[X86][AVX2] Broadcast subvectors

[RKSimon]

Summary

AVX2 can only broadcast from the zero'th element of a vector, but if the broadcastable element is the zero'th element of a 128-bit subvector its advantageous to extract the subvector and broadcast from that and often avoids the loading of shuffle mask data that would be needed for VPERMPS/VPERMD. The only exception being when the source type is 4f64 or 4i64 which can directly use the immediate shuffle VPERMPD/VPERMQ directly.

[RKSimon]

ping

[delena]

LGTM

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

if (V.getOpcode() == ISD::BUILD_VECTOR ||
     EVT SVT = BroadcastVT.getScalarType();
     unsigned Offset = BroadcastIdx * SVT.getStoreSize();
     SDValue NewAddr = DAG.getNode(
         ISD::ADD, DL, AddrVT, BaseAddr,
         DAG.getConstant(Offset, DL, AddrVT));
     V = DAG.getLoad(SVT, DL, Ld->getChain(), NewAddr,
                     DAG.getMachineFunction().getMachineMemOperand(
                         Ld->getMemOperand(), Offset, SVT.getStoreSize()));
-  } else if (BroadcastIdx != 0 || !Subtarget->hasAVX2()) {
-    // We can't broadcast from a vector register without AVX2, and we can only
-    // broadcast from the zero-element of a vector register.
+  } else if (!Subtarget->hasAVX2()) {
+    // We can't broadcast from a vector register without AVX2.
     return SDValue();
+  } else if (BroadcastIdx != 0) {
+    // We can only broadcast from the zero-element of a vector register,
+    // but it can be advantageous to broadcast from the zero-element of a
+    // subvector.
+    if (!VT.is256BitVector() && !VT.is512BitVector())
+      return SDValue();
+
+    // VPERMQ/VPERMPD can perform the cross-lane shuffle directly.
+    if (VT == MVT::v4f64 || VT == MVT::v4i64)
+      return SDValue();
+
+    // Only broadcast the zero-element of a 128-bit subvector.
+    unsigned EltSize = VT.getScalarSizeInBits();
+    if (((BroadcastIdx * EltSize) % 128) != 0)
+      return SDValue();
+
+    MVT ExtVT = MVT::getVectorVT(VT.getScalarType(), 128 / EltSize);
+    V = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, ExtVT, V,
+                    DAG.getIntPtrConstant(BroadcastIdx, DL));
   }
 
   V = DAG.getNode(X86ISD::VBROADCAST, DL, BroadcastVT, V);
   return DAG.getBitcast(VT, V);
 }
 
 // Check for whether we can use INSERTPS to perform the shuffle. We only use
 // INSERTPS when the V1 elements are already in the correct locations

llvm/trunk/test/CodeGen/X86/vector-shuffle-256-v16.ll

 ; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,1,2,1,4,5,6,7]
 ; AVX1-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,7,7]
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,6],xmm1[7]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v16i16_uu_uu_uu_01_uu_05_07_25_uu_uu_uu_09_uu_13_15_25:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpbroadcastd {{.*}}(%rip), %ymm2
-; AVX2-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %ymm1
 ; AVX2-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm0[0,1,2,1,4,5,6,7,8,9,10,9,12,13,14,15]
 ; AVX2-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,7,7,8,9,10,11,12,13,15,15]
 ; AVX2-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7],ymm0[8,9,10,11,12,13,14],ymm1[15]
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 undef, i32 undef, i32 undef, i32 1, i32 undef, i32 5, i32 7, i32 25, i32 undef, i32 undef, i32 undef, i32 9, i32 undef, i32 13, i32 15, i32 25>
   ret <16 x i16> %shuffle
 }
 

 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v16i16_8_8_8_8_8_8_8_8_8_8_8_8_8_8_8_8:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX2-NEXT:    vpbroadcastw %xmm0, %xmm0
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpbroadcastw %xmm0, %ymm0
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
   ret <16 x i16> %shuffle
 }
 
 define <16 x i16> @shuffle_v16i16_4_20_5_21_6_22_7_23_u_u_u_u_u_u_u_u(<16 x i16> %a, <16 x i16> %b) {
 ; ALL-LABEL: shuffle_v16i16_4_20_5_21_6_22_7_23_u_u_u_u_u_u_u_u:
 ; ALL:       # BB#0:

llvm/trunk/test/CodeGen/X86/vector-shuffle-256-v32.ll

 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16_16:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX2-NEXT:    vpbroadcastb %xmm0, %xmm0
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpbroadcastb %xmm0, %ymm0
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <32 x i8> %a, <32 x i8> %b, <32 x i32> <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
   ret <32 x i8> %shuffle
 }
 
 define <32 x i8> @shuffle_v32i8_15_15_15_15_15_15_15_15_32_32_32_32_32_32_32_32_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu(<32 x i8> %a, <32 x i8> %b) {
 ; AVX1-LABEL: shuffle_v32i8_15_15_15_15_15_15_15_15_32_32_32_32_32_32_32_32_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu_uu:
 ; AVX1:       # BB#0:

llvm/trunk/test/CodeGen/X86/vector-shuffle-256-v8.ll

 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v8f32_44444444:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vbroadcastss {{.*}}(%rip), %ymm1
-; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX2-NEXT:    vbroadcastss %xmm0, %ymm0
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4>
   ret <8 x float> %shuffle
 }
 
 define <8 x float> @shuffle_v8f32_1188uuuu(<8 x float> %a, <8 x float> %b) {
 ; ALL-LABEL: shuffle_v8f32_1188uuuu:
 ; ALL:       # BB#0:

 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v8i32_44444444:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpbroadcastd {{.*}}(%rip), %ymm1
-; AVX2-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX2-NEXT:    vbroadcastss %xmm0, %ymm0
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4>
   ret <8 x i32> %shuffle
 }
 
 define <8 x i32> @shuffle_v8i32_5555uuuu(<8 x i32> %a, <8 x i32> %b) {
 ; AVX1-LABEL: shuffle_v8i32_5555uuuu:
 ; AVX1:       # BB#0:

llvm/trunk/test/CodeGen/X86/vector-shuffle-512-v16.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx512f | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512F
 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW
 
 target triple = "x86_64-unknown-unknown"
 
+define <16 x float> @shuffle_v16f32_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00(<16 x float> %a, <16 x float> %b) {
+; ALL-LABEL: shuffle_v16f32_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; ALL:       # BB#0:
+; ALL-NEXT:    vbroadcastss %xmm0, %zmm0
+; ALL-NEXT:    retq
+  %shuffle = shufflevector <16 x float> %a, <16 x float> %b, <16 x i32><i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
+  ret <16 x float> %shuffle
+}
+
+define <16 x float> @shuffle_v16f32_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08(<16 x float> %a, <16 x float> %b) {
+; ALL-LABEL: shuffle_v16f32_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08:
+; ALL:       # BB#0:
+; ALL-NEXT:    vextractf32x4 $2, %zmm0, %xmm0
+; ALL-NEXT:    vbroadcastss %xmm0, %zmm0
+; ALL-NEXT:    retq
+  %shuffle = shufflevector <16 x float> %a, <16 x float> %b, <16 x i32><i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
+  ret <16 x float> %shuffle
+}
+
 define <16 x float> @shuffle_v16f32_00_10_01_11_04_14_05_15_08_18_09_19_0c_1c_0d_1d(<16 x float> %a, <16 x float> %b) {
 ; ALL-LABEL: shuffle_v16f32_00_10_01_11_04_14_05_15_08_18_09_19_0c_1c_0d_1d:
 ; ALL:       # BB#0:
 ; ALL-NEXT:    vunpcklps {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
 ; ALL-NEXT:    retq
   %shuffle = shufflevector <16 x float> %a, <16 x float> %b, <16 x i32><i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
   ret <16 x float> %shuffle
 }

 ; ALL:       # BB#0:
 ; ALL-NEXT:    vpxord %zmm0, %zmm0, %zmm0
 ; ALL-NEXT:    vunpckhps {{.*#+}} zmm0 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
 ; ALL-NEXT:    retq
   %shuffle = shufflevector <16 x float> zeroinitializer, <16 x float> %b, <16 x i32><i32 0, i32 18, i32 0, i32 19, i32 4, i32 22, i32 4, i32 23, i32 6, i32 26, i32 6, i32 27, i32 8, i32 30, i32 8, i32 31>
   ret <16 x float> %shuffle
 }
 
+define <16 x i32> @shuffle_v16i32_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00(<16 x i32> %a, <16 x i32> %b) {
+; ALL-LABEL: shuffle_v16i32_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; ALL:       # BB#0:
+; ALL-NEXT:    vpbroadcastd %xmm0, %zmm0
+; ALL-NEXT:    retq
+  %shuffle = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32><i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
+  ret <16 x i32> %shuffle
+}
+
+define <16 x i32> @shuffle_v16i32_04_04_04_04_04_04_04_04_04_04_04_04_04_04_04_04(<16 x i32> %a, <16 x i32> %b) {
+; ALL-LABEL: shuffle_v16i32_04_04_04_04_04_04_04_04_04_04_04_04_04_04_04_04:
+; ALL:       # BB#0:
+; ALL-NEXT:    vextracti32x4 $1, %zmm0, %xmm0
+; ALL-NEXT:    vpbroadcastd %xmm0, %zmm0
+; ALL-NEXT:    retq
+  %shuffle = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32><i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4>
+  ret <16 x i32> %shuffle
+}
+
 define <16 x i32> @shuffle_v16i32_02_12_03_13_06_16_07_17_0a_1a_0b_1b_0e_1e_0f_1f(<16 x i32> %a, <16 x i32> %b) {
 ; ALL-LABEL: shuffle_v16i32_02_12_03_13_06_16_07_17_0a_1a_0b_1b_0e_1e_0f_1f:
 ; ALL:       # BB#0:
 ; ALL-NEXT:    vpunpckhdq {{.*#+}} zmm0 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
 ; ALL-NEXT:    retq
   %shuffle = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32><i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
   ret <16 x i32> %shuffle
 }

llvm/trunk/test/CodeGen/X86/vector-shuffle-512-v32.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx512f -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW
 
 target triple = "x86_64-unknown-unknown"
 
+define <32 x i16> @shuffle_v32i16_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00(<32 x i16> %a)  {
+; ALL-LABEL: shuffle_v32i16_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; ALL:       # BB#0:
+; ALL-NEXT:    vpbroadcastw %xmm0, %zmm0
+; ALL-NEXT:    retq
+  %c = shufflevector <32 x i16> %a, <32 x i16> undef, <32 x i32> zeroinitializer
+  ret <32 x i16> %c
+}
+
+define <32 x i16> @shuffle_v32i16_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08(<32 x i16> %a)  {
+; ALL-LABEL: shuffle_v32i16_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08_08:
+; ALL:       # BB#0:
+; ALL-NEXT:    vextracti32x4 $1, %zmm0, %xmm0
+; ALL-NEXT:    vpbroadcastw %xmm0, %zmm0
+; ALL-NEXT:    retq
+  %c = shufflevector <32 x i16> %a, <32 x i16> undef, <32 x i32> <i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
+  ret <32 x i16> %c
+}
+
 define <32 x i16> @shuffle_v32i16_02_05_u_u_07_u_0a_01_00_05_u_04_07_u_0a_01_02_05_u_u_07_u_0a_01_00_05_u_04_07_u_0a_1f(<32 x i16> %a)  {
 ; ALL-LABEL: shuffle_v32i16_02_05_u_u_07_u_0a_01_00_05_u_04_07_u_0a_01_02_05_u_u_07_u_0a_01_00_05_u_04_07_u_0a_1f:
 ; ALL:       # BB#0:
 ; ALL-NEXT:    vmovdqu16 {{.*#+}} zmm1 = <2,5,u,u,7,u,10,1,0,5,u,4,7,u,10,1,2,5,u,u,7,u,10,1,0,5,u,4,7,u,10,31>
 ; ALL-NEXT:    vpermw %zmm0, %zmm1, %zmm0
 ; ALL-NEXT:    retq
   %c = shufflevector <32 x i16> %a, <32 x i16> undef, <32 x i32> <i32 2, i32 5, i32 undef, i32 undef, i32 7, i32 undef, i32 10, i32 1,  i32 0, i32 5, i32 undef, i32 4, i32 7, i32 undef, i32 10, i32 1, i32 2, i32 5, i32 undef, i32 undef, i32 7, i32 undef, i32 10, i32 1,  i32 0, i32 5, i32 undef, i32 4, i32 7, i32 undef, i32 10, i32 31>
   ret <32 x i16> %c

llvm/trunk/test/CodeGen/X86/vector-shuffle-512-v8.ll

 ; AVX512F-32-LABEL: shuffle_v8f64_00000000:
 ; AVX512F-32:       # BB#0:
 ; AVX512F-32-NEXT:    vbroadcastsd %xmm0, %zmm0
 ; AVX512F-32-NEXT:    retl
   %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   ret <8 x double> %shuffle
 }
 
+define <8 x double> @shuffle_v8f64_22222222(<8 x double> %a, <8 x double> %b) {
+; AVX512F-LABEL: shuffle_v8f64_22222222:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vextractf32x4 $1, %zmm0, %xmm0
+; AVX512F-NEXT:    vbroadcastsd %xmm0, %zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512F-32-LABEL: shuffle_v8f64_22222222:
+; AVX512F-32:       # BB#0:
+; AVX512F-32-NEXT:    vextractf32x4 $1, %zmm0, %xmm0
+; AVX512F-32-NEXT:    vbroadcastsd %xmm0, %zmm0
+; AVX512F-32-NEXT:    retl
+  %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
+  ret <8 x double> %shuffle
+}
+
+define <8 x double> @shuffle_v8f64_44444444(<8 x double> %a, <8 x double> %b) {
+; AVX512F-LABEL: shuffle_v8f64_44444444:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vextractf32x4 $2, %zmm0, %xmm0
+; AVX512F-NEXT:    vbroadcastsd %xmm0, %zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512F-32-LABEL: shuffle_v8f64_44444444:
+; AVX512F-32:       # BB#0:
+; AVX512F-32-NEXT:    vextractf32x4 $2, %zmm0, %xmm0
+; AVX512F-32-NEXT:    vbroadcastsd %xmm0, %zmm0
+; AVX512F-32-NEXT:    retl
+  %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4>
+  ret <8 x double> %shuffle
+}
+
 define <8 x double> @shuffle_v8f64_00000010(<8 x double> %a, <8 x double> %b) {
 ; AVX512F-LABEL: shuffle_v8f64_00000010:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [0,0,0,0,0,0,1,0]
 ; AVX512F-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512F-32-LABEL: shuffle_v8f64_00000010:

 ; AVX512F-32-LABEL: shuffle_v8i64_00000000:
 ; AVX512F-32:       # BB#0:
 ; AVX512F-32-NEXT:    vpbroadcastq %xmm0, %zmm0
 ; AVX512F-32-NEXT:    retl
   %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   ret <8 x i64> %shuffle
 }
 
+define <8 x i64> @shuffle_v8i64_44444444(<8 x i64> %a, <8 x i64> %b) {
+; AVX512F-LABEL: shuffle_v8i64_44444444:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vextracti32x4 $2, %zmm0, %xmm0
+; AVX512F-NEXT:    vpbroadcastq %xmm0, %zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512F-32-LABEL: shuffle_v8i64_44444444:
+; AVX512F-32:       # BB#0:
+; AVX512F-32-NEXT:    vextracti32x4 $2, %zmm0, %xmm0
+; AVX512F-32-NEXT:    vpbroadcastq %xmm0, %zmm0
+; AVX512F-32-NEXT:    retl
+  %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4>
+  ret <8 x i64> %shuffle
+}
+
+define <8 x i64> @shuffle_v8i64_66666666(<8 x i64> %a, <8 x i64> %b) {
+; AVX512F-LABEL: shuffle_v8i64_66666666:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
+; AVX512F-NEXT:    vpbroadcastq %xmm0, %zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512F-32-LABEL: shuffle_v8i64_66666666:
+; AVX512F-32:       # BB#0:
+; AVX512F-32-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
+; AVX512F-32-NEXT:    vpbroadcastq %xmm0, %zmm0
+; AVX512F-32-NEXT:    retl
+  %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6>
+  ret <8 x i64> %shuffle
+}
+
 define <8 x i64> @shuffle_v8i64_00000010(<8 x i64> %a, <8 x i64> %b) {
 ;
 ; AVX512F-LABEL: shuffle_v8i64_00000010:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [0,0,0,0,0,0,1,0]
 ; AVX512F-NEXT:    vpermq %zmm0, %zmm1, %zmm0
 ; AVX512F-NEXT:    retq
 ;

 ; AVX512F-NEXT:    vpsllq $63, %zmm2, %zmm2
 ; AVX512F-NEXT:    vptestmq %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vshuff64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5],zmm1[2,3,0,1]
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512F-32-LABEL: test_vshuff64x2_512_maskz:
 ; AVX512F-32:       # BB#0:
 ; AVX512F-32-NEXT:    vpmovsxwq %xmm2, %zmm2
-; AVX512F-32-NEXT:    vpsllvq .LCPI122_0, %zmm2, %zmm2
+; AVX512F-32-NEXT:    vpsllvq .LCPI126_0, %zmm2, %zmm2
 ; AVX512F-32-NEXT:    vptestmq %zmm2, %zmm2, %k1
 ; AVX512F-32-NEXT:    vshuff64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5],zmm1[2,3,0,1]
 ; AVX512F-32-NEXT:    retl
   %y = shufflevector <8 x double> %x, <8 x double> %x1, <8 x i32> <i32 0, i32 1, i32 4, i32 5, i32 10, i32 11, i32 8, i32 9>
   %res = select <8 x i1> %mask, <8 x double> %y, <8 x double> zeroinitializer
   ret <8 x double> %res
 }
 
 define <8 x i64> @test_vshufi64x2_512_mask(<8 x i64> %x, <8 x i64> %x1, <8 x i1> %mask) nounwind {
 ; AVX512F-LABEL: test_vshufi64x2_512_mask:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vpmovsxwq %xmm2, %zmm2
 ; AVX512F-NEXT:    vpsllq $63, %zmm2, %zmm2
 ; AVX512F-NEXT:    vptestmq %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5],zmm1[2,3,0,1]
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512F-32-LABEL: test_vshufi64x2_512_mask:
 ; AVX512F-32:       # BB#0:
 ; AVX512F-32-NEXT:    vpmovsxwq %xmm2, %zmm2
-; AVX512F-32-NEXT:    vpsllvq .LCPI123_0, %zmm2, %zmm2
+; AVX512F-32-NEXT:    vpsllvq .LCPI127_0, %zmm2, %zmm2
 ; AVX512F-32-NEXT:    vptestmq %zmm2, %zmm2, %k1
 ; AVX512F-32-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5],zmm1[2,3,0,1]
 ; AVX512F-32-NEXT:    retl
   %y = shufflevector <8 x i64> %x, <8 x i64> %x1, <8 x i32> <i32 0, i32 1, i32 4, i32 5, i32 10, i32 11, i32 8, i32 9>
   %res = select <8 x i1> %mask, <8 x i64> %y, <8 x i64> %x
   ret <8 x i64> %res
 }
 

 ; AVX512F-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k1
 ; AVX512F-NEXT:    vshuff64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5],mem[2,3,0,1]
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512F-32-LABEL: test_vshuff64x2_512_mem_mask:
 ; AVX512F-32:       # BB#0:
 ; AVX512F-32-NEXT:    vpmovsxwq %xmm1, %zmm1
-; AVX512F-32-NEXT:    vpsllvq .LCPI125_0, %zmm1, %zmm1
+; AVX512F-32-NEXT:    vpsllvq .LCPI129_0, %zmm1, %zmm1
 ; AVX512F-32-NEXT:    vptestmq %zmm1, %zmm1, %k1
 ; AVX512F-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; AVX512F-32-NEXT:    vshuff64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5],mem[2,3,0,1]
 ; AVX512F-32-NEXT:    retl
   %x1 = load <8 x double>,<8 x double> *%ptr,align 1
   %y = shufflevector <8 x double> %x, <8 x double> %x1, <8 x i32> <i32 0, i32 1, i32 4, i32 5, i32 10, i32 11, i32 8, i32 9>
   %res = select <8 x i1> %mask, <8 x double> %y, <8 x double> %x
   ret <8 x double> %res
 }
 
 define <8 x double> @test_vshuff64x2_512_mem_maskz(<8 x double> %x, <8 x double> *%ptr, <8 x i1> %mask) nounwind {
 ; AVX512F-LABEL: test_vshuff64x2_512_mem_maskz:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vpmovsxwq %xmm1, %zmm1
 ; AVX512F-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k1
 ; AVX512F-NEXT:    vshuff64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5],mem[2,3,0,1]
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512F-32-LABEL: test_vshuff64x2_512_mem_maskz:
 ; AVX512F-32:       # BB#0:
 ; AVX512F-32-NEXT:    vpmovsxwq %xmm1, %zmm1
-; AVX512F-32-NEXT:    vpsllvq .LCPI126_0, %zmm1, %zmm1
+; AVX512F-32-NEXT:    vpsllvq .LCPI130_0, %zmm1, %zmm1
 ; AVX512F-32-NEXT:    vptestmq %zmm1, %zmm1, %k1
 ; AVX512F-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; AVX512F-32-NEXT:    vshuff64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5],mem[2,3,0,1]
 ; AVX512F-32-NEXT:    retl
   %x1 = load <8 x double>,<8 x double> *%ptr,align 1
   %y = shufflevector <8 x double> %x, <8 x double> %x1, <8 x i32> <i32 0, i32 1, i32 4, i32 5, i32 10, i32 11, i32 8, i32 9>
   %res = select <8 x i1> %mask, <8 x double> %y, <8 x double> zeroinitializer
   ret <8 x double> %res

llvm/trunk/test/CodeGen/X86/vector-shuffle-v1.ll

 }
 
 define <8 x i1> @shuf8i1_u_2_u_u_2_u_2_u(i8 %a) {
 ; AVX512F-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    kmovw %edi, %k1
 ; AVX512F-NEXT:    vpbroadcastq {{.*}}(%rip), %zmm0
 ; AVX512F-NEXT:    vmovdqu64 %zmm0, %zmm1 {%k1} {z}
-; AVX512F-NEXT:    vpbroadcastq {{.*}}(%rip), %zmm2
-; AVX512F-NEXT:    vpermq %zmm1, %zmm2, %zmm1
+; AVX512F-NEXT:    vextracti32x4 $1, %zmm1, %xmm1
+; AVX512F-NEXT:    vpbroadcastq %xmm1, %zmm1
 ; AVX512F-NEXT:    vpsllq $63, %zmm1, %zmm1
 ; AVX512F-NEXT:    vptestmq %zmm1, %zmm1, %k1
 ; AVX512F-NEXT:    vmovdqu64 %zmm0, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    vpmovqw %zmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; VL_BW_DQ-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
 ; VL_BW_DQ:       # BB#0:
 ; VL_BW_DQ-NEXT:    kmovb %edi, %k0
 ; VL_BW_DQ-NEXT:    vpmovm2q %k0, %zmm0
-; VL_BW_DQ-NEXT:    vpbroadcastq {{.*}}(%rip), %zmm1
-; VL_BW_DQ-NEXT:    vpermq %zmm0, %zmm1, %zmm0
+; VL_BW_DQ-NEXT:    vextracti64x2 $1, %zmm0, %xmm0
+; VL_BW_DQ-NEXT:    vpbroadcastq %xmm0, %zmm0
 ; VL_BW_DQ-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; VL_BW_DQ-NEXT:    vpmovq2m %zmm0, %k0
 ; VL_BW_DQ-NEXT:    vpmovm2w %k0, %xmm0
 ; VL_BW_DQ-NEXT:    retq
   %b = bitcast i8 %a to <8 x i1>
   %c = shufflevector < 8 x i1> %b, <8 x i1>undef, <8 x i32> <i32 undef, i32 2, i32 undef, i32 undef, i32 2, i32 undef, i32 2, i32 undef>
   ret <8 x i1> %c
 }