Michael Krumland-DunningandClaude Opus 5.5 254d7b416f fix(shader): translate the fused f32 fma forms in f32 denormal mode 2 ...

v_fma_f32, v_fmac_f32, v_fmamk_f32 and v_fmaak_f32 threw "f32 denormal
mode 2 is not implemented" whenever FLOAT_MODE flushed f32 denormal
inputs but kept denormal results.

On RDNA2 (gfx1036) at round to nearest even, mode 2 flushes the inputs
and rounds the fused result once with gradual underflow, which the host's
Fma gives under DenormPreserve 32. The translation now matches all 5,596
measured triples for the four forms at IEEE 0 and 1. Mode 1 keeps
throwing: there the hardware flushes a result that is tiny after rounding
with an unbounded exponent even when the addend is nonzero.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

2026-10-27 22:05:57 -07:00

..

debt

fix(shader): translate the fused f32 fma forms in f32 denormal mode 2

2026-10-27 22:05:57 -07:00

ARCHITECTURE.md

fix(shader): keep the f32 floor and ceil flush where the device flushes by default

2026-10-26 08:57:06 +00:00

BUILD.md

Merge pull request #11062 from MarcZX10/marc/ci-nightly-sanitizers

2026-10-28 05:58:34 +13:00

CONVENTIONS.md

…

ENVIRONMENT.md

Merge pull request #11637 from B-Howell/codex/videoout-display-index-current-main

2026-10-28 03:49:48 +01:00

GAME_TESTING.md

docs(game-testing): say what exit code 1 of the import audit means

2026-10-26 23:24:10 +00:00

HW_ORACLE.md

…

PROGRESS.md

…

REVERSE_ENGINEERING.md

…

TechnicalDebt.md

Merge pull request #11868 from rooklz/fix/check-compile-save-temps-modes-v2

2026-10-28 12:59:01 +13:00

TRACING.md

feat(tools): split the periodic reports of an APS5_AB run by phase

2026-10-26 12:26:29 +01:00