platform/upstream/llvm.git
2 years ago[libcxx] [test] Fix the monetary locale pos/neg_format test for Windows and macOS
Martin Storsjö [Tue, 18 Jan 2022 09:20:04 +0000 (09:20 +0000)]
[libcxx] [test] Fix the monetary locale pos/neg_format test for Windows and macOS

The zh_CN.UTF-8 locale on Glibc has got `n_sign_posn == 4` (which means
having the negative sign just after the currency symbol), but has
`int_n_sign_posn == 1` (which means before the string).

On Windows, there's no separate `int_n_sign_posn` field, so the same
`n_sign_posn` (which is 4 there too) is used for international currency
formatting too. This makes the ordering for the international case on
Windows be the same as for the national one right above it.

On Apple platforms, the fr_FR.UTF-8 locale has got `n_sign_posn == 2`
but `p_sign_posn == 1`, giving a different order for the French locale
for the negative format.

On Apple platforms for the zh_CN.UTF-8 locale, both `n_sign_posn` and
`int_n_sign_posn` are 4, but `p_sign_posn` and `int_p_sign_posn` are 1.

Differential Revision: https://reviews.llvm.org/D120550

2 years ago[sancov] Add SPARC support for getPreviousInstructionPc
Xiaodong Liu [Sun, 27 Feb 2022 19:10:51 +0000 (19:10 +0000)]
[sancov] Add SPARC support for getPreviousInstructionPc

The SPARC and MIPS branching operations have a branch delay slot, 4 more bytes occupied.

Depends on D120381

Reviewed By: ro, MaskRay

Differential Revision: https://reviews.llvm.org/D120451

2 years ago[SLP] Remove field unused after 33ce97f to silence buildbots [NFC]
Philip Reames [Sun, 27 Feb 2022 18:18:03 +0000 (10:18 -0800)]
[SLP] Remove field unused after 33ce97f to silence buildbots [NFC]

2 years agoRevert "[VPlan] Introduce recipe to build scalar steps."
Florian Hahn [Sun, 27 Feb 2022 17:51:19 +0000 (17:51 +0000)]
Revert "[VPlan] Introduce recipe to build scalar steps."

This reverts commit 49b23f451cf713036c99573a35daed308d2ac894.

This appears to break some PPC build bots. Revert while I investigate.

2 years ago[SLP] Use BatchAA to reduce capture analysis cost [NFC]
Philip Reames [Sun, 27 Feb 2022 17:44:11 +0000 (09:44 -0800)]
[SLP] Use BatchAA to reduce capture analysis cost [NFC]

SLP makes very heavy use of aliasing queries to construct pointer dependencies for scheduling purposes.  AA internally usings pointerMayBeCaptured to prove some noalias results.  In a local profile, we were spending about 4% of total O2 time in capture tracking.  By using BatchAA interface - which caches capture results - this drops to 2%.

Note that there is no invalidation of BatchAA here.  This assumes that no transformation done by SLP invalidates alias or capture results.  This is the same assumption made by the existing AliasCache, so this is not a new assumption in the code.

2 years ago[VPlan] Introduce recipe to build scalar steps.
Florian Hahn [Sun, 27 Feb 2022 17:32:41 +0000 (17:32 +0000)]
[VPlan] Introduce recipe to build scalar steps.

This patch adds a new VPScalarIVStepsRecipe to handle building scalar
steps.

In the first patch, it only handles the case where there is no vector
induction variable needed.

Reviewed By: Ayal

Differential Revision: https://reviews.llvm.org/D115953

2 years ago[libc++] [test] Check the presence of "pragma include_instead" in newly added headers.
Arthur O'Dwyer [Sun, 27 Feb 2022 17:19:21 +0000 (12:19 -0500)]
[libc++] [test] Check the presence of "pragma include_instead" in newly added headers.

Unless/until we revert D106124, we should make sure that every newly added
detail header includes this line.

2 years ago[MLIR][Presburger] Move IntegerPolyhedron::reset to FlatAffineConstraints::reset
Groverkss [Sun, 27 Feb 2022 16:57:15 +0000 (22:27 +0530)]
[MLIR][Presburger] Move IntegerPolyhedron::reset to FlatAffineConstraints::reset

This patch moves IntegerPolyhedron::reset to FlatAffineConstraints::reset. This
function is not required in IntegerPolyhedron and creates ambiguity while
shifting implementations to IntegerRelation.

This patch is part of a series of patches to introduce relations in Presburger
library.

Reviewed By: arjunp

Differential Revision: https://reviews.llvm.org/D120628

2 years ago[SDAG] fold (rotate X) eq/ne (0/-1)
Sanjay Patel [Sun, 27 Feb 2022 16:26:26 +0000 (11:26 -0500)]
[SDAG] fold (rotate X) eq/ne (0/-1)

This is the SDAG equivalent of an instcombine transform added with:
fd807601a78

This is another step towards solving #49541 and part of an alternative
set of more general transforms than what is proposed in D111530.

https://alive2.llvm.org/ce/z/ToxaE8

2 years ago[x86] add tests for setcc eq/ne 0/-1 of rotate; NFC
Sanjay Patel [Sun, 27 Feb 2022 15:44:02 +0000 (10:44 -0500)]
[x86] add tests for setcc eq/ne 0/-1 of rotate; NFC

The tests are adapted from a similar file for instcombine:
fd807601a78

2 years ago[X86][SSE] Attempt to lower vec_reduce_add patterns with PSADBW for zero-extended...
Simon Pilgrim [Sun, 27 Feb 2022 15:17:42 +0000 (15:17 +0000)]
[X86][SSE] Attempt to lower vec_reduce_add patterns with PSADBW for zero-extended vXi8 sources

For i16/32/64 vectors, if the upper bits are known to be zero, then we can try to truncate to vXi8 (if its worth it) and perform this as a PSADBW to add+zext each v4i8 subvector to a i64 sum, which we can then reduce together.

This addresses some of the PR42674 test cases where the source data was vXi8 but had been extended to match a wider unsigned integer accumulator.

Differential Revision: https://reviews.llvm.org/D120193

2 years ago[SDAG] fold bitwise logic with shifted operands
Sanjay Patel [Sun, 27 Feb 2022 13:33:43 +0000 (08:33 -0500)]
[SDAG] fold bitwise logic with shifted operands

LOGIC (LOGIC (SH X0, Y), Z), (SH X1, Y) --> LOGIC (SH (LOGIC X0, X1), Y), Z

https://alive2.llvm.org/ce/z/QmR9rR

This is a reassociation + factoring fold. The common shift operation is moved
after a bitwise logic op on 2 input operands.
We get simpler cases of these patterns in IR, but I suspect we would miss all
of these exact tests in IR too. We also handle the simpler form of this plus
several other folds in DAGCombiner::hoistLogicOpWithSameOpcodeHands().

This is a partial implementation of a transform suggested in D111530
(only handles 'or' bitwise logic as a first step - need to stamp out more
tests for other opcodes).
Several of the same tests added for D111530 are altered here (but not
fully optimized). I'm not sure yet if this would help/hinder that patch,
but this should be an improvement for all tests added with ecf606cb4329ae
since it removes a shift operation in those examples.

Differential Revision: https://reviews.llvm.org/D120516

2 years agoRevert "[Support] Reset option to its default if its Default field is undefined"
Yevgeny Rouban [Sun, 27 Feb 2022 14:46:10 +0000 (21:46 +0700)]
Revert "[Support] Reset option to its default if its Default field is undefined"

This reverts commit 7fb39fb6d6665cd469557b43eb205cc32b0a7ac3 as clang buildbots failed.

2 years ago[Support] Reset option to its default if its Default field is undefined
Yevgeny Rouban [Sun, 27 Feb 2022 12:56:31 +0000 (19:56 +0700)]
[Support] Reset option to its default if its Default field is undefined

opt::setDefaultImpl() is changed to set the option value to the option
type's default if the Default field is not set. This results in option
value reset by Option::reset() or ResetAllOptionOccurrences() even if
the cl::init() is not specified.

Example:
  StackOption<std::string> Str("str"); // No cl::init().
  Str = "some value";
  cl::ResetAllOptionOccurrences();
  EXPECT_EQ("", Str); // The Str is reset.

Reviewed By: lattner
Differential Revision: https://reviews.llvm.org/D115433

2 years ago[VPlan] Add recipe to handle SCEV expansion (NFC).
Florian Hahn [Sun, 27 Feb 2022 12:44:07 +0000 (12:44 +0000)]
[VPlan] Add recipe to handle SCEV expansion (NFC).

This can be used to explicitly model VPValues that depend on SCEV
expansion, like the step for inductions.

Reviewed By: Ayal

Differential Revision: https://reviews.llvm.org/D116288

2 years ago[lldb] BreakpointOptions::CommandData::CreateFromStructuredData - remove dead code...
Simon Pilgrim [Sun, 27 Feb 2022 11:33:14 +0000 (11:33 +0000)]
[lldb] BreakpointOptions::CommandData::CreateFromStructuredData - remove dead code + variable. NFCI.

The found_something bool is only ever read after it has always been set to true.

Looks to be a leftover debugging variable.

Fixes static analyzer warning: https://llvm.org/reports/scan-build/report-BreakpointOptions.cpp-CreateFromStructuredData-8-4055b9.html#EndPath

2 years ago[DAG] Ensure type is legal for bswap(shl(x,c)) -> zext(bswap(trunc(shl(x,c-bw/2)...
Simon Pilgrim [Sun, 27 Feb 2022 11:25:17 +0000 (11:25 +0000)]
[DAG] Ensure type is legal for bswap(shl(x,c)) -> zext(bswap(trunc(shl(x,c-bw/2)))) fold

As reported on D120192

2 years ago[C++20][Modules][6/8] Record direct module imports.
Iain Sandoe [Sun, 30 Jan 2022 14:26:57 +0000 (14:26 +0000)]
[C++20][Modules][6/8] Record direct module imports.

This is a small cache to avoid having to check both Exports and
Imports.

Differential Revision: https://reviews.llvm.org/D118589

2 years ago[AMDGPU] Remove redundant isVALU in SIPreEmitPeephole. NFC
Carl Ritson [Sun, 27 Feb 2022 04:23:22 +0000 (13:23 +0900)]
[AMDGPU] Remove redundant isVALU in SIPreEmitPeephole. NFC

Remove redundant isVALU call added in D120202.

2 years ago[ELF] BitcodeFile: resolve defined symbols before undefined symbols
Fangrui Song [Sun, 27 Feb 2022 05:37:08 +0000 (05:37 +0000)]
[ELF] BitcodeFile: resolve defined symbols before undefined symbols

This ports D95985 for ELF relocatable object files to BitcodeFile.

2 years ago[ConstantFolding] Fix folding of constrained compare intrinsics
Serge Pavlov [Sat, 26 Feb 2022 16:39:12 +0000 (23:39 +0700)]
[ConstantFolding] Fix folding of constrained compare intrinsics

The change fixes treatment of constrained compare intrinsics if
compared values are of vector type.

Differential revision: https://reviews.llvm.org/D110322

2 years ago[bazel] Port g95b4e88b1db3
Benjamin Kramer [Sat, 26 Feb 2022 23:11:42 +0000 (00:11 +0100)]
[bazel] Port g95b4e88b1db3

2 years ago[clang] MarkVarDeclODRUsed - remove redundant nullptr check. NFCI.
Simon Pilgrim [Sat, 26 Feb 2022 21:24:15 +0000 (21:24 +0000)]
[clang] MarkVarDeclODRUsed - remove redundant nullptr check. NFCI.

The function has already been dereferenced the Var pointer

2 years ago[clangd] Test fixes missing from 257559ed9
Sam McCall [Sat, 26 Feb 2022 20:38:25 +0000 (21:38 +0100)]
[clangd] Test fixes missing from 257559ed9

2 years ago[clangd] Function return type hints: support lambdas, don't duplicate "->"
Sam McCall [Wed, 23 Feb 2022 15:59:19 +0000 (16:59 +0100)]
[clangd] Function return type hints: support lambdas, don't duplicate "->"

While here, fix an ugliness:
  auto foo()->auto { return 42; }
This (silly) code gains a "-> int" hint. While correct and useful, it renders as
  auto foo()->int->auto { return 42; }
which is confusing enough to do more harm than good I think.

Differential Revision: https://reviews.llvm.org/D120416

2 years ago[clang-format][NFC] Rename test and remove comments
Björn Schäpers [Wed, 23 Feb 2022 06:07:13 +0000 (07:07 +0100)]
[clang-format][NFC] Rename test and remove comments

Why put "InMacros" in the name? We test other things to, and I will add
more, withut macros.

Also all our tests are regression tests.

Differential Revision: https://reviews.llvm.org/D120401

2 years ago[clang-format][NFC] Remove redundant semi
Björn Schäpers [Tue, 22 Feb 2022 22:26:17 +0000 (23:26 +0100)]
[clang-format][NFC] Remove redundant semi

All "calls" have a semi, as they should, remove the one from the macro.

Differential Revision: https://reviews.llvm.org/D120359

2 years agoUse RegisterInfo::regsOverlaps instead of checking aliases
Benjamin Kramer [Sat, 26 Feb 2022 19:29:16 +0000 (20:29 +0100)]
Use RegisterInfo::regsOverlaps instead of checking aliases

This is both less code and faster since it doesn't have to expand all
the sub & superreg sets. NFCI.

2 years ago[PDLL] Properly error out on returning results from native constraints
River Riddle [Mon, 14 Feb 2022 21:39:06 +0000 (13:39 -0800)]
[PDLL] Properly error out on returning results from native constraints

PDL currently doesn't support result values from constraints, meaning we need
to error out until this is actually supported to avoid crashes.

Differential Revision: https://reviews.llvm.org/D119782

2 years ago[mlir:PDLL] Add support for C++ generation
River Riddle [Mon, 14 Feb 2022 07:51:52 +0000 (23:51 -0800)]
[mlir:PDLL] Add support for C++ generation

This commits adds a C++ generator to PDLL that generates wrapper PDL patterns
directly usable in C++ code, and also generates the definitions of native constraints/rewrites
that have code bodies specified in PDLL. This generator is effectively the PDLL equivalent of
the current DRR generator, and will allow easy replacement of DRR patterns with PDLL patterns.
A followup will start to utilize this for end-to-end integration testing and show case how to
use this as a drop-in replacement for DRR tablegen usage.

Differential Revision: https://reviews.llvm.org/D119781

2 years ago[mlir:PDLL] Fix handling of unspecified operands/results on operation expressions
River Riddle [Mon, 14 Feb 2022 07:51:46 +0000 (23:51 -0800)]
[mlir:PDLL] Fix handling of unspecified operands/results on operation expressions

If the operand list or result list of an operation expression is not specified, we interpret
this as meaning that the operands/results are "unconstraint" (i.e. "could be anything").
We currently don't properly handle differentiating this case from the case of
"no operands/results". This commit adds the insertion of implicit value/type range
variables when these lists are unspecified. This allows for adding proper support
for when zero operands or results are expected.

Differential Revision: https://reviews.llvm.org/D119780

2 years ago[mlir:PDLL] Add support for PDL MLIR code generation
River Riddle [Mon, 14 Feb 2022 07:51:21 +0000 (23:51 -0800)]
[mlir:PDLL] Add support for PDL MLIR code generation

This commits starts to plumb PDLL down into MLIR and adds an initial
PDL generator. After this commit, we will have conceptually support
end-to-end execution of PDLL. Followups will add CPP generation to
match the current DRR setup, and begin to add various end-to-end
tests to test PDLL execution.

Differential Revision: https://reviews.llvm.org/D119779

2 years ago[AArch64] Cleanup and extend cast costs. NFC
David Green [Sat, 26 Feb 2022 17:59:02 +0000 (17:59 +0000)]
[AArch64] Cleanup and extend cast costs. NFC

2 years ago[libc++] [test] Re-remove C++ comments from generated files.
Arthur O'Dwyer [Sat, 26 Feb 2022 17:44:26 +0000 (12:44 -0500)]
[libc++] [test] Re-remove C++ comments from generated files.

A merge conflict in D106124 accidentally reverted this part of
b82683b2e/D110794.

> Even if these comments have a benefit in .h files (for editors that
> care about language but can't be configured to treat .h as C++ code),
> they certainly have no benefit for files with the .cpp extension.

2 years ago[libcxx] [test] Remove an incorrect XFAIL, fix CI on main
Martin Storsjö [Sat, 26 Feb 2022 17:28:16 +0000 (12:28 -0500)]
[libcxx] [test] Remove an incorrect XFAIL, fix CI on main

9f5f084 (D119770) made this test pass in the Windows configuration too,
but didn't update the XFAIL accordingly.

Differential Revision: https://reviews.llvm.org/D120611

2 years ago[MLIR][Presburger] Remove redundant check from hasConsistentState
Groverkss [Sat, 26 Feb 2022 16:32:12 +0000 (22:02 +0530)]
[MLIR][Presburger] Remove redundant check from hasConsistentState

This patch removes a redundant check in hasConsistentState which is always true
after introduction of PresburgerSpace.

Reviewed By: arjunp

Differential Revision: https://reviews.llvm.org/D120615

2 years ago[VPlan] Remove dead header-phi recipes.
Florian Hahn [Sat, 26 Feb 2022 16:24:25 +0000 (16:24 +0000)]
[VPlan] Remove dead header-phi recipes.

This patch adds a new transform to remove dead recipes. For now, it only
removes dead recipes in the header, to keep the number tests that require
updating manageable. Future patches will extend this to remove dead
recipes across the whole plan.

Reviewed By: Ayal

Differential Revision: https://reviews.llvm.org/D118051

2 years ago[gn build] (manually) port 2e6ae1d3f2de more
Nico Weber [Sat, 26 Feb 2022 15:46:20 +0000 (10:46 -0500)]
[gn build] (manually) port 2e6ae1d3f2de more

b85f97bc00ae7 missed one file.

With this, all files listed in libcxx/include/CMakeLists.txt are
listed in this BUILD.gn file again.

2 years ago[gn build] (manually) port 01ace074fcb6 more
Nico Weber [Sat, 26 Feb 2022 15:43:11 +0000 (10:43 -0500)]
[gn build] (manually) port 01ace074fcb6 more

410d4492e395a only hit half the added files.

2 years ago[gn build] Define _LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER when building libcxx
Nico Weber [Sat, 26 Feb 2022 15:24:48 +0000 (10:24 -0500)]
[gn build] Define _LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER when building libcxx

Ports 87a82490fc6b89fd3 in a way. Has the effect of enabling warnings
when building libcxx itself, but more importantly this is required
after 5aaefa510ef0 to not get build errors when building libcxx itself.

2 years ago[LV] Add test with redundant cast in separate latch block.
Florian Hahn [Sat, 26 Feb 2022 14:48:45 +0000 (14:48 +0000)]
[LV] Add test with redundant cast in separate latch block.

Adds another interesting test for D118051.

2 years ago[clangd] Support IncludeFixer or base specifiers
Sam McCall [Sat, 26 Feb 2022 14:35:01 +0000 (15:35 +0100)]
[clangd] Support IncludeFixer or base specifiers

2 years agoRevert "[Driver] Default CLANG_DEFAULT_PIE_ON_LINUX to ON"
Nikita Popov [Sat, 26 Feb 2022 14:19:45 +0000 (15:19 +0100)]
Revert "[Driver] Default CLANG_DEFAULT_PIE_ON_LINUX to ON"

See post-commit discussion on https://reviews.llvm.org/D120305.
This change breaks the clang-ppc64le-rhel buildbot, though
there is suspicion that it's an issue with the bot. The change
also had a larger than expected impact on compile-time and
code-size.

This reverts commit 3c4ed02698afec021c6bca80740d1e58e3ee019e
and some followup changes.

2 years ago[MLIR][Presburger] Move IdKind specific insert/append
Groverkss [Sat, 26 Feb 2022 13:24:19 +0000 (18:54 +0530)]
[MLIR][Presburger] Move IdKind specific insert/append

This patch moves identifier kind specific insert/append functions like
`insertDimId`, `appendSymbolId`, etc. from IntegerPolyhedron to
FlatAffineConstraints.

This change allows for a smoother transition to IntegerRelation.

This change is part of a series of patches to introduce Relations in Presburger
library.

Reviewed By: arjunp

Differential Revision: https://reviews.llvm.org/D120576

2 years ago[clangd] Fix wrong included header. NFC
Sam McCall [Sat, 26 Feb 2022 13:21:52 +0000 (14:21 +0100)]
[clangd] Fix wrong included header. NFC

2 years ago[clangd] Fix include-cleaner false-positive bug
Sam McCall [Sat, 26 Feb 2022 13:08:12 +0000 (14:08 +0100)]
[clangd] Fix include-cleaner false-positive bug

For TemplateSpecializationType, we were checking the node's newness
twice, so it always failed the second test.

Fixes https://github.com/clangd/clangd/issues/1036

2 years ago[clangd] Qualify calls to std::move to silence -Wunqualified-std-cast-call. NFC.
Benjamin Kramer [Sat, 26 Feb 2022 12:34:35 +0000 (13:34 +0100)]
[clangd] Qualify calls to std::move to silence -Wunqualified-std-cast-call. NFC.

2 years ago[MLIR][Presburger] Factor out various Space equality checks to PresburgerSpace::isEqual
Groverkss [Sat, 26 Feb 2022 10:57:22 +0000 (16:27 +0530)]
[MLIR][Presburger] Factor out various Space equality checks to PresburgerSpace::isEqual

This patch factors out various checks for dimension compatibility to
PresburgerSpace::isEqual and PresburgerLocalSpace::isEqual (for local
identifiers).

Reviewed By: arjunp

Differential Revision: https://reviews.llvm.org/D120565

2 years ago[MLIR][Presburger] coalesce: add assert noting that locals are not yet supported
Arjun P [Sat, 26 Feb 2022 12:21:47 +0000 (12:21 +0000)]
[MLIR][Presburger] coalesce: add assert noting that locals are not yet supported

2 years ago[libc++] Remove recursion in basic_string::insert(const_iterator, ForwardIterator...
Nikolas Klauser [Sat, 26 Feb 2022 12:28:33 +0000 (13:28 +0100)]
[libc++] Remove recursion in basic_string::insert(const_iterator, ForwardIterator, ForwardIterator)

`__addr_in_range` is a non-constexpr function, so we can't call it during constant evaluation.

Reviewed By: Quuxplusone, #libc, miscco

Spies: miscco, libcxx-commits

Differential Revision: https://reviews.llvm.org/D119633

2 years ago[clangd] Add a missing include. NFC.
Benjamin Kramer [Sat, 26 Feb 2022 12:23:06 +0000 (13:23 +0100)]
[clangd] Add a missing include. NFC.

2 years ago[C++20][Modules][5/8] Diagnose wrong import/export for partition CMIs.
Iain Sandoe [Sun, 30 Jan 2022 14:11:57 +0000 (14:11 +0000)]
[C++20][Modules][5/8] Diagnose wrong import/export for partition CMIs.

We cannot export partition implementation CMIs, but we can export the content
of partition interface CMIs.

Differential Revision: https://reviews.llvm.org/D118588

2 years ago[clangd] Clean up unused includes. NFCI
Sam McCall [Sat, 26 Feb 2022 10:40:12 +0000 (11:40 +0100)]
[clangd] Clean up unused includes. NFCI

Add includes where needed to fix build.
Haven't systematically added used headers, so there is still accidental
dependency on transitive includes.

2 years ago[Verifier] Restore defined-resolver verification for IFuncs
Itay Bookstein [Sat, 29 Jan 2022 12:33:18 +0000 (14:33 +0200)]
[Verifier] Restore defined-resolver verification for IFuncs

Now that clang no longer emits GlobalIFunc-s with a
declaration for a resolver, we can restore that check.
In addition, add a linkage check like the one we have
on GlobalAlias-es, and a Verifier test for ifuncs.

Signed-off-by: Itay Bookstein <ibookstein@gmail.com>
Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D120267

2 years ago[MLIR][Presburger] fix double increment in coalesce
Michel Weber [Sat, 26 Feb 2022 10:30:42 +0000 (10:30 +0000)]
[MLIR][Presburger] fix double increment in coalesce

In the main-loop of the current coalesce implementation `i` was incremented
twice for some cases. This patch fixes this bug and adds a regression
testcase.

Reviewed By: arjunp

Differential Revision: https://reviews.llvm.org/D120613

2 years ago[clang][CodeGen] Avoid emitting ifuncs with undefined resolvers
Itay Bookstein [Sat, 29 Jan 2022 12:32:54 +0000 (14:32 +0200)]
[clang][CodeGen] Avoid emitting ifuncs with undefined resolvers

The purpose of this change is to fix the following codegen bug:

```
// main.c
__attribute__((cpu_specific(generic)))
int *foo(void) { static int z; return &z;}
int main() { return *foo() = 5; }

// other.c
__attribute__((cpu_dispatch(generic))) int *foo(void);

// run:
clang main.c other.c -o main; ./main
```

This will segfault prior to the change, and return the correct
exit code 5 after the change.

The underlying cause is that when a translation unit contains
a cpu_specific function without the corresponding cpu_dispatch
the generated code binds the reference to foo() against a
GlobalIFunc whose resolver is undefined. This is invalid: the
resolver must be defined in the same translation unit as the
ifunc, but historically the LLVM bitcode verifier did not check
that. The generated code then binds against the resolver rather
than the ifunc, so it ends up calling the resolver rather than
the resolvee. In the example above it treats its return value as
an int *, therefore trying to write to program text.

The root issue at the representation level is that GlobalIFunc,
like GlobalAlias, does not support a "declaration" state. The
object which provides the correct semantics in these cases
is a Function declaration, but unlike Functions, changing a
declaration to a definition in the GlobalIFunc case constitutes
a change of the object type, as opposed to simply emitting code
into a Function.

I think this limitation is unlikely to change, so I implemented
the fix by returning a function declaration rather than an ifunc
when encountering cpu_specific, and upgrading it to an ifunc
when emitting cpu_dispatch.
This uses `takeName` + `replaceAllUsesWith` in similar vein to
other places where the correct IR object type cannot be known
locally/up-front, like in `CodeGenModule::EmitAliasDefinition`.

Previous discussion in: https://reviews.llvm.org/D112349

Signed-off-by: Itay Bookstein <ibookstein@gmail.com>
Reviewed By: erichkeane

Differential Revision: https://reviews.llvm.org/D120266

2 years ago[libcxx][modules] protects users from relying on detail headers
Christopher Di Bella [Fri, 25 Feb 2022 18:59:32 +0000 (18:59 +0000)]
[libcxx][modules] protects users from relying on detail headers

libc++ has started splicing standard library headers into much more
fine-grained content for maintainability. It's very likely that outdated
and naive tooling (some of which is outside of LLVM's scope) will
suggest users include things such as <__ranges/access.h> instead of
<ranges>, and Hyrum's law suggests that users will eventually begin to
rely on this without the help of tooling. As such, this commit
intends to protect users from themselves, by making it a hard error for
anyone outside of the standard library to include libc++ detail headers.

Differential Revision: https://reviews.llvm.org/D106124

2 years ago[test] Precisely disable flaky sanitizer tests on clang-ppc64le-redhat
Fangrui Song [Sat, 26 Feb 2022 07:25:44 +0000 (07:25 +0000)]
[test] Precisely disable flaky sanitizer tests on clang-ppc64le-redhat

See https://reviews.llvm.org/D120305#3347094

2 years ago[SeparateConstOffsetFromGEP] Remove TargetMachine.h include. NFC
Craig Topper [Sat, 26 Feb 2022 05:39:03 +0000 (21:39 -0800)]
[SeparateConstOffsetFromGEP] Remove TargetMachine.h include. NFC

This doesn't appear to be used and it would be a layering violation
if it was.

2 years ago[SLP] "Normal" instructions should not go between PHI and Lading pad
Evgeniy Brevnov [Fri, 25 Feb 2022 11:09:07 +0000 (18:09 +0700)]
[SLP] "Normal" instructions should not go between PHI and Lading pad

Currently, SLP can insert "shuffle" instruction beween PHI and Landing pad instruction. The problem is demonstrated by LIT test. The solution is to adjust insertion point once we are done with PHI generation.

Differential Revision: https://reviews.llvm.org/D120552

2 years ago[crt][test] Add -no-pie to avoid wacky linking with crt1.o if (musl || CLANG_DEFAULT_...
Fangrui Song [Sat, 26 Feb 2022 04:41:49 +0000 (04:41 +0000)]
[crt][test] Add -no-pie to avoid wacky linking with crt1.o if (musl || CLANG_DEFAULT_PIE_ON_LINUX)

This works with glibc crt1.o (its crt1.o is essentially Scrt1.o plus (if static
PIE is supported for the arch) _dl_relocate_static_pie) but looks wacky.

musl crt1.o is not built with an explicit -fno-pic/-fpie/-fpic. If it was built
with a non-default-pie GCC/Clang, the linker might complain about absolute
relocations referencing _init/_fini for the -pie link:

```
ld: error: relocation R_X86_64_32 cannot be used against symbol '_fini'; recompile with -fPIC
>>> defined in obj/crt/crt1.o
>>> referenced by crt1.c
>>>               obj/crt/crt1.o:(_start_c)
```

2 years ago[ELF][X86] Don't create IBT .plt if there is no PLT entry
Joao Moreira [Sat, 26 Feb 2022 03:55:39 +0000 (03:55 +0000)]
[ELF][X86] Don't create IBT .plt if there is no PLT entry

https://github.com/ClangBuiltLinux/linux/issues/1606
When GNU_PROPERTY_X86_FEATURE_1_IBT is enabled, ld.lld will create .plt output
section even if there is no PLT entry. Fix this by implementing
IBTPltSection::isNeeded instead of using the default code path (which always
returns true).

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D120600

2 years ago[lldb/Host] Fix crash in FileSystem::IsLocal
Med Ismail Bennani [Sat, 26 Feb 2022 02:33:31 +0000 (18:33 -0800)]
[lldb/Host] Fix crash in FileSystem::IsLocal

This checks `m_fs` before dereferencing it to access its`isLocal` method.

rdar://67410058

Signed-off-by: Med Ismail Bennani <medismail.bennani@gmail.com>
2 years agoReland "[lldb/test] Fix TestProgressReporting.py race issue with the event listener"
Med Ismail Bennani [Sat, 26 Feb 2022 01:20:39 +0000 (17:20 -0800)]
Reland "[lldb/test] Fix TestProgressReporting.py race issue with the event listener"

This patch relands commit 3e3e79a9e4c378b59f5f393f556e6a84edcd8898, and
fixes the memory sanitizer issue described in D120284, by removing the
output arguments from the LLDB_INSTRUMENT_VA invocation.

Differential Revision: https://reviews.llvm.org/D120599

Signed-off-by: Med Ismail Bennani <medismail.bennani@gmail.com>
2 years ago[lldb/crashlog] Fix scripted_crashlog_json.test failure
Med Ismail Bennani [Sat, 26 Feb 2022 00:31:46 +0000 (16:31 -0800)]
[lldb/crashlog] Fix scripted_crashlog_json.test failure

This patch should fix the test failure on scripted_crashlog_json.test.

The failure is happening because crash reporter will obfuscate the
executable path in the crashlog, if it is located inside the user's
home directory and replace it with `/USER/*/` as a placeholder.

To fix that, we can patch the placeholder with the executable path
before loading the crashlog in lldb.

This also fixes a bug where we would create another target when loading
the crashlog in a scripted process, even if lldb already had a target
for it. Now, crashlog will only create a target if there is none in lldb.

Differential Revision: https://reviews.llvm.org/D120598

Signed-off-by: Med Ismail Bennani <medismail.bennani@gmail.com>
2 years ago[Driver][test] Clean up some AIX tests
Fangrui Song [Sat, 26 Feb 2022 01:06:24 +0000 (01:06 +0000)]
[Driver][test] Clean up some AIX tests

* For `-###`, `-o %t.o` is unnecessary if we don't specifically test the
  output filename.
* --target= is the canonical spelling. -target is a legacy spelling which
  unfortunately cannot be removed because there are too many uses.
* -no-canonical-prefixes uses the dereferenced absolute path for the cc1
  command. For most tests "-cc1" is sufficient to identify the command line, no
  need to specifically test the "clang" command, and -no-canonical-prefixes can
  removed.
* --unwindlib= is the preferred spelling. -u is a short option taking a value,
  which means a -uwindlib= typo cannot be detected.

I recommend that you take a look at linux-cross.cpp. Testing include paths and
library paths in one RUN line is sometimes more readable than having separate
include/library tests.

Having separate RUN lines for misc features like -fdata-sections
(`aix-data-sections.c`) is wasteful. It may be better testing multiple
options in a single RUN command.

Reviewed By: jsji

Differential Revision: https://reviews.llvm.org/D119309

2 years agoRevert "Don't append the working directory to absolute paths"
Adrian Prantl [Sat, 26 Feb 2022 01:00:10 +0000 (17:00 -0800)]
Revert "Don't append the working directory to absolute paths"

This reverts commit 2cd9a86da54f8be4eb2aff3e766b125cbdeb023f.

2 years ago[mlir][sparse] merge ifs in new sparse rewriting rules
Aart Bik [Thu, 24 Feb 2022 17:40:01 +0000 (09:40 -0800)]
[mlir][sparse] merge ifs in new sparse rewriting rules

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D120500

2 years ago[mlir][sparse] enable scalar test
Aart Bik [Fri, 25 Feb 2022 21:18:50 +0000 (13:18 -0800)]
[mlir][sparse] enable scalar test

Removed TODO now that we support scalars properly

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D120590

2 years ago[BPF] Fix tests that fail if /tmp/t.c exists
Reid Kleckner [Fri, 25 Feb 2022 22:52:38 +0000 (14:52 -0800)]
[BPF] Fix tests that fail if /tmp/t.c exists

IMO the BPF backend shouldn't read random source files referenced from
debug info. I filed llvm.org/pr54092 about this.

2 years ago[NewPM] Fix tests for OptimizerLast extension point
Elia Geretto [Fri, 25 Feb 2022 22:49:13 +0000 (14:49 -0800)]
[NewPM] Fix tests for OptimizerLast extension point

This PR fixes the tests for the `OptimizerLast` extension point in the new pass manager. The extension point uses module passes, but it was being tested with a function pass.

Reviewed By: aeubanks

Differential Revision: https://reviews.llvm.org/D120581

2 years ago[NewPM] Add extension points to LTO pipeline in PassBuilder
Elia Geretto [Fri, 25 Feb 2022 20:02:37 +0000 (12:02 -0800)]
[NewPM] Add extension points to LTO pipeline in PassBuilder

This PR adds two extension points to the default LTO pipeline in PassBuilder, one at the beginning and one at the end. These two extension points already existed in the old pass manager, the aim is to replicate the same functionality in the new one.

Reviewed By: aeubanks

Differential Revision: https://reviews.llvm.org/D120491

2 years ago[mlir][sparse][taco] Support reduction to scalar tensors.
Bixia Zheng [Fri, 25 Feb 2022 17:28:39 +0000 (09:28 -0800)]
[mlir][sparse][taco] Support reduction to scalar tensors.

The PyTACO DSL doesn't support reduction to scalars. This change
enhances the MLIR-PyTACO implementation to support reduction to scalars.

Extend an existing test to show the syntax of reduction to scalars and
two methods to retrieve the scalar values.

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D120572

2 years ago[Symbolizer] Move default ctor into .cpp file
Reid Kleckner [Thu, 24 Feb 2022 23:31:46 +0000 (15:31 -0800)]
[Symbolizer] Move default ctor into .cpp file

Follow up to 1e396affca6a0d21247d960c93a415e8f6fe0301.  On some standard
library configurations these have a dependency on the complete type of
SymbolizableModule.

2 years ago[Mangler] Mangle aliases to fastcall/vectorcall functions correctly
Amanieu d'Antras [Fri, 25 Feb 2022 22:06:47 +0000 (22:06 +0000)]
[Mangler] Mangle aliases to fastcall/vectorcall functions correctly

These aliases are produced by MergeFunctions and need to be mangled according to the calling convention of the function they are pointing to instead of defaulting to the C calling convention.

Reviewed By: rnk

Differential Revision: https://reviews.llvm.org/D120382

2 years ago[Triple] Add llvm::Triple::isSPARC{,32,64}
Xiaodong Liu [Fri, 25 Feb 2022 21:50:34 +0000 (21:50 +0000)]
[Triple] Add llvm::Triple::isSPARC{,32,64}

Reviewed By: ro, MaskRay

Differential Revision: https://reviews.llvm.org/D120381

2 years ago[libcxx] Fix the error checking for wctob_l, fixing locale narrow function on Windows
Martin Storsjö [Thu, 20 Jan 2022 22:38:52 +0000 (22:38 +0000)]
[libcxx] Fix the error checking for wctob_l, fixing locale narrow function on Windows

According to POSIX.1 (and Glibc docs, and Microsoft docs), the wctob
function returns EOF on error, not WEOF. (And wctob_l should consequently
do the same.)

The previous misconception about what this function returns on errors
seems to stem from incorrect documentation in macOS, stemming from BSD
docs with the same issue. The corresponding documentation bug in FreeBSD
was fixed in 2012 in
https://github.com/freebsd/freebsd-src/commit/945aab90991bdaeabeb6ef25112975a96c01dd4e,
but it hasn't been fixed for macOS yet.

The issue seems to only be a documentation issue; the implementation
on macOS actually does use EOF, not WEOF:
https://opensource.apple.com/source/Libc/Libc-1439.40.11/locale/FreeBSD/wctob.c.auto.html

On most Unices, EOF and WEOF are the same value, but on Windows,
EOF is -1, while WEOF is (unsigned short)0xFFFF. By fixing this,
two tests start passing on Windows.

Differential Revision: https://reviews.llvm.org/D120088

2 years ago[libcxx] [test] Fix the monetary locale negative_sign test for en_US.UTF-8 on Windows
Martin Storsjö [Thu, 13 Jan 2022 13:02:16 +0000 (13:02 +0000)]
[libcxx] [test] Fix the monetary locale negative_sign test for en_US.UTF-8 on Windows

On Windows, the en_US.UTF-8 locale returns `n_sign_posn == 0`, which
means that the sign for a negative currency is parentheses around
the whole value, instead of a leading minus.

Differential Revision: https://reviews.llvm.org/D120549

2 years ago[lldb] Fix check for TARGET_OS_IPHONE
Jonas Devlieghere [Fri, 25 Feb 2022 21:23:28 +0000 (13:23 -0800)]
[lldb] Fix check for TARGET_OS_IPHONE

Instead of checking whether TARGET_OS_IPHONE is set to 1, the current
code just check the existence of TARGET_OS_IPHONE, which either always
succeeds or always fails, depending on whether you have
TargetConditionals.h included.

2 years ago[dsymutil] Copy symbol table regardless of LINKEDIT segment
Jonas Devlieghere [Fri, 25 Feb 2022 19:37:03 +0000 (11:37 -0800)]
[dsymutil] Copy symbol table regardless of LINKEDIT segment

Ensure we copy the symbol table for MH_PRELOAD Mach-Os, which don't have
a LINKEDIT segment, but (can) have a symbol table.

rdar://88919473

Differential revision: https://reviews.llvm.org/D120583

2 years agoValidate chained fixup image formats
Adrian Prantl [Fri, 25 Feb 2022 19:43:44 +0000 (11:43 -0800)]
Validate chained fixup image formats

This is part of a series of patches to upstream support for Mach-O
chained fixups.

Differential Revision: https://reviews.llvm.org/D113725

2 years agoDon't append the working directory to absolute paths
Adrian Prantl [Fri, 25 Feb 2022 19:38:16 +0000 (11:38 -0800)]
Don't append the working directory to absolute paths

This fixes a bug that happens when using -fdebug-prefix-map to remap
an absolute path to a relative path. Since the path was absolute
before remapping, it is safe to assume that concatenating the remapped
working directory would be wrong.

Differential Revision: https://reviews.llvm.org/D113718

2 years ago[clang-format] Handle trailing comment for InsertBraces
owenca [Thu, 24 Feb 2022 18:15:11 +0000 (10:15 -0800)]
[clang-format] Handle trailing comment for InsertBraces

Differential Revision: https://reviews.llvm.org/D120503

2 years agoEnable tests from rG8e67982384d4a11892c04d16c2d10d7533e56094 that seem to work now
Jameson Nash [Fri, 25 Feb 2022 20:20:22 +0000 (15:20 -0500)]
Enable tests from rG8e67982384d4a11892c04d16c2d10d7533e56094 that seem to work now

I noticed randomly that the only reason these tests from
rG8e67982384d4a11892c04d16c2d10d7533e56094 seemed to still be failing is
that they are missing CHECK lines. I don't know anymore than that they
don't appear to crash or assert when I ran them today.

Reviewed By: fhahn

Differential Revision: https://reviews.llvm.org/D115377

2 years ago[HIP] Fix test hip-link-bundled-archive.hip
Yaxun (Sam) Liu [Fri, 25 Feb 2022 13:58:23 +0000 (08:58 -0500)]
[HIP] Fix test hip-link-bundled-archive.hip

match pattern should match lld.exe on windows

Reviewed by: Shangwu Yao

Differential Revision: https://reviews.llvm.org/D120563

2 years ago[flang] Lower logical comparison and logical operations
Valentin Clement [Fri, 25 Feb 2022 20:05:09 +0000 (21:05 +0100)]
[flang] Lower logical comparison and logical operations

This handles the lowering of the logical comparison
to `arith.cmpi` operation. The logical operations `.OR.`, `.AND.`
and `.NOT.` are lowered to `arith.ori`, `arith.andi` and `arith.xori`

This patch is part of the upstreaming effort from fir-dev branch.

Depends on D120559

Reviewed By: schweitz, rovka

Differential Revision: https://reviews.llvm.org/D120560

Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: Jean Perier <jperier@nvidia.com>
2 years ago[flang] Lower real comparison operations
Valentin Clement [Fri, 25 Feb 2022 20:02:53 +0000 (21:02 +0100)]
[flang] Lower real comparison operations

This patch handles the lowering of real
comparison operations. The real comparison operations
are lowered to `arith.cmpf` operation.

This patch is part of the upstreaming effort from fir-dev branch.

Reviewed By: PeteSteinfeld, schweitz

Differential Revision: https://reviews.llvm.org/D120561

Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: Jean Perier <jperier@nvidia.com>
2 years ago[flang] Lower integer comparison operation
Valentin Clement [Fri, 25 Feb 2022 20:00:07 +0000 (21:00 +0100)]
[flang] Lower integer comparison operation

This patch handles the lowering of comprison
operator between integers.
The comparison is lowered to a `arith.cmpi` operation.

This patch is part of the upstreaming effort from fir-dev branch.

Reviewed By: PeteSteinfeld, schweitz, rovka

Differential Revision: https://reviews.llvm.org/D120559

Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: Jean Perier <jperier@nvidia.com>
2 years ago[libunwind][test] remember_state_leak.pass.sh.s: link with -no-pie
Fangrui Song [Fri, 25 Feb 2022 19:55:56 +0000 (19:55 +0000)]
[libunwind][test] remember_state_leak.pass.sh.s: link with -no-pie

The no-pic large code model style `movabsq $callback, %rsi` does not work with -pie.

2 years ago[mlir][Linalg] Add support for tileFuseAndDistribute on tensors.
Hanhan Wang [Fri, 25 Feb 2022 18:52:08 +0000 (10:52 -0800)]
[mlir][Linalg] Add support for tileFuseAndDistribute on tensors.

This extends TileAndFuse to handle distribution on tensors.

Reviewed By: gysit

Differential Revision: https://reviews.llvm.org/D120441

2 years ago[RISCV] Remove tab character from test. Autogenerate CHECK lines. NFC
Craig Topper [Fri, 25 Feb 2022 19:36:23 +0000 (11:36 -0800)]
[RISCV] Remove tab character from test. Autogenerate CHECK lines. NFC

This was a test for an infinite loop so the CHECK lines don't really
matter, but they'd get generated the next time someone runs the script
on the file so might as well do it while I'm touching it.

2 years agomark getTargetTransformInfo and getTargetIRAnalysis as const
Jameson Nash [Fri, 25 Feb 2022 19:24:42 +0000 (14:24 -0500)]
mark getTargetTransformInfo and getTargetIRAnalysis as const

Seems like this can be const, since Passes shouldn't modify it.

Reviewed By: wsmoses

Differential Revision: https://reviews.llvm.org/D120518

2 years ago[mlir][Vector] Prevent AVX2 lowering for non-f32 transpose ops
Diego Caballero [Fri, 25 Feb 2022 18:27:43 +0000 (18:27 +0000)]
[mlir][Vector] Prevent AVX2 lowering for non-f32 transpose ops

The AVX2 lowering for transpose operations is only applicable to f32 vector types.

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D120427

2 years ago[mlir][Vector] Generalize AVX2 transpose lowering to n-D vectors
Diego Caballero [Fri, 25 Feb 2022 18:26:37 +0000 (18:26 +0000)]
[mlir][Vector] Generalize AVX2 transpose lowering to n-D vectors

The existing AVX2 lowering patterns for the transpose op only triggers if the
input vector is 2-D. This patch extends the patterns to trigger for n-D vectors
which are effectively 2-D vectors (e.g., vector<1x4x1x8x1). The main constraint
for the generalized AVX2 patterns to be applicable to these vectors is that the
dimensions that are greater than one must be transposed. Otherwise, the existing
patterns are not applicable.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D119505

2 years ago[ELF] Support some absolute/PC-relative relocation types for REL format
Fangrui Song [Fri, 25 Feb 2022 19:25:18 +0000 (19:25 +0000)]
[ELF] Support some absolute/PC-relative relocation types for REL format

ctfconvert seems to use REL-format `.rel.SUNW_dof` for 32-bit architectures.
```
Binary file usr/ports/lang/perl5.32/work/perl-5.32.1/dtrace_mini.o matches
[alfredo.junior@dell-a ~/tmp/llvm-bug]$ readelf -r dtrace_mini.o

Relocation section (.rel.SUNW_dof):
r_offset r_info   r_type              st_value st_name
00000184 0000281a R_PPC_REL32         00000000 $dtrace1772974259.Perl_dtrace_probe_load
```

Support R_PPC_REL32 to fix `ld.lld: error: drti.c:(.SUNW_dof+0x4E4): internal linker error: cannot read addend for relocation R_PPC_REL32`.
While here, add some common relocation types for AArch64, PPC, and PPC64.
We perform minimum tests.

Reviewed By: adalava, arichardson

Differential Revision: https://reviews.llvm.org/D120535

2 years ago[mlir] Support verification order (2/3)
Chia-hung Duan [Fri, 25 Feb 2022 18:17:30 +0000 (18:17 +0000)]
[mlir] Support verification order (2/3)

    This change gives explicit order of verifier execution and adds
    `hasRegionVerifier` and `verifyWithRegions` to increase the granularity
    of verifier classification. The orders are as below,

    1. InternalOpTrait will be verified first, they can be run independently.
    2. `verifyInvariants` which is constructed by ODS, it verifies the type,
       attributes, .etc.
    3. Other Traits/Interfaces that have marked their verifier as
       `verifyTrait` or `verifyWithRegions=0`.
    4. Custom verifier which is defined in the op and has marked
       `hasVerifier=1`

    If an operation has regions, then it may have the second phase,

    5. Traits/Interfaces that have marked their verifier as
       `verifyRegionTrait` or
       `verifyWithRegions=1`. This implies the verifier needs to access the
       operations in its regions.
    6. Custom verifier which is defined in the op and has marked
       `hasRegionVerifier=1`

    Note that the second phase will be run after the operations in the
    region are verified. Based on the verification order, you will be able to
    avoid verifying duplicate things.

Reviewed By: Mogball

Differential Revision: https://reviews.llvm.org/D116789

2 years ago[OPENMP]Fix PR50347: Mapping of global scope deep object fails.
Alexey Bataev [Thu, 1 Jul 2021 11:16:56 +0000 (04:16 -0700)]
[OPENMP]Fix PR50347: Mapping of global scope deep object fails.

Changed the we handle llvm::Constants in sizes arrays. ConstExprs and
GlobalValues cannot be used as initializers, need to put them at the
runtime, otherwise there wight be the compilation errors.

Differential Revision: https://reviews.llvm.org/D105297

2 years agoRevert "[lsan][test] Temporarily disable ppc64 and ppc64le to appease clang-ppc64le...
Fangrui Song [Fri, 25 Feb 2022 18:50:39 +0000 (18:50 +0000)]
Revert "[lsan][test] Temporarily disable ppc64 and ppc64le to appease clang-ppc64le-rhel"

This reverts commit cb76c4d71c41bbbae47852d7980e74b57c5a28df.

The failures were in test/sanitizer_common, not in test/lsan.

2 years ago[compiler-rt][test] Temporarily disable ppc64 and ppc64le test/sanitizer_common and...
Fangrui Song [Fri, 25 Feb 2022 18:49:10 +0000 (18:49 +0000)]
[compiler-rt][test] Temporarily disable ppc64 and ppc64le test/sanitizer_common and test/crt

to appease clang-ppc64le-rhel: https://github.com/llvm/llvm-project/issues/54084

2 years ago[AMDGPU][NFC]: Emit metadata for hidden_heap_v1 kernarg
Changpeng Fang [Fri, 25 Feb 2022 18:45:35 +0000 (10:45 -0800)]
[AMDGPU][NFC]: Emit metadata for hidden_heap_v1 kernarg

Summary:
  Emit metadata for hidden_heap_v1 kernarg

Reviewers:
  sameerds, b-sumner

Fixes:
  SWDEV-307188

Differential Revision:
  https://reviews.llvm.org/D119027