[llvm] [CodeGenPrepare] Reverse the canonicalization of isInf/isNanOrInf (PR #81572)

[Matt Arsenault via llvm-commits]

================
@@ -61,22 +61,22 @@ define amdgpu_kernel void @test_not_isinf_pattern_0(ptr addrspace(1) nocapture %
 ; SI-NEXT:    s_load_dword s0, s[0:1], 0xb
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, -1
-; SI-NEXT:    v_mov_b32_e32 v0, 0x7f800000
+; SI-NEXT:    v_mov_b32_e32 v0, 0x207
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    v_cmp_nlg_f32_e64 s[0:1], |s0|, v0
-; SI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s[0:1]
+; SI-NEXT:    v_cmp_class_f32_e32 vcc, s0, v0
+; SI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
----------------
arsenm wrote:

small code size improvement 

https://github.com/llvm/llvm-project/pull/81572