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

[Jeffrey Byrnes via llvm-commits]

================
@@ -652,10 +2379,26 @@ define amdgpu_kernel void @v8i8_multi_block(ptr addrspace(1) %src1, ptr addrspac
 ; GFX906-NEXT:    global_store_dwordx2 v0, v[3:4], s[12:13]
 ; GFX906-NEXT:  .LBB11_3: ; %Flow
 ; GFX906-NEXT:    s_or_b64 exec, exec, s[2:3]
+; GFX906-NEXT:    s_waitcnt vmcnt(0)
+; GFX906-NEXT:    v_lshrrev_b32_e32 v6, 24, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v7, 16, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v8, 8, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v9, 24, v1
+; GFX906-NEXT:    v_lshrrev_b32_e32 v10, 16, v1
+; GFX906-NEXT:    v_lshrrev_b32_e32 v11, 8, v1
 ; GFX906-NEXT:  .LBB11_4: ; %bb.3
 ; GFX906-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    global_store_dwordx2 v5, v[1:2], s[14:15]
+; GFX906-NEXT:    v_lshlrev_b16_e32 v0, 8, v11
+; GFX906-NEXT:    v_or_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX906-NEXT:    v_lshlrev_b16_e32 v1, 8, v9
+; GFX906-NEXT:    v_or_b32_sdwa v1, v10, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX906-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX906-NEXT:    v_lshlrev_b16_e32 v1, 8, v8
+; GFX906-NEXT:    v_or_b32_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX906-NEXT:    v_lshlrev_b16_e32 v2, 8, v6
+; GFX906-NEXT:    v_or_b32_sdwa v2, v7, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX906-NEXT:    v_or_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; GFX906-NEXT:    global_store_dwordx2 v5, v[0:1], s[14:15]
----------------
jrbyrnes wrote:

Ditto

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