Lines Matching refs:avx2
65 %1 = tail call <16 x i16> @llvm.x86.avx2.psrai.w(<16 x i16> %v, i32 0)
74 %1 = tail call <16 x i16> @llvm.x86.avx2.psrai.w(<16 x i16> %v, i32 15)
83 %1 = tail call <16 x i16> @llvm.x86.avx2.psrai.w(<16 x i16> %v, i32 64)
91 %1 = tail call <8 x i32> @llvm.x86.avx2.psrai.d(<8 x i32> %v, i32 0)
100 %1 = tail call <8 x i32> @llvm.x86.avx2.psrai.d(<8 x i32> %v, i32 15)
109 %1 = tail call <8 x i32> @llvm.x86.avx2.psrai.d(<8 x i32> %v, i32 64)
196 %1 = tail call <16 x i16> @llvm.x86.avx2.psrli.w(<16 x i16> %v, i32 0)
205 %1 = tail call <16 x i16> @llvm.x86.avx2.psrli.w(<16 x i16> %v, i32 15)
213 %1 = tail call <16 x i16> @llvm.x86.avx2.psrli.w(<16 x i16> %v, i32 64)
221 %1 = tail call <8 x i32> @llvm.x86.avx2.psrli.d(<8 x i32> %v, i32 0)
230 %1 = tail call <8 x i32> @llvm.x86.avx2.psrli.d(<8 x i32> %v, i32 15)
238 %1 = tail call <8 x i32> @llvm.x86.avx2.psrli.d(<8 x i32> %v, i32 64)
246 %1 = tail call <4 x i64> @llvm.x86.avx2.psrli.q(<4 x i64> %v, i32 0)
255 %1 = tail call <4 x i64> @llvm.x86.avx2.psrli.q(<4 x i64> %v, i32 15)
263 %1 = tail call <4 x i64> @llvm.x86.avx2.psrli.q(<4 x i64> %v, i32 64)
350 %1 = tail call <16 x i16> @llvm.x86.avx2.pslli.w(<16 x i16> %v, i32 0)
359 %1 = tail call <16 x i16> @llvm.x86.avx2.pslli.w(<16 x i16> %v, i32 15)
367 %1 = tail call <16 x i16> @llvm.x86.avx2.pslli.w(<16 x i16> %v, i32 64)
375 %1 = tail call <8 x i32> @llvm.x86.avx2.pslli.d(<8 x i32> %v, i32 0)
384 %1 = tail call <8 x i32> @llvm.x86.avx2.pslli.d(<8 x i32> %v, i32 15)
392 %1 = tail call <8 x i32> @llvm.x86.avx2.pslli.d(<8 x i32> %v, i32 64)
400 %1 = tail call <4 x i64> @llvm.x86.avx2.pslli.q(<4 x i64> %v, i32 0)
409 %1 = tail call <4 x i64> @llvm.x86.avx2.pslli.q(<4 x i64> %v, i32 15)
417 %1 = tail call <4 x i64> @llvm.x86.avx2.pslli.q(<4 x i64> %v, i32 64)
499 %1 = tail call <16 x i16> @llvm.x86.avx2.psra.w(<16 x i16> %v, <8 x i16> zeroinitializer)
508 …%1 = tail call <16 x i16> @llvm.x86.avx2.psra.w(<16 x i16> %v, <8 x i16> <i16 15, i16 0, i16 0, i1…
517 …%1 = tail call <16 x i16> @llvm.x86.avx2.psra.w(<16 x i16> %v, <8 x i16> <i16 15, i16 15, i16 15, …
526 …%1 = tail call <16 x i16> @llvm.x86.avx2.psra.w(<16 x i16> %v, <8 x i16> <i16 64, i16 0, i16 0, i1…
534 %1 = tail call <8 x i32> @llvm.x86.avx2.psra.d(<8 x i32> %v, <4 x i32> zeroinitializer)
543 …%1 = tail call <8 x i32> @llvm.x86.avx2.psra.d(<8 x i32> %v, <4 x i32> <i32 15, i32 0, i32 9999, i…
552 …%1 = tail call <8 x i32> @llvm.x86.avx2.psra.d(<8 x i32> %v, <4 x i32> <i32 15, i32 15, i32 15, i3…
561 …%1 = tail call <8 x i32> @llvm.x86.avx2.psra.d(<8 x i32> %v, <4 x i32> <i32 64, i32 0, i32 9999, i…
664 %1 = tail call <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16> %v, <8 x i16> zeroinitializer)
673 …%1 = tail call <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16> %v, <8 x i16> <i16 15, i16 0, i16 0, i1…
681 …%1 = tail call <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16> %v, <8 x i16> <i16 15, i16 15, i16 15, …
689 …%1 = tail call <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16> %v, <8 x i16> <i16 64, i16 0, i16 0, i1…
697 %1 = tail call <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32> %v, <4 x i32> zeroinitializer)
706 …%1 = tail call <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32> %v, <4 x i32> <i32 15, i32 0, i32 9999, i…
714 …%1 = tail call <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32> %v, <4 x i32> <i32 15, i32 15, i32 15, i3…
722 …%1 = tail call <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32> %v, <4 x i32> <i32 64, i32 0, i32 9999, i…
730 %1 = tail call <4 x i64> @llvm.x86.avx2.psrl.q(<4 x i64> %v, <2 x i64> zeroinitializer)
739 %1 = tail call <4 x i64> @llvm.x86.avx2.psrl.q(<4 x i64> %v, <2 x i64> <i64 15, i64 9999>)
747 %1 = tail call <4 x i64> @llvm.x86.avx2.psrl.q(<4 x i64> %v, <2 x i64> <i64 64, i64 9999>)
850 %1 = tail call <16 x i16> @llvm.x86.avx2.psll.w(<16 x i16> %v, <8 x i16> zeroinitializer)
859 …%1 = tail call <16 x i16> @llvm.x86.avx2.psll.w(<16 x i16> %v, <8 x i16> <i16 15, i16 0, i16 0, i1…
867 …%1 = tail call <16 x i16> @llvm.x86.avx2.psll.w(<16 x i16> %v, <8 x i16> <i16 15, i16 15, i16 15, …
875 …%1 = tail call <16 x i16> @llvm.x86.avx2.psll.w(<16 x i16> %v, <8 x i16> <i16 64, i16 0, i16 0, i1…
883 %1 = tail call <8 x i32> @llvm.x86.avx2.psll.d(<8 x i32> %v, <4 x i32> zeroinitializer)
892 …%1 = tail call <8 x i32> @llvm.x86.avx2.psll.d(<8 x i32> %v, <4 x i32> <i32 15, i32 0, i32 9999, i…
900 …%1 = tail call <8 x i32> @llvm.x86.avx2.psll.d(<8 x i32> %v, <4 x i32> <i32 15, i32 15, i32 15, i3…
908 …%1 = tail call <8 x i32> @llvm.x86.avx2.psll.d(<8 x i32> %v, <4 x i32> <i32 64, i32 0, i32 9999, i…
916 %1 = tail call <4 x i64> @llvm.x86.avx2.psll.q(<4 x i64> %v, <2 x i64> zeroinitializer)
925 %1 = tail call <4 x i64> @llvm.x86.avx2.psll.q(<4 x i64> %v, <2 x i64> <i64 15, i64 9999>)
933 %1 = tail call <4 x i64> @llvm.x86.avx2.psll.q(<4 x i64> %v, <2 x i64> <i64 64, i64 9999>)
945 %1 = tail call <4 x i32> @llvm.x86.avx2.psrav.d(<4 x i32> %v, <4 x i32> zeroinitializer)
953 %1 = tail call <8 x i32> @llvm.x86.avx2.psrav.d.256(<8 x i32> %v, <8 x i32> zeroinitializer)
962 …%1 = tail call <4 x i32> @llvm.x86.avx2.psrav.d(<4 x i32> %v, <4 x i32> <i32 0, i32 8, i32 16, i32…
971 …%1 = tail call <8 x i32> @llvm.x86.avx2.psrav.d.256(<8 x i32> %v, <8 x i32> <i32 0, i32 8, i32 16,…
980 …%1 = tail call <4 x i32> @llvm.x86.avx2.psrav.d(<4 x i32> %v, <4 x i32> <i32 32, i32 100, i32 -255…
989 …%1 = tail call <8 x i32> @llvm.x86.avx2.psrav.d.256(<8 x i32> %v, <8 x i32> <i32 undef, i32 100, i…
999 %2 = tail call <4 x i32> @llvm.x86.avx2.psrav.d(<4 x i32> %v, <4 x i32> %1)
1009 %2 = tail call <8 x i32> @llvm.x86.avx2.psrav.d.256(<8 x i32> %v, <8 x i32> %1)
1021 %1 = tail call <4 x i32> @llvm.x86.avx2.psrlv.d(<4 x i32> %v, <4 x i32> zeroinitializer)
1029 %1 = tail call <8 x i32> @llvm.x86.avx2.psrlv.d.256(<8 x i32> %v, <8 x i32> zeroinitializer)
1038 …%1 = tail call <4 x i32> @llvm.x86.avx2.psrlv.d(<4 x i32> %v, <4 x i32> <i32 0, i32 8, i32 16, i32…
1047 …%1 = tail call <8 x i32> @llvm.x86.avx2.psrlv.d.256(<8 x i32> %v, <8 x i32> <i32 0, i32 8, i32 16,…
1053 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x i32> @llvm.x86.avx2.psrlv.d(<4 x i32> %v, <4 x i32> …
1056 …%1 = tail call <4 x i32> @llvm.x86.avx2.psrlv.d(<4 x i32> %v, <4 x i32> <i32 0, i32 8, i32 16, i32…
1062 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x i32> @llvm.x86.avx2.psrlv.d.256(<8 x i32> %v, <8 x i…
1065 …%1 = tail call <8 x i32> @llvm.x86.avx2.psrlv.d.256(<8 x i32> %v, <8 x i32> <i32 0, i32 8, i32 16,…
1073 …%1 = tail call <4 x i32> @llvm.x86.avx2.psrlv.d(<4 x i32> %v, <4 x i32> <i32 32, i32 100, i32 -255…
1081 …%1 = tail call <8 x i32> @llvm.x86.avx2.psrlv.d.256(<8 x i32> %v, <8 x i32> <i32 undef, i32 100, i…
1091 %2 = tail call <4 x i32> @llvm.x86.avx2.psrlv.d(<4 x i32> %v, <4 x i32> %1)
1101 %2 = tail call <8 x i32> @llvm.x86.avx2.psrlv.d.256(<8 x i32> %v, <8 x i32> %1)
1109 %1 = tail call <2 x i64> @llvm.x86.avx2.psrlv.q(<2 x i64> %v, <2 x i64> zeroinitializer)
1117 %1 = tail call <4 x i64> @llvm.x86.avx2.psrlv.q.256(<4 x i64> %v, <4 x i64> zeroinitializer)
1126 %1 = tail call <2 x i64> @llvm.x86.avx2.psrlv.q(<2 x i64> %v, <2 x i64> <i64 0, i64 8>)
1135 …%1 = tail call <4 x i64> @llvm.x86.avx2.psrlv.q.256(<4 x i64> %v, <4 x i64> <i64 0, i64 8, i64 16,…
1141 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <2 x i64> @llvm.x86.avx2.psrlv.q(<2 x i64> %v, <2 x i64> …
1144 %1 = tail call <2 x i64> @llvm.x86.avx2.psrlv.q(<2 x i64> %v, <2 x i64> <i64 0, i64 128>)
1150 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x i64> @llvm.x86.avx2.psrlv.q.256(<4 x i64> %v, <4 x i…
1153 …%1 = tail call <4 x i64> @llvm.x86.avx2.psrlv.q.256(<4 x i64> %v, <4 x i64> <i64 0, i64 8, i64 16,…
1161 %1 = tail call <2 x i64> @llvm.x86.avx2.psrlv.q(<2 x i64> %v, <2 x i64> <i64 128, i64 -64>)
1169 …%1 = tail call <4 x i64> @llvm.x86.avx2.psrlv.q.256(<4 x i64> %v, <4 x i64> <i64 64, i64 undef, i6…
1179 %2 = tail call <2 x i64> @llvm.x86.avx2.psrlv.q(<2 x i64> %v, <2 x i64> %1)
1189 %2 = tail call <4 x i64> @llvm.x86.avx2.psrlv.q.256(<4 x i64> %v, <4 x i64> %1)
1201 %1 = tail call <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32> %v, <4 x i32> zeroinitializer)
1209 %1 = tail call <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32> %v, <8 x i32> zeroinitializer)
1218 …%1 = tail call <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32> %v, <4 x i32> <i32 0, i32 8, i32 16, i32…
1227 …%1 = tail call <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32> %v, <8 x i32> <i32 0, i32 8, i32 16,…
1233 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32> %v, <4 x i32> …
1236 …%1 = tail call <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32> %v, <4 x i32> <i32 0, i32 8, i32 16, i32…
1242 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32> %v, <8 x i…
1245 …%1 = tail call <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32> %v, <8 x i32> <i32 0, i32 8, i32 16,…
1253 …%1 = tail call <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32> %v, <4 x i32> <i32 32, i32 100, i32 -255…
1261 …%1 = tail call <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32> %v, <8 x i32> <i32 undef, i32 100, i…
1271 %2 = tail call <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32> %v, <4 x i32> %1)
1281 %2 = tail call <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32> %v, <8 x i32> %1)
1289 %1 = tail call <2 x i64> @llvm.x86.avx2.psllv.q(<2 x i64> %v, <2 x i64> zeroinitializer)
1297 %1 = tail call <4 x i64> @llvm.x86.avx2.psllv.q.256(<4 x i64> %v, <4 x i64> zeroinitializer)
1306 %1 = tail call <2 x i64> @llvm.x86.avx2.psllv.q(<2 x i64> %v, <2 x i64> <i64 0, i64 8>)
1315 …%1 = tail call <4 x i64> @llvm.x86.avx2.psllv.q.256(<4 x i64> %v, <4 x i64> <i64 0, i64 8, i64 16,…
1321 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <2 x i64> @llvm.x86.avx2.psllv.q(<2 x i64> %v, <2 x i64> …
1324 %1 = tail call <2 x i64> @llvm.x86.avx2.psllv.q(<2 x i64> %v, <2 x i64> <i64 0, i64 128>)
1330 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x i64> @llvm.x86.avx2.psllv.q.256(<4 x i64> %v, <4 x i…
1333 …%1 = tail call <4 x i64> @llvm.x86.avx2.psllv.q.256(<4 x i64> %v, <4 x i64> <i64 0, i64 8, i64 16,…
1341 %1 = tail call <2 x i64> @llvm.x86.avx2.psllv.q(<2 x i64> %v, <2 x i64> <i64 128, i64 -64>)
1349 …%1 = tail call <4 x i64> @llvm.x86.avx2.psllv.q.256(<4 x i64> %v, <4 x i64> <i64 64, i64 undef, i6…
1359 %2 = tail call <2 x i64> @llvm.x86.avx2.psllv.q(<2 x i64> %v, <2 x i64> %1)
1369 %2 = tail call <4 x i64> @llvm.x86.avx2.psllv.q.256(<4 x i64> %v, <4 x i64> %1)
1423 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x i16> @llvm.x86.avx2.psra.w(<16 x i16> %v, <8 x i16>…
1427 %2 = tail call <16 x i16> @llvm.x86.avx2.psra.w(<16 x i16> %v, <8 x i16> %1)
1433 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x i32> @llvm.x86.avx2.psra.d(<8 x i32> %v, <4 x i32> %…
1437 %2 = tail call <8 x i32> @llvm.x86.avx2.psra.d(<8 x i32> %v, <4 x i32> %1)
1473 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16> %v, <8 x i16>…
1477 %2 = tail call <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16> %v, <8 x i16> %1)
1484 ; CHECK-NEXT: [[TMP2:%.*]] = tail call <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16> %v, <8 x i16>…
1489 %3 = tail call <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16> %v, <8 x i16> %2)
1495 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32> %v, <4 x i32> %…
1499 %2 = tail call <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32> %v, <4 x i32> %1)
1506 ; CHECK-NEXT: [[TMP2:%.*]] = tail call <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32> %v, <4 x i32> […
1511 %3 = tail call <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32> %v, <4 x i32> %2)
1517 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x i64> @llvm.x86.avx2.psrl.q(<4 x i64> %v, <2 x i64> %…
1521 %2 = tail call <4 x i64> @llvm.x86.avx2.psrl.q(<4 x i64> %v, <2 x i64> %1)
1557 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x i16> @llvm.x86.avx2.psll.w(<16 x i16> %v, <8 x i16>…
1561 %2 = tail call <16 x i16> @llvm.x86.avx2.psll.w(<16 x i16> %v, <8 x i16> %1)
1567 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x i32> @llvm.x86.avx2.psll.d(<8 x i32> %v, <4 x i32> %…
1571 %2 = tail call <8 x i32> @llvm.x86.avx2.psll.d(<8 x i32> %v, <4 x i32> %1)
1577 ; CHECK-NEXT: [[TMP1:%.*]] = tail call <4 x i64> @llvm.x86.avx2.psll.q(<4 x i64> %v, <2 x i64> %…
1581 %2 = tail call <4 x i64> @llvm.x86.avx2.psll.q(<4 x i64> %v, <2 x i64> %1)
1635 %1 = tail call <16 x i16> @llvm.x86.avx2.psrai.w(<16 x i16> %A, i32 0)
1636 …%2 = tail call <16 x i16> @llvm.x86.avx2.psra.w(<16 x i16> %1, <8 x i16> <i16 0, i16 0, i16 0, i16…
1637 %3 = tail call <16 x i16> @llvm.x86.avx2.psrai.w(<16 x i16> %2, i32 0)
1646 %2 = tail call <16 x i16> @llvm.x86.avx2.psrai.w(<16 x i16> %1, i32 3)
1647 …%3 = tail call <16 x i16> @llvm.x86.avx2.psra.w(<16 x i16> %2, <8 x i16> <i16 3, i16 0, i16 0, i16…
1648 %4 = tail call <16 x i16> @llvm.x86.avx2.psrai.w(<16 x i16> %3, i32 2)
1656 %1 = tail call <8 x i32> @llvm.x86.avx2.psrai.d(<8 x i32> %A, i32 0)
1657 …%2 = tail call <8 x i32> @llvm.x86.avx2.psra.d(<8 x i32> %1, <4 x i32> <i32 0, i32 0, i32 7, i32 0…
1658 %3 = tail call <8 x i32> @llvm.x86.avx2.psrai.d(<8 x i32> %2, i32 0)
1667 %2 = tail call <8 x i32> @llvm.x86.avx2.psrai.d(<8 x i32> %1, i32 3)
1668 …%3 = tail call <8 x i32> @llvm.x86.avx2.psra.d(<8 x i32> %2, <4 x i32> <i32 3, i32 0, i32 7, i32 0…
1669 %4 = tail call <8 x i32> @llvm.x86.avx2.psrai.d(<8 x i32> %3, i32 2)
1710 …%5 = tail call <16 x i16> @llvm.x86.avx2.psll.w(<16 x i16> <i16 1, i16 0, i16 0, i16 0, i16 2, i16…
1713 %8 = tail call <8 x i32> @llvm.x86.avx2.psll.d(<8 x i32> %6, <4 x i32> %7)
1715 %10 = tail call <4 x i64> @llvm.x86.avx2.psll.q(<4 x i64> %9, <2 x i64> %3)
1717 %12 = tail call <16 x i16> @llvm.x86.avx2.pslli.w(<16 x i16> %11, i32 %S)
1719 %14 = tail call <8 x i32> @llvm.x86.avx2.pslli.d(<8 x i32> %13, i32 %S)
1721 %16 = tail call <4 x i64> @llvm.x86.avx2.pslli.q(<4 x i64> %15, i32 %S)
1758 …%5 = tail call <16 x i16> @llvm.x86.avx2.psll.w(<16 x i16> <i16 1, i16 0, i16 0, i16 0, i16 2, i16…
1761 %8 = tail call <8 x i32> @llvm.x86.avx2.psll.d(<8 x i32> %6, <4 x i32> %7)
1763 %10 = tail call <4 x i64> @llvm.x86.avx2.psll.q(<4 x i64> %9, <2 x i64> %3)
1765 %12 = tail call <16 x i16> @llvm.x86.avx2.pslli.w(<16 x i16> %11, i32 %S)
1767 %14 = tail call <8 x i32> @llvm.x86.avx2.pslli.d(<8 x i32> %13, i32 %S)
1769 %16 = tail call <4 x i64> @llvm.x86.avx2.pslli.q(<4 x i64> %15, i32 %S)
1805 …%5 = tail call <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16> <i16 1024, i16 0, i16 0, i16 0, i16 204…
1808 %8 = tail call <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32> %6, <4 x i32> %7)
1810 %10 = tail call <4 x i64> @llvm.x86.avx2.psrl.q(<4 x i64> %9, <2 x i64> %3)
1812 %12 = tail call <16 x i16> @llvm.x86.avx2.psrli.w(<16 x i16> %11, i32 %S)
1814 %14 = tail call <8 x i32> @llvm.x86.avx2.psrli.d(<8 x i32> %13, i32 %S)
1816 %16 = tail call <4 x i64> @llvm.x86.avx2.psrli.q(<4 x i64> %15, i32 %S)
1853 …%5 = tail call <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16> <i16 1024, i16 0, i16 0, i16 0, i16 204…
1856 %8 = tail call <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32> %6, <4 x i32> %7)
1858 %10 = tail call <4 x i64> @llvm.x86.avx2.psrl.q(<4 x i64> %9, <2 x i64> %3)
1860 %12 = tail call <16 x i16> @llvm.x86.avx2.psrli.w(<16 x i16> %11, i32 %S)
1862 %14 = tail call <8 x i32> @llvm.x86.avx2.psrli.d(<8 x i32> %13, i32 %S)
1864 %16 = tail call <4 x i64> @llvm.x86.avx2.psrli.q(<4 x i64> %15, i32 %S)
1868 declare <4 x i64> @llvm.x86.avx2.pslli.q(<4 x i64>, i32) #1
1869 declare <8 x i32> @llvm.x86.avx2.pslli.d(<8 x i32>, i32) #1
1870 declare <16 x i16> @llvm.x86.avx2.pslli.w(<16 x i16>, i32) #1
1871 declare <4 x i64> @llvm.x86.avx2.psll.q(<4 x i64>, <2 x i64>) #1
1872 declare <8 x i32> @llvm.x86.avx2.psll.d(<8 x i32>, <4 x i32>) #1
1873 declare <16 x i16> @llvm.x86.avx2.psll.w(<16 x i16>, <8 x i16>) #1
1881 declare <4 x i64> @llvm.x86.avx2.psrli.q(<4 x i64>, i32) #1
1882 declare <8 x i32> @llvm.x86.avx2.psrli.d(<8 x i32>, i32) #1
1883 declare <16 x i16> @llvm.x86.avx2.psrli.w(<16 x i16>, i32) #1
1884 declare <4 x i64> @llvm.x86.avx2.psrl.q(<4 x i64>, <2 x i64>) #1
1885 declare <8 x i32> @llvm.x86.avx2.psrl.d(<8 x i32>, <4 x i32>) #1
1886 declare <16 x i16> @llvm.x86.avx2.psrl.w(<16 x i16>, <8 x i16>) #1
1894 declare <8 x i32> @llvm.x86.avx2.psrai.d(<8 x i32>, i32) #1
1895 declare <16 x i16> @llvm.x86.avx2.psrai.w(<16 x i16>, i32) #1
1896 declare <8 x i32> @llvm.x86.avx2.psra.d(<8 x i32>, <4 x i32>) #1
1897 declare <16 x i16> @llvm.x86.avx2.psra.w(<16 x i16>, <8 x i16>) #1
1903 declare <4 x i32> @llvm.x86.avx2.psrav.d(<4 x i32>, <4 x i32>) #1
1904 declare <8 x i32> @llvm.x86.avx2.psrav.d.256(<8 x i32>, <8 x i32>) #1
1906 declare <4 x i32> @llvm.x86.avx2.psrlv.d(<4 x i32>, <4 x i32>) #1
1907 declare <8 x i32> @llvm.x86.avx2.psrlv.d.256(<8 x i32>, <8 x i32>) #1
1908 declare <2 x i64> @llvm.x86.avx2.psrlv.q(<2 x i64>, <2 x i64>) #1
1909 declare <4 x i64> @llvm.x86.avx2.psrlv.q.256(<4 x i64>, <4 x i64>) #1
1911 declare <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32>, <4 x i32>) #1
1912 declare <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32>, <8 x i32>) #1
1913 declare <2 x i64> @llvm.x86.avx2.psllv.q(<2 x i64>, <2 x i64>) #1
1914 declare <4 x i64> @llvm.x86.avx2.psllv.q.256(<4 x i64>, <4 x i64>) #1