[llvm] AMDGPU/GlobalISel: Partially move constant selection to patterns (PR #100786)

[Matt Arsenault via llvm-commits]

================
@@ -8,9 +8,9 @@ define double @v_floor_f64_ieee(double %x) {
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX6-NEXT:    v_fract_f64_e32 v[2:3], v[0:1]
-; GFX6-NEXT:    s_mov_b32 s4, -1
-; GFX6-NEXT:    s_mov_b32 s5, 0x3fefffff
-; GFX6-NEXT:    v_min_f64 v[2:3], v[2:3], s[4:5]
+; GFX6-NEXT:    v_mov_b32_e32 v4, -1
----------------
arsenm wrote:

Sort of yes and sort of no. It's not slower but it would probably be best to use SGPRs when possible 

https://github.com/llvm/llvm-project/pull/100786