platform/upstream/llvm.git
2 years ago[AMDGPU][MC] Corrected disassembly of s_waitcnt
Dmitry Preobrazhensky [Mon, 17 Jan 2022 17:16:18 +0000 (20:16 +0300)]
[AMDGPU][MC] Corrected disassembly of s_waitcnt

s_waitcnt with default expcnt, vmcnt and lgkmcnt values was disassembled without arguments.
See https://github.com/llvm/llvm-project/issues/52716

Differential Revision: https://reviews.llvm.org/D117305

2 years ago[DebugInfo] ValueMapper impl for DIArgList respects IgnoreMissingLocals
Stephen Tozer [Mon, 17 Jan 2022 11:42:35 +0000 (11:42 +0000)]
[DebugInfo] ValueMapper impl for DIArgList respects IgnoreMissingLocals

This patch fixes an issue in which SSA value reference within a
DIArgList would be unnecessarily dropped by llvm-link, even when
invoking on a single file (which should be a no-op). The reason for the
difference is that the ValueMapper does not refer to the
RF_IgnoreMissingLocals flag for LocalAsMetadata contained within a
DIArgList; this flag is used for direct LocalAsMetadata uses to preserve
SSA references even when the ValueMapper does not have an explicit
mapping for the referenced SSA value, which appears to always be the
case when using llvm-link in this manner.

Differential Revision: https://reviews.llvm.org/D114355

2 years ago[AArch64] Revive optimize add/sub with immediate through MIPeepholeOpt
Micah Weston [Mon, 17 Jan 2022 17:17:15 +0000 (17:17 +0000)]
[AArch64] Revive optimize add/sub with immediate through MIPeepholeOpt

Fixes the build issue with D111034, whose goal was to optimize
add/sub with long immediates.

Optimize ([add|sub] r, imm) -> ([ADD|SUB] ([ADD|SUB] r, #imm0, lsl #12), #imm1),
if imm == (imm0<<12)+imm1. and both imm0 and imm1 are non-zero 12-bit unsigned
integers.

Optimize ([add|sub] r, imm) -> ([SUB|ADD] ([SUB|ADD] r, #imm0, lsl #12), #imm1),
if imm == -(imm0<<12)-imm1, and both imm0 and imm1 are non-zero 12-bit unsigned
integers.

The change which fixed the build issue in D111034 was the use of new virtual
registers so that SSA form is maintained until deleting MI.

Differential Revision: https://reviews.llvm.org/D117429

2 years ago[test] Use opt -passes syntax in DeadArgElim lit tests. NFC
Bjorn Pettersson [Sun, 16 Jan 2022 00:18:29 +0000 (01:18 +0100)]
[test] Use opt -passes syntax in DeadArgElim lit tests. NFC

The legacy PM is deprecated, so update a bunch of lit tests running
opt to use the new PM syntax when specifying the pipeline.

2 years ago[mlir][Linalg] Relax PadTensor tiling constraints and expose it to strategies.
Nicolas Vasilache [Mon, 17 Jan 2022 17:07:46 +0000 (17:07 +0000)]
[mlir][Linalg] Relax PadTensor tiling constraints and expose it to strategies.

Reviewed By: ThomasRaoux

Differential Revision: https://reviews.llvm.org/D117334

2 years ago[libc++] [test] Improve the tests for std::assignable_from.
Arthur O'Dwyer [Sun, 16 Jan 2022 17:32:33 +0000 (12:32 -0500)]
[libc++] [test] Improve the tests for std::assignable_from.

Fixes #50060.

Differential Revision: https://reviews.llvm.org/D117434

2 years ago[mlir][Linalg] NFC - Drop vectorization reliance on ConvolutionOpInterface
Nicolas Vasilache [Fri, 14 Jan 2022 16:18:06 +0000 (16:18 +0000)]
[mlir][Linalg] NFC - Drop vectorization reliance on ConvolutionOpInterface

Differential Revision: https://reviews.llvm.org/D117323

2 years ago[AIX] Support of Big archive (read)
zhijian [Mon, 17 Jan 2022 15:37:08 +0000 (10:37 -0500)]
[AIX] Support of Big archive (read)

Summary:

The patch is based on the EGuesnet's implement of the "Support of Big archive (read)
the first commit of the patch is come from https://reviews.llvm.org/D100651.

the rest of commits of the patch

1  Addressed the comments on the https://reviews.llvm.org/D100651
2  according to https://www.ibm.com/docs/en/aix/7.2?topic=formats-ar-file-format-big
using the "fl_fstmoff" for the first object file number, using "char ar_nxtmem[20]" to get next object file ,
using the "char fl_lstmoff[20]" for the last of the object file will fix the following problems:
   2.1 can not correct reading a archive files which has padding data between too object file
   2.2 can not correct reading a archive files from which some object file has be deleted

3 introduce a new derived class BigArchive for big ar file.

Reviewers: James Henderson
Differential Revision: https://reviews.llvm.org/D111889

2 years agoRevert "[AIX] Support of Big archive (read)"
zhijian [Mon, 17 Jan 2022 16:38:01 +0000 (11:38 -0500)]
Revert "[AIX] Support of Big archive (read)"

This reverts commit 3130134d6e4823b5ee7619288a4b7e1e60831a82.

2 years ago[AIX][ZOS] Handle unsupported builtin function CFStringMakeConstantString
Jake Egan [Mon, 17 Jan 2022 16:22:33 +0000 (11:22 -0500)]
[AIX][ZOS] Handle unsupported builtin function CFStringMakeConstantString

This patch emits an error on AIX and z/OS because XCOFF and GOFF does not currently implement builtin function `CFStringMakeConstantString`. Tests that use this builtin were also disabled.

Reviewed By: SeanP

Differential Revision: https://reviews.llvm.org/D117315

2 years ago[clang-format] Treat ForEachMacros as loops
Marek Kurdej [Fri, 14 Jan 2022 20:59:40 +0000 (21:59 +0100)]
[clang-format] Treat ForEachMacros as loops

TT_ForEachMacro should be considered in rules AllowShortBlocksOnASingleLine
and AllowShortLoopsOnASingleLine.
Fixes https://github.com/llvm/llvm-project/issues/45432.

Reviewed By: MyDeveloperDay

Differential Revision: https://reviews.llvm.org/D94955

2 years ago[mlir][Vector] Generalize and improve folding of ExtractOp from Insert/Transpose...
Nicolas Vasilache [Fri, 14 Jan 2022 16:08:14 +0000 (16:08 +0000)]
[mlir][Vector] Generalize and improve folding of ExtractOp from Insert/Transpose chain.

This revision fixes a bug where the iterative algorithm would walk back def-use chains to an incorrect operand.
This exposed opportunities for a larger refactoring and behavior improvement.
The new algorithm has improved folding behavior and proceeds by tracking both the
permutation of the extraction position and the internal vector permutation.
Multiple partial intersection cases with a candidate insertOp are supported.

The refactoring of the implementation should also help it generalize to strided insert/extract op.

This also subsumes the previous `foldExtractOpFromTranspose` which is now a simple special case and can be deleted.

Differential Revision: https://reviews.llvm.org/D117322

2 years ago[MCA][X86] Add missing zero-idioms test file coverage
Simon Pilgrim [Mon, 17 Jan 2022 15:57:50 +0000 (15:57 +0000)]
[MCA][X86] Add missing zero-idioms test file coverage

atom/slm have no/limited zero-idioms handling but we should test all the common instructions anyhow

znver1/znver2 were just missing - I've copied the Haswell tests for consistent test coverage

2 years ago[ConstantFold] Don't fold load from non-byte-sized vector
Nikita Popov [Mon, 17 Jan 2022 15:59:23 +0000 (16:59 +0100)]
[ConstantFold] Don't fold load from non-byte-sized vector

Following up on https://github.com/llvm/llvm-project/commit/1470f94d71c544327f76b85c55cb6f7cb43a6cbb#r63981173:

The result here (probably) depends on endianness. Don't bother
trying to handle this exotic case, just bail out.

2 years ago[InstCombine] FP with reassoc FMF: (X * C) + X --> X * (MulC + 1.0)
Sanjay Patel [Mon, 17 Jan 2022 15:35:03 +0000 (10:35 -0500)]
[InstCombine] FP with reassoc FMF: (X * C) + X --> X * (MulC + 1.0)

This fold already exists for scalars via FAddCombine (and that's
why 2 of the tests are only changed cosmetically), but that code
misses vectors and has largely been replaced by simpler folds
over time, so this is another step towards removing it.

2 years ago[InstCombine] add tests for fadd+fmul factorization fold; NFC
Sanjay Patel [Mon, 17 Jan 2022 15:03:58 +0000 (10:03 -0500)]
[InstCombine] add tests for fadd+fmul factorization fold; NFC

As shown in the tests, this transform exists partially already.

2 years ago[AIX] Support of Big archive (read)
zhijian [Mon, 17 Jan 2022 15:37:08 +0000 (10:37 -0500)]
[AIX] Support of Big archive (read)

Summary:

The patch is based on the EGuesnet's implement of the "Support of Big archive (read)
the first commit of the patch is come from https://reviews.llvm.org/D100651.

the rest of commits of the patch

1  Addressed the comments on the https://reviews.llvm.org/D100651
2  according to https://www.ibm.com/docs/en/aix/7.2?topic=formats-ar-file-format-big
using the "fl_fstmoff" for the first object file number, using "char ar_nxtmem[20]" to get next object file ,
using the "char fl_lstmoff[20]" for the last of the object file will fix the following problems:
   2.1 can not correct reading a archive files which has padding data between too object file
   2.2 can not correct reading a archive files from which some object file has be deleted

3 introduce a new derived class BigArchive for big ar file.

Reviewers: James Henderson
Differential Revision: https://reviews.llvm.org/D111889

2 years ago[clang][dataflow] Remove TestingSupport's dependency on gtest
Yitzhak Mandelbaum [Fri, 14 Jan 2022 12:57:07 +0000 (12:57 +0000)]
[clang][dataflow] Remove TestingSupport's dependency on gtest

Users outside of the clang repo may use different googletest versions. So, it's
better not to depend on llvm's googletest. This patch removes the dependency by
having `checkDataflow` return an `llvm::Error` instead of calling googletest's
`FAIL` or `ASSERT...` macros.

Differential Revision: https://reviews.llvm.org/D117304

2 years ago[libc++] [test] Improve the test for `uniform_int_distribution<T>`.
Arthur O'Dwyer [Thu, 2 Dec 2021 00:55:26 +0000 (19:55 -0500)]
[libc++] [test] Improve the test for `uniform_int_distribution<T>`.

Extracted from https://reviews.llvm.org/D114920

2 years ago[DAG] Extend SearchForAndLoads with any_extend handling
David Green [Mon, 17 Jan 2022 15:25:11 +0000 (15:25 +0000)]
[DAG] Extend SearchForAndLoads with any_extend handling

This extends the code in SearchForAndLoads to be able to look through
ANY_EXTEND nodes, which can be created from mismatching IR types where
the AND node we begin from only demands the low parts of the register.
That turns zext and sext into any_extends as only the low bits are
demanded. To be able to look through ANY_EXTEND nodes we need to handle
mismatching types in a few places, potentially truncating the mask to
the size of the final load.

Differential Revision: https://reviews.llvm.org/D117457

2 years ago[libc++] [ABI BREAK] Conform lognormal_distribution::param_type.
Arthur O'Dwyer [Tue, 28 Dec 2021 22:51:55 +0000 (17:51 -0500)]
[libc++] [ABI BREAK] Conform lognormal_distribution::param_type.

Fixes #52906.

Differential Revision: https://reviews.llvm.org/D116344

2 years ago[libc++] [test] Check for another kind of modulemap typo in lint_modulemap.sh.py.
Arthur O'Dwyer [Sun, 16 Jan 2022 19:56:42 +0000 (14:56 -0500)]
[libc++] [test] Check for another kind of modulemap typo in lint_modulemap.sh.py.

Verify that the name of the private submodule matches the name of the detail header.

Differential Revision: https://reviews.llvm.org/D117438

2 years agoAMDGPU: Fix crashing on calls to C functions from graphics contexts
Matt Arsenault [Sat, 15 Jan 2022 20:12:18 +0000 (15:12 -0500)]
AMDGPU: Fix crashing on calls to C functions from graphics contexts

If we had one of the shader calling conventions calling a default
calling convention callee, this would crash when the caller did not
have anything to pass to the workitem ID.

This is illegal, but we still need to produce something
sensible. llvm-reduce likes to replace calls to intrinsics with calls
to null or undef, so this does appear and is helpful to avoid hard
erroring.

Pass undef in this case, as already happened for the other implicit
arguments. It might make sense to define the behavior here and pass
null for the pointers, and -1 for the workitem ID. We do have extra
bits in the workitem ID, so this wouldn't conflict with a valid value.

2 years agoAMDGPU/GlobalISel: Fix selection of constant 32-bit addrspace loads
Matt Arsenault [Wed, 12 Jan 2022 21:28:49 +0000 (16:28 -0500)]
AMDGPU/GlobalISel: Fix selection of constant 32-bit addrspace loads

Unfortunately the selection patterns still rely on the address space
from the memory operand instead of using the pointer type. Add this
address space to the list of cases supported by global-like loads.

Alternatively we would have to adjust the address space of the memory
operand to deviate from the underlying IR value, which looks ugly and
is more work in the legalizer.

This doesn't come up in the DAG path because it uses a different
selection strategy where the cast is inserted during the addressing
mode matching.

2 years agoAMDGPU: Correct getMaxNumSGPR treatment of flat_scratch
Matt Arsenault [Fri, 14 Jan 2022 21:21:24 +0000 (16:21 -0500)]
AMDGPU: Correct getMaxNumSGPR treatment of flat_scratch

This was approximating the entry point logic for flat_scratch_init,
which is not really the point. We need to account for whether we need
to reserve the SGPR pair used for flat_scratch, not whether we needed
the initialization kernel argument. If this was an arbitrary function,
we would end up over-reporting the number of potentially free
SGPRs. The logic for architected flat scratch also only applies to the
initialization in the kernel, not the reserved registers at the end.

Avoids compile failures in a future patch from allocating more SGPRs
than the subtarget supports.

2 years agoAMDGPU/GlobalISel: Fix legalization failure for s65 shifts
Matt Arsenault [Sun, 16 Jan 2022 15:45:38 +0000 (10:45 -0500)]
AMDGPU/GlobalISel: Fix legalization failure for s65 shifts

This was trying to clamp s65 down to s32, which wasn't handled so we
need to promote all the way to s128 first. Having to order the
legalization rules in just the right way is rather dissatisfying, but
I'm not sure how smart the legalizer should be in trying to interpret
the rules.

2 years ago[VE] select|vp.merge|vp.select v256 isel and tests
Simon Moll [Mon, 17 Jan 2022 13:56:12 +0000 (14:56 +0100)]
[VE] select|vp.merge|vp.select v256 isel and tests

Use the `VMRG` for all three operations for now. `vp_select` will be
used in passthru patterns.

Reviewed By: kaz7

Differential Revision: https://reviews.llvm.org/D117206

2 years agoAMDGPU/GlobalISel: Add failing ABI lowering testcases
Matt Arsenault [Sun, 16 Jan 2022 16:14:58 +0000 (11:14 -0500)]
AMDGPU/GlobalISel: Add failing ABI lowering testcases

2 years agoAMDGPU: Fix LiveVariables error after optimizing VGPR ranges
Matt Arsenault [Wed, 5 Jan 2022 17:09:09 +0000 (12:09 -0500)]
AMDGPU: Fix LiveVariables error after optimizing VGPR ranges

This was not removing the block from the live set depending on the
specific depth first visit order. Fixes a verifier error in the OpenCL
conformance tests.

2 years agoAMDGPU/GlobalISel: Fix test not matching test name
Matt Arsenault [Sun, 16 Jan 2022 19:24:03 +0000 (14:24 -0500)]
AMDGPU/GlobalISel: Fix test not matching test name

This was testing an s48 load instead of an s64 load as intended.

2 years ago[AArch64][SVE] Remove Redundant aarch64.sve.convert.to.svbool
Matt Devereau [Wed, 5 Jan 2022 13:42:01 +0000 (13:42 +0000)]
[AArch64][SVE] Remove Redundant aarch64.sve.convert.to.svbool

Generated code resulted in redundant aarch64.sve.convert.to.svbool
calls for AArch64 Binary Operations. Narrow the more precise operands
instead of widening the less precise operands

Differential Revision: https://reviews.llvm.org/D116730

2 years ago[LV] Add test coverage for reductions with odd interleave counts.
Florian Hahn [Mon, 17 Jan 2022 14:32:40 +0000 (14:32 +0000)]
[LV] Add test coverage for reductions with odd interleave counts.

Add test coverage for loops with reductions and odd (3, 5) interleave
counts.

2 years ago[Clang][AArch64][ARM] PMUv3 Option Added
Mubashar Ahmad [Wed, 5 Jan 2022 16:53:59 +0000 (16:53 +0000)]
[Clang][AArch64][ARM] PMUv3 Option Added

An option has been added to Clang to enable or disable
the PMU v3 architecture extension.

Differential Revision: https://reviews.llvm.org/D116748

2 years agoRe-land [clangd] Elide even more checks in SelectionTree.
Sam McCall [Mon, 17 Jan 2022 14:13:27 +0000 (15:13 +0100)]
Re-land [clangd] Elide even more checks in SelectionTree.

This reverts commit 1093b9f2e9842982d97534940a643e3a4657c60b.

Fix added for implicit-include case.

2 years ago[clangd] Don't rename on symbols from system headers.
Haojian Wu [Wed, 5 Jan 2022 09:42:45 +0000 (10:42 +0100)]
[clangd] Don't rename on symbols from system headers.

Fixes https://github.com/clangd/clangd/issues/963.

Differential Revision: https://reviews.llvm.org/D116643

2 years ago[DSE] Remove commented-out InvisibleToCallerBeforeRet. (NFC)
Florian Hahn [Mon, 17 Jan 2022 13:59:13 +0000 (13:59 +0000)]
[DSE] Remove commented-out InvisibleToCallerBeforeRet. (NFC)

This code was is a leftover from earlier changes and should be removed.

2 years ago[InstCombine] propagate IR flags from binop through select
Sanjay Patel [Sun, 16 Jan 2022 18:36:49 +0000 (13:36 -0500)]
[InstCombine] propagate IR flags from binop through select

The tests with constant folding that produces poison
could potentially remove the select entirely:
https://alive2.llvm.org/ce/z/e-WUqF
...but this patch just removes the FMF-only limitation on
propagation.

2 years ago[InstCombine] add tests for binop with flags and select op; NFC
Sanjay Patel [Sun, 16 Jan 2022 18:01:43 +0000 (13:01 -0500)]
[InstCombine] add tests for binop with flags and select op; NFC

2 years ago[InstCombine] auto-generate complete test check lines; NFC
Sanjay Patel [Sun, 16 Jan 2022 17:44:46 +0000 (12:44 -0500)]
[InstCombine] auto-generate complete test check lines; NFC

2 years ago[LV] Make test more robust by adding users of inductions.
Florian Hahn [Sat, 15 Jan 2022 15:57:35 +0000 (15:57 +0000)]
[LV] Make test more robust by adding users of inductions.

The modified tests didn't have actual users of all inductions, making it
trivial to eliminate them. Add users to make sure the inductions are
actually used in the vectorized version.

2 years ago[VPlan] Drop unnecessary uses of getVPSingleValue (NFC).
Florian Hahn [Mon, 17 Jan 2022 13:27:33 +0000 (13:27 +0000)]
[VPlan] Drop unnecessary uses of getVPSingleValue (NFC).

2 years ago[CodeGen][AArch64] Ensure isSExtCheaperThanZExt returns true for negative constants
David Sherwood [Mon, 22 Nov 2021 11:38:06 +0000 (11:38 +0000)]
[CodeGen][AArch64] Ensure isSExtCheaperThanZExt returns true for negative constants

When we know the value we're extending is a negative constant then it
makes sense to use SIGN_EXTEND because this may improve code quality in
some cases, particularly when doing a constant splat of an unpacked vector
type. For example, for SVE when splatting the value -1 into all elements
of a vector of type <vscale x 2 x i32> the element type will get promoted
from i32 -> i64. In this case we want the splat value to sign-extend from
(i32 -1) -> (i64 -1), whereas currently it zero-extends from
(i32 -1) -> (i64 0xFFFFFFFF). Sign-extending the constant means we can use
a single mov immediate instruction.

New tests added here:

  CodeGen/AArch64/sve-vector-splat.ll

I believe we see some code quality improvements in these existing
tests too:

  CodeGen/AArch64/reduce-and.ll
  CodeGen/AArch64/unfold-masked-merge-vector-variablemask.ll

The apparent regressions in CodeGen/AArch64/fast-isel-cmp-vec.ll only
occur because the test disables codegen prepare and branch folding.

Differential Revision: https://reviews.llvm.org/D114357

2 years ago[AArch64] Add tests for sinking mask And to smaller loads. NFC
David Green [Mon, 17 Jan 2022 11:04:14 +0000 (11:04 +0000)]
[AArch64] Add tests for sinking mask And to smaller loads. NFC

2 years ago[RISCV] Add tests for scalable-vector vwsub patterns
Fraser Cormack [Sat, 15 Jan 2022 11:46:30 +0000 (11:46 +0000)]
[RISCV] Add tests for scalable-vector vwsub patterns

This patch adds tests for patterns introduced in D117188.

Reviewed By: jacquesguan

Differential Revision: https://reviews.llvm.org/D117392

2 years ago[clangd] Add option to use dirty file contents when building preambles.
Nathan James [Mon, 17 Jan 2022 10:55:35 +0000 (10:55 +0000)]
[clangd] Add option to use dirty file contents when building preambles.

Adds a option `use-dirty-preambles` to enable using unsaved in editor contents when building pre-ambles.
This enables a more seamless user experience when switching between header and implementation files and forgetting to save inbetween.
It's also in line with the LSP spec that states open files in the editor should be used instead of on the contents on disk - https://microsoft.github.io/language-server-protocol/overviews/lsp/overview/
For now the option is defaulted to off and hidden, Though I have a feeling it should be moved into the `.clangd` config and possibly defaulted to true.

Addresses https://github.com/clangd/clangd/issues/488

Reviewed By: sammccall

Differential Revision: https://reviews.llvm.org/D95046

2 years ago[MLIR][Linalg] Handle Attribute in InitTensorOp
Lorenzo Chelini [Thu, 13 Jan 2022 07:43:29 +0000 (08:43 +0100)]
[MLIR][Linalg] Handle Attribute in InitTensorOp

In some cases, the result of an initTensorOp may have an attribute.
However, the Attribute was not passed to `inferResultType`, failing the
verifier. Therefore, propagate the Attribute to `inferResultType`.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D117192

2 years agoReland (2) "[AST] Add RParen loc for decltype AutoTypeloc.""
Haojian Wu [Fri, 14 Jan 2022 10:28:04 +0000 (11:28 +0100)]
Reland (2) "[AST] Add RParen loc for decltype AutoTypeloc.""

The patch was reverted because it caused a crash during PCH build -- we
missed to update the RParenLoc in TreeTransform<Derived>::TransformAutoType.

This relands 55d96ac and 37ec65e with a test and fix.

2 years agoRevert "[mlir] fix crash in PybindAdaptors.h"
Alex Zinenko [Mon, 17 Jan 2022 09:55:36 +0000 (10:55 +0100)]
Revert "[mlir] fix crash in PybindAdaptors.h"

This reverts commit 970cb57ef72c9045250e0492cb00127b49ddfea8.

Broke the buildbot.

2 years ago[MachineInstr] Don't include debug uses in bundle header (PR52817)
Nikita Popov [Mon, 10 Jan 2022 15:11:24 +0000 (16:11 +0100)]
[MachineInstr] Don't include debug uses in bundle header (PR52817)

Following the recommendation in
https://github.com/llvm/llvm-project/issues/52817#issuecomment-1007635426,
this excludes debug instructions when finalizing the bundle. As uses
in debug instructions don't have effects, they will no longer be
included in the BUNDLE header.

Fixes https://github.com/llvm/llvm-project/issues/52817.

Differential Revision: https://reviews.llvm.org/D116945

2 years ago[clangd] Avoid a code completion crash
Haojian Wu [Mon, 17 Jan 2022 09:35:29 +0000 (10:35 +0100)]
[clangd] Avoid a code completion crash

This is a workaround (adding a newline to the eof) in clangd to avoid the code
completion crash, see https://github.com/clangd/clangd/issues/332.

In principle, this is a clang bug, we should fix it in clang, but it is not
trivial.

Reviewed By: sammccall

Differential Revision: https://reviews.llvm.org/D117456

2 years ago[LLDB] Fix Python GIL-not-held issues
Ralf Grosse-Kunstleve [Tue, 11 Jan 2022 18:34:38 +0000 (19:34 +0100)]
[LLDB] Fix Python GIL-not-held issues

The GIL must be held when calling any Python C API functions. In multithreaded applications that use callbacks this requirement can easily be violated by accident. A general tool to ensure GIL health is not available, but patching Python Py_INCREF to add an assert provides a basic health check:

```
+int PyGILState_Check(void); /* Include/internal/pystate.h */
+
 #define Py_INCREF(op) (                         \
+    assert(PyGILState_Check()),                 \
     _Py_INC_REFTOTAL  _Py_REF_DEBUG_COMMA       \
     ((PyObject *)(op))->ob_refcnt++)

 #define Py_DECREF(op)                                   \
     do {                                                \
+        assert(PyGILState_Check());                     \
         PyObject *_py_decref_tmp = (PyObject *)(op);    \
         if (_Py_DEC_REFTOTAL  _Py_REF_DEBUG_COMMA       \
         --(_py_decref_tmp)->ob_refcnt != 0)             \
```

Adding this assertion causes around 50 test failures in LLDB. Adjusting the scope of things guarded by `py_lock` fixes them.

More background: https://docs.python.org/3/glossary.html#term-global-interpreter-lock

Patch by Ralf Grosse-Kunstleve

Differential Revision: https://reviews.llvm.org/D114722

2 years ago[ValueTracking] Remove ComputeMultiple() function
Nikita Popov [Mon, 17 Jan 2022 09:00:54 +0000 (10:00 +0100)]
[ValueTracking] Remove ComputeMultiple() function

This function is no longer used since
499f1ca79f232faae09b1793a994d1a22ba403cd.

2 years ago[mlir] fix crash in PybindAdaptors.h
Alex Zinenko [Fri, 14 Jan 2022 16:25:41 +0000 (17:25 +0100)]
[mlir] fix crash in PybindAdaptors.h

The constructor function was being defined without indicating its "__init__"
name, which made it interpret it as a regular fuction rather than a
constructor. When overload resolution failed, Pybind would attempt to print the
arguments actually passed to the function, including "self", which is not
initialized since the constructor couldn't be called. This would result in
"__repr__" being called with "self" referencing an uninitialized MLIR C API
object, which in turn would cause undefined behavior when attempting to print
in C++.

Fix this by specifying the correct name.

This in turn uncovers the fact the the mechanism used by PybindAdaptors.h to
bind constructors directly as "__init__" functions taking "self" is deprecated
by Pybind. The modern method requires using "py::init", which seems to rely on
the C++ equivalent of the bound class to be available, which is not the case in
PybindAdaptors.h. A deeper inspection shows that the deprecation concerns
old-style pybind11 constructors that had to allocate the object using
placement new with "self" as memory. The PybindAdaptors.h only provides
extension classes and never allocates (the object construction is delegated to
the base class), so it does not use the deprecated functionality. Use the
implementation detail tag class to convince pybind11 that we are using the
modern constructor binding method and suppress the warning.

On top of that, the definition of the function was incorrectly indicated as the
method on the "None" object instead of being the method of its parent class.
This would result in a second problem when Pybind would attempt to print
warnings pointing to the parent class since the "None" does not have a
"__name__" field or its C API equivalent.

Fix this by specifying the correct parent class by looking it up by name in the
parent module.

Reviewed By: stellaraccident

Differential Revision: https://reviews.llvm.org/D117325

2 years ago[RISCV] Add scheduler for bfp instruction in Zbf extension
Lian Wang [Mon, 17 Jan 2022 08:58:49 +0000 (08:58 +0000)]
[RISCV] Add scheduler for bfp instruction in Zbf extension

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D117290

2 years ago[GlobalOpt] Drop an incorrect check
Nikita Popov [Mon, 17 Jan 2022 09:09:44 +0000 (10:09 +0100)]
[GlobalOpt] Drop an incorrect check

This was a last-minute addition to D117249, and of course I ended
up inverting the condition in a way that caused an uninitialized
memory read.

I've dropped it entirely, as I don't think we actually care whether
the size is zero or not here. The previous code wasn't checking
this either.

2 years ago[GlobalOpt] Use generic type when converting malloc to global
Nikita Popov [Tue, 11 Jan 2022 13:24:49 +0000 (14:24 +0100)]
[GlobalOpt] Use generic type when converting malloc to global

The malloc to global transform currently determines the type of the
global by looking at bitcasts of the malloc. This is limited (the
transform fails if there are multiple different types) and
incompatible with opaque pointers.

My initial approach was to construct an appropriate struct type
based on usage in loads/stores. What this patch does instead is
to always create an [i8 x AllocSize] global, without trying to
guess types at all.

This does mean that other transforms that require a certain global
type may break. I fixed two of these in D117034 and D117223, which
I believe should be sufficient to avoid regressions. In particular,
the global SRA change should end up splitting the global into
naturally-typed sub-globals, at which point all other optimizations
should work.

Differential Revision: https://reviews.llvm.org/D117092

2 years ago[docs] Clarify & update JSONCompilationDatabase docs
Sam McCall [Sun, 16 Jan 2022 15:00:25 +0000 (16:00 +0100)]
[docs] Clarify & update JSONCompilationDatabase docs

- prefer `arguments` over `command`, and add example
- clarify that there's no shell-unescaping of `arguments`

Fixes https://github.com/llvm/llvm-project/issues/53143

Differential Revision: https://reviews.llvm.org/D117428

2 years ago[ELF] GnuHashTableSection: replace stable_sort with 2-key sort. NFC
Fangrui Song [Mon, 17 Jan 2022 08:34:42 +0000 (00:34 -0800)]
[ELF] GnuHashTableSection: replace stable_sort with 2-key sort. NFC

strTabOffset stabilizes llvm::sort. My x86-64 executable is 5+KiB smaller.

2 years ago[GlobalOpt] Make global SRA offset based
Nikita Popov [Thu, 13 Jan 2022 11:39:43 +0000 (12:39 +0100)]
[GlobalOpt] Make global SRA offset based

Currently global SRA uses the GEP structure to determine how to
split the global. This patch instead analyses the loads and stores
that are performed on the global, and collects which types are used
at which offset, and then splits the global according to those.

This is both more general, and works fine with opaque pointers.
This is also closer to how ordinary SROA is performed.

Differential Revision: https://reviews.llvm.org/D117223

2 years ago[clangd] Better handling `\n` in the synthesized diagnostic message.
Haojian Wu [Mon, 17 Jan 2022 08:17:55 +0000 (09:17 +0100)]
[clangd] Better handling `\n` in the synthesized diagnostic message.

The newline-eof fix was rendered as "insert '...'", this patch
special-case it.

Reviewed By: sammccall

Differential Revision: https://reviews.llvm.org/D117294

2 years ago[DSE] Remove alloc function check in canSkipDef()
Nikita Popov [Tue, 11 Jan 2022 11:34:11 +0000 (12:34 +0100)]
[DSE] Remove alloc function check in canSkipDef()

canSkipDef() currently skips inaccessiblememonly calls, but not
if they are allocation functions. This check was added in D103009,
but actually seems to be a leftover from a previous implementation
in D101440. canSkipDef() is not used on the storeIsNoop() path,
where the relevant transform ended up being implemented.

Differential Revision: https://reviews.llvm.org/D117005

2 years agoCycleInfo: Fix trivial typo. NFC.
Carl Ritson [Mon, 17 Jan 2022 08:04:28 +0000 (17:04 +0900)]
CycleInfo: Fix trivial typo. NFC.

2 years ago[ELF] RelocationScanner::scanOne: replace rel.r_offset with offset. NFC
Fangrui Song [Mon, 17 Jan 2022 08:05:27 +0000 (00:05 -0800)]
[ELF] RelocationScanner::scanOne: replace rel.r_offset with offset. NFC

2 years ago[clang-format] Add a BlockIndent option to AlignAfterOpenBracket
Cameron Mulhern [Mon, 17 Jan 2022 07:54:51 +0000 (08:54 +0100)]
[clang-format] Add a BlockIndent option to AlignAfterOpenBracket

This style is similar to AlwaysBreak, but places closing brackets on new lines.

For example, if you have a multiline parameter list, clang-format currently only supports breaking per-parameter, but places the closing bracket on the line of the last parameter.

Function(
    param1,
    param2,
    param3);

A style supported by other code styling tools (e.g. rustfmt) is to allow the closing brackets to be placed on their own line, aiding the user in being able to quickly infer the bounds of the block of code.

Function(
    param1,
    param2,
    param3
);

For prior work on a similar feature, see: https://reviews.llvm.org/D33029.

Note: This currently only supports block indentation for closing parentheses.

Differential Revision: https://reviews.llvm.org/D109557

2 years ago[ELF] Relocations: remove some cast<Undefined>. NFC
Fangrui Song [Mon, 17 Jan 2022 08:02:47 +0000 (00:02 -0800)]
[ELF] Relocations: remove some cast<Undefined>. NFC

2 years ago[ELF] De-template getAlternativeSpelling. NFC
Fangrui Song [Mon, 17 Jan 2022 07:56:24 +0000 (23:56 -0800)]
[ELF] De-template getAlternativeSpelling. NFC

2 years ago[RISCV] Add initial support for getRegUsageForType and getNumberOfRegisters
Kito Cheng [Sun, 9 Jan 2022 14:16:05 +0000 (22:16 +0800)]
[RISCV] Add initial support for getRegUsageForType and getNumberOfRegisters

Those two TTI hooks are used during vectorization for calculating
register pressure, the default implementation isn't consider for LMUL,
and that's also definitly wrong value for register number (all register class
are 8 registers).

So in this patch we tried to:

1. Calculate right register usage for vector type and scalar type.
2. Return right number of register for general purpose register and
   vector register.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D116890

2 years ago[PowerPC] Allow -mfloat128 option for VSX targets
Qiu Chaofan [Mon, 17 Jan 2022 07:12:33 +0000 (15:12 +0800)]
[PowerPC] Allow -mfloat128 option for VSX targets

Targets with VSX feature but without native float128 instructions can
also use that type with supplementary libcalls. We don't enable it by
default now because Glibc assumes long double and float128 can be
implicitly converted in between, which is not available under default
'ibmlongdouble' semantics in clang.

This commit partly relands cbd93ce.

2 years ago[ELF] Remove unneeded SyntheticSection memset(*, 0, *)
Fangrui Song [Mon, 17 Jan 2022 06:51:57 +0000 (22:51 -0800)]
[ELF] Remove unneeded SyntheticSection memset(*, 0, *)

After the D33630 fallout was properly fixed by a4c5db30be4e216834b44e31b47304ea1b92635f.

Tested by D37462/D44986 tests, the new --no-rosegment test in build-id.s, and a few --rosegment/--no-rosegment programs.

2 years ago[ELF] Remove redundant fillTrap and memset(*, 0, *). NFC
Fangrui Song [Mon, 17 Jan 2022 06:37:31 +0000 (22:37 -0800)]
[ELF] Remove redundant fillTrap and memset(*, 0, *). NFC

The new tests in build-id.s would catch problems if we made a mistake here.

2 years ago[ELF][test] Add --build-id tests for -z separate-loadable-segments and --no-rosegment
Fangrui Song [Mon, 17 Jan 2022 06:36:22 +0000 (22:36 -0800)]
[ELF][test] Add --build-id tests for -z separate-loadable-segments and --no-rosegment

2 years ago[ELF] RelocationSection<ELFT>::writeTo: use unstable partition
Fangrui Song [Mon, 17 Jan 2022 05:44:19 +0000 (21:44 -0800)]
[ELF] RelocationSection<ELFT>::writeTo: use unstable partition

2 years agoReland https://reviews.llvm.org/D113825 after fixing the test expectations.
esmeyi [Mon, 17 Jan 2022 05:28:25 +0000 (00:28 -0500)]
Reland https://reviews.llvm.org/D113825 after fixing the test expectations.

2 years ago[ELF] Change some DenseMap<StringRef, *> to DenseMap<CachedHashStringRef, *>. NFC
Fangrui Song [Mon, 17 Jan 2022 05:19:01 +0000 (21:19 -0800)]
[ELF] Change some DenseMap<StringRef, *> to DenseMap<CachedHashStringRef, *>. NFC

2 years ago[ELF] StringTableSection: Use DenseMap<CachedHashStringRef> to avoid redundant hash...
Fangrui Song [Mon, 17 Jan 2022 05:02:05 +0000 (21:02 -0800)]
[ELF] StringTableSection: Use DenseMap<CachedHashStringRef> to avoid redundant hash computation

5~6% speedup when linking clang and chrome.

2 years ago[Driver][FreeBSD] -r: imply -nostdlib like GCC
Fangrui Song [Mon, 17 Jan 2022 03:44:48 +0000 (19:44 -0800)]
[Driver][FreeBSD] -r: imply -nostdlib like GCC

Similar to D116843 for Gnu.cpp

Reviewed By: dim

Differential Revision: https://reviews.llvm.org/D117388

2 years ago[compiler-rt][cmake] Use `GNUInstallDirs` to support custom installation dirs
John Ericson [Sun, 16 Jan 2022 06:19:45 +0000 (06:19 +0000)]
[compiler-rt][cmake] Use `GNUInstallDirs` to support custom installation dirs

I am breaking apart D99484 so the cause of build failures is easier to
understand.

Differential Revision: https://reviews.llvm.org/D117420

2 years ago[RISCV] Add patterns for vector widening integer multiply-add instructions
eopXD [Sat, 15 Jan 2022 20:12:22 +0000 (12:12 -0800)]
[RISCV] Add patterns for vector widening integer multiply-add instructions

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D117404

2 years agoPreserve argument locations when cloning a region.
Dominik Grewe [Sun, 16 Jan 2022 21:16:57 +0000 (21:16 +0000)]
Preserve argument locations when cloning a region.

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D117403

2 years ago[AttrBuilder] Remove unused removeAttributes() overload
Nikita Popov [Sun, 16 Jan 2022 20:32:54 +0000 (21:32 +0100)]
[AttrBuilder] Remove unused removeAttributes() overload

The idiomatic way would be to call remove() with an AttributeMask
constructed from an AttributeSet.

2 years ago[AttrBuilder] Remove unused hasAttributes() overload
Nikita Popov [Sun, 16 Jan 2022 19:58:38 +0000 (20:58 +0100)]
[AttrBuilder] Remove unused hasAttributes() overload

This is unused, and doesn't make a lot of sense as an API. The
usual pattern would be to combine the AttrBuilder(AttributeSet)
constructor with the overlaps() method.

2 years ago[DAGCombine] Refactor DAGCombiner::ReduceLoadWidth. NFCI
Bjorn Pettersson [Tue, 11 Jan 2022 21:59:18 +0000 (22:59 +0100)]
[DAGCombine] Refactor DAGCombiner::ReduceLoadWidth. NFCI

Update code comments in DAGCombiner::ReduceLoadWidth and refactor
the handling of SRL a bit. The refactoring is done with the intent
of adding support for folding away SRA by using SEXTLOAD in a
follow-up patch.

The function is also renamed as DAGCombiner::reduceLoadWidth.

Differential Revision: https://reviews.llvm.org/D117104

2 years ago[clang][dataflow] Add transfer functions for bind temporary and static cast
Stanislav Gatev [Fri, 14 Jan 2022 18:27:39 +0000 (18:27 +0000)]
[clang][dataflow] Add transfer functions for bind temporary and static cast

This is part of the implementation of the dataflow analysis framework.
See "[RFC] A dataflow analysis framework for Clang AST" on cfe-dev.

Differential Revision: https://reviews.llvm.org/D117339

2 years ago[MLIR] Simplex::getRationalSample: return an optional, empty if Simplex is empty
Arjun P [Wed, 12 Jan 2022 16:36:47 +0000 (22:06 +0530)]
[MLIR] Simplex::getRationalSample: return an optional, empty if Simplex is empty

2 years ago[gn build] Port e6b153947dfa
LLVM GN Syncbot [Sun, 16 Jan 2022 16:03:49 +0000 (16:03 +0000)]
[gn build] Port e6b153947dfa

2 years agoRevert [LLD] Remove global state in lldCommon
Alexandre Ganea [Sun, 16 Jan 2022 16:03:06 +0000 (11:03 -0500)]
Revert [LLD] Remove global state in lldCommon

It seems to be causing issues on https://lab.llvm.org/buildbot/#/builders/123/builds/8383

2 years ago[LLD] Supplement with more comments. Clarify the intention in f860fe362282ed69b9d4503...
Alexandre Ganea [Sun, 16 Jan 2022 14:17:29 +0000 (09:17 -0500)]
[LLD] Supplement with more comments. Clarify the intention in f860fe362282ed69b9d4503a20e5d20b9a041189.

2 years agoResolve lit failures in clang-aarch64*
hyeongyu kim [Sun, 16 Jan 2022 14:02:55 +0000 (23:02 +0900)]
Resolve lit failures in clang-aarch64*

2 years ago[gn build] Port f860fe362282
LLVM GN Syncbot [Sun, 16 Jan 2022 13:58:27 +0000 (13:58 +0000)]
[gn build] Port f860fe362282

2 years ago[LLD] Remove global state in lldCommon
Alexandre Ganea [Sun, 16 Jan 2022 02:47:54 +0000 (21:47 -0500)]
[LLD] Remove global state in lldCommon

Move all variables at file-scope or function-static-scope into a hosting structure (lld::CommonLinkerContext) that lives at lldMain()-scope. Drivers will inherit from this structure and add their own global state, in the same way as for the existing COFFLinkerContext.

See discussion in https://lists.llvm.org/pipermail/llvm-dev/2021-June/151184.html

Differential Revision: https://reviews.llvm.org/D108850

2 years ago[LV] Restore metadata to disable runtime unrolling for epilogue loop.
Florian Hahn [Sun, 16 Jan 2022 13:14:16 +0000 (13:14 +0000)]
[LV] Restore metadata to disable runtime unrolling for epilogue loop.

After d4a8fc3a87a1 LV stopped adding metadata to disable runtime
unrolling to the vectorized epilogue loop. This was missed because
278aa65cc495 removed the relevant test coverage.

This patch fixes that by adding the relevant metadata after
vector loop generation.

2 years ago[msan] reflect the changed flag to the tests.
hyeongyu kim [Sun, 16 Jan 2022 11:42:35 +0000 (20:42 +0900)]
[msan] reflect the changed flag to the tests.

1b1c8d changed `enable-noundef-analysis` flag to
`disable-noundef-analysis`.  noundef_analysis.cpp was using old
`enable-noundef-analysis` flag and this patch fixes it.

2 years ago[clang-format] Add return code to git-clang-format
owenca [Sun, 16 Jan 2022 03:41:06 +0000 (19:41 -0800)]
[clang-format] Add return code to git-clang-format

https://github.com/llvm/llvm-project/issues/53220

Differential Revision: https://reviews.llvm.org/D117414

2 years ago[LV] Move AddRuntimeUnrollDisableMetaData so it can be used earlier (NFC)
Florian Hahn [Sun, 16 Jan 2022 10:28:55 +0000 (10:28 +0000)]
[LV] Move AddRuntimeUnrollDisableMetaData so it can be used earlier (NFC)

Move up the definition of AddRuntimeUnrollDisableMetaData, so it can be
re-used earlier in the file in a follow-up patch.

2 years ago[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn...
hyeongyu kim [Sun, 16 Jan 2022 09:53:11 +0000 (18:53 +0900)]
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default

Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions.
I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default.

Test updates are made as a separate patch: D108453

Reviewed By: eugenis

Differential Revision: https://reviews.llvm.org/D105169

2 years ago[ELF] Remove forEachRelSec. NFC
Fangrui Song [Sun, 16 Jan 2022 08:28:47 +0000 (00:28 -0800)]
[ELF] Remove forEachRelSec. NFC

2 years ago[ELF] Remove !isLazy() condition from computeBinding
Fangrui Song [Sun, 16 Jan 2022 07:58:15 +0000 (23:58 -0800)]
[ELF] Remove !isLazy() condition from computeBinding

Seems applicable since we demote lazy symbols to Undefined (D111365).

2 years ago[ELF] Remove config->relocatable condition from Symbol::computeBinding
Fangrui Song [Sun, 16 Jan 2022 07:49:47 +0000 (23:49 -0800)]
[ELF] Remove config->relocatable condition from Symbol::computeBinding

2 years ago[ELF] Speed up Symbol::computeBinding. NFC
Fangrui Song [Sun, 16 Jan 2022 07:40:43 +0000 (23:40 -0800)]
[ELF] Speed up Symbol::computeBinding. NFC

When computeBinding is inlined into includeInDynsym and computeIsPreemptible,
the optimizer can remove the config->gnuUnique load.

2 years ago[ELF] Slightly speed up Symbol::includeInDynsym. NFC
Fangrui Song [Sun, 16 Jan 2022 07:32:48 +0000 (23:32 -0800)]
[ELF] Slightly speed up Symbol::includeInDynsym. NFC