Michał Górny [Thu, 1 Apr 2021 12:17:47 +0000 (14:17 +0200)]
[lldb] [test] Mark lldb-server multiprocess tests as LLGS cat
Dmitry Preobrazhensky [Thu, 1 Apr 2021 11:21:00 +0000 (14:21 +0300)]
[AMDGPU][MC][GFX10][GFX90A] Corrected _e32/_e64 suffices
Fixed bugs https://bugs.llvm.org//show_bug.cgi?id=49643, https://bugs.llvm.org//show_bug.cgi?id=49644, https://bugs.llvm.org//show_bug.cgi?id=49645.
Differential Revision: https://reviews.llvm.org/D99413
Simon Pilgrim [Thu, 1 Apr 2021 09:08:08 +0000 (10:08 +0100)]
[X86][SSE] Fold HOP(HOP(X,X),HOP(Y,Y)) -> HOP(PERMUTE(HOP(X,Y)),PERMUTE(HOP(X,Y))
For slow-hop targets, attempt to merge HADD/SUB pairs used in chains.
Simon Pilgrim [Wed, 31 Mar 2021 16:36:34 +0000 (17:36 +0100)]
[X86][SSE] Enable (F)HADD/SUB handling to SimplifyMultipleUseDemandedVectorElts
Attempt to bypass unused horiz-op operands.
This is very similar to the PACKSS/PACKUS handling - we should try to merge these.
Simon Pilgrim [Wed, 31 Mar 2021 13:13:08 +0000 (14:13 +0100)]
[X86][SSE] Add isHorizOp helper function. NFCI.
Dmitry Preobrazhensky [Thu, 1 Apr 2021 10:41:09 +0000 (13:41 +0300)]
[AMDGPU][MC] Added flag to identify VOP instructions which have a single variant
By convention, VOP1/2/C instructions which can be promoted to VOP3 have _e32 suffix while promoted instructions have _e64 suffix. Instructions which have a single variant should have no _e32/_e64 suffix. Unfortunately there was no simple way to identify single variant instructions - it was implemented by a hack. See bug https://bugs.llvm.org/show_bug.cgi?id=39086.
This fix simplifies handling of single VOP instructions by adding a dedicated flag.
Differential Revision: https://reviews.llvm.org/D99408
Florian Hahn [Thu, 1 Apr 2021 10:11:58 +0000 (11:11 +0100)]
[SLP] Add test cases for missing SLP vectorization on AArch64.
Balázs Kéri [Thu, 1 Apr 2021 09:10:07 +0000 (11:10 +0200)]
[clang][Checkers] Extend PthreadLockChecker state dump (NFC).
Add printing of map 'DestroyRetVal'.
Reviewed By: steakhal
Differential Revision: https://reviews.llvm.org/D98502
David Sherwood [Tue, 23 Mar 2021 14:35:03 +0000 (14:35 +0000)]
[NFC] Add tests for scalable vectorization of loops with large stride acesses
This patch just adds tests that we can vectorize loop such as these:
for (i = 0; i < n; i++)
dst[i * 7] += 1;
and
for (i = 0; i < n; i++)
if (cond[i])
dst[i * 7] += 1;
using scalable vectors, where we expect to use gathers and scatters in the
vectorized loop. The vector of pointers used for the gather is identical
to those used for the scatter so there should be no memory dependences.
Tests are added here:
Transforms/LoopVectorize/AArch64/sve-large-strides.ll
Differential Revision: https://reviews.llvm.org/D99192
Vinayaka Bandishti [Thu, 1 Apr 2021 09:08:24 +0000 (14:38 +0530)]
[MLIR][Affine] Add utility to check if the slice is valid
Fixes a bug in affine fusion pipeline where an incorrect slice is computed.
After the slice computation is done, original domain of the the source is
compared with the new domain that will result if the fusion succeeds. If the
new domain must be a subset of the original domain for the slice to be
valid. If the slice computed is incorrect, fusion based on such a slice is
avoided.
Relevant test cases are added/edited.
Fixes https://bugs.llvm.org/show_bug.cgi?id=49203
Differential Revision: https://reviews.llvm.org/D98239
Muhammad Omair Javaid [Thu, 1 Apr 2021 09:15:00 +0000 (14:15 +0500)]
[LLDB] Fix sync issue in TestVSCode_launch.test_progress_events
This fixes flakiness in TestVSCode_launch.test_progress_events
vscode.progress_events some times failed to populate in time for
follow up iterations.
Adding a minor delay before the the for the loop fixes the issue.
Reviewed By: clayborg
Differential Revision: https://reviews.llvm.org/D99497
Muhammad Omair Javaid [Thu, 1 Apr 2021 09:10:14 +0000 (14:10 +0500)]
Revert "Revert "[LLDB] Arm64/Linux test case for MTE and Pointer Authentication regset""
This reverts commit
feb6f2c78fa9474e7329c4a809f175b1675d0975.
Muhammad Omair Javaid [Wed, 31 Mar 2021 18:01:38 +0000 (23:01 +0500)]
Revert "Revert "[LLDB] Arm64/Linux Add MTE and Pointer Authentication registers""
This reverts commit
71b648f7158c7a0b4918eaa3e94d307e4bbfce97.
There was a typo in the last commit which was causing LLDB AArch64 Linux
buildbot testsuite failures. Now fixed in current version.
Yevgeny Rouban [Thu, 1 Apr 2021 08:33:00 +0000 (15:33 +0700)]
[LoopFlatten] Do not report CFG analyses as up-to-date
Removes CFGAnalyses from the preserved analyses set
returned by LoopFlattenPass::run().
Reviewed By: Dave Green, Ta-Wei Tu
Differential Revision: https://reviews.llvm.org/D99700
Harald van Dijk [Thu, 1 Apr 2021 08:47:56 +0000 (09:47 +0100)]
[Driver] Fix architecture triplets and search paths for Linux x32
Currently, support for the x32 ABI is handled as a multilib to the
x86_64 target only. However, full self-hosting x32 systems treating it
as a separate architecture with its own architecture triplets as well as
search paths exist as well, in Debian's x32 port and elsewhere.
This adds the missing architecture triplets and search paths so that
clang can work as a native compiler on x32, and updates the tests so
that they pass when using an x32 libdir suffix.
Additionally, we would previously also assume that objects from any
x86_64-linux-gnu GCC installation could be used to target x32. This
changes the logic so that only GCC installations that include x32
support are used when targetting x32, meaning x86_64-linux-gnux32 GCC
installations, and x86_64-linux-gnu and i686-linux-gnu GCC installations
that include x32 multilib support.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D52050
Sam Parker [Wed, 31 Mar 2021 08:25:18 +0000 (09:25 +0100)]
[WebAssembly] Invert branch condition on xor input
A frequent pattern for floating point conditional branches use an xor
to invert the input for the branch. Instead we can fold away the xor
by swapping the branch target instead.
Differential Revision: https://reviews.llvm.org/D99171
Nicolas Vasilache [Wed, 31 Mar 2021 09:33:08 +0000 (09:33 +0000)]
[mlir][Python][Linalg] Add missing attributes to linalg ops
This revision tightens up the handling of attributes for both named
and generic linalg ops.
To demonstrate the IR validity, a working e2e Linalg example is added.
Differential Revision: https://reviews.llvm.org/D99430
Pavel Labath [Thu, 1 Apr 2021 07:01:35 +0000 (09:01 +0200)]
[lldb] Fix build errors from
3bea7306e8
The addition of the dummy constructors requires matching changes in os-
and arch-specific files, which I forgot about.
Pavel Labath [Thu, 1 Apr 2021 06:13:50 +0000 (08:13 +0200)]
[lldb] Fix compilation with gcc-6.5
This fixes (works around) two errors with gcc-6.5.
- in the RegisterContext_x86 files, gcc is unable to synthesize a
default constructor -- it thinks it needs to initialize the virtual
base class, even though said classes are abstract. I fix that by
providing a dummy constructor.
- In ReproducerInstrumentationTest, it is not able to deduce that the
TestingRegistry class is movable (it contains a map of unique
pointers). I change the type from Optional<TestingRegistry> to
unique_ptr<TestingRegistry), so that moving is not required
(copying/moving a polymorphic type is not a very good idea in any
case).
Marek Kurdej [Thu, 1 Apr 2021 06:29:55 +0000 (08:29 +0200)]
[libc++] Build and test with -Wundef warning. NFC.
This will avoid typos like `_LIBCPP_STD_VERS` (<future>) or using `#if TEST_STD_VER > 17` without including "test_macros.h".
Reviewed By: ldionne, #libc
Differential Revision: https://reviews.llvm.org/D99515
Max Kazantsev [Thu, 1 Apr 2021 06:10:10 +0000 (13:10 +0700)]
[NFC] Undo some erroneous renamings
Some vars renamed by mistake during auto-replacements. Undoing them.
Max Kazantsev [Thu, 1 Apr 2021 05:31:57 +0000 (12:31 +0700)]
[NFC] Disambiguate LI in GVN
Name GVN uses name 'LI' for two different unrelated things:
LoadInst and LoopInfo. This patch relates the variables with
former meaning into 'Load' to disambiguate the code.
Jonas Devlieghere [Thu, 1 Apr 2021 04:40:26 +0000 (21:40 -0700)]
[lldb] Remove references to LLDB_CAPTURE_REPRODUCER
Remove the remaining references to LLDB_CAPTURE_REPRODUCER. I removed
the functionality in an earlier commit but forgot that there was a
corresponding test and logic to unset it in our test suite.
Jonas Devlieghere [Thu, 1 Apr 2021 04:34:47 +0000 (21:34 -0700)]
[lldb-vscode] Use LLVM's ScopeExit to ensure we always terminate the debugger
Make sure we always terminate the debugger by using a RAII object.
Differential revision: https://reviews.llvm.org/D99702
Jonas Devlieghere [Thu, 1 Apr 2021 04:35:45 +0000 (21:35 -0700)]
[lldb-vscode] Consistently use return EXIT_SUCCESS and EXIT_FAILURE (NFC)
Consistently use return with EXIT_SUCCESS or EXIT_FAILURE instead of
mix-and-matching return, exit 0, 1 etc.
Differential revision: https://reviews.llvm.org/D99701
Chen Zheng [Fri, 26 Mar 2021 07:21:46 +0000 (03:21 -0400)]
[debug-info] support new tuning debugger type DBX for XCOFF DWARF
Based on this debugger type, for now, we plan to:
1: use inline string by default for XCOFF DWARF
2: generate no column info for debug line table.
Reviewed By: aprantl
Differential Revision: https://reviews.llvm.org/D99400
Jonas Devlieghere [Thu, 1 Apr 2021 04:08:55 +0000 (21:08 -0700)]
[lldb] Remove LLDB_CAPTURE_REPRODUCER override
Remove the LLDB_CAPTURE_REPRODUCER as it is inherently dangerous. The
reproducers require careful initialization which cannot be guaranteed by
overwriting the reproducer mode at this level.
If we want to provide this functionality, we should do it in the driver
instead. It was originally added to enable capture in CI, but we now
have a dedicated CI job that captures and replays the test suite.
Yang Fan [Thu, 1 Apr 2021 01:16:51 +0000 (09:16 +0800)]
[clang][APINotes] Fix -Wunused-function warning (NFC)
GCC warning:
```
/llvm-project/clang/lib/APINotes/APINotesYAMLCompiler.cpp:574:23: warning: ‘void {anonymous}::Module::dump()’ defined but not used [-Wunused-function]
574 | LLVM_DUMP_METHOD void Module::dump() {
| ^~~~~~
```
KAWASHIMA Takahiro [Wed, 31 Mar 2021 11:41:05 +0000 (20:41 +0900)]
[GVN] Propagate llvm.access.group metadata of loads
Before this change, the `llvm.access.group` metadata was dropped
when moving a load instruction in GVN. This prevents vectorizing
a C/C++ loop with `#pragma clang loop vectorize(assume_safety)`.
This change propagates the metadata as well as other metadata if
it is safe (the move-destination basic block and source basic
block belong to the same loop).
Differential Revision: https://reviews.llvm.org/D93503
KAWASHIMA Takahiro [Wed, 31 Mar 2021 11:39:44 +0000 (20:39 +0900)]
[GVN][NFC] Pre-commit test for D93503
qixingxue [Wed, 31 Mar 2021 23:08:23 +0000 (07:08 +0800)]
[GVN][NFC] Refactor analyzeLoadFromClobberingWrite
This commit adjusts the order of two swappable if statements to
make code cleaner.
Reviewed By: lattner, nikic
Differential Revision: https://reviews.llvm.org/D99648
Nick Desaulniers [Thu, 1 Apr 2021 00:02:13 +0000 (17:02 -0700)]
Revert "Add support to -Wa,--version in clang"
This reverts commit
3cc3c0f8352ec33ca2f2636f94cb1d85fc57ac16.
Breaks non-linux platforms.
https://reviews.llvm.org/D99556#2662706
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
Alexander Shaposhnikov [Wed, 31 Mar 2021 22:23:19 +0000 (15:23 -0700)]
[lld][MachO] Refactor handling of subsections
This diff is a preparation for fixing FunStabs (incorrect size calculation).
std::map<uint32_t, InputSection*> (SubsectionMap) is replaced with
a sorted vector + binary search. If .subsections_via_symbols is set
this vector will contain the list of subsections, otherwise,
the offsets will be used for calculating the symbols sizes.
Test plan: make check-all
Differential revision: https://reviews.llvm.org/D98837
Philip Reames [Wed, 31 Mar 2021 23:47:32 +0000 (16:47 -0700)]
[ValueTracking] Handle non-zero ashr/lshr recurrences
If we know we don't shift out bits (e.g. exact), all we need to know is that input is non-zero.
Philip Reames [Wed, 31 Mar 2021 23:36:19 +0000 (16:36 -0700)]
[tests] Add tests for ashr/lshr recurrences in isKnownNonZero
Jian Cai [Wed, 31 Mar 2021 22:37:19 +0000 (15:37 -0700)]
Add support to -Wa,--version in clang
Clang currently only supports -Wa,--version when -no-integrated-as is
used. This adds support to -Wa,--version with -integrated-as.
Link:
https://github.com/ClangBuiltLinux/linux/issues/1320
Reviewed By: nickdesaulniers, MaskRay
Differential Revision: https://reviews.llvm.org/D99556
Philip Reames [Wed, 31 Mar 2021 22:34:57 +0000 (15:34 -0700)]
Add debug printers for KnownBits [nfc]
Simonas Kazlauskas [Sat, 3 Oct 2020 17:25:02 +0000 (20:25 +0300)]
Support {S,U}REMEqFold before legalization
This allows these optimisations to apply to e.g. `urem i16` directly
before `urem` is promoted to i32 on architectures where i16 operations
are not intrinsically legal (such as on Aarch64). The legalization then
later can happen more directly and generated code gets a chance to avoid
wasting time on computing results in types wider than necessary, in the end.
Seems like mostly an improvement in terms of results at least as far as x86_64 and aarch64 are concerned, with a few regressions here and there. It also helps in preventing regressions in changes like {D87976}.
Reviewed By: lebedev.ri
Differential Revision: https://reviews.llvm.org/D88785
Craig Topper [Wed, 31 Mar 2021 22:06:07 +0000 (15:06 -0700)]
[RISCV] Add UnsupportedSchedZfh multiclass to reduce duplicate lines from RISCVSchedRocket.td and RISCVSchedSiFive7.td. NFC
YangKeao [Wed, 31 Mar 2021 20:44:54 +0000 (23:44 +0300)]
[X86] add dwarf annotation for inline stack probe
While probing stack, the stack register is moved without dwarf
information, which could cause panic if unwind the backtrace.
This commit only add annotation for the inline stack probe case.
Dwarf information for the loop case should be done in another
patch and need further discussion.
Reviewed By: nagisa
Differential Revision: https://reviews.llvm.org/D99579
natashaknk [Wed, 31 Mar 2021 21:10:18 +0000 (14:10 -0700)]
[mlir][tosa] Add tosa.reciprocal and tosa.sigmoid lowerings
Lowering reciprocal and sigmoid elementwise operations to linalg dialect.
Reviewed By: rsuderman
Differential Revision: https://reviews.llvm.org/D99676
Jianzhou Zhao [Wed, 31 Mar 2021 19:44:25 +0000 (19:44 +0000)]
[dfsan] Set sigemptyset's return label to be 0
This was not set from when the wrapper was introduced.
Reviewed By: gbalats
Differential Revision: https://reviews.llvm.org/D99678
Thomas Preud'homme [Sun, 28 Mar 2021 00:03:37 +0000 (00:03 +0000)]
[test, InferFunctionAttrs] Fix use of var defined in CHECK-NOT
LLVM test Transforms/InferFunctionAttrs/annotate contains two RUN
invokation (UNKNOWN and NVPTX lines) which involve a CHECK-NOT directive
with a variable not defined by the enabled CHECK prefixes. This commit
fixes that by:
- enabling CHECK prefix for unknown target with specialisation when it
differs from other targets
- checking for absence of bcmp with any attribute for NVPTX
Reviewed By: tra
Differential Revision: https://reviews.llvm.org/D99589
Roman Lebedev [Wed, 31 Mar 2021 20:23:57 +0000 (23:23 +0300)]
[NFC][LoopRotation] Count the number of instructions hoisted/cloned into preheader
Philip Reames [Wed, 31 Mar 2021 20:26:19 +0000 (13:26 -0700)]
Revert "Make TableGenGlobalISel an object library"
This reverts commit
2c3cf62d4a26de85aab180bb43a579c913b17f3e.
Causes build failures on x86_64, will respond to commit thread with link errors.
Aaron Puchert [Wed, 31 Mar 2021 20:14:22 +0000 (22:14 +0200)]
Make TableGenGlobalISel an object library
That's how it was originally intended but that wasn't possible because
we still needed to support older CMake versions.
The problem here is that the sources in TableGenGlobalISel are meant to
be linked into both llvm-tblgen and TableGenTests (a unit test), but not
be part of LLVM proper. So they shouldn't be an ordinary LLVM component.
Because they are used in llvm-tblgen, they can't draw in the LLVM dylib
dependency, but then we'd have to do the same thing in TableGenTests to
make sure we don't link both a static Support library and another copy
through the LLVM dylib.
With an object library we're just reusing the object files and don't
have to care about dependencies at all.
Reviewed By: beanz
Differential Revision: https://reviews.llvm.org/D74588
Alexey Bataev [Mon, 29 Mar 2021 17:26:42 +0000 (10:26 -0700)]
[OPENMP]Fix PR48885: Crash in passing firstprivate args to tasks on Apple M1.
Need to bitcast the function pointer passed as a parameter to the real
type to avoid possible problem with calling conventions.
Differential Revision: https://reviews.llvm.org/D99521
Alexey Bataev [Tue, 30 Mar 2021 17:36:23 +0000 (10:36 -0700)]
[OPENMP]Fix PR48658: [OpenMP 5.0] Compiler crash when OpenMP atomic sync hints used.
No need to consider hint clause kind as the main atomic clause kind at the
codegen.
Differential Revision: https://reviews.llvm.org/D99611
Philip Reames [Wed, 31 Mar 2021 19:48:14 +0000 (12:48 -0700)]
[tests] Exercise cases where SCEV can use trip counts to refine ashr/lshr recurrences
Jez Ng [Wed, 31 Mar 2021 19:48:18 +0000 (15:48 -0400)]
[lld-macho] Parallelize UUID hash computation
This reuses the approach (and some code) from LLD-ELF.
It's a decent win when linking chromium_framework on a Mac Pro (3.2 GHz 16-Core Intel Xeon W):
N Min Max Median Avg Stddev
x 20 4.58 4.83 4.66 4.6685 0.
066591844
+ 20 4.42 4.61 4.5 4.505 0.
04751731
Difference at 95.0% confidence
-0.1635 +/- 0.0370242
-3.5022% +/- 0.793064%
(Student's t, pooled s = 0.0578462)
The output binary is 381MB.
Reviewed By: #lld-macho, oontvoo
Differential Revision: https://reviews.llvm.org/D99279
Jez Ng [Wed, 31 Mar 2021 19:48:14 +0000 (15:48 -0400)]
[lld-macho][nfc] Test that -ObjC will import bitcode with category sections
The functionality was originally added in {D95265}, but the test in that
diff only checked if `-ObjC` would cause bitcode containing ObjC class
symbols to be loaded. It neglected to test for bitcode containing
categories but no class symbols.
This diff also changes the lto-archive.ll test to use `-why_load`
instead of inspecting the output binary's symbol table. This is
motivated by the stacked diff {D99105}, which will hide irrelevant
bitcode symbols.
Reviewed By: #lld-macho, oontvoo
Differential Revision: https://reviews.llvm.org/D99215
Alexey Bataev [Wed, 31 Mar 2021 19:35:38 +0000 (12:35 -0700)]
[SLP]Update test checks, NFC
Craig Topper [Wed, 31 Mar 2021 19:25:42 +0000 (12:25 -0700)]
[SelectionDAG] Remove unneeded vector resize from the end of FoldConstantArithmetic. NFC
There's an assert right before that makes sure the size already matches.
Earlier in this function's life, scalars and vectors shared more
code.
Andrew Young [Mon, 29 Mar 2021 02:25:32 +0000 (19:25 -0700)]
[mlir][cse] do not replace operands in previously simplified operations
If an operation has been inserted as a key in to the known values
hashtable, then it can not be modified in a way which changes its hash.
This change avoids modifying the operands of any previously recorded
operation, which prevents their hash from changing.
In an SSACFG region, it is impossible to visit an operation before
visiting its operands, so this is not a problem. This situation can only
happen in regions without strict dominance, such as graph regions.
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D99486
George Mitenkov [Wed, 31 Mar 2021 18:46:28 +0000 (21:46 +0300)]
[ConstantFolding] Fixing addo/subo with undef
When folding addo/subo with undef, the current
convention is to use { -1, false } for addo and
{ 0, false } for subo. This was fixed for InstSimplify in
https://reviews.llvm.org/rGf094d65beaa492e845b03561eddd75b5be653a01,
but not in ConstantFolding.
Reviewed By: nikic, lebedev.ri
Differential Revision: https://reviews.llvm.org/D99564
Alexey Bataev [Wed, 31 Mar 2021 18:21:54 +0000 (11:21 -0700)]
[SLP]Add a test for the bug in `getVectorElementSize()`, NFC.
peter klausler [Wed, 31 Mar 2021 16:12:28 +0000 (09:12 -0700)]
[flang] Refine checks for pointer initialization targets
f18 was emitting a bogus error message about the lack of a TARGET
attribute when a pointer was initialized with a component of a
variable that was a legitimate TARGET.
Differential Revision: https://reviews.llvm.org/D99665
Huihui Zhang [Wed, 31 Mar 2021 17:58:36 +0000 (10:58 -0700)]
[LoopVectorize] Use SetVector to track uniform uses to prevent non-determinism.
Use SetVector instead of SmallPtrSet to track values with uniform use. Doing this
can help avoid non-determinism caused by iterating over unordered containers.
This bug was found with reverse iteration turning on,
--extra-llvm-cmake-variables="-DLLVM_REVERSE_ITERATION=ON".
Failing LLVM test consecutive-ptr-uniforms.ll .
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D99549
Suraj Sudhir [Wed, 31 Mar 2021 17:48:21 +0000 (10:48 -0700)]
Move non-spec TOSA operators into TosaUtilOps.td
Reviewed By: stellaraccident
Differential Revision: https://reviews.llvm.org/D99628
Petr Hosek [Tue, 22 Sep 2020 08:01:16 +0000 (01:01 -0700)]
[Driver] Move detectLibcxxIncludePath to ToolChain
This helper method is useful even outside of Gnu toolchains, so move
it to ToolChain so it can be reused in other toolchains such as Fuchsia.
Differential Revision: https://reviews.llvm.org/D88452
Thomas Lively [Wed, 31 Mar 2021 17:46:17 +0000 (10:46 -0700)]
[WebAssembly] Implement i64x2 comparisons
Removes the prototype builtin and intrinsic for i64x2.eq and implements that
instruction as well as the other i64x2 comparison instructions in the final SIMD
spec. Unsigned comparisons were not included in the final spec, so they still
need to be scalarized via a custom lowering.
Differential Revision: https://reviews.llvm.org/D99623
Juneyoung Lee [Wed, 31 Mar 2021 17:01:17 +0000 (02:01 +0900)]
[ValueTracking] Add with.overflow intrinsics to poison analysis functions
This is a patch teaching ValueTracking that `s/u*.with.overflow` intrinsics do not
create undef/poison and they propagate poison.
I couldn't write a nice example like the one with ctpop; ValueTrackingTest.cpp were simply updated
to check these instead.
This patch helps reducing regression while fixing https://llvm.org/pr49688 .
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D99671
Philip Reames [Wed, 31 Mar 2021 17:29:21 +0000 (10:29 -0700)]
[SCEV] Handle unreachable binop when matching shift recurrence
This fixes an issue introduced with my change d4648e, and reported in pr49768.
The root problem is that dominance collapses in unreachable code, and that LoopInfo explicitly only models reachable code. Since the recurrence matcher doesn't filter by reachability (and can't easily because not all consumers have domtree), we need to bailout before assuming that finding a recurrence implies we found a loop.
Craig Topper [Wed, 31 Mar 2021 16:49:16 +0000 (09:49 -0700)]
[X86] Improve SMULO/UMULO codegen for vXi8 vectors.
The default expansion creates a MUL and either a MULHS/MULHU. Each
of those separately expand to sequences that use one or more
PMULLW instructions as well as additional instructions to
extend the types to vXi16. The MULHS/MULHU expansion computes the
whole 16-bit product, but only keeps the high part.
We can improve the lowering of SMULO/UMULO for some cases by using the MULHS/MULHU
expansion, but keep both the high and low parts. And we can use
those parts to calculate the overflow.
For AVX512 we might have vXi1 overflow outputs. We can improve those by using
vpcmpeqw to produce a k register if AVX512BW is enabled. This is a little better
than truncating the high result to use vpcmpeqb. If we don't have avx512bw we
can extend up to v16i32 to use vpcmpeqd to produce a k register.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D97624
Shimin Cui [Wed, 31 Mar 2021 16:36:49 +0000 (12:36 -0400)]
[PowerPC] [MLICM] Enable hoisting of caller preserved registers on AIX
On ppc64 linux , MachineLICM will hoist caller preserved registers, including TOC loads of the global variable address, out of loops. This is to enable this on AIX for both ppc64 and ppc32.
Differential Revision: https://reviews.llvm.org/D99076
Craig Topper [Wed, 31 Mar 2021 16:41:02 +0000 (09:41 -0700)]
[X86] Improve optimizeCompareInstr for signed comparisons after BMI/TBM instructions
We previously couldn't optimize out a TEST if the branch/setcc/cmov
used the overflow flag. This patches allows the TEST to be removed
if the flag producing instruction is known to clear the OF flag.
Thats what the TEST instruction would have done so that should be
equivalent.
Need to add test cases. I'll try to get back to this if I have bandwidth.
Fixes PR48768.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D94856
Wael Yehia [Wed, 31 Mar 2021 16:42:12 +0000 (16:42 +0000)]
[LTO][Legacy] Decouple option parsing from LTOCodeGenerator
in this patch we add a new libLTO API to specify debug options independent of an lto_code_gen_t.
This allows clients to pass codegen flags (through libLTO) which otherwise today are ignored.
Reviewed By: steven_wu
Differential Revision: https://reviews.llvm.org/D92611
Craig Topper [Wed, 31 Mar 2021 16:27:22 +0000 (09:27 -0700)]
[RISCV] Add RISCVISD opcodes for CLZW and CTZW.
Our CLZW isel pattern is quite easily broken by surrounding code
preventing it from matching sometimes. This usually results in
failing to remove the and X, 0xffffffff inserted by type
legalization. The add with -32 that type legalization also inserts
will often gets combined into other add/sub nodes. That doesn't
usually result in extra code when we don't use clzw.
CTTZ seems to be less fragile, but I wanted to keep it consistent
with CTLZ.
Reviewed By: asb, HsiangKai
Differential Revision: https://reviews.llvm.org/D99317
Jay Foad [Wed, 31 Mar 2021 15:54:21 +0000 (16:54 +0100)]
[AMDGPU] Add some image tests with enable-prt-strict-null disabled. NFC.
Jay Foad [Wed, 31 Mar 2021 15:51:16 +0000 (16:51 +0100)]
[AMDGPU] Use a common check prefix for some image tests. NFC.
Craig Topper [Wed, 31 Mar 2021 16:22:38 +0000 (09:22 -0700)]
[RISCV] Add isel patterns to select vsub_vx intrinsic to vadd.vi if it uses a small enough immediate
Also modify the simm5_plus1 check because Imm-1 is UB if Imm happens
to be INT64_MIN. I don't think the compiler would optimize based on that in this
usage, but it could fail UBSan or -ftrapv.
Reviewed By: HsiangKai, frasercrmck
Differential Revision: https://reviews.llvm.org/D99637
Arthur Eubanks [Tue, 30 Mar 2021 16:30:05 +0000 (09:30 -0700)]
[llvm-jitlink] Fix -Wunused-function on Windows
Reviewed By: sgraenitz
Differential Revision: https://reviews.llvm.org/D99604
Heejin Ahn [Wed, 31 Mar 2021 09:32:51 +0000 (02:32 -0700)]
[WebAssembly] Raname a test and fix comments
D99627 fixed a decoding bug, not an encoding bug. This renames the test
to correct it and fix comments.
Reviewed By: dschuff
Differential Revision: https://reviews.llvm.org/D99644
Sanjay Patel [Wed, 31 Mar 2021 15:26:22 +0000 (11:26 -0400)]
[InstCombine] fold abs(srem X, 2)
This is a missing optimization based on an example in:
https://llvm.org/PR49763
As noted there and the test here, we could add a more
general fold if that is shown useful.
https://alive2.llvm.org/ce/z/xEHdTv
https://alive2.llvm.org/ce/z/97dcY5
Sanjay Patel [Wed, 31 Mar 2021 13:29:49 +0000 (09:29 -0400)]
[InstCombine] add tests for srem+abs; NFC
Bradley Smith [Wed, 24 Mar 2021 11:16:20 +0000 (11:16 +0000)]
[AArch64][SVE] Add tests for UREM/SREM using fixed SVE types
Differential Revision: https://reviews.llvm.org/D99265
Timm Bäder [Wed, 3 Mar 2021 09:45:07 +0000 (10:45 +0100)]
[clang][parser] Allow GNU-style attributes in explicit template...
... instantiations
They are currently not being diagnosed because ProhibitAttributes() does
not handle attribute lists with an invalid source range. But once it
does, we need to allow GNU attributes in this place.
Additionally, start optionally diagnosing empty attr lists in
ProhibitCXX11Attributes(), since ProhibitAttribute() does it.
Differential Revision: https://reviews.llvm.org/D97362
Arthur O'Dwyer [Wed, 31 Mar 2021 14:18:51 +0000 (10:18 -0400)]
[libc++] Mark convert_copy.pass.cpp as UNSUPPORTED on clang-13 (i.e. trunk).
Because the constexpr-time codepath triggers a Clang bug. It seems
that Clang compiles it okay in release mode, but when Clang itself
is compiled in debug mode (with assertions turned on), this input
triggers an assertion failure in Clang itself. See comments on D96385
and Clang bug report https://bugs.llvm.org/show_bug.cgi?id=45879
This commit should get the debug-mode buildbots back to green.
Luís Marques [Wed, 31 Mar 2021 14:18:02 +0000 (15:18 +0100)]
[RISCV] Add XFAIL riscv32 for known issue with the old pass manager
See D80668, rG7b4832648a63 and https://bugs.llvm.org/show_bug.cgi?id=46117
for details of the issue.
Differential Revision: https://reviews.llvm.org/D99108
Sander de Smalen [Tue, 23 Feb 2021 17:10:12 +0000 (17:10 +0000)]
[SVE] Fix LoopVectorizer test scalalable-call.ll
This marks FSIN and other operations to EXPAND for scalable
vectors, so that they are not assumed to be legal by the cost-model.
Depends on D97470
Reviewed By: dmgreen, paulwalker-arm
Differential Revision: https://reviews.llvm.org/D97471
Sander de Smalen [Mon, 25 Jan 2021 13:59:37 +0000 (13:59 +0000)]
[CostModel] Align the cost model for intrinsics for scalable/fixed-width vectors.
Let getIntrinsicInstrCost call getTypeBasedIntrinsicInstrCost for scalable vectors,
similar to how this is done for fixed-width vectors, instead of falling back
on BaseT::getIntrinsicInstrCost().
If the intrinsic cannot be costed (or is not overloaded by the target),
it will return InstructionCost::getInvalid() instead.
Depends on D97469
Reviewed By: david-arm
Differential Revision: https://reviews.llvm.org/D97470
Anton Bikineev [Sat, 27 Mar 2021 16:27:21 +0000 (16:27 +0000)]
[C++2b] Support size_t literals
This adds support for C++2b's z/uz suffixes for size_t literals (P0330).
Joerg Sonnenberger [Thu, 18 Feb 2021 14:15:53 +0000 (15:15 +0100)]
[libc++] Avoid <climits> dependency in <thread>
The standard guarantees sleep durations of 2^63-1 nanoseconds to work.
Instead of depending on INT64_MAX or ULONGLONG_MAX to exist via the
header pollution, fold the constant directly. That has the additional
positive side effect that it avoids long double arithmetic bugs in GCC.
Differential Revision: https://reviews.llvm.org/D99516
Balázs Kéri [Wed, 31 Mar 2021 13:28:53 +0000 (15:28 +0200)]
Revert "[clang][Checkers] Extend PthreadLockChecker state dump (NFC)."
This reverts commit
49c0ab6d7611050ae902f6369fc878e17da453ad.
Test failures showed up because non-deterministic output.
Sander de Smalen [Fri, 22 Jan 2021 17:14:44 +0000 (17:14 +0000)]
NFC: Change getIntrinsicInstrCost to return InstructionCost
This patch migrates the TTI cost interfaces to return an InstructionCost.
See this patch for the introduction of the type: https://reviews.llvm.org/D91174
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2020-November/146408.html
Depends on D97468
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D97469
Sander de Smalen [Fri, 22 Jan 2021 17:34:10 +0000 (17:34 +0000)]
NFC: Change getTypeBasedIntrinsicCost to return InstructionCost
This patch migrates the TTI cost interfaces to return an InstructionCost.
See this patch for the introduction of the type: https://reviews.llvm.org/D91174
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2020-November/146408.html
Depends on D97466
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D97468
Max Kazantsev [Wed, 31 Mar 2021 11:20:19 +0000 (18:20 +0700)]
[Test] Auto-update test checks
Muhammad Omair Javaid [Wed, 31 Mar 2021 12:10:03 +0000 (17:10 +0500)]
Revert "[LLDB] Arm64/Linux Add MTE and Pointer Authentication registers"
This reverts commit
1164b4e2957290e814c3dd781a68e504dd39148e.
Reason: LLDB AArch64 Linux buildbot failure
Muhammad Omair Javaid [Wed, 31 Mar 2021 12:08:07 +0000 (17:08 +0500)]
Revert "[LLDB] Arm64/Linux test case for MTE and Pointer Authentication regset"
This reverts commit
9ab677180091a690cd99d4ac55d5fb9e1149b1ec.
Reason: LLDB AArch64/Linux buildbot failure.
Liqiang Tao [Wed, 31 Mar 2021 11:52:58 +0000 (19:52 +0800)]
[InlineCost] Remove TODO comment that consider other forms of savings in the cost-benefit analysis
Attempts to compute savings more accurately cannot impact the set of critically important call sites.
Reviewed By: kazu
Differential Revision: https://reviews.llvm.org/D98577
Roman Lebedev [Wed, 31 Mar 2021 11:14:13 +0000 (14:14 +0300)]
[X86] AMD Zen 3 has macro fusion
This is an improvement over Zen 2, where only branch fusion is supported,
as per Agner, 21.4 Instruction fusion.
AMD SOG 17h has no mention of fusion.
AMD SOG 19h, 2.9.3 Branch Fusion
The following flag writing instructions support branch fusion
with their reg/reg, reg/imm and reg/mem forms
* CMP
* TEST
* SUB
* ADD
* INC (no fusion with branches dependent on CF)
* DEC (no fusion with branches dependent on CF)
* OR
* AND
* XOR
Agner, 22.4 Instruction fusion
<...> This applies to CMP, TEST, ADD, SUB, AND, OR, XOR, INC, DEC and
all conditional jumps, except if the arithmetic or logic instruction has a rip-relative address or
both an address displacement and an immediate operand.
Balazs Benics [Wed, 31 Mar 2021 10:47:37 +0000 (12:47 +0200)]
[ASTImporter][NFC] Fix duplicated symbols in "Improve test coverage"
D99576 introduced a duplicate symbol, now im removing it.
Differential Revision: https://reviews.llvm.org/D99576
Fraser Cormack [Tue, 30 Mar 2021 13:49:25 +0000 (14:49 +0100)]
[RISCV] Add support for the stepvector intrinsic
This adds almost everything required for supporting the new stepvector
intrinsic on RVV. It is lowered to the existing VID_VL SDNode.
The only exception is a limitation that RV32 cannot yet lower the
intrinsic on i64 vectors. This is because the step operand is
(currently) required to be at least as large as the vector element type.
I will look into patching that out and loosening the requirement to only
an integer pointer type.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D99594
Muhammad Omair Javaid [Wed, 31 Mar 2021 09:36:38 +0000 (14:36 +0500)]
Revert "[LLDB] Skip TestVSCode_disconnect.test_launch arm/linux"
This reverts commit
73cf85e527f69c495daece7c74743b9073d4717c.
Jay Foad [Mon, 1 Mar 2021 17:24:57 +0000 (17:24 +0000)]
[AMDGPU][GlobalISel] Add support for global atomicrmw fadd
This includes gfx908 which only has a no-return version of the
global_atomic_add_f32 instruction, using the same hack that was
previously implemented for selecting from the
llvm.amdgcn.global.atomic.fadd intrinsic.
Differential Revision: https://reviews.llvm.org/D97767
Balazs Benics [Wed, 31 Mar 2021 10:10:23 +0000 (12:10 +0200)]
[ASTImporter][NFC] Improve test coverage
All three cases were imported correctly.
For BlockDecls, correctly means that we don't support importing them, thus an
error is the expected behaviour.
- BlockDecls were not yet covered. I know that they are not imported but the
test at least documents it.
- Default values for ParmVarDecls were also uncovered.
- Importing bitfield FieldDecls were imported correctly.
Reviewed By: martong, shafik
Differential Revision: https://reviews.llvm.org/D99576
Florian Hahn [Wed, 31 Mar 2021 08:38:12 +0000 (09:38 +0100)]
[AArch64] Avoid SCALAR_TO_VECTOR for single FP constant vector.
Currently the code only checks for integer constants (ConstantSDNode)
and triggers an infinite cycle for single-element floating point
vector constants. We need to check for both FP and integer constants.
Reviewed By: t.p.northover
Differential Revision: https://reviews.llvm.org/D99384
Sander de Smalen [Wed, 20 Jan 2021 17:17:23 +0000 (17:17 +0000)]
NFC: Change getUserCost to return InstructionCost
This patch migrates the TTI cost interfaces to return an InstructionCost.
See this patch for the introduction of the type: https://reviews.llvm.org/D91174
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2020-November/146408.html
Depends on D97382
Reviewed By: ctetreau, paulwalker-arm
Differential Revision: https://reviews.llvm.org/D97466
Balázs Kéri [Wed, 31 Mar 2021 08:36:06 +0000 (10:36 +0200)]
[clang][Checkers] Extend PthreadLockChecker state dump (NFC).
Add printing of map 'DestroyRetVal'.
Reviewed By: steakhal
Differential Revision: https://reviews.llvm.org/D98502
Matthias Springer [Wed, 31 Mar 2021 05:59:30 +0000 (14:59 +0900)]
[mlir] Change vector.transfer_read/write "masked" attribute to "in_bounds".
This is in preparation for adding a new "mask" operand. The existing "masked" attribute was used to specify dimensions that may be out-of-bounds. Such transfers can be lowered to masked load/stores. The new "in_bounds" attribute is used to specify dimensions that are guaranteed to be within bounds. (Semantics is inverted.)
Differential Revision: https://reviews.llvm.org/D99639