Skip to content

aco: fix packed 16bit fneg/fsat optimization

Daniel Schürmann requested to merge daniel-schuermann/mesa:aco_fix_packed_mod into main 
Make sure that the Operand is '1.0.xx'.

Fixes: b03be30e07546b813acc1156407b964e68892beb ('aco: optimize packed fneg')

Encountered in GTF-GL46.gtf21.GL.cross.cross_vec3_frag_xvaryyconst via Zink as

	vec1 16 div ssa_48 = load_const (0x3c00 = 1.000000)
	vec1 16 div ssa_47 = load_const (0x0000 = 0.000000)
	vec2 16 div ssa_49 = vec2 ssa_47, ssa_48
	vec2 16 div ssa_14 = vec2 ssa_12, ssa_13
	vec2 16 div ssa_15 = fmul ssa_14, ssa_49 // 1.0.yx

Merge request reports