[llvm-branch-commits] [llvm] AMDGPU/GlobalISel: Use integers for read-any-lane split type (PR #210098)

[Matt Arsenault via llvm-branch-commits]

================
@@ -414,12 +414,12 @@ define void @test_indirect_call_vgpr_ptr_arg(ptr %fptr) #0 {
 ; GISEL-NEXT:    s_mov_b64 s[38:39], s[6:7]
 ; GISEL-NEXT:    s_mov_b64 s[48:49], s[4:5]
 ; GISEL-NEXT:    s_mov_b64 s[54:55], exec
-; GISEL-NEXT:    v_mov_b32_e32 v2, 0x7b
 ; GISEL-NEXT:  .LBB3_1: ; =>This Inner Loop Header: Depth=1
 ; GISEL-NEXT:    v_readfirstlane_b32 s16, v0
 ; GISEL-NEXT:    v_readfirstlane_b32 s17, v1
 ; GISEL-NEXT:    v_cmp_eq_u64_e32 vcc, s[16:17], v[0:1]
 ; GISEL-NEXT:    s_and_saveexec_b64 s[64:65], vcc
+; GISEL-NEXT:    v_mov_b32_e32 v0, 0x7b
----------------
arsenm wrote:

Weird this had codegen changes 

https://github.com/llvm/llvm-project/pull/210098