[llvm] [AMDGPU] promote i1 arg type for amdgpu_cs (PR #82971)

[Matt Arsenault via llvm-commits]

================
@@ -2078,4 +2078,1169 @@ entry:
   ret void
 }
 
+define amdgpu_cs void @amdgpu_cs_i1(i1 %arg0) {
+; SI-LABEL: amdgpu_cs_i1:
+; SI:       ; %bb.0:
+; SI-NEXT:    v_and_b32_e32 v0, 1, v0
----------------
arsenm wrote:

Could also test the signext/zeroext cases, in which cases we can avoid the instruction 

https://github.com/llvm/llvm-project/pull/82971