Skip to content

Commit 8e68a51

Browse files
RKSimonRKSimon
committed
[X86] Add test coverage for #109790 
1 parent fe6a3d4 commit 8e68a51

File tree

1 file changed

+72
-0
lines changed

1 file changed

+72
-0
lines changed

‎llvm/test/CodeGen/X86/pmulh.ll

Lines changed: 72 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -937,6 +937,77 @@ define <16 x i32> @zext_mulhuw_v16i16_lshr(<16 x i16> %a, <16 x i16> %b) {
937937
ret <16 x i32> %d
938938
}
939939

940+
; PR109790
941+
define <16 x i16> @zext_mulhuw_v16i16_negative_constant(<16 x i16> %a) {
942+
; SSE-LABEL: zext_mulhuw_v16i16_negative_constant:
943+
; SSE: # %bb.0:
944+
; SSE-NEXT: movdqa {{.*#+}} xmm2 = [32767,32767,32767,32767,32767,32767,32767,32767]
945+
; SSE-NEXT: pand %xmm2, %xmm1
946+
; SSE-NEXT: pand %xmm2, %xmm0
947+
; SSE-NEXT: movdqa {{.*#+}} xmm2 = [64536,64536,64536,64536,64536,64536,64536,64536]
948+
; SSE-NEXT: pmulhw %xmm2, %xmm0
949+
; SSE-NEXT: pmulhw %xmm2, %xmm1
950+
; SSE-NEXT: retq
951+
;
952+
; AVX-LABEL: zext_mulhuw_v16i16_negative_constant:
953+
; AVX: # %bb.0:
954+
; AVX-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
955+
; AVX-NEXT: vpmulhw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [64536,64536,64536,64536,64536,64536,64536,64536,64536,64536,64536,64536,64536,64536,64536,64536]
956+
; AVX-NEXT: retq
957+
%k = and <16 x i16> %a, <i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767>
958+
%x = zext nneg <16 x i16> %k to <16 x i32>
959+
%m = mul nsw <16 x i32> %x, <i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000, i32 -1000>
960+
%s = lshr <16 x i32> %m, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
961+
%t = trunc nuw <16 x i32> %s to <16 x i16>
962+
ret <16 x i16> %t
963+
}
964+
965+
; PR109790
966+
define <16 x i16> @zext_mulhuw_v16i16_positive_constant(<16 x i16> %a) {
967+
; SSE-LABEL: zext_mulhuw_v16i16_positive_constant:
968+
; SSE: # %bb.0:
969+
; SSE-NEXT: movdqa {{.*#+}} xmm2 = [32767,32767,32767,32767,32767,32767,32767,32767]
970+
; SSE-NEXT: pand %xmm2, %xmm1
971+
; SSE-NEXT: pand %xmm2, %xmm0
972+
; SSE-NEXT: movdqa {{.*#+}} xmm2 = [1000,1000,1000,1000,1000,1000,1000,1000]
973+
; SSE-NEXT: pmulhw %xmm2, %xmm0
974+
; SSE-NEXT: pmulhw %xmm2, %xmm1
975+
; SSE-NEXT: retq
976+
;
977+
; AVX2-LABEL: zext_mulhuw_v16i16_positive_constant:
978+
; AVX2: # %bb.0:
979+
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
980+
; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
981+
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0
982+
; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
983+
; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm2 = [1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0]
984+
; AVX2-NEXT: vpmulhuw %ymm2, %ymm0, %ymm0
985+
; AVX2-NEXT: vpmulhuw %ymm2, %ymm1, %ymm1
986+
; AVX2-NEXT: vpackusdw %ymm0, %ymm1, %ymm0
987+
; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
988+
; AVX2-NEXT: retq
989+
;
990+
; AVX512F-LABEL: zext_mulhuw_v16i16_positive_constant:
991+
; AVX512F: # %bb.0:
992+
; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
993+
; AVX512F-NEXT: vpmulhw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [1000,1000,1000,1000,1000,1000,1000,1000,1000,1000,1000,1000,1000,1000,1000,1000]
994+
; AVX512F-NEXT: retq
995+
;
996+
; AVX512BW-LABEL: zext_mulhuw_v16i16_positive_constant:
997+
; AVX512BW: # %bb.0:
998+
; AVX512BW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
999+
; AVX512BW-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
1000+
; AVX512BW-NEXT: vpmulhuw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0 # [1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0,1000,0]
1001+
; AVX512BW-NEXT: vpmovdw %zmm0, %ymm0
1002+
; AVX512BW-NEXT: retq
1003+
%k = and <16 x i16> %a, <i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767, i16 32767>
1004+
%x = zext nneg <16 x i16> %k to <16 x i32>
1005+
%m = mul nuw nsw <16 x i32> %x, <i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000, i32 1000>
1006+
%s = lshr <16 x i32> %m, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
1007+
%t = trunc nuw nsw <16 x i32> %s to <16 x i16>
1008+
ret <16 x i16> %t
1009+
}
1010+
9401011
define <16 x i32> @mulhsw_v16i16_lshr(<16 x i16> %a, <16 x i16> %b) {
9411012
; SSE2-LABEL: mulhsw_v16i16_lshr:
9421013
; SSE2: # %bb.0:
@@ -2056,3 +2127,4 @@ define <8 x i16> @sse2_pmulhu_w_const(<8 x i16> %a0, <8 x i16> %a1) {
20562127
ret <8 x i16> %res
20572128
}
20582129
declare <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16>, <8 x i16>)
2130+

0 commit comments

Comments
 (0)