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

[Jeffrey Byrnes via llvm-commits]

================
@@ -130,19 +147,42 @@ define amdgpu_kernel void @v8i8_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 v4, 3, v0
+; GFX906-NEXT:    v_lshlrev_b32_e32 v10, 3, v0
 ; GFX906-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX906-NEXT:    v_cmp_gt_u32_e32 vcc, 15, v0
 ; GFX906-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX906-NEXT:    global_load_dwordx2 v[1:2], v4, s[0:1]
+; GFX906-NEXT:    global_load_dwordx2 v[1:2], v10, s[0:1]
+; GFX906-NEXT:    s_waitcnt vmcnt(0)
+; GFX906-NEXT:    v_lshrrev_b32_e32 v4, 24, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v5, 16, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v6, 8, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX906-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
+; GFX906-NEXT:    v_lshrrev_b32_e32 v9, 8, v1
 ; GFX906-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX906-NEXT:    s_cbranch_execz .LBB3_2
 ; GFX906-NEXT:  ; %bb.1: ; %bb.1
-; GFX906-NEXT:    global_load_dwordx2 v[1:2], v4, s[2:3]
+; GFX906-NEXT:    global_load_dwordx2 v[1:2], v10, s[2:3]
+; GFX906-NEXT:    s_waitcnt vmcnt(0)
+; GFX906-NEXT:    v_lshrrev_b32_e32 v4, 24, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v5, 16, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v6, 8, v2
+; GFX906-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX906-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
+; GFX906-NEXT:    v_lshrrev_b32_e32 v9, 8, v1
 ; GFX906-NEXT:  .LBB3_2: ; %bb.2
 ; GFX906-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX906-NEXT:    s_waitcnt vmcnt(0)
-; GFX906-NEXT:    global_store_dwordx2 v3, v[1:2], s[6:7]
+; GFX906-NEXT:    v_lshlrev_b16_e32 v0, 8, v9
+; 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, v7
+; GFX906-NEXT:    v_or_b32_sdwa v1, v8, 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, v6
+; 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, v4
+; GFX906-NEXT:    v_or_b32_sdwa v2, v5, 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 v3, v[0:1], s[6:7]
----------------
jrbyrnes wrote:

Ditto

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