's C++11 static_assert feature.
The __has_feature macro was only used here to determine whether to
define the static_assert macro.
Reviewed By: mstorsjo, #libunwind
Differential Revision: https://reviews.llvm.org/D86411
Julian Lettner [Mon, 24 Aug 2020 20:56:22 +0000 (13:56 -0700)]
[TSan][libdispatch] Add interceptors for dispatch_barrier_async_and_wait()
Add interceptors for the barrier variants of the "async and wait"
dispatch API family.
Sanjay Patel [Mon, 24 Aug 2020 19:31:31 +0000 (15:31 -0400)]
[InstCombine] improve demanded element analysis for vector insert-of-extract
InstCombine currently has odd rules for folding insert-extract chains to shuffles,
so we miss collapsing seemingly simple cases as shown in the tests here.
But poison makes this not quite as easy as we might have guessed. Alive2 tests to
show the subtle difference (similar to the regression tests):
https://alive2.llvm.org/ce/z/hp4hv3 (this is ok)
https://alive2.llvm.org/ce/z/ehEWaN (poison leakage)
SLP tends to create these patterns (as shown in the SLP tests), and this could
help with solving PR16739.
Differential Revision: https://reviews.llvm.org/D86460
Sanjay Patel [Mon, 24 Aug 2020 14:10:00 +0000 (10:10 -0400)]
[SLP] avoid 'tmp' names in regression tests; NFC
That can cause problems for update_test_checks.py (it warns when updating this file).
Sanjay Patel [Mon, 24 Aug 2020 13:27:30 +0000 (09:27 -0400)]
[InstCombine] add tests for insert+extract demanded elements; NFC
Shoaib Meenai [Fri, 21 Aug 2020 20:03:13 +0000 (13:03 -0700)]
[runtimes] Use llvm-libtool-darwin for runtimes build
It's full featured now and we can use it for the runtimes build instead
of relying on an external libtool, which means the CMAKE_HOST_APPLE
restriction serves no purpose either now. Restrict llvm-lipo to Darwin
targets while I'm here, since it's only needed there.
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D86367
Shoaib Meenai [Fri, 21 Aug 2020 19:57:16 +0000 (12:57 -0700)]
[runtimes] Remove TOOLCHAIN_TOOLS specialization
{builtin,runtime}_register_target passes a TOOLCHAIN_TOOLS list, whereas
{builtin,runtime}_default_target does notl. The explicit TOOLCHAIN_TOOLS
list matches what LLVMExternalProjectUtils would have set anyway,
barring some target-specific adjustments, and those target-specific
adjustments seem valuable, so let's drop the explicit TOOLCHAIN_TOOLS
list and let LLVMExternalProjectUtils take care of it.
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D86366
Shoaib Meenai [Fri, 21 Aug 2020 19:54:31 +0000 (12:54 -0700)]
[compiler-rt] Disable ranlib when using libtool
libtool already produces a table of contents, and ranlib just gives
spurious errors because it doesn't understand universal binaries.
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D86365
Shoaib Meenai [Fri, 21 Aug 2020 17:49:59 +0000 (10:49 -0700)]
[llvm-libtool-darwin] Add support for -V option
The -V option in cctools' libtool prints out the version number and
performs any specified operation. Add this option to LLVM's version.
cctools is more forgiving of invalid command lines when -V is specified,
but I think it's better to give errors instead of silently producing no
output.
Unfortunately, when -V is present, options that would otherwise be
required aren't anymore, so we need to perform some manual argument
validation.
Reviewed By: alexshap
Differential Revision: https://reviews.llvm.org/D86359
Dimitry Andric [Mon, 24 Aug 2020 20:45:02 +0000 (22:45 +0200)]
Add <stdarg.h> include to kmp_os.h, to get the va_list type, required
after
cde8f4c164a2. Sort system includes, while here.
Dimitry Andric [Sat, 22 Aug 2020 13:25:18 +0000 (15:25 +0200)]
Move special va_list handling to kmp_os.h
Instead of copying and pasting the same `#ifdef` expressions in multiple
places, define a type and a pair of macros in `kmp_os.h`, to handle
whether `va_list` is pointer-like or not:
* `kmp_va_list` is the type to use for `__kmp_fork_call()`
* `kmp_va_deref()` dereferences a `va_list`, if necessary
* `kmp_va_addr_of()` takes the address of a `va_list`, if necessary
Also add FreeBSD to the list of OSes that has a non pointer-like
va_list. This can now be easily extended to other OSes too.
Reviewed By: AndreyChurbanov
Differential Revision: https://reviews.llvm.org/D86397
Bjorn Pettersson [Mon, 24 Aug 2020 17:15:34 +0000 (19:15 +0200)]
[Scalarizer] Avoid updating the name of globals
The "takeName" logic at the end of ScalarizerVisitor::finish
could end up renaming global variables when having simplified
and extractelement instruction to simply pick a single vector
element. If the input vector to the extractelement instruction
held pointers to global variables we ended up renaming the global
variable.
The patch make sure we only take the name of the replaced Op when
we have added new instructions that might need a useful name.
Reviewed By: lebedev.ri
Differential Revision: https://reviews.llvm.org/D86472
Tim Keith [Mon, 24 Aug 2020 19:53:44 +0000 (12:53 -0700)]
[flang] Fix bug accessing implicit variable in specification expression
A specification expression can reference an implicitly declared variable
in the host procedure. Because we have to process specification parts
before execution parts, this may be the first time we encounter the
variable. We were assuming the variable was implicitly declared in the
scope where it was encountered, leading to an error because local
variables may not be referenced in specification expressions.
The fix is to tentatively create the implicit variable in the host
procedure because that is the only way the specification expression can
be valid. We mark it with the flag `ImplicitOrError` to indicate that
either it must be implicitly defined in the host (by being mentioned in
the execution part) or else its use turned out to be an error.
We need to apply the implicit type rules of the host, which requires
some changes to implicit typing.
Variables in common blocks are allowed to appear in specification expressions
(because they are not locals) but the common block definition may not appear
until after their use. To handle this we create common block symbols and object
entities for each common block object during the `PreSpecificationConstruct`
pass. This allows us to remove the corresponding code in the main visitor and
`commonBlockInfo_.curr`. The change in order of processing causes some
different error messages to be emitted.
Some cleanup is included with this change:
- In `ExpressionAnalyzer`, if an unresolved name is encountered but
no error has been reported, emit an internal error.
- Change `ImplicitRulesVisitor` to hide the `ImplicitRules` object
that implements it. Change the interface to pass in names rather
than having to get the first character of the name.
- Change `DeclareObjectEntity` to have the `attrs` argument default
to an empty set; that is the typical case.
- In `Pre(parser::SpecificationPart)` use "structured bindings" to
give names to the pieces that make up a specification-part.
- Enhance `parser::Unwrap` to unwrap `Statement` and `UnlabeledStatement`
and make use of that in PreSpecificationConstruct.
Differential Revision: https://reviews.llvm.org/D86322
Roman Lebedev [Mon, 24 Aug 2020 19:39:03 +0000 (22:39 +0300)]
[NFC][InstCombine] Multi-level aggregate test for phi-of-insertvalue pattern
See https://reviews.llvm.org/D86306
Roman Lebedev [Mon, 24 Aug 2020 19:33:58 +0000 (22:33 +0300)]
[NFC][InstCombine] Adjust naming for some methods to match coding standards
Requested as preparatory cleanup in https://reviews.llvm.org/D86306#inline-799065
Roland Froese [Mon, 24 Aug 2020 19:30:04 +0000 (15:30 -0400)]
[PowerPC] Extend custom lower of vector truncate to handle wider input
Current custom lowering of truncate vector handles a source of up to 128 bits, but that only uses one of the two shuffle vector operands. Extend it to use both operands to handle 256 bit sources.
Differential Revision: https://reviews.llvm.org/D68035
zhanghb97 [Mon, 24 Aug 2020 18:54:38 +0000 (18:54 +0000)]
[mlir] Add Index Type, Floating Point Type and None Type subclasses to python bindings.
Based on the PyType and PyConcreteType classes, this patch implements the bindings of Index Type, Floating Point Type and None Type subclasses.
These three subclasses share the same binding strategy:
- The function pointer `isaFunction` points to `mlirTypeIsA***`.
- The `mlir***TypeGet` C API is bound with the `***Type` constructor in the python side.
Reviewed By: stellaraccident
Differential Revision: https://reviews.llvm.org/D86466
Greg Clayton [Mon, 10 Aug 2020 22:07:47 +0000 (15:07 -0700)]
Add hashing of the .text section to ProcessMinidump.
Breakpad will always have a UUID for binaries when it creates minidump files. If an ELF files has a GNU build ID, it will use that. If it doesn't, it will create one by hashing up to the first 4096 bytes of the .text section. LLDB was not able to load these binaries even when we had the right binary because the UUID didn't match. LLDB will use the GNU build ID first as the main UUID for a binary and fallback onto a 8 byte CRC if a binary doesn't have one. With this fix, we will check for the Breakpad hash or the Facebook hash (a modified version of the breakpad hash that collides a bit less) and accept binaries when these hashes match.
Differential Revision: https://reviews.llvm.org/D86261
Fangrui Song [Mon, 24 Aug 2020 18:37:18 +0000 (11:37 -0700)]
Revert D85812 "[coroutine] should disable inline before calling coro split"
This reverts commit
2e43acfed89b1903de473f682c65878bdebc395a.
LLVMCoroutines (the library which contains Coroutines.h) depends on LLVMipo (the
library which contains SampleProfile.cpp). It is inappropriate for
SampleProfile.cpp to depent on Coroutines.h (circular dependency).
The test inverted dependencies as well:
llvm/test/Transforms/Coroutines/coro-inline.ll uses -sample-profile.
Matt Arsenault [Fri, 31 Jul 2020 18:50:39 +0000 (14:50 -0400)]
AMDGPU: Add flag to disable promotion of uniform i16 ops
This interferes with GlobalISel's much better handling of the
situation.
This should really be disable for GlobalISel. However, the fallback
only re-runs the selection passes, and doesn't go back and rerun any
codegen IR passes. I haven't come up with a good solution to this
problem.
Julian Lettner [Mon, 24 Aug 2020 18:30:58 +0000 (11:30 -0700)]
[TSan][Darwin] Handle NULL argument in interceptor
Handle NULL address argument in the `mach_vm_[de]allocate()`
interceptors and fix test: `Assignment 2` is not valid if we weren't
able to re-allocate memory.
rdar://
67680613
vnalamot [Mon, 24 Aug 2020 18:25:34 +0000 (23:55 +0530)]
[AMDGPU, docs] Fix typos
Reviewed By: t-tye, Flakebi
Differential Revision: https://reviews.llvm.org/D86340
Valentin Clement [Mon, 24 Aug 2020 18:22:34 +0000 (14:22 -0400)]
[flang][openacc] Fix reduction operator parsing/unparsing
Update the way reduction operator are defined for the OpenACC parser.
Reviewed By: klausler
Differential Revision: https://reviews.llvm.org/D86296
Gongyu Deng [Mon, 24 Aug 2020 16:58:17 +0000 (18:58 +0200)]
[lldb] type category name common completion
1. Added a new common completion TypeCategoryNames to provide a list of category names for completion;
2. Applied the completion to these commands: type category delete/enable/disable/list/define;
3. Added a related test case;
4. Bound the completion to the arguments of the type 'eArgTypeName'.
Reviewed By: teemperor, JDevlieghere
Differential Revision: https://reviews.llvm.org/D84124
Matt Arsenault [Sun, 2 Aug 2020 18:52:20 +0000 (14:52 -0400)]
TableGen/GlobalISel: Allow inst matcher to check multiple opcodes
This is to initially handleg immAllOnesV, which should match
G_BUILD_VECTOR or G_BUILD_VECTOR_TRUNC. In the future, it could be
used for other patterns cases that map to multiple G_* instructions,
such as G_ADD and G_PTR_ADD.
Craig Topper [Mon, 24 Aug 2020 17:06:07 +0000 (10:06 -0700)]
[LegalizeTypes][X86] Add ROTL/ROTR to WidenVectorResult.
We can widen these just like any other binary operation.
Added test cases for v2i32 for X86 for coverage.
Fixes failures seen after D77152.
Jay Foad [Mon, 24 Aug 2020 09:52:38 +0000 (10:52 +0100)]
[SDAG] Convert FSHL <--> FSHR if the target only supports one of them
D77152 tried to do this but got it wrong in the shift-by-zero case.
D86430 reverted the wrong code. Reimplement the optimization with
different code depending on whether the shift amount is known to be
non-zero (modulo bitwidth).
This improves code quality for fshl tests on AMDGPU, which only has an
fshr instruction.
Differential Revision: https://reviews.llvm.org/D86438
Sameeran joshi [Mon, 24 Aug 2020 14:21:45 +0000 (19:51 +0530)]
[FLANG] Pick `.md` files when building sphinx documentation.
Need to build sphinx using below flags to Cmake
`-DLLVM_ENABLE_SPHINX=ON -DSPHINX_WARNINGS_AS_ERRORS=OFF`.
Generate html docs using cmake target
`docs-flang-html`
Generated html files should be at `build/tools/flang/docs/html`.
Patch in series from the dicussion on review
https://reviews.llvm.org/D85828
After this patch the markdown docmentation must be written using guide in-
`llvm/docs/MarkdownQuickstartTemplate.md`
Reviewed By: sscalpone
Differential Revision: https://reviews.llvm.org/D86131
Gongyu Deng [Mon, 24 Aug 2020 15:34:32 +0000 (17:34 +0200)]
[lldb] Remote disk file/directory completion for platform commands
1. Extended the gdb-remote communication related classes with disk file/directory
completion functions;
2. Added two common completion functions RemoteDiskFiles and
RemoteDiskDirectories based on the functions above;
3. Added completion for these commands:
A. platform get-file <remote-file> <local-file>;
B. platform put-file <local-file> <remote-file>;
C. platform get-size <remote-file>;
D. platform settings -w <remote-dir>;
E. platform open file <remote-file>.
4. Added related tests for client and server;
5. Updated docs/lldb-platform-packets.txt.
Reviewed By: labath
Differential Revision: https://reviews.llvm.org/D85284
Gongyu Deng [Mon, 24 Aug 2020 12:23:23 +0000 (14:23 +0200)]
[lldb] common completion for process pids and process names
1. Added two common completions: `ProcessIDs` and `ProcessNames`, which are
refactored from their original dedicated option completions;
2. Removed the dedicated option completion functions of `process attach` and
`platform process attach`, so that they can use arg-type-bound common
completions instead;
3. Bound `eArgTypePid` to the pid completion, `eArgTypeProcessName` to the
process name completion in `CommandObject.cpp`;
4. Added a related test case.
Reviewed By: teemperor
Differential Revision: https://reviews.llvm.org/D80700
Florian Hahn [Tue, 18 Aug 2020 10:12:31 +0000 (11:12 +0100)]
[DSE,MemorySSA] Do not use callCapturesBefore in isReadClobber.
Using callCapturesBefore potentially improves the precision and the
number of stores we can remove. But in practice, it seems to have very
little impact in terms of stores removed. For example, for
SPEC2000/SPEC2006/MultiSource with -O3 -flto, ~50 more stores are
removed (out of ~26900 stores removed). But in terms of compile-time, it
is very expensive and the patch gives substantial compile-time
improvements: Geomean O3 -0.24%, ReleaseThinLTO -0.47%, ReleaseLTO-g
-0.39%.
http://llvm-compile-time-tracker.com/compare.php?from=
612a0bff88ed906c83b82f079d4c49e5fecfb9d0&to=
e6c86b96d20d97dd88e903a409bd8d39b6114312&stat=instructions
Raphael Isemann [Mon, 24 Aug 2020 15:07:56 +0000 (17:07 +0200)]
[clang][NFC] Fix a GCC warning in ASTImporterTest.cpp
Apparently only overriding one of the two CompleteType overloads causes
GCC to emit a warning with -Woverloaded-virtual .
Matt Arsenault [Wed, 19 Aug 2020 20:26:22 +0000 (16:26 -0400)]
AMDGPU/GlobalISel: Use unmerge instead of extract in addrspace queries
This is a bit more consistent with regular operation legalization.
AndreyChurbanov [Mon, 24 Aug 2020 14:50:38 +0000 (17:50 +0300)]
[OpenMP] Check if _MSC_VER is defined before using it
Patch by mati865@gmail.com
Differential Revision: https://reviews.llvm.org/D86448
Dokyung Song [Fri, 21 Aug 2020 22:40:26 +0000 (22:40 +0000)]
[libFuzzer] Make msan.test:SimpleCmpTest succeed with less trials.
Currently SimpleCmpTest passes after 9,831,994 trials on x86_64/Linux
when the number of given trials is 10,000,000, just a little bigger than
that. This patch modifies SimpleCmpTest.cpp so that the test passes with less
trials, reducing its chances of future failures as libFuzzer evolves. More
specifically, this patch changes a 32-bit equality check to a 8-bit equality
check, making this test pass at 4,635,303 trials.
Differential Revision: https://reviews.llvm.org/D86382
Thomas Preud'homme [Wed, 19 Aug 2020 14:47:42 +0000 (15:47 +0100)]
Test all CHECK-NOT in a block even if one fails
This commit makes FileCheck print all CHECK-NOT directive failure in a
CHECK-NOT block even if one fails. Prior to that, it would stop trying
to match CHECK-NOT directive as soon as one in the block fails.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D86315
Baptiste Saleil [Mon, 24 Aug 2020 14:14:16 +0000 (09:14 -0500)]
[PowerPC] Add clang options to control MMA support
This patch adds frontend and backend options to enable and disable
the PowerPC MMA operations added in ISA 3.1. Instructions using these
options will be added in subsequent patches.
Differential Revision: https://reviews.llvm.org/D81442
Eduardo Caldas [Mon, 24 Aug 2020 14:11:52 +0000 (14:11 +0000)]
[SyntaxTree] Split `DynamicExceptionSpecification` test
Eduardo Caldas [Mon, 24 Aug 2020 13:57:20 +0000 (13:57 +0000)]
[SyntaxTree] Split `ParametersAndQualifiers` tests
Differential Revision: https://reviews.llvm.org/D86459
Eduardo Caldas [Mon, 24 Aug 2020 10:30:09 +0000 (10:30 +0000)]
[SyntaxTree] Group tests related to `using`
Differential Revision: https://reviews.llvm.org/D86443
Eduardo Caldas [Mon, 24 Aug 2020 10:19:19 +0000 (10:19 +0000)]
[SyntaxTree] Split ExplicitTemplateInstantiation test
Differential Revision: https://reviews.llvm.org/D86441
Eduardo Caldas [Mon, 24 Aug 2020 10:07:27 +0000 (10:07 +0000)]
[SyntaxTree] Split FreeStandingClass tests
Differential Revision: https://reviews.llvm.org/D86440
Eduardo Caldas [Mon, 24 Aug 2020 09:54:58 +0000 (09:54 +0000)]
[SyntaxTree] Use annotations to reduce noise on member function tests
Differential Revision: https://reviews.llvm.org/D86439
Eduardo Caldas [Mon, 24 Aug 2020 09:50:03 +0000 (09:50 +0000)]
[SyntaxTree] Split array declarator tests
Differential Revision: https://reviews.llvm.org/D86437
dongAxis [Mon, 24 Aug 2020 14:22:08 +0000 (22:22 +0800)]
[coroutine] should disable inline before calling coro split
summary:
When callee coroutine function is inlined into caller coroutine
function before coro-split pass, llvm will emits "coroutine should
have exactly one defining @llvm.coro.begin". It seems that coro-early
pass can not handle this quiet well.
So we believe that unsplited coroutine function should not be inlined.
This patch fix such issue by not inlining function if it has attribute
"coroutine.presplit" (it means the function has not been splited) to
fix this issue
TestPlan: check-llvm
Reviewed By: wenlei
Differential Revision: https://reviews.llvm.org/D85812
Matt Arsenault [Sun, 23 Aug 2020 19:02:50 +0000 (15:02 -0400)]
GlobalISel: Improve dead instruction debug printing
This was printing the "Is dead" on a separate line from the
instruction, which was harder to follow.
Matt Arsenault [Thu, 16 Jul 2020 23:37:31 +0000 (19:37 -0400)]
AMDGPU/GlobalISel: Use different technique for sample v3s16 values
Avoid relying on implicit_def values, and odd sized G_INSERT/G_EXTRACT
Matt Arsenault [Wed, 19 Aug 2020 22:13:23 +0000 (18:13 -0400)]
AMDGPU/GlobalISel: Add baseline, failing unmerge tests
Alexey Bataev [Wed, 19 Aug 2020 19:23:00 +0000 (15:23 -0400)]
[OPENMP]Fix PR47158, case 3: allow devic_typein nested declare target region.
OpenMP 5.0 supports nested declare target regions. So, in general,it is
allow to mark a declarationas declare target with different device_type
or link type. Patch adds support for such kind of nesting.
Differential Revision: https://reviews.llvm.org/D86239
Francesco Petrogalli [Fri, 7 Aug 2020 22:03:24 +0000 (22:03 +0000)]
[llvm][LV] Replace `unsigned VF` with `ElementCount VF` [NFCI]
Changes:
* Change `ToVectorTy` to deal directly with `ElementCount` instances.
* `VF == 1` replaced with `VF.isScalar()`.
* `VF > 1` and `VF >=2` replaced with `VF.isVector()`.
* `VF <=1` is replaced with `VF.isZero() || VF.isScalar()`.
* Replaced the uses of `llvm::SmallSet<ElementCount, ...>` with
`llvm::SmallSetVector<ElementCount, ...>`. This avoids the need of an
ordering function for the `ElementCount` class.
* Bits and pieces around printing the `ElementCount` to string streams.
To guarantee that this change is a NFC, `VF.Min` and asserts are used
in the following places:
1. When it doesn't make sense to deal with the scalable property, for
example:
a. When computing unrolling factors.
b. When shuffle masks are built for fixed width vector types
In this cases, an
assert(!VF.Scalable && "<mgs>") has been added to make sure we don't
enter coepaths that don't make sense for scalable vectors.
2. When there is a conscious decision to use `FixedVectorType`. These
uses of `FixedVectorType` will likely be removed in favour of
`VectorType` once the vectorizer is generic enough to deal with both
fixed vector types and scalable vector types.
3. When dealing with building constants out of the value of VF, for
example when computing the vectorization `step`, or building vectors
of indices. These operation _make sense_ for scalable vectors too,
but changing the code in these places to be generic and make it work
for scalable vectors is to be submitted in a separate patch, as it is
a functional change.
4. When building the potential VFs in VPlan. Making the VPlan generic
enough to handle scalable vectorization factors is a functional change
that needs a separate patch. See for example `void
LoopVectorizationPlanner::buildVPlans(unsigned MinVF, unsigned
MaxVF)`.
5. The class `IntrinsicCostAttribute`: this class still uses `unsigned
VF` as updating the field to use `ElementCount` woudl require changes
that could result in changing the behavior of the compiler. Will be done
in a separate patch.
7. When dealing with user input for forcing the vectorization
factor. In this case, adding support for scalable vectorization is a
functional change that migh require changes at command line.
Note that in some places the idiom
```
unsigned VF = ...
auto VTy = FixedVectorType::get(ScalarTy, VF)
```
has been replaced with
```
ElementCount VF = ...
assert(!VF.Scalable && ...);
auto VTy = VectorType::get(ScalarTy, VF)
```
The assertion guarantees that the new code is (at least in debug mode)
functionally equivalent to the old version. Notice that this change had been
possible because none of the methods that are specific to `FixedVectorType`
were used after the instantiation of `VTy`.
Reviewed By: rengolin, ctetreau
Differential Revision: https://reviews.llvm.org/D85794
Matt Arsenault [Sun, 16 Aug 2020 17:42:55 +0000 (13:42 -0400)]
AMDGPU/GlobalISel: Start implementing computeKnownBitsForTargetInstr
Handle workitem intrinsics. There isn't really away to adequately test
this right now, since none of the known bits users are fine grained
enough to test the edge conditions. This triggers a number of
instances of the new 64-bit to 32-bit shift combine in the existing
tests.
Francesco Petrogalli [Mon, 24 Aug 2020 13:49:27 +0000 (13:49 +0000)]
Revert "[llvm][LV] Replace `unsigned VF` with `ElementCount VF` [NFCI]"
Reverting because the commit message doesn't reflect the one agreed on
phabricator at https://reviews.llvm.org/D85794.
This reverts commit
c8d2b065b98fa91139cc7bb1fd1407f032ef252e.
Matt Arsenault [Sat, 15 Aug 2020 17:38:29 +0000 (13:38 -0400)]
GlobalISel: Reduce G_SHL width if source is extension
shl ([sza]ext x, y) => zext (shl x, y).
Turns expensive 64 bit shifts into 32 bit if it does not overflow the
source type:
This is a port of an AMDGPU DAG combine added in
5fa289f0d8ff85b9e14d2f814a90761378ab54ae. InstCombine does this
already, but we need to do it again here to apply it to shifts
introduced for lowered getelementptrs. This will help matching
addressing modes that use 32-bit offsets in a future patch.
TableGen annoyingly assumes only a single match data operand, so
introduce a reusable struct. However, this still requires defining a
separate GIMatchData for every combine which is still annoying.
Adds a morally equivalent function to the existing
getShiftAmountTy. Without this, we would have to do try to repeatedly
query the legalizer info and guess at what type to use for the shift.
Francesco Petrogalli [Fri, 7 Aug 2020 22:03:24 +0000 (22:03 +0000)]
[llvm][LV] Replace `unsigned VF` with `ElementCount VF` [NFCI]
Changes:
* Change `ToVectorTy` to deal directly with `ElementCount` instances.
* `VF == 1` replaced with `VF.isScalar()`.
* `VF > 1` and `VF >=2` replaced with `VF.isVector()`.
* `VF <=1` is replaced with `VF.isZero() || VF.isScalar()`.
* Add `<` operator to `ElementCount` to be able to use
`llvm::SmallSetVector<ElementCount, ...>`.
* Bits and pieces around printing the ElementCount to string streams.
* Added a static method to `ElementCount` to represent a scalar.
To guarantee that this change is a NFC, `VF.Min` and asserts are used
in the following places:
1. When it doesn't make sense to deal with the scalable property, for
example:
a. When computing unrolling factors.
b. When shuffle masks are built for fixed width vector types
In this cases, an
assert(!VF.Scalable && "<mgs>") has been added to make sure we don't
enter coepaths that don't make sense for scalable vectors.
2. When there is a conscious decision to use `FixedVectorType`. These
uses of `FixedVectorType` will likely be removed in favour of
`VectorType` once the vectorizer is generic enough to deal with both
fixed vector types and scalable vector types.
3. When dealing with building constants out of the value of VF, for
example when computing the vectorization `step`, or building vectors
of indices. These operation _make sense_ for scalable vectors too,
but changing the code in these places to be generic and make it work
for scalable vectors is to be submitted in a separate patch, as it is
a functional change.
4. When building the potential VFs in VPlan. Making the VPlan generic
enough to handle scalable vectorization factors is a functional change
that needs a separate patch. See for example `void
LoopVectorizationPlanner::buildVPlans(unsigned MinVF, unsigned
MaxVF)`.
5. The class `IntrinsicCostAttribute`: this class still uses `unsigned
VF` as updating the field to use `ElementCount` woudl require changes
that could result in changing the behavior of the compiler. Will be done
in a separate patch.
7. When dealing with user input for forcing the vectorization
factor. In this case, adding support for scalable vectorization is a
functional change that migh require changes at command line.
Differential Revision: https://reviews.llvm.org/D85794
Valeriy Savchenko [Thu, 16 Jul 2020 10:28:29 +0000 (13:28 +0300)]
[analyzer][tests] Add a notion of project sizes
Summary:
Whith the number of projects growing, it is important to be able to
filter them in a more convenient way than by names. It is especially
important for benchmarks, when it is not viable to analyze big
projects 20 or 50 times in a row.
Because of this reason, this commit adds a notion of sizes and a
filtering interface that puts a limit on a maximum size of the project
to analyze or benchmark.
Sizes assigned to the projects in this commit, do not directly
correspond to the number of lines or files in the project. The key
factor that is important for the developers of the analyzer is the
time it takes to analyze the project. And for this very reason,
"size" basically helps to cluster projects based on their analysis
time.
Differential Revision: https://reviews.llvm.org/D83942
Florian Hahn [Fri, 21 Aug 2020 17:17:00 +0000 (18:17 +0100)]
[DSE,MemorySSA] Delay PointerMayBeCaptured calls until actually needed.
Avoid computing InvisibleToCallerBefore/AfterRet up front. In most
cases, this information is not really needed. Instead, introduce helper
functions to compute and cache the result on demand.
Notably, this also does not use PointerMayBeCapturedBefore for
isInvisibleToCallerBeforeRet, as it requires the killing MemoryDef as
starting instruction, making the caching ineffective. But it appears the
use of PointerMayBeCapturedBefore has very limited benefits in practice
(e.g. on SPEC2000/SPEC2006/MultiSource there are no binary changes with
-O3 -flto). Refrain from using it for now, to limit-compile-time.
This gives some nice compile-time improvements:
http://llvm-compile-time-tracker.com/compare.php?from=
db9345f6810f379a36752dc52caf5230585d0ebd&to=
b4d091047e1b8a3d377d200137b79d03aca65663&stat=instructions
Anna Welker [Mon, 24 Aug 2020 11:04:57 +0000 (12:04 +0100)]
[ARM][MVE] Allow tail predication for strides !=1 with gather/scatters
If gather/scatters are enabled, ARMTargetTransformInfo now allows
tail predication for loops with a much wider range of strides, up
to anything that is loop invariant.
Differential Revision: https://reviews.llvm.org/D85410
Raphael Isemann [Fri, 21 Aug 2020 12:21:21 +0000 (14:21 +0200)]
Reland "Correctly emit dwoIDs after ASTFileSignature refactoring (D81347)"
The orignal patch with the missing 'REQUIRES: asserts' as there is a debug-only
flag used in the test.
Original summary:
D81347 changes the ASTFileSignature to be an array of 20 uint8_t instead of 5
uint32_t. However, it didn't update the code in ObjectFilePCHContainerOperations
that creates the dwoID in the module from the ASTFileSignature
(`Buffer->Signature` being the array subclass that is now `std::array<uint8_t,
20>` instead of `std::array<uint32_t, 5>`).
```
uint64_t Signature = [..] (uint64_t)Buffer->Signature[1] << 32 | Buffer->Signature[0]
```
This code works with the old ASTFileSignature (where two uint32_t are enough to
fill the uint64_t), but after the patch this only took two bytes from the
ASTFileSignature and only partly filled the Signature uint64_t.
This caused that the dwoID in the module ref and the dwoID in the actual module
no longer match (which in turns causes that LLDB keeps warning about the dwoID's
not matching when debugging -gmodules-compiled binaries).
This patch just unifies the logic for turning the ASTFileSignature into an
uint64_t which makes the dwoID match again (and should prevent issues like that
in the future).
Reviewed By: aprantl, dang
Differential Revision: https://reviews.llvm.org/D84013
Bevin Hansson [Fri, 21 Aug 2020 08:26:52 +0000 (10:26 +0200)]
[Fixed Point] Add codegen for fixed-point shifts.
This patch adds codegen to Clang for fixed-point shift
operations.
Reviewed By: leonardchan
Differential Revision: https://reviews.llvm.org/D83294
Bevin Hansson [Thu, 20 Aug 2020 12:36:35 +0000 (14:36 +0200)]
[Fixed Point] Use FixedPointBuilder to codegen fixed-point IR.
This changes the methods in CGExprScalar to use
FixedPointBuilder to generate IR for fixed-point
conversions and operations.
Since FixedPointBuilder emits padded operations slightly
differently than the original code, some tests change.
Reviewed By: leonardchan
Differential Revision: https://reviews.llvm.org/D86282
Haojian Wu [Mon, 24 Aug 2020 12:20:36 +0000 (14:20 +0200)]
[AST][RecoveryAST] Preserve the type by default for recovery expression.
Differential Revision: https://reviews.llvm.org/D82657
Florian Hahn [Mon, 24 Aug 2020 12:24:44 +0000 (13:24 +0100)]
[DSE,MemorySSA] Regnerate some check lines.
The check lines where generated before align was added for all
instructions. Re-generate them, to reduce diff noise for actual
functional changes.
Jonas Paulsson [Mon, 24 Aug 2020 12:07:30 +0000 (14:07 +0200)]
[SystemZ] Preserve the MachineMemOperand in emitCondStore() in all cases.
Review: Ulrich Weigand
Florian Hahn [Sat, 22 Aug 2020 09:08:59 +0000 (10:08 +0100)]
[DSE,MemorySSA] Limit elimination at end of function to single UO.
Limit elimination of stores at the end of a function to MemoryDefs with
a single underlying object, to save compile time.
In practice, the case with multiple underlying objects seems not very
important in practice. For -O3 -flto on MultiSource/SPEC2000/SPEC2006
this results in a total of 2 more stores being eliminated.
We can always re-visit that in the future.