diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp --- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp +++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp @@ -25830,6 +25830,77 @@ } } +static SDValue GenerateFixedLengthSVETBL(SDValue Op, SDValue Op1, SDValue Op2, + ArrayRef ShuffleMask, EVT VT, + EVT ContainerVT, SelectionDAG &DAG) { + auto &Subtarget = DAG.getSubtarget(); + SDLoc DL(Op); + unsigned MinSVESize = Subtarget.getMinSVEVectorSizeInBits(); + unsigned MaxSVESize = Subtarget.getMaxSVEVectorSizeInBits(); + bool IsSingleOp = ShuffleVectorInst::isSingleSourceMask(ShuffleMask); + + // Ignore two operands if no SVE2 or all index numbers couldn't + // be represented. + if (!IsSingleOp && (!Subtarget.hasSVE2() || MinSVESize != MaxSVESize)) + return SDValue(); + + EVT VTOp1 = Op.getOperand(0).getValueType(); + unsigned BitsPerElt = VTOp1.getVectorElementType().getSizeInBits(); + unsigned IndexLen = MinSVESize / BitsPerElt; + unsigned ElementsPerVectorReg = VTOp1.getVectorNumElements(); + unsigned MaskSize = ShuffleMask.size(); + uint64_t MaxOffset = APInt(BitsPerElt, -1, false).getZExtValue(); + assert(ElementsPerVectorReg <= IndexLen && MaskSize <= IndexLen && + "Incorrectly legalised shuffle operation"); + + SmallVector TBLMask; + for (int Index : ShuffleMask) { + // Handling poison index value. + if (Index < 0) + Index = 0; + // If we refer to the second operand then we have to add elements + // number in hardware register minus number of elements in a type. + if ((unsigned)Index >= ElementsPerVectorReg) + Index += IndexLen - ElementsPerVectorReg; + // For 8-bit elements and 1024-bit SVE registers and MaxOffset equals + // to 255, this might point to the last element of in the second operand + // of the shufflevector, thus we are rejecting this transform. + if ((unsigned)Index >= MaxOffset) + return SDValue(); + TBLMask.push_back(DAG.getConstant(Index, DL, MVT::i64)); + } + + // Choosing an out-of-range index leads to the lane being zeroed vs zero + // value where it would perform first lane duplication for out of + // index elements. For i8 elements an out-of-range index could be a valid + // for 2048-bit vector register size. + for (unsigned i = 0; i < IndexLen - ElementsPerVectorReg; ++i) + TBLMask.push_back(DAG.getConstant((int)MaxOffset, DL, MVT::i64)); + + EVT MaskEltType = EVT::getIntegerVT(*DAG.getContext(), BitsPerElt); + EVT MaskType = EVT::getVectorVT(*DAG.getContext(), MaskEltType, IndexLen); + EVT MaskContainerVT = getContainerForFixedLengthVector(DAG, MaskType); + SDValue VecMask = + DAG.getBuildVector(MaskType, DL, ArrayRef(TBLMask.data(), IndexLen)); + SDValue SVEMask = convertToScalableVector(DAG, MaskContainerVT, VecMask); + + SDValue Shuffle; + if (IsSingleOp) + Shuffle = + DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, ContainerVT, + DAG.getConstant(Intrinsic::aarch64_sve_tbl, DL, MVT::i32), + Op1, SVEMask); + else if (Subtarget.hasSVE2()) + Shuffle = + DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, ContainerVT, + DAG.getConstant(Intrinsic::aarch64_sve_tbl2, DL, MVT::i32), + Op1, Op2, SVEMask); + else + llvm_unreachable("Cannot lower shuffle without SVE2 TBL"); + Shuffle = convertFromScalableVector(DAG, VT, Shuffle); + return DAG.getNode(ISD::BITCAST, DL, Op.getValueType(), Shuffle); +} + SDValue AArch64TargetLowering::LowerFixedLengthVECTOR_SHUFFLEToSVE( SDValue Op, SelectionDAG &DAG) const { EVT VT = Op.getValueType(); @@ -25975,6 +26046,11 @@ } } + // Avoid producing TBL instruction if we don't know SVE register minimal size. + if (MinSVESize) + return GenerateFixedLengthSVETBL(Op, Op1, Op2, ShuffleMask, VT, ContainerVT, + DAG); + return SDValue(); } diff --git a/llvm/test/CodeGen/AArch64/sve-fixed-length-permute-rev.ll b/llvm/test/CodeGen/AArch64/sve-fixed-length-permute-rev.ll --- a/llvm/test/CodeGen/AArch64/sve-fixed-length-permute-rev.ll +++ b/llvm/test/CodeGen/AArch64/sve-fixed-length-permute-rev.ll @@ -194,28 +194,13 @@ define void @test_rev_elts_fail(ptr %a) #1 { ; CHECK-LABEL: test_rev_elts_fail: ; CHECK: // %bb.0: -; CHECK-NEXT: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill -; CHECK-NEXT: sub x9, sp, #48 -; CHECK-NEXT: mov x29, sp -; CHECK-NEXT: and sp, x9, #0xffffffffffffffe0 -; CHECK-NEXT: .cfi_def_cfa w29, 16 -; CHECK-NEXT: .cfi_offset w30, -8 -; CHECK-NEXT: .cfi_offset w29, -16 ; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: adrp x8, .LCPI11_0 +; CHECK-NEXT: add x8, x8, :lo12:.LCPI11_0 ; CHECK-NEXT: ld1d { z0.d }, p0/z, [x0] -; CHECK-NEXT: mov z1.d, z0.d[2] -; CHECK-NEXT: mov z2.d, z0.d[3] -; CHECK-NEXT: mov x9, v0.d[1] -; CHECK-NEXT: fmov x8, d1 -; CHECK-NEXT: fmov x10, d2 -; CHECK-NEXT: stp x10, x8, [sp, #16] -; CHECK-NEXT: fmov x8, d0 -; CHECK-NEXT: stp x9, x8, [sp] -; CHECK-NEXT: mov x8, sp -; CHECK-NEXT: ld1d { z0.d }, p0/z, [x8] +; CHECK-NEXT: ld1d { z1.d }, p0/z, [x8] +; CHECK-NEXT: tbl z0.d, { z0.d }, z1.d ; CHECK-NEXT: st1d { z0.d }, p0, [x0] -; CHECK-NEXT: mov sp, x29 -; CHECK-NEXT: ldp x29, x30, [sp], #16 // 16-byte Folded Reload ; CHECK-NEXT: ret %tmp1 = load <4 x i64>, ptr %a %tmp2 = shufflevector <4 x i64> %tmp1, <4 x i64> undef, <4 x i32> @@ -260,39 +245,26 @@ ; sve-vector-bits-min=256, sve-vector-bits-max is not set, REV inst can't be generated. define void @test_revv8i32(ptr %a) #0 { -; CHECK-LABEL: test_revv8i32: -; CHECK: // %bb.0: -; CHECK-NEXT: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill -; CHECK-NEXT: sub x9, sp, #48 -; CHECK-NEXT: mov x29, sp -; CHECK-NEXT: and sp, x9, #0xffffffffffffffe0 -; CHECK-NEXT: .cfi_def_cfa w29, 16 -; CHECK-NEXT: .cfi_offset w30, -8 -; CHECK-NEXT: .cfi_offset w29, -16 -; CHECK-NEXT: ptrue p0.s, vl8 -; CHECK-NEXT: ld1w { z0.s }, p0/z, [x0] -; CHECK-NEXT: mov w8, v0.s[1] -; CHECK-NEXT: mov w9, v0.s[2] -; CHECK-NEXT: mov w10, v0.s[3] -; CHECK-NEXT: fmov w11, s0 -; CHECK-NEXT: mov z1.s, z0.s[4] -; CHECK-NEXT: mov z2.s, z0.s[5] -; CHECK-NEXT: mov z3.s, z0.s[6] -; CHECK-NEXT: mov z0.s, z0.s[7] -; CHECK-NEXT: stp w8, w11, [sp, #24] -; CHECK-NEXT: fmov w8, s1 -; CHECK-NEXT: stp w10, w9, [sp, #16] -; CHECK-NEXT: fmov w9, s2 -; CHECK-NEXT: stp w9, w8, [sp, #8] -; CHECK-NEXT: fmov w8, s3 -; CHECK-NEXT: fmov w9, s0 -; CHECK-NEXT: stp w9, w8, [sp] -; CHECK-NEXT: mov x8, sp -; CHECK-NEXT: ld1w { z0.s }, p0/z, [x8] -; CHECK-NEXT: st1w { z0.s }, p0, [x0] -; CHECK-NEXT: mov sp, x29 -; CHECK-NEXT: ldp x29, x30, [sp], #16 // 16-byte Folded Reload -; CHECK-NEXT: ret +; VBITS_GE_256-LABEL: test_revv8i32: +; VBITS_GE_256: // %bb.0: +; VBITS_GE_256-NEXT: ptrue p0.s, vl8 +; VBITS_GE_256-NEXT: index z0.s, #7, #-1 +; VBITS_GE_256-NEXT: ld1w { z1.s }, p0/z, [x0] +; VBITS_GE_256-NEXT: tbl z0.s, { z1.s }, z0.s +; VBITS_GE_256-NEXT: st1w { z0.s }, p0, [x0] +; VBITS_GE_256-NEXT: ret +; +; VBITS_GE_512-LABEL: test_revv8i32: +; VBITS_GE_512: // %bb.0: +; VBITS_GE_512-NEXT: ptrue p0.s, vl8 +; VBITS_GE_512-NEXT: adrp x8, .LCPI14_0 +; VBITS_GE_512-NEXT: add x8, x8, :lo12:.LCPI14_0 +; VBITS_GE_512-NEXT: ptrue p1.s, vl16 +; VBITS_GE_512-NEXT: ld1w { z0.s }, p0/z, [x0] +; VBITS_GE_512-NEXT: ld1w { z1.s }, p1/z, [x8] +; VBITS_GE_512-NEXT: tbl z0.s, { z0.s }, z1.s +; VBITS_GE_512-NEXT: st1w { z0.s }, p0, [x0] +; VBITS_GE_512-NEXT: ret %tmp1 = load <8 x i32>, ptr %a %tmp2 = shufflevector <8 x i32> %tmp1, <8 x i32> undef, <8 x i32> store <8 x i32> %tmp2, ptr %a @@ -379,60 +351,13 @@ define void @test_rev_fail(ptr %a) #1 { ; CHECK-LABEL: test_rev_fail: ; CHECK: // %bb.0: -; CHECK-NEXT: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill -; CHECK-NEXT: sub x9, sp, #48 -; CHECK-NEXT: mov x29, sp -; CHECK-NEXT: and sp, x9, #0xffffffffffffffe0 -; CHECK-NEXT: .cfi_def_cfa w29, 16 -; CHECK-NEXT: .cfi_offset w30, -8 -; CHECK-NEXT: .cfi_offset w29, -16 ; CHECK-NEXT: ptrue p0.h +; CHECK-NEXT: adrp x8, .LCPI20_0 +; CHECK-NEXT: add x8, x8, :lo12:.LCPI20_0 ; CHECK-NEXT: ld1h { z0.h }, p0/z, [x0] -; CHECK-NEXT: mov z1.h, z0.h[8] -; CHECK-NEXT: fmov w8, s0 -; CHECK-NEXT: mov z2.h, z0.h[9] -; CHECK-NEXT: mov z3.h, z0.h[10] -; CHECK-NEXT: mov z4.h, z0.h[11] -; CHECK-NEXT: strh w8, [sp, #14] -; CHECK-NEXT: fmov w8, s1 -; CHECK-NEXT: mov z1.h, z0.h[12] -; CHECK-NEXT: fmov w9, s2 -; CHECK-NEXT: mov z2.h, z0.h[13] -; CHECK-NEXT: strh w8, [sp, #30] -; CHECK-NEXT: fmov w8, s3 -; CHECK-NEXT: mov z3.h, z0.h[14] -; CHECK-NEXT: strh w9, [sp, #28] -; CHECK-NEXT: fmov w9, s4 -; CHECK-NEXT: mov z4.h, z0.h[15] -; CHECK-NEXT: fmov w10, s2 -; CHECK-NEXT: strh w8, [sp, #26] -; CHECK-NEXT: fmov w8, s1 -; CHECK-NEXT: fmov w11, s3 -; CHECK-NEXT: strh w9, [sp, #24] -; CHECK-NEXT: umov w9, v0.h[1] -; CHECK-NEXT: fmov w12, s4 -; CHECK-NEXT: strh w10, [sp, #20] -; CHECK-NEXT: umov w10, v0.h[3] -; CHECK-NEXT: strh w8, [sp, #22] -; CHECK-NEXT: umov w8, v0.h[2] -; CHECK-NEXT: strh w11, [sp, #18] -; CHECK-NEXT: umov w11, v0.h[4] -; CHECK-NEXT: strh w12, [sp, #16] -; CHECK-NEXT: umov w12, v0.h[5] -; CHECK-NEXT: strh w9, [sp, #12] -; CHECK-NEXT: umov w9, v0.h[6] -; CHECK-NEXT: strh w8, [sp, #10] -; CHECK-NEXT: umov w8, v0.h[7] -; CHECK-NEXT: strh w10, [sp, #8] -; CHECK-NEXT: strh w11, [sp, #6] -; CHECK-NEXT: strh w12, [sp, #4] -; CHECK-NEXT: strh w9, [sp, #2] -; CHECK-NEXT: strh w8, [sp] -; CHECK-NEXT: mov x8, sp -; CHECK-NEXT: ld1h { z0.h }, p0/z, [x8] +; CHECK-NEXT: ld1h { z1.h }, p0/z, [x8] +; CHECK-NEXT: tbl z0.h, { z0.h }, z1.h ; CHECK-NEXT: st1h { z0.h }, p0, [x0] -; CHECK-NEXT: mov sp, x29 -; CHECK-NEXT: ldp x29, x30, [sp], #16 // 16-byte Folded Reload ; CHECK-NEXT: ret %tmp1 = load <16 x i16>, ptr %a %tmp2 = shufflevector <16 x i16> %tmp1, <16 x i16> undef, <16 x i32> diff --git a/llvm/test/CodeGen/AArch64/sve-fixed-length-vector-shuffle-tbl.ll b/llvm/test/CodeGen/AArch64/sve-fixed-length-vector-shuffle-tbl.ll new file mode 100644 --- /dev/null +++ b/llvm/test/CodeGen/AArch64/sve-fixed-length-vector-shuffle-tbl.ll @@ -0,0 +1,254 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc -mattr=+sve2 -force-streaming-compatible-sve -aarch64-sve-vector-bits-min=128 -aarch64-sve-vector-bits-max=128 < %s | FileCheck %s -check-prefixes=CHECK,SVE2_128 +; RUN: llc -mattr=+sve2 -force-streaming-compatible-sve -aarch64-sve-vector-bits-min=128 < %s | FileCheck %s -check-prefixes=CHECK,SVE2_128_NOMAX + +target triple = "aarch64-unknown-linux-gnu" + +; SVE2_128: .LCPI0_0: +; SVE2_128-NEXT: .byte 0 // 0x0 +; SVE2_128-NEXT: .byte 7 // 0x7 +; SVE2_128-NEXT: .byte 2 // 0x2 +; SVE2_128-NEXT: .byte 3 // 0x3 +; SVE2_128-NEXT: .byte 4 // 0x4 +; SVE2_128-NEXT: .byte 5 // 0x5 +; SVE2_128-NEXT: .byte 6 // 0x6 +; SVE2_128-NEXT: .byte 7 // 0x7 +; SVE2_128-NEXT: .byte 255 // 0xff +; SVE2_128-NEXT: .byte 255 // 0xff +define <8 x i8> @shuffle_index_indices_from_op1(ptr %a, ptr %b) { +; CHECK-LABEL: shuffle_index_indices_from_op1: +; CHECK: // %bb.0: +; CHECK-NEXT: adrp x8, .LCPI0_0 +; CHECK-NEXT: ldr d0, [x0] +; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI0_0] +; CHECK-NEXT: tbl z0.b, { z0.b }, z1.b +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0 +; CHECK-NEXT: ret + %op1 = load <8 x i8>, ptr %a + %op2 = load <8 x i8>, ptr %b + %1 = shufflevector <8 x i8> %op1, <8 x i8> %op2, <8 x i32> + ret <8 x i8> %1 +} + +; SVE2_128: .LCPI1_0: +; SVE2_128-NEXT: .byte 0 // 0x0 +; SVE2_128-NEXT: .byte 1 // 0x1 +; SVE2_128-NEXT: .byte 1 // 0x1 +; SVE2_128-NEXT: .byte 3 // 0x3 +; SVE2_128-NEXT: .byte 4 // 0x4 +; SVE2_128-NEXT: .byte 7 // 0x7 +; SVE2_128-NEXT: .byte 6 // 0x6 +; SVE2_128-NEXT: .byte 7 // 0x7 +; SVE2_128-NEXT: .byte 255 // 0xff +; SVE2_128-NEXT: .byte 255 // 0xff +define <8 x i8> @shuffle_index_indices_from_op2(ptr %a, ptr %b) { +; CHECK-LABEL: shuffle_index_indices_from_op2: +; CHECK: // %bb.0: +; CHECK-NEXT: adrp x8, .LCPI1_0 +; CHECK-NEXT: ldr d0, [x1] +; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI1_0] +; CHECK-NEXT: tbl z0.b, { z0.b }, z1.b +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0 +; CHECK-NEXT: ret + %op1 = load <8 x i8>, ptr %a + %op2 = load <8 x i8>, ptr %b + %1 = shufflevector <8 x i8> %op1, <8 x i8> %op2, <8 x i32> + ret <8 x i8> %1 +} + +; SVE2_128: .LCPI2_0: +; SVE2_128-NEXT: .byte 1 // 0x1 +; SVE2_128-NEXT: .byte 17 // 0x11 +; SVE2_128-NEXT: .byte 18 // 0x12 +; SVE2_128-NEXT: .byte 19 // 0x13 +; SVE2_128-NEXT: .byte 20 // 0x14 +; SVE2_128-NEXT: .byte 20 // 0x14 +; SVE2_128-NEXT: .byte 22 // 0x16 +; SVE2_128-NEXT: .byte 23 // 0x17 +; SVE2_128-NEXT: .byte 255 // 0xff +; SVE2_128-NEXT: .byte 255 // 0xff +define <8 x i8> @shuffle_index_indices_from_both_ops(ptr %a, ptr %b) { +; SVE2_128-LABEL: shuffle_index_indices_from_both_ops: +; SVE2_128: // %bb.0: +; SVE2_128-NEXT: adrp x8, .LCPI2_0 +; SVE2_128-NEXT: ldr d0, [x0] +; SVE2_128-NEXT: ldr d1, [x1] +; SVE2_128-NEXT: ldr q2, [x8, :lo12:.LCPI2_0] +; SVE2_128-NEXT: tbl z0.b, { z0.b, z1.b }, z2.b +; SVE2_128-NEXT: // kill: def $d0 killed $d0 killed $z0 +; SVE2_128-NEXT: ret +; +; SVE2_128_NOMAX-LABEL: shuffle_index_indices_from_both_ops: +; SVE2_128_NOMAX: // %bb.0: +; SVE2_128_NOMAX-NEXT: sub sp, sp, #16 +; SVE2_128_NOMAX-NEXT: .cfi_def_cfa_offset 16 +; SVE2_128_NOMAX-NEXT: ldr d0, [x1] +; SVE2_128_NOMAX-NEXT: mov z1.b, z0.b[7] +; SVE2_128_NOMAX-NEXT: mov z2.b, z0.b[6] +; SVE2_128_NOMAX-NEXT: mov z3.b, z0.b[4] +; SVE2_128_NOMAX-NEXT: fmov w8, s1 +; SVE2_128_NOMAX-NEXT: ldr d1, [x0] +; SVE2_128_NOMAX-NEXT: fmov w9, s2 +; SVE2_128_NOMAX-NEXT: mov z2.b, z0.b[3] +; SVE2_128_NOMAX-NEXT: mov z1.b, z1.b[1] +; SVE2_128_NOMAX-NEXT: strb w8, [sp, #15] +; SVE2_128_NOMAX-NEXT: fmov w8, s3 +; SVE2_128_NOMAX-NEXT: mov z3.b, z0.b[2] +; SVE2_128_NOMAX-NEXT: strb w9, [sp, #14] +; SVE2_128_NOMAX-NEXT: mov z0.b, z0.b[1] +; SVE2_128_NOMAX-NEXT: fmov w9, s2 +; SVE2_128_NOMAX-NEXT: strb w8, [sp, #13] +; SVE2_128_NOMAX-NEXT: strb w8, [sp, #12] +; SVE2_128_NOMAX-NEXT: fmov w8, s3 +; SVE2_128_NOMAX-NEXT: strb w9, [sp, #11] +; SVE2_128_NOMAX-NEXT: fmov w9, s0 +; SVE2_128_NOMAX-NEXT: strb w8, [sp, #10] +; SVE2_128_NOMAX-NEXT: fmov w8, s1 +; SVE2_128_NOMAX-NEXT: strb w9, [sp, #9] +; SVE2_128_NOMAX-NEXT: strb w8, [sp, #8] +; SVE2_128_NOMAX-NEXT: ldr d0, [sp, #8] +; SVE2_128_NOMAX-NEXT: add sp, sp, #16 +; SVE2_128_NOMAX-NEXT: ret + %op1 = load <8 x i8>, ptr %a + %op2 = load <8 x i8>, ptr %b + %1 = shufflevector <8 x i8> %op1, <8 x i8> %op2, <8 x i32> + ret <8 x i8> %1 +} + +; SVE2_128: .LCPI3_0: +; SVE2_128-NEXT: .byte 1 // 0x1 +; SVE2_128-NEXT: .byte 17 // 0x11 +; SVE2_128-NEXT: .byte 18 // 0x12 +; SVE2_128-NEXT: .byte 19 // 0x13 +; SVE2_128-NEXT: .byte 20 // 0x14 +; SVE2_128-NEXT: .byte 20 // 0x14 +; SVE2_128-NEXT: .byte 22 // 0x16 +; SVE2_128-NEXT: .byte 0 // 0x0 +; SVE2_128-NEXT: .byte 255 // 0xff +; SVE2_128-NEXT: .byte 255 // 0xff +define <8 x i8> @shuffle_index_poison_value(ptr %a, ptr %b) { +; SVE2_128-LABEL: shuffle_index_poison_value: +; SVE2_128: // %bb.0: +; SVE2_128-NEXT: adrp x8, .LCPI3_0 +; SVE2_128-NEXT: ldr d0, [x0] +; SVE2_128-NEXT: ldr d1, [x1] +; SVE2_128-NEXT: ldr q2, [x8, :lo12:.LCPI3_0] +; SVE2_128-NEXT: tbl z0.b, { z0.b, z1.b }, z2.b +; SVE2_128-NEXT: // kill: def $d0 killed $d0 killed $z0 +; SVE2_128-NEXT: ret +; +; SVE2_128_NOMAX-LABEL: shuffle_index_poison_value: +; SVE2_128_NOMAX: // %bb.0: +; SVE2_128_NOMAX-NEXT: sub sp, sp, #16 +; SVE2_128_NOMAX-NEXT: .cfi_def_cfa_offset 16 +; SVE2_128_NOMAX-NEXT: ldr d0, [x1] +; SVE2_128_NOMAX-NEXT: ldr d3, [x0] +; SVE2_128_NOMAX-NEXT: mov z1.b, z0.b[6] +; SVE2_128_NOMAX-NEXT: mov z2.b, z0.b[4] +; SVE2_128_NOMAX-NEXT: fmov w8, s1 +; SVE2_128_NOMAX-NEXT: mov z1.b, z0.b[3] +; SVE2_128_NOMAX-NEXT: fmov w9, s2 +; SVE2_128_NOMAX-NEXT: mov z2.b, z0.b[2] +; SVE2_128_NOMAX-NEXT: mov z0.b, z0.b[1] +; SVE2_128_NOMAX-NEXT: strb w8, [sp, #14] +; SVE2_128_NOMAX-NEXT: fmov w8, s1 +; SVE2_128_NOMAX-NEXT: mov z1.b, z3.b[1] +; SVE2_128_NOMAX-NEXT: strb w9, [sp, #13] +; SVE2_128_NOMAX-NEXT: strb w9, [sp, #12] +; SVE2_128_NOMAX-NEXT: fmov w9, s2 +; SVE2_128_NOMAX-NEXT: strb w8, [sp, #11] +; SVE2_128_NOMAX-NEXT: fmov w8, s0 +; SVE2_128_NOMAX-NEXT: strb w9, [sp, #10] +; SVE2_128_NOMAX-NEXT: fmov w9, s1 +; SVE2_128_NOMAX-NEXT: strb w8, [sp, #9] +; SVE2_128_NOMAX-NEXT: strb w9, [sp, #8] +; SVE2_128_NOMAX-NEXT: ldr d0, [sp, #8] +; SVE2_128_NOMAX-NEXT: add sp, sp, #16 +; SVE2_128_NOMAX-NEXT: ret + %op1 = load <8 x i8>, ptr %a + %op2 = load <8 x i8>, ptr %b + %1 = shufflevector <8 x i8> %op1, <8 x i8> %op2, <8 x i32> + ret <8 x i8> %1 +} + +define <8 x i8> @shuffle_op1_poison(ptr %a, ptr %b) { +; CHECK-LABEL: shuffle_op1_poison: +; CHECK: // %bb.0: +; CHECK-NEXT: adrp x8, .LCPI4_0 +; CHECK-NEXT: ldr d0, [x1] +; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI4_0] +; CHECK-NEXT: tbl z0.b, { z0.b }, z1.b +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0 +; CHECK-NEXT: ret + %op2 = load <8 x i8>, ptr %b + %1 = shufflevector <8 x i8> poison, <8 x i8> %op2, <8 x i32> + ret <8 x i8> %1 +} + +; In this function, we could not represent indexes for the second operand +; because for i8 type, the maximum constant in the mask is 256. +define <8 x i8> @negative_test_shuffle_index_size_op_both_maxhw(ptr %a, ptr %b) "target-features"="+sve2" vscale_range(16,16) { +; CHECK-LABEL: negative_test_shuffle_index_size_op_both_maxhw: +; CHECK: // %bb.0: +; CHECK-NEXT: sub sp, sp, #16 +; CHECK-NEXT: .cfi_def_cfa_offset 16 +; CHECK-NEXT: ldr d0, [x1] +; CHECK-NEXT: mov z1.b, z0.b[7] +; CHECK-NEXT: mov z2.b, z0.b[6] +; CHECK-NEXT: mov z3.b, z0.b[4] +; CHECK-NEXT: fmov w8, s1 +; CHECK-NEXT: ldr d1, [x0] +; CHECK-NEXT: fmov w9, s2 +; CHECK-NEXT: mov z2.b, z0.b[3] +; CHECK-NEXT: mov z1.b, z1.b[1] +; CHECK-NEXT: strb w8, [sp, #15] +; CHECK-NEXT: fmov w8, s3 +; CHECK-NEXT: mov z3.b, z0.b[2] +; CHECK-NEXT: strb w9, [sp, #14] +; CHECK-NEXT: mov z0.b, z0.b[1] +; CHECK-NEXT: fmov w9, s2 +; CHECK-NEXT: strb w8, [sp, #13] +; CHECK-NEXT: strb w8, [sp, #12] +; CHECK-NEXT: fmov w8, s3 +; CHECK-NEXT: strb w9, [sp, #11] +; CHECK-NEXT: fmov w9, s0 +; CHECK-NEXT: strb w8, [sp, #10] +; CHECK-NEXT: fmov w8, s1 +; CHECK-NEXT: strb w9, [sp, #9] +; CHECK-NEXT: strb w8, [sp, #8] +; CHECK-NEXT: ldr d0, [sp, #8] +; CHECK-NEXT: add sp, sp, #16 +; CHECK-NEXT: ret + %op1 = load <8 x i8>, ptr %a + %op2 = load <8 x i8>, ptr %b + %1 = shufflevector <8 x i8> %op1, <8 x i8> %op2, <8 x i32> + ret <8 x i8> %1 +} + +; CHECK: .LCPI6_0: +; CHECK-NEXT: .byte 0 // 0x0 +; CHECK-NEXT: .byte 7 // 0x7 +; CHECK-NEXT: .byte 2 // 0x2 +; CHECK-NEXT: .byte 3 // 0x3 +; CHECK-NEXT: .byte 4 // 0x4 +; CHECK-NEXT: .byte 5 // 0x5 +; CHECK-NEXT: .byte 6 // 0x6 +; CHECK-NEXT: .byte 7 // 0x7 +; CHECK-NEXT: .byte 255 // 0xff +; CHECK-NEXT: .byte 255 // 0xff +define <8 x i8> @shuffle_index_size_op1_maxhw(ptr %a, ptr %b) "target-features"="+sve2" vscale_range(16,16) { +; CHECK-LABEL: shuffle_index_size_op1_maxhw: +; CHECK: // %bb.0: +; CHECK-NEXT: ptrue p0.b +; CHECK-NEXT: adrp x8, .LCPI6_0 +; CHECK-NEXT: add x8, x8, :lo12:.LCPI6_0 +; CHECK-NEXT: ldr d0, [x0] +; CHECK-NEXT: ld1b { z1.b }, p0/z, [x8] +; CHECK-NEXT: tbl z0.b, { z0.b }, z1.b +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $z0 +; CHECK-NEXT: ret + %op1 = load <8 x i8>, ptr %a + %op2 = load <8 x i8>, ptr %b + %1 = shufflevector <8 x i8> %op1, <8 x i8> %op2, <8 x i32> + ret <8 x i8> %1 +}