platform/upstream/llvm.git
16 months ago[SCCP][NFC] Regenerate test case
luxufan [Mon, 19 Jun 2023 06:15:01 +0000 (14:15 +0800)]
[SCCP][NFC] Regenerate test case

16 months ago[SCCP][NFC] Regenerate test case
luxufan [Mon, 19 Jun 2023 06:07:42 +0000 (14:07 +0800)]
[SCCP][NFC] Regenerate test case

16 months ago[LoopUnrollRuntime] Allow indirect transition to deopt non-latch exit blocks
Yevgeny Rouban [Thu, 8 Jun 2023 09:32:12 +0000 (16:32 +0700)]
[LoopUnrollRuntime] Allow indirect transition to deopt non-latch exit blocks

Relax condition on runtime trip count unrolling loops with 1 non-latch exit
that leads to a deop block.

There are cases when the deopt blocks are common exits for different loops.
LoopSimplify pass splits such edges to the common deopting blocks to make
sure that all exit nodes of the loop only have predecessors that are inside
of the loop (See simplifyOneLoop()). This breaks the current condition for
unrolling. This patch allows the split transitive blocks that still lead to
the deopting blocks.

Differential Revision: https://reviews.llvm.org/D152639

16 months agoRecommit [ABI] [C++20] [Modules] Don't generate vtable if the class is defined in...
Chuanqi Xu [Mon, 19 Jun 2023 02:35:16 +0000 (10:35 +0800)]
Recommit [ABI] [C++20] [Modules] Don't generate vtable if the class is defined in other module unit

Close https://github.com/llvm/llvm-project/issues/61940.

The root cause is that clang will generate vtable as strong symbol now
even if the corresponding class is defined in other module units. After
I check the wording in Itanium ABI, I find this is not inconsistent.
Itanium ABI 5.2.3
(https://itanium-cxx-abi.github.io/cxx-abi/abi.html#vague-vtable) says:

> The virtual table for a class is emitted in the same object containing
> the definition of its key function, i.e. the first non-pure virtual
> function that is not inline at the point of class definition.

So the current behavior is incorrect. This patch tries to address this.
Also I think we need to do a similar change for MSVC ABI. But I don't
find the formal wording. So I don't address this in this patch.

Reviewed By: rjmccall, iains, dblaikie

Differential Revision: https://reviews.llvm.org/D150023

16 months ago[XRay] Mark Mach-O xray_instr_map and xray_fn_idx as S_ATTR_LIVE_SUPPORT
Fangrui Song [Mon, 19 Jun 2023 02:30:16 +0000 (19:30 -0700)]
[XRay] Mark Mach-O xray_instr_map and xray_fn_idx as S_ATTR_LIVE_SUPPORT

Add the `S_ATTR_LIVE_SUPPORT` attribute to the sections so that `ld -dead_strip`
will retain subsections that reference live functions, once we we add linker
private "l" symbols as atoms.

16 months ago[RISCV] Match shl (ext v, splat 1) to vector widening add.
Jianjian GUAN [Fri, 16 Jun 2023 08:58:24 +0000 (16:58 +0800)]
[RISCV] Match shl (ext v, splat 1) to vector widening add.

Since we use match shl (v, splat 1) to vadd, we could also expand to widening add.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D153112

16 months ago[MC] flushPendingLabels: set Atom for new fragment after D71368
Fangrui Song [Mon, 19 Jun 2023 00:49:53 +0000 (17:49 -0700)]
[MC] flushPendingLabels: set Atom for new fragment after D71368

Fixes: c26c5e47ab9ca60835f191c90fa751e9a7dd0f3d (essentially a no-op)

The newly created MCDataFragment should inherit Atom (see
MCMachOStreamer::finishImpl). To the best of my knowledge, this change cannot be
tested at present, but this is important to ensure
MCExpr.cpp:AttemptToFoldSymbolOffsetDifference gives the same result in case we
evaluate the expression again with a MCAsmLayout.

In the following case,
```
.section __DATA,xray_instr_map
lxray_sleds_start1:
.space 16
Lxray_sleds_end1:
.section __DATA,xray_fn_idx
.quad (Lxray_sleds_end1-lxray_sleds_start1)>>4 // can be folded without a MCAsmLayout
```

When we have a MCAsmLayout, without this change, evaluating
(Lxray_sleds_end1-lxray_sleds_start1)>>4 again will fail due to
`FA->getAtom() == nullptr && FB.getAtom() != nullptr` in
MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl, called by
AttemptToFoldSymbolOffsetDifference.

16 months ago[MC] Fold A-B when A is a pending label or A/B are separated by a MCFillFragment
Fangrui Song [Mon, 19 Jun 2023 00:18:38 +0000 (17:18 -0700)]
[MC] Fold A-B when A is a pending label or A/B are separated by a MCFillFragment

When the MCAssembler is non-null and the MCAsmLayout is null, we can fold A-B
in these additional cases:

* when A is a pending label (will be reassigned to a real fragment in flushPendingLabels())
* A and B are separated by a MCFillFragment with a constant size

16 months ago[MC] Remove unneeded MCDataFragment check from AttemptToFoldSymbolOffsetDifference
Fangrui Song [Sun, 18 Jun 2023 23:00:18 +0000 (16:00 -0700)]
[MC] Remove unneeded MCDataFragment check from AttemptToFoldSymbolOffsetDifference

If FA == FB, we can use SA.getOffset() - SB.getOffset() even if FA is
not a MCDataFragment, as the only case this can be problematic
(different offsets for a variable-size fragment) is invalid/unreachable.
If FA != FB, the `if (FI->getKind() != MCFragment::FT_Data)` check below
can bail out correctly.

This change will help Mach-O fold more expressions. For ELF this is NFC,
unless evaluateFixup has a bug that would evaluate an expression
differently.

16 months ago[MC] flushPendingLabels: set Atom for new fragment after D71368
Fangrui Song [Sun, 18 Jun 2023 22:14:21 +0000 (15:14 -0700)]
[MC] flushPendingLabels: set Atom for new fragment after D71368

The newly created MCDataFragment should inherit Atom (see
MCMachOStreamer::finishImpl). I cannot think of a case to test the
behavior, but this is one step towards folding the Mach-O label
difference below and making Mach-O more similar to ELF.

```
.section __DATA,xray_instr_map
lxray_sleds_start1:
.space 16
Lxray_sleds_end1:
.section __DATA,xray_fn_idx
.quad (Lxray_sleds_end1-lxray_sleds_start1)>>4  // error: expected relocatable expression // Mach-O
```

16 months ago[libc] [NFC] malloc.h: fix include guard typo
Alfred Persson Forsberg [Sun, 18 Jun 2023 22:08:18 +0000 (23:08 +0100)]
[libc] [NFC] malloc.h: fix include guard typo

Differential Revision: https://reviews.llvm.org/D153231

16 months ago[XRay][test] Make tests less sensitive to .Ltmp/Ltmp label changes
Fangrui Song [Sun, 18 Jun 2023 20:32:40 +0000 (13:32 -0700)]
[XRay][test] Make tests less sensitive to .Ltmp/Ltmp label changes

16 months ago[Target] Use llvm::is_contained (NFC)
Kazu Hirata [Sun, 18 Jun 2023 19:44:00 +0000 (12:44 -0700)]
[Target] Use llvm::is_contained (NFC)

16 months ago[BOLT] Use llvm::is_contained (NFC)
Kazu Hirata [Sun, 18 Jun 2023 18:53:01 +0000 (11:53 -0700)]
[BOLT] Use llvm::is_contained (NFC)

16 months ago[AST] Use DenseMapBase::lookup (NFC)
Kazu Hirata [Sun, 18 Jun 2023 18:52:59 +0000 (11:52 -0700)]
[AST] Use DenseMapBase::lookup (NFC)

16 months ago[libc++][spaceship][NFC] P1612R2: Mark remove `operator!=` from "Ranges Library"...
Hristo Hristov [Mon, 12 Jun 2023 16:10:38 +0000 (19:10 +0300)]
[libc++][spaceship][NFC] P1612R2: Mark remove `operator!=` from "Ranges Library" items as "Complete"

Marked already implemented parts of P1612R2 as "Complete":
- `ranges::iota_view::iterator` https://reviews.llvm.org/D110774
- `iota_view::sentinel` https://reviews.llvm.org/D107396
- `filter_view::iterator` https://reviews.llvm.org/D109086
- `filter_view::sentinel` https://reviews.llvm.org/D109086
- `ranges::transform_view::iterator` https://reviews.llvm.org/D110774
- `transform_view::sentinel` https://reviews.llvm.org/D103056
- `take_view::sentinel` https://reviews.llvm.org/D123600
- `join_view::iterator` https://reviews.llvm.org/D107671
- `join_view::sentinel ` https://reviews.llvm.org/D107671
- `split_view::outer_iterator` https://reviews.llvm.org/D142063
- `split_view::inner_iterator` https://reviews.llvm.org/D142063

Note these operators were added and removed in C++20.

Reviewed By: Mordante, #libc

Differential Revision: https://reviews.llvm.org/D152721

16 months ago[Bazel][mlir] Port ee8b8d6b58e1ff45d1bcbad38c6cf458f872b38d
Pranav Kant [Sun, 18 Jun 2023 17:51:56 +0000 (17:51 +0000)]
[Bazel][mlir] Port ee8b8d6b58e1ff45d1bcbad38c6cf458f872b38d

16 months ago[MLIR] Provide bare pointer memref lowering option on gpu-to-nvvm pass
Uday Bondhugula [Sat, 17 Jun 2023 17:57:59 +0000 (23:27 +0530)]
[MLIR] Provide bare pointer memref lowering option on gpu-to-nvvm pass

Provide the bare pointer memref lowering option on gpu-to-nvvm pass.
This is needed whenever we lower memrefs on the host function side and
the kernel calls on the host-side (gpu-to-llvm) with the bare ptr
convention. The GPU module side of the lowering should also "align" and
use the bare pointer convention.

Reviewed By: krzysz00

Differential Revision: https://reviews.llvm.org/D152480

16 months agoclang/HIP: Use __builtin_fmaf16
Matt Arsenault [Tue, 22 Nov 2022 17:51:21 +0000 (12:51 -0500)]
clang/HIP: Use __builtin_fmaf16

Missed these in 43fd46fda3c90b014e8a73c62f67af9543ea4d59

16 months ago[Doc] Fix table layout
Serge Pavlov [Sun, 18 Jun 2023 16:46:08 +0000 (23:46 +0700)]
[Doc] Fix table layout

16 months ago[GlobalIsel][X86] selectDivRem - fix typo in 64-bit AH handling code
Simon Pilgrim [Sun, 18 Jun 2023 16:37:17 +0000 (17:37 +0100)]
[GlobalIsel][X86] selectDivRem - fix typo in 64-bit AH handling code

This function was lifted from fast-isel, and still referred to the Instruction::SRem/URrem opcodes, instead of the G_SREM/G_UREM opcodes.

But it turns out these aren't necessary at all as only the G_SREM/G_UREM codepaths will use the AH register for DivRemResultReg anyhow.

16 months ago[GlobalIsel][X86] Regenerate srem/urem select test coverage
Simon Pilgrim [Sun, 18 Jun 2023 16:06:32 +0000 (17:06 +0100)]
[GlobalIsel][X86] Regenerate srem/urem select test coverage

16 months ago[clang] Add __builtin_isfpclass
Serge Pavlov [Sun, 18 Jun 2023 15:53:32 +0000 (22:53 +0700)]
[clang] Add __builtin_isfpclass

A new builtin function __builtin_isfpclass is added. It is called as:

    __builtin_isfpclass(<floating point value>, <test>)

and returns an integer value, which is non-zero if the floating point
argument falls into one of the classes specified by the second argument,
and zero otherwise. The set of classes is an integer value, where each
value class is represented by a bit. There are ten data classes, as
defined by the IEEE-754 standard, they are represented by bits:

    0x0001 (__FPCLASS_SNAN)         - Signaling NaN
    0x0002 (__FPCLASS_QNAN)         - Quiet NaN
    0x0004 (__FPCLASS_NEGINF)       - Negative infinity
    0x0008 (__FPCLASS_NEGNORMAL)    - Negative normal
    0x0010 (__FPCLASS_NEGSUBNORMAL) - Negative subnormal
    0x0020 (__FPCLASS_NEGZERO)      - Negative zero
    0x0040 (__FPCLASS_POSZERO)      - Positive zero
    0x0080 (__FPCLASS_POSSUBNORMAL) - Positive subnormal
    0x0100 (__FPCLASS_POSNORMAL)    - Positive normal
    0x0200 (__FPCLASS_POSINF)       - Positive infinity

They have corresponding builtin macros to facilitate using the builtin
function:

    if (__builtin_isfpclass(x, __FPCLASS_NEGZERO | __FPCLASS_POSZERO) {
      // x is any zero.
    }

The data class encoding is identical to that used in llvm.is.fpclass
function.

Differential Revision: https://reviews.llvm.org/D152351

16 months ago[CodeGenPrepare][RISCV] Remove asserting VH references before erasing the dead GEP
Yingwei Zheng [Sun, 18 Jun 2023 15:40:19 +0000 (23:40 +0800)]
[CodeGenPrepare][RISCV] Remove asserting VH references before erasing the dead GEP

Fixes issue https://github.com/llvm/llvm-project/issues/63365

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D153194

16 months ago[SCCP][NFC] Regenerate test case
luxufan [Sun, 18 Jun 2023 13:52:19 +0000 (21:52 +0800)]
[SCCP][NFC] Regenerate test case

16 months ago[mlir][math] Uplift from arith to math.fma
Ivan Butygin [Sat, 10 Jun 2023 20:59:24 +0000 (22:59 +0200)]
[mlir][math] Uplift from arith to math.fma

Add pass to uplift from arith mulf + addf ops to math.fma if fastmath flags allow it.

Differential Revision: https://reviews.llvm.org/D152633

16 months ago[X86] Regenerate tls.ll and reuse common linux check prefixes
Simon Pilgrim [Sun, 18 Jun 2023 14:24:44 +0000 (15:24 +0100)]
[X86] Regenerate tls.ll and reuse common linux check prefixes

16 months ago[X86] Regenerate add32ri8.ll
Simon Pilgrim [Sun, 18 Jun 2023 14:08:44 +0000 (15:08 +0100)]
[X86] Regenerate add32ri8.ll

16 months ago[SVE][AArch64TTI] Fix invalid mla combine that miscomputes the value of inactive...
Paul Walker [Sat, 17 Jun 2023 16:51:49 +0000 (17:51 +0100)]
[SVE][AArch64TTI] Fix invalid mla combine that miscomputes the value of inactive lanes.

Consider: add(pg, a, mul_u(pg, b, c))

Although the multiply's inactive lanes are undefined, they don't
contribute to the final result.  The overall result of the inactive
lanes come from "a" and thus the above is another form of mla
rather than mla_u.

16 months ago[NFC][AArch64TTI] Breakout add/sub combines into discrete functions.
Paul Walker [Sat, 17 Jun 2023 15:48:09 +0000 (16:48 +0100)]
[NFC][AArch64TTI] Breakout add/sub combines into discrete functions.

16 months agoIncrease test coverage of Transforms/InstCombine/AArch64/sve-intrinsic-muladdsub.ll
Paul Walker [Sat, 17 Jun 2023 15:19:22 +0000 (16:19 +0100)]
Increase test coverage of Transforms/InstCombine/AArch64/sve-intrinsic-muladdsub.ll

16 months ago[clang-tidy] Improve `performance-move-const-arg` message when no move constructor...
AMS21 [Sun, 18 Jun 2023 11:40:47 +0000 (11:40 +0000)]
[clang-tidy] Improve `performance-move-const-arg` message when no move constructor is available

We now display a simple note if the reason is that the used class does not
support move semantics.

This fixes llvm#62550

Reviewed By: PiotrZSL

Differential Revision: https://reviews.llvm.org/D153220

16 months ago[clang-tidy] Fix `llvmlibc-inline-function-decl` false positives for templated functi...
AMS21 [Sun, 18 Jun 2023 11:40:32 +0000 (11:40 +0000)]
[clang-tidy] Fix `llvmlibc-inline-function-decl` false positives for templated function definitions

For a declaration the `FunctionDecl` begin location does not include the
template parameter lists, but for some reason if you have a separate
definitions to the declaration the begin location does include them.
With this patch we now correctly handle that case.

This fixes llvm#62746

Reviewed By: PiotrZSL

Differential Revision: https://reviews.llvm.org/D153218

16 months ago[gn build] Port 845618cf69e8
LLVM GN Syncbot [Sun, 18 Jun 2023 07:30:50 +0000 (07:30 +0000)]
[gn build] Port 845618cf69e8

16 months ago[clang-tidy] Refactor common code from the Noexcept*Checks into `NoexceptFunctionCheck`
AMS21 [Sun, 18 Jun 2023 06:50:05 +0000 (06:50 +0000)]
[clang-tidy] Refactor common code from the Noexcept*Checks into `NoexceptFunctionCheck`

As discussed in the https://reviews.llvm.org/D148697 review.

Reviewed By: PiotrZSL

Differential Revision: https://reviews.llvm.org/D153198

16 months agopr62660-normalization-failure.ll REQUIRES: asserts (#62660)
NAKAMURA Takumi [Sun, 18 Jun 2023 06:24:53 +0000 (15:24 +0900)]
pr62660-normalization-failure.ll REQUIRES: asserts (#62660)

16 months ago[clang] Replace uses of CGBuilderTy::CreateElementBitCast (NFC)
Youngsuk Kim [Sun, 18 Jun 2023 01:12:08 +0000 (04:12 +0300)]
[clang] Replace uses of CGBuilderTy::CreateElementBitCast (NFC)

* Add `Address::withElementType()` as a replacement for
  `CGBuilderTy::CreateElementBitCast`.

* Partial progress towards replacing `CreateElementBitCast`, as it no
  longer does what its name suggests. Either replace its uses with
  `Address::withElementType()`, or remove them if no longer needed.

* Remove unused parameter 'Name' of `CreateElementBitCast`

Reviewed By: barannikov88, nikic

Differential Revision: https://reviews.llvm.org/D153196

16 months ago[Hexagon] Add missing patterns for boolean [v]selects
Krzysztof Parzyszek [Sun, 18 Jun 2023 00:14:39 +0000 (17:14 -0700)]
[Hexagon] Add missing patterns for boolean [v]selects

Fixes https://github.com/llvm/llvm-project/issues/59663

16 months ago[Hexagon] Handle all compares of i1 and vNi1
Krzysztof Parzyszek [Sat, 17 Jun 2023 22:41:53 +0000 (15:41 -0700)]
[Hexagon] Handle all compares of i1 and vNi1

Fixes https://github.com/llvm/llvm-project/issues/63363

16 months ago[Hexagon] Add missing patterns for truncate to vNi1
Krzysztof Parzyszek [Sat, 17 Jun 2023 22:37:24 +0000 (15:37 -0700)]
[Hexagon] Add missing patterns for truncate to vNi1

16 months ago[gn build] Port cea4285949b5
LLVM GN Syncbot [Sat, 17 Jun 2023 23:02:48 +0000 (23:02 +0000)]
[gn build] Port cea4285949b5

16 months ago[gn] fix build after 992cb98462ab
Nico Weber [Sat, 17 Jun 2023 23:02:16 +0000 (19:02 -0400)]
[gn] fix build after 992cb98462ab

16 months ago[Pseudo Probe] Do not place functions in nodeduplicate COMDATs
Fangrui Song [Sat, 17 Jun 2023 22:40:19 +0000 (15:40 -0700)]
[Pseudo Probe] Do not place functions in nodeduplicate COMDATs

For a function not in an IR COMDAT, currently we place it into a nodeduplicate IR
COMDAT so that its text section and its associated .pseudo_probe section will be
in the same section group, which can be retained or discarded by the linker as a
unit. However, the section group wastes space.

After D153189 uses SHF_LINK_ORDER to ensure a .pseudo_probe section will be
discarded when its associated text section is discarded, we can remove the
nodeduplicate IR change.

In the following example, the .pseudo_probe associated with .text.f is discarded as expected.
```
clang -c -ffunction-sections -fpseudo-probe-for-profiling -xc =(printf 'void _start(){} void f(){}') -o a.o
ld.lld --gc-sections --print-gc-sections a.o
```

Reviewed By: hoy

Differential Revision: https://reviews.llvm.org/D153191

16 months ago[clang-format][NFC] Remove redundant getLLVMStyle() in unit tests
Owen Pan [Sat, 17 Jun 2023 22:19:01 +0000 (15:19 -0700)]
[clang-format][NFC] Remove redundant getLLVMStyle() in unit tests

16 months ago[Support/ELF] - Add OpenBSD PT_OPENBSD_NOBTCFI constant.
Brad Smith [Sat, 17 Jun 2023 21:38:28 +0000 (17:38 -0400)]
[Support/ELF] - Add OpenBSD PT_OPENBSD_NOBTCFI constant.

OpenBSD commit for reference:
https://github.com/openbsd/src/commit/7b407c478fab53a6d9a091887c828c3f7b3f8b46

16 months ago[LSR] Enable SCEV verification for test from f3a0ad2d and mark as XFAIL
Florian Hahn [Sat, 17 Jun 2023 20:06:21 +0000 (21:06 +0100)]
[LSR] Enable SCEV verification for test from f3a0ad2d and mark as XFAIL

The test fails SCEV verification, which cause the expensive check bots
to fail. Always run verification and mark as XFAIL until fixed.

16 months agoPPCAsmParser: Use parseOptionalToken
Fangrui Song [Sat, 17 Jun 2023 19:34:49 +0000 (12:34 -0700)]
PPCAsmParser: Use parseOptionalToken

to simplify code near __tls_get_addr parsing.

16 months ago[MLIR] Add support for bare pointer calling convention in gpu-to-llvm
Uday Bondhugula [Sat, 17 Jun 2023 17:13:43 +0000 (22:43 +0530)]
[MLIR] Add support for bare pointer calling convention in gpu-to-llvm

Add support for the bare pointer calling convention in the gpu-to-llvm
pass. This wasn't being exposed and is needed when GPU-compiled MLIR is
to be called with this convention.

Reviewed By: krzysz00

Differential Revision: https://reviews.llvm.org/D152477

16 months agoRevert "[LSR] Consider post-inc form when creating extends/truncates."
Florian Hahn [Sat, 17 Jun 2023 16:58:41 +0000 (17:58 +0100)]
Revert "[LSR] Consider post-inc form when creating extends/truncates."

This reverts commit abfeda5af329b5889db709ff74506e20e0b569e9.
and fe19036e1266d2a90b44725c82b898134906e4c3.

The added assertion triggers during clang bootstrap builds. Revert while
I investigate.

16 months ago[mlir][arith] Remove unused ODS class
Jeff Niu [Sat, 17 Jun 2023 16:54:13 +0000 (09:54 -0700)]
[mlir][arith] Remove unused ODS class

Differential Revision: https://reviews.llvm.org/D153203

16 months ago[LSR] Add test for #62660.
Florian Hahn [Sat, 17 Jun 2023 16:37:25 +0000 (17:37 +0100)]
[LSR] Add test for #62660.

Add test for LSR miscompile.

16 months ago[Coding style] Fix incorrect link syntax
Felipe de Azevedo Piovezan [Sat, 17 Jun 2023 12:26:05 +0000 (08:26 -0400)]
[Coding style] Fix incorrect link syntax

16 months ago[libc++][NFC] Granularise <thread> header
Hui [Wed, 31 May 2023 09:49:56 +0000 (10:49 +0100)]
[libc++][NFC] Granularise <thread> header

- This was to make implementing jthread easier and requested in https://reviews.llvm.org/D151559

Differential Revision: https://reviews.llvm.org/D151792

16 months ago[docs] Fix link for static constructors article
Prajwal S N [Sat, 17 Jun 2023 10:02:32 +0000 (15:32 +0530)]
[docs] Fix link for static constructors article

It was previously present in the inline code block and did not work as a
hyperlink.

Reviewed By: yassingh

Differential Revision: https://reviews.llvm.org/D153061

16 months ago[AMDGPU] Update test after abfeda5af329b58.
Florian Hahn [Sat, 17 Jun 2023 09:15:15 +0000 (10:15 +0100)]
[AMDGPU] Update test after abfeda5af329b58.

16 months ago[LSR] Consider post-inc form when creating extends/truncates.
Florian Hahn [Sat, 17 Jun 2023 08:58:37 +0000 (09:58 +0100)]
[LSR] Consider post-inc form when creating extends/truncates.

GenerateTruncates at the moment creates extends/truncates for post-inc
uses of normalized expressions. For example, if an add rec of the form
{1,+,-1} is used outside the loop, the normalized form will use {1,+,-1}
instead of {0,+,-1}. When naively sign-extending the normalized
expression, it will get extended incorrectly to {1,+,-1} for the wider
type, if the backedge-taken count of the loop is 1.

To address this, the patch updates GenerateTruncates to check if the
LSRUse contains any fixups with PostIncLoops. If that's the case, first
de-normalize the expression, then perform the extend/truncate, then
normalize again.

There may be other places where similar checks are needed and the helper
can be generalized for those cases. I'd not be surprised if other subtle
mis-compiles are caused by this.

Fixes #38847.
Fixes #58039.
Fixes #62852.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D153004

16 months ago[LSR] Add test case for #58039.
Florian Hahn [Sat, 17 Jun 2023 08:56:59 +0000 (09:56 +0100)]
[LSR] Add test case for #58039.

16 months ago[Clang] Add the list of core papers approved in Varna to the status page
Corentin Jabot [Sat, 17 Jun 2023 05:34:42 +0000 (08:34 +0300)]
[Clang] Add the list of core papers approved in Varna to the status page

16 months ago[Pseudo Probe] Make .pseudo_probe GC-able
Fangrui Song [Sat, 17 Jun 2023 06:46:36 +0000 (23:46 -0700)]
[Pseudo Probe] Make .pseudo_probe GC-able

* Add the SHF_LINK_ORDER flag so that the .pseudo_probe section is discarded when the associated text section is discarded.
* Add unique ID so that with `clang -ffunction-sections -fno-unique-section-names`, there is one separate .pseudo_probe for each text section (disambiguated by `.section ....,unique,id` in assembly)

The changes allow .pseudo_probe GC even if we don't place instrumented functions
in an IR comdat (see `getOrCreateFunctionComdat` in SampleProfileProbe.cpp).

Reviewed By: hoy

Differential Revision: https://reviews.llvm.org/D153189

16 months ago[AMDGPU] Generate checks for load-constant tests
Jay Foad [Fri, 16 Jun 2023 13:49:19 +0000 (14:49 +0100)]
[AMDGPU] Generate checks for load-constant tests

Differential Revision: https://reviews.llvm.org/D153139

16 months ago[bazel] Fix clang after D148094
Fangrui Song [Sat, 17 Jun 2023 05:19:32 +0000 (22:19 -0700)]
[bazel] Fix clang after D148094

16 months ago[clang][CodeGen] Break up TargetInfo.cpp [8/8]
Sergei Barannikov [Tue, 9 May 2023 15:55:59 +0000 (18:55 +0300)]
[clang][CodeGen] Break up TargetInfo.cpp [8/8]

This commit breaks up CodeGen/TargetInfo.cpp into a set of *.cpp files,
one file per target. There are no functional changes, mostly just code moving.

Non-code-moving changes are:
* A virtual destructor has been added to DefaultABIInfo to pin the vtable to a cpp file.
* A few methods of ABIInfo and DefaultABIInfo were split into declaration + definition
  in order to reduce the number of transitive includes.
* Several functions that used to be static have been placed in clang::CodeGen
  namespace so that they can be accessed from other cpp files.

RFC: https://discourse.llvm.org/t/rfc-splitting-clangs-targetinfo-cpp/69883

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D148094

16 months ago[clang][CodeGen] Break up TargetInfo.cpp [7/8]
Sergei Barannikov [Tue, 9 May 2023 15:41:05 +0000 (18:41 +0300)]
[clang][CodeGen] Break up TargetInfo.cpp [7/8]

Wrap calls to XXXTargetCodeGenInfo constructors into factory functions.
This allows moving implementations of TargetCodeGenInfo to dedicated cpp
files without a change.

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D150215

16 months ago[NFC] Fix potential dereferencing of null return value.
Sindhu Chittireddy [Tue, 13 Jun 2023 21:37:58 +0000 (14:37 -0700)]
[NFC] Fix potential dereferencing of null return value.

Replace getAs with castAs and add assert if needed.
Differential Revision: https://reviews.llvm.org/D152977

16 months ago[Bazel][mlir] Port 120cd5aafca2ccc8b04d208598e817ffc6d0b869
Pranav Kant [Sat, 17 Jun 2023 02:57:46 +0000 (02:57 +0000)]
[Bazel][mlir] Port 120cd5aafca2ccc8b04d208598e817ffc6d0b869

16 months ago[LoongArch] Fix handling of the chain of CSRWR and CSRXCHG nodes
Weining Lu [Sat, 17 Jun 2023 01:46:40 +0000 (09:46 +0800)]
[LoongArch] Fix handling of the chain of CSRWR and CSRXCHG nodes

`LoongArchISD::CSRWR` has two results. The first is the result of
`loongarch.csrwr.[wd]` intrinsic and the second is the chain. But
currently the chain is not processed correctly when creating this
node, resulting in the `csrwr` instruction being optimized out when
the result is not used by anyone [1]. `LoongArchISD::CSRXCHG` has
the same issue.

This patch addresses this issue.

[1]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/loongarch/include/asm/loongarch.h?h=v6.4-rc6#n219

Reviewed By: hev

Differential Revision: https://reviews.llvm.org/D153120

16 months ago[LoongArch][NFC] Precommit test for D153120 (the fix of CSRWR and CSRXCHG)
Weining Lu [Sat, 17 Jun 2023 01:46:29 +0000 (09:46 +0800)]
[LoongArch][NFC] Precommit test for D153120 (the fix of CSRWR and CSRXCHG)

Reviewed By: xry111

Differential Revision: https://reviews.llvm.org/D153119

16 months ago[MLIR] Register all extensions in CAPI's RegisterEverything
Ashay Rane [Fri, 16 Jun 2023 22:11:43 +0000 (17:11 -0500)]
[MLIR] Register all extensions in CAPI's RegisterEverything

The patch for promised interfaces (a5ef51d7) doesn't register all
extensions in the CAPI's `mlirRegisterAllDialects()` function.  This is
used by the MLIR Python bindings, causing downstream users of the Python
bindings to terminate abruptly.  This patch adds the call to register
all extensions.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D153174

16 months ago[RISCV] Fold special case (xor (setcc constant, y, setlt), 1) -> (setcc y, constant...
LiaoChunyu [Sat, 17 Jun 2023 01:33:48 +0000 (09:33 +0800)]
[RISCV] Fold special case (xor (setcc constant, y, setlt), 1) -> (setcc y, constant + 1, setlt)

Improve D151719.
(xor (setcc constant, y, setlt), 1) -> (setcc y, constant + 1, setlt)
https://alive2.llvm.org/ce/z/BZNEia

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D152128

16 months agoclang/AMDGPU: Emit atomicrmw for atomic_inc/dec builtins
Matt Arsenault [Sat, 5 Nov 2022 20:07:49 +0000 (13:07 -0700)]
clang/AMDGPU: Emit atomicrmw for atomic_inc/dec builtins

This makes the scope and ordering arguments actually do something.
Also add some new OpenCL tests since the existing HIP tests didn't
cover address spaces.

16 months ago[MC] Restore a special case to support limited A-B folding when A/B are in the same...
Fangrui Song [Sat, 17 Jun 2023 00:08:58 +0000 (17:08 -0700)]
[MC] Restore a special case to support limited A-B folding when A/B are in the same fragment being laided out

Add subsection-if.s to test what we can fold (in the same fragment) and what we cannot.

Fix https://github.com/ClangBuiltLinux/linux/issues/1876
Fixes: 4bdc7f7a331f82cca1637388cf68bdc5b32ab43b

16 months agoReland [clang-format] Fix overlapping whitespace replacements before PPDirective
Owen Pan [Wed, 14 Jun 2023 22:06:17 +0000 (15:06 -0700)]
Reland [clang-format] Fix overlapping whitespace replacements before PPDirective

If the first token of an annotated line already has a computed Newlines,
reuse it to avoid potential overlapping whitespace replacements before
preprocessor branching directives.

Fixes #62892.

Differential Revision: https://reviews.llvm.org/D151954

16 months ago[Fuchsia] Forward libedit flags to stage2
Daniel Thornburgh [Fri, 16 Jun 2023 23:54:43 +0000 (16:54 -0700)]
[Fuchsia] Forward libedit flags to stage2

16 months ago[RISCV] Fix a latent miscompile in doPeepholeMaskedRVV
Philip Reames [Fri, 16 Jun 2023 23:47:39 +0000 (16:47 -0700)]
[RISCV] Fix a latent miscompile in doPeepholeMaskedRVV

The code was using the tail policy being "agnostic" to select a instruction whose semantics were "undefined". This was almost always fine (as the pass through operand was usually implicit_def), but could in theory lead to a miscompile. I don't actually have a test case as it requires a later transform to exploit the wrong tail policy state, and I couldn't easily figure out to get vsetvli insertion to miscompile given the wrong state. This was spotted by inspection, and it may be a miscompile in theory only at the moment.

Note that this may cause regressions if there are instructions for which we either don't have a _TU pseudo form, or the _TU pseudo form is missing a policy operand. When I was first looking at this, I saw exactly that, and D153067 exists to add the missing policy operand I noticed.

As a later follow up, I want to always force the use of _TU, but it seemed good to fix the bug, then driven the _TU transition in a separate patch.

Differential Revision: https://reviews.llvm.org/D153070

16 months ago[RISCV] Add a policy operand to VPseudoBinaryNoMaskTU [NFC]
Philip Reames [Fri, 16 Jun 2023 23:41:09 +0000 (16:41 -0700)]
[RISCV] Add a policy operand to VPseudoBinaryNoMaskTU [NFC]

This change adds a policy operand to the helper class which is used for binary ops like vadd, but also, possibly surprisingly, some of the vslide variants. This allows us to represent the tail agnostic state with this pseudo family - previously, we could only represent tail undefined and tail undisturbed. (Since these don't have a mask, they're always mask undefined.)

This is NFC because no current producer uses the tail agnostic state. This will change in an upcoming change to doPeepholeMaskedRVV.

Differential Revision: https://reviews.llvm.org/D153067

16 months ago[mlir][Vector] Fix 0-D tensor vectorization in Linalg
Diego Caballero [Fri, 16 Jun 2023 23:21:24 +0000 (23:21 +0000)]
[mlir][Vector] Fix 0-D tensor vectorization in Linalg

It looks like scalable vector support broke vectorization for 0-D
tensors and we didn't have any test coverting that case. This patch
provides a fix and a test.

Differential Revision: https://reviews.llvm.org/D153181

16 months ago[clang-format][NFC] Use verifyGoogleFormat in FormatTest.cpp
Owen Pan [Fri, 16 Jun 2023 23:33:36 +0000 (16:33 -0700)]
[clang-format][NFC] Use verifyGoogleFormat in FormatTest.cpp

Replaces verifyFormat(..., getGoogleStyle()) with
verifyGoogleFormat(...) in FormatTest.cpp.

16 months ago[RISCV] Reduce alignment for __attribute__((riscv_rvv_vector_bits)) for LMUL<1 types.
Craig Topper [Fri, 16 Jun 2023 23:40:10 +0000 (16:40 -0700)]
[RISCV] Reduce alignment for __attribute__((riscv_rvv_vector_bits)) for LMUL<1 types.

Don't use an alignment larger than the vector size.

16 months ago[RISCV] Refactor vecPolicyOp skip logic in doPeepholeMaskedRVV. NFC
Luke Lau [Fri, 16 Jun 2023 23:31:35 +0000 (16:31 -0700)]
[RISCV] Refactor vecPolicyOp skip logic in doPeepholeMaskedRVV. NFC

We can just explicitly check if the new unmasked pseudo takes a policy
op, rather than implicitly relying on I->UnmaskedTUPseudo ==
I->UnmaskedPseudo. Split out from another patch to make the diff more
readable.

Differential Revision: https://reviews.llvm.org/D152961

16 months ago[RISCV] Reuse RISCVDAGToDAGISel member TTI in doPeepholeMaskedRVV. NFC
Luke Lau [Fri, 16 Jun 2023 23:30:38 +0000 (16:30 -0700)]
[RISCV] Reuse RISCVDAGToDAGISel member TTI in doPeepholeMaskedRVV. NFC

Differential Revision: https://reviews.llvm.org/D152960

16 months ago[Fuchsia] Forward terminfo flags to stage 2
Daniel Thornburgh [Fri, 16 Jun 2023 23:34:26 +0000 (16:34 -0700)]
[Fuchsia] Forward terminfo flags to stage 2

16 months ago[RISCV] Make all vector binops use the _TU pseudo form
Philip Reames [Fri, 16 Jun 2023 23:13:22 +0000 (16:13 -0700)]
[RISCV] Make all vector binops use the _TU pseudo form

This continues towards the goal spelled out in https://discourse.llvm.org/t/riscv-transition-in-vector-pseudo-structure-policy-variants/71295. This patch switches all the binary operations (no widen, no narrow, but both int and FP) to use the _TU + implicit_def passthrough form.  Change is mechanical.

This only changes the unmasked variants. Masked variants will still go through doPeepholeMaskedRVV and end up in the unsuffixed/TA form. Fixing that will be a separate change.

Differential Revision: https://reviews.llvm.org/D152940

16 months ago[RISCV] Introduce RISCVISD::VWMACC(U/SU)_VL opcode
Nitin John Raj [Thu, 15 Jun 2023 02:10:44 +0000 (19:10 -0700)]
[RISCV] Introduce RISCVISD::VWMACC(U/SU)_VL opcode

Differential Revision: https://reviews.llvm.org/D153057

16 months ago[clang-format][NFC] Clean up unit tests
Owen Pan [Fri, 16 Jun 2023 07:00:43 +0000 (00:00 -0700)]
[clang-format][NFC] Clean up unit tests

This patch adds a verifyNoChange macro to verify code that won't
change after being formatted. (The code will not be messed up before
being formatted.) It then replaces EXPECT_EQ with verifyFormat
wherever applicable so that the code will be messed up before being
formatted. When the replacement fails the unit test, verifyFormat is
replaced with verifyNoChange.

Differential Revision: https://reviews.llvm.org/D153109

16 months agoTTI: Add function to hasBranchDivergence
Matt Arsenault [Thu, 1 Jun 2023 22:11:24 +0000 (18:11 -0400)]
TTI: Add function to hasBranchDivergence

It my be possible to contextually ignore divergence in a function if
it's known to run single threaded.

16 months agoUniformityAnalysis: Skip computation with no branch divergence
Matt Arsenault [Fri, 2 Jun 2023 10:58:13 +0000 (06:58 -0400)]
UniformityAnalysis: Skip computation with no branch divergence

Check TTI before bothering to run the computation. Everything
will be assumed uniform by default.

16 months agoPrevent out of range fixup encoding on AArch64
Daniel Hoekwater [Tue, 13 Jun 2023 17:48:19 +0000 (10:48 -0700)]
Prevent out of range fixup encoding on AArch64

The range of a 21-bit signed integer is [-1048576, 1048575],
not [-2097152, 2097151].

Differential Revision: https://reviews.llvm.org/D152841

16 months ago[Hexagon] Generate correct instruction for store i1, ($Rs<<$u2 + $Rt)
Krzysztof Parzyszek [Fri, 16 Jun 2023 22:20:31 +0000 (15:20 -0700)]
[Hexagon] Generate correct instruction for store i1, ($Rs<<$u2 + $Rt)

Fixes https://github.com/llvm/llvm-project/issues/63359

16 months agoRevert "AMDGPU: Drop and auto-upgrade llvm.amdgcn.ldexp to llvm.ldexp"
Matt Arsenault [Fri, 16 Jun 2023 22:13:07 +0000 (18:13 -0400)]
Revert "AMDGPU: Drop and auto-upgrade llvm.amdgcn.ldexp to llvm.ldexp"

This reverts commit 1159c670d40e3ef302264c681fe7e0268a550874.

Accidentally pushed wrong patch

16 months agoAMDGPU: Drop and auto-upgrade llvm.amdgcn.ldexp to llvm.ldexp
Matt Arsenault [Fri, 28 Apr 2023 17:41:05 +0000 (13:41 -0400)]
AMDGPU: Drop and auto-upgrade llvm.amdgcn.ldexp to llvm.ldexp

16 months ago[lldb] Fix Python test formatting (NFC)
Jonas Devlieghere [Fri, 16 Jun 2023 21:48:37 +0000 (14:48 -0700)]
[lldb] Fix Python test formatting (NFC)

All Python files in the LLVM repository were reformatted with Black [1].
Files inside the LLDB subproject were reformatted in 2238dcc39358. This
patch updates a handful of tests that were added or modified since then
and weren't formatted with Black.

[1] https://discourse.llvm.org/t/rfc-document-and-standardize-python-code-style/68257

16 months agoFix diag for read-only target features
Yaxun (Sam) Liu [Fri, 16 Jun 2023 04:36:00 +0000 (00:36 -0400)]
Fix diag for read-only target features

Currently the diag is emitted even when there is no
target feature specified on command line for OpenMP.
This is because the function to initialize feature map
is also used with cached feature string. The fix is to
only diag when the feature map is initialized with
feature strings from command line options.

Reviewed by: Joseph Huber, Matt Arsenault, Johannes Doerfert

Differential Revision: https://reviews.llvm.org/D153123

16 months ago[lldb] Simplify logging in Process settings (NFC)
Jonas Devlieghere [Fri, 16 Jun 2023 21:41:00 +0000 (14:41 -0700)]
[lldb] Simplify logging in Process settings (NFC)

16 months ago[lldb][NFCI] Avoid a few unnecessary ConstString constructions in StructuredDataDarwinLog
Alex Langford [Fri, 16 Jun 2023 21:17:32 +0000 (14:17 -0700)]
[lldb][NFCI] Avoid a few unnecessary ConstString constructions in StructuredDataDarwinLog

Process::GetStructuredDataPlugin takes a StringRef directly, no need to
convert them to ConstString first.

16 months ago[dsymutil] Update DWARF 5 tests after 8119ab9bf78b
Jonas Devlieghere [Fri, 16 Jun 2023 20:57:20 +0000 (13:57 -0700)]
[dsymutil] Update DWARF 5 tests after 8119ab9bf78b

Update test affected by 8119ab9bf78b and address Adrian's post commit
review feedback.

16 months agoRevert "Revert "ValueTracking: Fix nan result handling for fmul""
Arthur Eubanks [Fri, 16 Jun 2023 20:48:27 +0000 (13:48 -0700)]
Revert "Revert "ValueTracking: Fix nan result handling for fmul""

This reverts commit 464dcab8a6c823c9cb462bf4107797b8173de088.

Going to fix forward size regression instead due to more dependent patches needing to be reverted otherwise.

16 months agoRevert "Revert "clang: Update tests after InstSimplify change""
Arthur Eubanks [Fri, 16 Jun 2023 20:48:20 +0000 (13:48 -0700)]
Revert "Revert "clang: Update tests after InstSimplify change""

This reverts commit dd1c79b03004ae88ebead51aa021ab229ab0175c.

Going to fix forward size regression instead due to more dependent patches needing to be reverted otherwise.

16 months agoRevert "Revert "InstSimplify: Remove null parent checks""
Arthur Eubanks [Fri, 16 Jun 2023 20:48:08 +0000 (13:48 -0700)]
Revert "Revert "InstSimplify: Remove null parent checks""

This reverts commit 935c8b6f3a4dda0ff881ed86faaad9fe5b276d70.

Going to fix forward size regression instead due to more dependent patches needing to be reverted otherwise.

16 months agoRevert "Revert "InstSimplify: Require instruction be parented""
Arthur Eubanks [Fri, 16 Jun 2023 20:47:30 +0000 (13:47 -0700)]
Revert "Revert "InstSimplify: Require instruction be parented""

This reverts commit 0c03f48480f69b854f86d31235425b5cb71ac921.

Going to fix forward size regression instead due to more dependent patches needing to be reverted otherwise.

16 months ago[scudo] Add mallopt to print stats to the log.
Christopher Ferris [Fri, 16 Jun 2023 01:52:59 +0000 (18:52 -0700)]
[scudo] Add mallopt to print stats to the log.

Reviewed By: Chia-hungDuan

Differential Revision: https://reviews.llvm.org/D153094