//=- X86SchedBroadwell.td - X86 Broadwell Scheduling ---------*- tablegen -*-=// // // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. // See https://llvm.org/LICENSE.txt for license information. // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception // //===----------------------------------------------------------------------===// // // This file defines the machine model for Broadwell to support instruction // scheduling and other instruction cost heuristics. // //===----------------------------------------------------------------------===// def BroadwellModel : SchedMachineModel { // All x86 instructions are modeled as a single micro-op, and BW can decode 4 // instructions per cycle. let IssueWidth = 4; let MicroOpBufferSize = 192; // Based on the reorder buffer. let LoadLatency = 5; let MispredictPenalty = 16; // Based on the LSD (loop-stream detector) queue size and benchmarking data. let LoopMicroOpBufferSize = 50; // This flag is set to allow the scheduler to assign a default model to // unrecognized opcodes. let CompleteModel = 0; } let SchedModel = BroadwellModel in { // Broadwell can issue micro-ops to 8 different ports in one cycle. // Ports 0, 1, 5, and 6 handle all computation. // Port 4 gets the data half of stores. Store data can be available later than // the store address, but since we don't model the latency of stores, we can // ignore that. // Ports 2 and 3 are identical. They handle loads and the address half of // stores. Port 7 can handle address calculations. def BWPort0 : ProcResource<1>; def BWPort1 : ProcResource<1>; def BWPort2 : ProcResource<1>; def BWPort3 : ProcResource<1>; def BWPort4 : ProcResource<1>; def BWPort5 : ProcResource<1>; def BWPort6 : ProcResource<1>; def BWPort7 : ProcResource<1>; // Many micro-ops are capable of issuing on multiple ports. def BWPort01 : ProcResGroup<[BWPort0, BWPort1]>; def BWPort23 : ProcResGroup<[BWPort2, BWPort3]>; def BWPort237 : ProcResGroup<[BWPort2, BWPort3, BWPort7]>; def BWPort04 : ProcResGroup<[BWPort0, BWPort4]>; def BWPort05 : ProcResGroup<[BWPort0, BWPort5]>; def BWPort06 : ProcResGroup<[BWPort0, BWPort6]>; def BWPort15 : ProcResGroup<[BWPort1, BWPort5]>; def BWPort16 : ProcResGroup<[BWPort1, BWPort6]>; def BWPort56 : ProcResGroup<[BWPort5, BWPort6]>; def BWPort015 : ProcResGroup<[BWPort0, BWPort1, BWPort5]>; def BWPort056 : ProcResGroup<[BWPort0, BWPort5, BWPort6]>; def BWPort0156: ProcResGroup<[BWPort0, BWPort1, BWPort5, BWPort6]>; // 60 Entry Unified Scheduler def BWPortAny : ProcResGroup<[BWPort0, BWPort1, BWPort2, BWPort3, BWPort4, BWPort5, BWPort6, BWPort7]> { let BufferSize=60; } // Integer division issued on port 0. def BWDivider : ProcResource<1>; // FP division and sqrt on port 0. def BWFPDivider : ProcResource<1>; // Integer loads are 5 cycles, so ReadAfterLd registers needn't be available until 5 // cycles after the memory operand. def : ReadAdvance; // Vector loads are 5/5/6 cycles, so ReadAfterVec*Ld registers needn't be available // until 5/5/6 cycles after the memory operand. def : ReadAdvance; def : ReadAdvance; def : ReadAdvance; def : ReadAdvance; // Many SchedWrites are defined in pairs with and without a folded load. // Instructions with folded loads are usually micro-fused, so they only appear // as two micro-ops when queued in the reservation station. // This multiclass defines the resource usage for variants with and without // folded loads. multiclass BWWriteResPair ExePorts, int Lat, list Res = [1], int UOps = 1, int LoadLat = 5> { // Register variant is using a single cycle on ExePort. def : WriteRes { let Latency = Lat; let ResourceCycles = Res; let NumMicroOps = UOps; } // Memory variant also uses a cycle on port 2/3 and adds LoadLat cycles to // the latency (default = 5). def : WriteRes { let Latency = !add(Lat, LoadLat); let ResourceCycles = !listconcat([1], Res); let NumMicroOps = !add(UOps, 1); } } // A folded store needs a cycle on port 4 for the store data, and an extra port // 2/3/7 cycle to recompute the address. def : WriteRes; // Arithmetic. defm : BWWriteResPair; // Simple integer ALU op. defm : BWWriteResPair; // Integer ALU + flags op. // Integer multiplication. defm : BWWriteResPair; defm : BWWriteResPair; defm : X86WriteRes; defm : X86WriteRes; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; def : WriteRes { let Latency = 3; } // TODO: Why isn't the BWDivider used consistently? defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : BWWriteResPair; def : WriteRes; // LEA instructions can't fold loads. defm : BWWriteResPair; // Conditional move. defm : X86WriteRes; // x87 conditional move. def : WriteRes; // Setcc. def : WriteRes { let Latency = 2; let NumMicroOps = 3; } defm : X86WriteRes; defm : X86WriteRes; // Bit Test instrs defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; // Bit Test + Set instrs defm : X86WriteRes; defm : X86WriteRes; // Bit counts. defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; // Integer shifts and rotates. defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; // SHLD/SHRD. defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; // BMI1 BEXTR/BLS, BMI2 BZHI defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; // Loads, stores, and moves, not folded with other operations. defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; // Idioms that clear a register, like xorps %xmm0, %xmm0. // These can often bypass execution ports completely. def : WriteRes; // Treat misc copies as a move. def : InstRW<[WriteMove], (instrs COPY)>; // Branches don't produce values, so they have no latency, but they still // consume resources. Indirect branches can fold loads. defm : BWWriteResPair; // Floating point. This covers both scalar and vector operations. defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : BWWriteResPair; // Floating point add/sub. defm : BWWriteResPair; // Floating point add/sub (XMM). defm : BWWriteResPair; // Floating point add/sub (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point double add/sub. defm : BWWriteResPair; // Floating point double add/sub (XMM). defm : BWWriteResPair; // Floating point double add/sub (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point compare. defm : BWWriteResPair; // Floating point compare (XMM). defm : BWWriteResPair; // Floating point compare (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point double compare. defm : BWWriteResPair; // Floating point double compare (XMM). defm : BWWriteResPair; // Floating point double compare (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point compare to flags. defm : BWWriteResPair; // Floating point multiplication. defm : BWWriteResPair; // Floating point multiplication (XMM). defm : BWWriteResPair; // Floating point multiplication (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point double multiplication. defm : BWWriteResPair; // Floating point double multiplication (XMM). defm : BWWriteResPair; // Floating point double multiplication (YMM/ZMM). defm : X86WriteResPairUnsupported; //defm : BWWriteResPair; // Floating point division. defm : BWWriteResPair; // Floating point division (XMM). defm : BWWriteResPair; // Floating point division (YMM). defm : X86WriteResPairUnsupported; //defm : BWWriteResPair; // Floating point division. defm : BWWriteResPair; // Floating point division (XMM). defm : BWWriteResPair; // Floating point division (YMM). defm : X86WriteResPairUnsupported; defm : X86WriteRes; // Floating point square root. defm : X86WriteRes; defm : BWWriteResPair; // Floating point square root (XMM). defm : BWWriteResPair; // Floating point square root (YMM). defm : X86WriteResPairUnsupported; defm : X86WriteRes; // Floating point double square root. defm : X86WriteRes; defm : BWWriteResPair; // Floating point double square root (XMM). defm : BWWriteResPair; // Floating point double square root (YMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point long double square root. defm : BWWriteResPair; // Floating point reciprocal estimate. defm : BWWriteResPair; // Floating point reciprocal estimate (XMM). defm : BWWriteResPair; // Floating point reciprocal estimate (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point reciprocal square root estimate. defm : BWWriteResPair; // Floating point reciprocal square root estimate (XMM). defm : BWWriteResPair; // Floating point reciprocal square root estimate (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Fused Multiply Add. defm : BWWriteResPair; // Fused Multiply Add (XMM). defm : BWWriteResPair; // Fused Multiply Add (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point double dot product. defm : BWWriteResPair; // Floating point single dot product. defm : BWWriteResPair; // Floating point single dot product (YMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point fabs/fchs. defm : X86WriteRes; // Floating point rounding. defm : X86WriteRes; // Floating point rounding (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : X86WriteRes; defm : X86WriteRes; defm : BWWriteResPair; // Floating point and/or/xor logicals. defm : BWWriteResPair; // Floating point and/or/xor logicals (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point TEST instructions. defm : BWWriteResPair; // Floating point TEST instructions (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point vector shuffles. defm : BWWriteResPair; // Floating point vector shuffles (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point vector variable shuffles. defm : BWWriteResPair; // Floating point vector variable shuffles. defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Floating point vector blends. defm : BWWriteResPair; // Floating point vector blends. defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Fp vector variable blends. defm : BWWriteResPair; // Fp vector variable blends. defm : X86WriteResPairUnsupported; // FMA Scheduling helper class. // class FMASC { X86FoldableSchedWrite Sched = WriteFAdd; } // Vector integer operations. defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteRes; defm : BWWriteResPair; // Vector integer ALU op, no logicals. defm : BWWriteResPair; // Vector integer ALU op, no logicals. defm : BWWriteResPair; // Vector integer ALU op, no logicals (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector integer and/or/xor. defm : BWWriteResPair; // Vector integer and/or/xor. defm : BWWriteResPair; // Vector integer and/or/xor (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector integer TEST instructions. defm : BWWriteResPair; // Vector integer TEST instructions (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector integer multiply. defm : BWWriteResPair; // Vector integer multiply. defm : BWWriteResPair; // Vector integer multiply. defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector PMULLD. defm : BWWriteResPair; // Vector PMULLD (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector shuffles. defm : BWWriteResPair; // Vector shuffles. defm : BWWriteResPair; // Vector shuffles (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector variable shuffles. defm : BWWriteResPair; // Vector variable shuffles. defm : BWWriteResPair; // Vector variable shuffles (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector blends. defm : BWWriteResPair; // Vector blends (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector variable blends. defm : BWWriteResPair; // Vector variable blends (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector MPSAD. defm : BWWriteResPair; // Vector MPSAD. defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector PSADBW. defm : BWWriteResPair; // Vector PSADBW. defm : BWWriteResPair; // Vector PSADBW (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Vector PHMINPOS. // Vector integer shifts. defm : BWWriteResPair; defm : BWWriteResPair; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteResPairUnsupported; defm : BWWriteResPair; defm : BWWriteResPair; // Vector integer immediate shifts (XMM). defm : BWWriteResPair; // Vector integer immediate shifts (YMM/ZMM). defm : X86WriteResPairUnsupported; defm : BWWriteResPair; // Variable vector shifts. defm : BWWriteResPair; // Variable vector shifts (YMM/ZMM). defm : X86WriteResPairUnsupported; // Vector insert/extract operations. def : WriteRes { let Latency = 2; let NumMicroOps = 2; let ResourceCycles = [2]; } def : WriteRes { let Latency = 6; let NumMicroOps = 2; } def : WriteRes { let Latency = 2; let NumMicroOps = 2; } def : WriteRes { let Latency = 2; let NumMicroOps = 3; } // Conversion between integer and float. defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : X86WriteResPairUnsupported; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : X86WriteResPairUnsupported; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : X86WriteResPairUnsupported; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : X86WriteResPairUnsupported; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : X86WriteResPairUnsupported; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : X86WriteResPairUnsupported; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteResUnsupported; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteResUnsupported; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteResUnsupported; defm : X86WriteRes; defm : X86WriteRes; defm : X86WriteResUnsupported; // Strings instructions. // Packed Compare Implicit Length Strings, Return Mask def : WriteRes { let Latency = 11; let NumMicroOps = 3; let ResourceCycles = [3]; } def : WriteRes { let Latency = 16; let NumMicroOps = 4; let ResourceCycles = [3,1]; } // Packed Compare Explicit Length Strings, Return Mask def : WriteRes { let Latency = 19; let NumMicroOps = 9; let ResourceCycles = [4,3,1,1]; } def : WriteRes { let Latency = 24; let NumMicroOps = 10; let ResourceCycles = [4,3,1,1,1]; } // Packed Compare Implicit Length Strings, Return Index def : WriteRes { let Latency = 11; let NumMicroOps = 3; let ResourceCycles = [3]; } def : WriteRes { let Latency = 16; let NumMicroOps = 4; let ResourceCycles = [3,1]; } // Packed Compare Explicit Length Strings, Return Index def : WriteRes { let Latency = 18; let NumMicroOps = 8; let ResourceCycles = [4,3,1]; } def : WriteRes { let Latency = 23; let NumMicroOps = 9; let ResourceCycles = [4,3,1,1]; } // MOVMSK Instructions. def : WriteRes { let Latency = 3; } def : WriteRes { let Latency = 3; } def : WriteRes { let Latency = 3; } def : WriteRes { let Latency = 1; } // AES instructions. def : WriteRes { // Decryption, encryption. let Latency = 7; let NumMicroOps = 1; let ResourceCycles = [1]; } def : WriteRes { let Latency = 12; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def : WriteRes { // InvMixColumn. let Latency = 14; let NumMicroOps = 2; let ResourceCycles = [2]; } def : WriteRes { let Latency = 19; let NumMicroOps = 3; let ResourceCycles = [2,1]; } def : WriteRes { // Key Generation. let Latency = 29; let NumMicroOps = 11; let ResourceCycles = [2,7,2]; } def : WriteRes { let Latency = 33; let NumMicroOps = 11; let ResourceCycles = [2,7,1,1]; } // Carry-less multiplication instructions. defm : BWWriteResPair; // Catch-all for expensive system instructions. def : WriteRes { let Latency = 100; } // def WriteSystem : SchedWrite; // AVX2. defm : BWWriteResPair; // Fp 256-bit width vector shuffles. defm : BWWriteResPair; // Fp 256-bit width vector variable shuffles. defm : BWWriteResPair; // 256-bit width vector shuffles. defm : BWWriteResPair; // 256-bit width vector variable shuffles. // Old microcoded instructions that nobody use. def : WriteRes { let Latency = 100; } // def WriteMicrocoded : SchedWrite; // Fence instructions. def : WriteRes; // Load/store MXCSR. def : WriteRes { let Latency = 7; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def : WriteRes { let Latency = 2; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } // Nop, not very useful expect it provides a model for nops! def : WriteRes; //////////////////////////////////////////////////////////////////////////////// // Horizontal add/sub instructions. //////////////////////////////////////////////////////////////////////////////// defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; defm : BWWriteResPair; // Remaining instrs. def BWWriteResGroup1 : SchedWriteRes<[BWPort0]> { let Latency = 1; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup1], (instregex "VPSLLVQ(Y?)rr", "VPSRLVQ(Y?)rr")>; def BWWriteResGroup2 : SchedWriteRes<[BWPort1]> { let Latency = 1; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup2], (instregex "COM(P?)_FST0r", "UCOM_F(P?)r")>; def BWWriteResGroup3 : SchedWriteRes<[BWPort5]> { let Latency = 1; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup3], (instrs MMX_MOVQ2DQrr)>; def BWWriteResGroup4 : SchedWriteRes<[BWPort6]> { let Latency = 1; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup4], (instregex "JMP(16|32|64)r")>; def BWWriteResGroup5 : SchedWriteRes<[BWPort01]> { let Latency = 1; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup5], (instrs FINCSTP, FNOP)>; def BWWriteResGroup6 : SchedWriteRes<[BWPort06]> { let Latency = 1; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup6], (instrs CDQ, CQO)>; def BWWriteResGroup7 : SchedWriteRes<[BWPort15]> { let Latency = 1; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup7], (instregex "ANDN(32|64)rr")>; def BWWriteResGroup8 : SchedWriteRes<[BWPort015]> { let Latency = 1; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup8], (instregex "VPBLENDD(Y?)rri")>; def BWWriteResGroup9 : SchedWriteRes<[BWPort0156]> { let Latency = 1; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup9], (instrs SGDT64m, SIDT64m, SMSW16m, STRm, SYSCALL)>; def BWWriteResGroup10 : SchedWriteRes<[BWPort4,BWPort237]> { let Latency = 1; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup10], (instrs FBSTPm)>; def: InstRW<[BWWriteResGroup10], (instregex "ST_FP(32|64|80)m")>; def BWWriteResGroup12 : SchedWriteRes<[BWPort01]> { let Latency = 2; let NumMicroOps = 2; let ResourceCycles = [2]; } def: InstRW<[BWWriteResGroup12], (instrs FDECSTP)>; def BWWriteResGroup14 : SchedWriteRes<[BWPort0156]> { let Latency = 2; let NumMicroOps = 2; let ResourceCycles = [2]; } def: InstRW<[BWWriteResGroup14], (instrs LFENCE, MFENCE, WAIT, XGETBV)>; def BWWriteResGroup15 : SchedWriteRes<[BWPort0,BWPort5]> { let Latency = 2; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup15], (instregex "(V?)CVTPS2PDrr", "(V?)CVTSS2SDrr")>; def BWWriteResGroup16 : SchedWriteRes<[BWPort6,BWPort0156]> { let Latency = 2; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup16], (instregex "CLFLUSH")>; def BWWriteResGroup17 : SchedWriteRes<[BWPort01,BWPort015]> { let Latency = 2; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup17], (instrs MMX_MOVDQ2Qrr)>; def BWWriteResGroup18 : SchedWriteRes<[BWPort237,BWPort0156]> { let Latency = 2; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup18], (instrs SFENCE)>; def BWWriteResGroup20 : SchedWriteRes<[BWPort06,BWPort0156]> { let Latency = 2; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup20], (instrs CWD, JCXZ, JECXZ, JRCXZ, ADC8i8, SBB8i8, ADC16i16, SBB16i16, ADC32i32, SBB32i32, ADC64i32, SBB64i32)>; def BWWriteResGroup22 : SchedWriteRes<[BWPort4,BWPort6,BWPort237]> { let Latency = 2; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup22], (instrs FNSTCW16m)>; def BWWriteResGroup24 : SchedWriteRes<[BWPort4,BWPort237,BWPort15]> { let Latency = 2; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup24], (instregex "MOVBE(16|32|64)mr")>; def BWWriteResGroup25 : SchedWriteRes<[BWPort4,BWPort237,BWPort0156]> { let Latency = 2; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup25], (instrs PUSH16r, PUSH32r, PUSH64r, PUSH64i8, STOSB, STOSL, STOSQ, STOSW)>; def: InstRW<[BWWriteResGroup25], (instregex "PUSH(16|32|64)rmr")>; def BWWriteResGroup27 : SchedWriteRes<[BWPort1]> { let Latency = 3; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup27], (instrs MMX_CVTPI2PSirr)>; def: InstRW<[BWWriteResGroup27], (instregex "P(DEP|EXT)(32|64)rr", "(V?)CVTDQ2PS(Y?)rr")>; def BWWriteResGroup28 : SchedWriteRes<[BWPort5]> { let Latency = 3; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup28], (instrs VPBROADCASTBrr, VPBROADCASTWrr)>; def BWWriteResGroup33 : SchedWriteRes<[BWPort5,BWPort0156]> { let Latency = 3; let NumMicroOps = 3; let ResourceCycles = [2,1]; } def: InstRW<[BWWriteResGroup33], (instrs MMX_PACKSSDWirr, MMX_PACKSSWBirr, MMX_PACKUSWBirr)>; def BWWriteResGroup34 : SchedWriteRes<[BWPort6,BWPort0156]> { let Latency = 3; let NumMicroOps = 3; let ResourceCycles = [1,2]; } def: InstRW<[BWWriteResGroup34], (instregex "CLD")>; def BWWriteResGroup35 : SchedWriteRes<[BWPort06,BWPort0156]> { let Latency = 3; let NumMicroOps = 3; let ResourceCycles = [1,2]; } def: InstRW<[BWWriteResGroup35], (instregex "RCL(8|16|32|64)r(1|i)", "RCR(8|16|32|64)r(1|i)")>; def BWWriteResGroup37 : SchedWriteRes<[BWPort4,BWPort6,BWPort237,BWPort0156]> { let Latency = 3; let NumMicroOps = 4; let ResourceCycles = [1,1,1,1]; } def: InstRW<[BWWriteResGroup37], (instregex "CALL(16|32|64)r")>; def BWWriteResGroup38 : SchedWriteRes<[BWPort4,BWPort237,BWPort06,BWPort0156]> { let Latency = 3; let NumMicroOps = 4; let ResourceCycles = [1,1,1,1]; } def: InstRW<[BWWriteResGroup38], (instrs CALL64pcrel32)>; def BWWriteResGroup39 : SchedWriteRes<[BWPort0,BWPort1]> { let Latency = 4; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup39], (instregex "(V?)CVT(T?)SD2SI64rr", "(V?)CVT(T?)SD2SIrr", "(V?)CVT(T?)SS2SI64rr", "(V?)CVT(T?)SS2SIrr")>; def BWWriteResGroup40 : SchedWriteRes<[BWPort0,BWPort5]> { let Latency = 4; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup40], (instrs VCVTPS2PDYrr)>; def BWWriteResGroup41 : SchedWriteRes<[BWPort0,BWPort0156]> { let Latency = 4; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup41], (instrs FNSTSW16r)>; def BWWriteResGroup42 : SchedWriteRes<[BWPort1,BWPort5]> { let Latency = 4; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup42], (instrs MMX_CVTPI2PDirr)>; def: InstRW<[BWWriteResGroup42], (instregex "MMX_CVT(T?)PD2PIirr", "MMX_CVT(T?)PS2PIirr", "(V?)CVTDQ2PDrr", "(V?)CVTPD2PSrr", "(V?)CVTSD2SSrr", "(V?)CVTSI642SDrr", "(V?)CVTSI2SDrr", "(V?)CVTSI2SSrr", "(V?)CVT(T?)PD2DQrr")>; def BWWriteResGroup43 : SchedWriteRes<[BWPort0,BWPort4,BWPort237]> { let Latency = 4; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup43], (instrs FNSTSWm)>; def BWWriteResGroup44 : SchedWriteRes<[BWPort1,BWPort4,BWPort237]> { let Latency = 4; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup44], (instregex "IST(T?)_FP(16|32|64)m", "IST_F(16|32)m")>; def BWWriteResGroup45 : SchedWriteRes<[BWPort0156]> { let Latency = 4; let NumMicroOps = 4; let ResourceCycles = [4]; } def: InstRW<[BWWriteResGroup45], (instrs FNCLEX)>; def BWWriteResGroup46 : SchedWriteRes<[]> { let Latency = 0; let NumMicroOps = 4; let ResourceCycles = []; } def: InstRW<[BWWriteResGroup46], (instrs VZEROUPPER)>; def BWWriteResGroup47 : SchedWriteRes<[BWPort0]> { let Latency = 5; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup47], (instregex "MUL_(FPrST0|FST0r|FrST0)")>; def BWWriteResGroup49 : SchedWriteRes<[BWPort23]> { let Latency = 5; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup49], (instregex "MOVSX(16|32|64)rm(8|16|32)", "MOVZX(16|32|64)rm(8|16)")>; def: InstRW<[BWWriteResGroup49], (instrs VBROADCASTSSrm, VMOVDDUPrm, MOVDDUPrm, VMOVSHDUPrm, MOVSHDUPrm, VMOVSLDUPrm, MOVSLDUPrm, VPBROADCASTDrm, VPBROADCASTQrm)>; def BWWriteResGroup50 : SchedWriteRes<[BWPort1,BWPort5]> { let Latency = 5; let NumMicroOps = 3; let ResourceCycles = [1,2]; } def: InstRW<[BWWriteResGroup50], (instregex "(V?)CVTSI642SSrr")>; def BWWriteResGroup51 : SchedWriteRes<[BWPort1,BWPort6,BWPort06]> { let Latency = 5; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup51], (instregex "STR(16|32|64)r")>; def BWWriteResGroup54 : SchedWriteRes<[BWPort6,BWPort0156]> { let Latency = 5; let NumMicroOps = 5; let ResourceCycles = [1,4]; } def: InstRW<[BWWriteResGroup54], (instrs PAUSE)>; def BWWriteResGroup55 : SchedWriteRes<[BWPort06,BWPort0156]> { let Latency = 5; let NumMicroOps = 5; let ResourceCycles = [1,4]; } def: InstRW<[BWWriteResGroup55], (instrs XSETBV)>; def BWWriteResGroup57 : SchedWriteRes<[BWPort4,BWPort237,BWPort0156]> { let Latency = 5; let NumMicroOps = 6; let ResourceCycles = [1,1,4]; } def: InstRW<[BWWriteResGroup57], (instregex "PUSHF(16|64)")>; def BWWriteResGroup58 : SchedWriteRes<[BWPort23]> { let Latency = 6; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup58], (instregex "LD_F(32|64|80)m")>; def: InstRW<[BWWriteResGroup58], (instrs VBROADCASTF128, VBROADCASTI128, VBROADCASTSDYrm, VBROADCASTSSYrm, VMOVDDUPYrm, VMOVSHDUPYrm, VMOVSLDUPYrm, VPBROADCASTDYrm, VPBROADCASTQYrm)>; def BWWriteResGroup59 : SchedWriteRes<[BWPort0,BWPort23]> { let Latency = 6; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup59], (instrs CVTPS2PDrm, VCVTPS2PDrm, CVTSS2SDrm, VCVTSS2SDrm, CVTSS2SDrm_Int, VCVTSS2SDrm_Int, VPSLLVQrm, VPSRLVQrm)>; def BWWriteResGroup60 : SchedWriteRes<[BWPort1,BWPort5]> { let Latency = 6; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup60], (instrs VCVTDQ2PDYrr, VCVTPD2PSYrr, VCVTPD2DQYrr, VCVTTPD2DQYrr)>; def BWWriteResGroup62 : SchedWriteRes<[BWPort6,BWPort23]> { let Latency = 6; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup62], (instrs FARJMP64)>; def: InstRW<[BWWriteResGroup62], (instregex "JMP(16|32|64)m")>; def BWWriteResGroup64 : SchedWriteRes<[BWPort23,BWPort15]> { let Latency = 6; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup64], (instregex "ANDN(32|64)rm", "MOVBE(16|32|64)rm")>; def BWWriteResGroup65 : SchedWriteRes<[BWPort23,BWPort015]> { let Latency = 6; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup65], (instrs VINSERTF128rm, VINSERTI128rm, VPBLENDDrmi)>; def BWWriteResGroup66 : SchedWriteRes<[BWPort23,BWPort0156]> { let Latency = 6; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup66], (instrs POP16r, POP32r, POP64r)>; def: InstRW<[BWWriteResGroup66], (instregex "POP(16|32|64)rmr")>; def BWWriteResGroup68 : SchedWriteRes<[BWPort1,BWPort6,BWPort06,BWPort0156]> { let Latency = 6; let NumMicroOps = 4; let ResourceCycles = [1,1,1,1]; } def: InstRW<[BWWriteResGroup68], (instregex "SLDT(16|32|64)r")>; def BWWriteResGroup69 : SchedWriteRes<[BWPort4,BWPort23,BWPort237,BWPort06]> { let Latency = 6; let NumMicroOps = 4; let ResourceCycles = [1,1,1,1]; } def: InstRW<[BWWriteResGroup69], (instregex "SAR(8|16|32|64)m(1|i)", "SHL(8|16|32|64)m(1|i)", "SHR(8|16|32|64)m(1|i)")>; def BWWriteResGroup70 : SchedWriteRes<[BWPort4,BWPort23,BWPort237,BWPort0156]> { let Latency = 6; let NumMicroOps = 4; let ResourceCycles = [1,1,1,1]; } def: InstRW<[BWWriteResGroup70], (instregex "POP(16|32|64)rmm", "PUSH(16|32|64)rmm")>; def BWWriteResGroup71 : SchedWriteRes<[BWPort6,BWPort0156]> { let Latency = 6; let NumMicroOps = 6; let ResourceCycles = [1,5]; } def: InstRW<[BWWriteResGroup71], (instrs STD)>; def BWWriteResGroup73 : SchedWriteRes<[BWPort0,BWPort23]> { let Latency = 7; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup73], (instrs VPSLLVQYrm, VPSRLVQYrm)>; def BWWriteResGroup74 : SchedWriteRes<[BWPort1,BWPort23]> { let Latency = 7; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup74], (instregex "FCOM(P?)(32|64)m")>; def BWWriteResGroup77 : SchedWriteRes<[BWPort23,BWPort015]> { let Latency = 7; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup77], (instrs VPBLENDDYrmi)>; def BWWriteResGroup79 : SchedWriteRes<[BWPort5,BWPort23]> { let Latency = 7; let NumMicroOps = 3; let ResourceCycles = [2,1]; } def: InstRW<[BWWriteResGroup79], (instrs MMX_PACKSSDWirm, MMX_PACKSSWBirm, MMX_PACKUSWBirm)>; def BWWriteResGroup80 : SchedWriteRes<[BWPort23,BWPort0156]> { let Latency = 7; let NumMicroOps = 3; let ResourceCycles = [1,2]; } def: InstRW<[BWWriteResGroup80], (instrs LEAVE, LEAVE64, SCASB, SCASL, SCASQ, SCASW)>; def BWWriteResGroup82 : SchedWriteRes<[BWPort0,BWPort01,BWPort23]> { let Latency = 7; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup82], (instrs FLDCW16m)>; def BWWriteResGroup84 : SchedWriteRes<[BWPort6,BWPort23,BWPort0156]> { let Latency = 7; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup84], (instrs LRETQ, RETQ)>; def BWWriteResGroup87 : SchedWriteRes<[BWPort4,BWPort23,BWPort237,BWPort06]> { let Latency = 7; let NumMicroOps = 5; let ResourceCycles = [1,1,1,2]; } def: InstRW<[BWWriteResGroup87], (instregex "ROL(8|16|32|64)m(1|i)", "ROR(8|16|32|64)m(1|i)")>; def BWWriteResGroup87_1 : SchedWriteRes<[BWPort06]> { let Latency = 2; let NumMicroOps = 2; let ResourceCycles = [2]; } def: InstRW<[BWWriteResGroup87_1], (instrs ROL8r1, ROL16r1, ROL32r1, ROL64r1, ROR8r1, ROR16r1, ROR32r1, ROR64r1)>; def BWWriteResGroup88 : SchedWriteRes<[BWPort4,BWPort23,BWPort237,BWPort0156]> { let Latency = 7; let NumMicroOps = 5; let ResourceCycles = [1,1,1,2]; } def: InstRW<[BWWriteResGroup88], (instregex "XADD(8|16|32|64)rm")>; def BWWriteResGroup89 : SchedWriteRes<[BWPort4,BWPort6,BWPort23,BWPort237,BWPort0156]> { let Latency = 7; let NumMicroOps = 5; let ResourceCycles = [1,1,1,1,1]; } def: InstRW<[BWWriteResGroup89], (instregex "CALL(16|32|64)m")>; def: InstRW<[BWWriteResGroup89], (instrs FARCALL64)>; def BWWriteResGroup90 : SchedWriteRes<[BWPort6,BWPort06,BWPort15,BWPort0156]> { let Latency = 7; let NumMicroOps = 7; let ResourceCycles = [2,2,1,2]; } def: InstRW<[BWWriteResGroup90], (instrs LOOP)>; def BWWriteResGroup91 : SchedWriteRes<[BWPort1,BWPort23]> { let Latency = 8; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup91], (instrs MMX_CVTPI2PSirm, CVTDQ2PSrm, VCVTDQ2PSrm)>; def: InstRW<[BWWriteResGroup91], (instregex "P(DEP|EXT)(32|64)rm")>; def BWWriteResGroup92 : SchedWriteRes<[BWPort5,BWPort23]> { let Latency = 8; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup92], (instrs VPMOVSXBDYrm, VPMOVSXBQYrm, VPMOVSXBWYrm, VPMOVSXDQYrm, VPMOVSXWDYrm, VPMOVSXWQYrm, VPMOVZXWDYrm)>; def BWWriteResGroup97 : SchedWriteRes<[BWPort23,BWPort237,BWPort06,BWPort0156]> { let Latency = 8; let NumMicroOps = 5; let ResourceCycles = [1,1,1,2]; } def: InstRW<[BWWriteResGroup97], (instregex "RCL(8|16|32|64)m(1|i)", "RCR(8|16|32|64)m(1|i)")>; def BWWriteResGroup99 : SchedWriteRes<[BWPort4,BWPort23,BWPort237,BWPort0156]> { let Latency = 8; let NumMicroOps = 6; let ResourceCycles = [1,1,1,3]; } def: InstRW<[BWWriteResGroup99], (instregex "XCHG(8|16|32|64)rm")>; def BWWriteResGroup100 : SchedWriteRes<[BWPort4,BWPort23,BWPort237,BWPort06,BWPort0156]> { let Latency = 8; let NumMicroOps = 6; let ResourceCycles = [1,1,1,2,1]; } def : SchedAlias; def: InstRW<[BWWriteResGroup100], (instregex "ROL(8|16|32|64)mCL", "ROR(8|16|32|64)mCL", "SAR(8|16|32|64)mCL", "SHL(8|16|32|64)mCL", "SHR(8|16|32|64)mCL")>; def BWWriteResGroup101 : SchedWriteRes<[BWPort1,BWPort23]> { let Latency = 9; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup101], (instregex "(ADD|SUB|SUBR)_F(32|64)m", "ILD_F(16|32|64)m")>; def: InstRW<[BWWriteResGroup101], (instrs VCVTPS2DQYrm, VCVTTPS2DQYrm)>; def BWWriteResGroup105 : SchedWriteRes<[BWPort0,BWPort1,BWPort23]> { let Latency = 9; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup105], (instregex "(V?)CVTSS2SI(64)?rm", "(V?)CVT(T?)SD2SI64rm", "(V?)CVT(T?)SD2SIrm", "VCVTTSS2SI64rm", "(V?)CVTTSS2SIrm")>; def BWWriteResGroup106 : SchedWriteRes<[BWPort0,BWPort5,BWPort23]> { let Latency = 9; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup106], (instrs VCVTPS2PDYrm)>; def BWWriteResGroup107 : SchedWriteRes<[BWPort1,BWPort5,BWPort23]> { let Latency = 9; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup107], (instrs CVTPD2PSrm, CVTPD2DQrm, CVTTPD2DQrm, MMX_CVTPI2PDirm)>; def: InstRW<[BWWriteResGroup107], (instregex "MMX_CVT(T?)PD2PIirm", "(V?)CVTDQ2PDrm", "(V?)CVTSD2SSrm")>; def BWWriteResGroup108 : SchedWriteRes<[BWPort5,BWPort23,BWPort015]> { let Latency = 9; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup108], (instregex "VPBROADCASTB(Y?)rm", "VPBROADCASTW(Y?)rm")>; def BWWriteResGroup112 : SchedWriteRes<[BWPort23,BWPort06,BWPort0156]> { let Latency = 9; let NumMicroOps = 5; let ResourceCycles = [1,1,3]; } def: InstRW<[BWWriteResGroup112], (instrs RDRAND16r, RDRAND32r, RDRAND64r)>; def BWWriteResGroup113 : SchedWriteRes<[BWPort1,BWPort6,BWPort23,BWPort0156]> { let Latency = 9; let NumMicroOps = 5; let ResourceCycles = [1,2,1,1]; } def: InstRW<[BWWriteResGroup113], (instregex "LAR(16|32|64)rm", "LSL(16|32|64)rm")>; def BWWriteResGroup115 : SchedWriteRes<[BWPort0,BWPort23]> { let Latency = 10; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup115], (instregex "(V?)PCMPGTQrm")>; def BWWriteResGroup117 : SchedWriteRes<[BWPort1,BWPort23]> { let Latency = 10; let NumMicroOps = 3; let ResourceCycles = [2,1]; } def: InstRW<[BWWriteResGroup117], (instregex "FICOM(P?)(16|32)m")>; def BWWriteResGroup120 : SchedWriteRes<[BWPort0,BWPort1,BWPort5,BWPort23]> { let Latency = 10; let NumMicroOps = 4; let ResourceCycles = [1,1,1,1]; } def: InstRW<[BWWriteResGroup120], (instregex "CVTTSS2SI64rm")>; def BWWriteResGroup122_1 : SchedWriteRes<[BWPort0,BWFPDivider]> { let Latency = 11; let NumMicroOps = 1; let ResourceCycles = [1,3]; // Really 2.5 cycle throughput } def : SchedAlias; // TODO - convert to ZnWriteResFpuPair def BWWriteResGroup123 : SchedWriteRes<[BWPort0,BWPort23]> { let Latency = 11; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup123], (instregex "MUL_F(32|64)m")>; def: InstRW<[BWWriteResGroup123], (instrs VPCMPGTQYrm)>; def BWWriteResGroup128 : SchedWriteRes<[BWPort1,BWPort5,BWPort23]> { let Latency = 11; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup128], (instrs VCVTDQ2PDYrm)>; def BWWriteResGroup131 : SchedWriteRes<[BWPort1,BWPort06,BWPort0156]> { let Latency = 11; let NumMicroOps = 7; let ResourceCycles = [2,2,3]; } def: InstRW<[BWWriteResGroup131], (instregex "RCL(16|32|64)rCL", "RCR(16|32|64)rCL")>; def BWWriteResGroup132 : SchedWriteRes<[BWPort1,BWPort06,BWPort15,BWPort0156]> { let Latency = 11; let NumMicroOps = 9; let ResourceCycles = [1,4,1,3]; } def: InstRW<[BWWriteResGroup132], (instrs RCL8rCL)>; def BWWriteResGroup133 : SchedWriteRes<[BWPort06,BWPort0156]> { let Latency = 11; let NumMicroOps = 11; let ResourceCycles = [2,9]; } def: InstRW<[BWWriteResGroup133], (instrs LOOPE)>; def: InstRW<[BWWriteResGroup133], (instrs LOOPNE)>; def BWWriteResGroup135 : SchedWriteRes<[BWPort1,BWPort23]> { let Latency = 12; let NumMicroOps = 3; let ResourceCycles = [2,1]; } def: InstRW<[BWWriteResGroup135], (instregex "(ADD|SUB|SUBR)_FI(16|32)m")>; def BWWriteResGroup139_1 : SchedWriteRes<[BWPort0,BWFPDivider]> { let Latency = 14; let NumMicroOps = 1; let ResourceCycles = [1,4]; } def : SchedAlias; // TODO - convert to ZnWriteResFpuPair def BWWriteResGroup141 : SchedWriteRes<[BWPort0,BWPort1,BWPort23]> { let Latency = 14; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup141], (instregex "MUL_FI(16|32)m")>; def BWWriteResGroup144 : SchedWriteRes<[BWPort1,BWPort6,BWPort23,BWPort0156]> { let Latency = 14; let NumMicroOps = 8; let ResourceCycles = [2,2,1,3]; } def: InstRW<[BWWriteResGroup144], (instregex "LAR(16|32|64)rr")>; def BWWriteResGroup145 : SchedWriteRes<[BWPort1,BWPort06,BWPort15,BWPort0156]> { let Latency = 14; let NumMicroOps = 10; let ResourceCycles = [2,3,1,4]; } def: InstRW<[BWWriteResGroup145], (instrs RCR8rCL)>; def BWWriteResGroup146 : SchedWriteRes<[BWPort0,BWPort1,BWPort6,BWPort0156]> { let Latency = 14; let NumMicroOps = 12; let ResourceCycles = [2,1,4,5]; } def: InstRW<[BWWriteResGroup146], (instrs XCH_F)>; def BWWriteResGroup147 : SchedWriteRes<[BWPort0]> { let Latency = 15; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup147], (instregex "DIVR_(FPrST0|FST0r|FrST0)")>; def BWWriteResGroup149 : SchedWriteRes<[BWPort1,BWPort23,BWPort237,BWPort06,BWPort15,BWPort0156]> { let Latency = 15; let NumMicroOps = 10; let ResourceCycles = [1,1,1,4,1,2]; } def: InstRW<[BWWriteResGroup149], (instregex "RCL(8|16|32|64)mCL")>; def BWWriteResGroup150 : SchedWriteRes<[BWPort0,BWPort23,BWFPDivider]> { let Latency = 16; let NumMicroOps = 2; let ResourceCycles = [1,1,5]; } def : SchedAlias; // TODO - convert to ZnWriteResFpuPair def BWWriteResGroup153 : SchedWriteRes<[BWPort4,BWPort23,BWPort237,BWPort06,BWPort15,BWPort0156]> { let Latency = 16; let NumMicroOps = 14; let ResourceCycles = [1,1,1,4,2,5]; } def: InstRW<[BWWriteResGroup153], (instrs CMPXCHG8B)>; def BWWriteResGroup154 : SchedWriteRes<[BWPort5,BWPort6]> { let Latency = 8; let NumMicroOps = 20; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup154], (instrs VZEROALL)>; def BWWriteResGroup159 : SchedWriteRes<[BWPort5,BWPort6,BWPort06,BWPort0156]> { let Latency = 18; let NumMicroOps = 8; let ResourceCycles = [1,1,1,5]; } def: InstRW<[BWWriteResGroup159], (instrs CPUID)>; def: InstRW<[BWWriteResGroup159], (instrs RDTSC)>; def BWWriteResGroup160 : SchedWriteRes<[BWPort1,BWPort23,BWPort237,BWPort06,BWPort15,BWPort0156]> { let Latency = 18; let NumMicroOps = 11; let ResourceCycles = [2,1,1,3,1,3]; } def: InstRW<[BWWriteResGroup160], (instregex "RCR(8|16|32|64)mCL")>; def BWWriteResGroup161 : SchedWriteRes<[BWPort0,BWPort23,BWFPDivider]> { let Latency = 19; let NumMicroOps = 2; let ResourceCycles = [1,1,8]; } def : SchedAlias; // TODO - convert to ZnWriteResFpuPair def BWWriteResGroup165 : SchedWriteRes<[BWPort0]> { let Latency = 20; let NumMicroOps = 1; let ResourceCycles = [1]; } def: InstRW<[BWWriteResGroup165], (instregex "DIV_(FPrST0|FST0r|FrST0)")>; def BWWriteResGroup167 : SchedWriteRes<[BWPort4,BWPort5,BWPort6,BWPort23,BWPort237,BWPort06,BWPort0156]> { let Latency = 20; let NumMicroOps = 8; let ResourceCycles = [1,1,1,1,1,1,2]; } def: InstRW<[BWWriteResGroup167], (instrs INSB, INSL, INSW)>; def BWWriteResGroup169 : SchedWriteRes<[BWPort0,BWPort23]> { let Latency = 21; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup169], (instregex "DIV_F(32|64)m")>; def BWWriteResGroup171 : SchedWriteRes<[BWPort0,BWPort4,BWPort5,BWPort23,BWPort237,BWPort06,BWPort0156]> { let Latency = 21; let NumMicroOps = 19; let ResourceCycles = [2,1,4,1,1,4,6]; } def: InstRW<[BWWriteResGroup171], (instrs CMPXCHG16B)>; def BWWriteResGroup172 : SchedWriteRes<[BWPort6,BWPort23,BWPort0156]> { let Latency = 22; let NumMicroOps = 18; let ResourceCycles = [1,1,16]; } def: InstRW<[BWWriteResGroup172], (instrs POPF64)>; def BWWriteResGroup176 : SchedWriteRes<[BWPort6,BWPort23,BWPort0156]> { let Latency = 23; let NumMicroOps = 19; let ResourceCycles = [3,1,15]; } def: InstRW<[BWWriteResGroup176], (instregex "XRSTOR(64)?")>; def BWWriteResGroup177 : SchedWriteRes<[BWPort0,BWPort1,BWPort23]> { let Latency = 24; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup177], (instregex "DIV_FI(16|32)m")>; def BWWriteResGroup180 : SchedWriteRes<[BWPort0,BWPort23]> { let Latency = 26; let NumMicroOps = 2; let ResourceCycles = [1,1]; } def: InstRW<[BWWriteResGroup180], (instregex "DIVR_F(32|64)m")>; def BWWriteResGroup182 : SchedWriteRes<[BWPort0,BWPort1,BWPort23]> { let Latency = 29; let NumMicroOps = 3; let ResourceCycles = [1,1,1]; } def: InstRW<[BWWriteResGroup182], (instregex "DIVR_FI(16|32)m")>; def BWWriteResGroup183_1 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> { let Latency = 22; let NumMicroOps = 7; let ResourceCycles = [1,3,2,1]; } def: InstRW<[BWWriteResGroup183_1], (instrs VGATHERQPDrm)>; def BWWriteResGroup183_2 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> { let Latency = 23; let NumMicroOps = 9; let ResourceCycles = [1,3,4,1]; } def: InstRW<[BWWriteResGroup183_2], (instrs VGATHERQPDYrm)>; def BWWriteResGroup183_3 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> { let Latency = 24; let NumMicroOps = 9; let ResourceCycles = [1,5,2,1]; } def: InstRW<[BWWriteResGroup183_3], (instrs VGATHERQPSYrm)>; def BWWriteResGroup183_4 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> { let Latency = 25; let NumMicroOps = 7; let ResourceCycles = [1,3,2,1]; } def: InstRW<[BWWriteResGroup183_4], (instrs VGATHERDPDrm, VGATHERDPSrm)>; def BWWriteResGroup183_5 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> { let Latency = 26; let NumMicroOps = 9; let ResourceCycles = [1,5,2,1]; } def: InstRW<[BWWriteResGroup183_5], (instrs VGATHERDPDYrm)>; def BWWriteResGroup183_6 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> { let Latency = 26; let NumMicroOps = 14; let ResourceCycles = [1,4,8,1]; } def: InstRW<[BWWriteResGroup183_6], (instrs VGATHERDPSYrm)>; def BWWriteResGroup183_7 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> { let Latency = 27; let NumMicroOps = 9; let ResourceCycles = [1,5,2,1]; } def: InstRW<[BWWriteResGroup183_7], (instrs VGATHERQPSrm)>; def BWWriteResGroup185 : SchedWriteRes<[BWPort4,BWPort6,BWPort23,BWPort237,BWPort0156]> { let Latency = 29; let NumMicroOps = 27; let ResourceCycles = [1,5,1,1,19]; } def: InstRW<[BWWriteResGroup185], (instrs XSAVE64)>; def BWWriteResGroup186 : SchedWriteRes<[BWPort4,BWPort6,BWPort23,BWPort237,BWPort0156]> { let Latency = 30; let NumMicroOps = 28; let ResourceCycles = [1,6,1,1,19]; } def: InstRW<[BWWriteResGroup186], (instrs XSAVE)>; def: InstRW<[BWWriteResGroup186], (instregex "XSAVEC", "XSAVES", "XSAVEOPT")>; def BWWriteResGroup191 : SchedWriteRes<[BWPort5,BWPort6,BWPort23,BWPort06,BWPort0156]> { let Latency = 34; let NumMicroOps = 23; let ResourceCycles = [1,5,3,4,10]; } def: InstRW<[BWWriteResGroup191], (instregex "IN(8|16|32)ri", "IN(8|16|32)rr")>; def BWWriteResGroup194 : SchedWriteRes<[BWPort5,BWPort6,BWPort23,BWPort237,BWPort06,BWPort0156]> { let Latency = 35; let NumMicroOps = 23; let ResourceCycles = [1,5,2,1,4,10]; } def: InstRW<[BWWriteResGroup194], (instregex "OUT(8|16|32)ir", "OUT(8|16|32)rr")>; def BWWriteResGroup196 : SchedWriteRes<[BWPort5,BWPort0156]> { let Latency = 42; let NumMicroOps = 22; let ResourceCycles = [2,20]; } def: InstRW<[BWWriteResGroup196], (instrs RDTSCP)>; def BWWriteResGroup197 : SchedWriteRes<[BWPort0,BWPort01,BWPort23,BWPort05,BWPort06,BWPort015,BWPort0156]> { let Latency = 60; let NumMicroOps = 64; let ResourceCycles = [2,2,8,1,10,2,39]; } def: InstRW<[BWWriteResGroup197], (instrs FLDENVm)>; def BWWriteResGroup198 : SchedWriteRes<[BWPort0,BWPort6,BWPort23,BWPort05,BWPort06,BWPort15,BWPort0156]> { let Latency = 63; let NumMicroOps = 88; let ResourceCycles = [4,4,31,1,2,1,45]; } def: InstRW<[BWWriteResGroup198], (instrs FXRSTOR64)>; def BWWriteResGroup199 : SchedWriteRes<[BWPort0,BWPort6,BWPort23,BWPort05,BWPort06,BWPort15,BWPort0156]> { let Latency = 63; let NumMicroOps = 90; let ResourceCycles = [4,2,33,1,2,1,47]; } def: InstRW<[BWWriteResGroup199], (instrs FXRSTOR)>; def BWWriteResGroup200 : SchedWriteRes<[BWPort5,BWPort01,BWPort0156]> { let Latency = 75; let NumMicroOps = 15; let ResourceCycles = [6,3,6]; } def: InstRW<[BWWriteResGroup200], (instrs FNINIT)>; def BWWriteResGroup202 : SchedWriteRes<[BWPort0,BWPort1,BWPort4,BWPort5,BWPort6,BWPort237,BWPort06,BWPort0156]> { let Latency = 115; let NumMicroOps = 100; let ResourceCycles = [9,9,11,8,1,11,21,30]; } def: InstRW<[BWWriteResGroup202], (instrs FSTENVm)>; def: InstRW<[WriteZero], (instrs CLC)>; // Intruction variants handled by the renamer. These might not need execution // ports in certain conditions. // See Agner's Fog "The microarchitecture of Intel, AMD and VIA CPUs", // section "Haswell and Broadwell Pipeline" > "Register allocation and // renaming". // These can be investigated with llvm-exegesis, e.g. // echo 'pxor %mm0, %mm0' | /tmp/llvm-exegesis -mode=uops -snippets-file=- // echo 'vxorpd %xmm0, %xmm0, %xmm1' | /tmp/llvm-exegesis -mode=uops -snippets-file=- def BWWriteZeroLatency : SchedWriteRes<[]> { let Latency = 0; } def BWWriteZeroIdiom : SchedWriteVariant<[ SchedVar, [BWWriteZeroLatency]>, SchedVar ]>; def : InstRW<[BWWriteZeroIdiom], (instrs SUB32rr, SUB64rr, XOR32rr, XOR64rr)>; def BWWriteFZeroIdiom : SchedWriteVariant<[ SchedVar, [BWWriteZeroLatency]>, SchedVar ]>; def : InstRW<[BWWriteFZeroIdiom], (instrs XORPSrr, VXORPSrr, XORPDrr, VXORPDrr)>; def BWWriteFZeroIdiomY : SchedWriteVariant<[ SchedVar, [BWWriteZeroLatency]>, SchedVar ]>; def : InstRW<[BWWriteFZeroIdiomY], (instrs VXORPSYrr, VXORPDYrr)>; def BWWriteVZeroIdiomLogicX : SchedWriteVariant<[ SchedVar, [BWWriteZeroLatency]>, SchedVar ]>; def : InstRW<[BWWriteVZeroIdiomLogicX], (instrs PXORrr, VPXORrr)>; def BWWriteVZeroIdiomLogicY : SchedWriteVariant<[ SchedVar, [BWWriteZeroLatency]>, SchedVar ]>; def : InstRW<[BWWriteVZeroIdiomLogicY], (instrs VPXORYrr)>; def BWWriteVZeroIdiomALUX : SchedWriteVariant<[ SchedVar, [BWWriteZeroLatency]>, SchedVar ]>; def : InstRW<[BWWriteVZeroIdiomALUX], (instrs PSUBBrr, VPSUBBrr, PSUBDrr, VPSUBDrr, PSUBQrr, VPSUBQrr, PSUBWrr, VPSUBWrr, PCMPGTBrr, VPCMPGTBrr, PCMPGTDrr, VPCMPGTDrr, PCMPGTWrr, VPCMPGTWrr)>; def BWWriteVZeroIdiomALUY : SchedWriteVariant<[ SchedVar, [BWWriteZeroLatency]>, SchedVar ]>; def : InstRW<[BWWriteVZeroIdiomALUY], (instrs VPSUBBYrr, VPSUBDYrr, VPSUBQYrr, VPSUBWYrr, VPCMPGTBYrr, VPCMPGTDYrr, VPCMPGTWYrr)>; def BWWritePCMPGTQ : SchedWriteRes<[BWPort0]> { let Latency = 5; let NumMicroOps = 1; let ResourceCycles = [1]; } def BWWriteVZeroIdiomPCMPGTQ : SchedWriteVariant<[ SchedVar, [BWWriteZeroLatency]>, SchedVar ]>; def : InstRW<[BWWriteVZeroIdiomPCMPGTQ], (instrs PCMPGTQrr, VPCMPGTQrr, VPCMPGTQYrr)>; // CMOVs that use both Z and C flag require an extra uop. def BWWriteCMOVA_CMOVBErr : SchedWriteRes<[BWPort06,BWPort0156]> { let Latency = 2; let ResourceCycles = [1,1]; let NumMicroOps = 2; } def BWWriteCMOVA_CMOVBErm : SchedWriteRes<[BWPort23,BWPort06,BWPort0156]> { let Latency = 7; let ResourceCycles = [1,1,1]; let NumMicroOps = 3; } def BWCMOVA_CMOVBErr : SchedWriteVariant<[ SchedVar, [BWWriteCMOVA_CMOVBErr]>, SchedVar ]>; def BWCMOVA_CMOVBErm : SchedWriteVariant<[ SchedVar, [BWWriteCMOVA_CMOVBErm]>, SchedVar ]>; def : InstRW<[BWCMOVA_CMOVBErr], (instrs CMOV16rr, CMOV32rr, CMOV64rr)>; def : InstRW<[BWCMOVA_CMOVBErm], (instrs CMOV16rm, CMOV32rm, CMOV64rm)>; // SETCCs that use both Z and C flag require an extra uop. def BWWriteSETA_SETBEr : SchedWriteRes<[BWPort06,BWPort0156]> { let Latency = 2; let ResourceCycles = [1,1]; let NumMicroOps = 2; } def BWWriteSETA_SETBEm : SchedWriteRes<[BWPort4,BWPort237,BWPort06,BWPort0156]> { let Latency = 3; let ResourceCycles = [1,1,1,1]; let NumMicroOps = 4; } def BWSETA_SETBErr : SchedWriteVariant<[ SchedVar, [BWWriteSETA_SETBEr]>, SchedVar ]>; def BWSETA_SETBErm : SchedWriteVariant<[ SchedVar, [BWWriteSETA_SETBEm]>, SchedVar ]>; def : InstRW<[BWSETA_SETBErr], (instrs SETCCr)>; def : InstRW<[BWSETA_SETBErm], (instrs SETCCm)>; } // SchedModel