platform/upstream/llvm.git
19 months ago[clang-tools-extra] Fix linking ClangdTests when using libclang-cpp
Michał Górny [Mon, 20 Mar 2023 15:00:10 +0000 (16:00 +0100)]
[clang-tools-extra] Fix linking ClangdTests when using libclang-cpp

Fix linking ClangdTests to specify the dependency on the private
clangTesting library via target_link_libraries() rather than
clang_target_link_libraries().  The latter uses libclang-cpp when
CLANG_LINK_CLANG_DYLIB is used, and clangTesting is not included
in this library.

This fixes d60d3455eb2b375d026a4aa74c4ba0c38f5d323c.

Differential Revision: https://reviews.llvm.org/D146427

19 months ago[libc] Add support for setjmp and longjmp in riscv
Mikhail R. Gadelha [Fri, 24 Mar 2023 19:13:19 +0000 (16:13 -0300)]
[libc] Add support for setjmp and longjmp in riscv

This patch implements setjmp and longjmp in riscv using inline asm. The
following changes were required:

* Omit frame pointer: otherwise gcc won't allow us to use s0
* Use __attribute__((naked)): otherwise both gcc and clang will generate
function prologue and epilogue in both functions. This doesn't happen
in x86_64, so we guard it to only riscv

Furthermore, using __attribute__((naked)) causes two problems: we
can't use `return 0` (both gcc and clang) and the function arguments in
the function body (clang only), so we had to use a0 and a1 directly.

Reviewed By: sivachandra

Differential Revision: https://reviews.llvm.org/D145584

19 months ago[ConstraintElim] Use GEPOperator instead of GetElementPtrInst.
Florian Hahn [Fri, 24 Mar 2023 19:14:34 +0000 (19:14 +0000)]
[ConstraintElim] Use GEPOperator instead of GetElementPtrInst.

The logic in ConstraintElimination should trivially apply to GEP
constant expressions as well, so update code to deal with GEPOperator
instead.

19 months ago[BOLT] Don't use section relocations when computing hash for data from other section
Denis Revunov [Fri, 24 Mar 2023 18:50:07 +0000 (21:50 +0300)]
[BOLT] Don't use section relocations when computing hash for data from other section

When computing symbol hashes in BinarySection::hash, we try to find relocations
in the section which reference the passed BinaryData. We do so by doing
lower_bound on data begin offset and upper_bound on data end offset. Since
offsets are relative to the current section, if it is a data from the previous
section, we get underflow when computing offset and lower_bound returns
Relocations.end(). If this data also ends where current section begins,
upper_bound on zero offset will return some valid iterator if we have any
relocations after the first byte. Then we'll try to iterate from lower_bound to
upper_bound, since they're not equal, which in that case means we'll dereference
Relocations.end(), increment it, and try to do so until we reach the second
valid iterator. Of course we reach segfault earlier. In this patch we stop BOLT
from searching relocations for symbols outside of the current section.

Reviewed By: rafauler

Differential Revision: https://reviews.llvm.org/D146620

19 months ago[SystemZ] Allow fp/int casting with inline assembly operands.
Jonas Paulsson [Tue, 14 Mar 2023 15:40:10 +0000 (16:40 +0100)]
[SystemZ] Allow fp/int casting with inline assembly operands.

Support bitcasting between int/fp/vector values and 'r'/'f'/'v' inline
assembly operands. This is intended to match GCCs beahvior.

Reviewed By: Ulrich Weigand

Differential Revision: https://reviews.llvm.org/D146059

19 months agoRevert "[Serialization] Place command line defines in the correct file"
Felipe de Azevedo Piovezan [Fri, 24 Mar 2023 18:49:47 +0000 (14:49 -0400)]
Revert "[Serialization] Place command line defines in the correct file"

This reverts commit 72073fc95cd4793a853925ddc8cc3fb2118808a5.

19 months ago[support] Temporarily disable Inf/NaN testing in PrintNumber
Paul Kirth [Fri, 24 Mar 2023 17:41:10 +0000 (17:41 +0000)]
[support] Temporarily disable Inf/NaN testing in PrintNumber

This is still breaking on some platforms. The underlying implementation
doesn't seem to be the cause, rather the test is not robust across
platforms. So, we'll just disable this for the time being, to unblock
builds until we have a proper fix.

Reviewed By: abhina.sreeskantharajan

Differential Revision: https://reviews.llvm.org/D146834

19 months agoAdd support for critical regions in device code.
Doru Bercea [Fri, 10 Mar 2023 23:03:51 +0000 (18:03 -0500)]
Add support for critical regions in device code.

Review: https://reviews.llvm.org/D145831

19 months ago[TSan] Refactor ExternalAccess() to avoid unnecessary pop/push tag [NFC]
Julian Lettner [Wed, 22 Mar 2023 22:07:37 +0000 (15:07 -0700)]
[TSan] Refactor ExternalAccess() to avoid unnecessary pop/push tag [NFC]

* Avoid unnecessary frame & tag push/pops if memory access is ignored
* Rename function and add comment to make it clearer what the code does
* Make helper functions static and move inside `#if !SANITIZER_GO`

Differential Revision: https://reviews.llvm.org/D146670

19 months ago[ARM] Add Thumb Attributes for thumb thunks created in SLSHarding
David Green [Fri, 24 Mar 2023 18:11:54 +0000 (18:11 +0000)]
[ARM] Add Thumb Attributes for thumb thunks created in SLSHarding

Without this the function will be use an Arm subtarget, meaning the
instructions in it will be invalid for the current subtarget.

Differential Revision: https://reviews.llvm.org/D144733

19 months ago[M68k] Fix CConvs for pointer type return values
Min-Yih Hsu [Fri, 24 Mar 2023 17:59:46 +0000 (10:59 -0700)]
[M68k] Fix CConvs for pointer type return values

Put the value into A0 instead of data registers. And remove the
redundant `RetCC_M68kCommon` as there aren't many rules shared between
existing CCs other than the pointer one.
This change is tested by existing tests.

19 months ago[lldb] Explicitly mark PlatformFreeBSD's dependency on PlatformPOSIX
Alex Langford [Fri, 24 Mar 2023 17:55:33 +0000 (10:55 -0700)]
[lldb] Explicitly mark PlatformFreeBSD's dependency on PlatformPOSIX

I accidentally broke the FreeBSD lldb-server build in 0c5cee7 because it
now depends on PlatformFreeBSD. PlatformFreeBSD depends on PlatformPOSIX
but this dependency was not explicitly tracked in CMake. As a result,
the FreeBSD lldb-server build broke.

Credit to John F. Carr <jfc@mit.edu> for pointing out the issue and
providing a fix.

19 months agoFix backtick handling in parsed commands.
Jim Ingham [Fri, 24 Mar 2023 17:38:56 +0000 (10:38 -0700)]
Fix backtick handling in parsed commands.

https://reviews.llvm.org/D146779

19 months ago[NVPTX] Fix integer overflow affecting array size calculation.
Artem Belevich [Thu, 23 Mar 2023 22:27:02 +0000 (15:27 -0700)]
[NVPTX] Fix integer overflow affecting array size calculation.

It turns out, 4GB+ large arrays are a thing and a 32-bit integer is just not
enough to handle them.

Differential Revision: https://reviews.llvm.org/D146767

19 months ago[RISCV] Model select and insertsubvector shuffle kinds
Luke Lau [Thu, 23 Mar 2023 17:45:30 +0000 (17:45 +0000)]
[RISCV] Model select and insertsubvector shuffle kinds

Selects get lowered to a vmerge with a mask, and insertsubvectors get
lowered to a vslideup.

Differential Revision: https://reviews.llvm.org/D146747

19 months ago[RISCV] Add test case for two equivalent reductions
Luke Lau [Fri, 24 Mar 2023 13:46:03 +0000 (13:46 +0000)]
[RISCV] Add test case for two equivalent reductions

They are functionally equivalent but currently one fails to vectorize
because the cost of an insert subvector shuffle is too expensive.
D146747 will update the cost of these types of shuffles, so add a test
case for it.

19 months ago[RISCV] Enable SLP in RISC-V SLP reduction tests
Luke Lau [Fri, 24 Mar 2023 13:43:32 +0000 (13:43 +0000)]
[RISCV] Enable SLP in RISC-V SLP reduction tests

Horizontal reduction can still kick in even when the max VF is set to 0,
but strange stuff can happen as it affects the cost model.
Enable it for these tests as eventually the goal will be to have SLP
enabled.

19 months ago[RISCV] Add test cases for modeling more shuffle kinds
Luke Lau [Fri, 24 Mar 2023 13:39:51 +0000 (13:39 +0000)]
[RISCV] Add test cases for modeling more shuffle kinds

These map to SK_InsertSubvector and SK_Select shuffle kinds

19 months ago[RISCV][NFC] Rename some test cases
Luke Lau [Fri, 24 Mar 2023 15:52:02 +0000 (15:52 +0000)]
[RISCV][NFC] Rename some test cases

Since they no longer involve a merge

19 months ago[RISCV] Lower insert subvector shuffles as vslideups
Luke Lau [Fri, 24 Mar 2023 12:03:41 +0000 (12:03 +0000)]
[RISCV] Lower insert subvector shuffles as vslideups

A shuffle with an insert subvector mask is functionally equivalent to:
(insert_subvector v0, (extract_subvector v1, len), index)
We can emulate by doing a vslideup on v1 into the right index, and
carefully selecting VL so that we don't overwrite any more destination
elements than what we have to.
This avoids the need for a select with a mask.

19 months ago[flang][driver][openmp] Write MLIR for -save-temps
Sergio Afonso [Tue, 14 Mar 2023 17:40:04 +0000 (17:40 +0000)]
[flang][driver][openmp] Write MLIR for -save-temps

This patch adds support for producing MLIR files when using -save-temps on
flang. One MLIR file will be produced before lowering and optimization passes,
containing the operations produced by the PFT-to-MLIR lowering bridge, and
another at the end of the process, just before LLVM IR generation.

This is accomplished by forwarding the -save-temps flag from the driver to the
frontend, and modifying it to output MLIR files accordingly.

Differential Revision: https://reviews.llvm.org/D146075

19 months ago[scudo] Use bytes-in-freelist as a hint of page release
Chia-hung Duan [Fri, 24 Mar 2023 17:06:29 +0000 (17:06 +0000)]
[scudo] Use bytes-in-freelist as a hint of page release

Tracking the pushed bytes between to releaseToOSMaybe calls may lead to
a overestimated case that if we do malloc 2KB -> free 2KB -> malloc 2KB
-> free 2KB, we may think we have released 4KB but it only releases 2KB
actually. Switch to use bytes-in-freelist excludes more cases that can't
release the pages

Reviewed By: cferris

Differential Revision: https://reviews.llvm.org/D146400

19 months agoRemove unused header file in VectorTransformOps.h
yijia1212 [Fri, 24 Mar 2023 17:05:07 +0000 (10:05 -0700)]
Remove unused header file in VectorTransformOps.h

Remove unused header file in VectorTransformOps.h

Differential Revision: https://reviews.llvm.org/D146825

19 months agoFix warnings
Kazu Hirata [Fri, 24 Mar 2023 16:52:17 +0000 (09:52 -0700)]
Fix warnings

This patch fixes:

  clang/lib/Driver/ToolChains/OHOS.cpp:410:18: warning: unused
  variable ‘A’ [-Wunused-variable]

  clang-tools-extra/clangd/refactor/tweaks/AddUsing.cpp:293:15:
  warning: unused variable ‘II’ [-Wunused-variable]

  llvm/lib/ProfileData/RawMemProfReader.cpp:364:68: warning: suggest
  parentheses around ‘&&’ within ‘||’ [-Wparentheses]

19 months agoMake test more explicit on failure.
Doru Bercea [Fri, 24 Mar 2023 14:18:27 +0000 (10:18 -0400)]
Make test more explicit on failure.

Patch: https://reviews.llvm.org/D146812

19 months ago[support] Attempt to fix PrintNumber test for Solaris
Paul Kirth [Fri, 24 Mar 2023 16:02:49 +0000 (16:02 +0000)]
[support] Attempt to fix PrintNumber test for Solaris

NaN and Inf are still causing some problems in a formatting test.
This patch makes the checked format string exactly match the internal
JSON format string. If there are still problems, we should disable
testing Inf and NaN values until we can come to a portable solution.

Reviewed By: abhina.sreeskantharajan

Differential Revision: https://reviews.llvm.org/D146818

19 months ago[extract_symbols.py] Export unique symbols
Tomas Matheson [Thu, 23 Mar 2023 15:18:34 +0000 (15:18 +0000)]
[extract_symbols.py] Export unique symbols

As described in a comment on D140637, PluginInlineOrderTest.NoInlineFoo
will fail with:

InlineOrderPlugin.so: undefined symbol: _ZN4llvm13AllAnalysesOnINS_6ModuleEE6SetKeyE

The symbol is unique in libLLVMCore and ends up local in AnalysisTests.
Fix this by exporting all unique symbols found in libraries.

AnalysisTests.symbols change in line count: 12464 -> 12499

Differential Revision: https://reviews.llvm.org/D146731

19 months ago[mlir][linalg] Use affine apply in im2col gather index calculations
Quinn Dawkins [Fri, 24 Mar 2023 15:44:51 +0000 (11:44 -0400)]
[mlir][linalg] Use affine apply in im2col gather index calculations

Differential Revision: https://reviews.llvm.org/D146816

19 months ago[libc] Update bazel build rules for 9c8bdbcbc502fac7d7d8da5c848cec448daf26ae
Augie Fackler [Fri, 24 Mar 2023 15:34:58 +0000 (11:34 -0400)]
[libc] Update bazel build rules for 9c8bdbcbc502fac7d7d8da5c848cec448daf26ae

19 months ago[AMDGPU] Add clang builtin for __builtin_amdgcn_ds_atomic_fadd_v2f16
Mariusz Sikora [Fri, 10 Mar 2023 07:50:16 +0000 (08:50 +0100)]
[AMDGPU] Add clang builtin for __builtin_amdgcn_ds_atomic_fadd_v2f16

Differential Revision: https://reviews.llvm.org/D146808

19 months ago[LICM] Require MSSA in SinkAndHoistLICMFlags (NFC)
Nikita Popov [Fri, 24 Mar 2023 15:08:37 +0000 (16:08 +0100)]
[LICM] Require MSSA in SinkAndHoistLICMFlags (NFC)

Nowadays MSSA is required for LICM/LoopSink, so drop the checks
for whether its available or not.

19 months ago[sanitizer][win] Change cmdline check to allow double backslashs
Alvin Wong [Wed, 22 Mar 2023 10:14:40 +0000 (18:14 +0800)]
[sanitizer][win] Change cmdline check to allow double backslashs

When `llvm-symbolizer.exe` is on the PATH in an entry containing two
consecutive backslashes, sanitizers will try to launch llvm-symbolizer
with its absolute path containing these consecutive backslashes. This
fails a sanity check in `sanitizer_symbolizer_win.cpp`.

According to the documentation of `CommandLineToArgvW` [1] and a MS blog
post [2], backslashes in general, regardless of how many of them in a
row, do not have any special effect, unless when immediately followed by
a double quote.

There already exists a check that fails when the command line arguments
contains double quote, therefore the check for double backslashes can
simply be removed.

[1]: https://learn.microsoft.com/en-us/windows/win32/api/shellapi/nf-shellapi-commandlinetoargvw
[2]: https://learn.microsoft.com/en-us/archive/blogs/twistylittlepassagesallalike/everyone-quotes-command-line-arguments-the-wrong-way

Differential Revision: https://reviews.llvm.org/D146621

19 months ago[LoongArch] Add Scalar link component
Xiaodong Liu [Fri, 24 Mar 2023 14:57:22 +0000 (09:57 -0500)]
[LoongArch] Add Scalar link component

Fix the lld link error on ppc64le builders:
```
ld.lld: error: undefined symbol: llvm::createLoopDataPrefetchPass()
>>> referenced by LoongArchTargetMachine.cpp
>>>               lib/Target/LoongArch/CMakeFiles/LLVMLoongArchCodeGen.dir/LoongArchTargetMachine.cpp.o:((anonymous namespace)::LoongArchPassConfig::addIRPasses())
collect2: error: ld returned 1 exit status`
```

Reviewed By: sdesmalen

Differential Revision: https://reviews.llvm.org/D146806

19 months ago[SelectionDAG] Correctly reduce BV to shuffle with zero on big endian
Nemanja Ivanovic [Fri, 24 Mar 2023 14:34:25 +0000 (10:34 -0400)]
[SelectionDAG] Correctly reduce BV to shuffle with zero on big endian

This DAG combine is correct on little endian targets but
is incorrect on big endian targets.
Add big endian code to correct it.

Differential revision: https://reviews.llvm.org/D146460

19 months ago[libc] Change RPC outbox stores to be relaxed
Joseph Huber [Fri, 24 Mar 2023 14:37:28 +0000 (09:37 -0500)]
[libc] Change RPC outbox stores to be relaxed

Summary:
These stored previously used `RELEASE`. This was done originally to
ensure that the stores to the shared memory buffer were flushed prior to
signaling that the other side can begin accessing it. However, this
should be accomplished by the memory fence above the store. This change
is required because NVPTX does not support non-relaxed atomics used on
unified shared memory.

19 months ago[AArch64][SME] SelectSMETileSlice should also match to 'reg+0' when slice is ADD...
Sander de Smalen [Wed, 22 Mar 2023 16:54:00 +0000 (16:54 +0000)]
[AArch64][SME] SelectSMETileSlice should also match to 'reg+0' when slice is ADD with non-constant RHS.

It would decompose an address into a `reg + 0` when the slice was not an ADD,
but when the RHS of the ADD was not a constant, it would simply not match.

This patch fixes that, by always resolving to a `reg + 0` slice.

19 months ago[clang] Extend pragma dump to support expressions
Vlad Serebrennikov [Fri, 24 Mar 2023 14:14:11 +0000 (17:14 +0300)]
[clang] Extend pragma dump to support expressions

Extend `#pragma clang __debug dump` to support not only single identifier, but an expression as well. This makes it possible to test ADL and overload resolution directly, without being creative to make them observable via diagnostics (e.g. when [[ http://eel.is/c++draft/over.match.best | over.match.best ]] is involved). This implementation has a known limitation of not supporting dependent expressions properly, but it's quite useful even without such support.

Differential Revision: https://reviews.llvm.org/D144115

19 months ago[LICM] Add tests for GEP reassociation (NFC)
Nikita Popov [Fri, 24 Mar 2023 14:24:21 +0000 (15:24 +0100)]
[LICM] Add tests for GEP reassociation (NFC)

19 months ago[Test] Regenerate test checks for some LSR tests (NFC)
Dmitry Makogon [Fri, 24 Mar 2023 13:34:55 +0000 (20:34 +0700)]
[Test] Regenerate test checks for some LSR tests (NFC)

19 months ago[Test] Add test to check that LCSSA is preserved by LSR (NFC)
Dmitry Makogon [Fri, 24 Mar 2023 12:02:25 +0000 (19:02 +0700)]
[Test] Add test to check that LCSSA is preserved by LSR (NFC)

Currently it fails as LSR doesn't preserve LCSSA in some cases.

19 months ago[mlir][Vector][Transforms] Improve the control over individual vector lowerings and...
Nicolas Vasilache [Fri, 24 Mar 2023 12:09:50 +0000 (12:09 +0000)]
[mlir][Vector][Transforms] Improve the control over individual vector lowerings and transforms

This revision adds vector transform operations that allow us to better inspect the composition
of various lowerings that were previously very opaque.

This commit is NFC in that it does not change patterns beyond adding `rewriter.notifyFailure` messages
and it does not change the tests beyond breaking them into pieces and using transforms instead of
throwaway opaque test passes.

Reviewed By: ftynse, springerm

Co-authored-by: Alex Zinenko <zinenko@google.com>
Differential Revision: https://reviews.llvm.org/D146755

19 months ago[SimplifyCFG] Don't merge invoke if this makes immarg non-constant (PR61265)
Nikita Popov [Thu, 23 Mar 2023 14:23:31 +0000 (15:23 +0100)]
[SimplifyCFG] Don't merge invoke if this makes immarg non-constant (PR61265)

Don't merge invokes if this replaces constant operands with phis
in a place where this is not legal.

This also disallows converting operand bundles from constant to
non-constant, in line with the restriction we use in other
transforms.

Fixes https://github.com/llvm/llvm-project/issues/61265.

Differential Revision: https://reviews.llvm.org/D146723

19 months ago[flang][Driver] Fix lto-flags.f90 on Solaris
Rainer Orth [Fri, 24 Mar 2023 13:24:07 +0000 (14:24 +0100)]
[flang][Driver] Fix lto-flags.f90 on Solaris

The `Flang :: Driver/lto-flags.f90` test `FAIL`s on Solaris:

  /vol/llvm/src/llvm-project/dist/flang/test/Driver/lto-flags.f90:30:13: error: THIN-LTO: expected string not found in input
  ! THIN-LTO: "-plugin-opt=thinlto"
              ^

This is no wonder since the native Solaris `ld` doesn't support the linker
plugin interface at all, so this patch marks the test as `UNSUPPORTED`.

Tested on `amd64-pc-solaris2.11` and `x86_64-pc-linux-gnu`.

Differential Revision: https://reviews.llvm.org/D146807

19 months ago[GlobalOpt] Fix dead const handling in pointer root user cleanup (PR61674)
Nikita Popov [Fri, 24 Mar 2023 13:17:26 +0000 (14:17 +0100)]
[GlobalOpt] Fix dead const handling in pointer root user cleanup (PR61674)

Rather than cleanup up dead constant expressions as we go along,
do this once at the end. This aligns it with the
CleanupConstantGlobalUsers() implementation and avoids
any invalidation issues.

Fixes https://github.com/llvm/llvm-project/issues/61674.

19 months ago[AggressiveInstCombine] use m_Deferred on funnel shift(NFC)
khei4 [Fri, 24 Mar 2023 10:21:25 +0000 (19:21 +0900)]
[AggressiveInstCombine] use m_Deferred on funnel shift(NFC)
Differential Revision: https://reviews.llvm.org/D146798
Reviewed By: nikic

19 months ago[AMDGPU] Create Subtarget Features for some of 16 bits atomic fadd instructions
Mariusz Sikora [Thu, 9 Mar 2023 10:29:33 +0000 (11:29 +0100)]
[AMDGPU] Create Subtarget Features for some of 16 bits atomic fadd instructions

Introducing Subtarget Features for instructions:
- ds_pk_add_bf16
- ds_pk_add_f16
- ds_pk_add_rtn_bf16
- ds_pk_add_rtn_f16
- flat_atomic_pk_add_f16
- flat_atomic_pk_add_bf16
- global_atomic_pk_add_f16
- global_atomic_pk_add_bf16
- buffer_atomic_pk_add_f16

Differential Revision: https://reviews.llvm.org/D146701

19 months ago[gn] somewhat port 0c5cee779929 (lldb-server platform plugin)
Nico Weber [Fri, 24 Mar 2023 12:06:37 +0000 (13:06 +0100)]
[gn] somewhat port 0c5cee779929 (lldb-server platform plugin)

19 months ago[Local] Check for null VH in RecursivelyDeleteTriviallyDeadInstructionsPermissive()
Nikita Popov [Fri, 24 Mar 2023 11:55:02 +0000 (12:55 +0100)]
[Local] Check for null VH in RecursivelyDeleteTriviallyDeadInstructionsPermissive()

Peculiarly, the non-permissive variant handled this gracefully,
but the permissive one did not.

19 months ago[Test] Use autogenerated checks in uglygep.ll test for LSR (NFC)
Dmitry Makogon [Fri, 24 Mar 2023 11:22:47 +0000 (18:22 +0700)]
[Test] Use autogenerated checks in uglygep.ll test for LSR (NFC)

19 months agoFix Clang sphinx build
Aaron Ballman [Fri, 24 Mar 2023 11:27:07 +0000 (07:27 -0400)]
Fix Clang sphinx build

This addresses the issues found in:
https://lab.llvm.org/buildbot/#/builders/92/builds/41772

19 months ago[Flang][OpenMP] Add TODO message for common block privatisation
Kiran Chandramohan [Fri, 24 Mar 2023 11:02:18 +0000 (11:02 +0000)]
[Flang][OpenMP] Add TODO message for common block privatisation

This is a temporary message until the feature is implemented and
merged.

Note: There is a proposed patch (https://reviews.llvm.org/D127215)

Reviewed By: peixin

Differential Revision: https://reviews.llvm.org/D146768

19 months ago[JITLink] Fix MSVC build error: fully qualify llvm::object::ELFFile
Stefan Gränitz [Fri, 24 Mar 2023 11:01:52 +0000 (12:01 +0100)]
[JITLink] Fix MSVC build error: fully qualify llvm::object::ELFFile

19 months ago[OpenCL] Emit EOL at end of generated header
Sven van Haastregt [Fri, 24 Mar 2023 11:03:10 +0000 (11:03 +0000)]
[OpenCL] Emit EOL at end of generated header

19 months ago[include-cleaner] Attribute references to explicit specializations
Kadir Cetinkaya [Thu, 23 Mar 2023 16:36:54 +0000 (17:36 +0100)]
[include-cleaner] Attribute references to explicit specializations

Fixes https://github.com/llvm/llvm-project/issues/61652

Differential Revision: https://reviews.llvm.org/D146732

19 months ago[compiler-rt] Fix signed integer overflow in int_mulo_impl.inc
Karl-Johan Karlsson [Fri, 24 Mar 2023 09:33:46 +0000 (10:33 +0100)]
[compiler-rt] Fix signed integer overflow in int_mulo_impl.inc

When compiling compiler-rt with -fsanitize=undefined and running testcases you
end up with the following warning:

UBSan:/repo/uabkaka/llvm-project/compiler-rt/lib/builtins/int_mulo_impl.inc:24:23: signed integer overflow: -1 * -2147483648 cannot be represented in type 'si_int' (aka 'long')

This can be avoided by doing the multiplication in a matching unsigned variant
of the type.

This was found in an out of tree target.

Reviewed By: phosek

Differential Revision: https://reviews.llvm.org/D146623

19 months ago[AVR] Do not emit 'LPM Rd, Z' on devices without FeatureLPMX
Ben Shi [Mon, 9 Jan 2023 04:05:00 +0000 (12:05 +0800)]
[AVR] Do not emit 'LPM Rd, Z' on devices without FeatureLPMX

The 'LPM' instruction has three forms:

------------------------
| form       | feature |
| ---------- | --------|
| LPM        | hasLPM  |
| LPM Rd, Z  | hasLPMX |
| LPM Rd, Z+ | hasLPMX |
------------------------

The second form is always selected in ISelDAGToDAG, even on devices
without FeatureLPMX. This patch emits "LPM + MOV" on devices with
only FeatureLPM.

Reviewed By: jacquesguan

Differential Revision: https://reviews.llvm.org/D141246

19 months ago[NFC][LoopVectorize] Change trip counts for some tests to guarantee a scalar tail
David Sherwood [Thu, 16 Mar 2023 11:39:00 +0000 (11:39 +0000)]
[NFC][LoopVectorize] Change trip counts for some tests to guarantee a scalar tail

Quite a few vectoriser tests were using a trip count of 1024,
which meant:

1. For fixed-length VFs we would never actually tail-fold, e.g.
see Transforms/LoopVectorize/RISCV/uniform-load-store.ll. This
is because we can prove at compile-time there will never be a
scalar tail.
2. As of D146199 the same optimisation mentioned above will also
apply to scalable VFs too.

I've changed all such trip counts to be 1025 instead.

Differential Revision: https://reviews.llvm.org/D146219

19 months ago[NFC] Fix auto usage to avoid copies
Akshay Khadse [Fri, 24 Mar 2023 09:14:30 +0000 (17:14 +0800)]
[NFC] Fix auto usage to avoid copies

Fixes some usages of the "auto" keyword to avoid creation of copies.

Reviewed By: LuoYuanke

Differential Revision: https://reviews.llvm.org/D146694

19 months ago[clang-format] NFC Format.h and ClangFormatStyleOptions.rst are out of date
mydeveloperday [Fri, 24 Mar 2023 09:27:23 +0000 (09:27 +0000)]
[clang-format] NFC Format.h and ClangFormatStyleOptions.rst are out of date

Regenerate the style documentation, requires some minor sphinx changes to avoid warnings

Reviewed By: klimek

Differential Revision: https://reviews.llvm.org/D146704

19 months agoReland "[JITLink] Initial AArch32 backend"
Stefan Gränitz [Thu, 23 Mar 2023 10:10:39 +0000 (11:10 +0100)]
Reland "[JITLink] Initial AArch32 backend"

This first version lays the foundations for AArch32 support in JITLink. ELFLinkGraphBuilder_aarch32 processes REL-type relocations and populates LinkGraphs from ELF object files for both big- and little-endian systems. The ArmCfg member controls subarchitecture-specific details throughout the linking process (i.e. it's passed to ELFJITLinker_aarch32).

Relocation types follow the ABI documentation's division into classes: Data (endian-sensitive), Arm (32-bit little-endian) and Thumb (2x 16-bit little-endian, "Thumb32" in the docs). The implementation of instruction encoding/decoding for relocation resolution is implemented symmetrically and is testable in isolation (see AArch32 category in JITLinkTests).

Callable Thumb functions are marked with a ThumbSymbol target-flag and stored in the LinkGraph with their real addresses. The thumb-bit is added back in when the owning JITDylib requests the address for such a symbol.

The StubsManager can generate (absolute) Thumb-state stubs for branch range extensions on v7+ targets. Proper GOT/PLT handling is not yet implemented.

This patch is based on the backend implementation in ez-clang and has just enough functionality to model the infrastructure and link a Thumb function `main()` that calls `printf()` to dump "Hello Arm!" on Armv7a. It was tested on Raspberry Pi with 32-bit Raspbian OS.

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D144083

19 months ago[RISCV] Add test for shuffles that could be done as vmerges
Luke Lau [Thu, 23 Mar 2023 10:59:59 +0000 (10:59 +0000)]
[RISCV] Add test for shuffles that could be done as vmerges

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D146710

19 months ago[mlir][arith] Extend the `floordivsi` converter
Andrzej Warzynski [Thu, 23 Mar 2023 17:39:48 +0000 (17:39 +0000)]
[mlir][arith] Extend the `floordivsi` converter

This patch extends the `createConst` method so that it can generate
constant vectors (it can already generate scalars). This change is
required to be able to apply the converter for `arith.floordivsi`
(i.e. `FloorDivSIOpConverter`) to vectors.

While `arith.floordivsi` is my main motivation for this change, this
patch should also allow other Arith ops to be converted in vector cases.
In my example, the Linalg vectorizer updates `arith.floordivsi` to
operate on vectors and hence the need for this change.

Differential Revision: https://reviews.llvm.org/D146741

19 months ago[NFC] Regenerate test of InstCombine/load-combine-metadata-dominance.ll
luxufan [Fri, 24 Mar 2023 08:35:56 +0000 (16:35 +0800)]
[NFC] Regenerate test of InstCombine/load-combine-metadata-dominance.ll

19 months ago[Test] Regenerate checks in test file
Max Kazantsev [Fri, 24 Mar 2023 08:42:23 +0000 (15:42 +0700)]
[Test] Regenerate checks in test file

19 months ago[CMake] Respect variables for specifying host tools even without LLVM_USE_HOST_TOOLS set
Martin Storsjö [Wed, 22 Mar 2023 22:44:18 +0000 (00:44 +0200)]
[CMake] Respect variables for specifying host tools even without LLVM_USE_HOST_TOOLS set

When LLVM_NATIVE_TOOL_DIR was introduced in
d3da9067d143f3d4ce59b6d9ab4606a8ef1dc937 / D131052, it consisted
of refactoring a couple cases of manual logic for tools in
clang-tools-extra/clang-tidy, clang-tools-extra/pseudo/include
and mlir/tools/mlir-linalg-ods-gen. The former two had the same
consistent behaviour while the latter was slightly different, so
the refactoring would end up slightly adjusting one or the other.

The difference was that the clang-tools-extra tools respected the
external variable for setting the tool name, regardless of the
LLVM_USE_HOST_TOOLS variable, while mlir-linalg-ods-gen tool
only checked its external variable if LLVM_USE_HOST_TOOLS was set.

LLVM_USE_HOST_TOOLS is supposed to be enabled automatically whenever
cross compiling, so this shouldn't have been an issue.

In https://github.com/llvm/llvm-project/issues/60784, it seems like
some users do cross compile LLVM, without CMake knowing about it
(without CMAKE_CROSSCOMPILING being set). In these cases, their
build broke, as the variables for pointing to external host tools
no longer were being respected.

The fact that CMAKE_CROSSCOMPILING wasn't set stems from a
non-obvious behaviour of CMake; CMAKE_CROSSCOMPILING isn't supposed
to be set by the user (and if it was, it gets overridden), but one
has to set CMAKE_SYSTEM_NAME to indicate that one is cross compiling,
even if the target OS is the same as the current host.

Skip the checks for LLVM_USE_HOST_TOOLS and always respect the
variables for pointing to external tools (both the old tool specific
variables, and the new LLVM_NATIVE_TOOL_DIR), if they're set. This
makes the logic within setup_host_tool more exactly match the
logic for the clang-tools-extra tools from before the refactoring
in d3da9067d143f3d4ce59b6d9ab4606a8ef1dc937. This makes the behaviour
consistent with that of the tablegen executables, which also respect
the externally set variables regardless of LLVM_USE_HOST_TOOLS.

This fixes
https://github.com/llvm/llvm-project/issues/60784.

Differential Revision: https://reviews.llvm.org/D146666

19 months ago[MLIR][LLVM] Add debug output to the LLVM inliner.
Johannes de Fine Licht [Fri, 24 Mar 2023 08:27:27 +0000 (09:27 +0100)]
[MLIR][LLVM] Add debug output to the LLVM inliner.

This revealed a test case that wasn't hitting the intended branch
because the inlinees had no function definition.

Depends on D146628

Reviewed By: gysit

Differential Revision: https://reviews.llvm.org/D146633

19 months ago[LLD] [test] Add a missing REQUIRES: x86 in a new test
Martin Storsjö [Fri, 24 Mar 2023 08:23:50 +0000 (10:23 +0200)]
[LLD] [test] Add a missing REQUIRES: x86 in a new test

This test somewhat unconventionally assembles both aarch64 and
x86 object files.

This fixes test failures in build configurations with the aarch64
target enabled but x86 target disabled.

19 months ago[NFC] Regenerate test NewGVN/metadata-nonnull.ll
luxufan [Fri, 24 Mar 2023 08:26:32 +0000 (16:26 +0800)]
[NFC] Regenerate test NewGVN/metadata-nonnull.ll

19 months agoRevert "Recommit [Modules] Remove unnecessary check when generating name lookup table...
Dmitry Chernenkov [Thu, 23 Mar 2023 16:20:42 +0000 (16:20 +0000)]
Revert "Recommit [Modules] Remove unnecessary check when generating name lookup table in ASTWriter"

This reverts commit 25557aa38a0dab76f5b7a4518942f69d879693c0.

19 months ago[mlir][llvm] Switch remaining LLVM dialect tests to opaque pointers.
Tobias Gysi [Fri, 24 Mar 2023 06:57:24 +0000 (07:57 +0100)]
[mlir][llvm] Switch remaining LLVM dialect tests to opaque pointers.

The revision switches the remaining LLVM dialect tests to use opaque
pointers. Selected tests are copied to a postfixed test file for the
time being.

A number of tests disappear once we fully switch to opaque pointers.
In particular, all tests that check verify a pointer element type
matches another type as well as tests of recursive types.

Part of https://discourse.llvm.org/t/rfc-switching-the-llvm-dialect-and-dialect-lowerings-to-opaque-pointers/68179

Reviewed By: Dinistro, zero9178

Differential Revision: https://reviews.llvm.org/D146726

19 months ago[clang-tidy][NFC] Improve naming convention in google-readability-avoid-underscore...
Carlos Galvez [Thu, 23 Mar 2023 12:16:40 +0000 (12:16 +0000)]
[clang-tidy][NFC] Improve naming convention in google-readability-avoid-underscore-in-googletest-name

According to the Google docs, the convention is
TEST(TestSuiteName, TestName). Apply that convention to the
source code, test and documentation of the check.

Differential Revision: https://reviews.llvm.org/D146713

19 months ago[Driver] Change multilib selection algorithm
Michael Platings [Tue, 14 Mar 2023 19:40:58 +0000 (19:40 +0000)]
[Driver] Change multilib selection algorithm

The new algorithm is:
1. Find all multilibs with flags that are a subset of the requested
   flags.
2. If more than one multilib matches, choose the last.

In addition a new selection mechanism is permitted via an overload of
MultilibSet::select() for which multiple multilibs are returned.
This allows layering multilibs on top of each other.

Since multilibs are now ordered within a list, they no longer need a
Priority field.

The new algorithm is different to the old algorithm, but in practise
the old algorithm was always used in such a way that the effect is the
same.
The old algorithm was to find the set intersection of the requested
flags (with the first character of each removed) with each multilib's
flags (ditto), and for that intersection check whether the first
character matched. However, ignoring the first characters, the
requested flags were always a superset of all the multilibs flags.
Therefore the new algorithm can be used as a drop-in replacement.

The exception is Fuchsia, which needs adjusting slightly to set both
fexceptions and fno-exceptions flags.

Differential Revision: https://reviews.llvm.org/D142905

19 months ago[X86] Precommit a test
Kazu Hirata [Fri, 24 Mar 2023 06:48:17 +0000 (23:48 -0700)]
[X86] Precommit a test

This patch precommits a test for:

https://github.com/llvm/llvm-project/issues/61365

19 months ago[lldb] Fix type of --apply-fixits (NFC)
Dave Lee [Fri, 24 Mar 2023 05:14:10 +0000 (22:14 -0700)]
[lldb] Fix type of --apply-fixits (NFC)

19 months ago[BugFix] Fix VSELECT ISel fail
Xiang1 Zhang [Tue, 21 Mar 2023 09:33:54 +0000 (17:33 +0800)]
[BugFix] Fix VSELECT ISel fail

Reviewed By: Luo yuanke

Differential Revision: https://reviews.llvm.org/D146683

19 months ago[AArch64] Add tests for umax(x, 1u)
Kazu Hirata [Fri, 24 Mar 2023 03:20:20 +0000 (20:20 -0700)]
[AArch64] Add tests for umax(x, 1u)

This patch adds tests for umax(x, 1u).

This patch fixes:

https://github.com/llvm/llvm-project/issues/60233

It turns out that commit 86b4d8645fc1b86693fef564cef68f24599c930f on
Feb 8, 2023 already performs the instcombine transformation proposed
in the issue, so the issue requires no change on the codegen side.

19 months ago[LoongArch] Enable LoopDataPrefetch pass
Xiaodong Liu [Fri, 24 Mar 2023 03:08:21 +0000 (11:08 +0800)]
[LoongArch] Enable LoopDataPrefetch pass

Keep `EnableLoopDataPrefetch` option off for now because
we need a few more TTIs and ISels.

This patch is inspired by http://reviews.llvm.org/D17943.

Reviewed By: SixWeining

Differential Revision: https://reviews.llvm.org/D146600

19 months ago[InstCombine] Try to recognize bswap pattern when calling funnel shifts
Jun Zhang [Fri, 24 Mar 2023 02:28:02 +0000 (10:28 +0800)]
[InstCombine] Try to recognize bswap pattern when calling funnel shifts

Alive2: https://alive2.llvm.org/ce/z/dxxD7B
Fixes: https://github.com/llvm/llvm-project/issues/60690

Signed-off-by: Jun Zhang <jun@junz.org>
Differential Revision: https://reviews.llvm.org/D146637

19 months agoPrecommit tests for #60690
Jun Zhang [Fri, 24 Mar 2023 02:27:02 +0000 (10:27 +0800)]
Precommit tests for #60690

Differential Revision: https://reviews.llvm.org/D146636

Signed-off-by: Jun Zhang <jun@junz.org>
19 months ago[NFC][X86]remove trailing space in X86InstrArithmetic.td
XinWang10 [Fri, 24 Mar 2023 02:32:18 +0000 (22:32 -0400)]
[NFC][X86]remove trailing space in X86InstrArithmetic.td

In this file, most of the line don't have trailing spaces,
but some of them have. To keep consistent, remove the trailing
spaces.

Reviewed By: skan

Differential Revision: https://reviews.llvm.org/D146697

19 months ago[InstCombine] Generate better code for std::bit_ceil
Kazu Hirata [Fri, 24 Mar 2023 02:26:43 +0000 (19:26 -0700)]
[InstCombine] Generate better code for std::bit_ceil

Without this patch, std::bit_ceil<uint32_t> is compiled as:

  %dec = add i32 %x, -1
  %lz = tail call i32 @llvm.ctlz.i32(i32 %dec, i1 false)
  %sub = sub i32 32, %lz
  %res = shl i32 1, %sub
  %ugt = icmp ugt i32 %x, 1
  %sel = select i1 %ugt, i32 %res, i32 1

With this patch, we generate:

  %dec = add i32 %x, -1
  %ctlz = tail call i32 @llvm.ctlz.i32(i32 %dec, i1 false)
  %sub = sub nsw i32 0, %ctlz
  %and = and i32 %1, 31
  %sel = shl nuw i32 1, %and
  ret i32 %sel

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

This patch recognizes the specific pattern from std::bit_ceil in
libc++ and libstdc++ and drops the conditional move.  In addition to
the LLVM IR generated for std::bit_ceil(X), this patch recognizes
variants like:

  std::bit_ceil(X - 1)
  std::bit_ceil(X + 1)
  std::bit_ceil(X + 2)
  std::bit_ceil(-X)
  std::bit_ceil(~X)

This patch fixes:

https://github.com/llvm/llvm-project/issues/60802

Differential Revision: https://reviews.llvm.org/D145299

19 months ago[SelectionDAG] Use isOneConstant (NFC)
Kazu Hirata [Fri, 24 Mar 2023 02:26:42 +0000 (19:26 -0700)]
[SelectionDAG] Use isOneConstant (NFC)

19 months ago[NFC] Fix Windows builds that use MSVC 14.x
Rahul Joshi [Thu, 23 Mar 2023 23:06:37 +0000 (16:06 -0700)]
[NFC] Fix Windows builds that use MSVC 14.x

Differential Revision: https://reviews.llvm.org/D146769

19 months ago[builtins][test] Fix divmodti4_test.c on Windows
Arthur Eubanks [Fri, 24 Mar 2023 02:08:53 +0000 (19:08 -0700)]
[builtins][test] Fix divmodti4_test.c on Windows

By making the 64 bit integer literals unsigned. Otherwise some of them
are unexpectedly sign extended (and the compiler rightly diagnosed this
with warnings)

Initially added in D80506.

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D146667

19 months ago[LegalizeTypes][RISCV] Add a special case for (add X, -1) to ExpandIntRes_ADDSUB
LiaoChunyu [Fri, 24 Mar 2023 01:04:59 +0000 (09:04 +0800)]
[LegalizeTypes][RISCV] Add a special case for (add X, -1) to ExpandIntRes_ADDSUB

 On targets without ADDCARRY or ADDE, we need to emit a separate
 SETCC to determine carry from the low half to the high half.
 The high half is calculated by a series of ADDs.

 When RHSLo and RHSHi are -1, without this patch, we get:
   Hi = (add (add LHSHi,(setult Lo, LHSLo), -1)
 Where as with the patch we get:
   Hi = (sub LHSHi, (seteq LHSLo, 0))

 Only RHSLo is -1 we can instead do (setne Lo, 0).

 Similar to gcc: https://godbolt.org/z/M83f6rz39

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D146635

19 months ago[mlir][linalg] Refactor convolution to img2col conversion to use gather semantics
Quinn Dawkins [Thu, 23 Feb 2023 16:30:20 +0000 (11:30 -0500)]
[mlir][linalg] Refactor convolution to img2col conversion to use gather semantics

Following up on the comments in https://reviews.llvm.org/D144108 this
patch refactors the im2col conversion patterns for `linalg.conv_2d_nhwc_hwcf`
and `linalg.conv_2d_nchw_fchw` convolutions to use gather semantics for the im2col
packing `linalg.generic`.

Follow up work can include a similar pattern for depthwise convolutions
and a generalization of the patterns here to work with any `LinalgOp` as
well.

Differential Revision: https://reviews.llvm.org/D144678

19 months ago[Builtins] Add __builtin_assume_separate_storage
Bruno Cardoso Lopes [Thu, 23 Mar 2023 21:34:12 +0000 (14:34 -0700)]
[Builtins] Add __builtin_assume_separate_storage

Plumbing from the language level to the assume intrinsics with
separate_storage operand bundles.

Patch by David Goldblatt (davidtgoldblatt)

Differential Revision: https://reviews.llvm.org/D136515

19 months ago[Driver] Remove remnant mips64el-linux-android code after D146565
Fangrui Song [Thu, 23 Mar 2023 23:00:16 +0000 (16:00 -0700)]
[Driver] Remove remnant mips64el-linux-android code after D146565

19 months agoAndroid.rules: remove mips* rules
Fangrui Song [Thu, 23 Mar 2023 22:58:42 +0000 (15:58 -0700)]
Android.rules: remove mips* rules

They have been obsoleted for a long time and D146565 recently removed
Clang support.

19 months ago[Driver][test] Remove remnant mips*-linux-android tests after 805f51f9fedf90d2aa0ad46...
Fangrui Song [Thu, 23 Mar 2023 22:49:38 +0000 (15:49 -0700)]
[Driver][test] Remove remnant mips*-linux-android tests after 805f51f9fedf90d2aa0ad46c61cb4c9c0c5bcfe9

19 months ago[WebAssembly] Tidy up DebugValueManager (NFC)
Heejin Ahn [Thu, 23 Mar 2023 02:14:00 +0000 (19:14 -0700)]
[WebAssembly] Tidy up DebugValueManager  (NFC)

Misc. cleanups for `WebAssemblyDebugValueManager`.
- Use `Register` for registers
- Simpler for loop iteration
- Rename a variable
- Reorder methods
- Reduce `SmallVector` size for `DBG_VALUE`s to 1; one def usually have
  a single `DBG_VALUE` attached to it in most cases
- Add a few more lines of comments

Reviewed By: dschuff

Differential Revision: https://reviews.llvm.org/D146743

19 months ago[clang][ExtractAPI]Fix Declaration fragments for instancetype in the type position...
NagaChaitanya Vellanki [Thu, 23 Mar 2023 21:38:37 +0000 (14:38 -0700)]
[clang][ExtractAPI]Fix Declaration fragments for instancetype in the type position degrade to id

Fixes https://github.com/llvm/llvm-project/issues/61481

Reviewed By: dang

Differential Revision: https://reviews.llvm.org/D146671

19 months ago[llvm][TextAPI] Handle implicitly upgraded deployment versions
Cyndy Ishida [Thu, 23 Mar 2023 21:51:37 +0000 (14:51 -0700)]
[llvm][TextAPI] Handle implicitly upgraded deployment versions

Sometimes the clang driver will receive a target triple where the
deployment version is too low to support the platform + arch. In those
cases, the compiler upgrades the final minOS which is what gets recorded
ultimately by the linker in LC_BUILD_VERSION. TextAPI should also reuse
this logic for capturing minOS in recorded TBDv5 files.

Reviewed By: ributzka

Differential Revision: https://reviews.llvm.org/D145690

19 months ago[JITLink] Add a jitlink::Section::empty operation.
Lang Hames [Thu, 23 Mar 2023 21:43:23 +0000 (14:43 -0700)]
[JITLink] Add a jitlink::Section::empty operation.

19 months agoRevert "[scudo] Add a Timer class to assist performance measurement"
Chia-hung Duan [Thu, 23 Mar 2023 21:49:02 +0000 (21:49 +0000)]
Revert "[scudo] Add a Timer class to assist performance measurement"

This reverts commit e0361396c2281a108a36d186161ace1843925431.

19 months ago[llvm] Handle duplicate call bases when applying branch funneling
Leonard Chan [Thu, 23 Mar 2023 21:44:59 +0000 (21:44 +0000)]
[llvm] Handle duplicate call bases when applying branch funneling

It's possible to segfault in `DevirtModule::applyICallBranchFunnel` when
attempting to call `getCaller` on a call base that was erased in a prior
iteration. This can occur when attempting to find devirtualizable calls
via `findDevirtualizableCallsForTypeTest` if the vtable passed to
llvm.type.test is a global and not a local. The function works by taking
the first argument of the llvm.type.test call (which is a vtable),
iterating through all uses of it, and adding any relevant all uses that
are calls associated with that intrinsic call to a vector. For most
cases where the vtable is actually a *local*, this wouldn't be an issue.
Take for example:

```
define i32 @fn(ptr %obj) #0 {
  %vtable = load ptr, ptr %obj
  %p = call i1 @llvm.type.test(ptr %vtable, metadata !"typeid2")
  call void @llvm.assume(i1 %p)
  %fptr = load ptr, ptr %vtable
  %result = call i32 %fptr(ptr %obj, i32 1)
  ret i32 %result
}
```

`findDevirtualizableCallsForTypeTest` will check the call base ` %result
= call i32 %fptr(ptr %obj, i32 1)`, find that it is associated with a
virtualizable call from `%vtable`, find all loads for `%vtable`, and add
any instances those load results are called into a vector. Now consider
the case where instead `%vtable` was the global itself rather than a
local:

```
define i32 @fn(ptr %obj) #0 {
  %p = call i1 @llvm.type.test(ptr @vtable, metadata !"typeid2")
  call void @llvm.assume(i1 %p)
  %fptr = load ptr, ptr @vtable
  %result = call i32 %fptr(ptr %obj, i32 1)
  ret i32 %result
}
```

`findDevirtualizableCallsForTypeTest` should work normally and add one
unique call instance to a vector. However, if there are multiple
instances where this same global is used for llvm.type.test, like with:

```
define i32 @fn(ptr %obj) #0 {
  %p = call i1 @llvm.type.test(ptr @vtable, metadata !"typeid2")
  call void @llvm.assume(i1 %p)
  %fptr = load ptr, ptr @vtable
  %result = call i32 %fptr(ptr %obj, i32 1)
  ret i32 %result
}

define i32 @fn2(ptr %obj) #0 {
  %p = call i1 @llvm.type.test(ptr @vtable, metadata !"typeid2")
  call void @llvm.assume(i1 %p)
  %fptr = load ptr, ptr @vtable
  %result = call i32 %fptr(ptr %obj, i32 1)
  ret i32 %result
}
```

Then each call base `%result = call i32 %fptr(ptr %obj, i32 1)` will be
added to the vector twice. This is because for either call base `%result
= call i32 %fptr(ptr %obj, i32 1) `, we determine it is associated with
a virtualizable call from `@vtable`, and then we iterate through all the
uses of `@vtable`, which is used across multiple functions. So when
scanning the first `%result = call i32 %fptr(ptr %obj, i32 1)`, then
both call bases will be added to the vector, but when scanning the
second one, both call bases are added again, resulting in duplicate call
bases in the CSInfo.CallSites vector.

Note this is actually accounted for in every other instance WPD iterates
over CallSites. What everything else does is actually add the call base
to the `OptimizedCalls` set and just check if it's already in the set.
We can't reuse that particular set since it serves a different purpose
marking which calls where devirtualized which `applyICallBranchFunnel`
explicitly says it doesn't. For this fix, we can just account for
duplicates with a map and do the actual replacements afterwards by
iterating over the map.

Differential Revision: https://reviews.llvm.org/D146267

19 months ago[OpenMP] Fix test after updating NVPTX atomic inlines
Joseph Huber [Thu, 23 Mar 2023 21:41:25 +0000 (16:41 -0500)]
[OpenMP] Fix test after updating NVPTX atomic inlines

Summary:
The previous patch fixed how we handle emitting atomics for targeting
NVPTX directly. This is the only other file that really does that and
has atomics and I forgot to update it.

19 months ago[libc][NFC] Fix misspelled variable name in cmake message
Joseph Huber [Thu, 23 Mar 2023 21:30:31 +0000 (16:30 -0500)]
[libc][NFC] Fix misspelled variable name in cmake message

19 months ago[NVPTX] Set the atomic inling threshold when targeting NVPTX directly
Joseph Huber [Thu, 23 Mar 2023 19:15:01 +0000 (14:15 -0500)]
[NVPTX] Set the atomic inling threshold when targeting NVPTX directly

Since Clang 16.0.0 users can target the `NVPTX` architecture directly
via `--target=nvptx64-nvidia-cuda`. However, this does not set the
atomic inlining size correctly. This leads to spurious warnings and
emission of runtime atomics that are never implemented. This patch
ensures that we set this to the appropriate pointer width. This will
always be 64 in the future as `nvptx64` will only be supported moving
forward.

Fixes: https://github.com/llvm/llvm-project/issues/61410

Reviewed By: tra

Differential Revision: https://reviews.llvm.org/D146750

19 months ago[lld][WebAssembly] Initial support for stub libraries
Sam Clegg [Wed, 7 Dec 2022 00:49:13 +0000 (16:49 -0800)]
[lld][WebAssembly] Initial support for stub libraries

See the docs in lld/docs/WebAssembly.rst for more on this.

This feature unlocks a lot of simplification in the emscripten toolchain
since we can represent the JS libraries to wasm-ld as stub libraries.

See https://github.com/emscripten-core/emscripten/issues/18875

Differential Revision: https://reviews.llvm.org/D145308