[llvm] [AMDGPU] Filter candidates of LiveRegOptimizer for profitable cases (PR #124624)

[Jeffrey Byrnes via llvm-commits]

================
@@ -54,19 +50,31 @@ define amdgpu_kernel void @v4i8_liveout(ptr addrspace(1) %src1, ptr addrspace(1)
 ; GFX906:       ; %bb.0: ; %entry
 ; GFX906-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
 ; GFX906-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
-; GFX906-NEXT:    v_lshlrev_b32_e32 v3, 2, v0
+; GFX906-NEXT:    v_lshlrev_b32_e32 v6, 2, v0
 ; GFX906-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX906-NEXT:    v_cmp_gt_u32_e32 vcc, 15, v0
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    global_load_dword v2, v3, s[0:1]
+; GFX906-NEXT:    global_load_dword v2, v6, s[0:1]
+; GFX906-NEXT:    s_waitcnt vmcnt(0)
+; GFX906-NEXT:    v_lshrrev_b32_e32 v3, 24, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v4, 16, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v5, 8, v2
 ; GFX906-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX906-NEXT:    s_cbranch_execz .LBB1_2
 ; GFX906-NEXT:  ; %bb.1: ; %bb.1
-; GFX906-NEXT:    global_load_dword v2, v3, s[2:3]
+; GFX906-NEXT:    global_load_dword v2, v6, s[2:3]
+; GFX906-NEXT:    s_waitcnt vmcnt(0)
+; GFX906-NEXT:    v_lshrrev_b32_e32 v3, 24, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v4, 16, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v5, 8, v2
 ; GFX906-NEXT:  .LBB1_2: ; %bb.2
 ; GFX906-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    global_store_dword v1, v2, s[6:7]
+; GFX906-NEXT:    v_lshlrev_b16_e32 v0, 8, v5
----------------
jrbyrnes wrote:

The LRO should coerce this.

https://github.com/llvm/llvm-project/pull/124624