[all-commits] [llvm/llvm-project] 755e4d: [SelectionDAG] Fix FP8 conversions to half for sub...

[Chinmay Deshpande via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 755e4d429f11a17ba386cdf4d4ff7883294368dd
      https://github.com/llvm/llvm-project/commit/755e4d429f11a17ba386cdf4d4ff7883294368dd
  Author: Chinmay Deshpande <chdeshpa at amd.com>
  Date:   2026-09-22 (Tue, 22 Sep 2026)

  Changed paths:
    M llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
    A llvm/test/CodeGen/X86/arbitrary-fp-to-half.ll

  Log Message:
  -----------
  [SelectionDAG] Fix FP8 conversions to half for subnormals and overflow (#224984)

This PR contains 2 fixes:

* Preserve subnormal values when the source format and destination have
the same exponent bias. Normalizing these inputs previously produced a
nonpositive destination exponent and an invalid floating-point encoding.

* Convert finite values beyond the destination exponent range to
infinity, while preserving the existing NaN handling. This fixes
Float8E5M3FNU to half conversions that previously produced NaNs on
overflow.

Assisted by: Codex (Astra)



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