[llvm] AMDGPU: Track AGPR pressure (PR #150288)

[Matt Arsenault via llvm-commits]

================
@@ -1935,25 +1935,25 @@ define <16 x i32> @v_saddsat_v16i32(<16 x i32> %lhs, <16 x i32> %rhs) {
 ; GFX6-NEXT:    v_min_i32_e32 v32, 0, v0
 ; GFX6-NEXT:    v_bfrev_b32_e32 v31, 1
 ; GFX6-NEXT:    v_sub_i32_e32 v32, vcc, v31, v32
-; GFX6-NEXT:    v_max_i32_e32 v32, v32, v16
 ; GFX6-NEXT:    v_max_i32_e32 v33, 0, v0
+; GFX6-NEXT:    v_max_i32_e32 v32, v32, v16
 ; GFX6-NEXT:    v_bfrev_b32_e32 v16, -2
 ; GFX6-NEXT:    v_sub_i32_e32 v33, vcc, v16, v33
 ; GFX6-NEXT:    v_min_i32_e32 v32, v32, v33
+; GFX6-NEXT:    v_min_i32_e32 v33, 0, v1
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v32
-; GFX6-NEXT:    v_min_i32_e32 v32, 0, v1
-; GFX6-NEXT:    v_sub_i32_e32 v32, vcc, v31, v32
+; GFX6-NEXT:    v_sub_i32_e32 v32, vcc, v31, v33
----------------
arsenm wrote:

Why are there so many test changes on targets that don't have AGPRs, and functions that don't have AGPRs 

https://github.com/llvm/llvm-project/pull/150288