[all-commits] [llvm/llvm-project] 7bb9b1: [AMDGPU] Rework pseudo-scalar transcendental inlin...

[Frederik Harwath via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 7bb9b13816ef9616425525651ea629f2138b194f
      https://github.com/llvm/llvm-project/commit/7bb9b13816ef9616425525651ea629f2138b194f
  Author: Frederik Harwath <frederik.harwath at amd.com>
  Date:   2026-09-09 (Wed, 09 Sep 2026)

  Changed paths:
    M llvm/lib/Target/AMDGPU/AMDGPU.td
    M llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
    M llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp
    M llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUInstPrinter.cpp
    M llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCCodeEmitter.cpp
    M llvm/lib/Target/AMDGPU/SIDefines.h
    M llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
    M llvm/lib/Target/AMDGPU/SIInstrInfo.h
    M llvm/lib/Target/AMDGPU/SIRegisterInfo.td
    M llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
    M llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
    M llvm/lib/Target/AMDGPU/VOP3Instructions.td
    M llvm/test/CodeGen/AMDGPU/pseudo-scalar-transcendental.mir
    M llvm/test/MC/AMDGPU/gfx12_asm_vop3-fake16.s
    M llvm/test/MC/AMDGPU/gfx12_asm_vop3.s
    M llvm/test/MC/AMDGPU/gfx13_asm_vop3.s
    M llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3.txt

  Log Message:
  -----------
  [AMDGPU] Rework pseudo-scalar transcendental inline constant restriction handling (#208459)

The f16 pseudo-scalar transcendental instructions do not support inline
constants on gfx12. The existing handling of this restriction is
incomplete. It involves checking the
hasNoF16PseudoScalarTransInlineConstants function in the right places.
For the SIInstrInfo::isInlineConstant function, for instance, this has
led to a situation where some overloads respect the restriction and
others do not. This approach also relies on completely disabling the
folding of constant operands. The machine code emission does still use
the inline constant encoding if it encounters a suitable operand.

Introduce a new "noinline" operand type and use it for the instruction
variant that is used on gfx12. This does also enable the use of literal
constants.



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