[all-commits] [llvm/llvm-project] 9c78bc: Revert "[LSV] Merge contiguous chains across scala...

[Drew Kersnar via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 9c78bc5de4fc2450d8fd5e5d52e8168ef653958e
      https://github.com/llvm/llvm-project/commit/9c78bc5de4fc2450d8fd5e5d52e8168ef653958e
  Author: Drew Kersnar <dkersnar at nvidia.com>
  Date:   2025-12-02 (Tue, 02 Dec 2025)

  Changed paths:
    M llvm/include/llvm/Transforms/Utils/Local.h
    M llvm/lib/Target/AMDGPU/AMDGPULowerBufferFatPointers.cpp
    M llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
    M llvm/lib/Transforms/Scalar/SROA.cpp
    M llvm/lib/Transforms/Utils/Local.cpp
    M llvm/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp
    M llvm/test/CodeGen/AMDGPU/GlobalISel/amdgpu-irtranslator.ll
    M llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
    M llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll
    M llvm/test/CodeGen/AMDGPU/build_vector.ll
    M llvm/test/CodeGen/AMDGPU/fabs.bf16.ll
    M llvm/test/CodeGen/AMDGPU/fabs.f16.ll
    M llvm/test/CodeGen/AMDGPU/fabs.ll
    M llvm/test/CodeGen/AMDGPU/fcopysign.f32.ll
    M llvm/test/CodeGen/AMDGPU/fdiv.ll
    M llvm/test/CodeGen/AMDGPU/fnearbyint.ll
    M llvm/test/CodeGen/AMDGPU/fneg-fabs.bf16.ll
    M llvm/test/CodeGen/AMDGPU/fneg-fabs.f16.ll
    M llvm/test/CodeGen/AMDGPU/fneg-fabs.ll
    M llvm/test/CodeGen/AMDGPU/fneg.ll
    M llvm/test/CodeGen/AMDGPU/fp_to_sint.ll
    M llvm/test/CodeGen/AMDGPU/fp_to_uint.ll
    M llvm/test/CodeGen/AMDGPU/fshl.ll
    M llvm/test/CodeGen/AMDGPU/fshr.ll
    M llvm/test/CodeGen/AMDGPU/global_atomics.ll
    M llvm/test/CodeGen/AMDGPU/half.ll
    M llvm/test/CodeGen/AMDGPU/insert_vector_dynelt.ll
    M llvm/test/CodeGen/AMDGPU/insert_vector_elt.ll
    M llvm/test/CodeGen/AMDGPU/kernel-args.ll
    M llvm/test/CodeGen/AMDGPU/llvm.amdgcn.mfma.gfx950.ll
    M llvm/test/CodeGen/AMDGPU/llvm.exp.ll
    M llvm/test/CodeGen/AMDGPU/llvm.exp10.ll
    M llvm/test/CodeGen/AMDGPU/llvm.exp2.ll
    M llvm/test/CodeGen/AMDGPU/llvm.log.ll
    M llvm/test/CodeGen/AMDGPU/llvm.log10.ll
    M llvm/test/CodeGen/AMDGPU/llvm.log2.ll
    M llvm/test/CodeGen/AMDGPU/max.ll
    M llvm/test/CodeGen/AMDGPU/min.ll
    M llvm/test/CodeGen/AMDGPU/packed-fp32.ll
    M llvm/test/CodeGen/AMDGPU/rotl.ll
    M llvm/test/CodeGen/AMDGPU/rotr.ll
    M llvm/test/CodeGen/AMDGPU/s_addk_i32.ll
    M llvm/test/CodeGen/AMDGPU/sminmax.v2i16.ll
    M llvm/test/CodeGen/AMDGPU/store-to-constant.ll
    M llvm/test/CodeGen/AMDGPU/udivrem.ll
    M llvm/test/CodeGen/AMDGPU/uint_to_fp.f64.ll
    R llvm/test/Transforms/InstCombine/copy-access-metadata.ll
    R llvm/test/Transforms/LoadStoreVectorizer/AMDGPU/copy-metadata-load-store.ll
    M llvm/test/Transforms/LoadStoreVectorizer/AMDGPU/merge-vectors-complex.ll
    M llvm/test/Transforms/LoadStoreVectorizer/AMDGPU/merge-vectors.ll

  Log Message:
  -----------
  Revert "[LSV] Merge contiguous chains across scalar types" (#170381)

Reverts llvm/llvm-project#154069. I pointed out a number of issues
post-merge, most importantly examples of miscompiles:
https://github.com/llvm/llvm-project/pull/154069#issuecomment-3603854626.

While the motivation of the change is clear, I think the implementation
approach is flawed. It seems like the goal is to allow elements like
`load <2xi16>` and `load i32` to be vectorized together despite the
current algorithm not grouping them into the same equivalence classes. I
personally think that if we want to attempt this it should be a more
wholistic approach, maybe even redefining the concept of an equivalence
class. This current solution seems like it would be really hard to do
bug-free, and even if the bugs were not present, it is only able to
merge chains that happen to be adjacent to each other after
`splitChainByContiguity`, which seems like it is leaving things up to
chance whether this optimization kicks in. But we can discuss more in
the re-land. Maybe the broader approach I'm proposing is too difficult,
and a narrow optimization is worthwhile. Regardless, this should be
reverted, it needs more iteration before it is correct.



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