[all-commits] [llvm/llvm-project] 3d9176: [X86] Fold FMSUBADD + FADD into VFCMADDC for fp16 ...

[Rohan Shenoy via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 3d91765f54e9c461c3f979a5ea7c302f59303489
      https://github.com/llvm/llvm-project/commit/3d91765f54e9c461c3f979a5ea7c302f59303489
  Author: Rohan Shenoy <RohanRadhesh.Shenoy at amd.com>
  Date:   2026-09-14 (Mon, 14 Sep 2026)

  Changed paths:
    M llvm/lib/Target/X86/X86ISelLowering.cpp
    A llvm/test/CodeGen/X86/avx512fp16-combine-fmsubadd-fadd.ll

  Log Message:
  -----------
  [X86] Fold FMSUBADD + FADD into VFCMADDC for fp16 (#222896)

Complex multiply by conjugate and accumulate fp16 (c += a * ~b) is
lowered into vfmsubadd + vaddph, failing to emit native vfcmaddcph in
avx512fp16. We can reuse the pattern for FSUBADD and extend
combineFaddCFmul to perform this fold.

Fixes #216290



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