[llvm] [AMDGPU] Allocate i1 argument to SGPRs (PR #72461)

[Matt Arsenault via llvm-commits]

================
@@ -2782,8 +2781,8 @@ define void @void_func_i1_inreg(i1 inreg %arg0) #0 {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK-NEXT:   liveins: $sgpr16
   ; CHECK-NEXT: {{  $}}
-  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr16
-  ; CHECK-NEXT:   [[TRUNC:%[0-9]+]]:_(s1) = G_TRUNC [[COPY]](s32)
+  ; CHECK-NEXT:   [[COPY:%[0-9]+]]:_(s64) = COPY $sgpr16_sgpr17
+  ; CHECK-NEXT:   [[TRUNC:%[0-9]+]]:_(s1) = G_TRUNC [[COPY]](s64)
----------------
arsenm wrote:

The inreg case shouldn't change 

https://github.com/llvm/llvm-project/pull/72461