diff options
Diffstat (limited to 'llvm')
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 60 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/sad.ll | 24 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/vector-shuffle-512-v32.ll | 18 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/vector-shuffle-512-v64.ll | 44 |
4 files changed, 111 insertions, 35 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 0b45d1d1f59..a8c5f5ed002 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -7768,6 +7768,7 @@ static SDValue lowerVectorShuffleAsByteRotate(SDLoc DL, MVT VT, SDValue V1, /// [ 1, 2, -1, -1, -1, -1, zz, zz] static SDValue lowerVectorShuffleAsShift(SDLoc DL, MVT VT, SDValue V1, SDValue V2, ArrayRef<int> Mask, + const X86Subtarget &Subtarget, SelectionDAG &DAG) { SmallBitVector Zeroable = computeZeroableShuffleElements(Mask, V1, V2); @@ -7821,7 +7822,8 @@ static SDValue lowerVectorShuffleAsShift(SDLoc DL, MVT VT, SDValue V1, // their width within the elements of the larger integer vector. Test each // multiple to see if we can find a match with the moved element indices // and that the shifted in elements are all zeroable. - for (int Scale = 2; Scale * VT.getScalarSizeInBits() <= 128; Scale *= 2) + unsigned MaxWidth = (VT.is512BitVector() && !Subtarget.hasBWI() ? 64 : 128); + for (int Scale = 2; Scale * VT.getScalarSizeInBits() <= MaxWidth; Scale *= 2) for (int Shift = 1; Shift != Scale; ++Shift) for (bool Left : {true, false}) if (CheckZeros(Shift, Scale, Left)) @@ -8917,8 +8919,8 @@ static SDValue lowerV2I64VectorShuffle(SDValue Op, SDValue V1, SDValue V2, : V2Pack.getOperand(1))); // Try to use shift instructions. - if (SDValue Shift = - lowerVectorShuffleAsShift(DL, MVT::v2i64, V1, V2, Mask, DAG)) + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v2i64, V1, V2, Mask, + Subtarget, DAG)) return Shift; // When loading a scalar and then shuffling it into a vector we can often do @@ -9211,8 +9213,8 @@ static SDValue lowerV4I32VectorShuffle(SDValue Op, SDValue V1, SDValue V2, } // Try to use shift instructions. - if (SDValue Shift = - lowerVectorShuffleAsShift(DL, MVT::v4i32, V1, V2, Mask, DAG)) + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v4i32, V1, V2, Mask, + Subtarget, DAG)) return Shift; // There are special ways we can lower some single-element blends. @@ -9837,8 +9839,8 @@ static SDValue lowerV8I16VectorShuffle(SDValue Op, SDValue V1, SDValue V2, return Broadcast; // Try to use shift instructions. - if (SDValue Shift = - lowerVectorShuffleAsShift(DL, MVT::v8i16, V1, V1, Mask, DAG)) + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v8i16, V1, V1, Mask, + Subtarget, DAG)) return Shift; // Use dedicated unpack instructions for masks that match their pattern. @@ -9860,8 +9862,8 @@ static SDValue lowerV8I16VectorShuffle(SDValue Op, SDValue V1, SDValue V2, "shuffles."); // Try to use shift instructions. - if (SDValue Shift = - lowerVectorShuffleAsShift(DL, MVT::v8i16, V1, V2, Mask, DAG)) + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v8i16, V1, V2, Mask, + Subtarget, DAG)) return Shift; // See if we can use SSE4A Extraction / Insertion. @@ -10006,8 +10008,8 @@ static SDValue lowerV16I8VectorShuffle(SDValue Op, SDValue V1, SDValue V2, assert(Mask.size() == 16 && "Unexpected mask size for v16 shuffle!"); // Try to use shift instructions. - if (SDValue Shift = - lowerVectorShuffleAsShift(DL, MVT::v16i8, V1, V2, Mask, DAG)) + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v16i8, V1, V2, Mask, + Subtarget, DAG)) return Shift; // Try to use byte rotation instructions. @@ -11217,8 +11219,8 @@ static SDValue lowerV4I64VectorShuffle(SDValue Op, SDValue V1, SDValue V2, getV4X86ShuffleImm8ForMask(Mask, DL, DAG)); // Try to use shift instructions. - if (SDValue Shift = - lowerVectorShuffleAsShift(DL, MVT::v4i64, V1, V2, Mask, DAG)) + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v4i64, V1, V2, Mask, + Subtarget, DAG)) return Shift; // Use dedicated unpack instructions for masks that match their pattern. @@ -11385,8 +11387,8 @@ static SDValue lowerV8I32VectorShuffle(SDValue Op, SDValue V1, SDValue V2, } // Try to use shift instructions. - if (SDValue Shift = - lowerVectorShuffleAsShift(DL, MVT::v8i32, V1, V2, Mask, DAG)) + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v8i32, V1, V2, Mask, + Subtarget, DAG)) return Shift; // Try to use byte rotation instructions. @@ -11459,8 +11461,8 @@ static SDValue lowerV16I16VectorShuffle(SDValue Op, SDValue V1, SDValue V2, return V; // Try to use shift instructions. - if (SDValue Shift = - lowerVectorShuffleAsShift(DL, MVT::v16i16, V1, V2, Mask, DAG)) + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v16i16, V1, V2, Mask, + Subtarget, DAG)) return Shift; // Try to use byte rotation instructions. @@ -11542,8 +11544,8 @@ static SDValue lowerV32I8VectorShuffle(SDValue Op, SDValue V1, SDValue V2, return V; // Try to use shift instructions. - if (SDValue Shift = - lowerVectorShuffleAsShift(DL, MVT::v32i8, V1, V2, Mask, DAG)) + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v32i8, V1, V2, Mask, + Subtarget, DAG)) return Shift; // Try to use byte rotation instructions. @@ -11771,6 +11773,11 @@ static SDValue lowerV8I64VectorShuffle(SDValue Op, SDValue V1, SDValue V2, lowerV4X128VectorShuffle(DL, MVT::v8i64, Mask, V1, V2, DAG)) return Shuf128; + // Try to use shift instructions. + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v8i64, V1, V2, Mask, + Subtarget, DAG)) + return Shift; + if (SDValue Unpck = lowerVectorShuffleWithUNPCK(DL, MVT::v8i64, Mask, V1, V2, DAG)) return Unpck; @@ -11789,6 +11796,11 @@ static SDValue lowerV16I32VectorShuffle(SDValue Op, SDValue V1, SDValue V2, ArrayRef<int> Mask = SVOp->getMask(); assert(Mask.size() == 16 && "Unexpected mask size for v16 shuffle!"); + // Try to use shift instructions. + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v16i32, V1, V2, Mask, + Subtarget, DAG)) + return Shift; + if (SDValue Unpck = lowerVectorShuffleWithUNPCK(DL, MVT::v16i32, Mask, V1, V2, DAG)) return Unpck; @@ -11814,6 +11826,11 @@ static SDValue lowerV32I16VectorShuffle(SDValue Op, SDValue V1, SDValue V2, assert(Mask.size() == 32 && "Unexpected mask size for v32 shuffle!"); assert(Subtarget.hasBWI() && "We can only lower v32i16 with AVX-512-BWI!"); + // Try to use shift instructions. + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v32i16, V1, V2, Mask, + Subtarget, DAG)) + return Shift; + // Try to use byte rotation instructions. if (SDValue Rotate = lowerVectorShuffleAsByteRotate( DL, MVT::v32i16, V1, V2, Mask, Subtarget, DAG)) @@ -11834,6 +11851,11 @@ static SDValue lowerV64I8VectorShuffle(SDValue Op, SDValue V1, SDValue V2, assert(Mask.size() == 64 && "Unexpected mask size for v64 shuffle!"); assert(Subtarget.hasBWI() && "We can only lower v64i8 with AVX-512-BWI!"); + // Try to use shift instructions. + if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v64i8, V1, V2, Mask, + Subtarget, DAG)) + return Shift; + // Try to use byte rotation instructions. if (SDValue Rotate = lowerVectorShuffleAsByteRotate( DL, MVT::v64i8, V1, V2, Mask, Subtarget, DAG)) diff --git a/llvm/test/CodeGen/X86/sad.ll b/llvm/test/CodeGen/X86/sad.ll index 4e764a7d8e7..8619badf666 100644 --- a/llvm/test/CodeGen/X86/sad.ll +++ b/llvm/test/CodeGen/X86/sad.ll @@ -78,9 +78,7 @@ define i32 @sad_16i8() nounwind { ; AVX512F-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512F-NEXT: vpunpckhqdq {{.*#+}} zmm1 = zmm0[1,1,3,3,5,5,7,7] ; AVX512F-NEXT: vpaddd %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: movl $1, %eax -; AVX512F-NEXT: vmovd %eax, %xmm1 -; AVX512F-NEXT: vpermd %zmm0, %zmm1, %zmm1 +; AVX512F-NEXT: vpsrlq $32, %zmm0, %zmm1 ; AVX512F-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512F-NEXT: vmovd %xmm0, %eax ; AVX512F-NEXT: retq @@ -103,9 +101,9 @@ define i32 @sad_16i8() nounwind { ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512BW-NEXT: vshufi64x2 {{.*#+}} zmm1 = zmm0[2,3,0,1,0,1,0,1] ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpunpckhqdq {{.*#+}} zmm1 = zmm0[1,1,3,3,5,5,7,7] +; AVX512BW-NEXT: vpsrldq $8, %zmm0, %zmm1 ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpalignr {{.*#+}} zmm1 = zmm0[4,5,6,7,8,9,10,11,12,13,14,15,0,1,2,3,20,21,22,23,24,25,26,27,28,29,30,31,16,17,18,19,36,37,38,39,40,41,42,43,44,45,46,47,32,33,34,35,52,53,54,55,56,57,58,59,60,61,62,63,48,49,50,51] +; AVX512BW-NEXT: vpsrlq $32, %zmm0, %zmm1 ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512BW-NEXT: vmovd %xmm0, %eax ; AVX512BW-NEXT: retq @@ -327,9 +325,7 @@ define i32 @sad_32i8() nounwind { ; AVX512F-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512F-NEXT: vpunpckhqdq {{.*#+}} zmm1 = zmm0[1,1,3,3,5,5,7,7] ; AVX512F-NEXT: vpaddd %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: movl $1, %eax -; AVX512F-NEXT: vmovd %eax, %xmm1 -; AVX512F-NEXT: vpermd %zmm0, %zmm1, %zmm1 +; AVX512F-NEXT: vpsrlq $32, %zmm0, %zmm1 ; AVX512F-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512F-NEXT: vmovd %xmm0, %eax ; AVX512F-NEXT: retq @@ -354,9 +350,9 @@ define i32 @sad_32i8() nounwind { ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512BW-NEXT: vshufi64x2 {{.*#+}} zmm1 = zmm0[2,3,0,1,0,1,0,1] ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpunpckhqdq {{.*#+}} zmm1 = zmm0[1,1,3,3,5,5,7,7] +; AVX512BW-NEXT: vpsrldq $8, %zmm0, %zmm1 ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpalignr {{.*#+}} zmm1 = zmm0[4,5,6,7,8,9,10,11,12,13,14,15,0,1,2,3,20,21,22,23,24,25,26,27,28,29,30,31,16,17,18,19,36,37,38,39,40,41,42,43,44,45,46,47,32,33,34,35,52,53,54,55,56,57,58,59,60,61,62,63,48,49,50,51] +; AVX512BW-NEXT: vpsrlq $32, %zmm0, %zmm1 ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512BW-NEXT: vmovd %xmm0, %eax ; AVX512BW-NEXT: retq @@ -811,9 +807,7 @@ define i32 @sad_avx64i8() nounwind { ; AVX512F-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512F-NEXT: vpunpckhqdq {{.*#+}} zmm1 = zmm0[1,1,3,3,5,5,7,7] ; AVX512F-NEXT: vpaddd %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: movl $1, %eax -; AVX512F-NEXT: vmovd %eax, %xmm1 -; AVX512F-NEXT: vpermd %zmm0, %zmm1, %zmm1 +; AVX512F-NEXT: vpsrlq $32, %zmm0, %zmm1 ; AVX512F-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512F-NEXT: vmovd %xmm0, %eax ; AVX512F-NEXT: retq @@ -839,9 +833,9 @@ define i32 @sad_avx64i8() nounwind { ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512BW-NEXT: vshufi64x2 {{.*#+}} zmm1 = zmm0[2,3,0,1,0,1,0,1] ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpunpckhqdq {{.*#+}} zmm1 = zmm0[1,1,3,3,5,5,7,7] +; AVX512BW-NEXT: vpsrldq $8, %zmm0, %zmm1 ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpalignr {{.*#+}} zmm1 = zmm0[4,5,6,7,8,9,10,11,12,13,14,15,0,1,2,3,20,21,22,23,24,25,26,27,28,29,30,31,16,17,18,19,36,37,38,39,40,41,42,43,44,45,46,47,32,33,34,35,52,53,54,55,56,57,58,59,60,61,62,63,48,49,50,51] +; AVX512BW-NEXT: vpsrlq $32, %zmm0, %zmm1 ; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm0 ; AVX512BW-NEXT: vmovd %xmm0, %eax ; AVX512BW-NEXT: retq diff --git a/llvm/test/CodeGen/X86/vector-shuffle-512-v32.ll b/llvm/test/CodeGen/X86/vector-shuffle-512-v32.ll index a3cdaf09b64..f2279ef7e39 100644 --- a/llvm/test/CodeGen/X86/vector-shuffle-512-v32.ll +++ b/llvm/test/CodeGen/X86/vector-shuffle-512-v32.ll @@ -61,3 +61,21 @@ define <32 x i16> @shuffle_v16i32_4_36_5_37_6_38_7_39_12_44_13_45_u_u_u_u(<32 x %c = shufflevector <32 x i16> %a, <32 x i16> %b, <32 x i32> <i32 4, i32 36, i32 5, i32 37, i32 6, i32 38, i32 7, i32 39, i32 12, i32 44, i32 13, i32 45, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef> ret <32 x i16> %c } + +define <32 x i16> @shuffle_v32i16_1_z_3_z_5_z_7_z_9_z_11_z_13_z_15_z_17_z_19_z_21_z_23_z_25_z_27_z_29_z_31_z(<32 x i16> %a, <32 x i16> %b) { +; ALL-LABEL: shuffle_v32i16_1_z_3_z_5_z_7_z_9_z_11_z_13_z_15_z_17_z_19_z_21_z_23_z_25_z_27_z_29_z_31_z: +; ALL: # BB#0: +; ALL-NEXT: vpsrld $16, %zmm0, %zmm0 +; ALL-NEXT: retq + %c = shufflevector <32 x i16> %a, <32 x i16> zeroinitializer, <32 x i32> <i32 1, i32 34, i32 3, i32 34, i32 5, i32 34, i32 7, i32 34, i32 9, i32 34, i32 11, i32 34, i32 13, i32 34, i32 15, i32 34, i32 17, i32 34, i32 19, i32 34, i32 21, i32 34, i32 23, i32 34, i32 25, i32 34, i32 27, i32 34, i32 29, i32 34, i32 31, i32 34> + ret <32 x i16> %c +} + +define <32 x i16> @shuffle_v32i16_z_0_z_2_z_4_z_6_z_8_z_10_z_12_z_14_z_16_z_18_z_20_z_22_z_24_z_26_z_28_z_30(<32 x i16> %a, <32 x i16> %b) { +; ALL-LABEL: shuffle_v32i16_z_0_z_2_z_4_z_6_z_8_z_10_z_12_z_14_z_16_z_18_z_20_z_22_z_24_z_26_z_28_z_30: +; ALL: # BB#0: +; ALL-NEXT: vpslld $16, %zmm0, %zmm0 +; ALL-NEXT: retq + %c = shufflevector <32 x i16> %a, <32 x i16> zeroinitializer, <32 x i32> <i32 34, i32 0, i32 34, i32 2, i32 34, i32 4, i32 34, i32 6, i32 34, i32 8, i32 34, i32 10, i32 34, i32 12, i32 34, i32 14, i32 34, i32 16, i32 34, i32 18, i32 34, i32 20, i32 34, i32 22, i32 34, i32 24, i32 34, i32 26, i32 34, i32 28, i32 34, i32 30> + ret <32 x i16> %c +} diff --git a/llvm/test/CodeGen/X86/vector-shuffle-512-v64.ll b/llvm/test/CodeGen/X86/vector-shuffle-512-v64.ll index 9991bab046b..7160713b7e3 100644 --- a/llvm/test/CodeGen/X86/vector-shuffle-512-v64.ll +++ b/llvm/test/CodeGen/X86/vector-shuffle-512-v64.ll @@ -13,7 +13,7 @@ define <64 x i8> @shuffle_v64i8_02_03_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_ ; ; AVX512BW-LABEL: shuffle_v64i8_02_03_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u: ; AVX512BW: # BB#0: -; AVX512BW-NEXT: vpalignr {{.*#+}} zmm0 = zmm0[2,3,4,5,6,7,8,9,10,11,12,13,14,15,0,1,18,19,20,21,22,23,24,25,26,27,28,29,30,31,16,17,34,35,36,37,38,39,40,41,42,43,44,45,46,47,32,33,50,51,52,53,54,55,56,57,58,59,60,61,62,63,48,49] +; AVX512BW-NEXT: vpsrld $16, %zmm0, %zmm0 ; AVX512BW-NEXT: retq ; ; AVX512DQ-LABEL: shuffle_v64i8_02_03_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u: @@ -24,6 +24,48 @@ define <64 x i8> @shuffle_v64i8_02_03_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_u_ ret <64 x i8> %b } +define <64 x i8> @shuffle_v64i8_zz_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14_zz_16_17_18_19_20_21_22_23_24_25_26_27_28_29_30_zz_32_33_34_35_36_37_38_39_40_41_42_43_44_45_46_zz_48_49_50_51_52_53_54_55_56_57_58_59_60_61_62(<64 x i8> %a, <64 x i8> %b) { +; AVX512F-LABEL: shuffle_v64i8_zz_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14_zz_16_17_18_19_20_21_22_23_24_25_26_27_28_29_30_zz_32_33_34_35_36_37_38_39_40_41_42_43_44_45_46_zz_48_49_50_51_52_53_54_55_56_57_58_59_60_61_62: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpslldq {{.*#+}} ymm0 = zero,ymm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14],zero,ymm0[16,17,18,19,20,21,22,23,24,25,26,27,28,29,30] +; AVX512F-NEXT: vpslldq {{.*#+}} ymm1 = zero,ymm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14],zero,ymm1[16,17,18,19,20,21,22,23,24,25,26,27,28,29,30] +; AVX512F-NEXT: retq +; +; AVX512BW-LABEL: shuffle_v64i8_zz_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14_zz_16_17_18_19_20_21_22_23_24_25_26_27_28_29_30_zz_32_33_34_35_36_37_38_39_40_41_42_43_44_45_46_zz_48_49_50_51_52_53_54_55_56_57_58_59_60_61_62: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpslldq $1, %zmm0, %zmm0 +; AVX512BW-NEXT: retq +; +; AVX512DQ-LABEL: shuffle_v64i8_zz_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14_zz_16_17_18_19_20_21_22_23_24_25_26_27_28_29_30_zz_32_33_34_35_36_37_38_39_40_41_42_43_44_45_46_zz_48_49_50_51_52_53_54_55_56_57_58_59_60_61_62: +; AVX512DQ: # BB#0: +; AVX512DQ-NEXT: vpslldq {{.*#+}} ymm0 = zero,ymm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14],zero,ymm0[16,17,18,19,20,21,22,23,24,25,26,27,28,29,30] +; AVX512DQ-NEXT: vpslldq {{.*#+}} ymm1 = zero,ymm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14],zero,ymm1[16,17,18,19,20,21,22,23,24,25,26,27,28,29,30] +; AVX512DQ-NEXT: retq + %shuffle = shufflevector <64 x i8> %a, <64 x i8> zeroinitializer, <64 x i32> <i32 79, 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 95, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 111, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 127, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62> + ret <64 x i8> %shuffle +} + +define <64 x i8> @shuffle_v64i8_02_03_04_05_06_07_08_09_10_11_12_13_14_15_zz_zz_18_19_20_21_22_23_24_25_26_27_28_29_30_31_zz_zz_34_35_36_37_38_39_40_41_42_43_44_45_46_47_zz_zz_50_51_52_53_54_55_56_57_58_59_60_61_62_63_zz_zz(<64 x i8> %a, <64 x i8> %b) { +; AVX512F-LABEL: shuffle_v64i8_02_03_04_05_06_07_08_09_10_11_12_13_14_15_zz_zz_18_19_20_21_22_23_24_25_26_27_28_29_30_31_zz_zz_34_35_36_37_38_39_40_41_42_43_44_45_46_47_zz_zz_50_51_52_53_54_55_56_57_58_59_60_61_62_63_zz_zz: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpsrldq {{.*#+}} ymm0 = ymm0[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero,ymm0[18,19,20,21,22,23,24,25,26,27,28,29,30,31],zero,zero +; AVX512F-NEXT: vpsrldq {{.*#+}} ymm1 = ymm1[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero,ymm1[18,19,20,21,22,23,24,25,26,27,28,29,30,31],zero,zero +; AVX512F-NEXT: retq +; +; AVX512BW-LABEL: shuffle_v64i8_02_03_04_05_06_07_08_09_10_11_12_13_14_15_zz_zz_18_19_20_21_22_23_24_25_26_27_28_29_30_31_zz_zz_34_35_36_37_38_39_40_41_42_43_44_45_46_47_zz_zz_50_51_52_53_54_55_56_57_58_59_60_61_62_63_zz_zz: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpsrldq $2, %zmm0, %zmm0 +; AVX512BW-NEXT: retq +; +; AVX512DQ-LABEL: shuffle_v64i8_02_03_04_05_06_07_08_09_10_11_12_13_14_15_zz_zz_18_19_20_21_22_23_24_25_26_27_28_29_30_31_zz_zz_34_35_36_37_38_39_40_41_42_43_44_45_46_47_zz_zz_50_51_52_53_54_55_56_57_58_59_60_61_62_63_zz_zz: +; AVX512DQ: # BB#0: +; AVX512DQ-NEXT: vpsrldq {{.*#+}} ymm0 = ymm0[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero,ymm0[18,19,20,21,22,23,24,25,26,27,28,29,30,31],zero,zero +; AVX512DQ-NEXT: vpsrldq {{.*#+}} ymm1 = ymm1[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero,ymm1[18,19,20,21,22,23,24,25,26,27,28,29,30,31],zero,zero +; AVX512DQ-NEXT: retq + %shuffle = shufflevector <64 x i8> %a, <64 x i8> zeroinitializer, <64 x i32> <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, i32 64, i32 64, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 64, i32 64, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 64, i32 64, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 64, i32 64> + ret <64 x i8> %shuffle +} + define <64 x i8> @shuffle_v64i8_79_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14_95_16_17_18_19_20_21_22_23_24_25_26_27_28_29_30_111_32_33_34_35_36_37_38_39_40_41_42_43_44_45_46_127_48_49_50_51_52_53_54_55_56_57_58_59_60_61_62(<64 x i8> %a, <64 x i8> %b) { ; AVX512F-LABEL: shuffle_v64i8_79_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14_95_16_17_18_19_20_21_22_23_24_25_26_27_28_29_30_111_32_33_34_35_36_37_38_39_40_41_42_43_44_45_46_127_48_49_50_51_52_53_54_55_56_57_58_59_60_61_62: ; AVX512F: # BB#0: |
