platform/upstream/llvm.git
18 months ago[ConstantRange] Test 1 bit ranges in exhaustive tests (NFC)
Nikita Popov [Mon, 9 Jan 2023 15:51:43 +0000 (16:51 +0100)]
[ConstantRange] Test 1 bit ranges in exhaustive tests (NFC)

There have been multiple cases where range calculations were wrong
in the 1 bit case. Make sure we catch these by not specifying the
bit width explicitly, and letting the test framework pick it (which
will now always test 1 and 4 bits both).

18 months ago[compiler-rt] Move up undefined macro checks
Alex Brachet [Mon, 9 Jan 2023 15:54:11 +0000 (15:54 +0000)]
[compiler-rt] Move up undefined macro checks

Previously HWCAP_ATOMIC and others were being used before checking if
they were defined. This moves up all the ifndef checks to define these
macros if they are not yet defined.

Differential Revision: https://reviews.llvm.org/D141285

18 months ago[mlir][spirv] Clean up transform pass definitions. NFC.
Jakub Kuderski [Mon, 9 Jan 2023 15:50:06 +0000 (10:50 -0500)]
[mlir][spirv] Clean up transform pass definitions. NFC.

- Make naming more consistent.
- Drop unnecessary custom constructors definitions.
- Move pass documentation to pass descriptions.

Reviewed By: antiagainst

Differential Revision: https://reviews.llvm.org/D141159

18 months ago[mlir][vector] Distribute vector.insert op
Matthias Springer [Mon, 9 Jan 2023 15:46:01 +0000 (16:46 +0100)]
[mlir][vector] Distribute vector.insert op

In case the distributed dim of the dest vector is also a dim of the src vector, each lane inserts a smaller part of the source vector. Otherwise, one lane inserts the entire src vector and the other lanes do nothing.

Differential Revision: https://reviews.llvm.org/D137953

18 months ago[AMDGPU] S_MULK_I32 does not define SCC. NFCI.
Jay Foad [Mon, 9 Jan 2023 14:47:09 +0000 (14:47 +0000)]
[AMDGPU] S_MULK_I32 does not define SCC. NFCI.

Differential Revision: https://reviews.llvm.org/D141281

18 months ago[mlir][vector] Distribute vector.insertelement op
Matthias Springer [Mon, 9 Jan 2023 15:40:32 +0000 (16:40 +0100)]
[mlir][vector] Distribute vector.insertelement op

In case of a distribution, only one lane inserts the scalar value. In case of a broadcast, every lane inserts the scalar.

Differential Revision: https://reviews.llvm.org/D137929

18 months ago[mlir][vector] Support vector.extract distribution of >1D vectors
Matthias Springer [Mon, 9 Jan 2023 15:35:29 +0000 (16:35 +0100)]
[mlir][vector] Support vector.extract distribution of >1D vectors

Ops such as `%1 = vector.extract %0[2] : vector<5x96xf32>`.

Distribute the source vector, then extract. In case of a 1d extract, rewrite to vector.extractelement.

Differential Revision: https://reviews.llvm.org/D137646

18 months ago[libc++] Add missing include in __format/unicode.h
Nikolas Klauser [Mon, 9 Jan 2023 15:37:31 +0000 (16:37 +0100)]
[libc++] Add missing include in __format/unicode.h

18 months ago[ConstantRange] Fix single bit abs range (PR59887)
Nikita Popov [Mon, 9 Jan 2023 15:29:37 +0000 (16:29 +0100)]
[ConstantRange] Fix single bit abs range (PR59887)

For a full range input, we would produce an empty range instead
of a full range. The change to the SMin.isNonNegative() branch is
an optimality fix, because we should account for the potentially
discarded SMin value in the IntMinIsPoison case.

Change TestUnaryOpExhaustive to test both 4 and 1 bits, to both
cover this specific case in unit tests, and make sure all other
unary operations deal with 1-bit inputs correctly.

Fixes https://github.com/llvm/llvm-project/issues/59887.

18 months ago[libc++][test] Fix missing include in `bit_ceil.fail.cpp`
Joe Loser [Mon, 9 Jan 2023 15:13:38 +0000 (08:13 -0700)]
[libc++][test] Fix missing include in `bit_ceil.fail.cpp`

The test uses `size_t` but does not include a header defining it.  Include
`<cstddef>` which provides `size_t`.

Differential Revision: https://reviews.llvm.org/D141284

18 months ago[libc++] Remove HIDE_FROM_ABI from virtual functions
Louis Dionne [Wed, 21 Dec 2022 15:08:54 +0000 (10:08 -0500)]
[libc++] Remove HIDE_FROM_ABI from virtual functions

_LIBCPP_HIDE_FROM_ABI (which is what _LIBCPP_INLINE_VISIBILITY is) uses
ABI tags to avoid ODR violations when linking together object files
compiled against different versions of libc++. However, pointer
authentication uses the mangled name of the function to sign the
function pointer in the vtable, which means that the ABI tag effectively
changes how the pointers are signed.

This leads to PAC failures when passing an object that holds one of these
pointers in its vtable across an ABI boundary: one side will sign the
pointer using one function mangling (with one ABI tag), and the other
side will authenticate the pointer expecting it to have a different
mangled name, which won't work.

To make sure this does not regress in the future, this patch also adds
a clang-query test to detect incorrect applications of _LIBCPP_HIDE_FROM_ABI.

Differential Revision: https://reviews.llvm.org/D140453

18 months ago[InstCombine] reduce smul.ov with i1 types to 'and'
Sanjay Patel [Mon, 9 Jan 2023 14:53:45 +0000 (09:53 -0500)]
[InstCombine] reduce smul.ov with i1 types to 'and'

https://alive2.llvm.org/ce/z/5tLkW6

There's still a miscompile bug as shown in issue #59876 / D141214 .

18 months ago[InstCombine] add tests for smul/umul with overflow with i1 types; NFC
Sanjay Patel [Mon, 9 Jan 2023 14:26:28 +0000 (09:26 -0500)]
[InstCombine] add tests for smul/umul with overflow with i1 types; NFC

More coverage for D141214 / issue #59876

18 months ago[AArch64] Add test showing reassociation potential.
Florian Hahn [Mon, 9 Jan 2023 15:20:55 +0000 (15:20 +0000)]
[AArch64] Add test showing reassociation potential.

Add a test case where some ops of a reassociate-able expression are in
an earlier block.

This can appear in practice, e.g. when computing the final reduction
value after vectorization.

18 months ago[CVP] Add test for PR59887 (NFC)
Nikita Popov [Mon, 9 Jan 2023 15:15:36 +0000 (16:15 +0100)]
[CVP] Add test for PR59887 (NFC)

Also fix all the incorrect intrinsic name mangling while here.

18 months ago[AArch64][SME] Improve streaming-compatible codegen for extending loads/truncating...
Sander de Smalen [Mon, 9 Jan 2023 09:45:50 +0000 (09:45 +0000)]
[AArch64][SME] Improve streaming-compatible codegen for extending loads/truncating stores.

This is another step in aligning addTypeForStreamingSVE with addTypeForFixedLengthSVE,
which also improves code quality for extending loads and truncating stores.

Reviewed By: hassnaa-arm

Differential Revision: https://reviews.llvm.org/D141266

18 months ago[clangd] Full support for #import insertions
David Goldman [Tue, 6 Dec 2022 21:33:12 +0000 (16:33 -0500)]
[clangd] Full support for #import insertions

These are still disabled by default, but will work in ObjC code if you
enable the `-import-insertions` flag.

Completion requires ASTSignals to be available; before ASTSignals are
available, we will always use #include. Once they are available, the
behavior varies as follows:

- For source files, use #import if the ObjC language flag is enabled
- For header files:
  - If the ObjC language flag is disabled, use #include
  - If the header file contains any #imports, use #import
  - If the header file references any ObjC decls, use #import
  - Otherwise, use #include

IncludeFixer support is similar, but it does not rely upon ASTSignals,
instead it does the above checks excluding the scan for ObjC symbols.

Differential Revision: https://reviews.llvm.org/D139458

18 months ago[clangd] Add flag to control #import include insertions
David Goldman [Tue, 6 Dec 2022 19:25:46 +0000 (14:25 -0500)]
[clangd] Add flag to control #import include insertions

This will be disabled by default, hopefully we can enable for the next
major release.

Differential Revision: https://reviews.llvm.org/D139446

18 months ago[InstCombine] Preserve alignment in atomicrmw -> store fold
Nikita Popov [Mon, 9 Jan 2023 14:36:32 +0000 (15:36 +0100)]
[InstCombine] Preserve alignment in atomicrmw -> store fold

Preserve the alignment of the original atomicrmw, rather than using
the ABI alignment.

The same problem exists for loads, but that code is being removed
in D141277 anyway.

18 months ago[flang] Disable LTO when building the flang runtime
David Truby [Fri, 18 Nov 2022 16:55:56 +0000 (16:55 +0000)]
[flang] Disable LTO when building the flang runtime

When building the flang runtime if LTO is enabled the archive file
contains LLVM IR rather than object code. Currently flang is not
LTO aware so cannot link this file to compiled Fortran code.

This patch disables LTO when building the flang runtime to avoid
this issue.

Differential Revision: https://reviews.llvm.org/D140016

18 months ago[InstCombine] Test alignment in atomicrmw -> store transform (NFC)
Nikita Popov [Mon, 9 Jan 2023 14:25:15 +0000 (15:25 +0100)]
[InstCombine] Test alignment in atomicrmw -> store transform (NFC)

And regenerate test checks. The current alignment is incorrect.

18 months ago[LLD] [COFF] Add a testcase for the type of autoexported symbols from LTO
Martin Storsjö [Mon, 9 Jan 2023 11:18:19 +0000 (13:18 +0200)]
[LLD] [COFF] Add a testcase for the type of autoexported symbols from LTO

This adds test coverage for the behaviour that broke in
7370ff624d217b0f8f7512ca5b651a9b8095a411.

18 months agoRevert "[LLD] Remove global state in lld/COFF"
Martin Storsjö [Mon, 9 Jan 2023 14:04:09 +0000 (16:04 +0200)]
Revert "[LLD] Remove global state in lld/COFF"

This reverts commit 7370ff624d217b0f8f7512ca5b651a9b8095a411.
(and 47fb8ae2f9a4075de05433ef24f459b6befd1730).

This commit broke the symbol type in import libraries generated
for mingw autoexported symbols, when the source files were built
with LTO. I'll commit a testcase that showcases this issue after
the revert.

18 months ago[RISC-V][HWASAN] Don't explicitly load GOT entry to call hwasan mismatch routine
Alexey Baturo [Wed, 31 Aug 2022 05:21:39 +0000 (08:21 +0300)]
[RISC-V][HWASAN] Don't explicitly load GOT entry to call hwasan mismatch routine

Reviewed by: luismarques

Differential Revision: https://reviews.llvm.org/D132994

18 months ago[ARM] Fold And/Or into CSel if possible
David Green [Mon, 9 Jan 2023 13:28:57 +0000 (13:28 +0000)]
[ARM] Fold And/Or into CSel if possible

This is the ARM equivalent of D141119, where we fold `and x, (csel 0, 1, cc)`
to `csel ZR, x, cc` if we know that x is 0/1 and for `or x, (csel 0, 1, cc)`
emit `csinc x, ZR, cc`. The or pattern gets recognized from a cmov under Arm.

Differential Revision: https://reviews.llvm.org/D141137

18 months ago[InstCombine] Fold zero check followed by decrement to usub.sat
Jamie Hill-Daniel [Mon, 9 Jan 2023 13:21:12 +0000 (14:21 +0100)]
[InstCombine] Fold zero check followed by decrement to usub.sat

Fold (a == 0) : 0 ? a - 1 into usub.sat(a, 1).

Differential Revision: https://reviews.llvm.org/D140798

18 months ago[InstCombine] Add tests for saturating subtract by one (NFC)
Jamie Hill-Daniel [Mon, 9 Jan 2023 13:07:55 +0000 (14:07 +0100)]
[InstCombine] Add tests for saturating subtract by one (NFC)

Tests for D140798.

18 months ago[mlir] verify against nullptr payload in transform dialect
Alex Zinenko [Mon, 9 Jan 2023 13:01:25 +0000 (14:01 +0100)]
[mlir] verify against nullptr payload in transform dialect

When establishing the correspondence between transform values and
payload operations or parameters, check that the latter are non-null and
report errors. This was previously allowed for exotic cases of partially
successfull transformations with "apply each" trait, but was dangerous.
The "apply each" implementation was reworked to remove the need for this
functionality, so this can now be hardned to avoid null pointer
dereferences.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D141142

18 months ago[include-cleaner] Don't count references to operators as uses
Haojian Wu [Thu, 22 Dec 2022 10:54:07 +0000 (11:54 +0100)]
[include-cleaner] Don't count references to operators as uses

Fixes https://github.com/llvm/llvm-project/issues/59655

Differential Revision: https://reviews.llvm.org/D140551

18 months ago[include-cleaner] Filter template instantiations from AST roots.
Viktoriia Bakalova [Mon, 9 Jan 2023 10:49:08 +0000 (10:49 +0000)]
[include-cleaner] Filter template instantiations from AST roots.

Fix: https://github.com/llvm/llvm-project/issues/59825
Differential Revision: https://reviews.llvm.org/D141271

18 months ago[mlir][linalg] Reuploading: Apply shortened printing/parsing form to linalg.reduce.
Aliia Khasanova [Mon, 9 Jan 2023 07:25:03 +0000 (08:25 +0100)]
[mlir][linalg] Reuploading: Apply shortened printing/parsing form to linalg.reduce.

Differential Revision: https://reviews.llvm.org/D141259

18 months ago[Test] One more test where check is not replaced to invariant
Max Kazantsev [Mon, 9 Jan 2023 12:23:46 +0000 (19:23 +0700)]
[Test] One more test where check is not replaced to invariant

Irrelevant constant check makes things even more difficult, surprisingly.

18 months agoLower math.cbrt to NVVM/ROCDL.
Johannes Reifferscheid [Mon, 9 Jan 2023 10:48:43 +0000 (11:48 +0100)]
Lower math.cbrt to NVVM/ROCDL.

Reviewed By: pifon2a

Differential Revision: https://reviews.llvm.org/D141270

18 months ago[Attributes] Avoid duplicate hasAttribute() query (NFC)
Nikita Popov [Mon, 9 Jan 2023 11:55:47 +0000 (12:55 +0100)]
[Attributes] Avoid duplicate hasAttribute() query (NFC)

removeAttribute() already performs a hasAttribute() check, so no
need to also do it in the caller. Instead check whether the
attribute set was changed.

This makes the implementations in line with removeAttributesAtIndex().

18 months ago[AArch64] Fold And/Or into CSel if possible
David Green [Mon, 9 Jan 2023 11:52:37 +0000 (11:52 +0000)]
[AArch64] Fold And/Or into CSel if possible

If we have `and x, (csel 0, 1, cc)` and we know that x is 0/1, then we
can emit a `csel ZR, x, cc`. Similarly for `or x, (csel 0, 1, cc)` we
can emit `csinc x, ZR, cc`. This can help where we can not otherwise
general ccmp instructions.

Differential Revision: https://reviews.llvm.org/D141119

18 months ago[InstCombine] Canonicalize (A & B_Pow2) eq/ne B_Pow2 patterns
Noah Goldstein [Mon, 9 Jan 2023 11:44:03 +0000 (12:44 +0100)]
[InstCombine] Canonicalize (A & B_Pow2) eq/ne B_Pow2 patterns

1. A & B_Pow2 != B_Pow2 -> A & B_Pow2 == 0
   https://alive2.llvm.org/ce/z/KVUej4

2. A & B_Pow2 == B_Pow2 -> A & B_Pow2 != 0
   https://alive2.llvm.org/ce/z/PVv9FR

This allows the patterns to more easily be analyzed elsewhere.

Differential Revision: https://reviews.llvm.org/D141090

18 months ago[lld][ARM] don't use short thumb thunks if no branch range extension
Ties Stuij [Mon, 9 Jan 2023 11:06:00 +0000 (11:06 +0000)]
[lld][ARM] don't use short thumb thunks if no branch range extension

In ThumbThunk::isCompatibleWith, we check if we can use short thunks if we are
within branch range. However these short thumb thunks will generate b.w
instructions, and these are not available on pre branch range extension
architectures.

On these architectures (v4, v5, and most of v6), we could replace the b.w with a
Thumb b (2) instruction, but that would in an ideal situation only give us an
extra range of 2048 bytes on top of the 4MB range of a BL, if a thunk section
happens to be placed on the outer range of a BL and the stars are aligned. It
doesn't seem worth it.

What would be worth it is a state change to Arm and a subsequent branch to
either Arm or Thumb code. But that's the subject of another patch.

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D140633

18 months ago[DebugInfo] Fix jump threading failing to update cloned dbg.values
Ben Mudd [Mon, 9 Jan 2023 11:04:39 +0000 (11:04 +0000)]
[DebugInfo] Fix jump threading failing to update cloned dbg.values

This is a patch to fix duplicated dbg.values in the JumpThreading pass not
pointing towards their local value, and instead towards the variable in the
original block.
JumpThreadingPass::cloneInstructions is the changed function to target metadata
as well as normal cloned values.

Reviewed By: jmorse, StephenTozer

Differential Revision: https://reviews.llvm.org/D140006

18 months ago[libc++] Fix transitive includes list for C++23
Nikolas Klauser [Mon, 9 Jan 2023 11:19:36 +0000 (12:19 +0100)]
[libc++] Fix transitive includes list for C++23

18 months agoMove from llvm::makeArrayRef to ArrayRef deduction guides - clang/ part
serge-sans-paille [Fri, 6 Jan 2023 15:56:23 +0000 (16:56 +0100)]
Move from llvm::makeArrayRef to ArrayRef deduction guides - clang/ part

This is a follow-up to https://reviews.llvm.org/D140896, split into
several parts as it touches a lot of files.

Differential Revision: https://reviews.llvm.org/D141139

18 months agoTailDuplication: do not remove trivial PHIs from addr-taken blocks.
Tim Northover [Mon, 9 Jan 2023 10:52:34 +0000 (10:52 +0000)]
TailDuplication: do not remove trivial PHIs from addr-taken blocks.

Unlike an anonymous block, it will not be removed even though we've resolved
all valid paths to get here. So removing a PHI can leave vregs with no
definition, violating SSA. Instead, this converts it to an IMPLICIT_DEF.

18 months ago[libc++][test] Fix missing include in `endian.pass.cpp`
Joe Loser [Mon, 9 Jan 2023 11:09:45 +0000 (12:09 +0100)]
[libc++][test] Fix missing include in `endian.pass.cpp`

`endian.pass.cpp` uses `std::is_enum` and friends but doesn't include
`<type_traits>`.  Add the missing include.

Reviewed By: philnik, #libc

Spies: libcxx-commits

Differential Revision: https://reviews.llvm.org/D141239

18 months ago[Attributes] Avoid repeated attribute set lookup (NFC)
Nikita Popov [Mon, 9 Jan 2023 10:58:06 +0000 (11:58 +0100)]
[Attributes] Avoid repeated attribute set lookup (NFC)

Perform the hasAttribute() check on the AttributeSet we need to
fetch anyway, rather than going through hasAttributeAtIndex().

18 months ago[InstCombine] Fix potentially buggy code in `((%x & C) == 0) --> %x u< (-C)` transform
Noah Goldstein [Mon, 9 Jan 2023 10:38:36 +0000 (11:38 +0100)]
[InstCombine] Fix potentially buggy code in `((%x & C) == 0) --> %x u< (-C)` transform

While demanded bits constant shrinking appears to prevent this in
practice right now, it is principally possible for C2 to have
set bits that are known not-needed (zeroable). See: D140858

`+` will overflow here, `|` will get the right logic.

Differential Revision: https://reviews.llvm.org/D141089

18 months ago[llvm] Fix an "unused variable" warning when assertions are disabled
Dmitri Gribenko [Mon, 9 Jan 2023 10:31:25 +0000 (11:31 +0100)]
[llvm] Fix an "unused variable" warning when assertions are disabled

18 months ago[TableGen][SourceMgr] Fix obvious mistake in D141220
Markus Böck [Mon, 9 Jan 2023 10:16:56 +0000 (11:16 +0100)]
[TableGen][SourceMgr] Fix obvious mistake in D141220

It now tried to open the IncludedFile instead of the Filename, which was not intended.

18 months ago[AArch64] NFC: Align addTypeForStreamingSVE and addTypeForFixedLengthSVE
Sander de Smalen [Mon, 9 Jan 2023 09:26:20 +0000 (09:26 +0000)]
[AArch64] NFC: Align addTypeForStreamingSVE and addTypeForFixedLengthSVE

This patch is NFC and just moves things around so their implementation is very similar.

18 months ago[bazel] Updates for https://github.com/llvm/llvm-project/commit/496f9a7d8d4b61b6f5dfa...
Dmitri Gribenko [Mon, 9 Jan 2023 09:27:15 +0000 (10:27 +0100)]
[bazel] Updates for https://github.com/llvm/llvm-project/commit/496f9a7d8d4b61b6f5dfa21e2614043a2b27cf67

18 months ago[bazel] Update the mpfr URL to a stable one
Dmitri Gribenko [Mon, 9 Jan 2023 09:26:25 +0000 (10:26 +0100)]
[bazel] Update the mpfr URL to a stable one

mpfr has released a new version, and now the old tar.gz URL does not
work anymore. Update the URL to a stable one.

18 months ago[mlir:LLVM] Rudimentary inlining support for LLVM load store.
Johannes de Fine Licht [Mon, 9 Jan 2023 09:28:10 +0000 (10:28 +0100)]
[mlir:LLVM] Rudimentary inlining support for LLVM load store.

Conservatively only allow inlining for loads and stores that don't carry
any attributes that require handling while inlining. This can later be
relaxed when proper handling is introduced.

Reviewed By: Dinistro, gysit

Differential Revision: https://reviews.llvm.org/D141115

18 months ago[mlir][llvm] Add branch weights to call and invoke
Christian Ulmann [Mon, 9 Jan 2023 09:14:21 +0000 (10:14 +0100)]
[mlir][llvm] Add branch weights to call and invoke

This commit introduces branch weight attributes to the LLVM::CallOp and
LLVM::InvokeOp and adds both import and export of them.

Reviewed By: gysit

Differential Revision: https://reviews.llvm.org/D141122

18 months ago[NFC] Missing whitespace in SSAUpdaterBulk debug output.
Thomas Symalla [Mon, 9 Jan 2023 09:11:08 +0000 (10:11 +0100)]
[NFC] Missing whitespace in SSAUpdaterBulk debug output.

Adds a whitespace in a debug message before printing out a
value in the SSAUpdaterBulk.
Without this, debugging can end up a bit cumbersome.

Differential Revision: https://reviews.llvm.org/D141262

18 months ago[JumpThreading] Preserve profile metadata during select unfolding
Max Kazantsev [Mon, 9 Jan 2023 09:04:00 +0000 (16:04 +0700)]
[JumpThreading] Preserve profile metadata during select unfolding

Jump threading can replace select and unconditional branch with
conditional branch, but when doing so loses profile information.

This destructive transform can eventually lead to a performance
degradation due to folding of branches in
shouldFoldCondBranchesToCommonDestination as branch probabilities
are no longer known.

Patch by Roman Paukner!

Differential Revision: https://reviews.llvm.org/D138132
Reviewed By: mkazantsev

18 months ago[gn build] Port 59b029238af2
LLVM GN Syncbot [Mon, 9 Jan 2023 09:00:23 +0000 (09:00 +0000)]
[gn build] Port 59b029238af2

18 months ago[DebugInfo] Produce variadic DBG_INSTR_REFs from ISel
Stephen Tozer [Tue, 3 Jan 2023 18:13:31 +0000 (18:13 +0000)]
[DebugInfo] Produce variadic DBG_INSTR_REFs from ISel

This patch modifies SelectionDAG and FastISel to produce DBG_INSTR_REFs with
variadic expressions, and produce DBG_INSTR_REFs for debug values with variadic
location expressions. The former essentially means just prepending
DW_OP_LLVM_arg, 0 to the existing expression. The latter is achieved in
MachineFunction::finalizeDebugInstrRefs and InstrEmitter::EmitDbgInstrRef.

Reviewed By: jmorse, Orlando

Differential Revision: https://reviews.llvm.org/D133929

18 months ago[clang][analyzer] Remove report of null stream from StreamChecker.
Balázs Kéri [Mon, 9 Jan 2023 08:16:18 +0000 (09:16 +0100)]
[clang][analyzer] Remove report of null stream from StreamChecker.

The case of NULL stream passed to stream functions was reported by StreamChecker.
The same condition is checked already by StdLibraryFunctionsChecker and it is
enough to check at one place. The StreamChecker stops now analysis if a passed NULL
stream is encountered but generates no report.
This change removes a dependency between StdCLibraryFunctionArgs checker and
StreamChecker. There is now no more specific message reported by StreamChecker,
the previous weak-dependency is not needed. And StreamChecker can be used
without StdCLibraryFunctions checker or its ModelPOSIX option.

Reviewed By: Szelethus

Differential Revision: https://reviews.llvm.org/D137790

18 months ago[reland][NFC] Vastly simplifies TypeSize
Guillaume Chatelet [Fri, 6 Jan 2023 16:33:56 +0000 (16:33 +0000)]
[reland][NFC] Vastly simplifies TypeSize

Simplifies the implementation of `TypeSize` while retaining its interface.
There is no need for abstract concepts like `LinearPolyBase`, `UnivariateLinearPolyBase` or `LinearPolySize`.

Differential Revision: https://reviews.llvm.org/D140263

18 months ago[AArch64][SelectionDAG] Eliminates redundant zero-extension for 32-bit popcount
zhongyunde [Mon, 9 Jan 2023 08:04:49 +0000 (16:04 +0800)]
[AArch64][SelectionDAG] Eliminates redundant zero-extension for 32-bit popcount

Fix https://github.com/llvm/llvm-project/issues/59597.
mov w8, w0 + fmov d0, x8 ==> fmov s0, w0

Reviewed By: dmgreen, efriedma

Differential Revision: https://reviews.llvm.org/D140649

18 months ago[TableGen][SourceMgr] Correctly append filename to include directories
Markus Böck [Sun, 8 Jan 2023 21:20:36 +0000 (22:20 +0100)]
[TableGen][SourceMgr] Correctly append filename to include directories

The current implementation unconditionally appends the system path separator with the filename to the include directory. This is not correct in edge cases however, such as when specifying `/` as include directory (on Unix systems) or just `\` on Windows.
This patch fixes that by using `sys::path::append`, which already has the required logic to correctly implement this.

While this is technically only a change in the `SourceMgr` class, I think the main user of that class, and the include mechanism, is TableGen.
No test attached because no behavioral difference is observable without trying to access the root directory of the users filesystem.

The motivation for this change is a rather funny story, as this actually fixes a performance problem when running `check-mlir` on Windows.
Some tests for `mlir-pdll-lsp-server` lead to adding `\` as include directory in TableGen (which is a valid absolute path on Windows!). Due to the unconditional append, the created filepath would then be of the form `\\<dir>\...` which is also a valid path on Windows, but is a network path. On my machine it'd then attempt to access the network and find a machine with the name `<dir>` and the file there. This call would take several seconds, leading to some tests in `mlir-pdll-lsp-server` taking 2 minutes on my machine.

Running `check-mlir` after this patch reduces the runtime on my machine from 161 seconds to 6 seconds.

Differential Revision: https://reviews.llvm.org/D141220

18 months ago[NFC] Add missing 'static' notion in createReplacement
Max Kazantsev [Mon, 9 Jan 2023 06:50:07 +0000 (13:50 +0700)]
[NFC] Add missing 'static' notion in createReplacement

18 months ago[Inline Spiller] Extend the snippet by statepoint uses
Serguei Katkov [Thu, 20 Oct 2022 05:05:33 +0000 (12:05 +0700)]
[Inline Spiller] Extend the snippet by statepoint uses

Snippet is a tiny live interval which has copy or fill like def
and copy or spill like use at the end (any of them might abcent).

Snippet has only one use/def inside interval and interval is located
in one basic block.

When inline spiller spills some reg around uses it also forces the
spilling of connected snippets those which got by splitting the
same original reg and its def is a full copy of our reg or its
last use is a full copy to our reg.

The definition of snippet is extended to allow not only one use/def
but more. However all other uses are statepoint instructions which will
fold fill into its operand. That way we do not introduce new fills/spills.

Reviewed By: qcolombet, dantrushin
Differential Revision: https://reviews.llvm.org/D138093

18 months ago[MLIR] NFC. Fix DEBUG_TYPE for AffineOps.cpp
Uday Bondhugula [Mon, 9 Jan 2023 05:40:44 +0000 (11:10 +0530)]
[MLIR] NFC. Fix DEBUG_TYPE for AffineOps.cpp

Fix DEBUG_TYPE for AffineOps.cpp. NFC.

Differential Revision: https://reviews.llvm.org/D141252

18 months ago[RISCV][CostModel] Add half type support for the cost model of sqrt/fabs
liqinweng [Mon, 9 Jan 2023 04:56:10 +0000 (12:56 +0800)]
[RISCV][CostModel] Add half type support for the cost model of sqrt/fabs

1. Refactor for costs of sqrt/fabs
2. Add half type support for the cost model of sqrt/fabs

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D132908

18 months agolldCOFF: Fix warnings for D110450 [-Wunused-private-field]
NAKAMURA Takumi [Mon, 9 Jan 2023 03:15:19 +0000 (12:15 +0900)]
lldCOFF: Fix warnings for D110450 [-Wunused-private-field]

18 months ago[NFC] Fix the test failure in 08f957808e5f7e44b11d in armv8
Chuanqi Xu [Mon, 9 Jan 2023 03:23:30 +0000 (11:23 +0800)]
[NFC] Fix the test failure in  08f957808e5f7e44b11d in armv8

The test in 08f957808e5f7e44b11d may fail in armv8 since the signature
of the constructor may be different. This patch tries to fix it.

18 months ago[RISCV][CostModel] Add cost model for integer abs
liqinweng [Mon, 9 Jan 2023 03:38:24 +0000 (11:38 +0800)]
[RISCV][CostModel] Add cost model for integer abs

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D132999

18 months ago[C++20] [Modules] Don't generate global ctors/dtors for variables which are available...
Chuanqi Xu [Mon, 9 Jan 2023 02:43:35 +0000 (10:43 +0800)]
[C++20] [Modules] Don't generate global ctors/dtors for variables which are available externally

Closes https://github.com/llvm/llvm-project/issues/59765.

Currently we will generate the global ctor/dtor for variables in
importing modules. It will cause multiple initialization/destructions.
It makes no sense. This patch tries to not generate global ctor/dtor for
variables which are available externally. Note that the variables in
header units and clang modules won't be available externally by default.

Reviewed By: iains

Differential Revision: https://reviews.llvm.org/D140867

18 months ago[C++20] [Modules] Make placement allocation functions always visible
Chuanqi Xu [Mon, 9 Jan 2023 02:39:35 +0000 (10:39 +0800)]
[C++20] [Modules] Make placement allocation functions always visible

Close https://github.com/llvm/llvm-project/issues/59601.

This is actually a workaround for the issue. See the comments and the
test for example. The proper fix should make the placement allocation
functions acceptable based on the context. But it is harder and more
complex on the one side. On the other side, such workaround won't be too
bad in practice since users rarely call the placement allocation
functions directly.

So personally I prefer to address such problems in the simpler way.

Reviewed By: royjacobson

Differential Revision: https://reviews.llvm.org/D141023

18 months ago[InstCombine] Fold logic-and/logic-or by distributive laws part2
chenglin.bi [Mon, 9 Jan 2023 02:21:05 +0000 (10:21 +0800)]
[InstCombine] Fold logic-and/logic-or by distributive laws part2

Follow up https://reviews.llvm.org/D139408, support `and/or+select` patterns
X && Z || Y && Z --> (X || Y) && Z
https://alive2.llvm.org/ce/z/EMCkBG
https://alive2.llvm.org/ce/z/Q-YRvr
https://alive2.llvm.org/ce/z/SFkVQc
https://alive2.llvm.org/ce/z/S9MCuJ
https://alive2.llvm.org/ce/z/KZ7zzz

(X || Z) && (Y || Z) --> (X && Y) || Z
https://alive2.llvm.org/ce/z/Ggpa8-
https://alive2.llvm.org/ce/z/nhQRLY
https://alive2.llvm.org/ce/z/zpmEnq
https://alive2.llvm.org/ce/z/7omsrf
https://alive2.llvm.org/ce/z/CWBzBp

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D139630

18 months ago[CodeGen] Fix a warning
Kazu Hirata [Mon, 9 Jan 2023 00:31:45 +0000 (16:31 -0800)]
[CodeGen] Fix a warning

This patch fixes:

  llvm/lib/CodeGen/AssignmentTrackingAnalysis.cpp:1220:13: error:
  unused function 'locStr' [-Werror,-Wunused-function]

18 months ago[libc++] Add _LIBCPP_HIDE_FROM_ABI to __constexpr_logb and __constexpr_scalbn
Nikolas Klauser [Mon, 9 Jan 2023 00:16:34 +0000 (01:16 +0100)]
[libc++] Add _LIBCPP_HIDE_FROM_ABI to __constexpr_logb and __constexpr_scalbn

18 months ago[LLD] Remove global state in lld/COFF
Amy Huang [Thu, 14 Jul 2022 19:53:32 +0000 (15:53 -0400)]
[LLD] Remove global state in lld/COFF

Remove globals from the lldCOFF library, by moving globals into a context class.
This patch mostly moves the config object into COFFLinkerContext.

See https://lists.llvm.org/pipermail/llvm-dev/2021-June/151184.html for
context about removing globals from LLD.

Reviewed By: aganea

Differential Revision: https://reviews.llvm.org/D110450

18 months ago[libc++][test] zip_view test cleanups
Casey Carter [Sun, 8 Jan 2023 07:12:21 +0000 (23:12 -0800)]
[libc++][test] zip_view test cleanups

* The reference type of `common_input_iterator<const int*>` can't be `int&`, because an lvalue of type `const int` _can't_ bind to an `int&`. Fix by changing the return type of `operator*` to `decltype(auto)` to make it fully generic.
* `range.zip/iterator/compare.pass.cpp` verifies that the iterators of a `zip_view` don't support `<=>` when the underlying iterators do not; this is not true after LWG-3692.
* libc++ doesn't yet implement P2165R4 "Compatibility between tuple, pair and tuple-like objects", so the tests expect `zip_view` to use `pair` in places where the working draft requires `tuple`.

Differential Revision: https://reviews.llvm.org/D141216

18 months ago[AVR] Support .reloc directive
Fangrui Song [Sun, 8 Jan 2023 23:27:21 +0000 (23:27 +0000)]
[AVR] Support .reloc directive

Reviewed By: benshi001

Differential Revision: https://reviews.llvm.org/D141176

18 months ago[libc][Obvious] Temporarily disable log10_test for further investigation.
Tue Ly [Sun, 8 Jan 2023 23:21:30 +0000 (18:21 -0500)]
[libc][Obvious] Temporarily disable log10_test for further investigation.

18 months ago[libc][Obvious] Remove constexpr fomr exact_mult in double_double.h
Tue Ly [Sun, 8 Jan 2023 22:52:46 +0000 (17:52 -0500)]
[libc][Obvious] Remove constexpr fomr exact_mult in double_double.h

18 months ago[libc][math] Implement log10 function correctly rounded for all rounding modes
Tue Ly [Sun, 8 Jan 2023 22:19:13 +0000 (17:19 -0500)]
[libc][math] Implement log10 function correctly rounded for all rounding modes

Implement double precision log10 function correctly rounded for all
rounding modes.  This implementation currently needs FMA instructions for
correctness.

Use 2 passes:
Fast pass:
- 1 step range reduction with a lookup table of `2^7 = 128` elements to reduce the ranges to `[-2^-7, 2^-7]`.
- Use a degree-7 minimax polynomial generated by Sollya, evaluated using a mixed of double-double and double precisions.
- Apply Ziv's test for accuracy.
Accurate pass:
- Apply 5 more range reduction steps to reduce the ranges further to [-2^-27, 2^-27].
- Use a degree-4 minimax polynomial generated by Sollya, evaluated using 192-bit precisions.
- By the result of Lefevre (add quote), this is more than enough for correct rounding to all rounding modes.

In progress: Adding detail documentations about the algorithm.

Depend on: https://reviews.llvm.org/D136799

Reviewed By: zimmermann6

Differential Revision: https://reviews.llvm.org/D139846

18 months ago[AAUnderlyingObjects] Introduce an AA for getting underlying objects of a pointer
Shilei Tian [Sun, 8 Jan 2023 21:45:42 +0000 (16:45 -0500)]
[AAUnderlyingObjects] Introduce an AA for getting underlying objects of a pointer

This patch introduces a new AA `AAUnderlyingObjects`. It is basically like a wrapper
AA of the function `AA::getAssumedUnderlyingObjects`, but it can recursively do
query if the underlying object is an indirect access, such as a phi node or a select
instruction.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D141164

18 months ago[mlir][analysis] Add an analysis for preserving symbol tables
Jeff Niu [Thu, 8 Dec 2022 20:22:40 +0000 (12:22 -0800)]
[mlir][analysis] Add an analysis for preserving symbol tables

This patch adds a `SymbolTableAnalysis` that can be used with the
analysis manager. It contains a symbol table collection. This analysis
allows symbol tables to be preserved across passes so that they do not
need to be recomputed. The analysis assumes it remains valid because
most transformations automatically keep symbol tables up-to-date using
its `insert` and `erase` methods.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D139666

18 months ago[AArch64][compiler-rt] Option to build compiler-rt without FMV support.
Pavel Iliin [Fri, 30 Dec 2022 13:34:45 +0000 (13:34 +0000)]
[AArch64][compiler-rt] Option to build compiler-rt without FMV support.

This commit adds compiler-rt cmake option COMPILER_RT_DISABLE_AARCH64_FMV
which, when enabled, doesn't include function multiversioning features
initilization code in 'builtins' build.

Differential Revision: https://reviews.llvm.org/D141199

18 months ago[X86] Add shuffle test coverage for Issue #59860
Simon Pilgrim [Sun, 8 Jan 2023 19:05:54 +0000 (19:05 +0000)]
[X86] Add shuffle test coverage for Issue #59860

18 months ago[AVR] Optimize 32-bit shifts: optimize REG_SEQUENCE
Ayke van Laethem [Tue, 6 Dec 2022 13:47:31 +0000 (14:47 +0100)]
[AVR] Optimize 32-bit shifts: optimize REG_SEQUENCE

This pseudo-instruction stores two small (8-bit) registers into one wide
(16-bit) register. But apparently the order matters a lot to the
register allocator.
This patch changes the order of inserting the registers to optimize for
the best register allocation in the tests of shift32.ll. It might be
detrimental in other cases, but keeping the registers in the same
physical register seems like it would be a common case.

Differential Revision: https://reviews.llvm.org/D140573

18 months ago[AVR] Optimize 32-bit shifts: reverse shift + move
Ayke van Laethem [Tue, 6 Dec 2022 13:39:37 +0000 (14:39 +0100)]
[AVR] Optimize 32-bit shifts: reverse shift + move

This optimization turns shifts of almost a multiple of 8 into a shift
into the opposite direction. Unfortunately it doesn't compose well with
the other optimizations (I've tried) so it's separate from them.

Differential Revision: https://reviews.llvm.org/D140572

18 months ago[AVR] Optimize 32-bit shifts: shift by 4 bits
Ayke van Laethem [Tue, 6 Dec 2022 13:26:01 +0000 (14:26 +0100)]
[AVR] Optimize 32-bit shifts: shift by 4 bits

This uses a complicated shift sequence that avr-gcc also uses, but
extended to work over any number of bytes and in both directions
(logical shift left and logical shift right). Unfortunately it can't be
used for an arithmetic shift right: I've tried to come up with a
sequence but couldn't.

Differential Revision: https://reviews.llvm.org/D140571

18 months ago[AVR] Optimize 32-bit shift: move bytes around
Ayke van Laethem [Tue, 6 Dec 2022 12:58:08 +0000 (13:58 +0100)]
[AVR] Optimize 32-bit shift: move bytes around

This patch optimizes 32-bit constant shifts by renaming registers. This
is very effective as the compiler would otherwise need to do a lot of
single bit shift instructions. Instead, the registers are renamed at the
SSA level which means the register allocator will insert the necessary
mov instructions.

Unfortunately, the register allocator will insert some unnecessary movs
with the current code. This will be fixed in a later patch.

Differential Revision: https://reviews.llvm.org/D140570

18 months ago[AVR] Custom lower 32-bit shift instructions
Ayke van Laethem [Tue, 6 Dec 2022 12:43:23 +0000 (13:43 +0100)]
[AVR] Custom lower 32-bit shift instructions

32-bit shift instructions were previously expanded using the default
SelectionDAG expander, which meant it used 16-bit constant shifts and
ORed them together. This works, but is far from optimal.

I've optimized 32-bit shifts on AVR using a custom inserter. This is
done using three new pseudo-instructions that take the upper and lower
bits of the value in two separate 16-bit registers and outputs two
16-bit registers.

This is the first commit in a series. When completed, shift instructions
will take around 31% less instructions on average for constant 32-bit
shifts, and is in all cases equal or better than the old behavior. It
also tends to match or outperform avr-gcc: the only cases where avr-gcc
does better is when it uses a loop to shift, or when the LLVM register
allocator inserts some unnecessary movs. But it even outperforms avr-gcc
in some cases where avr-gcc does not use a loop.

As a side effect, non-constant 32-bit shifts also become more efficient.

For some real-world differences: the build of compiler-rt I use in
TinyGo becomes 2.7% smaller and the build of picolibc I use becomes 0.9%
smaller. I think picolibc is a better representation of real-world code,
but even a ~1% reduction in code size is really significant.

The current patch just lays the groundwork. The result is actually a
regression in code size. Later patches will use this as a basis to
optimize these shift instructions.

Differential Revision: https://reviews.llvm.org/D140569

18 months ago[gn] port b712aef5b37e
Nico Weber [Sun, 8 Jan 2023 18:24:53 +0000 (13:24 -0500)]
[gn] port b712aef5b37e

18 months ago[gn build] Port 86aac87fe4b5
Nico Weber [Sun, 8 Jan 2023 18:22:18 +0000 (13:22 -0500)]
[gn build] Port 86aac87fe4b5

18 months ago[SelectionDAG][AVR] Add support for lrint and lround intrinsics
Ayke van Laethem [Sun, 1 Jan 2023 18:05:35 +0000 (19:05 +0100)]
[SelectionDAG][AVR] Add support for lrint and lround intrinsics

Integer legalization already supported splitting the output integer of
llround and llrint, but did not support this for lround and lrint yet.
This is not a problem for 32-bit architectures, but for 8/16-bit
architectures like AVR it results in a crash like this:

    ExpandIntegerResult #0: t7: i32 = lround t6

    LLVM ERROR: Do not know how to expand the result of this operator!

This patch simply add lrint/lround to the list of ISD opcodes to expand.

Fixes https://github.com/llvm/llvm-project/issues/59573.

Differential Revision: https://reviews.llvm.org/D140822

18 months ago[AVR] correctly declare __do_copy_data and __do_clear_bss
Ayke van Laethem [Mon, 2 Jan 2023 00:21:19 +0000 (01:21 +0100)]
[AVR] correctly declare __do_copy_data and __do_clear_bss

These two symbols are declared in object files to indicate whether .data
needs to be copied from flash or .bss needs to be cleared. They are
supported on avr-gcc and reduce firmware size a bit, which is especially
important on very small chips.

I checked the behavior of avr-gcc and matched it as well as possible.
From my investigation, it seems to work as follows:

__do_copy_data is set when the compiler finds a data symbol:
  * without a section name
  * with a section name starting with ".data" or ".gnu.linkonce.d"
  * with a section name starting with ".rodata" or ".gnu.linkonce.r" and
    flash and RAM are in the same address space

__do_clear_bss is set when the compiler finds a data symbol:
  * without a section name
  * with a section name that starts with .bss

Simply checking whether the calculated section name starts with ".data",
".rodata" or ".bss" should result in the same behavior.

Fixes: https://github.com/llvm/llvm-project/issues/58857

Differential Revision: https://reviews.llvm.org/D140830

18 months ago[X86][Disassembler][NFCI] Read bytes with support::endian::read
Benjamin Kramer [Sun, 8 Jan 2023 17:19:01 +0000 (18:19 +0100)]
[X86][Disassembler][NFCI] Read bytes with support::endian::read

18 months ago[InstCombine] fold not-shift of signbit to icmp+zext, part 2
Sanjay Patel [Sun, 8 Jan 2023 16:36:21 +0000 (11:36 -0500)]
[InstCombine] fold not-shift of signbit to icmp+zext, part 2

Follow-up to:
6c39a3aae1dc

That converted a pattern with ashr directly to icmp+zext, and
this updates the pattern that we used to convert to.

This canonicalizes to icmp for better analysis in the minimum case
and shortens patterns where the source type is not the same as dest type:
https://alive2.llvm.org/ce/z/tpXJ64
https://alive2.llvm.org/ce/z/dQ405O

This requires an adjustment to an icmp transform to avoid infinite looping.

18 months ago[OpenMP] Migrate OpenMPOffloadMappingFlags from Clang CodeGen to OMPConstants
Akash Banerjee [Sun, 8 Jan 2023 16:45:12 +0000 (16:45 +0000)]
[OpenMP] Migrate OpenMPOffloadMappingFlags from Clang CodeGen to OMPConstants

This patch moves the OpenMPOffloadMappingFlags enum definiition from Clang codegen to OMPConstants.h

Differential Revision: https://reviews.llvm.org/D140292

18 months ago[NFCI][Offload Bundler] Replace hand-rolled endian conversion with llvm::support
Benjamin Kramer [Sun, 8 Jan 2023 16:32:54 +0000 (17:32 +0100)]
[NFCI][Offload Bundler] Replace hand-rolled endian conversion with llvm::support

18 months ago[NFC] Hide implementation details in anonymous namespaces
Benjamin Kramer [Sun, 8 Jan 2023 16:25:29 +0000 (17:25 +0100)]
[NFC] Hide implementation details in anonymous namespaces

18 months ago[libc++] Granularize <bit> and remove <__bits>
Nikolas Klauser [Tue, 27 Dec 2022 01:31:38 +0000 (02:31 +0100)]
[libc++] Granularize <bit> and remove <__bits>

Reviewed By: Mordante, #libc

Spies: libcxx-commits

Differential Revision: https://reviews.llvm.org/D141225

18 months ago[VPlan] Keep entries in worklist in sinkScalarOperands.
Florian Hahn [Sun, 8 Jan 2023 15:52:00 +0000 (15:52 +0000)]
[VPlan] Keep entries in worklist in sinkScalarOperands.

Not removing the entries ensures that duplicates are avoided,
reducing the number of iterations.

18 months ago[libc++][CI] Fixes robust against ADL for C++03.
Mark de Wever [Wed, 28 Dec 2022 19:02:01 +0000 (20:02 +0100)]
[libc++][CI] Fixes robust against ADL for C++03.

This was disabled in D139545.

Reviewed By: philnik, #libc

Differential Revision: https://reviews.llvm.org/D140731

18 months ago[clang-tidy][NFC] Remove unused User argument in misc-misleading-bidirectional check
Carlos Galvez [Fri, 6 Jan 2023 11:02:33 +0000 (11:02 +0000)]
[clang-tidy][NFC] Remove unused User argument in misc-misleading-bidirectional check

It's not used anywhere.

Differential Revision: https://reviews.llvm.org/D141118

18 months ago[clangd] Fix an inlay-hint crash on a null deduced type.
Haojian Wu [Sat, 7 Jan 2023 21:59:50 +0000 (22:59 +0100)]
[clangd] Fix an inlay-hint crash on a null deduced type.