[all-commits] [llvm/llvm-project] ef4a19: Reland "[AMDGPU] PromoteAlloca: flatten homogeneou...

Louis Dionne via All-commits all-commits at lists.llvm.org
Wed Sep 30 10:21:23 PDT 2026


  Branch: refs/heads/users/ldionne/libcxx-benchmarks-inline-build
  Home:   https://github.com/llvm/llvm-project
  Commit: ef4a19d2cd7ae5d5e42003c6bc239af912eed611
      https://github.com/llvm/llvm-project/commit/ef4a19d2cd7ae5d5e42003c6bc239af912eed611
  Author: Domenic Nutile <domenic.nutile at gmail.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
    M llvm/test/CodeGen/AMDGPU/eliminate-frame-index-select.ll
    A llvm/test/CodeGen/AMDGPU/promote-alloca-homogeneous-struct.ll

  Log Message:
  -----------
  Reland "[AMDGPU] PromoteAlloca: flatten homogeneous structs to vectors" (#221058) (#223790)

This relands #217055

The original commit revealed a latent issue in eliminateFrameIndex in
SIRegisterInfo where SCC can be clobbered before reading it on
gfx900/gfx90a. This change itself has no known issues.


  Commit: 552ea17e0a93262e78813994c0abc6cea23bfb55
      https://github.com/llvm/llvm-project/commit/552ea17e0a93262e78813994c0abc6cea23bfb55
  Author: Ömer Sinan Ağacan <omer at osa1.net>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M clang/test/CodeGen/isfpclass.c
    M clang/test/Headers/__clang_hip_math.hip
    M clang/test/Headers/wasm.c
    M llvm/lib/Analysis/ValueTracking.cpp
    M llvm/test/Transforms/SimplifyCFG/switch-range-to-icmp.ll
    M llvm/unittests/Analysis/ValueTrackingTest.cpp

  Log Message:
  -----------
  [ValueTracking] Handle zext/sext in computeConstantRange (#227726)


  Commit: 24df73cb7311d9a66d132c7fcd20feac2fbae36d
      https://github.com/llvm/llvm-project/commit/24df73cb7311d9a66d132c7fcd20feac2fbae36d
  Author: Anshul Nigham <nigham at google.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M utils/bazel/llvm-project-overlay/llvm/BUILD.bazel

  Log Message:
  -----------
  [Bazel] Fixes commit 4f9feb85267c08ec29ea8bb2f80a3d0cd7ab925f (#227767)

Adds missing dependency


  Commit: 0a5b663d6474a958ef953cd6615de550957b8525
      https://github.com/llvm/llvm-project/commit/0a5b663d6474a958ef953cd6615de550957b8525
  Author: Farzon Lotfi <farzonlotfi at microsoft.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M clang/lib/CodeGen/CGExpr.cpp
    M clang/lib/CodeGen/CGExprScalar.cpp
    M clang/test/CodeGenHLSL/BasicFeatures/MatrixElementRowColFlags.hlsl
    M clang/test/CodeGenHLSL/BasicFeatures/MatrixElementTypeCast.hlsl
    M clang/test/CodeGenHLSL/matrix-layout-attr-overrides-default.hlsl

  Log Message:
  -----------
  [Clang][HLSL] Keep matrix SSA values in column-major order (#227087)

Keep matrix register values in canonical column-major order while
preserving
the declared matrix layout for memory accesses.

Construct elementwise matrix casts in column-major lane order and use
column-major indexing when extracting elements from SSA values. Continue
to
use layout-aware indexing for LValues and other physical memory
operations.

Materialize matrix prvalues through the matrix-aware scalar store path
so
row-major temporaries are transposed at the register-to-memory boundary.

Add coverage for matrix indexing, casts, swizzles, transposed prvalues,
and
explicit row-major out-parameter accesses.

Assisted by Copilot via GPT 5.6-sol


  Commit: 7e89fb5d646ef2d4b18520a8300818b5b94ef587
      https://github.com/llvm/llvm-project/commit/7e89fb5d646ef2d4b18520a8300818b5b94ef587
  Author: Dave Lee <davelee.com at gmail.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M lldb/source/DataFormatters/TypeSynthetic.cpp

  Log Message:
  -----------
  [lldb] Fix logic bugs in BytecodeSyntheticChildren (#227483)

Fix a pair of bugs in `BytecodeSyntheticChildren`.

1. Fix conditions that should compare with `>= 0` (not `> 0`)
2. Fix an inverted conditional when checking if a formatter implements
`@get_child_index`

Assisted-by: claude


  Commit: 85a5d0a53df92dc384058e8738777f81a78a0f69
      https://github.com/llvm/llvm-project/commit/85a5d0a53df92dc384058e8738777f81a78a0f69
  Author: Charles Zablit <c_zablit at apple.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M lldb/source/Core/IOHandler.cpp

  Log Message:
  -----------
  [REPL][Windows] Don't redraw the prompt into redirected REPL output (#227755)

On Windows without `libedit`, `IOHandlerEditline::PrintAsync` re-printed
the prompt after every async output chunk, even when stdout is not a
console. When REPL output is redirected, as in lit tests, `"> "` lands
in the middle of the program's output, which made `SwiftREPL` tests like
`ExclusivityREPL` flaky.

The prompt is now redrawn only when the handler is interactive and
stdout is a console that can erase it.


  Commit: 65051a3c32709971db58e2388f1eaae1edeb146a
      https://github.com/llvm/llvm-project/commit/65051a3c32709971db58e2388f1eaae1edeb146a
  Author: Farzon Lotfi <farzonlotfi at microsoft.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M clang/lib/CodeGen/CGExprConstant.cpp
    M clang/test/CodeGenHLSL/BasicFeatures/MatrixInitializerListOrder.hlsl

  Log Message:
  -----------
  [Clang][HLSL] Emit constant matrix values in column-major order (#227356)

Emit matrix APValues in the canonical column-major register layout,
independent of the selected matrix memory layout.

When a constant is used as a row-major memory initializer, reorder its
elements at the register-to-memory boundary. This preserves the physical
layout of global and static row-major matrices while ensuring constexpr
matrix expressions produce canonical register values.

Assisted by Cop-pilot GPT 5.6-Sol


  Commit: 6b9d37acbd572a8a07bf51c3784ddfd4ead2944c
      https://github.com/llvm/llvm-project/commit/6b9d37acbd572a8a07bf51c3784ddfd4ead2944c
  Author: Kazu Hirata <kazu at google.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M llvm/lib/CodeGen/CodeGenPrepare.cpp

  Log Message:
  -----------
  [CodeGen] Remove unused InstructionMoveBefore (NFC) (#227586)

The last use was removed on December 15, 2023 in commit
163aeca33d4adb97e8599584409457ca14b1419b.

Assisted-by: Antigravity


  Commit: db409ffcf81de44fb2b3644b2fe28eebf4b69df0
      https://github.com/llvm/llvm-project/commit/db409ffcf81de44fb2b3644b2fe28eebf4b69df0
  Author: Andy Kaylor <akaylor at nvidia.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M clang/test/CIR/CodeGen/constexpr-ptr-offset.cpp
    M clang/test/CIR/CodeGen/global-ptr-init.cpp
    M clang/test/CIR/CodeGen/globals.cpp
    M clang/test/CIR/CodeGen/self-ref-temporaries.cpp
    M clang/test/CIR/CodeGen/vtt.cpp
    M clang/test/CIR/CodeGenCXX/base-layout.cpp
    M clang/test/CIR/CodeGenCXX/vtable-linkage.cpp

  Log Message:
  -----------
  [CIR] Fix failing tests after #226904 (#227523)

A number of CIR tests were checking LLVM IR output for GEP statements
that contained `inbounds` and `nuw` flags that had been added by LLVM's
constant folder. Following
https://github.com/llvm/llvm-project/pull/226904 these are no longer
produced.

This change updates the tests to match the current output.

Assisted-by: Cursor / claude-opus-5.5


  Commit: b8afba677502d531efd5c6096e23b46ef488da96
      https://github.com/llvm/llvm-project/commit/b8afba677502d531efd5c6096e23b46ef488da96
  Author: Erich Keane <ekeane at nvidia.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M clang/include/clang/CIR/CIRGenerator.h
    M clang/include/clang/CIR/Dialect/IR/CIRDialect.td
    A clang/include/clang/CodeGenUtils/BackendDiagnosticHandler.h
    M clang/lib/CIR/CodeGen/CIRGenCall.cpp
    M clang/lib/CIR/CodeGen/CIRGenModule.cpp
    M clang/lib/CIR/CodeGen/CIRGenerator.cpp
    M clang/lib/CIR/FrontendAction/CIRGenAction.cpp
    M clang/lib/CIR/FrontendAction/CMakeLists.txt
    M clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
    M clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVMIR.cpp
    M clang/lib/CodeGen/BackendConsumer.h
    M clang/lib/CodeGen/CodeGenAction.cpp
    A clang/lib/CodeGenUtils/BackendDiagnosticHandler.cpp
    M clang/lib/CodeGenUtils/CMakeLists.txt
    A clang/test/CIR/CodeGen/attr-error.c
    A clang/test/CIR/Diagnostics/attr-error.c
    A clang/test/CIR/Diagnostics/inline-asm-backend-diagnostic.c

  Log Message:
  -----------
  [CodeGen][CIR] Split out Backend Diags to utils, impl CIR error-attr (#225966)

So first, the motivation for this is the `error` attribute, which
applies to a function and diagnoses if it 'survives' into the backend as
a call. In order to do this, we need to add the
dontcall-error/dontcall-warn attributes to functions.

The call operation ALSO needs the srcloc to be present, so this threads
that through as well. Note this is pretty fragile in classic-codegen, so
we inherit some fragility from that source location as well.

HOWEVER, in order to get diagnostics to work properly in the backend, we
need to consume them. This patch ALSO extracts the handling for that
from CodeGenAction.cpp into CodeGenUtils, plus uses it from both sides.

Disclaimer: I ended up using Claude to do a lot of the refactoring. It
was mostly copy/paste with some minor changes to generalize it, but
Cladue helped.


  Commit: ffb3120b0aae71d16c275bfb4a56b8633897fdae
      https://github.com/llvm/llvm-project/commit/ffb3120b0aae71d16c275bfb4a56b8633897fdae
  Author: Erich Keane <ekeane at nvidia.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M clang/lib/CIR/Dialect/Transforms/CIRSimplify.cpp
    M clang/test/CIR/Transforms/switch-fold.cir

  Log Message:
  -----------
  [CIR] Don't 'simplify' switch-case if there is something in the way. (#227437)

The simplify case code was assuming that cases were mergeable, and
didn't realize that code in the middle (like the goto in the test case)
meant they couldn't be re-joined. This patch adds that to the
CIRSimplify logic to make sure we con't try to combine cases that have
something 'in between'.


  Commit: ea8ba8302eea264706354d92e7c9027ceae4f0b4
      https://github.com/llvm/llvm-project/commit/ea8ba8302eea264706354d92e7c9027ceae4f0b4
  Author: Alex MacLean <amaclean at nvidia.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M llvm/lib/Transforms/Scalar/SeparateConstOffsetFromGEP.cpp
    M llvm/test/Transforms/SeparateConstOffsetFromGEP/NVPTX/split-gep.ll
    A llvm/test/Transforms/SeparateConstOffsetFromGEP/NVPTX/trunc-ext.ll

  Log Message:
  -----------
  [SeparateConstOffsetFromGEP] Fix offsets across trunc and ext (#225195)

Stop extracting offsets from arithmetic through a truncation followed by
sign or zero extension.

[Compiler Explorer reproducer](https://godbolt.org/z/xK1WTT7nY) and
[Alive2 counterexample](https://alive2.llvm.org/ce/z/KWcMNZ).


  Commit: 9fe2e2f546734461220bdf87805b26d58db3ffad
      https://github.com/llvm/llvm-project/commit/9fe2e2f546734461220bdf87805b26d58db3ffad
  Author: Louis Dionne <ldionne.2 at gmail.com>
  Date:   2026-09-30 (Wed, 30 Sep 2026)

  Changed paths:
    M clang/include/clang/CIR/CIRGenerator.h
    M clang/include/clang/CIR/Dialect/IR/CIRDialect.td
    A clang/include/clang/CodeGenUtils/BackendDiagnosticHandler.h
    M clang/lib/CIR/CodeGen/CIRGenCall.cpp
    M clang/lib/CIR/CodeGen/CIRGenModule.cpp
    M clang/lib/CIR/CodeGen/CIRGenerator.cpp
    M clang/lib/CIR/Dialect/Transforms/CIRSimplify.cpp
    M clang/lib/CIR/FrontendAction/CIRGenAction.cpp
    M clang/lib/CIR/FrontendAction/CMakeLists.txt
    M clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
    M clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVMIR.cpp
    M clang/lib/CodeGen/BackendConsumer.h
    M clang/lib/CodeGen/CGExpr.cpp
    M clang/lib/CodeGen/CGExprConstant.cpp
    M clang/lib/CodeGen/CGExprScalar.cpp
    M clang/lib/CodeGen/CodeGenAction.cpp
    A clang/lib/CodeGenUtils/BackendDiagnosticHandler.cpp
    M clang/lib/CodeGenUtils/CMakeLists.txt
    A clang/test/CIR/CodeGen/attr-error.c
    M clang/test/CIR/CodeGen/constexpr-ptr-offset.cpp
    M clang/test/CIR/CodeGen/global-ptr-init.cpp
    M clang/test/CIR/CodeGen/globals.cpp
    M clang/test/CIR/CodeGen/self-ref-temporaries.cpp
    M clang/test/CIR/CodeGen/vtt.cpp
    M clang/test/CIR/CodeGenCXX/base-layout.cpp
    M clang/test/CIR/CodeGenCXX/vtable-linkage.cpp
    A clang/test/CIR/Diagnostics/attr-error.c
    A clang/test/CIR/Diagnostics/inline-asm-backend-diagnostic.c
    M clang/test/CIR/Transforms/switch-fold.cir
    M clang/test/CodeGen/isfpclass.c
    M clang/test/CodeGenHLSL/BasicFeatures/MatrixElementRowColFlags.hlsl
    M clang/test/CodeGenHLSL/BasicFeatures/MatrixElementTypeCast.hlsl
    M clang/test/CodeGenHLSL/BasicFeatures/MatrixInitializerListOrder.hlsl
    M clang/test/CodeGenHLSL/matrix-layout-attr-overrides-default.hlsl
    M clang/test/Headers/__clang_hip_math.hip
    M clang/test/Headers/wasm.c
    M lldb/source/Core/IOHandler.cpp
    M lldb/source/DataFormatters/TypeSynthetic.cpp
    M llvm/lib/Analysis/ValueTracking.cpp
    M llvm/lib/CodeGen/CodeGenPrepare.cpp
    M llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
    M llvm/lib/Transforms/Scalar/SeparateConstOffsetFromGEP.cpp
    M llvm/test/CodeGen/AMDGPU/eliminate-frame-index-select.ll
    A llvm/test/CodeGen/AMDGPU/promote-alloca-homogeneous-struct.ll
    M llvm/test/Transforms/SeparateConstOffsetFromGEP/NVPTX/split-gep.ll
    A llvm/test/Transforms/SeparateConstOffsetFromGEP/NVPTX/trunc-ext.ll
    M llvm/test/Transforms/SimplifyCFG/switch-range-to-icmp.ll
    M llvm/unittests/Analysis/ValueTrackingTest.cpp
    M utils/bazel/llvm-project-overlay/llvm/BUILD.bazel

  Log Message:
  -----------
  Merge branch 'main' into users/ldionne/libcxx-benchmarks-inline-build


Compare: https://github.com/llvm/llvm-project/compare/9640e2fe517b...9fe2e2f54673

To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications


More information about the All-commits mailing list