@@ -1552,16 +1552,20 @@ define <4 x i16> @basic_smax_smin_v8i16_input_1(<8 x i16> %src) {
15521552; GISEL-GFX9-NEXT: v_pk_min_i16 v3, v4, v3
15531553; GISEL-GFX9-NEXT: v_pk_max_i16 v0, 0, v0
15541554; GISEL-GFX9-NEXT: v_mov_b32_e32 v4, 0xff
1555+ ; GISEL-GFX9-NEXT: v_and_b32_sdwa v5, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
15551556; GISEL-GFX9-NEXT: v_pk_max_i16 v1, 0, v1
1556- ; GISEL-GFX9-NEXT: v_and_b32_sdwa v5, v0, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
1557- ; GISEL-GFX9-NEXT: v_pk_max_i16 v2, 0, v2
1557+ ; GISEL-GFX9-NEXT: v_lshlrev_b16_e32 v5, 8, v5
15581558; GISEL-GFX9-NEXT: v_or_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
1559- ; GISEL-GFX9-NEXT: v_and_b32_sdwa v5, v1, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
1559+ ; GISEL-GFX9-NEXT: v_and_b32_sdwa v5, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
1560+ ; GISEL-GFX9-NEXT: v_pk_max_i16 v2, 0, v2
1561+ ; GISEL-GFX9-NEXT: v_lshlrev_b16_e32 v5, 8, v5
15601562; GISEL-GFX9-NEXT: v_pk_max_i16 v3, 0, v3
15611563; GISEL-GFX9-NEXT: v_or_b32_sdwa v1, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
1562- ; GISEL-GFX9-NEXT: v_and_b32_sdwa v5, v2, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
1564+ ; GISEL-GFX9-NEXT: v_and_b32_sdwa v5, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
1565+ ; GISEL-GFX9-NEXT: v_lshlrev_b16_e32 v5, 8, v5
1566+ ; GISEL-GFX9-NEXT: v_and_b32_sdwa v4, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
15631567; GISEL-GFX9-NEXT: v_or_b32_sdwa v2, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
1564- ; GISEL-GFX9-NEXT: v_and_b32_sdwa v4, v3 , v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
1568+ ; GISEL-GFX9-NEXT: v_lshlrev_b16_e32 v4, 8 , v4
15651569; GISEL-GFX9-NEXT: v_and_b32_e32 v0, 0xffff, v0
15661570; GISEL-GFX9-NEXT: v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
15671571; GISEL-GFX9-NEXT: v_lshl_or_b32 v0, v1, 16, v0
@@ -2208,9 +2212,11 @@ define <2 x i16> @basic_smax_smin_v4i16_input_1(<4 x i16> %src) {
22082212; GISEL-GFX9-NEXT: v_pk_max_i16 v0, 0, v0
22092213; GISEL-GFX9-NEXT: v_mov_b32_e32 v2, 0xff
22102214; GISEL-GFX9-NEXT: v_pk_max_i16 v1, 0, v1
2211- ; GISEL-GFX9-NEXT: v_and_b32_sdwa v3, v0, v2 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
2215+ ; GISEL-GFX9-NEXT: v_and_b32_sdwa v3, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
2216+ ; GISEL-GFX9-NEXT: v_lshlrev_b16_e32 v3, 8, v3
2217+ ; GISEL-GFX9-NEXT: v_and_b32_sdwa v2, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
22122218; GISEL-GFX9-NEXT: v_or_b32_sdwa v0, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
2213- ; GISEL-GFX9-NEXT: v_and_b32_sdwa v2, v1 , v2 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
2219+ ; GISEL-GFX9-NEXT: v_lshlrev_b16_e32 v2, 8 , v2
22142220; GISEL-GFX9-NEXT: v_or_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
22152221; GISEL-GFX9-NEXT: v_and_b32_e32 v0, 0xffff, v0
22162222; GISEL-GFX9-NEXT: v_lshl_or_b32 v0, v1, 16, v0
@@ -2535,7 +2541,8 @@ define i24 @basic_smax_smin_vec_v3i16(<3 x i16> %src) {
25352541; GISEL-VI-NEXT: v_max_i16_e32 v1, 0, v1
25362542; GISEL-VI-NEXT: v_or_b32_sdwa v0, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
25372543; GISEL-VI-NEXT: v_mov_b32_e32 v2, 0xffff
2538- ; GISEL-VI-NEXT: v_and_b32_sdwa v1, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
2544+ ; GISEL-VI-NEXT: v_and_b32_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
2545+ ; GISEL-VI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
25392546; GISEL-VI-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
25402547; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
25412548;
@@ -2547,7 +2554,8 @@ define i24 @basic_smax_smin_vec_v3i16(<3 x i16> %src) {
25472554; GISEL-GFX9-NEXT: v_mov_b32_e32 v2, 0xff
25482555; GISEL-GFX9-NEXT: v_pk_max_i16 v0, 0, v0
25492556; GISEL-GFX9-NEXT: v_med3_i16 v1, v1, 0, v2
2550- ; GISEL-GFX9-NEXT: v_and_b32_sdwa v2, v0, v2 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
2557+ ; GISEL-GFX9-NEXT: v_and_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
2558+ ; GISEL-GFX9-NEXT: v_lshlrev_b16_e32 v2, 8, v2
25512559; GISEL-GFX9-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
25522560; GISEL-GFX9-NEXT: s_mov_b32 s4, 0xffff
25532561; GISEL-GFX9-NEXT: v_and_b32_e32 v0, 0xffff, v0
0 commit comments