[llvm] InstSimplify: lookthru casts, binops in folding shuffles (PR #92668)

[Matt Arsenault via llvm-commits]

================
@@ -6,26 +6,11 @@ declare i32 @llvm.amdgcn.sffbh.i32(i32) nounwind readnone speculatable
 define amdgpu_kernel void @select_constant_cttz(ptr addrspace(1) noalias %out, ptr addrspace(1) nocapture readonly %arrayidx) nounwind {
 ; GCN-LABEL: select_constant_cttz:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_load_dword s2, s[2:3], 0x0
+; GCN-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
----------------
arsenm wrote:

The AMDGPU part doesn't matter, just make sure whatever simplify now happens, still doesn't happen. It's super frustruating that InstSimplify ends up breaking codegen tests like this. 

Is this just from the recursive depth increase? 

https://github.com/llvm/llvm-project/pull/92668