[all-commits] [llvm/llvm-project] 9b2af9: [AMDGPU] Price scalar integer to fp casts by sourc...

[Dmitry Sidorov via All-commits]

  Branch: refs/heads/users/mrsidims/fix-amdgpu-scalar-int-fp-cast-cost
  Home:   https://github.com/llvm/llvm-project
  Commit: 9b2af90382db17635aaa63a2423404a0fd3fefb9
      https://github.com/llvm/llvm-project/commit/9b2af90382db17635aaa63a2423404a0fd3fefb9
  Author: Dmitry Sidorov <Dmitry.Sidorov at amd.com>
  Date:   2026-09-21 (Mon, 21 Sep 2026)

  Changed paths:
    M llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
    M llvm/test/Analysis/CostModel/AMDGPU/narrow-int-to-bfloat.ll
    M llvm/test/Analysis/CostModel/AMDGPU/narrow-int-to-fp.ll

  Log Message:
  -----------
  [AMDGPU] Price scalar integer to fp casts by source width and sign

Scalar sources between a byte and 31 bits fell to the default cost of one
while the matching vector lanes were already priced, which skewed the
difference SLP weighs a bundle against. Such a source is extended before
the conversion, and what the extension takes depends on the width, on the
sign and on whether the subtarget has SDWA and 16 bit instructions.
Sources narrower than a byte are left alone, because their vector form is
not priced either.



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