[llvm] [DAGCombine] Propagate truncate to operands (PR #98666)

[Matt Arsenault via llvm-commits]

================
@@ -118,8 +118,8 @@ define amdgpu_kernel void @s_anyext_i16_i32(ptr addrspace(1) %out, ptr addrspace
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 1, v1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_load_ushort v2, v0, s[6:7]
-; GFX9-NEXT:    global_load_ushort v3, v1, s[0:1]
+; GFX9-NEXT:    global_load_ubyte v2, v0, s[6:7]
+; GFX9-NEXT:    global_load_ubyte v3, v1, s[0:1]
----------------
arsenm wrote:

Better

https://github.com/llvm/llvm-project/pull/98666