summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--llvm/lib/Target/X86/X86ISelLowering.cpp9
-rw-r--r--llvm/lib/Target/X86/X86InstrAVX512.td57
-rw-r--r--llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll4
-rwxr-xr-xllvm/test/CodeGen/X86/avx512-shuffle-schedule.ll32
-rw-r--r--llvm/test/CodeGen/X86/avx512-shuffles/shuffle-vec.ll16
-rw-r--r--llvm/test/CodeGen/X86/vector-shuffle-512-v16.ll4
6 files changed, 74 insertions, 48 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 8d01054b9ff..5205bf4f1b1 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -31395,15 +31395,6 @@ static bool combineBitcastForMaskedOp(SDValue OrigOp, SelectionDAG &DAG,
unsigned Opcode = Op.getOpcode();
switch (Opcode) {
- case X86ISD::SHUF128: {
- if (EltVT.getSizeInBits() != 32 && EltVT.getSizeInBits() != 64)
- return false;
- // Only change element size, not type.
- if (VT.isInteger() != Op.getSimpleValueType().isInteger())
- return false;
- return BitcastAndCombineShuffle(Opcode, Op.getOperand(0), Op.getOperand(1),
- Op.getOperand(2));
- }
case X86ISD::SUBV_BROADCAST: {
unsigned EltSize = EltVT.getSizeInBits();
if (EltSize != 32 && EltSize != 64)
diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 7b1cc56b43f..c6a404f4dfb 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -9361,25 +9361,60 @@ def : Pat<(v4f64 (ftrunc VR256X:$src)),
(VRNDSCALEPDZ256rri VR256X:$src, (i32 0xB))>;
}
+multiclass avx512_shuff_packed_128_common<bits<8> opc, string OpcodeStr,
+ OpndItins itins, X86VectorVTInfo _,
+ X86VectorVTInfo CastInfo> {
+ let ExeDomain = _.ExeDomain in {
+ defm rri : AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.RC:$src2, u8imm:$src3),
+ OpcodeStr, "$src3, $src2, $src1", "$src1, $src2, $src3",
+ (_.VT (bitconvert
+ (CastInfo.VT (X86Shuf128 _.RC:$src1, _.RC:$src2,
+ (i8 imm:$src3))))),
+ itins.rr>, Sched<[itins.Sched]>;
+ defm rmi : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.MemOp:$src2, u8imm:$src3),
+ OpcodeStr, "$src3, $src2, $src1", "$src1, $src2, $src3",
+ (_.VT
+ (bitconvert
+ (CastInfo.VT (X86Shuf128 _.RC:$src1,
+ (bitconvert (_.LdFrag addr:$src2)),
+ (i8 imm:$src3))))), itins.rm>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+ defm rmbi : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.ScalarMemOp:$src2, u8imm:$src3),
+ OpcodeStr, "$src3, ${src2}"##_.BroadcastStr##", $src1",
+ "$src1, ${src2}"##_.BroadcastStr##", $src3",
+ (_.VT
+ (bitconvert
+ (CastInfo.VT
+ (X86Shuf128 _.RC:$src1,
+ (X86VBroadcast (_.ScalarLdFrag addr:$src2)),
+ (i8 imm:$src3))))), itins.rm>, EVEX_B,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+ }
+}
+
multiclass avx512_shuff_packed_128<string OpcodeStr, OpndItins itins,
- AVX512VLVectorVTInfo _, bits<8> opc>{
- let Predicates = [HasAVX512] in {
- defm Z : avx512_3Op_imm8<opc, OpcodeStr, X86Shuf128, itins, _.info512>, EVEX_V512;
+ AVX512VLVectorVTInfo _,
+ AVX512VLVectorVTInfo CastInfo, bits<8> opc>{
+ let Predicates = [HasAVX512] in
+ defm Z : avx512_shuff_packed_128_common<opc, OpcodeStr, itins,
+ _.info512, CastInfo.info512>, EVEX_V512;
- }
- let Predicates = [HasAVX512, HasVLX] in {
- defm Z256 : avx512_3Op_imm8<opc, OpcodeStr, X86Shuf128, itins, _.info256>, EVEX_V256;
- }
+ let Predicates = [HasAVX512, HasVLX] in
+ defm Z256 : avx512_shuff_packed_128_common<opc, OpcodeStr, itins,
+ _.info256, CastInfo.info256>, EVEX_V256;
}
defm VSHUFF32X4 : avx512_shuff_packed_128<"vshuff32x4", SSE_SHUFP,
- avx512vl_f32_info, 0x23>, AVX512AIi8Base, EVEX_4V, EVEX_CD8<32, CD8VF>;
+ avx512vl_f32_info, avx512vl_f64_info, 0x23>, AVX512AIi8Base, EVEX_4V, EVEX_CD8<32, CD8VF>;
defm VSHUFF64X2 : avx512_shuff_packed_128<"vshuff64x2", SSE_SHUFP,
- avx512vl_f64_info, 0x23>, AVX512AIi8Base, EVEX_4V, EVEX_CD8<64, CD8VF>, VEX_W;
+ avx512vl_f64_info, avx512vl_f64_info, 0x23>, AVX512AIi8Base, EVEX_4V, EVEX_CD8<64, CD8VF>, VEX_W;
defm VSHUFI32X4 : avx512_shuff_packed_128<"vshufi32x4", SSE_SHUFP,
- avx512vl_i32_info, 0x43>, AVX512AIi8Base, EVEX_4V, EVEX_CD8<32, CD8VF>;
+ avx512vl_i32_info, avx512vl_i64_info, 0x43>, AVX512AIi8Base, EVEX_4V, EVEX_CD8<32, CD8VF>;
defm VSHUFI64X2 : avx512_shuff_packed_128<"vshufi64x2", SSE_SHUFP,
- avx512vl_i64_info, 0x43>, AVX512AIi8Base, EVEX_4V, EVEX_CD8<64, CD8VF>, VEX_W;
+ avx512vl_i64_info, avx512vl_i64_info, 0x43>, AVX512AIi8Base, EVEX_4V, EVEX_CD8<64, CD8VF>, VEX_W;
let Predicates = [HasAVX512] in {
// Provide fallback in case the load node that is used in the broadcast
diff --git a/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
index 27a57d5825b..e5dce2d7248 100644
--- a/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
+++ b/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
@@ -58,12 +58,12 @@ entry:
define <16 x float> @test_mm512_shuffle_f32x4(<16 x float> %__A, <16 x float> %__B) {
; X32-LABEL: test_mm512_shuffle_f32x4:
; X32: # %bb.0: # %entry
-; X32-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[0,1,2,3],zmm1[0,1,0,1]
+; X32-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[0,1,2,3,4,5,6,7],zmm1[0,1,2,3,0,1,2,3]
; X32-NEXT: retl
;
; X64-LABEL: test_mm512_shuffle_f32x4:
; X64: # %bb.0: # %entry
-; X64-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[0,1,2,3],zmm1[0,1,0,1]
+; X64-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[0,1,2,3,4,5,6,7],zmm1[0,1,2,3,0,1,2,3]
; X64-NEXT: retq
entry:
%shuffle = shufflevector <16 x float> %__A, <16 x float> %__B, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 16, i32 17, i32 18, i32 19>
diff --git a/llvm/test/CodeGen/X86/avx512-shuffle-schedule.ll b/llvm/test/CodeGen/X86/avx512-shuffle-schedule.ll
index 544d7e2f1a8..bd2ec4a22f6 100755
--- a/llvm/test/CodeGen/X86/avx512-shuffle-schedule.ll
+++ b/llvm/test/CodeGen/X86/avx512-shuffle-schedule.ll
@@ -9114,12 +9114,12 @@ define <8 x float> @test_8xfloat_zero_masked_shuff_mem_mask3(<8 x float> %vec1,
define <16 x float> @test_16xfloat_shuff_mask0(<16 x float> %vec1, <16 x float> %vec2, <16 x i32> %mask) {
; GENERIC-LABEL: test_16xfloat_shuff_mask0:
; GENERIC: # %bb.0:
-; GENERIC-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[6,7,0,1],zmm1[2,3,6,7] sched: [1:1.00]
+; GENERIC-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[12,13,14,15,0,1,2,3],zmm1[4,5,6,7,12,13,14,15] sched: [1:1.00]
; GENERIC-NEXT: retq # sched: [1:1.00]
;
; SKX-LABEL: test_16xfloat_shuff_mask0:
; SKX: # %bb.0:
-; SKX-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[6,7,0,1],zmm1[2,3,6,7] sched: [3:1.00]
+; SKX-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[12,13,14,15,0,1,2,3],zmm1[4,5,6,7,12,13,14,15] sched: [3:1.00]
; SKX-NEXT: retq # sched: [7:1.00]
%res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 12, i32 13, i32 14, i32 15, i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 28, i32 29, i32 30, i32 31>
ret <16 x float> %res
@@ -9238,12 +9238,12 @@ define <16 x float> @test_16xfloat_zero_masked_shuff_mask2(<16 x float> %vec1, <
define <16 x float> @test_16xfloat_shuff_mask3(<16 x float> %vec1, <16 x float> %vec2) {
; GENERIC-LABEL: test_16xfloat_shuff_mask3:
; GENERIC: # %bb.0:
-; GENERIC-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[4,5,6,7],zmm1[0,1,4,5] sched: [1:1.00]
+; GENERIC-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[8,9,10,11,12,13,14,15],zmm1[0,1,2,3,8,9,10,11] sched: [1:1.00]
; GENERIC-NEXT: retq # sched: [1:1.00]
;
; SKX-LABEL: test_16xfloat_shuff_mask3:
; SKX: # %bb.0:
-; SKX-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[4,5,6,7],zmm1[0,1,4,5] sched: [3:1.00]
+; SKX-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[8,9,10,11,12,13,14,15],zmm1[0,1,2,3,8,9,10,11] sched: [3:1.00]
; SKX-NEXT: retq # sched: [7:1.00]
%res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 24, i32 25, i32 26, i32 27>
ret <16 x float> %res
@@ -9288,12 +9288,12 @@ define <16 x float> @test_16xfloat_zero_masked_shuff_mask3(<16 x float> %vec1, <
define <16 x float> @test_16xfloat_shuff_mem_mask0(<16 x float> %vec1, <16 x float>* %vec2p) {
; GENERIC-LABEL: test_16xfloat_shuff_mem_mask0:
; GENERIC: # %bb.0:
-; GENERIC-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[6,7,4,5],mem[4,5,2,3] sched: [5:1.00]
+; GENERIC-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[12,13,14,15,8,9,10,11],mem[8,9,10,11,4,5,6,7] sched: [5:1.00]
; GENERIC-NEXT: retq # sched: [1:1.00]
;
; SKX-LABEL: test_16xfloat_shuff_mem_mask0:
; SKX: # %bb.0:
-; SKX-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[6,7,4,5],mem[4,5,2,3] sched: [10:1.00]
+; SKX-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[12,13,14,15,8,9,10,11],mem[8,9,10,11,4,5,6,7] sched: [10:1.00]
; SKX-NEXT: retq # sched: [7:1.00]
%vec2 = load <16 x float>, <16 x float>* %vec2p
%res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 24, i32 25, i32 26, i32 27, i32 20, i32 21, i32 22, i32 23>
@@ -9422,12 +9422,12 @@ define <16 x float> @test_16xfloat_zero_masked_shuff_mem_mask2(<16 x float> %vec
define <16 x float> @test_16xfloat_shuff_mem_mask3(<16 x float> %vec1, <16 x float>* %vec2p) {
; GENERIC-LABEL: test_16xfloat_shuff_mem_mask3:
; GENERIC: # %bb.0:
-; GENERIC-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[2,3,0,1],mem[6,7,6,7] sched: [5:1.00]
+; GENERIC-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,0,1,2,3],mem[12,13,14,15,12,13,14,15] sched: [5:1.00]
; GENERIC-NEXT: retq # sched: [1:1.00]
;
; SKX-LABEL: test_16xfloat_shuff_mem_mask3:
; SKX: # %bb.0:
-; SKX-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[2,3,0,1],mem[6,7,6,7] sched: [10:1.00]
+; SKX-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,0,1,2,3],mem[12,13,14,15,12,13,14,15] sched: [10:1.00]
; SKX-NEXT: retq # sched: [7:1.00]
%vec2 = load <16 x float>, <16 x float>* %vec2p
%res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 28, i32 29, i32 30, i32 31, i32 28, i32 29, i32 30, i32 31>
@@ -10562,12 +10562,12 @@ define <8 x i32> @test_8xi32_zero_masked_shuff_mem_mask3(<8 x i32> %vec1, <8 x i
define <16 x i32> @test_16xi32_shuff_mask0(<16 x i32> %vec1, <16 x i32> %vec2) {
; GENERIC-LABEL: test_16xi32_shuff_mask0:
; GENERIC: # %bb.0:
-; GENERIC-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[2,3,2,3],zmm1[2,3,6,7] sched: [1:1.00]
+; GENERIC-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,4,5,6,7],zmm1[4,5,6,7,12,13,14,15] sched: [1:1.00]
; GENERIC-NEXT: retq # sched: [1:1.00]
;
; SKX-LABEL: test_16xi32_shuff_mask0:
; SKX: # %bb.0:
-; SKX-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[2,3,2,3],zmm1[2,3,6,7] sched: [3:1.00]
+; SKX-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,4,5,6,7],zmm1[4,5,6,7,12,13,14,15] sched: [3:1.00]
; SKX-NEXT: retq # sched: [7:1.00]
%res = shufflevector <16 x i32> %vec1, <16 x i32> %vec2, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 20, i32 21, i32 22, i32 23, i32 28, i32 29, i32 30, i32 31>
ret <16 x i32> %res
@@ -10686,12 +10686,12 @@ define <16 x i32> @test_16xi32_zero_masked_shuff_mask2(<16 x i32> %vec1, <16 x i
define <16 x i32> @test_16xi32_shuff_mask3(<16 x i32> %vec1, <16 x i32> %vec2) {
; GENERIC-LABEL: test_16xi32_shuff_mask3:
; GENERIC: # %bb.0:
-; GENERIC-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[2,3,0,1],zmm1[4,5,2,3] sched: [1:1.00]
+; GENERIC-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,0,1,2,3],zmm1[8,9,10,11,4,5,6,7] sched: [1:1.00]
; GENERIC-NEXT: retq # sched: [1:1.00]
;
; SKX-LABEL: test_16xi32_shuff_mask3:
; SKX: # %bb.0:
-; SKX-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[2,3,0,1],zmm1[4,5,2,3] sched: [3:1.00]
+; SKX-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,0,1,2,3],zmm1[8,9,10,11,4,5,6,7] sched: [3:1.00]
; SKX-NEXT: retq # sched: [7:1.00]
%res = shufflevector <16 x i32> %vec1, <16 x i32> %vec2, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 24, i32 25, i32 26, i32 27, i32 20, i32 21, i32 22, i32 23>
ret <16 x i32> %res
@@ -10736,12 +10736,12 @@ define <16 x i32> @test_16xi32_zero_masked_shuff_mask3(<16 x i32> %vec1, <16 x i
define <16 x i32> @test_16xi32_shuff_mem_mask0(<16 x i32> %vec1, <16 x i32>* %vec2p) {
; GENERIC-LABEL: test_16xi32_shuff_mem_mask0:
; GENERIC: # %bb.0:
-; GENERIC-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[4,5,2,3],mem[4,5,0,1] sched: [5:1.00]
+; GENERIC-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[8,9,10,11,4,5,6,7],mem[8,9,10,11,0,1,2,3] sched: [5:1.00]
; GENERIC-NEXT: retq # sched: [1:1.00]
;
; SKX-LABEL: test_16xi32_shuff_mem_mask0:
; SKX: # %bb.0:
-; SKX-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[4,5,2,3],mem[4,5,0,1] sched: [10:1.00]
+; SKX-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[8,9,10,11,4,5,6,7],mem[8,9,10,11,0,1,2,3] sched: [10:1.00]
; SKX-NEXT: retq # sched: [7:1.00]
%vec2 = load <16 x i32>, <16 x i32>* %vec2p
%res = shufflevector <16 x i32> %vec1, <16 x i32> %vec2, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7, i32 24, i32 25, i32 26, i32 27, i32 16, i32 17, i32 18, i32 19>
@@ -10870,12 +10870,12 @@ define <16 x i32> @test_16xi32_zero_masked_shuff_mem_mask2(<16 x i32> %vec1, <16
define <16 x i32> @test_16xi32_shuff_mem_mask3(<16 x i32> %vec1, <16 x i32>* %vec2p) {
; GENERIC-LABEL: test_16xi32_shuff_mem_mask3:
; GENERIC: # %bb.0:
-; GENERIC-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[2,3,2,3],mem[2,3,6,7] sched: [5:1.00]
+; GENERIC-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,4,5,6,7],mem[4,5,6,7,12,13,14,15] sched: [5:1.00]
; GENERIC-NEXT: retq # sched: [1:1.00]
;
; SKX-LABEL: test_16xi32_shuff_mem_mask3:
; SKX: # %bb.0:
-; SKX-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[2,3,2,3],mem[2,3,6,7] sched: [10:1.00]
+; SKX-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,4,5,6,7],mem[4,5,6,7,12,13,14,15] sched: [10:1.00]
; SKX-NEXT: retq # sched: [7:1.00]
%vec2 = load <16 x i32>, <16 x i32>* %vec2p
%res = shufflevector <16 x i32> %vec1, <16 x i32> %vec2, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 20, i32 21, i32 22, i32 23, i32 28, i32 29, i32 30, i32 31>
diff --git a/llvm/test/CodeGen/X86/avx512-shuffles/shuffle-vec.ll b/llvm/test/CodeGen/X86/avx512-shuffles/shuffle-vec.ll
index d4f12747028..e6a1c05c59c 100644
--- a/llvm/test/CodeGen/X86/avx512-shuffles/shuffle-vec.ll
+++ b/llvm/test/CodeGen/X86/avx512-shuffles/shuffle-vec.ll
@@ -260,7 +260,7 @@ define <8 x float> @test_8xfloat_zero_masked_shuff_mem_mask3(<8 x float> %vec1,
define <16 x float> @test_16xfloat_shuff_mask0(<16 x float> %vec1, <16 x float> %vec2) {
; CHECK-LABEL: test_16xfloat_shuff_mask0:
; CHECK: # %bb.0:
-; CHECK-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[6,7,0,1],zmm1[2,3,6,7]
+; CHECK-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[12,13,14,15,0,1,2,3],zmm1[4,5,6,7,12,13,14,15]
; CHECK-NEXT: retq
%res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 12, i32 13, i32 14, i32 15, i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 28, i32 29, i32 30, i32 31>
ret <16 x float> %res
@@ -346,7 +346,7 @@ define <16 x float> @test_16xfloat_zero_masked_shuff_mask2(<16 x float> %vec1, <
define <16 x float> @test_16xfloat_shuff_mask3(<16 x float> %vec1, <16 x float> %vec2) {
; CHECK-LABEL: test_16xfloat_shuff_mask3:
; CHECK: # %bb.0:
-; CHECK-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[4,5,6,7],zmm1[0,1,4,5]
+; CHECK-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[8,9,10,11,12,13,14,15],zmm1[0,1,2,3,8,9,10,11]
; CHECK-NEXT: retq
%res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 24, i32 25, i32 26, i32 27>
ret <16 x float> %res
@@ -380,7 +380,7 @@ define <16 x float> @test_16xfloat_zero_masked_shuff_mask3(<16 x float> %vec1, <
define <16 x float> @test_16xfloat_shuff_mem_mask0(<16 x float> %vec1, <16 x float>* %vec2p) {
; CHECK-LABEL: test_16xfloat_shuff_mem_mask0:
; CHECK: # %bb.0:
-; CHECK-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[6,7,4,5],mem[4,5,2,3]
+; CHECK-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[12,13,14,15,8,9,10,11],mem[8,9,10,11,4,5,6,7]
; CHECK-NEXT: retq
%vec2 = load <16 x float>, <16 x float>* %vec2p
%res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 24, i32 25, i32 26, i32 27, i32 20, i32 21, i32 22, i32 23>
@@ -476,7 +476,7 @@ define <16 x float> @test_16xfloat_zero_masked_shuff_mem_mask2(<16 x float> %vec
define <16 x float> @test_16xfloat_shuff_mem_mask3(<16 x float> %vec1, <16 x float>* %vec2p) {
; CHECK-LABEL: test_16xfloat_shuff_mem_mask3:
; CHECK: # %bb.0:
-; CHECK-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[2,3,0,1],mem[6,7,6,7]
+; CHECK-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,0,1,2,3],mem[12,13,14,15,12,13,14,15]
; CHECK-NEXT: retq
%vec2 = load <16 x float>, <16 x float>* %vec2p
%res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 28, i32 29, i32 30, i32 31, i32 28, i32 29, i32 30, i32 31>
@@ -1260,7 +1260,7 @@ define <8 x i32> @test_8xi32_zero_masked_shuff_mem_mask3(<8 x i32> %vec1, <8 x i
define <16 x i32> @test_16xi32_shuff_mask0(<16 x i32> %vec1, <16 x i32> %vec2) {
; CHECK-LABEL: test_16xi32_shuff_mask0:
; CHECK: # %bb.0:
-; CHECK-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[2,3,2,3],zmm1[2,3,6,7]
+; CHECK-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,4,5,6,7],zmm1[4,5,6,7,12,13,14,15]
; CHECK-NEXT: retq
%res = shufflevector <16 x i32> %vec1, <16 x i32> %vec2, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 20, i32 21, i32 22, i32 23, i32 28, i32 29, i32 30, i32 31>
ret <16 x i32> %res
@@ -1340,7 +1340,7 @@ define <16 x i32> @test_16xi32_zero_masked_shuff_mask2(<16 x i32> %vec1, <16 x i
define <16 x i32> @test_16xi32_shuff_mask3(<16 x i32> %vec1, <16 x i32> %vec2) {
; CHECK-LABEL: test_16xi32_shuff_mask3:
; CHECK: # %bb.0:
-; CHECK-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[2,3,0,1],zmm1[4,5,2,3]
+; CHECK-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,0,1,2,3],zmm1[8,9,10,11,4,5,6,7]
; CHECK-NEXT: retq
%res = shufflevector <16 x i32> %vec1, <16 x i32> %vec2, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 24, i32 25, i32 26, i32 27, i32 20, i32 21, i32 22, i32 23>
ret <16 x i32> %res
@@ -1372,7 +1372,7 @@ define <16 x i32> @test_16xi32_zero_masked_shuff_mask3(<16 x i32> %vec1, <16 x i
define <16 x i32> @test_16xi32_shuff_mem_mask0(<16 x i32> %vec1, <16 x i32>* %vec2p) {
; CHECK-LABEL: test_16xi32_shuff_mem_mask0:
; CHECK: # %bb.0:
-; CHECK-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[4,5,2,3],mem[4,5,0,1]
+; CHECK-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[8,9,10,11,4,5,6,7],mem[8,9,10,11,0,1,2,3]
; CHECK-NEXT: retq
%vec2 = load <16 x i32>, <16 x i32>* %vec2p
%res = shufflevector <16 x i32> %vec1, <16 x i32> %vec2, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7, i32 24, i32 25, i32 26, i32 27, i32 16, i32 17, i32 18, i32 19>
@@ -1462,7 +1462,7 @@ define <16 x i32> @test_16xi32_zero_masked_shuff_mem_mask2(<16 x i32> %vec1, <16
define <16 x i32> @test_16xi32_shuff_mem_mask3(<16 x i32> %vec1, <16 x i32>* %vec2p) {
; CHECK-LABEL: test_16xi32_shuff_mem_mask3:
; CHECK: # %bb.0:
-; CHECK-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[2,3,2,3],mem[2,3,6,7]
+; CHECK-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[4,5,6,7,4,5,6,7],mem[4,5,6,7,12,13,14,15]
; CHECK-NEXT: retq
%vec2 = load <16 x i32>, <16 x i32>* %vec2p
%res = shufflevector <16 x i32> %vec1, <16 x i32> %vec2, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 20, i32 21, i32 22, i32 23, i32 28, i32 29, i32 30, i32 31>
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-512-v16.ll b/llvm/test/CodeGen/X86/vector-shuffle-512-v16.ll
index 8242019f1a9..2511c85e791 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-512-v16.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-512-v16.ll
@@ -547,7 +547,7 @@ define <16 x i32> @maskz_shuffle_v16i32_02_03_04_05_06_07_08_09_10_11_12_13_14_1
define <16 x float> @test_vshuff32x4_512(<16 x float> %x, <16 x float> %x1) nounwind {
; ALL-LABEL: test_vshuff32x4_512:
; ALL: # %bb.0:
-; ALL-NEXT: vshuff64x2 {{.*#+}} zmm0 = zmm0[0,1,2,3],zmm1[2,3,0,1]
+; ALL-NEXT: vshuff32x4 {{.*#+}} zmm0 = zmm0[0,1,2,3,4,5,6,7],zmm1[4,5,6,7,0,1,2,3]
; ALL-NEXT: retq
%res = shufflevector <16 x float> %x, <16 x float> %x1, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 20, i32 21, i32 22, i32 23, i32 16, i32 17, i32 18, i32 19>
ret <16 x float> %res
@@ -556,7 +556,7 @@ define <16 x float> @test_vshuff32x4_512(<16 x float> %x, <16 x float> %x1) noun
define <16 x i32> @test_vshufi32x4_512(<16 x i32> %x, <16 x i32> %x1) nounwind {
; ALL-LABEL: test_vshufi32x4_512:
; ALL: # %bb.0:
-; ALL-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,2,3],zmm1[2,3,0,1]
+; ALL-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[0,1,2,3,4,5,6,7],zmm1[4,5,6,7,0,1,2,3]
; ALL-NEXT: retq
%res = shufflevector <16 x i32> %x, <16 x i32> %x1, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 20, i32 21, i32 22, i32 23, i32 16, i32 17, i32 18, i32 19>
ret <16 x i32> %res
OpenPOWER on IntegriCloud