Chris Bieneman [Tue, 27 Sep 2022 02:44:13 +0000 (21:44 -0500)]
[CMake] Add `CLANG_ENABLE_HLSL` CMake option
The HLSL support in clang is in progress and not fully functioning. As
such we don't want to install the related optional build components by
default (yet), but we do need an option to build and install them
locally for testing and for some key users.
This adds the `CLANG_ENABLE_HLSL` option which is off by default and can
be enabled to install the HLSL clang headers and the clang-dxc symlink.
Reviewed By: phosek
Differential Revision: https://reviews.llvm.org/D134693
Krzysztof Parzyszek [Tue, 27 Sep 2022 17:03:44 +0000 (10:03 -0700)]
[Hexagon] Move function to a different class, NFC
"Sector" is a concept from AlignVectors, so the check for it
should be there.
Craig Topper [Tue, 27 Sep 2022 17:25:18 +0000 (10:25 -0700)]
[Driver] Silence a -Wparentheses error. NFC
Doru Bercea [Tue, 27 Sep 2022 01:51:17 +0000 (01:51 +0000)]
Move allocas converted from __kmpc_alloc_shared to entry block.
Gergely Nagy [Tue, 27 Sep 2022 11:59:55 +0000 (07:59 -0400)]
[libcxx] Make stdatomic.h work when included from a C source file
If a C source file includes the libc++ stdatomic.h, compilation will
break because (a) the C++ standard check will fail (which is expected),
and (b) `_LIBCPP_COMPILER_CLANG_BASED` won't be defined because the
logic defining it in `__config` is guarded by a `__cplusplus` check, so
we'll end up with a blank header. Move the detection logic outside of
the `__cplusplus` check to make the second check pass even in a C context
when you're using Clang. Note that `_LIBCPP_STD_VER` is not defined when
in C mode, hence stdatomic.h needs to check if in C++ mode before using
that macro to avoid a warning.
In an ideal world, a C source file wouldn't be including the libc++
header directory in its search path, so we'd never have this issue.
Unfortunately, certain build environments make this hard to guarantee,
and in this case it's easy to tweak this header to make it work in a C
context, so I'm hoping this is acceptable.
Fixes https://github.com/llvm/llvm-project/issues/57710.
Differential Revision: https://reviews.llvm.org/D134591
Lei Zhang [Tue, 27 Sep 2022 17:08:22 +0000 (13:08 -0400)]
[mlir][spirv] Use "SPIR-V" in the doc after dialect renaming
Craig Topper [Tue, 27 Sep 2022 17:04:30 +0000 (10:04 -0700)]
[Driver] Prevent Mips specific code from claiming -mabi argument on other targets.
Fixes PR57976.
Reviewed By: erichkeane, arichardson, MaskRay
Differential Revision: https://reviews.llvm.org/D134671
Lang Hames [Tue, 27 Sep 2022 15:59:58 +0000 (08:59 -0700)]
[ORC][JITLink] Retain Weak flags in JITDylib interfaces, propagate to LinkGraph.
Previously we stripped Weak flags from JITDylib symbol table entries once they
were resolved (there was no particularly good reason for this). Now we want to
retain them and query them when setting the Linkage on external symbols in
LinkGraphs during symbol resolution (this was the motivation for
75404e9ef88).
Making weak linkage of external definitions discoverable in the LinkGraph will
in turn allow future plugins to implement correct handling for them (by
recording locations that depend on exported weak definitions and pointing all
of these at one chosen definition at runtime).
Lang Hames [Tue, 27 Sep 2022 15:37:15 +0000 (08:37 -0700)]
[JITLink] Fix typo in debugging output.
Michael Buch [Mon, 26 Sep 2022 16:38:37 +0000 (18:38 +0200)]
[lldb][TypeSystemClang] Honor DW_AT_rvalue_reference when creating C++ FunctionPrototypes
Currently funciton lookup in the expression evaluator
fails to disambiguate member functions the are overloaded
on lvalue/rvalue reference-qualifiers. This happens because
we unconditionally set a `FunctionPrototype`s
`ExtProtoInfo::RefQualifier` to `RQ_None`. We lose
the ref-qualifiers in the synthesized AST and `clang::Sema`
fails to pick a correct overload candidate.
DWARF emits information about a function's ref-qualifiers
in the form of a boolean `DW_AT_rvalue_reference` (for rvalues)
and `DW_AT_reference` (for lvalues).
This patch sets the `FunctionPrototype::ExtProtoInfo::RefQualifier`
based on the DWARF attributes above.
**Testing**
* Added API test
llvm/llvm-project issue #57866
Differential Revision: https://reviews.llvm.org/D134661
Craig Topper [Tue, 27 Sep 2022 16:35:21 +0000 (09:35 -0700)]
[VP][RISCV] Add vp.fmuladd.
Expanded in SelectionDAGBuilder similar to llvm.fmuladd.
Reviewed By: frasercrmck, simoll
Differential Revision: https://reviews.llvm.org/D134474
Min-Yih Hsu [Wed, 3 Aug 2022 21:08:33 +0000 (14:08 -0700)]
[mlir][LLVMIR] Fix incorrect result type from llvm.fcmp
If any of the operands for FCmpOp is a vector, returns a vector<Nxi1>,
rather than an i1 type result.
Differential Revision: https://reviews.llvm.org/D134449
Krzysztof Drewniak [Tue, 27 Sep 2022 16:13:07 +0000 (16:13 +0000)]
[mlir][ExecutionEngine] Remove ScopedContext from ROCm wrappers
The push/pop context APIs are deprecated in HIP, and keeping the
default device set is handled in IHP using hipSetDevice().
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D134747
Stefan Gränitz [Tue, 27 Sep 2022 16:43:26 +0000 (18:43 +0200)]
[ObjC][ARC] Fix target register for call expanded from CALL_RVMARKER on Windows
Fix regression https://github.com/llvm/llvm-project/issues/56952 for Clang CodeGen on Windows. In the Windows ABI the instruction sequence that is expanded from CALL_RVMARKER should use RCX as target register and not RDI.
Reviewed By: rnk, fhahn
Differential Revision: https://reviews.llvm.org/D134441
Alex Zinenko [Tue, 27 Sep 2022 16:08:44 +0000 (16:08 +0000)]
[mlir] specify the values when notifying about op replacement
It is useful for PatternRewriter listeners to know the values that are
replacing the op in addition to only the fact of the op being replaced
for being able to keep track of changes or for debugging.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D134748
David Green [Tue, 27 Sep 2022 16:08:17 +0000 (17:08 +0100)]
[AArch64] Remove incorrect zero element insert-bitcast patterns
These two patterns are not working as intended, as shown in D134022.
They need to insert the value into the new register, not override it.
Joseph Huber [Mon, 26 Sep 2022 18:39:11 +0000 (13:39 -0500)]
[LinkerWrapper] Fix optimized debugging builds for NVPTX LTO
The ptxas assembler does not allow the `-g` flag along with
optimizations. Normally this is degraded to line info in the driver, but
when using LTO we did not have this step and the linker wrapper was not
correctly degrading the option. Note that this will not work if the user
does not pass `-g` again to the linker invocation. That will require
setting some flags in the binary to indicate that debugging was used
when building.
This fixes #57990
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D134660
Philip Reames [Tue, 27 Sep 2022 15:20:49 +0000 (08:20 -0700)]
[RISCV] Rename getVectorImmCost to getStoreImmCost [nfc]
My original intent had been to reuse this for arithmetic instructions as well, but due to the availability of a immediate splat encoding there, we will need different heuristics. So specialize the existing code for the store case.
Lorenzo Chelini [Tue, 27 Sep 2022 15:16:43 +0000 (11:16 -0400)]
[MLIR] Expose `getAsValues` in `StaticValueUtils.h` (NFC) [reland]
The utility function should live in `StaticValueUtils.h` as it provides
a convenient way to convert a vector of OpFoldResults into a vector of
Values.
Reviewed By: nicolasvasilache, cota
Differential Revision: https://reviews.llvm.org/D134451
Philip Reames [Tue, 27 Sep 2022 14:49:54 +0000 (07:49 -0700)]
[RISCV][SLP] Add test coverage for stores of constants
Simon Pilgrim [Tue, 27 Sep 2022 14:44:21 +0000 (15:44 +0100)]
[TTI] getExtractWithExtendCost - remove default Index = -1 value
We currently always specify the index (even if we set it to -1), and a future patch will need to adjust this method to make it more compatible with getScalarizationOverhead
Bernhard Urban-Forster [Tue, 27 Sep 2022 14:35:57 +0000 (10:35 -0400)]
[lld-macho] Fix -force_symbols_{,not_}weak_list arg type
This is still undocumented and unsupported, but if someone passed it before you
would end up with a missing file error since this takes an argument that
wouldn't be handled.
Differential Revision: https://reviews.llvm.org/D133854
Krzysztof Drewniak [Mon, 26 Sep 2022 20:46:17 +0000 (20:46 +0000)]
[mlir] Correctly handle HIP_PATH when ROCM_PATH is set
With the CMake file as written, if code elsewhere had set ROCM_PATH,
then HIP_PATH would not be set, breaking the rest of the ROCm
execution utility handling.
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D134674
Oleg Shyshkov [Tue, 27 Sep 2022 14:26:56 +0000 (14:26 +0000)]
[mlir][linalg] Add getIteratorTypesArray to LinalgInterface.
Summary:
Most of the code that gets `iterator_types` from LinalgInterface is forced to
extract values from an `Attribute`. As a result, the usage pattern looks like
this:
```
SmallVector<StringRef> iterators = llvm::to_vector<4>(linalgOp.iterator_types().getAsValueRange<StringAttr>());
```
It also forces all operations that implement LinalgOp interface to have
`iterator_types` attribute even when the information can be easily infered from
other parameters.
In perfect future, `getIteratorTypeArray` should be the only method to get
iterator types from the interface. The default implementation can rely on
`iterator_types` attribute though.
The name `getIteratorTypeArray` was picked to be consistent with existing
`getIndexingMapsArray`.
This patch add a few sample usages. More cleanups will follow.
Differential Revision: https://reviews.llvm.org/D134729
Philip Reames [Tue, 27 Sep 2022 14:22:41 +0000 (07:22 -0700)]
[LV] Adjust cost model to use uniform store lowering for unpredicated uniform stores
Follow up to D133580; adjust the cost model to prefer uniform store lowering for scalable stores which are unpredicated.
The impact here isn't in the uniform store lowering quality itself. InstCombine happily converts the scatter form into the single store form. The main impact is in letting the rest of the cost model make choices based on the knowledge that the vector will be scalarized on use.
Differential Revision: https://reviews.llvm.org/D134460
Simon Pilgrim [Tue, 27 Sep 2022 14:15:43 +0000 (15:15 +0100)]
Fix 'unused-lambda-capture' gcc warning. NFCI.
Simon Pilgrim [Tue, 27 Sep 2022 13:48:59 +0000 (14:48 +0100)]
[SLP] Add ScalarizationOverheadBuilder helper to track vector extractions
Instead of accumulating all extraction costs separately and then adjusting for repeated subvector extractions, this patch collects all the extractions and then converts to calls to getScalarizationOverhead to improve the accuracy of the costs.
I'm not entirely satisfied with the getExtractWithExtendCost handling yet - this still just adds all the getExtractWithExtendCost costs together - it really needs to be replaced with a "getScalarizationOverheadWithExtend", but that will require further refactoring first.
This replaces my initial attempt in D124769.
Differential Revision: https://reviews.llvm.org/D134605
Erich Keane [Mon, 26 Sep 2022 18:02:39 +0000 (11:02 -0700)]
Fix regression from Deferred Concepts with lambda in var init
As reported in GH #57945, this would crash because the decl context for
the lambda was being loaded via 'getNonClosureContext', which only gets
CODE contexts, so a global lambda was getting 'nullptr' here instead.
This patch does some work to make sure we get a valid/valuable
declcontext here instead.
Michael Buch [Tue, 27 Sep 2022 10:40:13 +0000 (11:40 +0100)]
[lldb][test] Remove failing ValueCheck on deprecated libcxx entity
A recent libcxx change renamed all internal variables starting with
`__`. As such, `std::reverse_iterator::__t` was renamed to
`std::reverse_iterator::__t_`. This breaks the `import-std-module`
tests with newer libcxx versions. Since this variable is deprecated
in libcxx anyway, this patch simply removes the explicit `ValueCheck`
on the variable name. We don't lose any interesting test-case here
since the purpose of the test is to see if we can call functions
from the `std` module.
We can now re-enable the tests on Darwin for all buildbot Clang
compiler variants.
Differential Revision: https://reviews.llvm.org/D134727
Michael Buch [Tue, 27 Sep 2022 13:00:24 +0000 (14:00 +0100)]
[lldb][test] Disable TestSBValueUnsignedEnumBitField.py for old DWARF versions
With older DWARF versions we don't encode the enum's underlying
type in DWARF. In those cases LLDB sign-extends the bitfield as
a signed integer. Without the actual enum type being present in
DWARF there's not much we can do.
Differential Revision: https://reviews.llvm.org/D134734
Amaury Séchet [Fri, 5 Aug 2022 13:33:07 +0000 (13:33 +0000)]
[DAG] select Cond, -1, C --> or (sext Cond), C if Cond is MVT::i1
This seems to be beneficial overall, except for midpoint-int.ll .
The X86 backend seems to generate zeroing that are not necesary.
Reviewed By: shchenz
Differential Revision: https://reviews.llvm.org/D131260
wanglei [Tue, 27 Sep 2022 11:03:18 +0000 (19:03 +0800)]
[LoongArch] Add some comments for expand pseudo-inst pass. NFC
Differential Revision: https://reviews.llvm.org/D134708
Carlos Alberto Enciso [Tue, 27 Sep 2022 11:48:44 +0000 (12:48 +0100)]
[ADT] Add IntervalTree - light tree data structure to hold intervals.
Fix build failure in:
https://lab.llvm.org/buildbot/#/builders/36/builds/25424
error: comparison of integers of different signs:
'const unsigned long' and 'const int' [-Werror,-Wsign-compare]
Reviewed By: Orlando
Differential Revision: https://reviews.llvm.org/D125776
Kazushi (Jam) Marukawa [Mon, 26 Sep 2022 23:36:02 +0000 (08:36 +0900)]
[VE] Change to expand FPOW
VE doesn't have FPOW instruction, so this patch makes llvm expand it.
Reviewed By: efocht
Differential Revision: https://reviews.llvm.org/D134695
WANG Xuerui [Tue, 27 Sep 2022 09:50:14 +0000 (17:50 +0800)]
[LoongArch] Support lowering frames larger than 2048 bytes
Differential Revision: https://reviews.llvm.org/D134582
Guray Ozen [Tue, 27 Sep 2022 10:23:18 +0000 (12:23 +0200)]
[mlir] Refactor transform dialect's gpu block func
This revision refactors gpu block id generator lambda that is used in the transform dialect. It removes the lambda and instead uses a static function that's name generateGpuBlockIds.
It also simplifies arguments that the function takes.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D134724
Alvin Wong [Tue, 27 Sep 2022 10:09:59 +0000 (13:09 +0300)]
[lldb] Improve display of absolute symbol lookup
When running `target module lookup` command, show the name of absolute
symbols. Also fix indentation issue after printing an absolute symbol.
Reviewed By: clayborg, DavidSpickett
Differential Revision: https://reviews.llvm.org/D134516
Alvin Wong [Tue, 27 Sep 2022 10:09:45 +0000 (13:09 +0300)]
[lldb] Add newline in output of `target modules lookup`
This adds a line break between each result address in the output of the
lldb command `target modules lookup`. Before this change, a new address
result will be printed on the same line as the summary of the last
result, making the output difficult to view.
Also adds a test for this command.
Reviewed By: labath
Differential Revision: https://reviews.llvm.org/D134111
Simon Pilgrim [Tue, 27 Sep 2022 10:01:17 +0000 (11:01 +0100)]
[X86] Clean up prefixes for avoid-sfb.ll
Simplifies the diff for D134697
Florian Hahn [Tue, 27 Sep 2022 08:58:48 +0000 (09:58 +0100)]
[LSR] Preserve LCSSA in expander when rewriting loop exit values.
The expanded values when rewriting exit values need to preserve LCSSA.
Ask SCEVExpander to preserve LCSSA to ensure that.
Fixes #58007.
Nikita Popov [Tue, 27 Sep 2022 08:25:58 +0000 (10:25 +0200)]
[BasicAA] Use ScopeExit to clear Visited set (NFC)
Daniel Kiss [Tue, 27 Sep 2022 08:22:20 +0000 (10:22 +0200)]
[AArch64] Add all predecessor archs in target info
A given function is compatible with all previous arch versions.
To avoid compering values of the attribute this logic adds all predecessor
architecture values.
Reviewed By: dmgreen, DavidSpickett
Differential Revision: https://reviews.llvm.org/D134353
Nikita Popov [Fri, 23 Sep 2022 11:03:59 +0000 (13:03 +0200)]
[FunctionAttrs] Infer precise FMRB
This updates checkFunctionMemoryAccess() to infer a precise
FunctionModRefBehavior, rather than an approximation split into
read/write and argmemonly.
Afterwards, we still map this back to imprecise function attributes.
This still allows us to infer some cases that we previously did not
handle, namely inaccessiblememonly and inaccessiblemem_or_argmemonly.
In practice, this means we get better memory attributes in the
presence of intrinsics like @llvm.assume.
Differential Revision: https://reviews.llvm.org/D134527
Nikita Popov [Mon, 26 Sep 2022 13:42:35 +0000 (15:42 +0200)]
[libcxx] Support LIBCXX_STATICALLY_LINK_ABI_IN_STATIC_LIBRARY + LIBCXX_ENABLE_ABI_LINKER_SCRIPT
Currently if both LIBCXX_STATICALLY_LINK_ABI_IN_STATIC_LIBRARY (only
applies to static library) and LIBCXX_ENABLE_ABI_LINKER_SCRIPT (only
applies to shared library) are enabled, the former will be silently
ignored. The shared library will use a linker script, while the
static library fails to link libc++abi.a entirely.
This is caused by what appears to be an implementation bug: The
LIBCXX_STATICALLY_LINK_ABI_IN_*_LIBRARY options are declared as
dependent options of LIBCXX_ENABLE_STATIC_ABI_LIBRARY, rather than
simply using it as the default value.
Of course, the combination of
LIBCXX_STATICALLY_LINK_ABI_IN_SHARED_LIBRARY and
LIBCXX_ENABLE_ABI_LINKER_SCRIPT still results in a cmake error,
because these would be conflicting requests for the shared library.
Differential Revision: https://reviews.llvm.org/D134644
David Sherwood [Thu, 15 Sep 2022 12:18:14 +0000 (12:18 +0000)]
[AArch64] Add Neoverse V2 CPU support
Adds support for the Neoverse V2 CPU to the AArch64 backend.
Differential Revision: https://reviews.llvm.org/D134352
Florian Hahn [Tue, 27 Sep 2022 07:47:19 +0000 (08:47 +0100)]
[LoopUnroll] Forget block and loop dispositions during unrolling.
After unrolling a loop, the block and loop dispositions need to be
cleared. As we don't know which SCEVs in the loop/blocks may be
impacted, completely clear the cache. This should also fix some cases
where deleted loops remained in the LoopDispositions cache.
This fixes a verification failure surfaced by D134531.
I am planning on reviewing/updating the existing uses of
forgetLoopDispositions to check if they should be replaced by
forgetBlockAndLoopDispositions.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D134612
LLVM GN Syncbot [Tue, 27 Sep 2022 07:24:34 +0000 (07:24 +0000)]
[gn build] Port
6584d1f9309b
Carlos Alberto Enciso [Tue, 27 Sep 2022 07:22:28 +0000 (08:22 +0100)]
[ADT] Add IntervalTree - light tree data structure to hold intervals.
It allows finding all intervals that overlap with any given point.
At this time, it does not support any deletion or rebalancing
operations.
The IntervalTree is designed to be set up once, and then queried
without any further additions.
Reviewed By: psamolysov, probinson
Differential Revision: https://reviews.llvm.org/D125776
Paulo Matos [Tue, 27 Sep 2022 07:08:43 +0000 (09:08 +0200)]
[WebAssembly] Use intrinsics for table.get/set instructions
Initial table.get/set implementation would match and lower combinations
of GEP+load/store to table.get/set instructions. However, this is error
prone due to potential combinations of GEP+load/store we don't implement,
and load/store optimizations. By changing the code to using intrinsics, we
avoid both issues and simplify the code.
New builtins implemented:
* @llvm.wasm.table.get.externref
* @llvm.wasm.table.get.funcref
* @llvm.wasm.table.set.externref
* @llvm.wasm.table.set.funcref
Reviewed By: asb, tlively
Differential Revision: https://reviews.llvm.org/D134436
Fraser Cormack [Tue, 27 Sep 2022 06:31:21 +0000 (07:31 +0100)]
[Support][NFC] Fix typo in comment
Vitaly Buka [Tue, 27 Sep 2022 05:50:49 +0000 (22:50 -0700)]
[test][clangd] Try to deflake with sanitizers
Test sometimes fail with:
https://lab.llvm.org/buildbot/#/builders/168/builds/9110
Yeting Kuo [Mon, 26 Sep 2022 12:41:35 +0000 (20:41 +0800)]
[VP][RISCV] Add vp.maxnum and vp.minnum intrinsics and RISC-V support.
Add vp.maxnum and vp.minnum which are vector predicted intrinsics of llvm.maxnum
and llvm.minnum.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D134639
Vitaly Buka [Tue, 27 Sep 2022 05:16:34 +0000 (22:16 -0700)]
Revert "[AMDGPU] Move SIModeRegisterDefaults to SI MFI"
Break msan bots. Details in D134666.
This reverts commit
0ce96e06ee0226938e723bd0c8e16e3d2d51f203.
Katherine Rasmussen [Mon, 19 Sep 2022 17:37:01 +0000 (10:37 -0700)]
[flang] Write semantics test for atomic_define
Write a semantics test for the atomic intrinsic subroutine,
atomic_define.
Reviewed By: rouson
Differential Revision: https://reviews.llvm.org/D134288
jacquesguan [Mon, 26 Sep 2022 07:49:32 +0000 (15:49 +0800)]
[RISCV][NFC] Merge rv32 and rv64 insertelt/extract fp tests.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D134631
Paul Scoropan [Tue, 27 Sep 2022 02:03:49 +0000 (22:03 -0400)]
[PowerPC] XCOFF exception section support on the direct assembler path
This feature implements support for making entries in the exception section
on XCOFF on the direct assembly path using the ".except" pseudo-op. It also
provides functionality to lower entries (comprised of language and reason
codes) into the exception section through the use of annotation metadata
attached to llvm.ppc.trap/trapd/tw/tdw intrinsics. Integrated assembler
support will be provided in another review. https://reviews.llvm.org/D133030
needs to merge first for LIT tests
Reviewed By: shchenz, RKSimon
Differential Revision: https://reviews.llvm.org/D132146
Yaxun (Sam) Liu [Sat, 10 Sep 2022 01:54:49 +0000 (21:54 -0400)]
[HIP] Fix unbundling archive
HIP is able to unbundle archive of bundled bitcode.
However currently there are two bugs:
1. archives passed by -l: are not unbundled.
2. archives passed as input files are not unbundled
The actual file name of an archive passed by -l: should
not be prefixed with lib and appended with '.a',
but the file path is prefixed with paths in '-L' options.
The actual file name of an archive passed as an input file
stays the same, not affected by the '-L' options.
Sebastian Peryt [Tue, 27 Sep 2022 01:31:32 +0000 (18:31 -0700)]
[NFC][2/n] Remove PrunePH pass
Second patch in the series to remove legacy PM and
associated -enable-new-pm=0 flag targets pass that
has not been ported to new PM - PruneEH.
Discussion about this can be found in D44415.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D134686
Tue Ly [Fri, 23 Sep 2022 23:12:38 +0000 (19:12 -0400)]
[libc][math] Simplify tanf implementation and improve its performance.
Simplify `tanf` implementation and improve its performance.
Completely reuse the implementation of `sinf`, `cosf`, `sincosf` and use
the definition `tan(x) = sin(x)/cos(x)`.
Performance benchmark using perf tool from the CORE-MATH project on Ryzen 1700:
```
$ CORE_MATH_PERF_MODE="rdtsc" ./perf.sh tanf
GNU libc version: 2.35
GNU libc release: stable
CORE-MATH reciprocal throughput : 18.558
System LIBC reciprocal throughput : 49.919
BEFORE:
LIBC reciprocal throughput : 36.480
LIBC reciprocal throughput : 27.217 (with `-msse4.2` flag)
LIBC reciprocal throughput : 20.205 (with `-mfma` flag)
AFTER:
LIBC reciprocal throughput : 30.337
LIBC reciprocal throughput : 21.072 (with `-msse4.2` flag)
LIBC reciprocal throughput : 15.804 (with `-mfma` flag)
$ CORE_MATH_PERF_MODE="rdtsc" ./perf.sh tanf --latency
GNU libc version: 2.35
GNU libc release: stable
CORE-MATH latency : 56.702
System LIBC latency : 107.206
BEFORE
LIBC latency : 97.598
LIBC latency : 91.119 (with `-msse4.2` flag)
LIBC latency : 82.655 (with `-mfma` flag)
AFTER
LIBC latency : 74.560
LIBC latency : 66.575 (with `-msse4.2` flag)
LIBC latency : 61.636 (with `-mfma` flag)
```
Reviewed By: zimmermann6
Differential Revision: https://reviews.llvm.org/D134575
Yuanfang Chen [Tue, 27 Sep 2022 00:40:46 +0000 (17:40 -0700)]
[NFC][Clang] Remove dead code
And correct a few typos.
Xiang Li [Tue, 27 Sep 2022 01:08:41 +0000 (18:08 -0700)]
[NFC] legalize triple for dxil tests.
Vitaly Buka [Tue, 27 Sep 2022 01:06:41 +0000 (18:06 -0700)]
[test][msan] Fix more tests for D134669
Vitaly Buka [Sat, 27 Aug 2022 01:04:17 +0000 (18:04 -0700)]
[tests][msan] Use -fno-sanitize-memory-param-retval in tests
Supports either default of -f[no-]sanitize-memory-param-retval.
Reviewed By: aeubanks, MaskRay
Differential Revision: https://reviews.llvm.org/D134683
Adrian Prantl [Tue, 27 Sep 2022 00:16:51 +0000 (17:16 -0700)]
Revert "Skip crashing test"
This reverts commit
c1ce19021da0cf1c88722024e6ff9cee7aabc7b6.
Fangrui Song [Tue, 27 Sep 2022 00:16:44 +0000 (17:16 -0700)]
[AArch64] Check target feature support for __builtin_arm_crc*
This is the AArch64 counterpart of D134127.
Daniel Kiss will change more `BUILTIN` to `TARGET_BUILTIN`.
Fix #57802
Alex Yermolovich [Mon, 26 Sep 2022 23:23:37 +0000 (16:23 -0700)]
[DWARF DUMP] Fix infinite recursion in Type Printer.
There is an implicit circular dependency in a debug information coming from GCC.
This results in a coredump.
It tries to resolve scope for DIE TAGs that are scopeless.
Reviewed By: dblaikie
Differential Revision: https://reviews.llvm.org/D132900
Tony Tye [Thu, 22 Sep 2022 18:53:10 +0000 (18:53 +0000)]
[AMDGPU][NFC] Correct name of DW_OP_LLVM_aspace_bregx
Correct naming of DW_OP_LLVM_aspace_bregx in
AMDGPUDwarfExtensionsForHeterogeneousDebugging.rst .
Reviewed By: dblaikie
Differential Revision: https://reviews.llvm.org/D134467
Aart Bik [Mon, 26 Sep 2022 20:40:08 +0000 (13:40 -0700)]
[mlir][sparse] minor cleanup of sparse tensor dialect ops
Removed some whitespace, improved some of the ops doc.
Reviewed By: wrengr
Differential Revision: https://reviews.llvm.org/D134672
Mircea Trofin [Mon, 26 Sep 2022 19:20:37 +0000 (12:20 -0700)]
[nfc][mlgo] Lazily compute the regalloc reward
Differential Revision: https://reviews.llvm.org/D134664
Min-Yih Hsu [Mon, 25 Jul 2022 20:05:13 +0000 (13:05 -0700)]
[mlir][LLVMIR] Add translation of AtomicRMW/CmpXchg from LLVM IR
Add support for translating AtomicRMWInst and AtomicCmpXchgInst from
LLVM IR.
Differential Revision: https://reviews.llvm.org/D134450
Craig Topper [Mon, 26 Sep 2022 21:53:06 +0000 (14:53 -0700)]
[IR][InstCombine] Support scalable vector splats ConstantExprs in Constant::getUniqueInteger().
I've left the getAggregateElement as a fast path for non-ConstantExprs
to avoid a call to getSplatValue in release builds.
Fixes PR57989.
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D134670
Akira Hatanaka [Mon, 26 Sep 2022 21:19:53 +0000 (14:19 -0700)]
Check whether the allocated type is an array type before calling
checkArrayElementAlignment in Sema::BuildCXXNew
This commit fixes a bug that was introduced by
adaf62ced and reported
here: https://reviews.llvm.org/D133711#3814717
Fangrui Song [Mon, 26 Sep 2022 21:20:27 +0000 (14:20 -0700)]
[ELF] Rewrite R_RISCV_ALIGN nops when r.addend%4 != 0
For RVC, GNU assembler and LLVM integrated assembler add c.nop followed by a
sequence of 4-byte nops. Even if remove % 4 == 0, we have to split one 4-byte
nop and therefore need to write the code sequence, otherwise we create an
incorrect c.unimp.
Florian Hahn [Mon, 26 Sep 2022 21:11:32 +0000 (22:11 +0100)]
[LoopDeletion] Regenerate check lines for test added in
f076a7097999d.
Florian Hahn [Mon, 26 Sep 2022 20:59:40 +0000 (21:59 +0100)]
[LoopDeletion] Add cache verification failure test case.
Extra test coverage for D134613.
Francis Visoiu Mistrih [Sun, 11 Sep 2022 05:13:41 +0000 (22:13 -0700)]
[Matrix] Add tests for addition transpose optimizations
Tests before transpose optimizations around additions.
Differential Revision: https://reviews.llvm.org/D133656
Stanislav Mekhanoshin [Mon, 26 Sep 2022 19:40:22 +0000 (12:40 -0700)]
[AMDGPU] Move SIModeRegisterDefaults to SI MFI
It does not belong to a general AMDGPU MFI.
Differential Revision: https://reviews.llvm.org/D134666
Lorenzo Chelini [Mon, 26 Sep 2022 20:11:40 +0000 (22:11 +0200)]
Revert "[MLIR] Expose `getAsValues` in `StaticValueUtils.h` (NFC)"
It introduces a circular build dependence: DialectUtils <-
ArithmeticUtils <- ArithDialect <- DialectUtils
This reverts commit
27224fe7272a791bcc9f28c997ce322f7d3856cd.
Sanjay Patel [Mon, 26 Sep 2022 19:59:16 +0000 (15:59 -0400)]
[InstCombine] add assert/test for zext to i1
This is a test to verify that we do not crash with the
problem noted in issue #57986. The root problem should
be fixed with a prior change to InstSimplify.
Sanjay Patel [Mon, 26 Sep 2022 19:15:20 +0000 (15:15 -0400)]
[InstSimplify] don't commute constant expression operand in min/max calls
The test shows that we would fail to consistently fold the
instruction based on the max value operand. This is also
the root cause for issue #57986, but I'll add an instcombine
test + assert for that exact problem in another commit.
Sanjay Patel [Mon, 26 Sep 2022 19:11:30 +0000 (15:11 -0400)]
[InstSimplify] add tests for max intrinsic with constant expression; NFC
This shows the root problem that leads to the crash in issue #57986.
Jan Svoboda [Mon, 26 Sep 2022 19:40:01 +0000 (12:40 -0700)]
[clang][deps] Fix test on AIX
The "-fno-integrated-as" on AIX expands the driver invocation into multiple jobs. This patch makes new test accommodates for that.
Krzysztof Parzyszek [Mon, 26 Sep 2022 19:27:56 +0000 (12:27 -0700)]
[Hexagon] Avoid some unnecessary sign-extend instructions
Simplify (sext_inreg (extractu ...)) -> (extract ...) where appropriate.
Mahesh Ravishankar [Thu, 22 Sep 2022 06:01:45 +0000 (06:01 +0000)]
[mlir][TilingInterface] Fix `iter_args` handling in tile (and fuse).
The current approach for handling `iter_args` was to replace all uses
of the value that is used as `init` value with the corresponding
region block argument within the `scf.for`. This is not always
correct. Instead a more deliberate approach needs to be taken to
handle these. If the slice being fused represents a slice of the
destination operand of the untiled op, then
- Make the destination of the fused producer the `init` value of the
loop nest
- For the tiled and fused producer op created, replace the slice of
the destination operand with a slice of the corresponding region
iter arg of the innermost loop of the generated loop nest
Differential Revision: https://reviews.llvm.org/D134411
Sanjay Patel [Mon, 26 Sep 2022 16:47:05 +0000 (12:47 -0400)]
[InstSimplify] rearrange matching for select-of-min/max folds; NFC
This makes the code a little shorter and should be easier to extend
for a pattern like in issue #42100.
Sanjay Patel [Mon, 26 Sep 2022 14:37:23 +0000 (10:37 -0400)]
[InstSimplify] remove redundant predicate check; NFC
It's still possible that there's a simpler way to specify
the conditions needed for this set of folds, but "getStrictPred"
converts >= to > for example, so there's no need to explicitly
check that.
Michael Buch [Fri, 23 Sep 2022 22:39:01 +0000 (00:39 +0200)]
[lldb][test] 2 - Add gmodules test category explicitly where previously done implicitly
Since we don't compile with `gmodules` implicitly via
debug-info test replication, we should mark all implicit
`gmodules` tests with the appropriate category so the API
tests get actually run as intended.
Differential Revision: https://reviews.llvm.org/D134574
Michael Buch [Fri, 23 Sep 2022 09:39:43 +0000 (11:39 +0200)]
[lldb][test] 1 - Don't do test-replication for gmodules debug_info variant
Currently, by default LLDB runs an API test with 3 variants,
one of which, depending on platform, is `gmodules`. However,
most tests don't actually make use of any specific `gmodules`
feature since they compile a single `main.cpp` file without
imporint types from other modules.
Instead of running all tests an extra time with `-gmodules`
enabled, we plan to test `gmodules` features with dedicated
API tests that explicitly opt-into compiling with `-gmodules`.
One of the main benefits of this will be a reduction in total
API test-suite run-time (by around 1/3).
This patch adds a flag to `debug_info_categories` that indicates
whether a category is eligible to be replicated by `lldbtest`.
Keeping `gmodules` a debug-info category is desirable because
`builder.py` knows how to inject the appropriate Makefile flags
into the build command for debug-info categories already. Whereas
for non-debug-info categories we'd have to teach it how to. The
category is inferred from the test-name debug-info suffix currently.
Differential Revision: https://reviews.llvm.org/D134524
Jakub Kuderski [Mon, 26 Sep 2022 18:53:21 +0000 (14:53 -0400)]
[mlir][spirv] Format a define. NFC.
Matt Arsenault [Mon, 19 Sep 2022 20:53:47 +0000 (16:53 -0400)]
GuardWidening: Pass through AssumptionCache (NFC)
Matt Arsenault [Mon, 19 Sep 2022 21:50:00 +0000 (17:50 -0400)]
LoopPeel: Pass through AssumptionCache (NFC)
Matt Arsenault [Mon, 19 Sep 2022 21:50:22 +0000 (17:50 -0400)]
LoopUnroll: Pass through AssumptionCache (NFC)
Using these queries with a context instruction and without a cache
seems to be about 2x slower than with it so this theoretically
improves compile time.
Jakub Kuderski [Mon, 26 Sep 2022 17:53:58 +0000 (13:53 -0400)]
Reland "[mlir][spirv] Change td def/class prefix from SPV to SPIRV"
This reverts commit
ce82530cd03d884a2c9fa48e1f9eb931064b2553.
Undo accidental Vulkan/SPIR-V name changes.
Tested with `ninja check-mlir check-mlir-mlir-spirv-cpu-runner check-mlir-mlir-vulkan-runner`.
Thomas Raoux [Fri, 23 Sep 2022 23:00:18 +0000 (23:00 +0000)]
[mlir][vector] Handle subview correctly in sotre to load opt on memref
Make sure we consider other subviews of the same buffer when doing store
to load forwarding or dead store elimination.
Differential Revision: https://reviews.llvm.org/D134576
Dave Lee [Fri, 23 Sep 2022 04:03:31 +0000 (21:03 -0700)]
[lldb] Fix completion of 'settings set' values
Some time ago, a refactor (
1153dc960) broke completion for assigning settings
values (`settings set`). This was most annoying for enum settings, where you'd
have to get the valid enum names separately.
This restores the logic in the post-refactor completion function, as well as
adding a test to catch future regressions.
Differential Revision: https://reviews.llvm.org/D134515
Krzysztof Parzyszek [Mon, 26 Sep 2022 18:13:41 +0000 (11:13 -0700)]
[Hexagon] Make sure we can still shift scalar vectors by non-splats
Thomas Raoux [Mon, 26 Sep 2022 05:22:06 +0000 (05:22 +0000)]
[mlir] Relax transfer op hoisting on tensor
Improve hoisting logic to support cases where the read being hoisted
comes from a transfer_write with disjoint indices.
Differential Revision: https://reviews.llvm.org/D134624
Lang Hames [Mon, 26 Sep 2022 14:31:07 +0000 (07:31 -0700)]
[JITLink] Remove overly-restrictive visibility check introduced in
75404e9ef88.
ELF undefined symbols can have non-default visibility.
Craig Topper [Mon, 26 Sep 2022 18:09:31 +0000 (11:09 -0700)]
[LegalizeTypes][RISCV] Support f16 in ExpandIntRes_LLROUND_LLRINT.
Promote f16 to f32 and use the f32 libcall.
I deleted rv64zfh-half-intrinsics-strict.ll because it only existed due to this issue breaking rv32.
Differential Revision: https://reviews.llvm.org/D134579
Jan Svoboda [Mon, 26 Sep 2022 16:40:02 +0000 (09:40 -0700)]
[clang][modules] Over-align the `Module` class
This makes `llvm::PointerIntPair<Module *, 3>` from
f35230ae work across platforms.
Reviewed By: srj
Differential Revision: https://reviews.llvm.org/D134653