[llvm] [AMDGPU] Prefer v_madak_f32 over v_madmk_f32 to reduce vgpr pressure (PR #72506)

[Jay Foad via llvm-commits]

================
@@ -37,41 +36,42 @@ define amdgpu_ps float @_amdgpu_ps_main() #0 {
 ; GFX10-NEXT:    v_fma_f32 v1, v1, v5, s28
 ; GFX10-NEXT:    v_max_f32_e64 v6, s0, s0 clamp
 ; GFX10-NEXT:    v_add_f32_e64 v5, s29, -1.0
-; GFX10-NEXT:    v_sub_f32_e32 v9, s0, v1
-; GFX10-NEXT:    v_fma_f32 v8, -s2, v6, s6
+; GFX10-NEXT:    v_sub_f32_e32 v8, s0, v1
+; GFX10-NEXT:    v_fma_f32 v7, -s2, v6, s6
 ; GFX10-NEXT:    v_fma_f32 v5, v6, v5, 1.0
-; GFX10-NEXT:    v_mad_f32 v11, s2, v6, v2
-; GFX10-NEXT:    v_fmac_f32_e32 v1, v6, v9
-; GFX10-NEXT:    v_fmac_f32_e32 v11, v8, v6
+; GFX10-NEXT:    v_mad_f32 v10, s2, v6, v2
+; GFX10-NEXT:    s_mov_b32 s0, 0x3c23d70a
----------------
jayfoad wrote:

Now: literal materialized in sgpr

https://github.com/llvm/llvm-project/pull/72506