[llvm] [InstSimplify] Simplify fcmp implied by dominating fcmp (PR #161090)

[Yingwei Zheng via llvm-commits]

================
@@ -256,7 +256,7 @@ endif:                                            ; preds = %else, %if
 define amdgpu_kernel void @copy1(ptr addrspace(1) %out, ptr addrspace(1) %in0) {
 entry:
   %tmp = load float, ptr addrspace(1) %in0
-  %tmp1 = fcmp oeq float %tmp, 0.000000e+00
+  %tmp1 = fcmp one float %tmp, 0.000000e+00
----------------
dtcxzyw wrote:

Without this change, a cmov pattern will be emitted, and `v_add` is missing.

https://github.com/llvm/llvm-project/pull/161090