diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td index e189b12d88b..a7aaef6957a 100644 --- a/lib/Target/ARM/ARMInstrNEON.td +++ b/lib/Target/ARM/ARMInstrNEON.td @@ -184,6 +184,7 @@ def VecListOneDWordIndexed : Operand { let ParserMatchClass = VecListOneDWordIndexAsmOperand; let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx); } + // Register list of two D registers with byte lane subscripting. def VecListTwoDByteIndexAsmOperand : AsmOperandClass { let Name = "VecListTwoDByteIndexed"; @@ -235,6 +236,59 @@ def VecListTwoQWordIndexed : Operand { let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx); } + +// Register list of three D registers with byte lane subscripting. +def VecListThreeDByteIndexAsmOperand : AsmOperandClass { + let Name = "VecListThreeDByteIndexed"; + let ParserMethod = "parseVectorList"; + let RenderMethod = "addVecListIndexedOperands"; +} +def VecListThreeDByteIndexed : Operand { + let ParserMatchClass = VecListThreeDByteIndexAsmOperand; + let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx); +} +// ...with half-word lane subscripting. +def VecListThreeDHWordIndexAsmOperand : AsmOperandClass { + let Name = "VecListThreeDHWordIndexed"; + let ParserMethod = "parseVectorList"; + let RenderMethod = "addVecListIndexedOperands"; +} +def VecListThreeDHWordIndexed : Operand { + let ParserMatchClass = VecListThreeDHWordIndexAsmOperand; + let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx); +} +// ...with word lane subscripting. +def VecListThreeDWordIndexAsmOperand : AsmOperandClass { + let Name = "VecListThreeDWordIndexed"; + let ParserMethod = "parseVectorList"; + let RenderMethod = "addVecListIndexedOperands"; +} +def VecListThreeDWordIndexed : Operand { + let ParserMatchClass = VecListThreeDWordIndexAsmOperand; + let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx); +} +// Register list of three Q registers with half-word lane subscripting. +def VecListThreeQHWordIndexAsmOperand : AsmOperandClass { + let Name = "VecListThreeQHWordIndexed"; + let ParserMethod = "parseVectorList"; + let RenderMethod = "addVecListIndexedOperands"; +} +def VecListThreeQHWordIndexed : Operand { + let ParserMatchClass = VecListThreeQHWordIndexAsmOperand; + let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx); +} +// ...with word lane subscripting. +def VecListThreeQWordIndexAsmOperand : AsmOperandClass { + let Name = "VecListThreeQWordIndexed"; + let ParserMethod = "parseVectorList"; + let RenderMethod = "addVecListIndexedOperands"; +} +def VecListThreeQWordIndexed : Operand { + let ParserMatchClass = VecListThreeQWordIndexAsmOperand; + let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx); +} + + //===----------------------------------------------------------------------===// // NEON-specific DAG Nodes. //===----------------------------------------------------------------------===// @@ -5914,6 +5968,55 @@ def VST2LNqWB_register_Asm_32 : rGPR:$Rm, pred:$p)>; +// VLD3 single-lane pseudo-instructions. These need special handling for +// the lane index that an InstAlias can't handle, so we use these instead. +def VLD3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr", + (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>; +def VLD3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr", + (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>; +def VLD3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr", + (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>; +def VLD3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr", + (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>; +def VLD3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr", + (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>; + +def VLD3LNdWB_fixed_Asm_8 : + NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!", + (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>; +def VLD3LNdWB_fixed_Asm_16 : + NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!", + (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>; +def VLD3LNdWB_fixed_Asm_32 : + NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!", + (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>; +def VLD3LNqWB_fixed_Asm_16 : + NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!", + (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>; +def VLD3LNqWB_fixed_Asm_32 : + NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!", + (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>; +def VLD3LNdWB_register_Asm_8 : + NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm", + (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, + rGPR:$Rm, pred:$p)>; +def VLD3LNdWB_register_Asm_16 : + NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm", + (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, + rGPR:$Rm, pred:$p)>; +def VLD3LNdWB_register_Asm_32 : + NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm", + (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, + rGPR:$Rm, pred:$p)>; +def VLD3LNqWB_register_Asm_16 : + NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm", + (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, + rGPR:$Rm, pred:$p)>; +def VLD3LNqWB_register_Asm_32 : + NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm", + (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, + rGPR:$Rm, pred:$p)>; + // VMOV takes an optional datatype suffix defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm", (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>; diff --git a/lib/Target/ARM/AsmParser/ARMAsmParser.cpp b/lib/Target/ARM/AsmParser/ARMAsmParser.cpp index 6cdfc56102a..c685a266b8a 100644 --- a/lib/Target/ARM/AsmParser/ARMAsmParser.cpp +++ b/lib/Target/ARM/AsmParser/ARMAsmParser.cpp @@ -1168,6 +1168,31 @@ public: return VectorList.Count == 2 && VectorList.LaneIndex <= 1; } + bool isVecListThreeDByteIndexed() const { + if (!isSingleSpacedVectorIndexed()) return false; + return VectorList.Count == 3 && VectorList.LaneIndex <= 7; + } + + bool isVecListThreeDHWordIndexed() const { + if (!isSingleSpacedVectorIndexed()) return false; + return VectorList.Count == 3 && VectorList.LaneIndex <= 3; + } + + bool isVecListThreeQWordIndexed() const { + if (!isDoubleSpacedVectorIndexed()) return false; + return VectorList.Count == 3 && VectorList.LaneIndex <= 1; + } + + bool isVecListThreeQHWordIndexed() const { + if (!isDoubleSpacedVectorIndexed()) return false; + return VectorList.Count == 3 && VectorList.LaneIndex <= 3; + } + + bool isVecListThreeDWordIndexed() const { + if (!isSingleSpacedVectorIndexed()) return false; + return VectorList.Count == 3 && VectorList.LaneIndex <= 1; + } + bool isVectorIndex8() const { if (Kind != k_VectorIndex) return false; return VectorIndex.Val < 8; @@ -5304,6 +5329,53 @@ static unsigned getRealVLDLNOpcode(unsigned Opc, unsigned &Spacing) { case ARM::VLD2LNqAsm_32: Spacing = 2; return ARM::VLD2LNq32; + + // VLD3LN + case ARM::VLD3LNdWB_fixed_Asm_8: + Spacing = 1; + return ARM::VLD3LNd8_UPD; + case ARM::VLD3LNdWB_fixed_Asm_16: + Spacing = 1; + return ARM::VLD3LNd16_UPD; + case ARM::VLD3LNdWB_fixed_Asm_32: + Spacing = 1; + return ARM::VLD3LNd32_UPD; + case ARM::VLD3LNqWB_fixed_Asm_16: + Spacing = 1; + return ARM::VLD3LNq16_UPD; + case ARM::VLD3LNqWB_fixed_Asm_32: + Spacing = 2; + return ARM::VLD3LNq32_UPD; + case ARM::VLD3LNdWB_register_Asm_8: + Spacing = 1; + return ARM::VLD3LNd8_UPD; + case ARM::VLD3LNdWB_register_Asm_16: + Spacing = 1; + return ARM::VLD3LNd16_UPD; + case ARM::VLD3LNdWB_register_Asm_32: + Spacing = 1; + return ARM::VLD3LNd32_UPD; + case ARM::VLD3LNqWB_register_Asm_16: + Spacing = 2; + return ARM::VLD3LNq16_UPD; + case ARM::VLD3LNqWB_register_Asm_32: + Spacing = 2; + return ARM::VLD3LNq32_UPD; + case ARM::VLD3LNdAsm_8: + Spacing = 1; + return ARM::VLD3LNd8; + case ARM::VLD3LNdAsm_16: + Spacing = 1; + return ARM::VLD3LNd16; + case ARM::VLD3LNdAsm_32: + Spacing = 1; + return ARM::VLD3LNd32; + case ARM::VLD3LNqAsm_16: + Spacing = 2; + return ARM::VLD3LNq16; + case ARM::VLD3LNqAsm_32: + Spacing = 2; + return ARM::VLD3LNq32; } } @@ -5502,6 +5574,37 @@ processInstruction(MCInst &Inst, return true; } + case ARM::VLD3LNdWB_register_Asm_8: + case ARM::VLD3LNdWB_register_Asm_16: + case ARM::VLD3LNdWB_register_Asm_32: + case ARM::VLD3LNqWB_register_Asm_16: + case ARM::VLD3LNqWB_register_Asm_32: { + MCInst TmpInst; + // Shuffle the operands around so the lane index operand is in the + // right place. + unsigned Spacing; + TmpInst.setOpcode(getRealVLDLNOpcode(Inst.getOpcode(), Spacing)); + TmpInst.addOperand(Inst.getOperand(0)); // Vd + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(Inst.getOperand(2)); // Rn_wb + TmpInst.addOperand(Inst.getOperand(2)); // Rn + TmpInst.addOperand(Inst.getOperand(3)); // alignment + TmpInst.addOperand(Inst.getOperand(4)); // Rm + TmpInst.addOperand(Inst.getOperand(0)); // Tied operand src (== Vd) + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(Inst.getOperand(1)); // lane + TmpInst.addOperand(Inst.getOperand(5)); // CondCode + TmpInst.addOperand(Inst.getOperand(6)); + Inst = TmpInst; + return true; + } + case ARM::VLD1LNdWB_fixed_Asm_8: case ARM::VLD1LNdWB_fixed_Asm_16: case ARM::VLD1LNdWB_fixed_Asm_32: { @@ -5550,6 +5653,37 @@ processInstruction(MCInst &Inst, return true; } + case ARM::VLD3LNdWB_fixed_Asm_8: + case ARM::VLD3LNdWB_fixed_Asm_16: + case ARM::VLD3LNdWB_fixed_Asm_32: + case ARM::VLD3LNqWB_fixed_Asm_16: + case ARM::VLD3LNqWB_fixed_Asm_32: { + MCInst TmpInst; + // Shuffle the operands around so the lane index operand is in the + // right place. + unsigned Spacing; + TmpInst.setOpcode(getRealVLDLNOpcode(Inst.getOpcode(), Spacing)); + TmpInst.addOperand(Inst.getOperand(0)); // Vd + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(Inst.getOperand(2)); // Rn_wb + TmpInst.addOperand(Inst.getOperand(2)); // Rn + TmpInst.addOperand(Inst.getOperand(3)); // alignment + TmpInst.addOperand(MCOperand::CreateReg(0)); // Rm + TmpInst.addOperand(Inst.getOperand(0)); // Tied operand src (== Vd) + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(Inst.getOperand(1)); // lane + TmpInst.addOperand(Inst.getOperand(4)); // CondCode + TmpInst.addOperand(Inst.getOperand(5)); + Inst = TmpInst; + return true; + } + case ARM::VLD1LNdAsm_8: case ARM::VLD1LNdAsm_16: case ARM::VLD1LNdAsm_32: { @@ -5593,6 +5727,36 @@ processInstruction(MCInst &Inst, Inst = TmpInst; return true; } + + case ARM::VLD3LNdAsm_8: + case ARM::VLD3LNdAsm_16: + case ARM::VLD3LNdAsm_32: + case ARM::VLD3LNqAsm_16: + case ARM::VLD3LNqAsm_32: { + MCInst TmpInst; + // Shuffle the operands around so the lane index operand is in the + // right place. + unsigned Spacing; + TmpInst.setOpcode(getRealVLDLNOpcode(Inst.getOpcode(), Spacing)); + TmpInst.addOperand(Inst.getOperand(0)); // Vd + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(Inst.getOperand(2)); // Rn + TmpInst.addOperand(Inst.getOperand(3)); // alignment + TmpInst.addOperand(Inst.getOperand(0)); // Tied operand src (== Vd) + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(MCOperand::CreateReg(Inst.getOperand(0).getReg() + + Spacing)); + TmpInst.addOperand(Inst.getOperand(1)); // lane + TmpInst.addOperand(Inst.getOperand(4)); // CondCode + TmpInst.addOperand(Inst.getOperand(5)); + Inst = TmpInst; + return true; + } + // Handle the Thumb2 mode MOV complex aliases. case ARM::t2MOVsr: case ARM::t2MOVSsr: { diff --git a/test/MC/ARM/neon-vst-encoding.s b/test/MC/ARM/neon-vst-encoding.s index 57c7fe15d08..19aefee2ca7 100644 --- a/test/MC/ARM/neon-vst-encoding.s +++ b/test/MC/ARM/neon-vst-encoding.s @@ -133,17 +133,39 @@ @ CHECK: vst2.32 {d5[0], d7[0]}, [r4, :64], r7 @ encoding: [0x57,0x59,0x84,0xf4] -@ vst3.8 {d16[1], d17[1], d18[1]}, [r0] -@ vst3.16 {d16[1], d17[1], d18[1]}, [r0] -@ vst3.32 {d16[1], d17[1], d18[1]}, [r0] -@ vst3.16 {d17[2], d19[2], d21[2]}, [r0] -@ vst3.32 {d16[0], d18[0], d20[0]}, [r0] + vld3.8 {d16[1], d17[1], d18[1]}, [r1] + vld3.16 {d6[1], d7[1], d8[1]}, [r2] + vld3.32 {d1[1], d2[1], d3[1]}, [r3] + vld3.u16 {d27[2], d29[2], d31[2]}, [r4] + vld3.i32 {d6[0], d8[0], d10[0]}, [r5] -@ FIXME: vst3.8 {d16[1], d17[1], d18[1]}, [r0] @ encoding: [0x2f,0x02,0xc0,0xf4] -@ FIXME: vst3.16 {d16[1], d17[1], d18[1]}, [r0]@ encoding: [0x4f,0x06,0xc0,0xf4] -@ FIXME: vst3.32 {d16[1], d17[1], d18[1]}, [r0]@ encoding: [0x8f,0x0a,0xc0,0xf4] -@ FIXME: vst3.16 {d17[2], d19[2], d21[2]}, [r0]@ encoding: [0xaf,0x16,0xc0,0xf4] -@ FIXME: vst3.32 {d16[0], d18[0], d20[0]}, [r0]@ encoding: [0x4f,0x0a,0xc0,0xf4] + vld3.i8 {d12[3], d13[3], d14[3]}, [r6], r1 + vld3.i16 {d11[2], d12[2], d13[2]}, [r7], r2 + vld3.u32 {d2[1], d3[1], d4[1]}, [r8], r3 + vld3.u16 {d14[2], d16[2], d18[2]}, [r9], r4 + vld3.i32 {d16[0], d18[0], d20[0]}, [r10], r5 + + vld3.p8 {d6[6], d7[6], d8[6]}, [r8]! + vld3.16 {d9[2], d10[2], d11[2]}, [r7]! + vld3.f32 {d1[1], d2[1], d3[1]}, [r6]! + vld3.p16 {d20[2], d22[2], d24[2]}, [r5]! + vld3.32 {d5[0], d7[0], d9[0]}, [r4]! + +@ CHECK: vld3.8 {d16[1], d17[1], d17[1]}, [r1] @ encoding: [0x2f,0x02,0xe1,0xf4] +@ CHECK: vld3.16 {d6[1], d7[1], d7[1]}, [r2] @ encoding: [0x4f,0x66,0xa2,0xf4] +@ CHECK: vld3.32 {d1[1], d2[1], d2[1]}, [r3] @ encoding: [0x8f,0x1a,0xa3,0xf4] +@ CHECK: vld3.16 {d27[2], d29[2], d29[2]}, [r4] @ encoding: [0xaf,0xb6,0xe4,0xf4] +@ CHECK: vld3.32 {d6[0], d8[0], d8[0]}, [r5] @ encoding: [0x4f,0x6a,0xa5,0xf4] +@ CHECK: vld3.8 {d12[3], d13[3], d13[3]}, [r6], r1 @ encoding: [0x61,0xc2,0xa6,0xf4] +@ CHECK: vld3.16 {d11[2], d12[2], d12[2]}, [r7], r2 @ encoding: [0x82,0xb6,0xa7,0xf4] +@ CHECK: vld3.32 {d2[1], d3[1], d3[1]}, [r8], r3 @ encoding: [0x83,0x2a,0xa8,0xf4] +@ CHECK: vld3.16 {d14[2], d16[2], d16[2]}, [r9], r4 @ encoding: [0xa4,0xe6,0xa9,0xf4] +@ CHECK: vld3.32 {d16[0], d18[0], d18[0]}, [r10], r5 @ encoding: [0x45,0x0a,0xea,0xf4] +@ CHECK: vld3.8 {d6[6], d7[6], d7[6]}, [r8]! @ encoding: [0xcd,0x62,0xa8,0xf4] +@ CHECK: vld3.16 {d9[2], d10[2], d10[2]}, [r7]! @ encoding: [0x8d,0x96,0xa7,0xf4] +@ CHECK: vld3.32 {d1[1], d2[1], d2[1]}, [r6]! @ encoding: [0x8d,0x1a,0xa6,0xf4] +@ CHECK: vld3.16 {d20[2], d21[2], d21[2]}, [r5]! @ encoding: [0xad,0x46,0xe5,0xf4] +@ CHECK: vld3.32 {d5[0], d7[0], d7[0]}, [r4]! @ encoding: [0x4d,0x5a,0xa4,0xf4] @ vst4.8 {d16[1], d17[1], d18[1], d19[1]}, [r0, :32]