Francesco Petrogalli [Tue, 8 Sep 2020 08:08:59 +0000 (08:08 +0000)]
[llvm][CodeGen] Do not scalarize `llvm.masked.[gather|scatter]` operating on scalable vectors.
This patch prevents the `llvm.masked.gather` and `llvm.masked.scatter` intrinsics to be scalarized when invoked on scalable vectors.
The change in `Function.cpp` is needed to prevent the warning that is raised when `getNumElements` is used in place of `getElementCount` on `VectorType` instances. The tests guards for regressions on this change.
The tests makes sure that calls to `llvm.masked.[gather|scatter]` are still scalarized when:
# the intrinsics are operating on fixed size vectors, and
# the compiler is not targeting fixed length SVE code generation.
Reviewed By: efriedma, sdesmalen
Differential Revision: https://reviews.llvm.org/D86249
Arthur Eubanks [Wed, 16 Sep 2020 05:06:50 +0000 (22:06 -0700)]
[NPM] Translate alias analysis into require<> as well
'require<globals-aa>' is needed to make globals-aa work in NPM, since
globals-aa is a module analysis but function passes cannot run module
analyses on demand.
So don't skip translating alias analyses to 'require<>'.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D87743
Dmitry Preobrazhensky [Wed, 16 Sep 2020 15:51:26 +0000 (18:51 +0300)]
[AMDGPU] Corrected directive to use for ELF weak refs
WeakRefDirective should specify a directive to declare "a global as being a weak undefined symbol".
The directive used by AMDGPU was incorrect - ".weakref" was intended for other purposes.
The correct directive is ".weak" and it is already defined as default for ELF.
So the redefinition was removed.
Reviewers: arsenm, rampitec
Differential Revision: https://reviews.llvm.org/D87762
Benjamin Kramer [Wed, 16 Sep 2020 15:28:59 +0000 (17:28 +0200)]
[ASTMatchers] Add missing definition for decompositionDecl
Otherwise we'd get a linker error whenever decompositionDecl is ODR
used.
Simon Pilgrim [Wed, 16 Sep 2020 15:26:13 +0000 (16:26 +0100)]
[X86] EmitInstrWithCustomInserter - remove redundant getDebugLoc() calls. NFCI.
Use the same DebugLoc that is called at the top of the method.
Fixes some Wshadow static analyzer warnings.
Mircea Trofin [Tue, 15 Sep 2020 21:54:38 +0000 (14:54 -0700)]
[NFC][Regalloc] accessors for 'reg' and 'weight'
Also renamed the fields to follow style guidelines.
Accessors help with readability - weight mutation, in particular,
is easier to follow this way.
Differential Revision: https://reviews.llvm.org/D87725
Matt Arsenault [Sun, 30 Aug 2020 21:28:48 +0000 (17:28 -0400)]
AMDGPU: Improve <2 x i24> arguments and return value handling
This was asserting for GlobalISel. For SelectionDAG, this was
passing this on the stack. Instead, scalarize this as if it were a
32-bit vector.
Sebastian Neubauer [Thu, 23 Jul 2020 14:59:00 +0000 (16:59 +0200)]
[AMDGPU] Add v3f16/v3i16 support to SDag
Fix lowering and instruction selection for v3x16 types
and enable InstCombine to emit them.
This patch only implements it for the selection dag.
GlobalISel tests in GlobalISel/llvm.amdgcn.image.load.1d.d16.ll and
GlobalISel/llvm.amdgcn.image.store.2d.d16.ll still don't work.
Differential Revision: https://reviews.llvm.org/D84420
Simon Pilgrim [Wed, 16 Sep 2020 15:17:35 +0000 (16:17 +0100)]
[X86] Assert that we've found a terminator instruction. NFCI.
Fixes clang static analayzer null dereference warning.
Jay Foad [Wed, 9 Sep 2020 16:21:36 +0000 (17:21 +0100)]
[AMDGPU] Enable scheduling around FP MODE-setting instructions
Pre-gfx10 all MODE-setting instructions were S_SETREG_B32 which is
marked as having unmodeled side effects, which makes the machine
scheduler treat it as a barrier. Now that we have proper implicit $mode
operands we can use a no-side-effects S_SETREG_B32_mode pseudo instead
for setregs that only touch the FP MODE bits, to give the scheduler more
freedom.
Differential Revision: https://reviews.llvm.org/D87446
Jay Foad [Tue, 15 Sep 2020 11:00:38 +0000 (12:00 +0100)]
[AMDGPU] Add -show-mc-encoding to setreg tests
This is a pre-commit for D87446 "[AMDGPU] Enable scheduling around FP MODE-setting instructions"
Simon Pilgrim [Wed, 16 Sep 2020 14:46:23 +0000 (15:46 +0100)]
[X86][SSE] Move VZEXT_MOVL(INSERT_SUBVECTOR(UNDEF,X,0)) handling into combineTargetShuffle.
Now that we're getting better at combining shuffles of different vector widths, this can now be performed as part of the standard target shuffle combines and isn't required for cleanup.
Exposed a minor issue in combineX86ShufflesRecursively where we failed to check if a shuffle's src ops were simple types.
Valentin Clement [Wed, 16 Sep 2020 14:48:51 +0000 (10:48 -0400)]
[mlir][openacc] Add missing operands for acc.parallel operation
Add missing operands to represent copin with readonly modifier, copyout with zero
modifier, create with zero modifier and default clause.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D87733
Eugene Zhulenev [Wed, 16 Sep 2020 14:03:35 +0000 (10:03 -0400)]
Enable inlining for Linalg dialect
Enable inlining for Linalg dialect.
Differential Revision: https://reviews.llvm.org/D87567
Dangeti Tharun kumar [Wed, 16 Sep 2020 14:11:24 +0000 (15:11 +0100)]
[Partial Inliner] Compute intrinsic cost through TTI
https://bugs.llvm.org/show_bug.cgi?id=45932
assert(OutlinedFunctionCost >= Cloner.OutlinedRegionCost && "Outlined function cost should be no less than the outlined region") getting triggered in computeBBInlineCost.
Intrinsics like "assume" are considered regular function calls while computing costs.
This patch enables computeBBInlineCost to queries TTI for intrinsic call cost.
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D87132
Alex Zinenko [Tue, 15 Sep 2020 10:04:59 +0000 (12:04 +0200)]
[mlir] Model StringRef in C API
Numerous MLIR functions return instances of `StringRef` to refer to a
non-owning fragment of a string (usually owned by the context). This is a
relatively simple class that is defined in LLVM. Provide a simple wrapper in
the MLIR C API that contains the pointer and length of the string fragment and
use it for Standard attribute functions that return StringRef instead of the
previous, callback-based mechanism.
Reviewed By: stellaraccident
Differential Revision: https://reviews.llvm.org/D87677
Florian Hahn [Wed, 16 Sep 2020 13:49:44 +0000 (14:49 +0100)]
[DSE] Add another test cases with loop carried dependence.
Alexey Bataev [Wed, 16 Sep 2020 12:10:55 +0000 (08:10 -0400)]
[OPENMP]Do not allow threadprivates as base for array-like reduction.
The base must be shared between the threads, threadprivates are not
allowed to be bases for array-like reductions.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D85762
Paul C. Anagnostopoulos [Tue, 15 Sep 2020 18:18:51 +0000 (14:18 -0400)]
Add section with details about DAGs.
Sanjay Patel [Wed, 16 Sep 2020 12:47:35 +0000 (08:47 -0400)]
[SLP] fix formatting; NFC
Also move variable declarations closer to usage and add code comments.
mydeveloperday [Wed, 16 Sep 2020 12:45:45 +0000 (13:45 +0100)]
[clang-format] [NFC] Fix spelling mistake in the documentation
Ensure ClangFormatStyleOptions.rst can be regenerated from Format.h
Patch By: YangZhihui
Reviewed By: MyDeveloperDay
Differential Revision: https://reviews.llvm.org/D87352
Sam Parker [Wed, 16 Sep 2020 12:38:36 +0000 (13:38 +0100)]
[ARM] Reorder some logic
Re-order some checks in ValidateMVEInst.
Sanjay Patel [Wed, 16 Sep 2020 12:26:21 +0000 (08:26 -0400)]
[SLP] remove uses of 'auto' that obscure functionality; NFC
Sanjay Patel [Wed, 16 Sep 2020 12:11:19 +0000 (08:11 -0400)]
[SLP] remove redundant size check; NFC
We bail out on small array size anyway.
Sanjay Patel [Wed, 16 Sep 2020 11:44:03 +0000 (07:44 -0400)]
[SLP] move loop index variable declaration to its use; NFC
Sanjay Patel [Wed, 16 Sep 2020 11:40:15 +0000 (07:40 -0400)]
[SLP] change poorly named variable; NFC
'V' shadows a function argument.
Sam Parker [Wed, 16 Sep 2020 11:42:58 +0000 (12:42 +0100)]
[RDA] Fix getUniqueReachingDef for self loops
We've fixed the case where this could return an instruction after the
given instruction, but also means that we can falsely return a
'unique' def when they could be one coming from the backedge of a
loop.
Differential Revision: https://reviews.llvm.org/D87751
Simon Pilgrim [Wed, 16 Sep 2020 11:40:15 +0000 (12:40 +0100)]
[Sema] isOpenMPCapturedDecl - assert we locate CapturedRegionScopeInfo. NFCI.
Fixes clang static analayzer null dereference warning.
Sam McCall [Wed, 16 Sep 2020 11:42:01 +0000 (13:42 +0200)]
Update dead links to Itanium and ARM ABIs. NFC
Simon Pilgrim [Wed, 16 Sep 2020 11:29:50 +0000 (12:29 +0100)]
[AST] ASTReader::ReadModuleMapFileBlock - assert non-null Module. NFCI.
At this stage the Module* shouldn't be null - add an assert to fix a clang static analyzer warning.
Simon Pilgrim [Wed, 16 Sep 2020 11:17:44 +0000 (12:17 +0100)]
TokenAnnotator.cpp - remove useless pointer null test. NFCI.
We dereference the Left pointer throughout the parseParens() function apart from this single case - just add an non-null assertion and drop the check.
Fixes clang static analayzer null dereference warning.
Simon Pilgrim [Wed, 16 Sep 2020 11:11:29 +0000 (12:11 +0100)]
CGBlocks.cpp - assert non-null CGF pointer. NFCI.
Fixes static analyzer warning.
Adam Balogh [Fri, 11 Sep 2020 17:04:38 +0000 (19:04 +0200)]
[ASTMatchers] Fix `hasBody` for the descendants of `FunctionDecl`
//AST Matcher// `hasBody` is a polymorphic matcher that behaves
differently for loop statements and function declarations. The main
difference is the for functions declarations it does not only call
`FunctionDecl::getBody()` but first checks whether the declaration in
question is that specific declaration which has the body by calling
`FunctionDecl::doesThisDeclarationHaveABody()`. This is achieved by
specialization of the template `GetBodyMatcher`. Unfortunately template
specializations do not catch the descendants of the class for which the
template is specialized. Therefore it does not work correcly for the
descendants of `FunctionDecl`, such as `CXXMethodDecl`,
`CXXConstructorDecl`, `CXXDestructorDecl` etc. This patch fixes this
issue by using a template metaprogram.
The patch also introduces a new matcher `hasAnyBody` which matches
declarations which have a body present in the AST but not necessarily
belonging to that particular declaration.
Differential Revision: https://reviews.llvm.org/D87527
Adam Balogh [Tue, 15 Sep 2020 10:53:13 +0000 (12:53 +0200)]
[clang-tidy] Crash fix for bugprone-misplaced-pointer-arithmetic-in-alloc
Placement new operators on non-object types cause crash in
`bugprone-misplaced-pointer-arithmetic-in-alloc`. This patch fixes this
issue.
Differential Revision: https://reviews.llvm.org/D87683
Adam Czachorowski [Tue, 15 Sep 2020 18:13:00 +0000 (20:13 +0200)]
[clangd] Actually parse Index section of the YAML file.
This fixes a bug in
dbf486c0de92c76df77c1a1f815cf16533ecbb3a, which
introduced the Index section of the config, but did not register the
parse method, so it didn't work in a YAML file (but did in a test).
Differential Revision: https://reviews.llvm.org/D87710
Sam Parker [Wed, 16 Sep 2020 10:47:26 +0000 (11:47 +0100)]
[ARM] Fix tail predication predicate tracking
Clear the CurrentPredicate when we find an instruction which would
completely overwrite the VPR. This fix essentially means we're back
to not really being able to handle VPT instructions when tail
predicating.
Differential Revision: https://reviews.llvm.org/D87610
Sam Parker [Wed, 16 Sep 2020 10:17:13 +0000 (11:17 +0100)]
[ARM] Add more validForTailPredication
Modify the unit test to inspect all MVE instructions and mark the
load/store/move of vpr/p0 as valid, as well as the remaining scalar
shifts.
Differential Revision: https://reviews.llvm.org/D87753
Kirill Bobyrev [Wed, 16 Sep 2020 10:27:20 +0000 (12:27 +0200)]
Fix table formatting after D87686
Simon Pilgrim [Wed, 16 Sep 2020 10:02:09 +0000 (11:02 +0100)]
[DAG] Remover getOperand() call. NFCI.
Kirill Bobyrev [Wed, 16 Sep 2020 10:09:29 +0000 (12:09 +0200)]
[clang-tidy] Improve documentation on Clangd integration
The integration is already complete; this patch updates information as well as
suggests using Clang-Tidy via Clangd integration that is vastly available
in most editors through LSP client plugins.
Reviewed By: hokein
Differential Revision: https://reviews.llvm.org/D87686
Sam Tebbs [Wed, 16 Sep 2020 09:59:19 +0000 (10:59 +0100)]
[ARM][LowOverheadLoops] Fix tests after ef0b9f3
ef0b9f3 didn't update the tests that it affected.
Georgii Rymar [Mon, 14 Sep 2020 13:38:29 +0000 (16:38 +0300)]
[llvm-readobj][test] - Improve section-symbols.test
`section-symbols.test` tests how we print section symbols in
different situations.
We might have 2 different cases:
1) A named STT_SECTION symbol.
2) An unnamed STT_SECTION symbol.
Usually section symbols have no name and then `--symbols` uses their
section names when prints them. If symbol has a name, then it is used.
For `--relocations` we also want to have this logic probably,
but currently we always ignore symbol names and always use section names.
It is not consistent with GNU readelf and with our logic for `--symbols`.
This patch refines testing to document the existent behavior and improve
coverage.
Differential revision: https://reviews.llvm.org/D87612
Andrew Ng [Wed, 9 Sep 2020 09:48:21 +0000 (10:48 +0100)]
[LLD][ELF] Optimize linker script filename glob pattern matching NFC
Optimize the filename glob pattern matching in
LinkerScript::computeInputSections() and LinkerScript::shouldKeep().
Add InputFile::getNameForScript() which gets and if required caches the
Inputfile's name used for linker script matching. This avoids the
overhead of name creation that was in getFilename() in LinkerScript.cpp.
Add InputSectionDescription::matchesFile() and
SectionPattern::excludesFile() which perform the glob pattern matching
for an InputFile and make use of a cache of the previous result. As both
computeInputSections() and shouldKeep() process sections in order and
the sections of the same InputFile are contiguous, these single entry
caches can significantly speed up performance for more complex glob
patterns.
These changes have been seen to reduce link time with --gc-sections by
up to ~40% with linker scripts that contain KEEP filename glob patterns
such as "*crtbegin*.o".
Differential Revision: https://reviews.llvm.org/D87469
Andrew Ng [Mon, 7 Sep 2020 12:22:12 +0000 (13:22 +0100)]
[Support] Add GlobPattern::isTrivialMatchAll()
GlobPattern::isTrivialMatchAll() returns true for the GlobPattern "*"
which will match all inputs.
This can be used to avoid performing expensive preparation of the input
for match() when the result of the match will always be true.
Differential Revision: https://reviews.llvm.org/D87468
Kirill Bobyrev [Wed, 16 Sep 2020 09:11:31 +0000 (11:11 +0200)]
[clangd] Implement hot index reloading for clangd-index-server
This patch adds a mechanism to load new versions of index into
clangd-index-server using SwapIndex and FileStatus information about last
modification time without downtime.
Reviewed By: kadircet
Differential Revision: https://reviews.llvm.org/D87450
Georgii Rymar [Wed, 16 Sep 2020 08:50:14 +0000 (11:50 +0300)]
[llvm-readobj][test] - Address a forgotten review comment for D86923.
Seems I've forgot to address this bit and this looks like a reason
of a failture on mac (http://45.33.8.238/mac/20491/step_11.txt).
Alok Kumar Sharma [Thu, 10 Sep 2020 06:23:43 +0000 (11:53 +0530)]
[DebugInfo][flang] DISubrange support for fortran assumed size array
This is needed to support assumed size array of fortran which can have missing upperBound/count
, contrary to current DISubrange support.
Example:
subroutine sub (array1, array2)
integer :: array1 (*)
integer :: array2 (4:9, 10:*)
array1(7:8) = 9
array2(5, 10) = 10
end subroutine
Now the validation check is relaxed for fortran.
Reviewed By: aprantl
Differential Revision: https://reviews.llvm.org/D87500
Sjoerd Meijer [Wed, 16 Sep 2020 08:34:31 +0000 (09:34 +0100)]
Follow up rG635b87511ec3: forgot to add/commit the new test file. NFC.
Sam Tebbs [Mon, 14 Sep 2020 14:44:54 +0000 (15:44 +0100)]
[ARM][LowOverheadLoops] Combine a VCMP and VPST into a VPT
This patch combines a VCMP followed by a VPST into a VPT, which has the
same semantics as the combination of the former two.
Aleksandr Platonov [Wed, 16 Sep 2020 08:04:53 +0000 (11:04 +0300)]
[clangd] Don't use zlib when it's unavailable.
Without this patch `clangd` crashes at try to load compressed string table when `zlib` is not available.
Example:
- Build `clangd` with MinGW (`zlib` found)
- Build index
- Build `clangd` with Visual Studio compiler (`zlib` not found)
- Try to load index
Reviewed By: sammccall, adamcz
Differential Revision: https://reviews.llvm.org/D87673
Yvan Roux [Wed, 16 Sep 2020 07:54:26 +0000 (09:54 +0200)]
[ARM][MachineOutliner] Add calls handling.
Handles calls inside outlined regions, by saving and restoring the link
register.
Differential Revision: https://reviews.llvm.org/D87136
Vitaly Buka [Wed, 16 Sep 2020 07:27:13 +0000 (00:27 -0700)]
Revert "[Asan] Accept __lsan_ignore_object for redzone pointer"
We still keep AddrIsInside.
This reverts commit
1d70984fa220f966ddcecd7906c5f10368fe1b93.
Vitaly Buka [Wed, 16 Sep 2020 07:03:07 +0000 (00:03 -0700)]
Revert "[Asan] Fix false leak report"
Additional investigated confirmed that issue is not about
AddrIsInside, but missing registers.
This reverts commit
9d01612db48fa27d18c6320974b8d711572e5c67.
Max Kazantsev [Wed, 16 Sep 2020 07:24:00 +0000 (14:24 +0700)]
[Test] Add positive range checks tests in addition to negative
Richard Barton [Wed, 16 Sep 2020 07:18:08 +0000 (08:18 +0100)]
[flang] Fix docs build
Apply a local fix to an issue with recommonmark's AutoStructify extension
when used with certain versions of sphinx.
See https://github.com/readthedocs/recommonmark/issues/93
Reviewed By: hans
Differential Revision: https://reviews.llvm.org/D87714
Max Kazantsev [Wed, 16 Sep 2020 06:59:41 +0000 (13:59 +0700)]
[Test] Some more potential range check elimination opportunities
Alina Sbirlea [Wed, 16 Sep 2020 05:52:42 +0000 (22:52 -0700)]
[MemorySSA] Report unoptimized as None, not MayAlias.
Xing GUO [Wed, 16 Sep 2020 06:46:12 +0000 (14:46 +0800)]
[obj2yaml] Add support for dumping the .debug_addr(v5) section.
This patch adds support for dumping the .debug_addr(v5) section to
obj2yaml.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D87601
Martin Storsjö [Wed, 5 Aug 2020 08:00:21 +0000 (11:00 +0300)]
[llvm-rc] Lowercase the option definitions. NFC.
This matches how such options are most commonly defined in other tools.
This was pointed out in an earlier review a few months ago, that
the llvm-rc td entries felt shouty.
The INCLUDE option is renamed to includepath, to avoid clashing with
the tablegen include directive.
Martin Storsjö [Fri, 31 Jul 2020 14:41:20 +0000 (17:41 +0300)]
[llvm-rc] Update a comment. NFC.
Fix a typo and mention one missing step.
Martin Storsjö [Sat, 12 Sep 2020 21:24:26 +0000 (00:24 +0300)]
[llvm-rc] Allow omitting components from VERSIONINFO versions
MS rc.exe doesn't require specifying all 4 components.
Differential Revision: https://reviews.llvm.org/D87570
Alina Sbirlea [Wed, 16 Sep 2020 00:42:03 +0000 (17:42 -0700)]
[MemorySSA] Set MustDominate to true for PhiTranslation.
Craig Topper [Wed, 16 Sep 2020 06:22:53 +0000 (23:22 -0700)]
[X86] Don't scalarize gather/scatters with non-power of 2 element counts. Widen instead.
We can pad the mask with zeros in order to widen. We already do
this for power 2 types that are smaller than a legal type.
Craig Topper [Wed, 16 Sep 2020 05:45:50 +0000 (22:45 -0700)]
[X86] Add test case for non-power of 2 scatter. NFC
Mircea Trofin [Wed, 16 Sep 2020 05:40:13 +0000 (22:40 -0700)]
[ThinLTO] Relax thinlto_embed_bitcode.ll check
Fixes fuscia test [1] - the thinlto annotations may not always be there.
[1] http://lab.llvm.org:8011/builders/fuchsia-x86_64-linux/builds/11312
Daniel Stone [Tue, 15 Sep 2020 17:01:04 +0000 (13:01 -0400)]
libclc: Add a __builtin to let SPIRV targets select between SW and HW FMA
Reviewer: jenatali jvesely
Differential Revision: https://reviews.llvm.org/D85910
Uday Bondhugula [Tue, 15 Sep 2020 05:28:45 +0000 (10:58 +0530)]
[MLIR][NFC] Value print update for block arguments
Emit some more information when printing/dumping `Value`s of
`BlockArgument` kind. This is purely to help for debugging purposes.
Differential Revision: https://reviews.llvm.org/D87670
Dave Lee [Mon, 14 Sep 2020 20:53:50 +0000 (13:53 -0700)]
[lldb] Reword CompilerType docstring to not say "generic type"
Since "generic type" has a precise meaning in some languages, reword the docstring of `CompilerType` to avoid ambiguity.
Differential Revision: https://reviews.llvm.org/D87633
Zinovy Nis [Mon, 14 Sep 2020 19:08:00 +0000 (22:08 +0300)]
[clang-tidy] Fix crash in modernize-use-noexcept on uninstantiated throw class
Bug: https://bugs.llvm.org/show_bug.cgi?id=47446
Differential Revision: https://reviews.llvm.org/D87627
Max Kazantsev [Wed, 16 Sep 2020 04:30:21 +0000 (11:30 +0700)]
[Test] Add signed version of a test
Serguei Katkov [Mon, 7 Sep 2020 05:56:34 +0000 (12:56 +0700)]
[InstCombine] Add tests for statepoint simplification
This tests increase coverage for change introduced in D85959
Reviewers: reames, reames
Reviewed By: reames
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D87224
Mehdi Amini [Wed, 16 Sep 2020 03:40:36 +0000 (03:40 +0000)]
Fully qualify some more namespace in MLIR ODS to be more friendly to dialects not defined under the mlir namespace (NFC)
Arthur Eubanks [Wed, 16 Sep 2020 03:29:20 +0000 (20:29 -0700)]
[NewPM] Fix opt-hot-cold-split.ll under NPM
Pin to legacy PM, there are already NPM RUN lines.
Arthur Eubanks [Wed, 16 Sep 2020 03:25:35 +0000 (20:25 -0700)]
[NewPM][SCEV] Fix constant-fold-gep.ll under NPM
Arthur Eubanks [Wed, 16 Sep 2020 03:21:45 +0000 (20:21 -0700)]
[NewPM] Fix 2003-02-19-LoopInfoNestingBug.ll under NPM
Also move it to a more appropriate directory.
Craig Topper [Wed, 16 Sep 2020 02:31:48 +0000 (19:31 -0700)]
[X86] Always use 16-bit displacement in 16-bit mode when there is no base or index register.
Previously we only did this if the immediate fit in 16 bits, but
the GNU assembler seems to just truncate.
Fixes PR46952
Alina Sbirlea [Wed, 16 Sep 2020 02:12:10 +0000 (19:12 -0700)]
Fix test after D86156.
Reid Kleckner [Wed, 16 Sep 2020 01:50:34 +0000 (18:50 -0700)]
[PDB] Drop LF_PRECOMP from debugTypes earlier
This is a minor simplification to avoid firing up a BinaryStreamReader
and CVType parser.
Krzysztof Parzyszek [Wed, 16 Sep 2020 01:32:09 +0000 (20:32 -0500)]
[Hexagon] Replace incorrect pattern for vpackl HWI32 -> HVi8
V6_vdealb4w is not correct for pairs, use V6_vpackeh/V6_vpackeb instead.
Arthur Eubanks [Mon, 14 Sep 2020 21:37:46 +0000 (14:37 -0700)]
[NewPM] Port strip* passes to NPM
strip-nondebug and strip-debug-declare have no existing associated tests
Reviewed By: ychen
Differential Revision: https://reviews.llvm.org/D87639
Arthur Eubanks [Tue, 15 Sep 2020 22:02:23 +0000 (15:02 -0700)]
[LowerSwitch][NewPM] Port lowerswitch to NPM
Reviewed By: ychen
Differential Revision: https://reviews.llvm.org/D87726
Michael Kitzan [Wed, 16 Sep 2020 00:50:48 +0000 (17:50 -0700)]
Test commit
Wenlei He [Wed, 16 Sep 2020 00:29:32 +0000 (17:29 -0700)]
SVML support for log10, sqrt
Although LLVM supports vectorization of loops containing log10/sqrt, it did not support using SVML implementation of it. Added support so that when clang is invoked with -fveclib=SVML now an appropriate SVML library log2 implementation will be invoked.
Follow up on: https://reviews.llvm.org/D77114
Tests:
Added unit tests to svml-calls.ll, svml-calls-finite.ll. Can be run with llvm-lint.
Created a simple c++ file that tests log10/sqrt, and used clang+ to build it, and output final assembly.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D87169
Wenlei He [Wed, 16 Sep 2020 00:21:32 +0000 (17:21 -0700)]
[LICM] Make Loop ICM profile aware again
D65060 was reverted because it introduced non-determinism by using BFI counts from already freed blocks. The parent of this revision fixes that by using a VH callback on blocks to prevent this from happening and makes sure BFI data is passed correctly in LoopStandardAnalysisResults.
This re-introduces the previous optimization of using BFI data to prevent LICM from hoisting/sinking if the instruction will end up moving to a colder block.
Internally at Facebook this change results in a ~7% win in a CPU related metric in one of our big services by preventing hoisting cold code into a hot pre-header like the added test case demonstrates.
Testing:
ninja check
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D87551
Jessica Paquette [Thu, 10 Sep 2020 20:34:15 +0000 (13:34 -0700)]
[AArch64][GlobalISel] Refactor + improve CMN, ADDS, and ADD emit functions
These functions were extremely similar:
- `emitADD`
- `emitADDS`
- `emitCMN`
Refactor them a little, introducing a more generic `emitInstr` function to
do most of the work.
Also add support for the immediate + shifted register addressing modes in each
of them.
Update select-uaddo.mir to show that selecing ADDS now supports folding
immediates + shifts. (I don't think this can impact CMN, because the CMN checks
require a G_SUB with a non-constant on the RHS.)
This is around a 0.02% code size improvement on CTMark at -O3.
Differential Revision: https://reviews.llvm.org/D87529
Arthur Eubanks [Mon, 14 Sep 2020 18:06:36 +0000 (11:06 -0700)]
[CGSCC][NewPM] Fix adding mutually recursive new functions
When adding a new function via addNewFunctionIntoRefSCC(), it creates a
new node and immediately populates the edges. Since populateSlow() calls
G->get() on all referenced functions, it will create a node (but not
populate it) for functions that haven't yet been added. If we add two
mutually recursive functions, the assert that the node should never have
been created will fire when the second function is added. So here we
remove that assert since the node may have already been created (but not
yet populated).
createNode() is only called from addNewFunctionInto{,Ref}SCC().
https://bugs.llvm.org/show_bug.cgi?id=47502
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D87623
Volkan Keles [Tue, 15 Sep 2020 23:40:38 +0000 (16:40 -0700)]
GlobalISel: Fix a failing combiner test
test/CodeGen/AArch64/GlobalISel/combine-trunc.mir was failing
due to the different order for evaluating function arguments.
This patch updates the related code to fix the issue.
Alexandre Ganea [Tue, 15 Sep 2020 23:18:24 +0000 (19:18 -0400)]
[llvm][cmake] Change LLVM_INTEGRATED_CRT_ALLOC to a path instead of a boolean
Differential Revision: https://reviews.llvm.org/D87609
Wenlei He [Tue, 15 Sep 2020 23:09:30 +0000 (16:09 -0700)]
[BFI] Make BFI information available through loop passes inside LoopStandardAnalysisResults
~~D65060 uncovered that trying to use BFI in loop passes can lead to non-deterministic behavior when blocks are re-used while retaining old BFI data.~~
~~To make sure BFI is preserved through loop passes a Value Handle (VH) callback is registered on blocks themselves. When a block is freed it now also wipes out the accompanying BFI entry such that stale BFI data can no longer persist resolving the determinism issue. ~~
~~An optimistic approach would be to incrementally update BFI information throughout the loop passes rather than only invalidating them on removed blocks. The issues with that are:~~
~~1. It is not clear how BFI information should be incrementally updated: If a block is duplicated does its BFI information come with? How about if it's split/modified/moved around? ~~
~~2. Assuming we can address these problems the implementation here will be a massive undertaking. ~~
~~There's a known need of BFI in LICM analysis which requires correct but not incrementally updated BFI data. A follow-up change can register BFI in all loop passes so this preserved but potentially lossy data is available to any loop pass that wants it.~~
See: D75341 for an identical implementation of preserving BFI via VH callbacks. The previous statements do still apply but this change no longer has to be in this diff because it's already upstream 😄 .
This diff also moves BFI to be a part of LoopStandardAnalysisResults since the previous method using getCachedResults now (correctly!) statically asserts (D72893) that this data isn't static through the loop passes.
Testing
Ninja check
Reviewed By: asbirlea, nikic
Differential Revision: https://reviews.llvm.org/D86156
Aditya Nandakumar [Tue, 15 Sep 2020 23:06:55 +0000 (16:06 -0700)]
[GISel] Add new GISel combiners for G_MUL
https://reviews.llvm.org/D87668
Patch adds two new GICombinerRules, one for G_MUL(X, 1) and another for G_MUL(X, -1).
G_MUL(X, 1) is an identity combine, and G_MUL(X, -1) gets replaced with G_SUB(0, X).
Patch additionally adds new combiner tests for the AArch64 target to test these
new combiner rules, as well as updates AMDGPU GISel tests.
Patch by mkitzan
Mircea Trofin [Mon, 14 Sep 2020 17:45:00 +0000 (10:45 -0700)]
[ThinLTO] add post-thinlto-merge option to -lto-embed-bitcode
This will embed bitcode after (Thin)LTO merge, but before optimizations.
In the case the thinlto backend is called from clang, the .llvmcmd
section is also produced. Doing so in the case where the caller is the
linker doesn't yet have a motivation, and would require plumbing through
command line args.
Differential Revision: https://reviews.llvm.org/D87636
Jan Korous [Wed, 19 Aug 2020 05:36:16 +0000 (22:36 -0700)]
[SourceManager] Explicitly check for potential iterator underflow
Differential Revision: https://reviews.llvm.org/D86231
Volkan Keles [Tue, 15 Sep 2020 22:50:34 +0000 (15:50 -0700)]
GlobalISel: Add combines for G_TRUNC
https://reviews.llvm.org/D87050
Stanislav Mekhanoshin [Thu, 10 Sep 2020 22:10:52 +0000 (15:10 -0700)]
[AMDGPU] Unify intrinsic ret/nortn interface
We have a single noret intrinsic an a lot of special handling
around it. Declare it just as any other but do not define rtn
instructions itself instead.
Differential Revision: https://reviews.llvm.org/D87719
Xun Li [Tue, 15 Sep 2020 22:19:57 +0000 (15:19 -0700)]
[TSAN] Handle musttail call properly in EscapeEnumerator (and TSAN)
Call instructions with musttail tag must be optimized as a tailcall, otherwise could lead to incorrect program behavior.
When TSAN is instrumenting functions, it broke the contract by adding a call to the tsan exit function inbetween the musttail call and return instruction, and also inserted exception handling code.
This happend throguh EscapeEnumerator, which adds exception handling code and returns ret instructions as the place to insert instrumentation calls.
This becomes especially problematic for coroutines, because coroutines rely on tail calls to do symmetric transfers properly.
To fix this, this patch moves the location to insert instrumentation calls prior to the musttail call for ret instructions that are following musttail calls, and also does not handle exception for musttail calls.
Differential Revision: https://reviews.llvm.org/D87620
Joseph Huber [Tue, 15 Sep 2020 19:04:37 +0000 (15:04 -0400)]
[OpenMP] Additional Information for Libomptarget Mappings
Summary:
This patch adds additonal support for priting infromation from Libomptarget for
already existing maps and printing the final data mapped on the device at
device destruction.
Reviewers: jdoerfort gkistanova
Subscribers: guansong openmp-commits sstefan1 yaxunl
Tags: #OpenMP
Differential Revision: https://reviews.llvm.org/D87722
Alexey Bataev [Tue, 15 Sep 2020 19:57:11 +0000 (15:57 -0400)]
[OPENMP]Fix codegen for is_device_ptr component, captured by reference.
Need to map the component as TO instead of the literal, because need to
pass a reference to a component if the pointer is overaligned.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D84887
Diego Caballero [Fri, 4 Sep 2020 18:44:32 +0000 (11:44 -0700)]
[mlir] Rename 'setInsertionPointAfter' to avoid ambiguity
Rename 'setInsertionPointAfter(Value)' API to avoid ambiguity with
'setInsertionPointAfter(Operation *)' for SingleResult operations which
implicitly convert to Value (see D86756).
Differential Revision: https://reviews.llvm.org/D87155
Guillaume Chatelet [Tue, 15 Sep 2020 20:48:08 +0000 (20:48 +0000)]
[libc] Remove special case for 8 and 16 bytes
They don't seem to gain much in real apps and its better to favor less branches and smaller code.
Saleem Abdulrasool [Tue, 8 Sep 2020 22:49:41 +0000 (22:49 +0000)]
Sema: add support for `__attribute__((__swift_bridged_typedef__))`
Extend the semantic attributes that clang processes for Swift to include
`swift_bridged_typedef`. This attribute enables typedefs to be bridged
into Swift with a bridged name.
This is based on the work of the original changes in
https://github.com/llvm/llvm-project-staging/commit/
8afaf3aad2af43cfedca7a24cd817848c4e95c0c
Differential Revision: https://reviews.llvm.org/D87396
Reviewed By: Aaron Ballman