//===----------------------------------------------------------------------===// // Vector Instructions //===----------------------------------------------------------------------===// // Pseudo instructions for VM/VM512 spill/restore // // These pseudo instructions are used for only spill/restore since // InlineSpiller assumes storeRegToStackSlot/loadRegFromStackSlot // functions emit only single instruction. Those functions emit a // single store/load instruction or one of these pseudo store/load // instructions. // // Specifies hasSideEffects = 0 to disable UnmodeledSideEffects. let mayLoad = 1, hasSideEffects = 0 in { def LDVMrii : Pseudo< (outs VM:$vmx), (ins MEMrii:$addr), "# pseudo ldvm $vmx, $addr", []>; def LDVM512rii : Pseudo< (outs VM512:$vmx), (ins MEMrii:$addr), "# pseudo ldvm512 $vmx, $addr", []>; } let mayStore = 1, hasSideEffects = 0 in { def STVMrii : Pseudo< (outs), (ins MEMrii:$addr, VM:$vmx), "# pseudo stvm $addr, $vmx", []>; def STVM512rii : Pseudo< (outs), (ins MEMrii:$addr, VM512:$vmx), "# pseudo stvm512 $addr, $vmx", []>; } //===----------------------------------------------------------------------===// // Pseudo instructions for VM512 modifications //===----------------------------------------------------------------------===// // LVM/SVM instructions using VM512 let hasSideEffects = 0, isCodeGenOnly = 1 in { let Constraints = "$vx = $vd", DisableEncoding = "$vd" in { def LVMyir_y : Pseudo<(outs VM512:$vx), (ins uimm3:$sy, I64:$sz, VM512:$vd), "# pseudo LVM $vx, $sy, $sz, $vd">; def LVMyim_y : Pseudo<(outs VM512:$vx), (ins uimm3:$sy, mimm:$sz, VM512:$vd), "# pseudo LVM $vx, $sy, $sz, $vd">; } def LVMyir : Pseudo<(outs VM512:$vx), (ins uimm3:$sy, I64:$sz), "# pseudo LVM $vx, $sy, $sz">; def LVMyim : Pseudo<(outs VM512:$vx), (ins uimm3:$sy, mimm:$sz), "# pseudo LVM $vx, $sy, $sz">; def SVMyi : Pseudo<(outs I64:$sx), (ins VM512:$vz, uimm3:$sy), "# pseudo SVM $sx, $vz, $sy">; } // VFMK/VFMKW/VFMKS instructions using VM512 let hasSideEffects = 0, isCodeGenOnly = 1, DisableEncoding = "$vl" in { def VFMKyal : Pseudo<(outs VM512:$vmx), (ins I32:$vl), "# pseudo-vfmk.at $vmx">; def VFMKynal : Pseudo<(outs VM512:$vmx), (ins I32:$vl), "# pseudo-vfmk.af $vmx">; def VFMKWyvl : Pseudo<(outs VM512:$vmx), (ins CCOp:$cf, V64:$vz, I32:$vl), "# pseudo-vfmk.w.$cf $vmx, $vz">; def VFMKWyvyl : Pseudo<(outs VM512:$vmx), (ins CCOp:$cf, V64:$vz, VM512:$vm, I32:$vl), "# pseudo-vfmk.w.$cf $vmx, $vz, $vm">; def VFMKSyvl : Pseudo<(outs VM512:$vmx), (ins CCOp:$cf, V64:$vz, I32:$vl), "# pseudo-vfmk.s.$cf $vmx, $vz">; def VFMKSyvyl : Pseudo<(outs VM512:$vmx), (ins CCOp:$cf, V64:$vz, VM512:$vm, I32:$vl), "# pseudo-vfmk.s.$cf $vmx, $vz, $vm">; } // ANDM/ORM/XORM/EQVM/NNDM/NEGM instructions using VM512 let hasSideEffects = 0, isCodeGenOnly = 1 in { def ANDMyy : Pseudo<(outs VM512:$vmx), (ins VM512:$vmy, VM512:$vmz), "# andm $vmx, $vmy, $vmz">; def ORMyy : Pseudo<(outs VM512:$vmx), (ins VM512:$vmy, VM512:$vmz), "# orm $vmx, $vmy, $vmz">; def XORMyy : Pseudo<(outs VM512:$vmx), (ins VM512:$vmy, VM512:$vmz), "# xorm $vmx, $vmy, $vmz">; def EQVMyy : Pseudo<(outs VM512:$vmx), (ins VM512:$vmy, VM512:$vmz), "# eqvm $vmx, $vmy, $vmz">; def NNDMyy : Pseudo<(outs VM512:$vmx), (ins VM512:$vmy, VM512:$vmz), "# nndm $vmx, $vmy, $vmz">; def NEGMy : Pseudo<(outs VM512:$vmx), (ins VM512:$vmy), "# negm $vmx, $vmy">; } //===----------------------------------------------------------------------===// // Instructions // // Define all vector instructions defined in SX-Aurora TSUBASA Architecture // Guide here. As those mnemonics, we use mnemonics defined in Vector Engine // Assembly Language Reference Manual. // // Some instructions can update existing data by following instructions // sequence. // // lea %s0, 256 // lea %s1, 128 // lvl %s0 // vbrd %v0, 2 # v0 = { 2, 2, 2, ..., 2, 2, 2 } // lvl %s1 // vbrd %v0, 3 # v0 = { 3, 3, 3, ..., 3, 2, 2, 2, ..., 2, 2, 2 } // // In order to represent above with a virtual register, we defines instructions // with an additional base register and `_v` suffiex in mnemonic. // // lea t0, 256 // lea t1, 128 // lea t0 // vbrd tv0, 2 // lvl t1 // vbrd_v tv1, 2, tv0 // // We also have some instructions uses VL register with an pseudo VL value // with following suffixes in mnemonic. // // l: have an additional I32 register to represent the VL value. // L: have an additional VL register to represent the VL value. //===----------------------------------------------------------------------===// //----------------------------------------------------------------------------- // Section 8.9 - Vector Load/Store and Move Instructions //----------------------------------------------------------------------------- // Multiclass for VLD instructions let mayLoad = 1, hasSideEffects = 0, Uses = [VL] in multiclass VLDbmopc, RegisterClass RC, dag dag_in, string disEnc = ""> { let DisableEncoding = disEnc in def "" : RVM; let Constraints = "$vx = $base", DisableEncoding = disEnc#"$base", isCodeGenOnly = 1 in def _v : RVM; } multiclass VLDlmopc, RegisterClass RC, dag dag_in> { defm "" : VLDbm; let isCodeGenOnly = 1, VE_VLInUse = 1 in { defm l : VLDbm; defm L : VLDbm; } } let VE_VLIndex = 3 in multiclass VLDtgmopc, RegisterClass RC> { defm rr : VLDlm; let cy = 0 in defm ir : VLDlm; let cz = 0 in defm rz : VLDlm; let cy = 0, cz = 0 in defm iz : VLDlm; } multiclass VLDmopc, RegisterClass RC> { let vc = 1 in defm "" : VLDtgm; let vc = 0 in defm NC : VLDtgm; } // Section 8.9.1 - VLD (Vector Load) defm VLD : VLDm<"vld", 0x81, V64>; // Section 8.9.2 - VLDU (Vector Load Upper) defm VLDU : VLDm<"vldu", 0x82, V64>; // Section 8.9.3 - VLDL (Vector Load Lower) defm VLDLSX : VLDm<"vldl.sx", 0x83, V64>; let cx = 1 in defm VLDLZX : VLDm<"vldl.zx", 0x83, V64>; // Section 8.9.4 - VLD2D (Vector Load 2D) defm VLD2D : VLDm<"vld2d", 0xc1, V64>; // Section 8.9.5 - VLDU2D (Vector Load Upper 2D) defm VLDU2D : VLDm<"vldu2d", 0xc2, V64>; // Section 8.9.6 - VLDL2D (Vector Load Lower 2D) defm VLDL2DSX : VLDm<"vldl2d.sx", 0xc3, V64>; let cx = 1 in defm VLDL2DZX : VLDm<"vldl2d.zx", 0xc3, V64>; // Multiclass for VST instructions let mayStore = 1, hasSideEffects = 0, Uses = [VL] in multiclass VSTbmopc, dag dag_in> { def "" : RVM; let DisableEncoding = "$vl", isCodeGenOnly = 1, VE_VLInUse = 1 in { def l : RVM; def L : RVM; } } multiclass VSTmmopc, dag dag_in> { defm "" : VSTbm; let m = ?, VE_VLWithMask = 1 in defm m : VSTbm; } let VE_VLIndex = 3 in multiclass VSTtgmopc, RegisterClass RC> { defm rrv : VSTmm; let cy = 0 in defm irv : VSTmm; let cz = 0 in defm rzv : VSTmm; let cy = 0, cz = 0 in defm izv : VSTmm; } multiclass VSTmopc, RegisterClass RC> { let vc = 1, cx = 0 in defm "" : VSTtgm; let vc = 0, cx = 0 in defm NC : VSTtgm; let vc = 1, cx = 1 in defm OT : VSTtgm; let vc = 0, cx = 1 in defm NCOT : VSTtgm; } // Section 8.9.7 - VST (Vector Store) defm VST : VSTm<"vst", 0x91, V64>; // Section 8.9.8 - VST (Vector Store Upper) defm VSTU : VSTm<"vstu", 0x92, V64>; // Section 8.9.9 - VSTL (Vector Store Lower) defm VSTL : VSTm<"vstl", 0x93, V64>; // Section 8.9.10 - VST2D (Vector Store 2D) defm VST2D : VSTm<"vst2d", 0xd1, V64>; // Section 8.9.11 - VSTU2D (Vector Store Upper 2D) defm VSTU2D : VSTm<"vstu2d", 0xd2, V64>; // Section 8.9.12 - VSTL2D (Vector Store Lower 2D) defm VSTL2D : VSTm<"vstl2d", 0xd3, V64>; // Multiclass for VGT instructions let mayLoad = 1, hasSideEffects = 0, Uses = [VL] in multiclass VGTbmopc, RegisterClass RC, dag dag_in, string disEnc = ""> { let DisableEncoding = disEnc in def "" : RVM; let Constraints = "$vx = $base", DisableEncoding = disEnc#"$base", isCodeGenOnly = 1 in def _v : RVM; } multiclass VGTlmopc, RegisterClass RC, dag dag_in> { defm "" : VGTbm; let isCodeGenOnly = 1, VE_VLInUse = 1 in { defm l : VGTbm; defm L : VGTbm; } } multiclass VGTmmopc, RegisterClass RC, dag dag_in> { defm "" : VGTlm; let m = ?, VE_VLWithMask = 1 in defm m : VGTlm; } let VE_VLIndex = 4 in multiclass VGTlhmopc, RegisterClass RC, dag dag_in> { defm rr : VGTmm; let cy = 0 in defm ir : VGTmm; let cz = 0 in defm rz : VGTmm; let cy = 0, cz = 0 in defm iz : VGTmm; } multiclass VGTtgmopc, RegisterClass RC> { let vy = ? in defm v : VGTlhm; let cs = 1, sw = ? in defm s : VGTlhm; } multiclass VGTmopc, RegisterClass RC> { let vc = 1 in defm "" : VGTtgm; let vc = 0 in defm NC : VGTtgm; } // Section 8.9.13 - VGT (Vector Gather) defm VGT : VGTm<"vgt", 0xa1, V64>; // Section 8.9.14 - VGTU (Vector Gather Upper) defm VGTU : VGTm<"vgtu", 0xa2, V64>; // Section 8.9.15 - VGTL (Vector Gather Lower) defm VGTLSX : VGTm<"vgtl.sx", 0xa3, V64>; let cx = 1 in defm VGTLZX : VGTm<"vgtl.zx", 0xa3, V64>; def : MnemonicAlias<"vgtl", "vgtl.zx">; def : MnemonicAlias<"vgtl.nc", "vgtl.zx.nc">; // Multiclass for VSC instructions let mayStore = 1, hasSideEffects = 0, Uses = [VL] in multiclass VSCbmopc, dag dag_in> { def "" : RVM; let DisableEncoding = "$vl", isCodeGenOnly = 1, VE_VLInUse = 1 in { def l : RVM; def L : RVM; } } multiclass VSCmmopc, dag dag_in> { defm "" : VSCbm; let m = ?, VE_VLWithMask = 1 in defm m : VSCbm; } let VE_VLIndex = 4 in multiclass VSClhmopc, RegisterClass RC, dag dag_in> { defm rrv : VSCmm; let cy = 0 in defm irv : VSCmm; let cz = 0 in defm rzv : VSCmm; let cy = 0, cz = 0 in defm izv : VSCmm; } multiclass VSCtgmopc, RegisterClass RC> { let vy = ? in defm v : VSClhm; let cs = 1, sw = ? in defm s : VSClhm; } multiclass VSCmopc, RegisterClass RC> { let vc = 1, cx = 0 in defm "" : VSCtgm; let vc = 0, cx = 0 in defm NC : VSCtgm; let vc = 1, cx = 1 in defm OT : VSCtgm; let vc = 0, cx = 1 in defm NCOT : VSCtgm; } // Section 8.9.16 - VSC (Vector Scatter) defm VSC : VSCm<"vsc", 0xb1, V64>; // Section 8.9.17 - VSCU (Vector Scatter Upper) defm VSCU : VSCm<"vscu", 0xb2, V64>; // Section 8.9.18 - VSCL (Vector Scatter Lower) defm VSCL : VSCm<"vscl", 0xb3, V64>; // Section 8.9.19 - PFCHV (Prefetch Vector) let Uses = [VL] in multiclass PFCHVbmopc, dag dag_in> { def "" : RVM; let DisableEncoding = "$vl", isCodeGenOnly = 1, VE_VLInUse = 1 in { def l : RVM; def L : RVM; } } let VE_VLIndex = 2 in multiclass PFCHVmopc> { defm rr : PFCHVbm; let cy = 0 in defm ir : PFCHVbm; let cz = 0 in defm rz : PFCHVbm; let cy = 0, cz = 0 in defm iz : PFCHVbm; } let vc = 1, vx = 0 in defm PFCHV : PFCHVm<"pfchv", 0x80>; let vc = 0, vx = 0 in defm PFCHVNC : PFCHVm<"pfchv.nc", 0x80>; // Section 8.9.20 - LSV (Load S to V) let sx = 0, vx = ?, hasSideEffects = 0 in multiclass LSVbmopc, RegisterClass RC, dag dag_in> { def "" : RR; let Constraints = "$vx = $base", DisableEncoding = "$base", isCodeGenOnly = 1 in def _v : RR; } multiclass LSVmopc, RegisterClass RC> { defm rr : LSVbm; let cy = 0 in defm ir : LSVbm; let cz = 0 in defm rm : LSVbm; let cy = 0, cz = 0 in defm im : LSVbm; } defm LSV : LSVm<"lsv", 0x8e, V64>; // Section 8.9.21 - LVS (Load V to S) let cz = 0, sz = 0, vx = ?, hasSideEffects = 0 in multiclass LVSmopc, RegisterClass RC> { def vr : RR; let cy = 0 in def vi : RR; } defm LVS : LVSm<"lvs", 0x9e, V64>; // Section 8.9.22 - LVM (Load VM) let sx = 0, vx = ?, hasSideEffects = 0 in multiclass LVMbmopc, RegisterClass RCM, dag dag_in> { def "" : RR; let Constraints = "$vx = $base", DisableEncoding = "$base", isCodeGenOnly = 1 in { def _m : RR; } } multiclass LVMomopc, RegisterClass RCM> { defm rr : LVMbm; let cy = 0 in defm ir : LVMbm; let cz = 0 in defm rm : LVMbm; let cy = 0, cz = 0 in defm im : LVMbm; } multiclass LVMmopc, RegisterClass RCM> { defm "" : LVMom; } defm LVM : LVMm<"lvm", 0xb7, VM>; // Section 8.9.23 - SVM (Save VM) let cz = 0, sz = 0, vz = ?, hasSideEffects = 0 in multiclass SVMmopc, RegisterClass RCM> { def mr : RR; let cy = 0 in def mi : RR; } defm SVM : SVMm<"svm", 0xa7, VM>; // Section 8.9.24 - VBRD (Vector Broadcast) let vx = ?, hasSideEffects = 0, Uses = [VL] in multiclass VBRDbmopc, RegisterClass RC, dag dag_in, string disEnc = ""> { let DisableEncoding = disEnc in def "" : RV; let Constraints = "$vx = $base", DisableEncoding = disEnc#"$base", isCodeGenOnly = 1 in def _v : RV; } multiclass VBRDlmopc, RegisterClass RC, dag dag_in> { defm "" : VBRDbm; let isCodeGenOnly = 1, VE_VLInUse = 1 in { defm l : VBRDbm; defm L : VBRDbm; } } multiclass VBRDmmopc, RegisterClass RC, RegisterClass RCM, dag dag_in> { defm "" : VBRDlm; let m = ?, VE_VLWithMask = 1 in defm m : VBRDlm; } let VE_VLIndex = 2 in multiclass VBRDmopc, RegisterClass VRC, RegisterClass RC, RegisterClass RCM> { defm r : VBRDmm; let cy = 0 in defm i : VBRDmm; } let cx = 0, cx2 = 0 in defm VBRD : VBRDm<"vbrd", 0x8c, V64, I64, VM>; let cx = 0, cx2 = 1 in defm VBRDL : VBRDm<"vbrdl", 0x8c, V64, I32, VM>; let cx = 1, cx2 = 0 in defm VBRDU : VBRDm<"vbrdu", 0x8c, V64, F32, VM>; let cx = 1, cx2 = 1 in defm PVBRD : VBRDm<"pvbrd", 0x8c, V64, I64, VM512>; // Section 8.9.25 - VMV (Vector Move) let vx = ?, vz = ?, hasSideEffects = 0, Uses = [VL] in multiclass VMVbmopc, RegisterClass RC, dag dag_in, string disEnc = ""> { let DisableEncoding = disEnc in def "" : RV; let Constraints = "$vx = $base", DisableEncoding = disEnc#"$base", isCodeGenOnly = 1 in def _v : RV; } multiclass VMVlmopc, RegisterClass RC, dag dag_in> { defm "" : VMVbm; let isCodeGenOnly = 1, VE_VLInUse = 1 in { defm l : VMVbm; defm L : VMVbm; } } multiclass VMVmmopc, RegisterClass RC, RegisterClass RCM, dag dag_in> { defm "" : VMVlm; let m = ?, VE_VLWithMask = 1 in defm m : VMVlm; } let VE_VLIndex = 3 in multiclass VMVmopc, RegisterClass RC, RegisterClass RCM> { defm rv : VMVmm; let cy = 0 in defm iv : VMVmm; } defm VMV : VMVm<"vmv", 0x9c, V64, VM>; //----------------------------------------------------------------------------- // Section 8.10 - Vector Fixed-Point Arithmetic Instructions //----------------------------------------------------------------------------- // Multiclass for generic vector calculation let vx = ?, hasSideEffects = 0, Uses = [VL] in multiclass RVbmopc, RegisterClass RC, dag dag_in, string disEnc = ""> { let DisableEncoding = disEnc in def "" : RV; let Constraints = "$vx = $base", DisableEncoding = disEnc#"$base", isCodeGenOnly = 1 in def _v : RV; } multiclass RVlmopc, RegisterClass RC, dag dag_in> { defm "" : RVbm; let isCodeGenOnly = 1, VE_VLInUse = 1 in { defm l : RVbm; defm L : RVbm; } } multiclass RVmmopc, RegisterClass RC, RegisterClass RCM, dag dag_in> { defm "" : RVlm; let m = ?, VE_VLWithMask = 1 in defm m : RVlm; } // Generic RV multiclass with 2 arguments. // e.g. VADD, VSUB, VMPY, and etc. let VE_VLIndex = 3 in multiclass RVmopc, RegisterClass VRC, RegisterClass RC, RegisterClass RCM, Operand SIMM = simm7> { let cy = 0, sy = 0, vy = ?, vz = ? in defm vv : RVmm; let cs = 1, vz = ? in defm rv : RVmm; let cs = 1, cy = 0, vz = ? in defm iv : RVmm; } // Special RV multiclass with 2 arguments using cs2. // e.g. VDIV, VDVS, and VDVX. let VE_VLIndex = 3 in multiclass RVDIVmopc, RegisterClass VRC, RegisterClass RC, RegisterClass RCM, Operand SIMM = simm7> { let cy = 0, sy = 0, vy = ?, vz = ? in defm vv : RVmm; let cs2 = 1, vy = ? in defm vr : RVmm; let cs2 = 1, cy = 0, vy = ? in defm vi : RVmm; let cs = 1, vz = ? in defm rv : RVmm; let cs = 1, cy = 0, vz = ? in defm iv : RVmm; } // Generic RV multiclass with 2 arguments for logical operations. // e.g. VAND, VOR, VXOR, and etc. let VE_VLIndex = 3 in multiclass RVLmopc, RegisterClass ScaRC, RegisterClass RC, RegisterClass RCM> { let cy = 0, sy = 0, vy = ?, vz = ? in defm vv : RVmm; let cs = 1, vz = ? in defm rv : RVmm; let cs = 1, cy = 0, vz = ? in defm mv : RVmm; } // Generic RV multiclass with 1 argument. // e.g. VLDZ, VPCNT, and VBRV. let VE_VLIndex = 2 in multiclass RV1mopc, RegisterClass RC, RegisterClass RCM> { let cy = 0, sy = 0, vz = ? in defm v : RVmm; } // Generic RV multiclass with no argument. // e.g. VSEQ. let VE_VLIndex = 1 in multiclass RV0mopc, RegisterClass RC, RegisterClass RCM> { let cy = 0, sy = 0 in defm "" : RVmm; } // Generic RV multiclass with 2 arguments for shift operations. // e.g. VSLL, VSRL, VSLA, and etc. let VE_VLIndex = 3 in multiclass RVSmopc, RegisterClass ScaRC, RegisterClass RC, RegisterClass RCM> { let cy = 0, sy = 0, vy = ?, vz = ? in defm vv : RVmm; let cs = 1, vz = ? in defm vr : RVmm; let cs = 1, cy = 0, vz = ? in defm vi : RVmm; } // Generic RV multiclass with 3 arguments for shift operations. // e.g. VSLD and VSRD. let VE_VLIndex = 4 in multiclass RVSDmopc, RegisterClass RC, RegisterClass RCM> { let vy = ?, vz = ? in defm vvr : RVmm; let cy = 0, vy = ?, vz = ? in defm vvi : RVmm; } // Special RV multiclass with 3 arguments. // e.g. VSFA let VE_VLIndex = 4 in multiclass RVSAmopc, RegisterClass RC, RegisterClass RCM> { let cz = 1, sz = ?, vz = ? in defm vrr : RVmm; let cz = 0, sz = ?, vz = ? in defm vrm : RVmm; let cy = 0, cz = 1, sz = ?, vz = ? in defm vir : RVmm; let cy = 0, cz = 0, sz = ?, vz = ? in defm vim : RVmm; } // Generic RV multiclass with 1 argument using vy field. // e.g. VFSQRT, VRCP, and VRSQRT. let VE_VLIndex = 2 in multiclass RVF1mopc, RegisterClass RC, RegisterClass RCM> { let cy = 0, sy = 0, vy = ? in defm v : RVmm; } // Special RV multiclass with 3 arguments using cs2. // e.g. VFMAD, VFMSB, VFNMAD, and etc. let VE_VLIndex = 4 in multiclass RVMmopc, RegisterClass VRC, RegisterClass RC, RegisterClass RCM, Operand SIMM = simm7> { let cy = 0, sy = 0, vy = ?, vz = ?, vw = ? in defm vvv : RVmm; let cs2 = 1, vy = ?, vw = ? in defm vrv : RVmm; let cs2 = 1, cy = 0, vy = ?, vw = ? in defm viv : RVmm; let cs = 1, vz = ?, vw = ? in defm rvv : RVmm; let cs = 1, cy = 0, vz = ?, vw = ? in defm ivv : RVmm; } // Special RV multiclass with 2 arguments for floating point conversions. // e.g. VFIX and VFIXX let hasSideEffects = 0, VE_VLIndex = 3 in multiclass RVFIXm opc, RegisterClass RC, RegisterClass RCM> { let cy = 0, sy = 0, vy = ?, vz = ? in defm v : RVmm; } // Multiclass for generic iterative vector calculation let vx = ?, hasSideEffects = 0, Uses = [VL] in multiclass RVIbmopc, RegisterClass RC, dag dag_in, string disEnc = ""> { let DisableEncoding = disEnc in def "" : RV; let isCodeGenOnly = 1, Constraints = "$vx = $base", DisableEncoding = disEnc#"$base" in def _v : RV; } multiclass RVIlmopc, RegisterClass RC, dag dag_in> { defm "" : RVIbm; let isCodeGenOnly = 1, VE_VLInUse = 1 in { defm l : RVIbm; defm L : RVIbm; } } // Generic RV multiclass for iterative operation with 2 argument. // e.g. VFIA, VFIS, and VFIM let VE_VLIndex = 3 in multiclass RVI2mopc, RegisterClass VRC, RegisterClass RC> { let vy = ? in defm vr : RVIlm; let cy = 0, vy = ? in defm vi : RVIlm; } // Generic RV multiclass for iterative operation with 3 argument. // e.g. VFIAM, VFISM, VFIMA, and etc. let VE_VLIndex = 4 in multiclass RVI3mopc, RegisterClass VRC, RegisterClass RC> { let vy = ?, vz = ? in defm vvr : RVIlm; let cy = 0, vy = ?, vz = ? in defm vvi : RVIlm; } // special RV multiclass with 3 arguments for VSHF. // e.g. VSHF let vy = ?, vz = ?, VE_VLIndex = 4 in multiclass RVSHFmopc, RegisterClass RC, Operand SIMM = uimm4> { defm vvr : RVlm; let cy = 0 in defm vvi : RVlm; } // Multiclass for generic mask calculation let vx = ?, hasSideEffects = 0, Uses = [VL] in multiclass RVMKbmopc, dag dag_out, dag dag_in> { def "" : RV; let DisableEncoding = "$vl", isCodeGenOnly = 1, VE_VLInUse = 1 in { def l : RV; def L : RV; } } multiclass RVMKlmopc, RegisterClass RCM, dag dag_in> { defm "" : RVMKbm; let m = ?, VE_VLWithMask = 1 in defm m : RVMKbm; } // Generic RV multiclass for mask calculation with a condition. // e.g. VFMK, VFMS, and VFMF let cy = 0, sy = 0 in multiclass RVMKom opc, RegisterClass RC, RegisterClass RCM> { let vy = ?, vz = ?, VE_VLIndex = 3 in defm v : RVMKlm; let vy = 15 /* AT */, VE_VLIndex = 1 in defm a : RVMKlm; let vy = 0 /* AF */, VE_VLIndex = 1 in defm na : RVMKlm; } multiclass RVMKm opc, RegisterClass RC, RegisterClass RCM> { defm "" : RVMKom; } // Generic RV multiclass for mask calculation with 2 arguments. // e.g. ANDM, ORM, XORM, and etc. let cy = 0, sy = 0, vx = ?, vy = ?, vz = ?, hasSideEffects = 0 in multiclass RVM2m opc, RegisterClass RCM> { def mm : RV; } // Generic RV multiclass for mask calculation with 1 argument. // e.g. NEGM let cy = 0, sy = 0, vx = ?, vy = ?, hasSideEffects = 0 in multiclass RVM1m opc, RegisterClass RCM> { def m : RV; } // Generic RV multiclass for mask calculation with 1 argument. // e.g. PCVM, LZVM, and TOVM let cy = 0, sy = 0, vy = ?, hasSideEffects = 0, Uses = [VL] in multiclass RVMSbmopc, dag dag_in> { def "" : RV { bits<7> sx; let Inst{54-48} = sx; } let DisableEncoding = "$vl", isCodeGenOnly = 1, VE_VLInUse = 1 in { def l : RV { bits<7> sx; let Inst{54-48} = sx; } def L : RV { bits<7> sx; let Inst{54-48} = sx; } } } let VE_VLIndex = 2 in multiclass RVMSm opc, RegisterClass RCM> { defm m : RVMSbm; } // Section 8.10.1 - VADD (Vector Add) let cx = 0, cx2 = 0 in defm VADDUL : RVm<"vaddu.l", 0xc8, V64, I64, VM>; let cx = 0, cx2 = 1 in { defm PVADDULO : RVm<"pvaddu.lo", 0xc8, V64, I32, VM>; let isCodeGenOnly = 1 in defm VADDUW : RVm<"vaddu.w", 0xc8, V64, I32, VM>; } let cx = 1, cx2 = 0 in defm PVADDUUP : RVm<"pvaddu.up", 0xc8, V64, I64, VM>; let cx = 1, cx2 = 1 in defm PVADDU : RVm<"pvaddu", 0xc8, V64, I64, VM512>; def : MnemonicAlias<"vaddu.w", "pvaddu.lo">; // Section 8.10.2 - VADS (Vector Add Single) let cx = 0, cx2 = 0 in defm VADDSWSX : RVm<"vadds.w.sx", 0xca, V64, I32, VM>; let cx = 0, cx2 = 1 in { defm PVADDSLO : RVm<"pvadds.lo", 0xca, V64, I32, VM>; let isCodeGenOnly = 1 in defm VADDSWZX : RVm<"vadds.w.zx", 0xca, V64, I32, VM>; } let cx = 1, cx2 = 0 in defm PVADDSUP : RVm<"pvadds.up", 0xca, V64, I64, VM>; let cx = 1, cx2 = 1 in defm PVADDS : RVm<"pvadds", 0xca, V64, I64, VM512>; def : MnemonicAlias<"pvadds.lo.sx", "vadds.w.sx">; def : MnemonicAlias<"vadds.w.zx", "pvadds.lo">; def : MnemonicAlias<"vadds.w", "pvadds.lo">; def : MnemonicAlias<"pvadds.lo.zx", "pvadds.lo">; // Section 8.10.3 - VADX (Vector Add) defm VADDSL : RVm<"vadds.l", 0x8b, V64, I64, VM>; // Section 8.10.4 - VSUB (Vector Subtract) let cx = 0, cx2 = 0 in defm VSUBUL : RVm<"vsubu.l", 0xd8, V64, I64, VM>; let cx = 0, cx2 = 1 in { defm PVSUBULO : RVm<"pvsubu.lo", 0xd8, V64, I32, VM>; let isCodeGenOnly = 1 in defm VSUBUW : RVm<"vsubu.w", 0xd8, V64, I32, VM>; } let cx = 1, cx2 = 0 in defm PVSUBUUP : RVm<"pvsubu.up", 0xd8, V64, I64, VM>; let cx = 1, cx2 = 1 in defm PVSUBU : RVm<"pvsubu", 0xd8, V64, I64, VM512>; def : MnemonicAlias<"vsubu.w", "pvsubu.lo">; // Section 8.10.5 - VSBS (Vector Subtract Single) let cx = 0, cx2 = 0 in defm VSUBSWSX : RVm<"vsubs.w.sx", 0xda, V64, I32, VM>; let cx = 0, cx2 = 1 in { defm PVSUBSLO : RVm<"pvsubs.lo", 0xda, V64, I32, VM>; let isCodeGenOnly = 1 in defm VSUBSWZX : RVm<"vsubs.w.zx", 0xda, V64, I32, VM>; } let cx = 1, cx2 = 0 in defm PVSUBSUP : RVm<"pvsubs.up", 0xda, V64, I64, VM>; let cx = 1, cx2 = 1 in defm PVSUBS : RVm<"pvsubs", 0xda, V64, I64, VM512>; def : MnemonicAlias<"pvsubs.lo.sx", "vsubs.w.sx">; def : MnemonicAlias<"vsubs.w.zx", "pvsubs.lo">; def : MnemonicAlias<"vsubs.w", "pvsubs.lo">; def : MnemonicAlias<"pvsubs.lo.zx", "pvsubs.lo">; // Section 8.10.6 - VSBX (Vector Subtract) defm VSUBSL : RVm<"vsubs.l", 0x9b, V64, I64, VM>; // Section 8.10.7 - VMPY (Vector Multiply) let cx2 = 0 in defm VMULUL : RVm<"vmulu.l", 0xc9, V64, I64, VM>; let cx2 = 1 in defm VMULUW : RVm<"vmulu.w", 0xc9, V64, I32, VM>; // Section 8.10.8 - VMPS (Vector Multiply Single) let cx2 = 0 in defm VMULSWSX : RVm<"vmuls.w.sx", 0xcb, V64, I32, VM>; let cx2 = 1 in defm VMULSWZX : RVm<"vmuls.w.zx", 0xcb, V64, I32, VM>; def : MnemonicAlias<"vmuls.w", "vmuls.w.zx">; // Section 8.10.9 - VMPX (Vector Multiply) defm VMULSL : RVm<"vmuls.l", 0xdb, V64, I64, VM>; // Section 8.10.10 - VMPD (Vector Multiply) defm VMULSLW : RVm<"vmuls.l.w", 0xd9, V64, I32, VM>; // Section 8.10.11 - VDIV (Vector Divide) let cx2 = 0 in defm VDIVUL : RVDIVm<"vdivu.l", 0xe9, V64, I64, VM>; let cx2 = 1 in defm VDIVUW : RVDIVm<"vdivu.w", 0xe9, V64, I32, VM>; // Section 8.10.12 - VDVS (Vector Divide Single) let cx2 = 0 in defm VDIVSWSX : RVDIVm<"vdivs.w.sx", 0xeb, V64, I32, VM>; let cx2 = 1 in defm VDIVSWZX : RVDIVm<"vdivs.w.zx", 0xeb, V64, I32, VM>; def : MnemonicAlias<"vdivs.w", "vdivs.w.zx">; // Section 8.10.13 - VDVX (Vector Divide) defm VDIVSL : RVDIVm<"vdivs.l", 0xfb, V64, I64, VM>; // Section 8.10.14 - VCMP (Vector Compare) let cx = 0, cx2 = 0 in defm VCMPUL : RVm<"vcmpu.l", 0xb9, V64, I64, VM>; let cx = 0, cx2 = 1 in { defm PVCMPULO : RVm<"pvcmpu.lo", 0xb9, V64, I32, VM>; let isCodeGenOnly = 1 in defm VCMPUW : RVm<"vcmpu.w", 0xb9, V64, I32, VM>; } let cx = 1, cx2 = 0 in defm PVCMPUUP : RVm<"pvcmpu.up", 0xb9, V64, I64, VM>; let cx = 1, cx2 = 1 in defm PVCMPU : RVm<"pvcmpu", 0xb9, V64, I64, VM512>; def : MnemonicAlias<"vcmpu.w", "pvcmpu.lo">; // Section 8.10.15 - VCPS (Vector Compare Single) let cx = 0, cx2 = 0 in defm VCMPSWSX : RVm<"vcmps.w.sx", 0xfa, V64, I32, VM>; let cx = 0, cx2 = 1 in { defm PVCMPSLO : RVm<"pvcmps.lo", 0xfa, V64, I32, VM>; let isCodeGenOnly = 1 in defm VCMPSWZX : RVm<"vcmps.w.zx", 0xfa, V64, I32, VM>; } let cx = 1, cx2 = 0 in defm PVCMPSUP : RVm<"pvcmps.up", 0xfa, V64, I64, VM>; let cx = 1, cx2 = 1 in defm PVCMPS : RVm<"pvcmps", 0xfa, V64, I64, VM512>; def : MnemonicAlias<"pvcmps.lo.sx", "vcmps.w.sx">; def : MnemonicAlias<"vcmps.w.zx", "pvcmps.lo">; def : MnemonicAlias<"vcmps.w", "pvcmps.lo">; def : MnemonicAlias<"pvcmps.lo.zx", "pvcmps.lo">; // Section 8.10.16 - VCPX (Vector Compare) defm VCMPSL : RVm<"vcmps.l", 0xba, V64, I64, VM>; // Section 8.10.17 - VCMS (Vector Compare and Select Maximum/Minimum Single) let cx = 0, cx2 = 0 in defm VMAXSWSX : RVm<"vmaxs.w.sx", 0x8a, V64, I32, VM>; let cx = 0, cx2 = 1 in { defm PVMAXSLO : RVm<"pvmaxs.lo", 0x8a, V64, I32, VM>; let isCodeGenOnly = 1 in defm VMAXSWZX : RVm<"vmaxs.w.zx", 0x8a, V64, I32, VM>; } let cx = 1, cx2 = 0 in defm PVMAXSUP : RVm<"pvmaxs.up", 0x8a, V64, I64, VM>; let cx = 1, cx2 = 1 in defm PVMAXS : RVm<"pvmaxs", 0x8a, V64, I64, VM512>; let cs2 = 1 in { let cx = 0, cx2 = 0 in defm VMINSWSX : RVm<"vmins.w.sx", 0x8a, V64, I32, VM>; let cx = 0, cx2 = 1 in { defm PVMINSLO : RVm<"pvmins.lo", 0x8a, V64, I32, VM>; let isCodeGenOnly = 1 in defm VMINSWZX : RVm<"vmins.w.zx", 0x8a, V64, I32, VM>; } let cx = 1, cx2 = 0 in defm PVMINSUP : RVm<"pvmins.up", 0x8a, V64, I64, VM>; let cx = 1, cx2 = 1 in defm PVMINS : RVm<"pvmins", 0x8a, V64, I64, VM512>; } def : MnemonicAlias<"pvmaxs.lo.sx", "vmaxs.w.sx">; def : MnemonicAlias<"vmaxs.w.zx", "pvmaxs.lo">; def : MnemonicAlias<"vmaxs.w", "pvmaxs.lo">; def : MnemonicAlias<"pvmaxs.lo.zx", "pvmaxs.lo">; def : MnemonicAlias<"pvmins.lo.sx", "vmins.w.sx">; def : MnemonicAlias<"vmins.w.zx", "pvmins.lo">; def : MnemonicAlias<"vmins.w", "pvmins.lo">; def : MnemonicAlias<"pvmins.lo.zx", "pvmins.lo">; // Section 8.10.18 - VCMX (Vector Compare and Select Maximum/Minimum) defm VMAXSL : RVm<"vmaxs.l", 0x9a, V64, I64, VM>; let cs2 = 1 in defm VMINSL : RVm<"vmins.l", 0x9a, V64, I64, VM>; //----------------------------------------------------------------------------- // Section 8.11 - Vector Logical Operation Instructions //----------------------------------------------------------------------------- // Section 8.11.1 - VAND (Vector And) let cx = 0, cx2 = 0 in defm VAND : RVLm<"vand", 0xc4, I64, V64, VM>; let cx = 0, cx2 = 1 in defm PVANDLO : RVLm<"pvand.lo", 0xc4, I32, V64, VM>; let cx = 1, cx2 = 0 in defm PVANDUP : RVLm<"pvand.up", 0xc4, F32, V64, VM>; let cx = 1, cx2 = 1 in defm PVAND : RVLm<"pvand", 0xc4, I64, V64, VM512>; // Section 8.11.2 - VOR (Vector Or) let cx = 0, cx2 = 0 in defm VOR : RVLm<"vor", 0xc5, I64, V64, VM>; let cx = 0, cx2 = 1 in defm PVORLO : RVLm<"pvor.lo", 0xc5, I32, V64, VM>; let cx = 1, cx2 = 0 in defm PVORUP : RVLm<"pvor.up", 0xc5, F32, V64, VM>; let cx = 1, cx2 = 1 in defm PVOR : RVLm<"pvor", 0xc5, I64, V64, VM512>; // Section 8.11.3 - VXOR (Vector Exclusive Or) let cx = 0, cx2 = 0 in defm VXOR : RVLm<"vxor", 0xc6, I64, V64, VM>; let cx = 0, cx2 = 1 in defm PVXORLO : RVLm<"pvxor.lo", 0xc6, I32, V64, VM>; let cx = 1, cx2 = 0 in defm PVXORUP : RVLm<"pvxor.up", 0xc6, F32, V64, VM>; let cx = 1, cx2 = 1 in defm PVXOR : RVLm<"pvxor", 0xc6, I64, V64, VM512>; // Section 8.11.4 - VEQV (Vector Equivalence) let cx = 0, cx2 = 0 in defm VEQV : RVLm<"veqv", 0xc7, I64, V64, VM>; let cx = 0, cx2 = 1 in defm PVEQVLO : RVLm<"pveqv.lo", 0xc7, I32, V64, VM>; let cx = 1, cx2 = 0 in defm PVEQVUP : RVLm<"pveqv.up", 0xc7, F32, V64, VM>; let cx = 1, cx2 = 1 in defm PVEQV : RVLm<"pveqv", 0xc7, I64, V64, VM512>; // Section 8.11.5 - VLDZ (Vector Leading Zero Count) let cx = 0, cx2 = 0 in defm VLDZ : RV1m<"vldz", 0xe7, V64, VM>; let cx = 0, cx2 = 1 in defm PVLDZLO : RV1m<"pvldz.lo", 0xe7, V64, VM>; let cx = 1, cx2 = 0 in defm PVLDZUP : RV1m<"pvldz.up", 0xe7, V64, VM>; let cx = 1, cx2 = 1 in defm PVLDZ : RV1m<"pvldz", 0xe7, V64, VM512>; // Section 8.11.6 - VPCNT (Vector Population Count) let cx = 0, cx2 = 0 in defm VPCNT : RV1m<"vpcnt", 0xac, V64, VM>; let cx = 0, cx2 = 1 in defm PVPCNTLO : RV1m<"pvpcnt.lo", 0xac, V64, VM>; let cx = 1, cx2 = 0 in defm PVPCNTUP : RV1m<"pvpcnt.up", 0xac, V64, VM>; let cx = 1, cx2 = 1 in defm PVPCNT : RV1m<"pvpcnt", 0xac, V64, VM512>; // Section 8.11.7 - VBRV (Vector Bit Reverse) let cx = 0, cx2 = 0 in defm VBRV : RV1m<"vbrv", 0xf7, V64, VM>; let cx = 0, cx2 = 1 in defm PVBRVLO : RV1m<"pvbrv.lo", 0xf7, V64, VM>; let cx = 1, cx2 = 0 in defm PVBRVUP : RV1m<"pvbrv.up", 0xf7, V64, VM>; let cx = 1, cx2 = 1 in defm PVBRV : RV1m<"pvbrv", 0xf7, V64, VM512>; // Section 8.11.8 - VSEQ (Vector Sequential Number) let cx = 0, cx2 = 0 in defm VSEQ : RV0m<"vseq", 0x99, V64, VM>; let cx = 0, cx2 = 1 in defm PVSEQLO : RV0m<"pvseq.lo", 0x99, V64, VM>; let cx = 1, cx2 = 0 in defm PVSEQUP : RV0m<"pvseq.up", 0x99, V64, VM>; let cx = 1, cx2 = 1 in defm PVSEQ : RV0m<"pvseq", 0x99, V64, VM512>; //----------------------------------------------------------------------------- // Section 8.12 - Vector Shift Operation Instructions //----------------------------------------------------------------------------- // Section 8.12.1 - VSLL (Vector Shift Left Logical) let cx = 0, cx2 = 0 in defm VSLL : RVSm<"vsll", 0xe5, I64, V64, VM>; let cx = 0, cx2 = 1 in defm PVSLLLO : RVSm<"pvsll.lo", 0xe5, I32, V64, VM>; let cx = 1, cx2 = 0 in defm PVSLLUP : RVSm<"pvsll.up", 0xe5, F32, V64, VM>; let cx = 1, cx2 = 1 in defm PVSLL : RVSm<"pvsll", 0xe5, I64, V64, VM512>; // Section 8.12.2 - VSLD (Vector Shift Left Double) defm VSLD : RVSDm<"vsld", 0xe4, V64, VM>; // Section 8.12.3 - VSRL (Vector Shift Right Logical) let cx = 0, cx2 = 0 in defm VSRL : RVSm<"vsrl", 0xf5, I64, V64, VM>; let cx = 0, cx2 = 1 in defm PVSRLLO : RVSm<"pvsrl.lo", 0xf5, I32, V64, VM>; let cx = 1, cx2 = 0 in defm PVSRLUP : RVSm<"pvsrl.up", 0xf5, F32, V64, VM>; let cx = 1, cx2 = 1 in defm PVSRL : RVSm<"pvsrl", 0xf5, I64, V64, VM512>; // Section 8.12.4 - VSRD (Vector Shift Right Double) defm VSRD : RVSDm<"vsrd", 0xf4, V64, VM>; // Section 8.12.5 - VSLA (Vector Shift Left Arithmetic) let cx = 0, cx2 = 0 in defm VSLAWSX : RVSm<"vsla.w.sx", 0xe6, I32, V64, VM>; let cx = 0, cx2 = 1 in { defm PVSLALO : RVSm<"pvsla.lo", 0xe6, I32, V64, VM>; let isCodeGenOnly = 1 in defm VSLAWZX : RVSm<"vsla.w.zx", 0xe6, I32, V64, VM>; } let cx = 1, cx2 = 0 in defm PVSLAUP : RVSm<"pvsla.up", 0xe6, F32, V64, VM>; let cx = 1, cx2 = 1 in defm PVSLA : RVSm<"pvsla", 0xe6, I64, V64, VM512>; def : MnemonicAlias<"pvsla.lo.sx", "vsla.w.sx">; def : MnemonicAlias<"vsla.w.zx", "pvsla.lo">; def : MnemonicAlias<"vsla.w", "pvsla.lo">; def : MnemonicAlias<"pvsla.lo.zx", "pvsla.lo">; // Section 8.12.6 - VSLAX (Vector Shift Left Arithmetic) defm VSLAL : RVSm<"vsla.l", 0xd4, I64, V64, VM>; // Section 8.12.7 - VSRA (Vector Shift Right Arithmetic) let cx = 0, cx2 = 0 in defm VSRAWSX : RVSm<"vsra.w.sx", 0xf6, I32, V64, VM>; let cx = 0, cx2 = 1 in { defm PVSRALO : RVSm<"pvsra.lo", 0xf6, I32, V64, VM>; let isCodeGenOnly = 1 in defm VSRAWZX : RVSm<"vsra.w.zx", 0xf6, I32, V64, VM>; } let cx = 1, cx2 = 0 in defm PVSRAUP : RVSm<"pvsra.up", 0xf6, F32, V64, VM>; let cx = 1, cx2 = 1 in defm PVSRA : RVSm<"pvsra", 0xf6, I64, V64, VM512>; def : MnemonicAlias<"pvsra.lo.sx", "vsra.w.sx">; def : MnemonicAlias<"vsra.w.zx", "pvsra.lo">; def : MnemonicAlias<"vsra.w", "pvsra.lo">; def : MnemonicAlias<"pvsra.lo.zx", "pvsra.lo">; // Section 8.12.8 - VSRAX (Vector Shift Right Arithmetic) defm VSRAL : RVSm<"vsra.l", 0xd5, I64, V64, VM>; // Section 8.12.9 - VSFA (Vector Shift Left and Add) defm VSFA : RVSAm<"vsfa", 0xd7, V64, VM>; //----------------------------------------------------------------------------- // Section 8.13 - Vector Floating-Point Arithmetic Instructions //----------------------------------------------------------------------------- // Section 8.13.1 - VFAD (Vector Floating Add) let cx = 0, cx2 = 0 in defm VFADDD : RVm<"vfadd.d", 0xcc, V64, I64, VM, simm7fp>; let cx = 0, cx2 = 1 in defm PVFADDLO : RVm<"pvfadd.lo", 0xcc, V64, I64, VM, simm7fp>; let cx = 1, cx2 = 0 in { defm PVFADDUP : RVm<"pvfadd.up", 0xcc, V64, F32, VM, simm7fp>; let isCodeGenOnly = 1 in defm VFADDS : RVm<"vfadd.s", 0xcc, V64, F32, VM, simm7fp>; } let cx = 1, cx2 = 1 in defm PVFADD : RVm<"pvfadd", 0xcc, V64, I64, VM512, simm7fp>; def : MnemonicAlias<"vfadd.s", "pvfadd.up">; // Section 8.13.2 - VFSB (Vector Floating Subtract) let cx = 0, cx2 = 0 in defm VFSUBD : RVm<"vfsub.d", 0xdc, V64, I64, VM, simm7fp>; let cx = 0, cx2 = 1 in defm PVFSUBLO : RVm<"pvfsub.lo", 0xdc, V64, I64, VM, simm7fp>; let cx = 1, cx2 = 0 in { defm PVFSUBUP : RVm<"pvfsub.up", 0xdc, V64, F32, VM, simm7fp>; let isCodeGenOnly = 1 in defm VFSUBS : RVm<"vfsub.s", 0xdc, V64, F32, VM, simm7fp>; } let cx = 1, cx2 = 1 in defm PVFSUB : RVm<"pvfsub", 0xdc, V64, I64, VM512, simm7fp>; def : MnemonicAlias<"vfsub.s", "pvfsub.up">; // Section 8.13.3 - VFMP (Vector Floating Multiply) let cx = 0, cx2 = 0 in defm VFMULD : RVm<"vfmul.d", 0xcd, V64, I64, VM, simm7fp>; let cx = 0, cx2 = 1 in defm PVFMULLO : RVm<"pvfmul.lo", 0xcd, V64, I64, VM, simm7fp>; let cx = 1, cx2 = 0 in { defm PVFMULUP : RVm<"pvfmul.up", 0xcd, V64, F32, VM, simm7fp>; let isCodeGenOnly = 1 in defm VFMULS : RVm<"vfmul.s", 0xcd, V64, F32, VM, simm7fp>; } let cx = 1, cx2 = 1 in defm PVFMUL : RVm<"pvfmul", 0xcd, V64, I64, VM512, simm7fp>; def : MnemonicAlias<"vfmul.s", "pvfmul.up">; // Section 8.13.4 - VFDV (Vector Floating Divide) defm VFDIVD : RVDIVm<"vfdiv.d", 0xdd, V64, I64, VM, simm7fp>; let cx = 1 in defm VFDIVS : RVDIVm<"vfdiv.s", 0xdd, V64, F32, VM, simm7fp>; // Section 8.13.5 - VFSQRT (Vector Floating Square Root) defm VFSQRTD : RVF1m<"vfsqrt.d", 0xed, V64, VM>; let cx = 1 in defm VFSQRTS : RVF1m<"vfsqrt.s", 0xed, V64, VM>; // Section 8.13.6 - VFCP (Vector Floating Compare) let cx = 0, cx2 = 0 in defm VFCMPD : RVm<"vfcmp.d", 0xfc, V64, I64, VM, simm7fp>; let cx = 0, cx2 = 1 in defm PVFCMPLO : RVm<"pvfcmp.lo", 0xfc, V64, I64, VM, simm7fp>; let cx = 1, cx2 = 0 in { defm PVFCMPUP : RVm<"pvfcmp.up", 0xfc, V64, F32, VM, simm7fp>; let isCodeGenOnly = 1 in defm VFCMPS : RVm<"vfcmp.s", 0xfc, V64, F32, VM, simm7fp>; } let cx = 1, cx2 = 1 in defm PVFCMP : RVm<"pvfcmp", 0xfc, V64, I64, VM512, simm7fp>; def : MnemonicAlias<"vfcmp.s", "pvfcmp.up">; // Section 8.13.7 - VFCM (Vector Floating Compare and Select Maximum/Minimum) let cx = 0, cx2 = 0 in defm VFMAXD : RVm<"vfmax.d", 0xbd, V64, I64, VM, simm7fp>; let cx = 0, cx2 = 1 in defm PVFMAXLO : RVm<"pvfmax.lo", 0xbd, V64, I64, VM, simm7fp>; let cx = 1, cx2 = 0 in { defm PVFMAXUP : RVm<"pvfmax.up", 0xbd, V64, F32, VM, simm7fp>; let isCodeGenOnly = 1 in defm VFMAXS : RVm<"vfmax.s", 0xbd, V64, F32, VM, simm7fp>; } let cx = 1, cx2 = 1 in defm PVFMAX : RVm<"pvfmax", 0xbd, V64, I64, VM512, simm7fp>; let cs2 = 1 in { let cx = 0, cx2 = 0 in defm VFMIND : RVm<"vfmin.d", 0xbd, V64, I64, VM, simm7fp>; let cx = 0, cx2 = 1 in defm PVFMINLO : RVm<"pvfmin.lo", 0xbd, V64, I64, VM, simm7fp>; let cx = 1, cx2 = 0 in { defm PVFMINUP : RVm<"pvfmin.up", 0xbd, V64, F32, VM, simm7fp>; let isCodeGenOnly = 1 in defm VFMINS : RVm<"vfmin.s", 0xbd, V64, F32, VM, simm7fp>; } let cx = 1, cx2 = 1 in defm PVFMIN : RVm<"pvfmin", 0xbd, V64, I64, VM512, simm7fp>; } def : MnemonicAlias<"vfmax.s", "pvfmax.up">; def : MnemonicAlias<"vfmin.s", "pvfmin.up">; // Section 8.13.8 - VFMAD (Vector Floating Fused Multiply Add) let cx = 0, cx2 = 0 in defm VFMADD : RVMm<"vfmad.d", 0xe2, V64, I64, VM, simm7fp>; let cx = 0, cx2 = 1 in defm PVFMADLO : RVMm<"pvfmad.lo", 0xe2, V64, I64, VM, simm7fp>; let cx = 1, cx2 = 0 in { defm PVFMADUP : RVMm<"pvfmad.up", 0xe2, V64, F32, VM, simm7fp>; let isCodeGenOnly = 1 in defm VFMADS : RVMm<"vfmad.s", 0xe2, V64, F32, VM, simm7fp>; } let cx = 1, cx2 = 1 in defm PVFMAD : RVMm<"pvfmad", 0xe2, V64, I64, VM512, simm7fp>; def : MnemonicAlias<"vfmad.s", "pvfmad.up">; // Section 8.13.9 - VFMSB (Vector Floating Fused Multiply Subtract) let cx = 0, cx2 = 0 in defm VFMSBD : RVMm<"vfmsb.d", 0xf2, V64, I64, VM, simm7fp>; let cx = 0, cx2 = 1 in defm PVFMSBLO : RVMm<"pvfmsb.lo", 0xf2, V64, I64, VM, simm7fp>; let cx = 1, cx2 = 0 in { defm PVFMSBUP : RVMm<"pvfmsb.up", 0xf2, V64, F32, VM, simm7fp>; let isCodeGenOnly = 1 in defm VFMSBS : RVMm<"vfmsb.s", 0xf2, V64, F32, VM, simm7fp>; } let cx = 1, cx2 = 1 in defm PVFMSB : RVMm<"pvfmsb", 0xf2, V64, I64, VM512, simm7fp>; def : MnemonicAlias<"vfmsb.s", "pvfmsb.up">; // Section 8.13.10 - VFNMAD (Vector Floating Fused Negative Multiply Add) let cx = 0, cx2 = 0 in defm VFNMADD : RVMm<"vfnmad.d", 0xe3, V64, I64, VM, simm7fp>; let cx = 0, cx2 = 1 in defm PVFNMADLO : RVMm<"pvfnmad.lo", 0xe3, V64, I64, VM, simm7fp>; let cx = 1, cx2 = 0 in { defm PVFNMADUP : RVMm<"pvfnmad.up", 0xe3, V64, F32, VM, simm7fp>; let isCodeGenOnly = 1 in defm VFNMADS : RVMm<"vfnmad.s", 0xe3, V64, F32, VM, simm7fp>; } let cx = 1, cx2 = 1 in defm PVFNMAD : RVMm<"pvfnmad", 0xe3, V64, I64, VM512, simm7fp>; def : MnemonicAlias<"vfnmad.s", "pvfnmad.up">; // Section 8.13.11 - VFNMSB (Vector Floating Fused Negative Multiply Subtract) let cx = 0, cx2 = 0 in defm VFNMSBD : RVMm<"vfnmsb.d", 0xf3, V64, I64, VM, simm7fp>; let cx = 0, cx2 = 1 in defm PVFNMSBLO : RVMm<"pvfnmsb.lo", 0xf3, V64, I64, VM, simm7fp>; let cx = 1, cx2 = 0 in { defm PVFNMSBUP : RVMm<"pvfnmsb.up", 0xf3, V64, F32, VM, simm7fp>; let isCodeGenOnly = 1 in defm VFNMSBS : RVMm<"vfnmsb.s", 0xf3, V64, F32, VM, simm7fp>; } let cx = 1, cx2 = 1 in defm PVFNMSB : RVMm<"pvfnmsb", 0xf3, V64, I64, VM512, simm7fp>; def : MnemonicAlias<"vfnmsb.s", "pvfnmsb.up">; // Section 8.13.12 - VRCP (Vector Floating Reciprocal) let cx = 0, cx2 = 0 in defm VRCPD : RVF1m<"vrcp.d", 0xe1, V64, VM>; let cx = 0, cx2 = 1 in defm PVRCPLO : RVF1m<"pvrcp.lo", 0xe1, V64, VM>; let cx = 1, cx2 = 0 in { defm PVRCPUP : RVF1m<"pvrcp.up", 0xe1, V64, VM>; let isCodeGenOnly = 1 in defm VRCPS : RVF1m<"vrcp.s", 0xe1, V64, VM>; } let cx = 1, cx2 = 1 in defm PVRCP : RVF1m<"pvrcp", 0xe1, V64, VM512>; def : MnemonicAlias<"vrcp.s", "pvrcp.up">; // Section 8.13.13 - VRSQRT (Vector Floating Reciprocal Square Root) let cx = 0, cx2 = 0 in defm VRSQRTD : RVF1m<"vrsqrt.d", 0xf1, V64, VM>; let cx = 0, cx2 = 1 in defm PVRSQRTLO : RVF1m<"pvrsqrt.lo", 0xf1, V64, VM>; let cx = 1, cx2 = 0 in { defm PVRSQRTUP : RVF1m<"pvrsqrt.up", 0xf1, V64, VM>; let isCodeGenOnly = 1 in defm VRSQRTS : RVF1m<"vrsqrt.s", 0xf1, V64, VM>; } let cx = 1, cx2 = 1 in defm PVRSQRT : RVF1m<"pvrsqrt", 0xf1, V64, VM512>; let cs2 = 1 in { let cx = 0, cx2 = 0 in defm VRSQRTDNEX : RVF1m<"vrsqrt.d.nex", 0xf1, V64, VM>; let cx = 0, cx2 = 1 in defm PVRSQRTLONEX : RVF1m<"pvrsqrt.lo.nex", 0xf1, V64, VM>; let cx = 1, cx2 = 0 in { defm PVRSQRTUPNEX : RVF1m<"pvrsqrt.up.nex", 0xf1, V64, VM>; let isCodeGenOnly = 1 in defm VRSQRTSNEX : RVF1m<"vrsqrt.s.nex", 0xf1, V64, VM>; } let cx = 1, cx2 = 1 in defm PVRSQRTNEX : RVF1m<"pvrsqrt.nex", 0xf1, V64, VM512>; } def : MnemonicAlias<"vrsqrt.s", "pvrsqrt.up">; def : MnemonicAlias<"vrsqrt.s.nex", "pvrsqrt.up.nex">; // Section 8.13.14 - VFIX (Vector Convert to Fixed Pointer) let cx = 0, cx2 = 0, cs2 = 0 in defm VCVTWDSX : RVFIXm<"vcvt.w.d.sx", 0xe8, V64, VM>; let cx = 0, cx2 = 1, cs2 = 0 in defm VCVTWDZX : RVFIXm<"vcvt.w.d.zx", 0xe8, V64, VM>; let cx = 1, cx2 = 0, cs2 = 0 in defm VCVTWSSX : RVFIXm<"vcvt.w.s.sx", 0xe8, V64, VM>; let cx = 1, cx2 = 1, cs2 = 0 in defm VCVTWSZX : RVFIXm<"vcvt.w.s.zx", 0xe8, V64, VM>; let cx = 0, cx2 = 1, cs2 = 1 in defm PVCVTWSLO : RVFIXm<"pvcvt.w.s.lo", 0xe8, V64, VM>; let cx = 1, cx2 = 0, cs2 = 1 in defm PVCVTWSUP : RVFIXm<"pvcvt.w.s.up", 0xe8, V64, VM>; let cx = 1, cx2 = 1, cs2 = 1 in defm PVCVTWS : RVFIXm<"pvcvt.w.s", 0xe8, V64, VM512>; // Section 8.13.15 - VFIXX (Vector Convert to Fixed Pointer) defm VCVTLD : RVFIXm<"vcvt.l.d", 0xa8, V64, VM>; // Section 8.13.16 - VFLT (Vector Convert to Floating Pointer) let cx = 0, cx2 = 0, cs2 = 0 in defm VCVTDW : RVF1m<"vcvt.d.w", 0xf8, V64, VM>; let cx = 1, cx2 = 0, cs2 = 0 in defm VCVTSW : RVF1m<"vcvt.s.w", 0xf8, V64, VM>; let cx = 0, cx2 = 1, cs2 = 1 in defm PVCVTSWLO : RVF1m<"pvcvt.s.w.lo", 0xf8, V64, VM>; let cx = 1, cx2 = 0, cs2 = 1 in defm PVCVTSWUP : RVF1m<"pvcvt.s.w.up", 0xf8, V64, VM>; let cx = 1, cx2 = 1, cs2 = 1 in defm PVCVTSW : RVF1m<"pvcvt.s.w", 0xf8, V64, VM512>; // Section 8.13.17 - VFLTX (Vector Convert to Floating Pointer) defm VCVTDL : RVF1m<"vcvt.d.l", 0xb8, V64, VM>; // Section 8.13.18 - VCVS (Vector Convert to Single-format) defm VCVTSD : RVF1m<"vcvt.s.d", 0x9f, V64, VM>; // Section 8.13.19 - VCVD (Vector Convert to Double-format) defm VCVTDS : RVF1m<"vcvt.d.s", 0x8f, V64, VM>; //----------------------------------------------------------------------------- // Section 8.14 - Vector Reduction Instructions //----------------------------------------------------------------------------- // Section 8.14.1 - VSUMS (Vector Sum Single) defm VSUMWSX : RVF1m<"vsum.w.sx", 0xea, V64, VM>; let cx2 = 1 in defm VSUMWZX : RVF1m<"vsum.w.zx", 0xea, V64, VM>; // Section 8.14.2 - VSUMX (Vector Sum) defm VSUML : RVF1m<"vsum.l", 0xaa, V64, VM>; // Section 8.14.3 - VFSUM (Vector Floating Sum) defm VFSUMD : RVF1m<"vfsum.d", 0xec, V64, VM>; let cx = 1 in defm VFSUMS : RVF1m<"vfsum.s", 0xec, V64, VM>; // Section 8.14.4 - VMAXS (Vector Maximum/Minimum Single) let cx2 = 0 in defm VRMAXSWFSTSX : RVF1m<"vrmaxs.w.fst.sx", 0xbb, V64, VM>; let cx2 = 1 in defm VRMAXSWFSTZX : RVF1m<"vrmaxs.w.fst.zx", 0xbb, V64, VM>; let cs = 1 in { let cx2 = 0 in defm VRMAXSWLSTSX : RVF1m<"vrmaxs.w.lst.sx", 0xbb, V64, VM>; let cx2 = 1 in defm VRMAXSWLSTZX : RVF1m<"vrmaxs.w.lst.zx", 0xbb, V64, VM>; } let cs2 = 1 in { let cx2 = 0 in defm VRMINSWFSTSX : RVF1m<"vrmins.w.fst.sx", 0xbb, V64, VM>; let cx2 = 1 in defm VRMINSWFSTZX : RVF1m<"vrmins.w.fst.zx", 0xbb, V64, VM>; let cs = 1 in { let cx2 = 0 in defm VRMINSWLSTSX : RVF1m<"vrmins.w.lst.sx", 0xbb, V64, VM>; let cx2 = 1 in defm VRMINSWLSTZX : RVF1m<"vrmins.w.lst.zx", 0xbb, V64, VM>; } } // Section 8.14.5 - VMAXX (Vector Maximum/Minimum) let cs = 0 in defm VRMAXSLFST : RVF1m<"vrmaxs.l.fst", 0xab, V64, VM>; let cs = 1 in defm VRMAXSLLST : RVF1m<"vrmaxs.l.lst", 0xab, V64, VM>; let cs2 = 1 in { let cs = 0 in defm VRMINSLFST : RVF1m<"vrmins.l.fst", 0xab, V64, VM>; let cs = 1 in defm VRMINSLLST : RVF1m<"vrmins.l.lst", 0xab, V64, VM>; } // Section 8.14.6 - VFMAX (Vector Floating Maximum/Minimum) let cs = 0 in defm VFRMAXDFST : RVF1m<"vfrmax.d.fst", 0xad, V64, VM>; let cs = 1 in defm VFRMAXDLST : RVF1m<"vfrmax.d.lst", 0xad, V64, VM>; let cs2 = 1 in { let cs = 0 in defm VFRMINDFST : RVF1m<"vfrmin.d.fst", 0xad, V64, VM>; let cs = 1 in defm VFRMINDLST : RVF1m<"vfrmin.d.lst", 0xad, V64, VM>; } let cx = 1 in { let cs = 0 in defm VFRMAXSFST : RVF1m<"vfrmax.s.fst", 0xad, V64, VM>; let cs = 1 in defm VFRMAXSLST : RVF1m<"vfrmax.s.lst", 0xad, V64, VM>; let cs2 = 1 in { let cs = 0 in defm VFRMINSFST : RVF1m<"vfrmin.s.fst", 0xad, V64, VM>; let cs = 1 in defm VFRMINSLST : RVF1m<"vfrmin.s.lst", 0xad, V64, VM>; } } // Section 8.14.7 - VRAND (Vector Reduction And) defm VRAND : RVF1m<"vrand", 0x88, V64, VM>; // Section 8.14.8 - VROR (Vector Reduction Or) defm VROR : RVF1m<"vror", 0x98, V64, VM>; // Section 8.14.9 - VRXOR (Vector Reduction Exclusive Or) defm VRXOR : RVF1m<"vrxor", 0x89, V64, VM>; //----------------------------------------------------------------------------- // Section 8.15 - Vector Iterative Operation Instructions //----------------------------------------------------------------------------- // Section 8.15.1 - VFIA (Vector Floating Iteration Add) let cx = 0 in defm VFIAD : RVI2m<"vfia.d", 0xce, V64, I64>; let cx = 1 in defm VFIAS : RVI2m<"vfia.s", 0xce, V64, F32>; // Section 8.15.2 - VFIS (Vector Floating Iteration Subtract) let cx = 0 in defm VFISD : RVI2m<"vfis.d", 0xde, V64, I64>; let cx = 1 in defm VFISS : RVI2m<"vfis.s", 0xde, V64, F32>; // Section 8.15.3 - VFIM (Vector Floating Iteration Multiply) let cx = 0 in defm VFIMD : RVI2m<"vfim.d", 0xcf, V64, I64>; let cx = 1 in defm VFIMS : RVI2m<"vfim.s", 0xcf, V64, F32>; // Section 8.15.4 - VFIAM (Vector Floating Iteration Add and Multiply) let cx = 0 in defm VFIAMD : RVI3m<"vfiam.d", 0xee, V64, I64>; let cx = 1 in defm VFIAMS : RVI3m<"vfiam.s", 0xee, V64, F32>; // Section 8.15.5 - VFISM (Vector Floating Iteration Subtract and Multiply) let cx = 0 in defm VFISMD : RVI3m<"vfism.d", 0xfe, V64, I64>; let cx = 1 in defm VFISMS : RVI3m<"vfism.s", 0xfe, V64, F32>; // Section 8.15.6 - VFIMA (Vector Floating Iteration Multiply and Add) let cx = 0 in defm VFIMAD : RVI3m<"vfima.d", 0xef, V64, I64>; let cx = 1 in defm VFIMAS : RVI3m<"vfima.s", 0xef, V64, F32>; // Section 8.15.7 - VFIMS (Vector Floating Iteration Multiply and Subtract) let cx = 0 in defm VFIMSD : RVI3m<"vfims.d", 0xff, V64, I64>; let cx = 1 in defm VFIMSS : RVI3m<"vfims.s", 0xff, V64, F32>; //----------------------------------------------------------------------------- // Section 8.16 - Vector Merger Operation Instructions //----------------------------------------------------------------------------- // Section 8.16.1 - VMRG (Vector Merge) let cx = 0 in defm VMRG : RVm<"vmrg", 0xd6, V64, I64, VM>; // FIXME: vmrg.w should be called as pvmrg, but following assembly manual. let cx = 1 in defm VMRGW : RVm<"vmrg.w", 0xd6, V64, I64, VM512>; def : MnemonicAlias<"vmrg.l", "vmrg">; // Section 8.16.2 - VSHF (Vector Shuffle) defm VSHF : RVSHFm<"vshf", 0xbc, V64>; // Section 8.16.3 - VCP (Vector Compress) defm VCP : RV1m<"vcp", 0x8d, V64, VM>; // Section 8.16.4 - VEX (Vector Expand) defm VEX : RV1m<"vex", 0x9d, V64, VM>; //----------------------------------------------------------------------------- // Section 8.17 - Vector Mask Operation Instructions //----------------------------------------------------------------------------- // Section 8.17.1 - VFMK (Vector Form Mask) defm VFMKL : RVMKm<"vfmk.l.", 0xb4, V64, VM>; def : MnemonicAlias<"vfmk.l", "vfmk.l.at">; // Section 8.17.2 - VFMS (Vector Form Mask Single) defm VFMKW : RVMKm<"vfmk.w.", 0xb5, V64, VM>; let isCodeGenOnly = 1 in defm PVFMKWLO : RVMKm<"vfmk.w.", 0xb5, V64, VM>; let cx = 1 in defm PVFMKWUP : RVMKm<"pvfmk.w.up.", 0xb5, V64, VM>; def : MnemonicAlias<"vfmk.w", "vfmk.w.at">; def : MnemonicAlias<"pvfmk.w.up", "pvfmk.w.up.at">; def : MnemonicAlias<"pvfmk.w.lo", "vfmk.w.at">; foreach CC = [ "af", "gt", "lt", "ne", "eq", "ge", "le", "at" ] in { def : MnemonicAlias<"pvfmk.w.lo."#CC, "vfmk.w."#CC>; } // Section 8.17.3 - VFMF (Vector Form Mask Floating Point) defm VFMKD : RVMKm<"vfmk.d.", 0xb6, V64, VM>; let cx2 = 1 in defm PVFMKSLO : RVMKm<"pvfmk.s.lo.", 0xb6, V64, VM>; let cx = 1 in { defm PVFMKSUP : RVMKm<"pvfmk.s.up.", 0xb6, V64, VM>; let isCodeGenOnly = 1 in defm VFMKS : RVMKm<"vfmk.s.", 0xb6, V64, VM>; } def : MnemonicAlias<"vfmk.d", "vfmk.d.at">; def : MnemonicAlias<"pvfmk.s.lo", "pvfmk.s.lo.at">; def : MnemonicAlias<"pvfmk.s.up", "pvfmk.s.up.at">; def : MnemonicAlias<"vfmk.s", "pvfmk.s.up.at">; foreach CC = [ "af", "gt", "lt", "ne", "eq", "ge", "le", "at", "num", "nan", "gtnan", "ltnan", "nenan", "eqnan", "genan", "lenan" ] in { def : MnemonicAlias<"vfmk.s."#CC, "pvfmk.s.up."#CC>; } // Section 8.17.4 - ANDM (And VM) defm ANDM : RVM2m<"andm", 0x84, VM>; // Section 8.17.5 - ORM (Or VM) defm ORM : RVM2m<"orm", 0x85, VM>; // Section 8.17.6 - XORM (Exclusive Or VM) defm XORM : RVM2m<"xorm", 0x86, VM>; // Section 8.17.7 - EQVM (Equivalence VM) defm EQVM : RVM2m<"eqvm", 0x87, VM>; // Section 8.17.8 - NNDM (Negate And VM) defm NNDM : RVM2m<"nndm", 0x94, VM>; // Section 8.17.9 - NEGM (Negate VM) defm NEGM : RVM1m<"negm", 0x95, VM>; // Section 8.17.10 - PCVM (Population Count of VM) defm PCVM : RVMSm<"pcvm", 0xa4, VM>; // Section 8.17.11 - LZVM (Leading Zero of VM) defm LZVM : RVMSm<"lzvm", 0xa5, VM>; // Section 8.17.12 - TOVM (Trailing One of VM) defm TOVM : RVMSm<"tovm", 0xa6, VM>; //----------------------------------------------------------------------------- // Section 8.18 - Vector Control Instructions //----------------------------------------------------------------------------- // Section 8.18.1 - LVL (Load VL) let sx = 0, cz = 0, sz = 0, hasSideEffects = 0, Defs = [VL] in { def LVLr : RR<0xbf, (outs), (ins I64:$sy), "lvl $sy">; let cy = 0 in def LVLi : RR<0xbf, (outs), (ins simm7:$sy), "lvl $sy">; } // Section 8.18.2 - SVL (Save VL) let cy = 0, sy = 0, cz = 0, sz = 0, hasSideEffects = 0, Uses = [VL] in def SVL : RR<0x2f, (outs I64:$sx), (ins), "svl $sx">; // Section 8.18.3 - SMVL (Save Maximum Vector Length) let cy = 0, sy = 0, cz = 0, sz = 0, hasSideEffects = 0 in def SMVL : RR<0x2e, (outs I64:$sx), (ins), "smvl $sx">; // Section 8.18.4 - LVIX (Load Vector Data Index) let sx = 0, cz = 0, sz = 0, hasSideEffects = 0, Defs = [VIX] in { def LVIXr : RR<0xaf, (outs), (ins I64:$sy), "lvix $sy">; let cy = 0 in def LVIXi : RR<0xaf, (outs), (ins uimm6:$sy), "lvix $sy">; }