[llvm] 88e0014 - [PhaseOrdering][X86] Add additional hadd/hsub test coverage
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Wed Jan 29 09:26:17 PST 2025
Author: Simon Pilgrim
Date: 2025-01-29T17:25:54Z
New Revision: 88e00141f81c4dfd48bca58fda15d078a138b586
URL: https://github.com/llvm/llvm-project/commit/88e00141f81c4dfd48bca58fda15d078a138b586
DIFF: https://github.com/llvm/llvm-project/commit/88e00141f81c4dfd48bca58fda15d078a138b586.diff
LOG: [PhaseOrdering][X86] Add additional hadd/hsub test coverage
Add v16i16 coverage and "reverse order hadd/hsub" tests
Added:
Modified:
llvm/test/Transforms/PhaseOrdering/X86/hadd.ll
llvm/test/Transforms/PhaseOrdering/X86/hsub.ll
Removed:
################################################################################
diff --git a/llvm/test/Transforms/PhaseOrdering/X86/hadd.ll b/llvm/test/Transforms/PhaseOrdering/X86/hadd.ll
index 0c9f279c01bae0..9b139a4a4e234c 100644
--- a/llvm/test/Transforms/PhaseOrdering/X86/hadd.ll
+++ b/llvm/test/Transforms/PhaseOrdering/X86/hadd.ll
@@ -119,6 +119,409 @@ define <8 x i16> @add_v8i16_u1234567(<8 x i16> %a, <8 x i16> %b) {
ret <8 x i16> %result
}
+define <8 x i16> @add_v8i16_76u43210(<8 x i16> %a, <8 x i16> %b) {
+; SSE2-LABEL: @add_v8i16_76u43210(
+; SSE2-NEXT: [[SHIFT:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> poison, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP1:%.*]] = add <8 x i16> [[A]], [[SHIFT]]
+; SSE2-NEXT: [[SHIFT2:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 5, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP2:%.*]] = add <8 x i16> [[B]], [[SHIFT2]]
+; SSE2-NEXT: [[SHIFT3:%.*]] = shufflevector <8 x i16> [[B]], <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 6>
+; SSE2-NEXT: [[TMP3:%.*]] = add <8 x i16> [[SHIFT3]], [[B]]
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 2, i32 4, i32 6, i32 8, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP5:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 3, i32 5, i32 7, i32 9, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP6:%.*]] = add <8 x i16> [[TMP4]], [[TMP5]]
+; SSE2-NEXT: [[HADD41:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP6]], <8 x i32> <i32 0, i32 8, i32 9, i32 10, i32 11, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[HADD6:%.*]] = shufflevector <8 x i16> [[HADD41]], <8 x i16> [[TMP2]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 poison, i32 12, i32 poison>
+; SSE2-NEXT: [[RESULT:%.*]] = shufflevector <8 x i16> [[TMP3]], <8 x i16> [[HADD6]], <8 x i32> <i32 7, i32 14, i32 poison, i32 12, i32 11, i32 10, i32 9, i32 8>
+; SSE2-NEXT: ret <8 x i16> [[RESULT]]
+;
+; SSE4-LABEL: @add_v8i16_76u43210(
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> poison, <8 x i32> <i32 0, i32 3, i32 4, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> poison, <8 x i32> <i32 1, i32 2, i32 5, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[HADD22:%.*]] = add <8 x i16> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: [[TMP3:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B:%.*]], <8 x i32> <i32 6, i32 8, i32 poison, i32 12, i32 14, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 7, i32 9, i32 poison, i32 13, i32 15, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP5:%.*]] = add <8 x i16> [[TMP3]], [[TMP4]]
+; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <8 x i16> [[TMP5]], <8 x i16> [[HADD22]], <8 x i32> <i32 4, i32 3, i32 poison, i32 1, i32 0, i32 10, i32 9, i32 8>
+; SSE4-NEXT: ret <8 x i16> [[RESULT]]
+;
+; AVX-LABEL: @add_v8i16_76u43210(
+; AVX-NEXT: [[TMP1:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> poison, <8 x i32> <i32 0, i32 3, i32 4, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX-NEXT: [[TMP2:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> poison, <8 x i32> <i32 1, i32 2, i32 5, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX-NEXT: [[HADD22:%.*]] = add <8 x i16> [[TMP1]], [[TMP2]]
+; AVX-NEXT: [[TMP3:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B:%.*]], <8 x i32> <i32 6, i32 8, i32 poison, i32 12, i32 14, i32 poison, i32 poison, i32 poison>
+; AVX-NEXT: [[TMP4:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 7, i32 9, i32 poison, i32 13, i32 15, i32 poison, i32 poison, i32 poison>
+; AVX-NEXT: [[TMP5:%.*]] = add <8 x i16> [[TMP3]], [[TMP4]]
+; AVX-NEXT: [[RESULT:%.*]] = shufflevector <8 x i16> [[TMP5]], <8 x i16> [[HADD22]], <8 x i32> <i32 4, i32 3, i32 poison, i32 1, i32 0, i32 10, i32 9, i32 8>
+; AVX-NEXT: ret <8 x i16> [[RESULT]]
+;
+ %a0 = extractelement <8 x i16> %a, i32 0
+ %a1 = extractelement <8 x i16> %a, i32 1
+ %a2 = extractelement <8 x i16> %a, i32 2
+ %a3 = extractelement <8 x i16> %a, i32 3
+ %a4 = extractelement <8 x i16> %a, i32 4
+ %a5 = extractelement <8 x i16> %a, i32 5
+ %a6 = extractelement <8 x i16> %a, i32 6
+ %a7 = extractelement <8 x i16> %a, i32 7
+ %a01 = add i16 %a0, %a1
+ %a23 = add i16 %a2, %a3
+ %a45 = add i16 %a4, %a5
+ %a67 = add i16 %a6, %a7
+ %b0 = extractelement <8 x i16> %b, i32 0
+ %b1 = extractelement <8 x i16> %b, i32 1
+ %b2 = extractelement <8 x i16> %b, i32 2
+ %b3 = extractelement <8 x i16> %b, i32 3
+ %b4 = extractelement <8 x i16> %b, i32 4
+ %b5 = extractelement <8 x i16> %b, i32 5
+ %b6 = extractelement <8 x i16> %b, i32 6
+ %b7 = extractelement <8 x i16> %b, i32 7
+ %b01 = add i16 %b0, %b1
+ %b23 = add i16 %b2, %b3
+ %b45 = add i16 %b4, %b5
+ %b67 = add i16 %b6, %b7
+ %hadd0 = insertelement <8 x i16> poison, i16 %a01, i32 0
+ %hadd1 = insertelement <8 x i16> %hadd0, i16 %a23, i32 1
+ %hadd2 = insertelement <8 x i16> %hadd1, i16 %a45, i32 2
+ %hadd3 = insertelement <8 x i16> %hadd2, i16 %a67, i32 3
+ %hadd4 = insertelement <8 x i16> %hadd3, i16 %b01, i32 4
+ %hadd5 = insertelement <8 x i16> %hadd4, i16 %b23, i32 5
+ %hadd6 = insertelement <8 x i16> %hadd5, i16 %b45, i32 6
+ %hadd7 = insertelement <8 x i16> %hadd6, i16 %b67, i32 7
+ %result = shufflevector <8 x i16> %hadd7, <8 x i16> %a, <8 x i32> <i32 7, i32 6, i32 poison, i32 4, i32 3, i32 2, i32 1, i32 0>
+ ret <8 x i16> %result
+}
+
+;
+; v16i16
+;
+
+define <16 x i16> @add_v16i16_0123456789ABCDEF(<16 x i16> %a, <16 x i16> %b) {
+; CHECK-LABEL: @add_v16i16_0123456789ABCDEF(
+; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 8, i32 10, i32 12, i32 14, i32 24, i32 26, i32 28, i32 30>
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 9, i32 11, i32 13, i32 15, i32 25, i32 27, i32 29, i32 31>
+; CHECK-NEXT: [[TMP3:%.*]] = add <16 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT: ret <16 x i16> [[TMP3]]
+;
+ %a0 = extractelement <16 x i16> %a, i32 0
+ %a1 = extractelement <16 x i16> %a, i32 1
+ %a2 = extractelement <16 x i16> %a, i32 2
+ %a3 = extractelement <16 x i16> %a, i32 3
+ %a4 = extractelement <16 x i16> %a, i32 4
+ %a5 = extractelement <16 x i16> %a, i32 5
+ %a6 = extractelement <16 x i16> %a, i32 6
+ %a7 = extractelement <16 x i16> %a, i32 7
+ %a8 = extractelement <16 x i16> %a, i32 8
+ %a9 = extractelement <16 x i16> %a, i32 9
+ %aA = extractelement <16 x i16> %a, i32 10
+ %aB = extractelement <16 x i16> %a, i32 11
+ %aC = extractelement <16 x i16> %a, i32 12
+ %aD = extractelement <16 x i16> %a, i32 13
+ %aE = extractelement <16 x i16> %a, i32 14
+ %aF = extractelement <16 x i16> %a, i32 15
+ %a01 = add i16 %a0, %a1
+ %a23 = add i16 %a2, %a3
+ %a45 = add i16 %a4, %a5
+ %a67 = add i16 %a6, %a7
+ %a89 = add i16 %a8, %a9
+ %aAB = add i16 %aA, %aB
+ %aCD = add i16 %aC, %aD
+ %aEF = add i16 %aE, %aF
+ %b0 = extractelement <16 x i16> %b, i32 0
+ %b1 = extractelement <16 x i16> %b, i32 1
+ %b2 = extractelement <16 x i16> %b, i32 2
+ %b3 = extractelement <16 x i16> %b, i32 3
+ %b4 = extractelement <16 x i16> %b, i32 4
+ %b5 = extractelement <16 x i16> %b, i32 5
+ %b6 = extractelement <16 x i16> %b, i32 6
+ %b7 = extractelement <16 x i16> %b, i32 7
+ %b8 = extractelement <16 x i16> %b, i32 8
+ %b9 = extractelement <16 x i16> %b, i32 9
+ %bA = extractelement <16 x i16> %b, i32 10
+ %bB = extractelement <16 x i16> %b, i32 11
+ %bC = extractelement <16 x i16> %b, i32 12
+ %bD = extractelement <16 x i16> %b, i32 13
+ %bE = extractelement <16 x i16> %b, i32 14
+ %bF = extractelement <16 x i16> %b, i32 15
+ %b01 = add i16 %b0, %b1
+ %b23 = add i16 %b2, %b3
+ %b45 = add i16 %b4, %b5
+ %b67 = add i16 %b6, %b7
+ %b89 = add i16 %b8, %b9
+ %bAB = add i16 %bA, %bB
+ %bCD = add i16 %bC, %bD
+ %bEF = add i16 %bE, %bF
+ %hadd0 = insertelement <16 x i16> poison, i16 %a01, i32 0
+ %hadd1 = insertelement <16 x i16> %hadd0, i16 %a23, i32 1
+ %hadd2 = insertelement <16 x i16> %hadd1, i16 %a45, i32 2
+ %hadd3 = insertelement <16 x i16> %hadd2, i16 %a67, i32 3
+ %hadd4 = insertelement <16 x i16> %hadd3, i16 %b01, i32 4
+ %hadd5 = insertelement <16 x i16> %hadd4, i16 %b23, i32 5
+ %hadd6 = insertelement <16 x i16> %hadd5, i16 %b45, i32 6
+ %hadd7 = insertelement <16 x i16> %hadd6, i16 %b67, i32 7
+ %hadd8 = insertelement <16 x i16> %hadd7, i16 %a89, i32 8
+ %hadd9 = insertelement <16 x i16> %hadd8, i16 %aAB, i32 9
+ %haddA = insertelement <16 x i16> %hadd9, i16 %aCD, i32 10
+ %haddB = insertelement <16 x i16> %haddA, i16 %aEF, i32 11
+ %haddC = insertelement <16 x i16> %haddB, i16 %b89, i32 12
+ %haddD = insertelement <16 x i16> %haddC, i16 %bAB, i32 13
+ %haddE = insertelement <16 x i16> %haddD, i16 %bCD, i32 14
+ %haddF = insertelement <16 x i16> %haddE, i16 %bEF, i32 15
+ %result = shufflevector <16 x i16> %haddF, <16 x i16> %a, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+ ret <16 x i16> %result
+}
+
+define <16 x i16> @add_v16i16_0123u56789uBCDEF(<16 x i16> %a, <16 x i16> %b) {
+; SSE2-LABEL: @add_v16i16_0123u56789uBCDEF(
+; SSE2-NEXT: [[A8:%.*]] = extractelement <16 x i16> [[A:%.*]], i64 8
+; SSE2-NEXT: [[A9:%.*]] = extractelement <16 x i16> [[A]], i64 9
+; SSE2-NEXT: [[A89:%.*]] = add i16 [[A8]], [[A9]]
+; SSE2-NEXT: [[BC:%.*]] = extractelement <16 x i16> [[B:%.*]], i64 12
+; SSE2-NEXT: [[BD:%.*]] = extractelement <16 x i16> [[B]], i64 13
+; SSE2-NEXT: [[BE:%.*]] = extractelement <16 x i16> [[B]], i64 14
+; SSE2-NEXT: [[BF:%.*]] = extractelement <16 x i16> [[B]], i64 15
+; SSE2-NEXT: [[BCD:%.*]] = add i16 [[BC]], [[BD]]
+; SSE2-NEXT: [[BEF:%.*]] = add i16 [[BE]], [[BF]]
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 poison, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 poison, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP3:%.*]] = add <16 x i16> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: [[HADD8:%.*]] = insertelement <16 x i16> [[TMP3]], i16 [[A89]], i64 8
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 10, i32 poison, i32 14, i32 24, i32 26, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP5:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 11, i32 poison, i32 15, i32 25, i32 27, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP6:%.*]] = add <16 x i16> [[TMP4]], [[TMP5]]
+; SSE2-NEXT: [[HADDD1:%.*]] = shufflevector <16 x i16> [[HADD8]], <16 x i16> [[TMP6]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 16, i32 poison, i32 18, i32 19, i32 20, i32 poison, i32 poison>
+; SSE2-NEXT: [[HADDE:%.*]] = insertelement <16 x i16> [[HADDD1]], i16 [[BCD]], i64 14
+; SSE2-NEXT: [[HADDF:%.*]] = insertelement <16 x i16> [[HADDE]], i16 [[BEF]], i64 15
+; SSE2-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[HADDF]], <16 x i16> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 9, i32 poison, i32 11, i32 12, i32 13, i32 14, i32 15>
+; SSE2-NEXT: ret <16 x i16> [[RESULT]]
+;
+; SSE4-LABEL: @add_v16i16_0123u56789uBCDEF(
+; SSE4-NEXT: [[A8:%.*]] = extractelement <16 x i16> [[A:%.*]], i64 8
+; SSE4-NEXT: [[A9:%.*]] = extractelement <16 x i16> [[A]], i64 9
+; SSE4-NEXT: [[A89:%.*]] = add i16 [[A8]], [[A9]]
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 poison, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 poison, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP3:%.*]] = add <16 x i16> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: [[HADD8:%.*]] = insertelement <16 x i16> [[TMP3]], i16 [[A89]], i64 8
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> poison, <16 x i32> <i32 10, i32 poison, i32 14, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> poison, <16 x i32> <i32 11, i32 poison, i32 15, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP6:%.*]] = add <16 x i16> [[TMP4]], [[TMP5]]
+; SSE4-NEXT: [[HADDB2:%.*]] = shufflevector <16 x i16> [[HADD8]], <16 x i16> [[TMP6]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 16, i32 poison, i32 18, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP7:%.*]] = shufflevector <16 x i16> [[B]], <16 x i16> poison, <16 x i32> <i32 8, i32 10, i32 12, i32 14, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP8:%.*]] = shufflevector <16 x i16> [[B]], <16 x i16> poison, <16 x i32> <i32 9, i32 11, i32 13, i32 15, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP9:%.*]] = add <16 x i16> [[TMP7]], [[TMP8]]
+; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[HADDB2]], <16 x i16> [[TMP9]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 9, i32 poison, i32 11, i32 16, i32 17, i32 18, i32 19>
+; SSE4-NEXT: ret <16 x i16> [[RESULT]]
+;
+; AVX2-LABEL: @add_v16i16_0123u56789uBCDEF(
+; AVX2-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 poison, i32 18, i32 20, i32 22, i32 9, i32 10, i32 poison, i32 14, i32 24, i32 26, i32 28, i32 30>
+; AVX2-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 poison, i32 19, i32 21, i32 23, i32 8, i32 11, i32 poison, i32 15, i32 25, i32 27, i32 29, i32 31>
+; AVX2-NEXT: [[RESULT:%.*]] = add <16 x i16> [[TMP1]], [[TMP2]]
+; AVX2-NEXT: ret <16 x i16> [[RESULT]]
+;
+; AVX512-LABEL: @add_v16i16_0123u56789uBCDEF(
+; AVX512-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 poison, i32 18, i32 20, i32 22, i32 8, i32 10, i32 poison, i32 14, i32 24, i32 26, i32 28, i32 30>
+; AVX512-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 poison, i32 19, i32 21, i32 23, i32 9, i32 11, i32 poison, i32 15, i32 25, i32 27, i32 29, i32 31>
+; AVX512-NEXT: [[RESULT:%.*]] = add <16 x i16> [[TMP1]], [[TMP2]]
+; AVX512-NEXT: ret <16 x i16> [[RESULT]]
+;
+ %a0 = extractelement <16 x i16> %a, i32 0
+ %a1 = extractelement <16 x i16> %a, i32 1
+ %a2 = extractelement <16 x i16> %a, i32 2
+ %a3 = extractelement <16 x i16> %a, i32 3
+ %a4 = extractelement <16 x i16> %a, i32 4
+ %a5 = extractelement <16 x i16> %a, i32 5
+ %a6 = extractelement <16 x i16> %a, i32 6
+ %a7 = extractelement <16 x i16> %a, i32 7
+ %a8 = extractelement <16 x i16> %a, i32 8
+ %a9 = extractelement <16 x i16> %a, i32 9
+ %aA = extractelement <16 x i16> %a, i32 10
+ %aB = extractelement <16 x i16> %a, i32 11
+ %aC = extractelement <16 x i16> %a, i32 12
+ %aD = extractelement <16 x i16> %a, i32 13
+ %aE = extractelement <16 x i16> %a, i32 14
+ %aF = extractelement <16 x i16> %a, i32 15
+ %a01 = add i16 %a0, %a1
+ %a23 = add i16 %a2, %a3
+ %a45 = add i16 %a4, %a5
+ %a67 = add i16 %a6, %a7
+ %a89 = add i16 %a8, %a9
+ %aAB = add i16 %aA, %aB
+ %aCD = add i16 %aC, %aD
+ %aEF = add i16 %aE, %aF
+ %b0 = extractelement <16 x i16> %b, i32 0
+ %b1 = extractelement <16 x i16> %b, i32 1
+ %b2 = extractelement <16 x i16> %b, i32 2
+ %b3 = extractelement <16 x i16> %b, i32 3
+ %b4 = extractelement <16 x i16> %b, i32 4
+ %b5 = extractelement <16 x i16> %b, i32 5
+ %b6 = extractelement <16 x i16> %b, i32 6
+ %b7 = extractelement <16 x i16> %b, i32 7
+ %b8 = extractelement <16 x i16> %b, i32 8
+ %b9 = extractelement <16 x i16> %b, i32 9
+ %bA = extractelement <16 x i16> %b, i32 10
+ %bB = extractelement <16 x i16> %b, i32 11
+ %bC = extractelement <16 x i16> %b, i32 12
+ %bD = extractelement <16 x i16> %b, i32 13
+ %bE = extractelement <16 x i16> %b, i32 14
+ %bF = extractelement <16 x i16> %b, i32 15
+ %b01 = add i16 %b0, %b1
+ %b23 = add i16 %b2, %b3
+ %b45 = add i16 %b4, %b5
+ %b67 = add i16 %b6, %b7
+ %b89 = add i16 %b8, %b9
+ %bAB = add i16 %bA, %bB
+ %bCD = add i16 %bC, %bD
+ %bEF = add i16 %bE, %bF
+ %hadd0 = insertelement <16 x i16> poison, i16 %a01, i32 0
+ %hadd1 = insertelement <16 x i16> %hadd0, i16 %a23, i32 1
+ %hadd2 = insertelement <16 x i16> %hadd1, i16 %a45, i32 2
+ %hadd3 = insertelement <16 x i16> %hadd2, i16 %a67, i32 3
+ %hadd4 = insertelement <16 x i16> %hadd3, i16 %b01, i32 4
+ %hadd5 = insertelement <16 x i16> %hadd4, i16 %b23, i32 5
+ %hadd6 = insertelement <16 x i16> %hadd5, i16 %b45, i32 6
+ %hadd7 = insertelement <16 x i16> %hadd6, i16 %b67, i32 7
+ %hadd8 = insertelement <16 x i16> %hadd7, i16 %a89, i32 8
+ %hadd9 = insertelement <16 x i16> %hadd8, i16 %aAB, i32 9
+ %haddA = insertelement <16 x i16> %hadd9, i16 %aCD, i32 10
+ %haddB = insertelement <16 x i16> %haddA, i16 %aEF, i32 11
+ %haddC = insertelement <16 x i16> %haddB, i16 %b89, i32 12
+ %haddD = insertelement <16 x i16> %haddC, i16 %bAB, i32 13
+ %haddE = insertelement <16 x i16> %haddD, i16 %bCD, i32 14
+ %haddF = insertelement <16 x i16> %haddE, i16 %bEF, i32 15
+ %result = shufflevector <16 x i16> %haddF, <16 x i16> %a, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 9, i32 poison, i32 11, i32 12, i32 13, i32 14, i32 15>
+ ret <16 x i16> %result
+}
+
+define <16 x i16> @add_v16i16_FEuCBA98765432u0(<16 x i16> %a, <16 x i16> %b) {
+; SSE2-LABEL: @add_v16i16_FEuCBA98765432u0(
+; SSE2-NEXT: [[A8:%.*]] = extractelement <16 x i16> [[A:%.*]], i64 8
+; SSE2-NEXT: [[A9:%.*]] = extractelement <16 x i16> [[A]], i64 9
+; SSE2-NEXT: [[A89:%.*]] = add i16 [[A8]], [[A9]]
+; SSE2-NEXT: [[BC:%.*]] = extractelement <16 x i16> [[B:%.*]], i64 12
+; SSE2-NEXT: [[BD:%.*]] = extractelement <16 x i16> [[B]], i64 13
+; SSE2-NEXT: [[BE:%.*]] = extractelement <16 x i16> [[B]], i64 14
+; SSE2-NEXT: [[BF:%.*]] = extractelement <16 x i16> [[B]], i64 15
+; SSE2-NEXT: [[BCD:%.*]] = add i16 [[BC]], [[BD]]
+; SSE2-NEXT: [[BEF:%.*]] = add i16 [[BE]], [[BF]]
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 0, i32 poison, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 poison, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP3:%.*]] = add <16 x i16> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: [[HADD8:%.*]] = insertelement <16 x i16> [[TMP3]], i16 [[A89]], i64 8
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 10, i32 12, i32 14, i32 24, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP5:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 11, i32 13, i32 15, i32 25, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP6:%.*]] = add <16 x i16> [[TMP4]], [[TMP5]]
+; SSE2-NEXT: [[HADDC1:%.*]] = shufflevector <16 x i16> [[HADD8]], <16 x i16> [[TMP6]], <16 x i32> <i32 0, i32 poison, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 16, i32 17, i32 18, i32 19, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[HADDE:%.*]] = insertelement <16 x i16> [[HADDC1]], i16 [[BCD]], i64 14
+; SSE2-NEXT: [[HADDF:%.*]] = insertelement <16 x i16> [[HADDE]], i16 [[BEF]], i64 15
+; SSE2-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[HADDF]], <16 x i16> poison, <16 x i32> <i32 15, i32 14, i32 poison, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 poison, i32 0>
+; SSE2-NEXT: ret <16 x i16> [[RESULT]]
+;
+; SSE4-LABEL: @add_v16i16_FEuCBA98765432u0(
+; SSE4-NEXT: [[A8:%.*]] = extractelement <16 x i16> [[A:%.*]], i64 8
+; SSE4-NEXT: [[A9:%.*]] = extractelement <16 x i16> [[A]], i64 9
+; SSE4-NEXT: [[A89:%.*]] = add i16 [[A8]], [[A9]]
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 poison, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 poison, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP3:%.*]] = add <16 x i16> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: [[HADD8:%.*]] = insertelement <16 x i16> [[TMP3]], i16 [[A89]], i64 8
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> poison, <16 x i32> <i32 11, i32 12, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> poison, <16 x i32> <i32 10, i32 13, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP6:%.*]] = add <16 x i16> [[TMP4]], [[TMP5]]
+; SSE4-NEXT: [[HADDA2:%.*]] = shufflevector <16 x i16> [[HADD8]], <16 x i16> [[TMP6]], <16 x i32> <i32 0, i32 poison, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 16, i32 17, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP7:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 14, i32 24, i32 poison, i32 28, i32 30, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP8:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 15, i32 25, i32 poison, i32 29, i32 31, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP9:%.*]] = add <16 x i16> [[TMP7]], [[TMP8]]
+; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[TMP9]], <16 x i16> [[HADDA2]], <16 x i32> <i32 4, i32 3, i32 poison, i32 1, i32 0, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 poison, i32 16>
+; SSE4-NEXT: ret <16 x i16> [[RESULT]]
+;
+; AVX2-LABEL: @add_v16i16_FEuCBA98765432u0(
+; AVX2-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 poison, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 9, i32 10, i32 12, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX2-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 poison, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 8, i32 11, i32 13, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX2-NEXT: [[HADDA:%.*]] = add <16 x i16> [[TMP1]], [[TMP2]]
+; AVX2-NEXT: [[TMP3:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 14, i32 24, i32 poison, i32 28, i32 30, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX2-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 15, i32 25, i32 poison, i32 29, i32 31, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX2-NEXT: [[TMP5:%.*]] = add <16 x i16> [[TMP3]], [[TMP4]]
+; AVX2-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[TMP5]], <16 x i16> [[HADDA]], <16 x i32> <i32 4, i32 3, i32 poison, i32 1, i32 0, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 poison, i32 16>
+; AVX2-NEXT: ret <16 x i16> [[RESULT]]
+;
+; AVX512-LABEL: @add_v16i16_FEuCBA98765432u0(
+; AVX512-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 poison, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 8, i32 11, i32 12, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX512-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 poison, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 9, i32 10, i32 13, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX512-NEXT: [[HADDA2:%.*]] = add <16 x i16> [[TMP1]], [[TMP2]]
+; AVX512-NEXT: [[TMP3:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 14, i32 24, i32 poison, i32 28, i32 30, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX512-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 15, i32 25, i32 poison, i32 29, i32 31, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX512-NEXT: [[TMP5:%.*]] = add <16 x i16> [[TMP3]], [[TMP4]]
+; AVX512-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[TMP5]], <16 x i16> [[HADDA2]], <16 x i32> <i32 4, i32 3, i32 poison, i32 1, i32 0, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 poison, i32 16>
+; AVX512-NEXT: ret <16 x i16> [[RESULT]]
+;
+ %a0 = extractelement <16 x i16> %a, i32 0
+ %a1 = extractelement <16 x i16> %a, i32 1
+ %a2 = extractelement <16 x i16> %a, i32 2
+ %a3 = extractelement <16 x i16> %a, i32 3
+ %a4 = extractelement <16 x i16> %a, i32 4
+ %a5 = extractelement <16 x i16> %a, i32 5
+ %a6 = extractelement <16 x i16> %a, i32 6
+ %a7 = extractelement <16 x i16> %a, i32 7
+ %a8 = extractelement <16 x i16> %a, i32 8
+ %a9 = extractelement <16 x i16> %a, i32 9
+ %aA = extractelement <16 x i16> %a, i32 10
+ %aB = extractelement <16 x i16> %a, i32 11
+ %aC = extractelement <16 x i16> %a, i32 12
+ %aD = extractelement <16 x i16> %a, i32 13
+ %aE = extractelement <16 x i16> %a, i32 14
+ %aF = extractelement <16 x i16> %a, i32 15
+ %a01 = add i16 %a0, %a1
+ %a23 = add i16 %a2, %a3
+ %a45 = add i16 %a4, %a5
+ %a67 = add i16 %a6, %a7
+ %a89 = add i16 %a8, %a9
+ %aAB = add i16 %aA, %aB
+ %aCD = add i16 %aC, %aD
+ %aEF = add i16 %aE, %aF
+ %b0 = extractelement <16 x i16> %b, i32 0
+ %b1 = extractelement <16 x i16> %b, i32 1
+ %b2 = extractelement <16 x i16> %b, i32 2
+ %b3 = extractelement <16 x i16> %b, i32 3
+ %b4 = extractelement <16 x i16> %b, i32 4
+ %b5 = extractelement <16 x i16> %b, i32 5
+ %b6 = extractelement <16 x i16> %b, i32 6
+ %b7 = extractelement <16 x i16> %b, i32 7
+ %b8 = extractelement <16 x i16> %b, i32 8
+ %b9 = extractelement <16 x i16> %b, i32 9
+ %bA = extractelement <16 x i16> %b, i32 10
+ %bB = extractelement <16 x i16> %b, i32 11
+ %bC = extractelement <16 x i16> %b, i32 12
+ %bD = extractelement <16 x i16> %b, i32 13
+ %bE = extractelement <16 x i16> %b, i32 14
+ %bF = extractelement <16 x i16> %b, i32 15
+ %b01 = add i16 %b0, %b1
+ %b23 = add i16 %b2, %b3
+ %b45 = add i16 %b4, %b5
+ %b67 = add i16 %b6, %b7
+ %b89 = add i16 %b8, %b9
+ %bAB = add i16 %bA, %bB
+ %bCD = add i16 %bC, %bD
+ %bEF = add i16 %bE, %bF
+ %hadd0 = insertelement <16 x i16> poison, i16 %a01, i32 0
+ %hadd1 = insertelement <16 x i16> %hadd0, i16 %a23, i32 1
+ %hadd2 = insertelement <16 x i16> %hadd1, i16 %a45, i32 2
+ %hadd3 = insertelement <16 x i16> %hadd2, i16 %a67, i32 3
+ %hadd4 = insertelement <16 x i16> %hadd3, i16 %b01, i32 4
+ %hadd5 = insertelement <16 x i16> %hadd4, i16 %b23, i32 5
+ %hadd6 = insertelement <16 x i16> %hadd5, i16 %b45, i32 6
+ %hadd7 = insertelement <16 x i16> %hadd6, i16 %b67, i32 7
+ %hadd8 = insertelement <16 x i16> %hadd7, i16 %a89, i32 8
+ %hadd9 = insertelement <16 x i16> %hadd8, i16 %aAB, i32 9
+ %haddA = insertelement <16 x i16> %hadd9, i16 %aCD, i32 10
+ %haddB = insertelement <16 x i16> %haddA, i16 %aEF, i32 11
+ %haddC = insertelement <16 x i16> %haddB, i16 %b89, i32 12
+ %haddD = insertelement <16 x i16> %haddC, i16 %bAB, i32 13
+ %haddE = insertelement <16 x i16> %haddD, i16 %bCD, i32 14
+ %haddF = insertelement <16 x i16> %haddE, i16 %bEF, i32 15
+ %result = shufflevector <16 x i16> %haddF, <16 x i16> %a, <16 x i32> <i32 15, i32 14, i32 poison, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 poison, i32 0>
+ ret <16 x i16> %result
+}
+
;
; v4i32
;
@@ -330,6 +733,51 @@ define <4 x i32> @add_v4i32_01uu(<4 x i32> %a, <4 x i32> %b) {
ret <4 x i32> %result
}
+define <4 x i32> @add_v4i32_32u0(<4 x i32> %a, <4 x i32> %b) {
+; SSE2-LABEL: @add_v4i32_32u0(
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[A:%.*]], <4 x i32> <i32 2, i32 0, i32 poison, i32 4>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> [[A]], <4 x i32> <i32 3, i32 1, i32 poison, i32 5>
+; SSE2-NEXT: [[RESULT1:%.*]] = add <4 x i32> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: ret <4 x i32> [[RESULT1]]
+;
+; SSE4-LABEL: @add_v4i32_32u0(
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[A:%.*]], <4 x i32> <i32 2, i32 0, i32 poison, i32 4>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> [[A]], <4 x i32> <i32 3, i32 1, i32 poison, i32 5>
+; SSE4-NEXT: [[RESULT:%.*]] = add <4 x i32> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: ret <4 x i32> [[RESULT]]
+;
+; AVX2-LABEL: @add_v4i32_32u0(
+; AVX2-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[A:%.*]], <4 x i32> <i32 2, i32 0, i32 poison, i32 4>
+; AVX2-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> [[A]], <4 x i32> <i32 3, i32 1, i32 poison, i32 5>
+; AVX2-NEXT: [[RESULT:%.*]] = add <4 x i32> [[TMP1]], [[TMP2]]
+; AVX2-NEXT: ret <4 x i32> [[RESULT]]
+;
+; AVX512-LABEL: @add_v4i32_32u0(
+; AVX512-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[A:%.*]], <4 x i32> <i32 2, i32 0, i32 poison, i32 4>
+; AVX512-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> [[A]], <4 x i32> <i32 3, i32 1, i32 poison, i32 5>
+; AVX512-NEXT: [[RESULT1:%.*]] = add <4 x i32> [[TMP1]], [[TMP2]]
+; AVX512-NEXT: ret <4 x i32> [[RESULT1]]
+;
+ %a0 = extractelement <4 x i32> %a, i32 0
+ %a1 = extractelement <4 x i32> %a, i32 1
+ %a2 = extractelement <4 x i32> %a, i32 2
+ %a3 = extractelement <4 x i32> %a, i32 3
+ %a01 = add i32 %a0, %a1
+ %a23 = add i32 %a2, %a3
+ %b0 = extractelement <4 x i32> %b, i32 0
+ %b1 = extractelement <4 x i32> %b, i32 1
+ %b2 = extractelement <4 x i32> %b, i32 2
+ %b3 = extractelement <4 x i32> %b, i32 3
+ %b01 = add i32 %b0, %b1
+ %b23 = add i32 %b2, %b3
+ %hadd0 = insertelement <4 x i32> poison, i32 %a01, i32 0
+ %hadd1 = insertelement <4 x i32> %hadd0, i32 %a23, i32 1
+ %hadd2 = insertelement <4 x i32> %hadd1, i32 %b01, i32 2
+ %hadd3 = insertelement <4 x i32> %hadd2, i32 %b23, i32 3
+ %result = shufflevector <4 x i32> %hadd3, <4 x i32> %a, <4 x i32> <i32 3, i32 2, i32 poison, i32 0>
+ ret <4 x i32> %result
+}
+
;
; v8i32
;
@@ -379,30 +827,30 @@ define <8 x i32> @add_v8i32_01234567(<8 x i32> %a, <8 x i32> %b) {
define <8 x i32> @add_v8i32_01234u67(<8 x i32> %a, <8 x i32> %b) {
; SSE2-LABEL: @add_v8i32_01234u67(
-; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 poison, i32 13, i32 14>
-; SSE2-NEXT: [[TMP3:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 poison, i32 12, i32 15>
-; SSE2-NEXT: [[TMP4:%.*]] = add <8 x i32> [[TMP2]], [[TMP3]]
-; SSE2-NEXT: ret <8 x i32> [[TMP4]]
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 poison, i32 13, i32 14>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 poison, i32 12, i32 15>
+; SSE2-NEXT: [[RESULT:%.*]] = add <8 x i32> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: ret <8 x i32> [[RESULT]]
;
; SSE4-LABEL: @add_v8i32_01234u67(
; SSE4-NEXT: [[A4:%.*]] = extractelement <8 x i32> [[A:%.*]], i64 4
; SSE4-NEXT: [[A5:%.*]] = extractelement <8 x i32> [[A]], i64 5
; SSE4-NEXT: [[A45:%.*]] = add i32 [[A4]], [[A5]]
-; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP3:%.*]] = add <8 x i32> [[TMP2]], [[TMP5]]
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP3:%.*]] = add <8 x i32> [[TMP1]], [[TMP2]]
; SSE4-NEXT: [[HADD4:%.*]] = insertelement <8 x i32> [[TMP3]], i32 [[A45]], i64 4
; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x i32> [[B]], <8 x i32> poison, <8 x i32> <i32 5, i32 6, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP7:%.*]] = shufflevector <8 x i32> [[B]], <8 x i32> poison, <8 x i32> <i32 4, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP6:%.*]] = add <8 x i32> [[TMP4]], [[TMP7]]
+; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <8 x i32> [[B]], <8 x i32> poison, <8 x i32> <i32 4, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP6:%.*]] = add <8 x i32> [[TMP4]], [[TMP5]]
; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <8 x i32> [[HADD4]], <8 x i32> [[TMP6]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 poison, i32 8, i32 9>
; SSE4-NEXT: ret <8 x i32> [[RESULT]]
;
; AVX-LABEL: @add_v8i32_01234u67(
-; AVX-NEXT: [[TMP5:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 poison, i32 13, i32 14>
-; AVX-NEXT: [[TMP6:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 poison, i32 12, i32 15>
-; AVX-NEXT: [[TMP7:%.*]] = add <8 x i32> [[TMP5]], [[TMP6]]
-; AVX-NEXT: ret <8 x i32> [[TMP7]]
+; AVX-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 poison, i32 13, i32 14>
+; AVX-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 poison, i32 12, i32 15>
+; AVX-NEXT: [[RESULT:%.*]] = add <8 x i32> [[TMP1]], [[TMP2]]
+; AVX-NEXT: ret <8 x i32> [[RESULT]]
;
%a0 = extractelement <8 x i32> %a, i32 0
%a1 = extractelement <8 x i32> %a, i32 1
@@ -599,8 +1047,8 @@ define <4 x float> @add_v4f32_012u(<4 x float> %a, <4 x float> %b) {
define <4 x float> @add_v4f32_uu23(<4 x float> %a, <4 x float> %b) {
; CHECK-LABEL: @add_v4f32_uu23(
-; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[TMP1:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 1, i32 2>
-; CHECK-NEXT: [[RESULT1:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 0, i32 3>
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 1, i32 2>
+; CHECK-NEXT: [[RESULT1:%.*]] = shufflevector <4 x float> [[B]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 0, i32 3>
; CHECK-NEXT: [[RESULT2:%.*]] = fadd <4 x float> [[TMP2]], [[RESULT1]]
; CHECK-NEXT: ret <4 x float> [[RESULT2]]
;
@@ -626,8 +1074,8 @@ define <4 x float> @add_v4f32_uu23(<4 x float> %a, <4 x float> %b) {
define <4 x float> @add_v4f32_01uu(<4 x float> %a, <4 x float> %b) {
; CHECK-LABEL: @add_v4f32_01uu(
-; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[TMP1:%.*]], <4 x float> poison, <4 x i32> <i32 1, i32 2, i32 poison, i32 poison>
-; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> poison, <4 x i32> <i32 0, i32 3, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 1, i32 2, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[A]], <4 x float> poison, <4 x i32> <i32 0, i32 3, i32 poison, i32 poison>
; CHECK-NEXT: [[TMP4:%.*]] = fadd <4 x float> [[TMP2]], [[TMP3]]
; CHECK-NEXT: ret <4 x float> [[TMP4]]
;
@@ -718,14 +1166,14 @@ define <8 x float> @add_v8f32_012u4567(<8 x float> %a, <8 x float> %b) {
; SSE4-NEXT: [[A6:%.*]] = extractelement <8 x float> [[A:%.*]], i64 6
; SSE4-NEXT: [[A7:%.*]] = extractelement <8 x float> [[A]], i64 7
; SSE4-NEXT: [[A67:%.*]] = fadd float [[A6]], [[A7]]
-; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[A]], <8 x float> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 poison, i32 4, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[A]], <8 x float> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 poison, i32 5, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP3:%.*]] = fadd <8 x float> [[TMP2]], [[TMP5]]
-; SSE4-NEXT: [[HADD5:%.*]] = insertelement <8 x float> [[TMP3]], float [[A67]], i64 5
-; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 5, i32 6, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP7:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 4, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x float> [[A]], <8 x float> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 poison, i32 4, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP7:%.*]] = shufflevector <8 x float> [[A]], <8 x float> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 poison, i32 5, i32 poison, i32 poison, i32 poison>
; SSE4-NEXT: [[TMP6:%.*]] = fadd <8 x float> [[TMP4]], [[TMP7]]
-; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <8 x float> [[HADD5]], <8 x float> [[TMP6]], <8 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 4, i32 5, i32 8, i32 9>
+; SSE4-NEXT: [[HADD5:%.*]] = insertelement <8 x float> [[TMP6]], float [[A67]], i64 5
+; SSE4-NEXT: [[TMP8:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 5, i32 6, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 4, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP9:%.*]] = fadd <8 x float> [[TMP8]], [[TMP5]]
+; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <8 x float> [[HADD5]], <8 x float> [[TMP9]], <8 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 4, i32 5, i32 8, i32 9>
; SSE4-NEXT: ret <8 x float> [[RESULT]]
;
; AVX-LABEL: @add_v8f32_012u4567(
@@ -770,6 +1218,79 @@ define <8 x float> @add_v8f32_012u4567(<8 x float> %a, <8 x float> %b) {
ret <8 x float> %result
}
+define <8 x float> @add_v8f32_76u43210(<8 x float> %a, <8 x float> %b) {
+; SSE2-LABEL: @add_v8f32_76u43210(
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <8 x float> [[A:%.*]], <8 x float> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[A]], <8 x float> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; SSE2-NEXT: [[TMP3:%.*]] = fadd <4 x float> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: [[B0:%.*]] = extractelement <8 x float> [[B:%.*]], i64 0
+; SSE2-NEXT: [[B1:%.*]] = extractelement <8 x float> [[B]], i64 1
+; SSE2-NEXT: [[B01:%.*]] = fadd float [[B0]], [[B1]]
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <2 x i32> <i32 5, i32 6>
+; SSE2-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <2 x i32> <i32 4, i32 7>
+; SSE2-NEXT: [[TMP6:%.*]] = fadd <2 x float> [[TMP4]], [[TMP5]]
+; SSE2-NEXT: [[TMP7:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[HADD4:%.*]] = insertelement <8 x float> [[TMP7]], float [[B01]], i64 4
+; SSE2-NEXT: [[TMP8:%.*]] = shufflevector <2 x float> [[TMP6]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[RESULT:%.*]] = shufflevector <8 x float> [[TMP8]], <8 x float> [[HADD4]], <8 x i32> <i32 1, i32 0, i32 poison, i32 12, i32 11, i32 10, i32 9, i32 8>
+; SSE2-NEXT: ret <8 x float> [[RESULT]]
+;
+; SSE4-LABEL: @add_v8f32_76u43210(
+; SSE4-NEXT: [[B0:%.*]] = extractelement <8 x float> [[B:%.*]], i64 0
+; SSE4-NEXT: [[B1:%.*]] = extractelement <8 x float> [[B]], i64 1
+; SSE4-NEXT: [[B01:%.*]] = fadd float [[B0]], [[B1]]
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <8 x float> [[A:%.*]], <8 x float> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[A]], <8 x float> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[RESULT:%.*]] = fadd <8 x float> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: [[HADD4:%.*]] = insertelement <8 x float> [[RESULT]], float [[B01]], i64 4
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 5, i32 6, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 4, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP6:%.*]] = fadd <8 x float> [[TMP4]], [[TMP5]]
+; SSE4-NEXT: [[RESULT1:%.*]] = shufflevector <8 x float> [[TMP6]], <8 x float> [[HADD4]], <8 x i32> <i32 1, i32 0, i32 poison, i32 12, i32 11, i32 10, i32 9, i32 8>
+; SSE4-NEXT: ret <8 x float> [[RESULT1]]
+;
+; AVX-LABEL: @add_v8f32_76u43210(
+; AVX-NEXT: [[TMP1:%.*]] = shufflevector <8 x float> [[B:%.*]], <8 x float> [[A:%.*]], <8 x i32> <i32 6, i32 5, i32 poison, i32 0, i32 14, i32 12, i32 10, i32 8>
+; AVX-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[B]], <8 x float> [[A]], <8 x i32> <i32 7, i32 4, i32 poison, i32 1, i32 15, i32 13, i32 11, i32 9>
+; AVX-NEXT: [[RESULT:%.*]] = fadd <8 x float> [[TMP1]], [[TMP2]]
+; AVX-NEXT: ret <8 x float> [[RESULT]]
+;
+ %a0 = extractelement <8 x float> %a, i32 0
+ %a1 = extractelement <8 x float> %a, i32 1
+ %a2 = extractelement <8 x float> %a, i32 2
+ %a3 = extractelement <8 x float> %a, i32 3
+ %a4 = extractelement <8 x float> %a, i32 4
+ %a5 = extractelement <8 x float> %a, i32 5
+ %a6 = extractelement <8 x float> %a, i32 6
+ %a7 = extractelement <8 x float> %a, i32 7
+ %a01 = fadd float %a0, %a1
+ %a23 = fadd float %a2, %a3
+ %a45 = fadd float %a4, %a5
+ %a67 = fadd float %a6, %a7
+ %b0 = extractelement <8 x float> %b, i32 0
+ %b1 = extractelement <8 x float> %b, i32 1
+ %b2 = extractelement <8 x float> %b, i32 2
+ %b3 = extractelement <8 x float> %b, i32 3
+ %b4 = extractelement <8 x float> %b, i32 4
+ %b5 = extractelement <8 x float> %b, i32 5
+ %b6 = extractelement <8 x float> %b, i32 6
+ %b7 = extractelement <8 x float> %b, i32 7
+ %b01 = fadd float %b0, %b1
+ %b23 = fadd float %b2, %b3
+ %b45 = fadd float %b4, %b5
+ %b67 = fadd float %b6, %b7
+ %hadd0 = insertelement <8 x float> poison, float %a01, i32 0
+ %hadd1 = insertelement <8 x float> %hadd0, float %a23, i32 1
+ %hadd2 = insertelement <8 x float> %hadd1, float %a45, i32 2
+ %hadd3 = insertelement <8 x float> %hadd2, float %a67, i32 3
+ %hadd4 = insertelement <8 x float> %hadd3, float %b01, i32 4
+ %hadd5 = insertelement <8 x float> %hadd4, float %b23, i32 5
+ %hadd6 = insertelement <8 x float> %hadd5, float %b45, i32 6
+ %hadd7 = insertelement <8 x float> %hadd6, float %b67, i32 7
+ %result = shufflevector <8 x float> %hadd7, <8 x float> %a, <8 x i32> <i32 7, i32 6, i32 poison, i32 4, i32 3, i32 2, i32 1, i32 0>
+ ret <8 x float> %result
+}
+
;
; v2f64
;
@@ -924,8 +1445,8 @@ define <4 x double> @add_v4f64_0u23(<4 x double> %a, <4 x double> %b) {
; SSE4-NEXT: [[B2:%.*]] = extractelement <4 x double> [[B:%.*]], i64 2
; SSE4-NEXT: [[B3:%.*]] = extractelement <4 x double> [[B]], i64 3
; SSE4-NEXT: [[B23:%.*]] = fadd double [[B2]], [[B3]]
-; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <4 x double> [[TMP1:%.*]], <4 x double> poison, <4 x i32> <i32 1, i32 poison, i32 2, i32 poison>
-; SSE4-NEXT: [[TMP3:%.*]] = shufflevector <4 x double> [[TMP1]], <4 x double> poison, <4 x i32> <i32 0, i32 poison, i32 3, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <4 x double> [[A:%.*]], <4 x double> poison, <4 x i32> <i32 1, i32 poison, i32 2, i32 poison>
+; SSE4-NEXT: [[TMP3:%.*]] = shufflevector <4 x double> [[A]], <4 x double> poison, <4 x i32> <i32 0, i32 poison, i32 3, i32 poison>
; SSE4-NEXT: [[TMP4:%.*]] = fadd <4 x double> [[TMP2]], [[TMP3]]
; SSE4-NEXT: [[RESULT:%.*]] = insertelement <4 x double> [[TMP4]], double [[B23]], i64 3
; SSE4-NEXT: ret <4 x double> [[RESULT]]
@@ -1131,3 +1652,51 @@ define <4 x double> @add_v4f64_01uu(<4 x double> %a, <4 x double> %b) {
%result = shufflevector <4 x double> %hadd3, <4 x double> %a, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
ret <4 x double> %result
}
+
+define <4 x double> @add_v4f64_32u0(<4 x double> %a, <4 x double> %b) {
+; SSE2-LABEL: @add_v4f64_32u0(
+; SSE2-NEXT: [[A0:%.*]] = extractelement <4 x double> [[A:%.*]], i64 0
+; SSE2-NEXT: [[A1:%.*]] = extractelement <4 x double> [[A]], i64 1
+; SSE2-NEXT: [[A01:%.*]] = fadd double [[A0]], [[A1]]
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[B:%.*]], <4 x double> poison, <2 x i32> <i32 2, i32 0>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <4 x double> [[B]], <4 x double> poison, <2 x i32> <i32 3, i32 1>
+; SSE2-NEXT: [[TMP3:%.*]] = fadd <2 x double> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <2 x double> [[TMP3]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
+; SSE2-NEXT: [[RESULT:%.*]] = insertelement <4 x double> [[TMP4]], double [[A01]], i64 3
+; SSE2-NEXT: ret <4 x double> [[RESULT]]
+;
+; SSE4-LABEL: @add_v4f64_32u0(
+; SSE4-NEXT: [[A0:%.*]] = extractelement <4 x double> [[A:%.*]], i64 0
+; SSE4-NEXT: [[A1:%.*]] = extractelement <4 x double> [[A]], i64 1
+; SSE4-NEXT: [[A01:%.*]] = fadd double [[A0]], [[A1]]
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[B:%.*]], <4 x double> poison, <4 x i32> <i32 2, i32 0, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <4 x double> [[B]], <4 x double> poison, <4 x i32> <i32 3, i32 1, i32 poison, i32 poison>
+; SSE4-NEXT: [[RESULT:%.*]] = fadd <4 x double> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: [[RESULT1:%.*]] = insertelement <4 x double> [[RESULT]], double [[A01]], i64 3
+; SSE4-NEXT: ret <4 x double> [[RESULT1]]
+;
+; AVX-LABEL: @add_v4f64_32u0(
+; AVX-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[B:%.*]], <4 x double> [[A:%.*]], <4 x i32> <i32 2, i32 0, i32 poison, i32 4>
+; AVX-NEXT: [[TMP2:%.*]] = shufflevector <4 x double> [[B]], <4 x double> [[A]], <4 x i32> <i32 3, i32 1, i32 poison, i32 5>
+; AVX-NEXT: [[RESULT:%.*]] = fadd <4 x double> [[TMP1]], [[TMP2]]
+; AVX-NEXT: ret <4 x double> [[RESULT]]
+;
+ %a0 = extractelement <4 x double> %a, i32 0
+ %a1 = extractelement <4 x double> %a, i32 1
+ %a2 = extractelement <4 x double> %a, i32 2
+ %a3 = extractelement <4 x double> %a, i32 3
+ %a01 = fadd double %a0, %a1
+ %a23 = fadd double %a2, %a3
+ %b0 = extractelement <4 x double> %b, i32 0
+ %b1 = extractelement <4 x double> %b, i32 1
+ %b2 = extractelement <4 x double> %b, i32 2
+ %b3 = extractelement <4 x double> %b, i32 3
+ %b01 = fadd double %b0, %b1
+ %b23 = fadd double %b2, %b3
+ %hadd0 = insertelement <4 x double> poison, double %a01, i32 0
+ %hadd1 = insertelement <4 x double> %hadd0, double %a23, i32 1
+ %hadd2 = insertelement <4 x double> %hadd1, double %b01, i32 2
+ %hadd3 = insertelement <4 x double> %hadd2, double %b23, i32 3
+ %result = shufflevector <4 x double> %hadd3, <4 x double> %a, <4 x i32> <i32 3, i32 2, i32 poison, i32 0>
+ ret <4 x double> %result
+}
diff --git a/llvm/test/Transforms/PhaseOrdering/X86/hsub.ll b/llvm/test/Transforms/PhaseOrdering/X86/hsub.ll
index ae05f6470e5636..998877e246b164 100644
--- a/llvm/test/Transforms/PhaseOrdering/X86/hsub.ll
+++ b/llvm/test/Transforms/PhaseOrdering/X86/hsub.ll
@@ -61,10 +61,10 @@ define <8 x i16> @sub_v8i16_u1234567(<8 x i16> %a, <8 x i16> %b) {
; SSE2-LABEL: @sub_v8i16_u1234567(
; SSE2-NEXT: [[SHIFT3:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 7, i32 poison>
; SSE2-NEXT: [[TMP6:%.*]] = sub <8 x i16> [[A]], [[SHIFT3]]
-; SSE2-NEXT: [[TMP7:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> poison, <8 x i32> <i32 poison, i32 2, i32 4, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> poison, <8 x i32> <i32 poison, i32 3, i32 5, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE2-NEXT: [[HSUB1:%.*]] = sub <8 x i16> [[TMP7]], [[TMP4]]
-; SSE2-NEXT: [[HSUB3:%.*]] = shufflevector <8 x i16> [[HSUB1]], <8 x i16> [[TMP6]], <8 x i32> <i32 poison, i32 1, i32 2, i32 14, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> poison, <8 x i32> <i32 poison, i32 2, i32 4, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP5:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> poison, <8 x i32> <i32 poison, i32 3, i32 5, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[HSUB22:%.*]] = sub <8 x i16> [[TMP4]], [[TMP5]]
+; SSE2-NEXT: [[HSUB3:%.*]] = shufflevector <8 x i16> [[HSUB22]], <8 x i16> [[TMP6]], <8 x i32> <i32 poison, i32 1, i32 2, i32 14, i32 poison, i32 poison, i32 poison, i32 poison>
; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 poison, i32 poison, i32 poison, i32 poison>
; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <8 x i16> [[B]], <8 x i16> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
; SSE2-NEXT: [[TMP3:%.*]] = sub <8 x i16> [[TMP1]], [[TMP2]]
@@ -119,6 +119,403 @@ define <8 x i16> @sub_v8i16_u1234567(<8 x i16> %a, <8 x i16> %b) {
ret <8 x i16> %result
}
+define <8 x i16> @sub_v8i16_76u43210(<8 x i16> %a, <8 x i16> %b) {
+; SSE2-LABEL: @sub_v8i16_76u43210(
+; SSE2-NEXT: [[SHIFT:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> poison, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP1:%.*]] = sub <8 x i16> [[A]], [[SHIFT]]
+; SSE2-NEXT: [[SHIFT2:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 5, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP2:%.*]] = sub <8 x i16> [[B]], [[SHIFT2]]
+; SSE2-NEXT: [[SHIFT3:%.*]] = shufflevector <8 x i16> [[B]], <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 6>
+; SSE2-NEXT: [[TMP3:%.*]] = sub <8 x i16> [[SHIFT3]], [[B]]
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 2, i32 4, i32 6, i32 8, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP5:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 3, i32 5, i32 7, i32 9, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP6:%.*]] = sub <8 x i16> [[TMP4]], [[TMP5]]
+; SSE2-NEXT: [[HSUB41:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP6]], <8 x i32> <i32 0, i32 8, i32 9, i32 10, i32 11, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[HSUB6:%.*]] = shufflevector <8 x i16> [[HSUB41]], <8 x i16> [[TMP2]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 poison, i32 12, i32 poison>
+; SSE2-NEXT: [[RESULT:%.*]] = shufflevector <8 x i16> [[TMP3]], <8 x i16> [[HSUB6]], <8 x i32> <i32 7, i32 14, i32 poison, i32 12, i32 11, i32 10, i32 9, i32 8>
+; SSE2-NEXT: ret <8 x i16> [[RESULT]]
+;
+; SSE4-LABEL: @sub_v8i16_76u43210(
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[HSUB22:%.*]] = sub <8 x i16> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: [[TMP3:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B:%.*]], <8 x i32> <i32 6, i32 8, i32 poison, i32 12, i32 14, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 7, i32 9, i32 poison, i32 13, i32 15, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP5:%.*]] = sub <8 x i16> [[TMP3]], [[TMP4]]
+; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <8 x i16> [[TMP5]], <8 x i16> [[HSUB22]], <8 x i32> <i32 4, i32 3, i32 poison, i32 1, i32 0, i32 10, i32 9, i32 8>
+; SSE4-NEXT: ret <8 x i16> [[RESULT]]
+;
+; AVX-LABEL: @sub_v8i16_76u43210(
+; AVX-NEXT: [[TMP1:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX-NEXT: [[TMP2:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX-NEXT: [[HSUB22:%.*]] = sub <8 x i16> [[TMP1]], [[TMP2]]
+; AVX-NEXT: [[TMP3:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B:%.*]], <8 x i32> <i32 6, i32 8, i32 poison, i32 12, i32 14, i32 poison, i32 poison, i32 poison>
+; AVX-NEXT: [[TMP4:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 7, i32 9, i32 poison, i32 13, i32 15, i32 poison, i32 poison, i32 poison>
+; AVX-NEXT: [[TMP5:%.*]] = sub <8 x i16> [[TMP3]], [[TMP4]]
+; AVX-NEXT: [[RESULT:%.*]] = shufflevector <8 x i16> [[TMP5]], <8 x i16> [[HSUB22]], <8 x i32> <i32 4, i32 3, i32 poison, i32 1, i32 0, i32 10, i32 9, i32 8>
+; AVX-NEXT: ret <8 x i16> [[RESULT]]
+;
+ %a0 = extractelement <8 x i16> %a, i32 0
+ %a1 = extractelement <8 x i16> %a, i32 1
+ %a2 = extractelement <8 x i16> %a, i32 2
+ %a3 = extractelement <8 x i16> %a, i32 3
+ %a4 = extractelement <8 x i16> %a, i32 4
+ %a5 = extractelement <8 x i16> %a, i32 5
+ %a6 = extractelement <8 x i16> %a, i32 6
+ %a7 = extractelement <8 x i16> %a, i32 7
+ %a01 = sub i16 %a0, %a1
+ %a23 = sub i16 %a2, %a3
+ %a45 = sub i16 %a4, %a5
+ %a67 = sub i16 %a6, %a7
+ %b0 = extractelement <8 x i16> %b, i32 0
+ %b1 = extractelement <8 x i16> %b, i32 1
+ %b2 = extractelement <8 x i16> %b, i32 2
+ %b3 = extractelement <8 x i16> %b, i32 3
+ %b4 = extractelement <8 x i16> %b, i32 4
+ %b5 = extractelement <8 x i16> %b, i32 5
+ %b6 = extractelement <8 x i16> %b, i32 6
+ %b7 = extractelement <8 x i16> %b, i32 7
+ %b01 = sub i16 %b0, %b1
+ %b23 = sub i16 %b2, %b3
+ %b45 = sub i16 %b4, %b5
+ %b67 = sub i16 %b6, %b7
+ %hsub0 = insertelement <8 x i16> poison, i16 %a01, i32 0
+ %hsub1 = insertelement <8 x i16> %hsub0, i16 %a23, i32 1
+ %hsub2 = insertelement <8 x i16> %hsub1, i16 %a45, i32 2
+ %hsub3 = insertelement <8 x i16> %hsub2, i16 %a67, i32 3
+ %hsub4 = insertelement <8 x i16> %hsub3, i16 %b01, i32 4
+ %hsub5 = insertelement <8 x i16> %hsub4, i16 %b23, i32 5
+ %hsub6 = insertelement <8 x i16> %hsub5, i16 %b45, i32 6
+ %hsub7 = insertelement <8 x i16> %hsub6, i16 %b67, i32 7
+ %result = shufflevector <8 x i16> %hsub7, <8 x i16> %a, <8 x i32> <i32 7, i32 6, i32 poison, i32 4, i32 3, i32 2, i32 1, i32 0>
+ ret <8 x i16> %result
+}
+
+;
+; v16i16
+;
+
+define <16 x i16> @sub_v16i16_0123456789ABCDEF(<16 x i16> %a, <16 x i16> %b) {
+; CHECK-LABEL: @sub_v16i16_0123456789ABCDEF(
+; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 8, i32 10, i32 12, i32 14, i32 24, i32 26, i32 28, i32 30>
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 9, i32 11, i32 13, i32 15, i32 25, i32 27, i32 29, i32 31>
+; CHECK-NEXT: [[TMP3:%.*]] = sub <16 x i16> [[TMP1]], [[TMP2]]
+; CHECK-NEXT: ret <16 x i16> [[TMP3]]
+;
+ %a0 = extractelement <16 x i16> %a, i32 0
+ %a1 = extractelement <16 x i16> %a, i32 1
+ %a2 = extractelement <16 x i16> %a, i32 2
+ %a3 = extractelement <16 x i16> %a, i32 3
+ %a4 = extractelement <16 x i16> %a, i32 4
+ %a5 = extractelement <16 x i16> %a, i32 5
+ %a6 = extractelement <16 x i16> %a, i32 6
+ %a7 = extractelement <16 x i16> %a, i32 7
+ %a8 = extractelement <16 x i16> %a, i32 8
+ %a9 = extractelement <16 x i16> %a, i32 9
+ %aA = extractelement <16 x i16> %a, i32 10
+ %aB = extractelement <16 x i16> %a, i32 11
+ %aC = extractelement <16 x i16> %a, i32 12
+ %aD = extractelement <16 x i16> %a, i32 13
+ %aE = extractelement <16 x i16> %a, i32 14
+ %aF = extractelement <16 x i16> %a, i32 15
+ %a01 = sub i16 %a0, %a1
+ %a23 = sub i16 %a2, %a3
+ %a45 = sub i16 %a4, %a5
+ %a67 = sub i16 %a6, %a7
+ %a89 = sub i16 %a8, %a9
+ %aAB = sub i16 %aA, %aB
+ %aCD = sub i16 %aC, %aD
+ %aEF = sub i16 %aE, %aF
+ %b0 = extractelement <16 x i16> %b, i32 0
+ %b1 = extractelement <16 x i16> %b, i32 1
+ %b2 = extractelement <16 x i16> %b, i32 2
+ %b3 = extractelement <16 x i16> %b, i32 3
+ %b4 = extractelement <16 x i16> %b, i32 4
+ %b5 = extractelement <16 x i16> %b, i32 5
+ %b6 = extractelement <16 x i16> %b, i32 6
+ %b7 = extractelement <16 x i16> %b, i32 7
+ %b8 = extractelement <16 x i16> %b, i32 8
+ %b9 = extractelement <16 x i16> %b, i32 9
+ %bA = extractelement <16 x i16> %b, i32 10
+ %bB = extractelement <16 x i16> %b, i32 11
+ %bC = extractelement <16 x i16> %b, i32 12
+ %bD = extractelement <16 x i16> %b, i32 13
+ %bE = extractelement <16 x i16> %b, i32 14
+ %bF = extractelement <16 x i16> %b, i32 15
+ %b01 = sub i16 %b0, %b1
+ %b23 = sub i16 %b2, %b3
+ %b45 = sub i16 %b4, %b5
+ %b67 = sub i16 %b6, %b7
+ %b89 = sub i16 %b8, %b9
+ %bAB = sub i16 %bA, %bB
+ %bCD = sub i16 %bC, %bD
+ %bEF = sub i16 %bE, %bF
+ %hsub0 = insertelement <16 x i16> poison, i16 %a01, i32 0
+ %hsub1 = insertelement <16 x i16> %hsub0, i16 %a23, i32 1
+ %hsub2 = insertelement <16 x i16> %hsub1, i16 %a45, i32 2
+ %hsub3 = insertelement <16 x i16> %hsub2, i16 %a67, i32 3
+ %hsub4 = insertelement <16 x i16> %hsub3, i16 %b01, i32 4
+ %hsub5 = insertelement <16 x i16> %hsub4, i16 %b23, i32 5
+ %hsub6 = insertelement <16 x i16> %hsub5, i16 %b45, i32 6
+ %hsub7 = insertelement <16 x i16> %hsub6, i16 %b67, i32 7
+ %hsub8 = insertelement <16 x i16> %hsub7, i16 %a89, i32 8
+ %hsub9 = insertelement <16 x i16> %hsub8, i16 %aAB, i32 9
+ %hsubA = insertelement <16 x i16> %hsub9, i16 %aCD, i32 10
+ %hsubB = insertelement <16 x i16> %hsubA, i16 %aEF, i32 11
+ %hsubC = insertelement <16 x i16> %hsubB, i16 %b89, i32 12
+ %hsubD = insertelement <16 x i16> %hsubC, i16 %bAB, i32 13
+ %hsubE = insertelement <16 x i16> %hsubD, i16 %bCD, i32 14
+ %hsubF = insertelement <16 x i16> %hsubE, i16 %bEF, i32 15
+ %result = shufflevector <16 x i16> %hsubF, <16 x i16> %a, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+ ret <16 x i16> %result
+}
+
+define <16 x i16> @sub_v16i16_0123u56789uBCDEF(<16 x i16> %a, <16 x i16> %b) {
+; SSE2-LABEL: @sub_v16i16_0123u56789uBCDEF(
+; SSE2-NEXT: [[A8:%.*]] = extractelement <16 x i16> [[A:%.*]], i64 8
+; SSE2-NEXT: [[A9:%.*]] = extractelement <16 x i16> [[A]], i64 9
+; SSE2-NEXT: [[A89:%.*]] = sub i16 [[A8]], [[A9]]
+; SSE2-NEXT: [[BC:%.*]] = extractelement <16 x i16> [[B:%.*]], i64 12
+; SSE2-NEXT: [[BD:%.*]] = extractelement <16 x i16> [[B]], i64 13
+; SSE2-NEXT: [[BE:%.*]] = extractelement <16 x i16> [[B]], i64 14
+; SSE2-NEXT: [[BF:%.*]] = extractelement <16 x i16> [[B]], i64 15
+; SSE2-NEXT: [[BCD:%.*]] = sub i16 [[BC]], [[BD]]
+; SSE2-NEXT: [[BEF:%.*]] = sub i16 [[BE]], [[BF]]
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 poison, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 poison, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP3:%.*]] = sub <16 x i16> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: [[HSUB8:%.*]] = insertelement <16 x i16> [[TMP3]], i16 [[A89]], i64 8
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 10, i32 poison, i32 14, i32 24, i32 26, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP5:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 11, i32 poison, i32 15, i32 25, i32 27, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP6:%.*]] = sub <16 x i16> [[TMP4]], [[TMP5]]
+; SSE2-NEXT: [[HSUBD1:%.*]] = shufflevector <16 x i16> [[HSUB8]], <16 x i16> [[TMP6]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 16, i32 poison, i32 18, i32 19, i32 20, i32 poison, i32 poison>
+; SSE2-NEXT: [[HSUBE:%.*]] = insertelement <16 x i16> [[HSUBD1]], i16 [[BCD]], i64 14
+; SSE2-NEXT: [[HSUBF:%.*]] = insertelement <16 x i16> [[HSUBE]], i16 [[BEF]], i64 15
+; SSE2-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[HSUBF]], <16 x i16> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 9, i32 poison, i32 11, i32 12, i32 13, i32 14, i32 15>
+; SSE2-NEXT: ret <16 x i16> [[RESULT]]
+;
+; SSE4-LABEL: @sub_v16i16_0123u56789uBCDEF(
+; SSE4-NEXT: [[A8:%.*]] = extractelement <16 x i16> [[A:%.*]], i64 8
+; SSE4-NEXT: [[A9:%.*]] = extractelement <16 x i16> [[A]], i64 9
+; SSE4-NEXT: [[A89:%.*]] = sub i16 [[A8]], [[A9]]
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 poison, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 poison, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP3:%.*]] = sub <16 x i16> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: [[HSUB8:%.*]] = insertelement <16 x i16> [[TMP3]], i16 [[A89]], i64 8
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> poison, <16 x i32> <i32 10, i32 poison, i32 14, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> poison, <16 x i32> <i32 11, i32 poison, i32 15, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP6:%.*]] = sub <16 x i16> [[TMP4]], [[TMP5]]
+; SSE4-NEXT: [[HSUBB2:%.*]] = shufflevector <16 x i16> [[HSUB8]], <16 x i16> [[TMP6]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 16, i32 poison, i32 18, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP7:%.*]] = shufflevector <16 x i16> [[B]], <16 x i16> poison, <16 x i32> <i32 8, i32 10, i32 12, i32 14, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP8:%.*]] = shufflevector <16 x i16> [[B]], <16 x i16> poison, <16 x i32> <i32 9, i32 11, i32 13, i32 15, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP9:%.*]] = sub <16 x i16> [[TMP7]], [[TMP8]]
+; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[HSUBB2]], <16 x i16> [[TMP9]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 9, i32 poison, i32 11, i32 16, i32 17, i32 18, i32 19>
+; SSE4-NEXT: ret <16 x i16> [[RESULT]]
+;
+; AVX-LABEL: @sub_v16i16_0123u56789uBCDEF(
+; AVX-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 poison, i32 18, i32 20, i32 22, i32 8, i32 10, i32 poison, i32 14, i32 24, i32 26, i32 28, i32 30>
+; AVX-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 poison, i32 19, i32 21, i32 23, i32 9, i32 11, i32 poison, i32 15, i32 25, i32 27, i32 29, i32 31>
+; AVX-NEXT: [[RESULT:%.*]] = sub <16 x i16> [[TMP1]], [[TMP2]]
+; AVX-NEXT: ret <16 x i16> [[RESULT]]
+;
+ %a0 = extractelement <16 x i16> %a, i32 0
+ %a1 = extractelement <16 x i16> %a, i32 1
+ %a2 = extractelement <16 x i16> %a, i32 2
+ %a3 = extractelement <16 x i16> %a, i32 3
+ %a4 = extractelement <16 x i16> %a, i32 4
+ %a5 = extractelement <16 x i16> %a, i32 5
+ %a6 = extractelement <16 x i16> %a, i32 6
+ %a7 = extractelement <16 x i16> %a, i32 7
+ %a8 = extractelement <16 x i16> %a, i32 8
+ %a9 = extractelement <16 x i16> %a, i32 9
+ %aA = extractelement <16 x i16> %a, i32 10
+ %aB = extractelement <16 x i16> %a, i32 11
+ %aC = extractelement <16 x i16> %a, i32 12
+ %aD = extractelement <16 x i16> %a, i32 13
+ %aE = extractelement <16 x i16> %a, i32 14
+ %aF = extractelement <16 x i16> %a, i32 15
+ %a01 = sub i16 %a0, %a1
+ %a23 = sub i16 %a2, %a3
+ %a45 = sub i16 %a4, %a5
+ %a67 = sub i16 %a6, %a7
+ %a89 = sub i16 %a8, %a9
+ %aAB = sub i16 %aA, %aB
+ %aCD = sub i16 %aC, %aD
+ %aEF = sub i16 %aE, %aF
+ %b0 = extractelement <16 x i16> %b, i32 0
+ %b1 = extractelement <16 x i16> %b, i32 1
+ %b2 = extractelement <16 x i16> %b, i32 2
+ %b3 = extractelement <16 x i16> %b, i32 3
+ %b4 = extractelement <16 x i16> %b, i32 4
+ %b5 = extractelement <16 x i16> %b, i32 5
+ %b6 = extractelement <16 x i16> %b, i32 6
+ %b7 = extractelement <16 x i16> %b, i32 7
+ %b8 = extractelement <16 x i16> %b, i32 8
+ %b9 = extractelement <16 x i16> %b, i32 9
+ %bA = extractelement <16 x i16> %b, i32 10
+ %bB = extractelement <16 x i16> %b, i32 11
+ %bC = extractelement <16 x i16> %b, i32 12
+ %bD = extractelement <16 x i16> %b, i32 13
+ %bE = extractelement <16 x i16> %b, i32 14
+ %bF = extractelement <16 x i16> %b, i32 15
+ %b01 = sub i16 %b0, %b1
+ %b23 = sub i16 %b2, %b3
+ %b45 = sub i16 %b4, %b5
+ %b67 = sub i16 %b6, %b7
+ %b89 = sub i16 %b8, %b9
+ %bAB = sub i16 %bA, %bB
+ %bCD = sub i16 %bC, %bD
+ %bEF = sub i16 %bE, %bF
+ %hsub0 = insertelement <16 x i16> poison, i16 %a01, i32 0
+ %hsub1 = insertelement <16 x i16> %hsub0, i16 %a23, i32 1
+ %hsub2 = insertelement <16 x i16> %hsub1, i16 %a45, i32 2
+ %hsub3 = insertelement <16 x i16> %hsub2, i16 %a67, i32 3
+ %hsub4 = insertelement <16 x i16> %hsub3, i16 %b01, i32 4
+ %hsub5 = insertelement <16 x i16> %hsub4, i16 %b23, i32 5
+ %hsub6 = insertelement <16 x i16> %hsub5, i16 %b45, i32 6
+ %hsub7 = insertelement <16 x i16> %hsub6, i16 %b67, i32 7
+ %hsub8 = insertelement <16 x i16> %hsub7, i16 %a89, i32 8
+ %hsub9 = insertelement <16 x i16> %hsub8, i16 %aAB, i32 9
+ %hsubA = insertelement <16 x i16> %hsub9, i16 %aCD, i32 10
+ %hsubB = insertelement <16 x i16> %hsubA, i16 %aEF, i32 11
+ %hsubC = insertelement <16 x i16> %hsubB, i16 %b89, i32 12
+ %hsubD = insertelement <16 x i16> %hsubC, i16 %bAB, i32 13
+ %hsubE = insertelement <16 x i16> %hsubD, i16 %bCD, i32 14
+ %hsubF = insertelement <16 x i16> %hsubE, i16 %bEF, i32 15
+ %result = shufflevector <16 x i16> %hsubF, <16 x i16> %a, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 5, i32 6, i32 7, i32 8, i32 9, i32 poison, i32 11, i32 12, i32 13, i32 14, i32 15>
+ ret <16 x i16> %result
+}
+
+define <16 x i16> @sub_v16i16_FEuCBA98765432u0(<16 x i16> %a, <16 x i16> %b) {
+; SSE2-LABEL: @sub_v16i16_FEuCBA98765432u0(
+; SSE2-NEXT: [[A8:%.*]] = extractelement <16 x i16> [[A:%.*]], i64 8
+; SSE2-NEXT: [[A9:%.*]] = extractelement <16 x i16> [[A]], i64 9
+; SSE2-NEXT: [[A89:%.*]] = sub i16 [[A8]], [[A9]]
+; SSE2-NEXT: [[BC:%.*]] = extractelement <16 x i16> [[B:%.*]], i64 12
+; SSE2-NEXT: [[BD:%.*]] = extractelement <16 x i16> [[B]], i64 13
+; SSE2-NEXT: [[BE:%.*]] = extractelement <16 x i16> [[B]], i64 14
+; SSE2-NEXT: [[BF:%.*]] = extractelement <16 x i16> [[B]], i64 15
+; SSE2-NEXT: [[BCD:%.*]] = sub i16 [[BC]], [[BD]]
+; SSE2-NEXT: [[BEF:%.*]] = sub i16 [[BE]], [[BF]]
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 0, i32 poison, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 poison, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP3:%.*]] = sub <16 x i16> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: [[HSUB8:%.*]] = insertelement <16 x i16> [[TMP3]], i16 [[A89]], i64 8
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 10, i32 12, i32 14, i32 24, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP5:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 11, i32 13, i32 15, i32 25, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[TMP6:%.*]] = sub <16 x i16> [[TMP4]], [[TMP5]]
+; SSE2-NEXT: [[HSUBC1:%.*]] = shufflevector <16 x i16> [[HSUB8]], <16 x i16> [[TMP6]], <16 x i32> <i32 0, i32 poison, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 16, i32 17, i32 18, i32 19, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[HSUBE:%.*]] = insertelement <16 x i16> [[HSUBC1]], i16 [[BCD]], i64 14
+; SSE2-NEXT: [[HSUBF:%.*]] = insertelement <16 x i16> [[HSUBE]], i16 [[BEF]], i64 15
+; SSE2-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[HSUBF]], <16 x i16> poison, <16 x i32> <i32 15, i32 14, i32 poison, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 poison, i32 0>
+; SSE2-NEXT: ret <16 x i16> [[RESULT]]
+;
+; SSE4-LABEL: @sub_v16i16_FEuCBA98765432u0(
+; SSE4-NEXT: [[A8:%.*]] = extractelement <16 x i16> [[A:%.*]], i64 8
+; SSE4-NEXT: [[A9:%.*]] = extractelement <16 x i16> [[A]], i64 9
+; SSE4-NEXT: [[A89:%.*]] = sub i16 [[A8]], [[A9]]
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 poison, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 poison, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP3:%.*]] = sub <16 x i16> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: [[HSUB8:%.*]] = insertelement <16 x i16> [[TMP3]], i16 [[A89]], i64 8
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> poison, <16 x i32> <i32 10, i32 12, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> poison, <16 x i32> <i32 11, i32 13, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP6:%.*]] = sub <16 x i16> [[TMP4]], [[TMP5]]
+; SSE4-NEXT: [[HSUBA2:%.*]] = shufflevector <16 x i16> [[HSUB8]], <16 x i16> [[TMP6]], <16 x i32> <i32 0, i32 poison, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 16, i32 17, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP7:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 14, i32 24, i32 poison, i32 28, i32 30, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP8:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 15, i32 25, i32 poison, i32 29, i32 31, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP9:%.*]] = sub <16 x i16> [[TMP7]], [[TMP8]]
+; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[TMP9]], <16 x i16> [[HSUBA2]], <16 x i32> <i32 4, i32 3, i32 poison, i32 1, i32 0, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 poison, i32 16>
+; SSE4-NEXT: ret <16 x i16> [[RESULT]]
+;
+; AVX2-LABEL: @sub_v16i16_FEuCBA98765432u0(
+; AVX2-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 poison, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 8, i32 10, i32 12, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX2-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 poison, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 9, i32 11, i32 13, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX2-NEXT: [[HSUBA:%.*]] = sub <16 x i16> [[TMP1]], [[TMP2]]
+; AVX2-NEXT: [[TMP3:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 14, i32 24, i32 poison, i32 28, i32 30, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX2-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 15, i32 25, i32 poison, i32 29, i32 31, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX2-NEXT: [[TMP5:%.*]] = sub <16 x i16> [[TMP3]], [[TMP4]]
+; AVX2-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[TMP5]], <16 x i16> [[HSUBA]], <16 x i32> <i32 4, i32 3, i32 poison, i32 1, i32 0, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 poison, i32 16>
+; AVX2-NEXT: ret <16 x i16> [[RESULT]]
+;
+; AVX512-LABEL: @sub_v16i16_FEuCBA98765432u0(
+; AVX512-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 poison, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 8, i32 10, i32 12, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX512-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 1, i32 poison, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 9, i32 11, i32 13, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX512-NEXT: [[HSUBA2:%.*]] = sub <16 x i16> [[TMP1]], [[TMP2]]
+; AVX512-NEXT: [[TMP3:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 14, i32 24, i32 poison, i32 28, i32 30, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX512-NEXT: [[TMP4:%.*]] = shufflevector <16 x i16> [[A]], <16 x i16> [[B]], <16 x i32> <i32 15, i32 25, i32 poison, i32 29, i32 31, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; AVX512-NEXT: [[TMP5:%.*]] = sub <16 x i16> [[TMP3]], [[TMP4]]
+; AVX512-NEXT: [[RESULT:%.*]] = shufflevector <16 x i16> [[TMP5]], <16 x i16> [[HSUBA2]], <16 x i32> <i32 4, i32 3, i32 poison, i32 1, i32 0, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 poison, i32 16>
+; AVX512-NEXT: ret <16 x i16> [[RESULT]]
+;
+ %a0 = extractelement <16 x i16> %a, i32 0
+ %a1 = extractelement <16 x i16> %a, i32 1
+ %a2 = extractelement <16 x i16> %a, i32 2
+ %a3 = extractelement <16 x i16> %a, i32 3
+ %a4 = extractelement <16 x i16> %a, i32 4
+ %a5 = extractelement <16 x i16> %a, i32 5
+ %a6 = extractelement <16 x i16> %a, i32 6
+ %a7 = extractelement <16 x i16> %a, i32 7
+ %a8 = extractelement <16 x i16> %a, i32 8
+ %a9 = extractelement <16 x i16> %a, i32 9
+ %aA = extractelement <16 x i16> %a, i32 10
+ %aB = extractelement <16 x i16> %a, i32 11
+ %aC = extractelement <16 x i16> %a, i32 12
+ %aD = extractelement <16 x i16> %a, i32 13
+ %aE = extractelement <16 x i16> %a, i32 14
+ %aF = extractelement <16 x i16> %a, i32 15
+ %a01 = sub i16 %a0, %a1
+ %a23 = sub i16 %a2, %a3
+ %a45 = sub i16 %a4, %a5
+ %a67 = sub i16 %a6, %a7
+ %a89 = sub i16 %a8, %a9
+ %aAB = sub i16 %aA, %aB
+ %aCD = sub i16 %aC, %aD
+ %aEF = sub i16 %aE, %aF
+ %b0 = extractelement <16 x i16> %b, i32 0
+ %b1 = extractelement <16 x i16> %b, i32 1
+ %b2 = extractelement <16 x i16> %b, i32 2
+ %b3 = extractelement <16 x i16> %b, i32 3
+ %b4 = extractelement <16 x i16> %b, i32 4
+ %b5 = extractelement <16 x i16> %b, i32 5
+ %b6 = extractelement <16 x i16> %b, i32 6
+ %b7 = extractelement <16 x i16> %b, i32 7
+ %b8 = extractelement <16 x i16> %b, i32 8
+ %b9 = extractelement <16 x i16> %b, i32 9
+ %bA = extractelement <16 x i16> %b, i32 10
+ %bB = extractelement <16 x i16> %b, i32 11
+ %bC = extractelement <16 x i16> %b, i32 12
+ %bD = extractelement <16 x i16> %b, i32 13
+ %bE = extractelement <16 x i16> %b, i32 14
+ %bF = extractelement <16 x i16> %b, i32 15
+ %b01 = sub i16 %b0, %b1
+ %b23 = sub i16 %b2, %b3
+ %b45 = sub i16 %b4, %b5
+ %b67 = sub i16 %b6, %b7
+ %b89 = sub i16 %b8, %b9
+ %bAB = sub i16 %bA, %bB
+ %bCD = sub i16 %bC, %bD
+ %bEF = sub i16 %bE, %bF
+ %hsub0 = insertelement <16 x i16> poison, i16 %a01, i32 0
+ %hsub1 = insertelement <16 x i16> %hsub0, i16 %a23, i32 1
+ %hsub2 = insertelement <16 x i16> %hsub1, i16 %a45, i32 2
+ %hsub3 = insertelement <16 x i16> %hsub2, i16 %a67, i32 3
+ %hsub4 = insertelement <16 x i16> %hsub3, i16 %b01, i32 4
+ %hsub5 = insertelement <16 x i16> %hsub4, i16 %b23, i32 5
+ %hsub6 = insertelement <16 x i16> %hsub5, i16 %b45, i32 6
+ %hsub7 = insertelement <16 x i16> %hsub6, i16 %b67, i32 7
+ %hsub8 = insertelement <16 x i16> %hsub7, i16 %a89, i32 8
+ %hsub9 = insertelement <16 x i16> %hsub8, i16 %aAB, i32 9
+ %hsubA = insertelement <16 x i16> %hsub9, i16 %aCD, i32 10
+ %hsubB = insertelement <16 x i16> %hsubA, i16 %aEF, i32 11
+ %hsubC = insertelement <16 x i16> %hsubB, i16 %b89, i32 12
+ %hsubD = insertelement <16 x i16> %hsubC, i16 %bAB, i32 13
+ %hsubE = insertelement <16 x i16> %hsubD, i16 %bCD, i32 14
+ %hsubF = insertelement <16 x i16> %hsubE, i16 %bEF, i32 15
+ %result = shufflevector <16 x i16> %hsubF, <16 x i16> %a, <16 x i32> <i32 15, i32 14, i32 poison, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 poison, i32 0>
+ ret <16 x i16> %result
+}
+
;
; v4i32
;
@@ -330,6 +727,51 @@ define <4 x i32> @sub_v4i32_01uu(<4 x i32> %a, <4 x i32> %b) {
ret <4 x i32> %result
}
+define <4 x i32> @sub_v4i32_32u0(<4 x i32> %a, <4 x i32> %b) {
+; SSE2-LABEL: @sub_v4i32_32u0(
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[A:%.*]], <4 x i32> <i32 2, i32 0, i32 poison, i32 4>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> [[A]], <4 x i32> <i32 3, i32 1, i32 poison, i32 5>
+; SSE2-NEXT: [[RESULT1:%.*]] = sub <4 x i32> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: ret <4 x i32> [[RESULT1]]
+;
+; SSE4-LABEL: @sub_v4i32_32u0(
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[A:%.*]], <4 x i32> <i32 2, i32 0, i32 poison, i32 4>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> [[A]], <4 x i32> <i32 3, i32 1, i32 poison, i32 5>
+; SSE4-NEXT: [[RESULT:%.*]] = sub <4 x i32> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: ret <4 x i32> [[RESULT]]
+;
+; AVX2-LABEL: @sub_v4i32_32u0(
+; AVX2-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[A:%.*]], <4 x i32> <i32 2, i32 0, i32 poison, i32 4>
+; AVX2-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> [[A]], <4 x i32> <i32 3, i32 1, i32 poison, i32 5>
+; AVX2-NEXT: [[RESULT:%.*]] = sub <4 x i32> [[TMP1]], [[TMP2]]
+; AVX2-NEXT: ret <4 x i32> [[RESULT]]
+;
+; AVX512-LABEL: @sub_v4i32_32u0(
+; AVX512-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[A:%.*]], <4 x i32> <i32 2, i32 0, i32 poison, i32 4>
+; AVX512-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> [[A]], <4 x i32> <i32 3, i32 1, i32 poison, i32 5>
+; AVX512-NEXT: [[RESULT1:%.*]] = sub <4 x i32> [[TMP1]], [[TMP2]]
+; AVX512-NEXT: ret <4 x i32> [[RESULT1]]
+;
+ %a0 = extractelement <4 x i32> %a, i32 0
+ %a1 = extractelement <4 x i32> %a, i32 1
+ %a2 = extractelement <4 x i32> %a, i32 2
+ %a3 = extractelement <4 x i32> %a, i32 3
+ %a01 = sub i32 %a0, %a1
+ %a23 = sub i32 %a2, %a3
+ %b0 = extractelement <4 x i32> %b, i32 0
+ %b1 = extractelement <4 x i32> %b, i32 1
+ %b2 = extractelement <4 x i32> %b, i32 2
+ %b3 = extractelement <4 x i32> %b, i32 3
+ %b01 = sub i32 %b0, %b1
+ %b23 = sub i32 %b2, %b3
+ %hsub0 = insertelement <4 x i32> poison, i32 %a01, i32 0
+ %hsub1 = insertelement <4 x i32> %hsub0, i32 %a23, i32 1
+ %hsub2 = insertelement <4 x i32> %hsub1, i32 %b01, i32 2
+ %hsub3 = insertelement <4 x i32> %hsub2, i32 %b23, i32 3
+ %result = shufflevector <4 x i32> %hsub3, <4 x i32> %a, <4 x i32> <i32 3, i32 2, i32 poison, i32 0>
+ ret <4 x i32> %result
+}
+
;
; v8i32
;
@@ -379,30 +821,30 @@ define <8 x i32> @sub_v8i32_01234567(<8 x i32> %a, <8 x i32> %b) {
define <8 x i32> @sub_v8i32_01234u67(<8 x i32> %a, <8 x i32> %b) {
; SSE2-LABEL: @sub_v8i32_01234u67(
-; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 poison, i32 12, i32 14>
-; SSE2-NEXT: [[TMP3:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 poison, i32 13, i32 15>
-; SSE2-NEXT: [[TMP4:%.*]] = sub <8 x i32> [[TMP2]], [[TMP3]]
-; SSE2-NEXT: ret <8 x i32> [[TMP4]]
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 poison, i32 12, i32 14>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 poison, i32 13, i32 15>
+; SSE2-NEXT: [[RESULT:%.*]] = sub <8 x i32> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: ret <8 x i32> [[RESULT]]
;
; SSE4-LABEL: @sub_v8i32_01234u67(
; SSE4-NEXT: [[A4:%.*]] = extractelement <8 x i32> [[A:%.*]], i64 4
; SSE4-NEXT: [[A5:%.*]] = extractelement <8 x i32> [[A]], i64 5
; SSE4-NEXT: [[A45:%.*]] = sub i32 [[A4]], [[A5]]
-; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP3:%.*]] = sub <8 x i32> [[TMP2]], [[TMP4]]
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP3:%.*]] = sub <8 x i32> [[TMP1]], [[TMP2]]
; SSE4-NEXT: [[HSUB4:%.*]] = insertelement <8 x i32> [[TMP3]], i32 [[A45]], i64 4
-; SSE4-NEXT: [[TMP7:%.*]] = shufflevector <8 x i32> [[B]], <8 x i32> poison, <8 x i32> <i32 4, i32 6, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x i32> [[B]], <8 x i32> poison, <8 x i32> <i32 4, i32 6, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <8 x i32> [[B]], <8 x i32> poison, <8 x i32> <i32 5, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP6:%.*]] = sub <8 x i32> [[TMP7]], [[TMP5]]
+; SSE4-NEXT: [[TMP6:%.*]] = sub <8 x i32> [[TMP4]], [[TMP5]]
; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <8 x i32> [[HSUB4]], <8 x i32> [[TMP6]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 poison, i32 8, i32 9>
; SSE4-NEXT: ret <8 x i32> [[RESULT]]
;
; AVX-LABEL: @sub_v8i32_01234u67(
-; AVX-NEXT: [[TMP5:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 poison, i32 12, i32 14>
-; AVX-NEXT: [[TMP6:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 poison, i32 13, i32 15>
-; AVX-NEXT: [[TMP7:%.*]] = sub <8 x i32> [[TMP5]], [[TMP6]]
-; AVX-NEXT: ret <8 x i32> [[TMP7]]
+; AVX-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 poison, i32 12, i32 14>
+; AVX-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 poison, i32 13, i32 15>
+; AVX-NEXT: [[RESULT:%.*]] = sub <8 x i32> [[TMP1]], [[TMP2]]
+; AVX-NEXT: ret <8 x i32> [[RESULT]]
;
%a0 = extractelement <8 x i32> %a, i32 0
%a1 = extractelement <8 x i32> %a, i32 1
@@ -599,8 +1041,8 @@ define <4 x float> @sub_v4f32_012u(<4 x float> %a, <4 x float> %b) {
define <4 x float> @sub_v4f32_uu23(<4 x float> %a, <4 x float> %b) {
; CHECK-LABEL: @sub_v4f32_uu23(
-; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[TMP1:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 0, i32 2>
-; CHECK-NEXT: [[RESULT1:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 1, i32 3>
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 0, i32 2>
+; CHECK-NEXT: [[RESULT1:%.*]] = shufflevector <4 x float> [[B]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 1, i32 3>
; CHECK-NEXT: [[RESULT2:%.*]] = fsub <4 x float> [[TMP2]], [[RESULT1]]
; CHECK-NEXT: ret <4 x float> [[RESULT2]]
;
@@ -626,8 +1068,8 @@ define <4 x float> @sub_v4f32_uu23(<4 x float> %a, <4 x float> %b) {
define <4 x float> @sub_v4f32_01uu(<4 x float> %a, <4 x float> %b) {
; CHECK-LABEL: @sub_v4f32_01uu(
-; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[TMP1:%.*]], <4 x float> poison, <4 x i32> <i32 0, i32 2, i32 poison, i32 poison>
-; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> poison, <4 x i32> <i32 1, i32 3, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 0, i32 2, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[A]], <4 x float> poison, <4 x i32> <i32 1, i32 3, i32 poison, i32 poison>
; CHECK-NEXT: [[TMP4:%.*]] = fsub <4 x float> [[TMP2]], [[TMP3]]
; CHECK-NEXT: ret <4 x float> [[TMP4]]
;
@@ -718,14 +1160,14 @@ define <8 x float> @sub_v8f32_012u4567(<8 x float> %a, <8 x float> %b) {
; SSE4-NEXT: [[A6:%.*]] = extractelement <8 x float> [[A:%.*]], i64 6
; SSE4-NEXT: [[A7:%.*]] = extractelement <8 x float> [[A]], i64 7
; SSE4-NEXT: [[A67:%.*]] = fsub float [[A6]], [[A7]]
-; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[A]], <8 x float> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 poison, i32 4, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[A]], <8 x float> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 poison, i32 5, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP3:%.*]] = fsub <8 x float> [[TMP2]], [[TMP5]]
-; SSE4-NEXT: [[HSUB5:%.*]] = insertelement <8 x float> [[TMP3]], float [[A67]], i64 5
-; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 4, i32 6, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-; SSE4-NEXT: [[TMP7:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 5, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x float> [[A]], <8 x float> [[B:%.*]], <8 x i32> <i32 0, i32 2, i32 8, i32 poison, i32 4, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP7:%.*]] = shufflevector <8 x float> [[A]], <8 x float> [[B]], <8 x i32> <i32 1, i32 3, i32 9, i32 poison, i32 5, i32 poison, i32 poison, i32 poison>
; SSE4-NEXT: [[TMP6:%.*]] = fsub <8 x float> [[TMP4]], [[TMP7]]
-; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <8 x float> [[HSUB5]], <8 x float> [[TMP6]], <8 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 4, i32 5, i32 8, i32 9>
+; SSE4-NEXT: [[HSUB5:%.*]] = insertelement <8 x float> [[TMP6]], float [[A67]], i64 5
+; SSE4-NEXT: [[TMP8:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 4, i32 6, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 5, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP9:%.*]] = fsub <8 x float> [[TMP8]], [[TMP5]]
+; SSE4-NEXT: [[RESULT:%.*]] = shufflevector <8 x float> [[HSUB5]], <8 x float> [[TMP9]], <8 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 4, i32 5, i32 8, i32 9>
; SSE4-NEXT: ret <8 x float> [[RESULT]]
;
; AVX-LABEL: @sub_v8f32_012u4567(
@@ -770,6 +1212,79 @@ define <8 x float> @sub_v8f32_012u4567(<8 x float> %a, <8 x float> %b) {
ret <8 x float> %result
}
+define <8 x float> @sub_v8f32_76u43210(<8 x float> %a, <8 x float> %b) {
+; SSE2-LABEL: @sub_v8f32_76u43210(
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <8 x float> [[A:%.*]], <8 x float> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[A]], <8 x float> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; SSE2-NEXT: [[TMP3:%.*]] = fsub <4 x float> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: [[B0:%.*]] = extractelement <8 x float> [[B:%.*]], i64 0
+; SSE2-NEXT: [[B1:%.*]] = extractelement <8 x float> [[B]], i64 1
+; SSE2-NEXT: [[B01:%.*]] = fsub float [[B0]], [[B1]]
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <2 x i32> <i32 4, i32 6>
+; SSE2-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <2 x i32> <i32 5, i32 7>
+; SSE2-NEXT: [[TMP6:%.*]] = fsub <2 x float> [[TMP4]], [[TMP5]]
+; SSE2-NEXT: [[TMP7:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[HSUB4:%.*]] = insertelement <8 x float> [[TMP7]], float [[B01]], i64 4
+; SSE2-NEXT: [[TMP8:%.*]] = shufflevector <2 x float> [[TMP6]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE2-NEXT: [[RESULT:%.*]] = shufflevector <8 x float> [[TMP8]], <8 x float> [[HSUB4]], <8 x i32> <i32 1, i32 0, i32 poison, i32 12, i32 11, i32 10, i32 9, i32 8>
+; SSE2-NEXT: ret <8 x float> [[RESULT]]
+;
+; SSE4-LABEL: @sub_v8f32_76u43210(
+; SSE4-NEXT: [[B0:%.*]] = extractelement <8 x float> [[B:%.*]], i64 0
+; SSE4-NEXT: [[B1:%.*]] = extractelement <8 x float> [[B]], i64 1
+; SSE4-NEXT: [[B01:%.*]] = fsub float [[B0]], [[B1]]
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <8 x float> [[A:%.*]], <8 x float> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[A]], <8 x float> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[RESULT:%.*]] = fsub <8 x float> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: [[HSUB4:%.*]] = insertelement <8 x float> [[RESULT]], float [[B01]], i64 4
+; SSE4-NEXT: [[TMP4:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 4, i32 6, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[B]], <8 x float> poison, <8 x i32> <i32 5, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP6:%.*]] = fsub <8 x float> [[TMP4]], [[TMP5]]
+; SSE4-NEXT: [[RESULT1:%.*]] = shufflevector <8 x float> [[TMP6]], <8 x float> [[HSUB4]], <8 x i32> <i32 1, i32 0, i32 poison, i32 12, i32 11, i32 10, i32 9, i32 8>
+; SSE4-NEXT: ret <8 x float> [[RESULT1]]
+;
+; AVX-LABEL: @sub_v8f32_76u43210(
+; AVX-NEXT: [[TMP1:%.*]] = shufflevector <8 x float> [[B:%.*]], <8 x float> [[A:%.*]], <8 x i32> <i32 6, i32 4, i32 poison, i32 0, i32 14, i32 12, i32 10, i32 8>
+; AVX-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[B]], <8 x float> [[A]], <8 x i32> <i32 7, i32 5, i32 poison, i32 1, i32 15, i32 13, i32 11, i32 9>
+; AVX-NEXT: [[RESULT:%.*]] = fsub <8 x float> [[TMP1]], [[TMP2]]
+; AVX-NEXT: ret <8 x float> [[RESULT]]
+;
+ %a0 = extractelement <8 x float> %a, i32 0
+ %a1 = extractelement <8 x float> %a, i32 1
+ %a2 = extractelement <8 x float> %a, i32 2
+ %a3 = extractelement <8 x float> %a, i32 3
+ %a4 = extractelement <8 x float> %a, i32 4
+ %a5 = extractelement <8 x float> %a, i32 5
+ %a6 = extractelement <8 x float> %a, i32 6
+ %a7 = extractelement <8 x float> %a, i32 7
+ %a01 = fsub float %a0, %a1
+ %a23 = fsub float %a2, %a3
+ %a45 = fsub float %a4, %a5
+ %a67 = fsub float %a6, %a7
+ %b0 = extractelement <8 x float> %b, i32 0
+ %b1 = extractelement <8 x float> %b, i32 1
+ %b2 = extractelement <8 x float> %b, i32 2
+ %b3 = extractelement <8 x float> %b, i32 3
+ %b4 = extractelement <8 x float> %b, i32 4
+ %b5 = extractelement <8 x float> %b, i32 5
+ %b6 = extractelement <8 x float> %b, i32 6
+ %b7 = extractelement <8 x float> %b, i32 7
+ %b01 = fsub float %b0, %b1
+ %b23 = fsub float %b2, %b3
+ %b45 = fsub float %b4, %b5
+ %b67 = fsub float %b6, %b7
+ %hsub0 = insertelement <8 x float> poison, float %a01, i32 0
+ %hsub1 = insertelement <8 x float> %hsub0, float %a23, i32 1
+ %hsub2 = insertelement <8 x float> %hsub1, float %a45, i32 2
+ %hsub3 = insertelement <8 x float> %hsub2, float %a67, i32 3
+ %hsub4 = insertelement <8 x float> %hsub3, float %b01, i32 4
+ %hsub5 = insertelement <8 x float> %hsub4, float %b23, i32 5
+ %hsub6 = insertelement <8 x float> %hsub5, float %b45, i32 6
+ %hsub7 = insertelement <8 x float> %hsub6, float %b67, i32 7
+ %result = shufflevector <8 x float> %hsub7, <8 x float> %a, <8 x i32> <i32 7, i32 6, i32 poison, i32 4, i32 3, i32 2, i32 1, i32 0>
+ ret <8 x float> %result
+}
+
;
; v2f64
;
@@ -924,9 +1439,9 @@ define <4 x double> @sub_v4f64_0u23(<4 x double> %a, <4 x double> %b) {
; SSE4-NEXT: [[B2:%.*]] = extractelement <4 x double> [[B:%.*]], i64 2
; SSE4-NEXT: [[B3:%.*]] = extractelement <4 x double> [[B]], i64 3
; SSE4-NEXT: [[B23:%.*]] = fsub double [[B2]], [[B3]]
-; SSE4-NEXT: [[TMP3:%.*]] = shufflevector <4 x double> [[TMP1:%.*]], <4 x double> poison, <4 x i32> <i32 1, i32 poison, i32 3, i32 poison>
-; SSE4-NEXT: [[TMP2:%.*]] = fsub <4 x double> [[TMP1]], [[TMP3]]
-; SSE4-NEXT: [[RESULT:%.*]] = insertelement <4 x double> [[TMP2]], double [[B23]], i64 3
+; SSE4-NEXT: [[TMP3:%.*]] = shufflevector <4 x double> [[TMP2:%.*]], <4 x double> poison, <4 x i32> <i32 1, i32 poison, i32 3, i32 poison>
+; SSE4-NEXT: [[TMP4:%.*]] = fsub <4 x double> [[TMP2]], [[TMP3]]
+; SSE4-NEXT: [[RESULT:%.*]] = insertelement <4 x double> [[TMP4]], double [[B23]], i64 3
; SSE4-NEXT: ret <4 x double> [[RESULT]]
;
; AVX-LABEL: @sub_v4f64_0u23(
@@ -1130,3 +1645,51 @@ define <4 x double> @sub_v4f64_01uu(<4 x double> %a, <4 x double> %b) {
%result = shufflevector <4 x double> %hsub3, <4 x double> %a, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
ret <4 x double> %result
}
+
+define <4 x double> @sub_v4f64_32u0(<4 x double> %a, <4 x double> %b) {
+; SSE2-LABEL: @sub_v4f64_32u0(
+; SSE2-NEXT: [[A0:%.*]] = extractelement <4 x double> [[A:%.*]], i64 0
+; SSE2-NEXT: [[A1:%.*]] = extractelement <4 x double> [[A]], i64 1
+; SSE2-NEXT: [[A01:%.*]] = fsub double [[A0]], [[A1]]
+; SSE2-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[B:%.*]], <4 x double> poison, <2 x i32> <i32 2, i32 0>
+; SSE2-NEXT: [[TMP2:%.*]] = shufflevector <4 x double> [[B]], <4 x double> poison, <2 x i32> <i32 3, i32 1>
+; SSE2-NEXT: [[TMP3:%.*]] = fsub <2 x double> [[TMP1]], [[TMP2]]
+; SSE2-NEXT: [[TMP4:%.*]] = shufflevector <2 x double> [[TMP3]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
+; SSE2-NEXT: [[RESULT:%.*]] = insertelement <4 x double> [[TMP4]], double [[A01]], i64 3
+; SSE2-NEXT: ret <4 x double> [[RESULT]]
+;
+; SSE4-LABEL: @sub_v4f64_32u0(
+; SSE4-NEXT: [[A0:%.*]] = extractelement <4 x double> [[A:%.*]], i64 0
+; SSE4-NEXT: [[A1:%.*]] = extractelement <4 x double> [[A]], i64 1
+; SSE4-NEXT: [[A01:%.*]] = fsub double [[A0]], [[A1]]
+; SSE4-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[B:%.*]], <4 x double> poison, <4 x i32> <i32 2, i32 0, i32 poison, i32 poison>
+; SSE4-NEXT: [[TMP2:%.*]] = shufflevector <4 x double> [[B]], <4 x double> poison, <4 x i32> <i32 3, i32 1, i32 poison, i32 poison>
+; SSE4-NEXT: [[RESULT:%.*]] = fsub <4 x double> [[TMP1]], [[TMP2]]
+; SSE4-NEXT: [[RESULT1:%.*]] = insertelement <4 x double> [[RESULT]], double [[A01]], i64 3
+; SSE4-NEXT: ret <4 x double> [[RESULT1]]
+;
+; AVX-LABEL: @sub_v4f64_32u0(
+; AVX-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[B:%.*]], <4 x double> [[A:%.*]], <4 x i32> <i32 2, i32 0, i32 poison, i32 4>
+; AVX-NEXT: [[TMP2:%.*]] = shufflevector <4 x double> [[B]], <4 x double> [[A]], <4 x i32> <i32 3, i32 1, i32 poison, i32 5>
+; AVX-NEXT: [[RESULT:%.*]] = fsub <4 x double> [[TMP1]], [[TMP2]]
+; AVX-NEXT: ret <4 x double> [[RESULT]]
+;
+ %a0 = extractelement <4 x double> %a, i32 0
+ %a1 = extractelement <4 x double> %a, i32 1
+ %a2 = extractelement <4 x double> %a, i32 2
+ %a3 = extractelement <4 x double> %a, i32 3
+ %a01 = fsub double %a0, %a1
+ %a23 = fsub double %a2, %a3
+ %b0 = extractelement <4 x double> %b, i32 0
+ %b1 = extractelement <4 x double> %b, i32 1
+ %b2 = extractelement <4 x double> %b, i32 2
+ %b3 = extractelement <4 x double> %b, i32 3
+ %b01 = fsub double %b0, %b1
+ %b23 = fsub double %b2, %b3
+ %hsub0 = insertelement <4 x double> poison, double %a01, i32 0
+ %hsub1 = insertelement <4 x double> %hsub0, double %a23, i32 1
+ %hsub2 = insertelement <4 x double> %hsub1, double %b01, i32 2
+ %hsub3 = insertelement <4 x double> %hsub2, double %b23, i32 3
+ %result = shufflevector <4 x double> %hsub3, <4 x double> %a, <4 x i32> <i32 3, i32 2, i32 poison, i32 0>
+ ret <4 x double> %result
+}
More information about the llvm-commits
mailing list