Skip to content

aco: disable mul(cndmask(0, 1, b), a) optimization sometimes

Rhys Perry requested to merge pendingchaos/mesa:aco_small_mul_b2f_fixes into main

This optimization doesn't work for SDWA or DPP multiplications and we can't do it if denormal flushing is required because v_cndmask_b32 doesn't do that and we can't do it if we can't assume operands are finite because 0.0 * inf is NaN, not 0.

Merge request reports