069f8121e0
We cannot actually retire an infinite number of uops per cycle. This patch adds a RCU to the skylake scheduling model to fix this. I'm purposefully using a loose upper bound here. We're unlikely to actually get four fused uops per cycle, but this is better than not setting anything. Most realistic code I've put through uiCA will retire up to ~6 uops per cycle. Information taken from https://en.wikichip.org/wiki/intel/microarchitectures/skylake_(client). This requires modification of the two zero idiom tests because we do not currently model the CPU frontend which would likely be the actual bottleneck in that case. Related to #153747.
1762 lines
71 KiB
TableGen
1762 lines
71 KiB
TableGen
//=- X86SchedSkylake.td - X86 Skylake Client 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 Skylake Client to support
|
|
// instruction scheduling and other instruction cost heuristics.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
def SkylakeClientModel : SchedMachineModel {
|
|
// All x86 instructions are modeled as a single micro-op, and SKylake can
|
|
// decode 6 instructions per cycle.
|
|
let IssueWidth = 6;
|
|
let MicroOpBufferSize = 224; // Based on the reorder buffer.
|
|
let LoadLatency = 5;
|
|
let MispredictPenalty = 14;
|
|
|
|
// 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 = SkylakeClientModel in {
|
|
|
|
// Skylake Client 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 SKLPort0 : ProcResource<1>;
|
|
def SKLPort1 : ProcResource<1>;
|
|
def SKLPort2 : ProcResource<1>;
|
|
def SKLPort3 : ProcResource<1>;
|
|
def SKLPort4 : ProcResource<1>;
|
|
def SKLPort5 : ProcResource<1>;
|
|
def SKLPort6 : ProcResource<1>;
|
|
def SKLPort7 : ProcResource<1>;
|
|
|
|
// Many micro-ops are capable of issuing on multiple ports.
|
|
def SKLPort01 : ProcResGroup<[SKLPort0, SKLPort1]>;
|
|
def SKLPort23 : ProcResGroup<[SKLPort2, SKLPort3]>;
|
|
def SKLPort237 : ProcResGroup<[SKLPort2, SKLPort3, SKLPort7]>;
|
|
def SKLPort04 : ProcResGroup<[SKLPort0, SKLPort4]>;
|
|
def SKLPort05 : ProcResGroup<[SKLPort0, SKLPort5]>;
|
|
def SKLPort06 : ProcResGroup<[SKLPort0, SKLPort6]>;
|
|
def SKLPort15 : ProcResGroup<[SKLPort1, SKLPort5]>;
|
|
def SKLPort16 : ProcResGroup<[SKLPort1, SKLPort6]>;
|
|
def SKLPort56 : ProcResGroup<[SKLPort5, SKLPort6]>;
|
|
def SKLPort015 : ProcResGroup<[SKLPort0, SKLPort1, SKLPort5]>;
|
|
def SKLPort056 : ProcResGroup<[SKLPort0, SKLPort5, SKLPort6]>;
|
|
def SKLPort0156: ProcResGroup<[SKLPort0, SKLPort1, SKLPort5, SKLPort6]>;
|
|
|
|
def SKLDivider : ProcResource<1>; // Integer division issued on port 0.
|
|
// FP division and sqrt on port 0.
|
|
def SKLFPDivider : ProcResource<1>;
|
|
|
|
// 60 Entry Unified Scheduler
|
|
def SKLPortAny : ProcResGroup<[SKLPort0, SKLPort1, SKLPort2, SKLPort3, SKLPort4,
|
|
SKLPort5, SKLPort6, SKLPort7]> {
|
|
let BufferSize=60;
|
|
}
|
|
|
|
// Skylake can retire up to four (potentially fused) uops per cycle. Set the
|
|
// limit to twice that given we do not model fused uops as only taking up one
|
|
// retirement slot. I could not find any documented sources on how many
|
|
// in-flight micro-ops can be tracked.
|
|
def SKRCU : RetireControlUnit<0, 8>;
|
|
|
|
// Integer loads are 5 cycles, so ReadAfterLd registers needn't be available until 5
|
|
// cycles after the memory operand.
|
|
def : ReadAdvance<ReadAfterLd, 5>;
|
|
|
|
// Vector loads are 5/6/7 cycles, so ReadAfterVec*Ld registers needn't be available
|
|
// until 5/6/7 cycles after the memory operand.
|
|
def : ReadAdvance<ReadAfterVecLd, 5>;
|
|
def : ReadAdvance<ReadAfterVecXLd, 6>;
|
|
def : ReadAdvance<ReadAfterVecYLd, 7>;
|
|
|
|
def : ReadAdvance<ReadInt2Fpu, 0>;
|
|
|
|
// 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 SKLWriteResPair<X86FoldableSchedWrite SchedRW,
|
|
list<ProcResourceKind> ExePorts,
|
|
int Lat, list<int> Res = [1], int UOps = 1,
|
|
int LoadLat = 5, int LoadUOps = 1> {
|
|
// Register variant is using a single cycle on ExePort.
|
|
def : WriteRes<SchedRW, ExePorts> {
|
|
let Latency = Lat;
|
|
let ReleaseAtCycles = 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<SchedRW.Folded, !listconcat([SKLPort23], ExePorts)> {
|
|
let Latency = !add(Lat, LoadLat);
|
|
let ReleaseAtCycles = !listconcat([1], Res);
|
|
let NumMicroOps = !add(UOps, LoadUOps);
|
|
}
|
|
}
|
|
|
|
// 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<WriteRMW, [SKLPort237,SKLPort4]>;
|
|
|
|
// Arithmetic.
|
|
defm : SKLWriteResPair<WriteALU, [SKLPort0156], 1>; // Simple integer ALU op.
|
|
defm : SKLWriteResPair<WriteADC, [SKLPort06], 1>; // Integer ALU + flags op.
|
|
|
|
// Integer multiplication.
|
|
defm : SKLWriteResPair<WriteIMul8, [SKLPort1], 3>;
|
|
defm : SKLWriteResPair<WriteIMul16, [SKLPort1,SKLPort06,SKLPort0156], 4, [1,1,2], 4>;
|
|
defm : X86WriteRes<WriteIMul16Imm, [SKLPort1,SKLPort0156], 4, [1,1], 2>;
|
|
defm : X86WriteRes<WriteIMul16ImmLd, [SKLPort1,SKLPort0156,SKLPort23], 8, [1,1,1], 3>;
|
|
defm : SKLWriteResPair<WriteIMul16Reg, [SKLPort1], 3>;
|
|
defm : SKLWriteResPair<WriteIMul32, [SKLPort1,SKLPort06,SKLPort0156], 4, [1,1,1], 3>;
|
|
defm : SKLWriteResPair<WriteMULX32, [SKLPort1,SKLPort06,SKLPort0156], 3, [1,1,1], 3>;
|
|
defm : SKLWriteResPair<WriteIMul32Imm, [SKLPort1], 3>;
|
|
defm : SKLWriteResPair<WriteIMul32Reg, [SKLPort1], 3>;
|
|
defm : SKLWriteResPair<WriteIMul64, [SKLPort1,SKLPort5], 4, [1,1], 2>;
|
|
defm : SKLWriteResPair<WriteMULX64, [SKLPort1,SKLPort5], 3, [1,1], 2>;
|
|
defm : SKLWriteResPair<WriteIMul64Imm, [SKLPort1], 3>;
|
|
defm : SKLWriteResPair<WriteIMul64Reg, [SKLPort1], 3>;
|
|
def SKLWriteIMulH : WriteRes<WriteIMulH, []> { let Latency = 4; }
|
|
def : WriteRes<WriteIMulHLd, []> {
|
|
let Latency = !add(SKLWriteIMulH.Latency, SkylakeClientModel.LoadLatency);
|
|
}
|
|
|
|
defm : X86WriteRes<WriteBSWAP32, [SKLPort15], 1, [1], 1>;
|
|
defm : X86WriteRes<WriteBSWAP64, [SKLPort06, SKLPort15], 2, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCMPXCHG,[SKLPort06, SKLPort0156], 5, [2,3], 5>;
|
|
defm : X86WriteRes<WriteCMPXCHGRMW,[SKLPort23,SKLPort06,SKLPort0156,SKLPort237,SKLPort4], 8, [1,2,1,1,1], 6>;
|
|
defm : X86WriteRes<WriteXCHG, [SKLPort0156], 2, [3], 3>;
|
|
|
|
// TODO: Why isn't the SKLDivider used?
|
|
defm : SKLWriteResPair<WriteDiv8, [SKLPort0,SKLDivider], 25, [1,10], 1, 4>;
|
|
defm : X86WriteRes<WriteDiv16, [SKLPort0,SKLPort1,SKLPort5,SKLPort6,SKLPort05,SKLPort0156], 76, [7,2,8,3,1,11], 32>;
|
|
defm : X86WriteRes<WriteDiv32, [SKLPort0,SKLPort1,SKLPort5,SKLPort6,SKLPort05,SKLPort0156], 76, [7,2,8,3,1,11], 32>;
|
|
defm : X86WriteRes<WriteDiv64, [SKLPort0,SKLPort1,SKLPort5,SKLPort6,SKLPort05,SKLPort0156], 76, [7,2,8,3,1,11], 32>;
|
|
defm : X86WriteRes<WriteDiv16Ld, [SKLPort0,SKLPort23,SKLDivider], 29, [1,1,10], 2>;
|
|
defm : X86WriteRes<WriteDiv32Ld, [SKLPort0,SKLPort23,SKLDivider], 29, [1,1,10], 2>;
|
|
defm : X86WriteRes<WriteDiv64Ld, [SKLPort0,SKLPort23,SKLDivider], 29, [1,1,10], 2>;
|
|
|
|
defm : X86WriteRes<WriteIDiv8, [SKLPort0,SKLDivider], 25, [1,10], 1>;
|
|
defm : X86WriteRes<WriteIDiv16, [SKLPort0,SKLPort1,SKLPort5,SKLPort6,SKLPort06,SKLPort0156], 102, [4,2,4,8,14,34], 66>;
|
|
defm : X86WriteRes<WriteIDiv32, [SKLPort0,SKLPort1,SKLPort5,SKLPort6,SKLPort06,SKLPort0156], 102, [4,2,4,8,14,34], 66>;
|
|
defm : X86WriteRes<WriteIDiv64, [SKLPort0,SKLPort1,SKLPort5,SKLPort6,SKLPort06,SKLPort0156], 102, [4,2,4,8,14,34], 66>;
|
|
defm : X86WriteRes<WriteIDiv8Ld, [SKLPort0,SKLPort5,SKLPort23,SKLPort0156], 28, [2,4,1,1], 8>;
|
|
defm : X86WriteRes<WriteIDiv16Ld, [SKLPort0,SKLPort5,SKLPort23,SKLPort0156], 28, [2,4,1,1], 8>;
|
|
defm : X86WriteRes<WriteIDiv32Ld, [SKLPort0,SKLPort5,SKLPort23,SKLPort0156], 28, [2,4,1,1], 8>;
|
|
defm : X86WriteRes<WriteIDiv64Ld, [SKLPort0,SKLPort5,SKLPort23,SKLPort0156], 28, [2,4,1,1], 8>;
|
|
|
|
defm : SKLWriteResPair<WriteCRC32, [SKLPort1], 3>;
|
|
|
|
def : WriteRes<WriteLEA, [SKLPort15]>; // LEA instructions can't fold loads.
|
|
|
|
defm : SKLWriteResPair<WriteCMOV, [SKLPort06], 1, [1], 1>; // Conditional move.
|
|
defm : X86WriteRes<WriteFCMOV, [SKLPort1], 3, [1], 1>; // x87 conditional move.
|
|
def : WriteRes<WriteSETCC, [SKLPort06]>; // Setcc.
|
|
def : WriteRes<WriteSETCCStore, [SKLPort06,SKLPort4,SKLPort237]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 3;
|
|
}
|
|
|
|
defm : X86WriteRes<WriteLAHFSAHF, [SKLPort06], 1, [1], 1>;
|
|
defm : X86WriteRes<WriteBitTest, [SKLPort06], 1, [1], 1>;
|
|
defm : X86WriteRes<WriteBitTestImmLd, [SKLPort06,SKLPort23], 6, [1,1], 2>;
|
|
defm : X86WriteRes<WriteBitTestRegLd, [SKLPort0156,SKLPort23], 6, [1,1], 2>;
|
|
defm : X86WriteRes<WriteBitTestSet, [SKLPort06], 1, [1], 1>;
|
|
defm : X86WriteRes<WriteBitTestSetImmLd, [SKLPort06,SKLPort23], 5, [1,1], 3>;
|
|
defm : X86WriteRes<WriteBitTestSetRegLd, [SKLPort0156,SKLPort23], 5, [1,1], 2>;
|
|
|
|
// Bit counts.
|
|
defm : SKLWriteResPair<WriteBSF, [SKLPort1], 3>;
|
|
defm : SKLWriteResPair<WriteBSR, [SKLPort1], 3>;
|
|
defm : SKLWriteResPair<WriteLZCNT, [SKLPort1], 3>;
|
|
defm : SKLWriteResPair<WriteTZCNT, [SKLPort1], 3>;
|
|
defm : SKLWriteResPair<WritePOPCNT, [SKLPort1], 3>;
|
|
|
|
// Integer shifts and rotates.
|
|
defm : SKLWriteResPair<WriteShift, [SKLPort06], 1>;
|
|
defm : SKLWriteResPair<WriteShiftCL, [SKLPort06], 3, [3], 3>;
|
|
defm : SKLWriteResPair<WriteRotate, [SKLPort06], 1, [1], 1>;
|
|
defm : SKLWriteResPair<WriteRotateCL, [SKLPort06], 3, [3], 3>;
|
|
|
|
// SHLD/SHRD.
|
|
defm : X86WriteRes<WriteSHDrri, [SKLPort1], 3, [1], 1>;
|
|
defm : X86WriteRes<WriteSHDrrcl,[SKLPort1,SKLPort06,SKLPort0156], 6, [1, 2, 1], 4>;
|
|
defm : X86WriteRes<WriteSHDmri, [SKLPort1,SKLPort23,SKLPort237,SKLPort0156], 9, [1, 1, 1, 1], 4>;
|
|
defm : X86WriteRes<WriteSHDmrcl,[SKLPort1,SKLPort23,SKLPort237,SKLPort06,SKLPort0156], 11, [1, 1, 1, 2, 1], 6>;
|
|
|
|
// BMI1 BEXTR/BLS, BMI2 BZHI
|
|
defm : SKLWriteResPair<WriteBEXTR, [SKLPort06,SKLPort15], 2, [1,1], 2>;
|
|
defm : SKLWriteResPair<WriteBLS, [SKLPort15], 1>;
|
|
defm : SKLWriteResPair<WriteBZHI, [SKLPort15], 1>;
|
|
|
|
// Loads, stores, and moves, not folded with other operations.
|
|
defm : X86WriteRes<WriteLoad, [SKLPort23], 5, [1], 1>;
|
|
defm : X86WriteRes<WriteStore, [SKLPort237, SKLPort4], 1, [1,1], 1>;
|
|
defm : X86WriteRes<WriteStoreNT, [SKLPort237, SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteMove, [SKLPort0156], 1, [1], 1>;
|
|
|
|
// Model the effect of clobbering the read-write mask operand of the GATHER operation.
|
|
// Does not cost anything by itself, only has latency, matching that of the WriteLoad,
|
|
defm : X86WriteRes<WriteVecMaskedGatherWriteback, [], 5, [], 0>;
|
|
|
|
// Idioms that clear a register, like xorps %xmm0, %xmm0.
|
|
// These can often bypass execution ports completely.
|
|
def : WriteRes<WriteZero, []>;
|
|
|
|
// Branches don't produce values, so they have no latency, but they still
|
|
// consume resources. Indirect branches can fold loads.
|
|
defm : SKLWriteResPair<WriteJump, [SKLPort06], 1>;
|
|
|
|
// Floating point. This covers both scalar and vector operations.
|
|
defm : X86WriteRes<WriteFLD0, [SKLPort05], 1, [1], 1>;
|
|
defm : X86WriteRes<WriteFLD1, [SKLPort05], 1, [2], 2>;
|
|
defm : X86WriteRes<WriteFLDC, [SKLPort05], 1, [2], 2>;
|
|
defm : X86WriteRes<WriteFLoad, [SKLPort23], 5, [1], 1>;
|
|
defm : X86WriteRes<WriteFLoadX, [SKLPort23], 6, [1], 1>;
|
|
defm : X86WriteRes<WriteFLoadY, [SKLPort23], 7, [1], 1>;
|
|
defm : X86WriteRes<WriteFMaskedLoad, [SKLPort23,SKLPort015], 7, [1,1], 2>;
|
|
defm : X86WriteRes<WriteFMaskedLoadY, [SKLPort23,SKLPort015], 8, [1,1], 2>;
|
|
defm : X86WriteRes<WriteFStore, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteFStoreX, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteFStoreY, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteFStoreNT, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteFStoreNTX, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteFStoreNTY, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
|
|
defm : X86WriteRes<WriteFMaskedStore32, [SKLPort23,SKLPort0,SKLPort4], 2, [1,1,1], 3>;
|
|
defm : X86WriteRes<WriteFMaskedStore32Y, [SKLPort23,SKLPort0,SKLPort4], 2, [1,1,1], 3>;
|
|
defm : X86WriteRes<WriteFMaskedStore64, [SKLPort23,SKLPort0,SKLPort4], 2, [1,1,1], 3>;
|
|
defm : X86WriteRes<WriteFMaskedStore64Y, [SKLPort23,SKLPort0,SKLPort4], 2, [1,1,1], 3>;
|
|
|
|
defm : X86WriteRes<WriteFMove, [SKLPort015], 1, [1], 1>;
|
|
defm : X86WriteRes<WriteFMoveX, [SKLPort015], 1, [1], 1>;
|
|
defm : X86WriteRes<WriteFMoveY, [SKLPort015], 1, [1], 1>;
|
|
defm : X86WriteResUnsupported<WriteFMoveZ>;
|
|
defm : X86WriteRes<WriteEMMS, [SKLPort05,SKLPort0156], 10, [9,1], 10>;
|
|
|
|
defm : SKLWriteResPair<WriteFAdd, [SKLPort01], 4, [1], 1, 5>; // Floating point add/sub.
|
|
defm : SKLWriteResPair<WriteFAddX, [SKLPort01], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFAddY, [SKLPort01], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFAddZ>;
|
|
defm : SKLWriteResPair<WriteFAdd64, [SKLPort01], 4, [1], 1, 5>; // Floating point double add/sub.
|
|
defm : SKLWriteResPair<WriteFAdd64X, [SKLPort01], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFAdd64Y, [SKLPort01], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFAdd64Z>;
|
|
|
|
defm : SKLWriteResPair<WriteFCmp, [SKLPort01], 4, [1], 1, 5>; // Floating point compare.
|
|
defm : SKLWriteResPair<WriteFCmpX, [SKLPort01], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFCmpY, [SKLPort01], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFCmpZ>;
|
|
defm : SKLWriteResPair<WriteFCmp64, [SKLPort01], 4, [1], 1, 5>; // Floating point double compare.
|
|
defm : SKLWriteResPair<WriteFCmp64X, [SKLPort01], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFCmp64Y, [SKLPort01], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFCmp64Z>;
|
|
|
|
defm : SKLWriteResPair<WriteFCom, [SKLPort0], 2>; // Floating point compare to flags (X87).
|
|
defm : SKLWriteResPair<WriteFComX, [SKLPort0], 2>; // Floating point compare to flags (SSE).
|
|
|
|
defm : SKLWriteResPair<WriteFMul, [SKLPort01], 4, [1], 1, 5>; // Floating point multiplication.
|
|
defm : SKLWriteResPair<WriteFMulX, [SKLPort01], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFMulY, [SKLPort01], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFMulZ>;
|
|
defm : SKLWriteResPair<WriteFMul64, [SKLPort01], 4, [1], 1, 5>; // Floating point double multiplication.
|
|
defm : SKLWriteResPair<WriteFMul64X, [SKLPort01], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFMul64Y, [SKLPort01], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFMul64Z>;
|
|
|
|
defm : SKLWriteResPair<WriteFDiv, [SKLPort0,SKLFPDivider], 11, [1,3], 1, 5>; // Floating point division.
|
|
defm : SKLWriteResPair<WriteFDivX, [SKLPort0,SKLFPDivider], 11, [1,3], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFDivY, [SKLPort0,SKLFPDivider], 11, [1,5], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFDivZ>;
|
|
defm : SKLWriteResPair<WriteFDiv64, [SKLPort0,SKLFPDivider], 14, [1,4], 1, 5>; // Floating point double division.
|
|
defm : SKLWriteResPair<WriteFDiv64X, [SKLPort0,SKLFPDivider], 14, [1,4], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFDiv64Y, [SKLPort0,SKLFPDivider], 14, [1,8], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFDiv64Z>;
|
|
|
|
defm : SKLWriteResPair<WriteFSqrt, [SKLPort0,SKLFPDivider], 12, [1,3], 1, 5>; // Floating point square root.
|
|
defm : SKLWriteResPair<WriteFSqrtX, [SKLPort0,SKLFPDivider], 12, [1,3], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFSqrtY, [SKLPort0,SKLFPDivider], 12, [1,6], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFSqrtZ>;
|
|
defm : SKLWriteResPair<WriteFSqrt64, [SKLPort0,SKLFPDivider], 18, [1,6], 1, 5>; // Floating point double square root.
|
|
defm : SKLWriteResPair<WriteFSqrt64X, [SKLPort0,SKLFPDivider], 18, [1,6], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFSqrt64Y, [SKLPort0,SKLFPDivider], 18, [1,12],1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFSqrt64Z>;
|
|
defm : SKLWriteResPair<WriteFSqrt80, [SKLPort0,SKLFPDivider], 21, [1,7]>; // Floating point long double square root.
|
|
|
|
defm : SKLWriteResPair<WriteFRcp, [SKLPort0], 4, [1], 1, 5>; // Floating point reciprocal estimate.
|
|
defm : SKLWriteResPair<WriteFRcpX, [SKLPort0], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFRcpY, [SKLPort0], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFRcpZ>;
|
|
|
|
defm : SKLWriteResPair<WriteFRsqrt, [SKLPort0], 4, [1], 1, 5>; // Floating point reciprocal square root estimate.
|
|
defm : SKLWriteResPair<WriteFRsqrtX,[SKLPort0], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFRsqrtY,[SKLPort0], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFRsqrtZ>;
|
|
|
|
defm : SKLWriteResPair<WriteFMA, [SKLPort01], 4, [1], 1, 5>; // Fused Multiply Add.
|
|
defm : SKLWriteResPair<WriteFMAX, [SKLPort01], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteFMAY, [SKLPort01], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFMAZ>;
|
|
defm : SKLWriteResPair<WriteDPPD, [SKLPort5,SKLPort01], 9, [1,2], 3, 6>; // Floating point double dot product.
|
|
defm : X86WriteRes<WriteDPPS, [SKLPort5,SKLPort01], 13, [1,3], 4>;
|
|
defm : X86WriteRes<WriteDPPSY, [SKLPort5,SKLPort01], 13, [1,3], 4>;
|
|
defm : X86WriteRes<WriteDPPSLd, [SKLPort5,SKLPort01,SKLPort06,SKLPort23], 19, [1,3,1,1], 6>;
|
|
defm : X86WriteRes<WriteDPPSYLd, [SKLPort5,SKLPort01,SKLPort06,SKLPort23], 20, [1,3,1,1], 6>;
|
|
defm : SKLWriteResPair<WriteFSign, [SKLPort0], 1>; // Floating point fabs/fchs.
|
|
defm : SKLWriteResPair<WriteFRnd, [SKLPort01], 8, [2], 2, 6>; // Floating point rounding.
|
|
defm : SKLWriteResPair<WriteFRndY, [SKLPort01], 8, [2], 2, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFRndZ>;
|
|
defm : SKLWriteResPair<WriteFLogic, [SKLPort015], 1, [1], 1, 6>; // Floating point and/or/xor logicals.
|
|
defm : SKLWriteResPair<WriteFLogicY, [SKLPort015], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFLogicZ>;
|
|
defm : SKLWriteResPair<WriteFTest, [SKLPort0], 2, [1], 1, 6>; // Floating point TEST instructions.
|
|
defm : SKLWriteResPair<WriteFTestY, [SKLPort0], 2, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFTestZ>;
|
|
defm : SKLWriteResPair<WriteFShuffle, [SKLPort5], 1, [1], 1, 6>; // Floating point vector shuffles.
|
|
defm : SKLWriteResPair<WriteFShuffleY, [SKLPort5], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFShuffleZ>;
|
|
defm : SKLWriteResPair<WriteFVarShuffle, [SKLPort5], 1, [1], 1, 6>; // Floating point vector shuffles.
|
|
defm : SKLWriteResPair<WriteFVarShuffleY, [SKLPort5], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFVarShuffleZ>;
|
|
defm : SKLWriteResPair<WriteFBlend, [SKLPort015], 1, [1], 1, 6>; // Floating point vector blends.
|
|
defm : SKLWriteResPair<WriteFBlendY, [SKLPort015], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFBlendZ>;
|
|
defm : SKLWriteResPair<WriteFVarBlend, [SKLPort015], 2, [2], 2, 6>; // Fp vector variable blends.
|
|
defm : SKLWriteResPair<WriteFVarBlendY,[SKLPort015], 2, [2], 2, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteFVarBlendZ>;
|
|
|
|
// FMA Scheduling helper class.
|
|
// class FMASC { X86FoldableSchedWrite Sched = WriteFAdd; }
|
|
|
|
// Vector integer operations.
|
|
defm : X86WriteRes<WriteVecLoad, [SKLPort23], 5, [1], 1>;
|
|
defm : X86WriteRes<WriteVecLoadX, [SKLPort23], 6, [1], 1>;
|
|
defm : X86WriteRes<WriteVecLoadY, [SKLPort23], 7, [1], 1>;
|
|
defm : X86WriteRes<WriteVecLoadNT, [SKLPort23,SKLPort015], 7, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecLoadNTY, [SKLPort23,SKLPort015], 8, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecMaskedLoad, [SKLPort23,SKLPort015], 7, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecMaskedLoadY, [SKLPort23,SKLPort015], 8, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecStore, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecStoreX, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecStoreY, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecStoreNT, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecStoreNTY, [SKLPort237,SKLPort4], 1, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecMaskedStore32, [SKLPort23,SKLPort0,SKLPort4], 2, [1,1,1], 3>;
|
|
defm : X86WriteRes<WriteVecMaskedStore32Y, [SKLPort23,SKLPort0,SKLPort4], 2, [1,1,1], 3>;
|
|
defm : X86WriteRes<WriteVecMaskedStore64, [SKLPort23,SKLPort0,SKLPort4], 2, [1,1,1], 3>;
|
|
defm : X86WriteRes<WriteVecMaskedStore64Y, [SKLPort23,SKLPort0,SKLPort4], 2, [1,1,1], 3>;
|
|
defm : X86WriteRes<WriteVecMove, [SKLPort05], 1, [1], 1>;
|
|
defm : X86WriteRes<WriteVecMoveX, [SKLPort015], 1, [1], 1>;
|
|
defm : X86WriteRes<WriteVecMoveY, [SKLPort015], 1, [1], 1>;
|
|
defm : X86WriteResUnsupported<WriteVecMoveZ>;
|
|
defm : X86WriteRes<WriteVecMoveToGpr, [SKLPort0], 2, [1], 1>;
|
|
defm : X86WriteRes<WriteVecMoveFromGpr, [SKLPort5], 1, [1], 1>;
|
|
|
|
defm : SKLWriteResPair<WriteVecALU, [SKLPort05], 1, [1], 1, 5>; // Vector integer ALU op, no logicals.
|
|
defm : SKLWriteResPair<WriteVecALUX, [SKLPort01], 1, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteVecALUY, [SKLPort01], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteVecALUZ>;
|
|
defm : SKLWriteResPair<WriteVecLogic, [SKLPort05], 1, [1], 1, 5>; // Vector integer and/or/xor.
|
|
defm : SKLWriteResPair<WriteVecLogicX,[SKLPort015], 1, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteVecLogicY,[SKLPort015], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteVecLogicZ>;
|
|
defm : SKLWriteResPair<WriteVecTest, [SKLPort0,SKLPort5], 3, [1,1], 2, 6>; // Vector integer TEST instructions.
|
|
defm : SKLWriteResPair<WriteVecTestY, [SKLPort0,SKLPort5], 3, [1,1], 2, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteVecTestZ>;
|
|
defm : SKLWriteResPair<WriteVecIMul, [SKLPort0] , 5, [1], 1, 5>; // Vector integer multiply.
|
|
defm : SKLWriteResPair<WriteVecIMulX, [SKLPort01], 5, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteVecIMulY, [SKLPort01], 5, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteVecIMulZ>;
|
|
defm : SKLWriteResPair<WritePMULLD, [SKLPort01], 10, [2], 2, 6>; // Vector PMULLD.
|
|
defm : SKLWriteResPair<WritePMULLDY, [SKLPort01], 10, [2], 2, 7>;
|
|
defm : X86WriteResPairUnsupported<WritePMULLDZ>;
|
|
defm : SKLWriteResPair<WriteShuffle, [SKLPort5], 1, [1], 1, 5>; // Vector shuffles.
|
|
defm : SKLWriteResPair<WriteShuffleX, [SKLPort5], 1, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteShuffleY, [SKLPort5], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteShuffleZ>;
|
|
defm : SKLWriteResPair<WriteVarShuffle, [SKLPort0,SKLPort5], 1, [1,1], 2, 5>; // Vector shuffles.
|
|
defm : SKLWriteResPair<WriteVarShuffleX, [SKLPort5], 1, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteVarShuffleY, [SKLPort5], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteVarShuffleZ>;
|
|
defm : SKLWriteResPair<WriteBlend, [SKLPort5], 1, [1], 1, 6>; // Vector blends.
|
|
defm : SKLWriteResPair<WriteBlendY, [SKLPort5], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteBlendZ>;
|
|
defm : SKLWriteResPair<WriteVarBlend, [SKLPort015], 2, [2], 2, 6>; // Vector variable blends.
|
|
defm : SKLWriteResPair<WriteVarBlendY, [SKLPort015], 2, [2], 2, 6>;
|
|
defm : X86WriteResPairUnsupported<WriteVarBlendZ>;
|
|
defm : SKLWriteResPair<WriteMPSAD, [SKLPort5], 4, [2], 2, 6>; // Vector MPSAD.
|
|
defm : SKLWriteResPair<WriteMPSADY, [SKLPort5], 4, [2], 2, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteMPSADZ>;
|
|
defm : SKLWriteResPair<WritePSADBW, [SKLPort5], 3, [1], 1, 5>; // Vector PSADBW.
|
|
defm : SKLWriteResPair<WritePSADBWX, [SKLPort5], 3, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WritePSADBWY, [SKLPort5], 3, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WritePSADBWZ>;
|
|
defm : SKLWriteResPair<WritePHMINPOS, [SKLPort0], 4, [1], 1, 6>; // Vector PHMINPOS.
|
|
|
|
// Vector integer shifts.
|
|
defm : SKLWriteResPair<WriteVecShift, [SKLPort0], 1, [1], 1, 5>;
|
|
defm : X86WriteRes<WriteVecShiftX, [SKLPort5,SKLPort01], 2, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecShiftY, [SKLPort5,SKLPort01], 4, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecShiftXLd, [SKLPort01,SKLPort23], 7, [1,1], 2>;
|
|
defm : X86WriteRes<WriteVecShiftYLd, [SKLPort01,SKLPort23], 8, [1,1], 2>;
|
|
defm : X86WriteResPairUnsupported<WriteVecShiftZ>;
|
|
|
|
defm : SKLWriteResPair<WriteVecShiftImm, [SKLPort0], 1, [1], 1, 5>; // Vector integer immediate shifts.
|
|
defm : SKLWriteResPair<WriteVecShiftImmX, [SKLPort01], 1, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteVecShiftImmY, [SKLPort01], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteVecShiftImmZ>;
|
|
defm : SKLWriteResPair<WriteVarVecShift, [SKLPort01], 1, [1], 1, 6>; // Variable vector shifts.
|
|
defm : SKLWriteResPair<WriteVarVecShiftY, [SKLPort01], 1, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteVarVecShiftZ>;
|
|
|
|
// Vector insert/extract operations.
|
|
def : WriteRes<WriteVecInsert, [SKLPort5]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [2];
|
|
}
|
|
def : WriteRes<WriteVecInsertLd, [SKLPort5,SKLPort23]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 2;
|
|
}
|
|
def: InstRW<[WriteVecInsertLd], (instregex "(V?)MOV(H|L)(PD|PS)rm")>;
|
|
|
|
def : WriteRes<WriteVecExtract, [SKLPort0,SKLPort5]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 2;
|
|
}
|
|
def : WriteRes<WriteVecExtractSt, [SKLPort4,SKLPort5,SKLPort237]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 3;
|
|
}
|
|
|
|
// Conversion between integer and float.
|
|
defm : SKLWriteResPair<WriteCvtSS2I, [SKLPort0,SKLPort01], 6, [1,1], 2, 5>;
|
|
defm : SKLWriteResPair<WriteCvtPS2I, [SKLPort01], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteCvtPS2IY, [SKLPort01], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteCvtPS2IZ>;
|
|
defm : SKLWriteResPair<WriteCvtSD2I, [SKLPort0,SKLPort01], 6, [1,1], 2, 5>;
|
|
defm : SKLWriteResPair<WriteCvtPD2I, [SKLPort5,SKLPort01], 5, [1,1], 2, 6>;
|
|
defm : SKLWriteResPair<WriteCvtPD2IY, [SKLPort5,SKLPort01], 7, [1,1], 2, 6>;
|
|
defm : X86WriteResPairUnsupported<WriteCvtPD2IZ>;
|
|
|
|
defm : X86WriteRes<WriteCvtI2SS, [SKLPort5,SKLPort01], 5, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtI2SSLd, [SKLPort23,SKLPort01], 10, [1,1], 2>;
|
|
defm : SKLWriteResPair<WriteCvtI2PS, [SKLPort01], 4, [1], 1, 6>;
|
|
defm : SKLWriteResPair<WriteCvtI2PSY, [SKLPort01], 4, [1], 1, 7>;
|
|
defm : X86WriteResPairUnsupported<WriteCvtI2PSZ>;
|
|
defm : X86WriteRes<WriteCvtI2SD, [SKLPort5,SKLPort01], 5, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtI2SDLd, [SKLPort23,SKLPort01], 10, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtI2PD, [SKLPort5,SKLPort01], 5, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtI2PDLd, [SKLPort23,SKLPort01], 10, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtI2PDY, [SKLPort5,SKLPort01], 7, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtI2PDYLd, [SKLPort23,SKLPort01], 11, [1,1], 2>;
|
|
defm : X86WriteResPairUnsupported<WriteCvtI2PDZ>;
|
|
|
|
defm : X86WriteRes<WriteCvtSS2SD, [SKLPort5,SKLPort01], 5, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtSS2SDLd, [SKLPort23,SKLPort01], 10, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtPS2PD, [SKLPort5,SKLPort01], 5, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtPS2PDLd, [SKLPort23,SKLPort01], 9, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtPS2PDY, [SKLPort5,SKLPort01], 7, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtPS2PDYLd, [SKLPort23,SKLPort01], 11, [1,1], 2>;
|
|
defm : X86WriteResPairUnsupported<WriteCvtPS2PDZ>;
|
|
defm : SKLWriteResPair<WriteCvtSD2SS, [SKLPort5,SKLPort01], 5, [1,1], 2, 5>;
|
|
defm : SKLWriteResPair<WriteCvtPD2PS, [SKLPort5,SKLPort01], 5, [1,1], 2, 6>;
|
|
defm : SKLWriteResPair<WriteCvtPD2PSY, [SKLPort5,SKLPort01], 7, [1,1], 2, 6>;
|
|
defm : X86WriteResPairUnsupported<WriteCvtPD2PSZ>;
|
|
|
|
defm : X86WriteRes<WriteCvtPH2PS, [SKLPort5,SKLPort01], 5, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtPH2PSY, [SKLPort5,SKLPort01], 7, [1,1], 2>;
|
|
defm : X86WriteResUnsupported<WriteCvtPH2PSZ>;
|
|
defm : X86WriteRes<WriteCvtPH2PSLd, [SKLPort23,SKLPort01], 9, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtPH2PSYLd, [SKLPort23,SKLPort01], 10, [1,1], 2>;
|
|
defm : X86WriteResUnsupported<WriteCvtPH2PSZLd>;
|
|
|
|
defm : X86WriteRes<WriteCvtPS2PH, [SKLPort5,SKLPort01], 5, [1,1], 2>;
|
|
defm : X86WriteRes<WriteCvtPS2PHY, [SKLPort5,SKLPort01], 7, [1,1], 2>;
|
|
defm : X86WriteResUnsupported<WriteCvtPS2PHZ>;
|
|
defm : X86WriteRes<WriteCvtPS2PHSt, [SKLPort4,SKLPort5,SKLPort237,SKLPort01], 6, [1,1,1,1], 4>;
|
|
defm : X86WriteRes<WriteCvtPS2PHYSt, [SKLPort4,SKLPort5,SKLPort237,SKLPort01], 8, [1,1,1,1], 4>;
|
|
defm : X86WriteResUnsupported<WriteCvtPS2PHZSt>;
|
|
|
|
// Strings instructions.
|
|
|
|
// Packed Compare Implicit Length Strings, Return Mask
|
|
def : WriteRes<WritePCmpIStrM, [SKLPort0]> {
|
|
let Latency = 10;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [3];
|
|
}
|
|
def : WriteRes<WritePCmpIStrMLd, [SKLPort0, SKLPort23]> {
|
|
let Latency = 16;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [3,1];
|
|
}
|
|
|
|
// Packed Compare Explicit Length Strings, Return Mask
|
|
def : WriteRes<WritePCmpEStrM, [SKLPort0, SKLPort5, SKLPort015, SKLPort0156]> {
|
|
let Latency = 19;
|
|
let NumMicroOps = 9;
|
|
let ReleaseAtCycles = [4,3,1,1];
|
|
}
|
|
def : WriteRes<WritePCmpEStrMLd, [SKLPort0, SKLPort5,SKLPort23, SKLPort015, SKLPort0156]> {
|
|
let Latency = 25;
|
|
let NumMicroOps = 10;
|
|
let ReleaseAtCycles = [4,3,1,1,1];
|
|
}
|
|
|
|
// Packed Compare Implicit Length Strings, Return Index
|
|
def : WriteRes<WritePCmpIStrI, [SKLPort0]> {
|
|
let Latency = 10;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [3];
|
|
}
|
|
def : WriteRes<WritePCmpIStrILd, [SKLPort0, SKLPort23]> {
|
|
let Latency = 16;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [3,1];
|
|
}
|
|
|
|
// Packed Compare Explicit Length Strings, Return Index
|
|
def : WriteRes<WritePCmpEStrI, [SKLPort0, SKLPort5, SKLPort0156]> {
|
|
let Latency = 18;
|
|
let NumMicroOps = 8;
|
|
let ReleaseAtCycles = [4,3,1];
|
|
}
|
|
def : WriteRes<WritePCmpEStrILd, [SKLPort0, SKLPort5, SKLPort23, SKLPort0156]> {
|
|
let Latency = 24;
|
|
let NumMicroOps = 9;
|
|
let ReleaseAtCycles = [4,3,1,1];
|
|
}
|
|
|
|
// MOVMSK Instructions.
|
|
def : WriteRes<WriteFMOVMSK, [SKLPort0]> { let Latency = 2; }
|
|
def : WriteRes<WriteVecMOVMSK, [SKLPort0]> { let Latency = 2; }
|
|
def : WriteRes<WriteVecMOVMSKY, [SKLPort0]> { let Latency = 2; }
|
|
def : WriteRes<WriteMMXMOVMSK, [SKLPort0]> { let Latency = 2; }
|
|
|
|
// AES instructions.
|
|
def : WriteRes<WriteAESDecEnc, [SKLPort0]> { // Decryption, encryption.
|
|
let Latency = 4;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def : WriteRes<WriteAESDecEncLd, [SKLPort0, SKLPort23]> {
|
|
let Latency = 10;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
|
|
def : WriteRes<WriteAESIMC, [SKLPort0]> { // InvMixColumn.
|
|
let Latency = 8;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [2];
|
|
}
|
|
def : WriteRes<WriteAESIMCLd, [SKLPort0, SKLPort23]> {
|
|
let Latency = 14;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [2,1];
|
|
}
|
|
|
|
def : WriteRes<WriteAESKeyGen, [SKLPort0, SKLPort5, SKLPort015]> { // Key Generation.
|
|
let Latency = 20;
|
|
let NumMicroOps = 11;
|
|
let ReleaseAtCycles = [3,6,2];
|
|
}
|
|
def : WriteRes<WriteAESKeyGenLd, [SKLPort0, SKLPort5, SKLPort23, SKLPort015]> {
|
|
let Latency = 25;
|
|
let NumMicroOps = 11;
|
|
let ReleaseAtCycles = [3,6,1,1];
|
|
}
|
|
|
|
// Carry-less multiplication instructions.
|
|
def : WriteRes<WriteCLMul, [SKLPort5]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def : WriteRes<WriteCLMulLd, [SKLPort5, SKLPort23]> {
|
|
let Latency = 12;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
|
|
// Catch-all for expensive system instructions.
|
|
def : WriteRes<WriteSystem, [SKLPort0156]> { let Latency = 100; } // def WriteSystem : SchedWrite;
|
|
|
|
// AVX2.
|
|
defm : SKLWriteResPair<WriteFShuffle256, [SKLPort5], 3, [1], 1, 7>; // Fp 256-bit width vector shuffles.
|
|
defm : SKLWriteResPair<WriteFVarShuffle256, [SKLPort5], 3, [1], 1, 7>; // Fp 256-bit width vector variable shuffles.
|
|
defm : SKLWriteResPair<WriteShuffle256, [SKLPort5], 3, [1], 1, 7>; // 256-bit width vector shuffles.
|
|
defm : SKLWriteResPair<WriteVPMOV256, [SKLPort5], 3, [1], 1, 7>; // 256-bit width packed vector width-changing move.
|
|
defm : SKLWriteResPair<WriteVarShuffle256, [SKLPort5], 3, [1], 1, 7>; // 256-bit width vector variable shuffles.
|
|
|
|
// Old microcoded instructions that nobody use.
|
|
def : WriteRes<WriteMicrocoded, [SKLPort0156]> { let Latency = 100; } // def WriteMicrocoded : SchedWrite;
|
|
|
|
// Fence instructions.
|
|
def : WriteRes<WriteFence, [SKLPort23, SKLPort4]> { let NumMicroOps = 2; let ReleaseAtCycles = [1,1]; }
|
|
|
|
// Load/store MXCSR.
|
|
def : WriteRes<WriteLDMXCSR, [SKLPort0,SKLPort23,SKLPort0156]> { let Latency = 7; let NumMicroOps = 3; let ReleaseAtCycles = [1,1,1]; }
|
|
def : WriteRes<WriteSTMXCSR, [SKLPort4,SKLPort5,SKLPort237]> { let Latency = 2; let NumMicroOps = 3; let ReleaseAtCycles = [1,1,1]; }
|
|
|
|
// Nop, not very useful expect it provides a model for nops!
|
|
def : WriteRes<WriteNop, []>;
|
|
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
// Horizontal add/sub instructions.
|
|
////////////////////////////////////////////////////////////////////////////////
|
|
|
|
defm : SKLWriteResPair<WriteFHAdd, [SKLPort5,SKLPort01], 6, [2,1], 3, 6>;
|
|
defm : SKLWriteResPair<WriteFHAddY, [SKLPort5,SKLPort01], 6, [2,1], 3, 7>;
|
|
defm : SKLWriteResPair<WritePHAdd, [SKLPort5,SKLPort05], 3, [2,1], 3, 5>;
|
|
defm : SKLWriteResPair<WritePHAddX, [SKLPort5,SKLPort015], 3, [2,1], 3, 6>;
|
|
defm : SKLWriteResPair<WritePHAddY, [SKLPort5,SKLPort015], 3, [2,1], 3, 7>;
|
|
|
|
// Remaining instrs.
|
|
|
|
def SKLWriteResGroup1 : SchedWriteRes<[SKLPort0]> {
|
|
let Latency = 1;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup1], (instregex "MMX_PADDS(B|W)rr",
|
|
"MMX_PADDUS(B|W)rr",
|
|
"MMX_PAVG(B|W)rr",
|
|
"MMX_PCMPEQ(B|D|W)rr",
|
|
"MMX_PCMPGT(B|D|W)rr",
|
|
"MMX_P(MAX|MIN)SWrr",
|
|
"MMX_P(MAX|MIN)UBrr",
|
|
"MMX_PSUBS(B|W)rr",
|
|
"MMX_PSUBUS(B|W)rr")>;
|
|
|
|
def SKLWriteResGroup3 : SchedWriteRes<[SKLPort5]> {
|
|
let Latency = 1;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup3], (instregex "COM(P?)_FST0r",
|
|
"UCOM_F(P?)r")>;
|
|
|
|
def SKLWriteResGroup4 : SchedWriteRes<[SKLPort6]> {
|
|
let Latency = 1;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup4], (instregex "JMP(16|32|64)r")>;
|
|
|
|
def SKLWriteResGroup6 : SchedWriteRes<[SKLPort05]> {
|
|
let Latency = 1;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup6], (instrs FINCSTP, FNOP)>;
|
|
|
|
def SKLWriteResGroup7 : SchedWriteRes<[SKLPort06]> {
|
|
let Latency = 1;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup7], (instrs CDQ, CQO, CLAC, STAC)>;
|
|
|
|
def SKLWriteResGroup8 : SchedWriteRes<[SKLPort15]> {
|
|
let Latency = 1;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup8], (instregex "ANDN(32|64)rr")>;
|
|
|
|
def SKLWriteResGroup9 : SchedWriteRes<[SKLPort015]> {
|
|
let Latency = 1;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup9], (instregex "(V?)PADD(B|D|Q|W)(Y?)rr",
|
|
"VPBLENDD(Y?)rri")>;
|
|
|
|
def SKLWriteResGroup10 : SchedWriteRes<[SKLPort0156]> {
|
|
let Latency = 1;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup10], (instrs SGDT64m,
|
|
SIDT64m,
|
|
SMSW16m,
|
|
STRm,
|
|
SYSCALL)>;
|
|
|
|
def SKLWriteResGroup11 : SchedWriteRes<[SKLPort4,SKLPort237]> {
|
|
let Latency = 1;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup11], (instrs FBSTPm, VMPTRSTm)>;
|
|
def: InstRW<[SKLWriteResGroup11], (instregex "ST_FP(32|64|80)m")>;
|
|
|
|
def SKLWriteResGroup13 : SchedWriteRes<[SKLPort5]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup13], (instrs MMX_MOVQ2DQrr)>;
|
|
|
|
def SKLWriteResGroup14 : SchedWriteRes<[SKLPort05]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup14], (instrs FDECSTP,
|
|
MMX_MOVDQ2Qrr)>;
|
|
|
|
def SKLWriteResGroup17 : SchedWriteRes<[SKLPort0156]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup17], (instrs LFENCE,
|
|
WAIT,
|
|
XGETBV)>;
|
|
|
|
def SKLWriteResGroup20 : SchedWriteRes<[SKLPort6,SKLPort0156]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup20], (instregex "CLFLUSH")>;
|
|
|
|
def SKLWriteResGroup23 : SchedWriteRes<[SKLPort06,SKLPort0156]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup23], (instrs CWD,
|
|
JCXZ, JECXZ, JRCXZ,
|
|
ADC8i8, SBB8i8,
|
|
ADC16i16, SBB16i16,
|
|
ADC32i32, SBB32i32,
|
|
ADC64i32, SBB64i32)>;
|
|
|
|
def SKLWriteResGroup25 : SchedWriteRes<[SKLPort4,SKLPort6,SKLPort237]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup25], (instrs FNSTCW16m)>;
|
|
|
|
def SKLWriteResGroup27 : SchedWriteRes<[SKLPort4,SKLPort237,SKLPort15]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup27], (instregex "MOVBE(16|32|64)mr")>;
|
|
|
|
def SKLWriteResGroup28 : SchedWriteRes<[SKLPort4,SKLPort237,SKLPort0156]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup28], (instrs PUSH16r, PUSH32r, PUSH64r, PUSH64i8,
|
|
STOSB, STOSL, STOSQ, STOSW)>;
|
|
def: InstRW<[SKLWriteResGroup28], (instregex "PUSH(16|32|64)rmr")>;
|
|
|
|
def SKLWriteResGroup29 : SchedWriteRes<[SKLPort1]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup29], (instregex "PDEP(32|64)rr",
|
|
"PEXT(32|64)rr")>;
|
|
|
|
def SKLWriteResGroup30 : SchedWriteRes<[SKLPort5]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup30], (instregex "(ADD|SUB|SUBR)_(FPrST0|FST0r|FrST0)",
|
|
"VPBROADCAST(B|W)rr")>;
|
|
|
|
def SKLWriteResGroup32 : SchedWriteRes<[SKLPort0,SKLPort0156]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup32], (instrs FNSTSW16r)>;
|
|
|
|
def SKLWriteResGroup35 : SchedWriteRes<[SKLPort0,SKLPort5]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup35], (instregex "MMX_PH(ADD|SUB)SWrr")>;
|
|
|
|
def SKLWriteResGroup36 : SchedWriteRes<[SKLPort5,SKLPort01]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [2,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup36], (instregex "(V?)PHADDSW(Y?)rr",
|
|
"(V?)PHSUBSW(Y?)rr")>;
|
|
|
|
def SKLWriteResGroup39 : SchedWriteRes<[SKLPort5]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup39], (instrs MMX_PACKSSDWrr,
|
|
MMX_PACKSSWBrr,
|
|
MMX_PACKUSWBrr)>;
|
|
|
|
def SKLWriteResGroup40 : SchedWriteRes<[SKLPort6,SKLPort0156]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup40], (instregex "CLD")>;
|
|
|
|
def SKLWriteResGroup42 : SchedWriteRes<[SKLPort06,SKLPort0156]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup42], (instrs RCL8r1, RCL16r1, RCL32r1, RCL64r1,
|
|
RCR8r1, RCR16r1, RCR32r1, RCR64r1)>;
|
|
|
|
def SKLWriteResGroup42b : SchedWriteRes<[SKLPort1,SKLPort06,SKLPort0156]> {
|
|
let Latency = 5;
|
|
let NumMicroOps = 8;
|
|
let ReleaseAtCycles = [2,4,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup42b], (instrs RCR8ri, RCR16ri, RCR32ri, RCR64ri)>;
|
|
|
|
def SKLWriteResGroup42c : SchedWriteRes<[SKLPort1,SKLPort06,SKLPort0156]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 8;
|
|
let ReleaseAtCycles = [2,4,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup42c], (instrs RCL8ri, RCL16ri, RCL32ri, RCL64ri)>;
|
|
|
|
def SKLWriteResGroup43 : SchedWriteRes<[SKLPort0,SKLPort4,SKLPort237]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup43], (instrs FNSTSWm)>;
|
|
|
|
def SKLWriteResGroup45 : SchedWriteRes<[SKLPort4,SKLPort6,SKLPort237,SKLPort0156]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [1,1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup45], (instregex "CALL(16|32|64)r")>;
|
|
|
|
def SKLWriteResGroup46 : SchedWriteRes<[SKLPort4,SKLPort237,SKLPort06,SKLPort0156]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [1,1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup46], (instrs CALL64pcrel32)>;
|
|
|
|
def SKLWriteResGroup47 : SchedWriteRes<[SKLPort0]> {
|
|
let Latency = 4;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup47], (instregex "MUL_(FPrST0|FST0r|FrST0)")>;
|
|
|
|
def SKLWriteResGroup53 : SchedWriteRes<[SKLPort4,SKLPort5,SKLPort237]> {
|
|
let Latency = 4;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup53], (instregex "IST(T?)_FP(16|32|64)m",
|
|
"IST_F(16|32)m")>;
|
|
|
|
def SKLWriteResGroup54 : SchedWriteRes<[SKLPort0156]> {
|
|
let Latency = 4;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [4];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup54], (instrs FNCLEX)>;
|
|
|
|
def SKLWriteResGroup55 : SchedWriteRes<[SKLPort6,SKLPort0156]> {
|
|
let Latency = 4;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [1,3];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup55], (instrs PAUSE)>;
|
|
|
|
def SKLWriteResGroup56 : SchedWriteRes<[]> {
|
|
let Latency = 0;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup56], (instrs VZEROUPPER)>;
|
|
|
|
def SKLWriteResGroup57 : SchedWriteRes<[SKLPort1,SKLPort6,SKLPort0156]> {
|
|
let Latency = 4;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [1,1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup57], (instregex "LAR(16|32|64)rr")>;
|
|
|
|
def SKLWriteResGroup60 : SchedWriteRes<[SKLPort5,SKLPort0]> {
|
|
let Latency = 5;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup60], (instregex "MMX_CVT(T?)PS2PIrr")>;
|
|
|
|
def SKLWriteResGroup61 : SchedWriteRes<[SKLPort1,SKLPort6,SKLPort06]> {
|
|
let Latency = 5;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup61], (instregex "STR(16|32|64)r")>;
|
|
|
|
def SKLWriteResGroup63 : SchedWriteRes<[SKLPort06,SKLPort0156]> {
|
|
let Latency = 5;
|
|
let NumMicroOps = 5;
|
|
let ReleaseAtCycles = [1,4];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup63], (instrs XSETBV)>;
|
|
|
|
def SKLWriteResGroup65 : SchedWriteRes<[SKLPort4,SKLPort237,SKLPort0156]> {
|
|
let Latency = 5;
|
|
let NumMicroOps = 6;
|
|
let ReleaseAtCycles = [1,1,4];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup65], (instregex "PUSHF(16|64)")>;
|
|
|
|
def SKLWriteResGroup67 : SchedWriteRes<[SKLPort23]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup67], (instrs VBROADCASTSSrm,
|
|
VPBROADCASTDrm,
|
|
VPBROADCASTQrm)>;
|
|
def: InstRW<[SKLWriteResGroup67], (instregex "(V?)MOVSHDUPrm",
|
|
"(V?)MOVSLDUPrm",
|
|
"(V?)MOVDDUPrm")>;
|
|
|
|
def SKLWriteResGroup68 : SchedWriteRes<[SKLPort01]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup68], (instrs MMX_CVTPI2PSrr)>;
|
|
|
|
def SKLWriteResGroup69 : SchedWriteRes<[SKLPort0,SKLPort23]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup69], (instrs MMX_PADDSBrm,
|
|
MMX_PADDSWrm,
|
|
MMX_PADDUSBrm,
|
|
MMX_PADDUSWrm,
|
|
MMX_PAVGBrm,
|
|
MMX_PAVGWrm,
|
|
MMX_PCMPEQBrm,
|
|
MMX_PCMPEQDrm,
|
|
MMX_PCMPEQWrm,
|
|
MMX_PCMPGTBrm,
|
|
MMX_PCMPGTDrm,
|
|
MMX_PCMPGTWrm,
|
|
MMX_PMAXSWrm,
|
|
MMX_PMAXUBrm,
|
|
MMX_PMINSWrm,
|
|
MMX_PMINUBrm,
|
|
MMX_PSUBSBrm,
|
|
MMX_PSUBSWrm,
|
|
MMX_PSUBUSBrm,
|
|
MMX_PSUBUSWrm)>;
|
|
|
|
def SKLWriteResGroup72 : SchedWriteRes<[SKLPort6,SKLPort23]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup72], (instrs FARJMP64m)>;
|
|
def: InstRW<[SKLWriteResGroup72], (instregex "JMP(16|32|64)m")>;
|
|
|
|
def SKLWriteResGroup75 : SchedWriteRes<[SKLPort23,SKLPort15]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup75], (instregex "ANDN(32|64)rm",
|
|
"MOVBE(16|32|64)rm")>;
|
|
|
|
def SKLWriteResGroup76 : SchedWriteRes<[SKLPort23,SKLPort0156]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup76], (instrs POP16r, POP32r, POP64r)>;
|
|
def: InstRW<[SKLWriteResGroup76], (instregex "POP(16|32|64)rmr")>;
|
|
|
|
def SKLWriteResGroup78 : SchedWriteRes<[SKLPort5,SKLPort01]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [2,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup78], (instregex "(V?)CVTSI642SSrr")>;
|
|
|
|
def SKLWriteResGroup80 : SchedWriteRes<[SKLPort1,SKLPort6,SKLPort06,SKLPort0156]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [1,1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup80], (instregex "SLDT(16|32|64)r")>;
|
|
|
|
def SKLWriteResGroup82 : SchedWriteRes<[SKLPort4,SKLPort23,SKLPort237,SKLPort06]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [1,1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup82], (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 SKLWriteResGroup83 : SchedWriteRes<[SKLPort4,SKLPort23,SKLPort237,SKLPort0156]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [1,1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup83], (instregex "POP(16|32|64)rmm",
|
|
"PUSH(16|32|64)rmm")>;
|
|
|
|
def SKLWriteResGroup84 : SchedWriteRes<[SKLPort6,SKLPort0156]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 6;
|
|
let ReleaseAtCycles = [1,5];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup84], (instrs STD)>;
|
|
|
|
def SKLWriteResGroup85 : SchedWriteRes<[SKLPort23]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup85], (instregex "LD_F(32|64|80)m")>;
|
|
def: InstRW<[SKLWriteResGroup85], (instrs VBROADCASTF128rm,
|
|
VBROADCASTI128rm,
|
|
VBROADCASTSDYrm,
|
|
VBROADCASTSSYrm,
|
|
VMOVDDUPYrm,
|
|
VMOVSHDUPYrm,
|
|
VMOVSLDUPYrm,
|
|
VPBROADCASTDYrm,
|
|
VPBROADCASTQYrm)>;
|
|
|
|
def SKLWriteResGroup88 : SchedWriteRes<[SKLPort5,SKLPort23]> {
|
|
let Latency = 6;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup88], (instregex "(V?)PMOV(SX|ZX)BDrm",
|
|
"(V?)PMOV(SX|ZX)BQrm",
|
|
"(V?)PMOV(SX|ZX)BWrm",
|
|
"(V?)PMOV(SX|ZX)DQrm",
|
|
"(V?)PMOV(SX|ZX)WDrm",
|
|
"(V?)PMOV(SX|ZX)WQrm")>;
|
|
|
|
def SKLWriteResGroup91 : SchedWriteRes<[SKLPort23,SKLPort015]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup91], (instrs VINSERTF128rmi,
|
|
VINSERTI128rmi,
|
|
VPBLENDDrmi)>;
|
|
def: InstRW<[SKLWriteResGroup91, ReadAfterVecXLd],
|
|
(instregex "(V?)PADD(B|D|Q|W)rm",
|
|
"(V?)PSUB(B|D|Q|W)rm")>;
|
|
|
|
def SKLWriteResGroup92 : SchedWriteRes<[SKLPort5,SKLPort23]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [2,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup92], (instrs MMX_PACKSSDWrm,
|
|
MMX_PACKSSWBrm,
|
|
MMX_PACKUSWBrm)>;
|
|
|
|
def SKLWriteResGroup94 : SchedWriteRes<[SKLPort23,SKLPort0156]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup94], (instrs LEAVE, LEAVE64,
|
|
SCASB, SCASL, SCASQ, SCASW)>;
|
|
|
|
def SKLWriteResGroup95 : SchedWriteRes<[SKLPort0,SKLPort5,SKLPort01]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup95], (instregex "(V?)CVT(T?)SS2SI64rr")>;
|
|
|
|
def SKLWriteResGroup96 : SchedWriteRes<[SKLPort0,SKLPort23,SKLPort05]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup96], (instrs FLDCW16m)>;
|
|
|
|
def SKLWriteResGroup98 : SchedWriteRes<[SKLPort6,SKLPort23,SKLPort0156]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup98], (instrs LRET64, RET64)>;
|
|
|
|
def SKLWriteResGroup100 : SchedWriteRes<[SKLPort4,SKLPort23,SKLPort237,SKLPort06]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 5;
|
|
let ReleaseAtCycles = [1,1,1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup100], (instregex "ROL(8|16|32|64)m(1|i)",
|
|
"ROR(8|16|32|64)m(1|i)")>;
|
|
|
|
def SKLWriteResGroup100_1 : SchedWriteRes<[SKLPort06]> {
|
|
let Latency = 2;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup100_1], (instrs ROL8r1, ROL16r1, ROL32r1, ROL64r1,
|
|
ROR8r1, ROR16r1, ROR32r1, ROR64r1)>;
|
|
|
|
def SKLWriteResGroup101 : SchedWriteRes<[SKLPort4,SKLPort23,SKLPort237,SKLPort0156]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 5;
|
|
let ReleaseAtCycles = [1,1,1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup101], (instregex "XADD(8|16|32|64)rm")>;
|
|
|
|
def SKLWriteResGroup102 : SchedWriteRes<[SKLPort4,SKLPort6,SKLPort23,SKLPort237,SKLPort0156]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 5;
|
|
let ReleaseAtCycles = [1,1,1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup102], (instregex "CALL(16|32|64)m")>;
|
|
def: InstRW<[SKLWriteResGroup102], (instrs FARCALL64m)>;
|
|
|
|
def SKLWriteResGroup103 : SchedWriteRes<[SKLPort6,SKLPort06,SKLPort15,SKLPort0156]> {
|
|
let Latency = 7;
|
|
let NumMicroOps = 7;
|
|
let ReleaseAtCycles = [1,3,1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup103], (instrs LOOP)>;
|
|
|
|
def SKLWriteResGroup107 : SchedWriteRes<[SKLPort1,SKLPort23]> {
|
|
let Latency = 8;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup107], (instregex "PDEP(32|64)rm",
|
|
"PEXT(32|64)rm")>;
|
|
|
|
def SKLWriteResGroup108 : SchedWriteRes<[SKLPort5,SKLPort23]> {
|
|
let Latency = 8;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup108], (instregex "FCOM(P?)(32|64)m")>;
|
|
def: InstRW<[SKLWriteResGroup108], (instrs VPBROADCASTBYrm,
|
|
VPBROADCASTWYrm,
|
|
VPMOVSXBDYrm,
|
|
VPMOVSXBQYrm,
|
|
VPMOVSXWQYrm)>;
|
|
|
|
def SKLWriteResGroup110 : SchedWriteRes<[SKLPort23,SKLPort015]> {
|
|
let Latency = 8;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup110], (instrs VPBLENDDYrmi)>;
|
|
def: InstRW<[SKLWriteResGroup110, ReadAfterVecYLd],
|
|
(instregex "VPADD(B|D|Q|W)Yrm",
|
|
"VPSUB(B|D|Q|W)Yrm")>;
|
|
|
|
def SKLWriteResGroup112 : SchedWriteRes<[SKLPort0,SKLPort5,SKLPort23]> {
|
|
let Latency = 8;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [1,2,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup112], (instregex "MMX_PH(ADD|SUB)SWrm")>;
|
|
|
|
def SKLWriteResGroup116 : SchedWriteRes<[SKLPort23,SKLPort237,SKLPort06,SKLPort0156]> {
|
|
let Latency = 8;
|
|
let NumMicroOps = 5;
|
|
let ReleaseAtCycles = [1,1,1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup116], (instregex "RCL(8|16|32|64)m(1|i)",
|
|
"RCR(8|16|32|64)m(1|i)")>;
|
|
|
|
def SKLWriteResGroup117 : SchedWriteRes<[SKLPort4,SKLPort23,SKLPort237,SKLPort06]> {
|
|
let Latency = 8;
|
|
let NumMicroOps = 6;
|
|
let ReleaseAtCycles = [1,1,1,3];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup117], (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 SKLWriteResGroup119 : SchedWriteRes<[SKLPort4,SKLPort23,SKLPort237,SKLPort06,SKLPort0156]> {
|
|
let Latency = 8;
|
|
let NumMicroOps = 6;
|
|
let ReleaseAtCycles = [1,1,1,2,1];
|
|
}
|
|
def: SchedAlias<WriteADCRMW, SKLWriteResGroup119>;
|
|
|
|
def SKLWriteResGroup121 : SchedWriteRes<[SKLPort5,SKLPort23]> {
|
|
let Latency = 9;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup121], (instrs PCMPGTQrm,
|
|
VPCMPGTQrm,
|
|
VPMOVSXBWYrm,
|
|
VPMOVSXDQYrm,
|
|
VPMOVSXWDYrm,
|
|
VPMOVZXWDYrm)>;
|
|
|
|
def SKLWriteResGroup123 : SchedWriteRes<[SKLPort23,SKLPort0]> {
|
|
let Latency = 9;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup123], (instregex "MMX_CVT(T?)PS2PIrm")>;
|
|
|
|
def SKLWriteResGroup128 : SchedWriteRes<[SKLPort5,SKLPort01,SKLPort23]> {
|
|
let Latency = 9;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [2,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup128], (instregex "(V?)PHADDSWrm",
|
|
"(V?)PHSUBSWrm")>;
|
|
|
|
def SKLWriteResGroup131 : SchedWriteRes<[SKLPort1,SKLPort6,SKLPort23,SKLPort0156]> {
|
|
let Latency = 9;
|
|
let NumMicroOps = 5;
|
|
let ReleaseAtCycles = [1,2,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup131], (instregex "LAR(16|32|64)rm",
|
|
"LSL(16|32|64)rm")>;
|
|
|
|
def SKLWriteResGroup133 : SchedWriteRes<[SKLPort5,SKLPort23]> {
|
|
let Latency = 10;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup133], (instregex "(ADD|SUB|SUBR)_F(32|64)m",
|
|
"ILD_F(16|32|64)m")>;
|
|
def: InstRW<[SKLWriteResGroup133], (instrs VPCMPGTQYrm)>;
|
|
|
|
def SKLWriteResGroup140 : SchedWriteRes<[SKLPort5,SKLPort01,SKLPort23]> {
|
|
let Latency = 10;
|
|
let NumMicroOps = 4;
|
|
let ReleaseAtCycles = [2,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup140], (instrs VPHADDSWYrm,
|
|
VPHSUBSWYrm)>;
|
|
|
|
def SKLWriteResGroup143 : SchedWriteRes<[SKLPort4,SKLPort6,SKLPort23,SKLPort237,SKLPort06,SKLPort0156]> {
|
|
let Latency = 10;
|
|
let NumMicroOps = 8;
|
|
let ReleaseAtCycles = [1,1,1,1,1,3];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup143], (instregex "XCHG(8|16|32|64)rm")>;
|
|
|
|
def SKLWriteResGroup146 : SchedWriteRes<[SKLPort0,SKLPort23]> {
|
|
let Latency = 11;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup146], (instregex "MUL_F(32|64)m")>;
|
|
|
|
def SKLWriteResGroup149 : SchedWriteRes<[SKLPort5,SKLPort23]> {
|
|
let Latency = 11;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [2,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup149], (instregex "FICOM(P?)(16|32)m")>;
|
|
|
|
def SKLWriteResGroup154 : SchedWriteRes<[SKLPort1,SKLPort06,SKLPort0156]> {
|
|
let Latency = 11;
|
|
let NumMicroOps = 7;
|
|
let ReleaseAtCycles = [2,3,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup154], (instregex "RCL(16|32|64)rCL",
|
|
"RCR(16|32|64)rCL")>;
|
|
|
|
def SKLWriteResGroup155 : SchedWriteRes<[SKLPort1,SKLPort06,SKLPort15,SKLPort0156]> {
|
|
let Latency = 11;
|
|
let NumMicroOps = 9;
|
|
let ReleaseAtCycles = [1,5,1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup155], (instrs RCL8rCL)>;
|
|
|
|
def SKLWriteResGroup156 : SchedWriteRes<[SKLPort06,SKLPort0156]> {
|
|
let Latency = 11;
|
|
let NumMicroOps = 11;
|
|
let ReleaseAtCycles = [2,9];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup156], (instrs LOOPE, LOOPNE)>;
|
|
|
|
def SKLWriteResGroup162 : SchedWriteRes<[SKLPort5,SKLPort23]> {
|
|
let Latency = 13;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [2,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup162], (instregex "(ADD|SUB|SUBR)_FI(16|32)m")>;
|
|
|
|
def SKLWriteResGroup169 : SchedWriteRes<[SKLPort0,SKLPort5,SKLPort23]> {
|
|
let Latency = 14;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup169], (instregex "MUL_FI(16|32)m")>;
|
|
|
|
def SKLWriteResGroup170 : SchedWriteRes<[SKLPort1,SKLPort06,SKLPort15,SKLPort0156]> {
|
|
let Latency = 14;
|
|
let NumMicroOps = 10;
|
|
let ReleaseAtCycles = [2,4,1,3];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup170], (instrs RCR8rCL)>;
|
|
|
|
def SKLWriteResGroup171 : SchedWriteRes<[SKLPort0]> {
|
|
let Latency = 15;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup171], (instregex "DIVR_(FPrST0|FST0r|FrST0)")>;
|
|
|
|
def SKLWriteResGroup174 : SchedWriteRes<[SKLPort1,SKLPort23,SKLPort237,SKLPort06,SKLPort15,SKLPort0156]> {
|
|
let Latency = 15;
|
|
let NumMicroOps = 10;
|
|
let ReleaseAtCycles = [1,1,1,5,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup174], (instregex "RCL(8|16|32|64)mCL")>;
|
|
|
|
def SKLWriteResGroup177 : SchedWriteRes<[SKLPort4,SKLPort23,SKLPort237,SKLPort06,SKLPort15,SKLPort0156]> {
|
|
let Latency = 16;
|
|
let NumMicroOps = 14;
|
|
let ReleaseAtCycles = [1,1,1,4,2,5];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup177], (instrs CMPXCHG8B)>;
|
|
|
|
def SKLWriteResGroup178 : SchedWriteRes<[SKLPort0156]> {
|
|
let Latency = 16;
|
|
let NumMicroOps = 16;
|
|
let ReleaseAtCycles = [16];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup178], (instrs VZEROALL)>;
|
|
|
|
def SKLWriteResGroup180 : SchedWriteRes<[SKLPort0,SKLPort1,SKLPort5,SKLPort6,SKLPort05,SKLPort0156]> {
|
|
let Latency = 17;
|
|
let NumMicroOps = 15;
|
|
let ReleaseAtCycles = [2,1,2,4,2,4];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup180], (instrs XCH_F)>;
|
|
|
|
def SKLWriteResGroup184 : SchedWriteRes<[SKLPort5,SKLPort6,SKLPort06,SKLPort0156]> {
|
|
let Latency = 18;
|
|
let NumMicroOps = 8;
|
|
let ReleaseAtCycles = [1,1,1,5];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup184], (instrs CPUID, RDTSC)>;
|
|
|
|
def SKLWriteResGroup185 : SchedWriteRes<[SKLPort1,SKLPort23,SKLPort237,SKLPort06,SKLPort15,SKLPort0156]> {
|
|
let Latency = 18;
|
|
let NumMicroOps = 11;
|
|
let ReleaseAtCycles = [2,1,1,4,1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup185], (instregex "RCR(8|16|32|64)mCL")>;
|
|
|
|
def SKLWriteResGroup189 : SchedWriteRes<[SKLPort0]> {
|
|
let Latency = 20;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup189], (instregex "DIV_(FPrST0|FST0r|FrST0)")>;
|
|
|
|
def SKLWriteResGroup192 : SchedWriteRes<[SKLPort4,SKLPort5,SKLPort6,SKLPort23,SKLPort237,SKLPort06,SKLPort0156]> {
|
|
let Latency = 20;
|
|
let NumMicroOps = 8;
|
|
let ReleaseAtCycles = [1,1,1,1,1,1,2];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup192], (instrs INSB, INSL, INSW)>;
|
|
|
|
def SKLWriteResGroup193 : SchedWriteRes<[SKLPort5,SKLPort6,SKLPort0156]> {
|
|
let Latency = 20;
|
|
let NumMicroOps = 10;
|
|
let ReleaseAtCycles = [1,2,7];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup193], (instrs MWAITrr)>;
|
|
|
|
def SKLWriteResGroup196 : SchedWriteRes<[SKLPort0,SKLPort23]> {
|
|
let Latency = 22;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup196], (instregex "DIV_F(32|64)m")>;
|
|
|
|
def SKLWriteResGroupVEX2 : SchedWriteRes<[SKLPort0, SKLPort23, SKLPort5, SKLPort015]> {
|
|
let Latency = 18;
|
|
let NumMicroOps = 5; // 2 uops perform multiple loads
|
|
let ReleaseAtCycles = [1,2,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroupVEX2], (instrs VGATHERDPDrm, VPGATHERDQrm,
|
|
VGATHERQPDrm, VPGATHERQQrm,
|
|
VGATHERQPSrm, VPGATHERQDrm)>;
|
|
|
|
def SKLWriteResGroupVEX4 : SchedWriteRes<[SKLPort0, SKLPort23, SKLPort5, SKLPort015]> {
|
|
let Latency = 20;
|
|
let NumMicroOps = 5; // 2 uops peform multiple loads
|
|
let ReleaseAtCycles = [1,4,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroupVEX4], (instrs VGATHERDPDYrm, VPGATHERDQYrm,
|
|
VGATHERDPSrm, VPGATHERDDrm,
|
|
VGATHERQPDYrm, VPGATHERQQYrm,
|
|
VGATHERQPSYrm, VPGATHERQDYrm)>;
|
|
|
|
def SKLWriteResGroupVEX8 : SchedWriteRes<[SKLPort0, SKLPort23, SKLPort5, SKLPort015]> {
|
|
let Latency = 22;
|
|
let NumMicroOps = 5; // 2 uops perform multiple loads
|
|
let ReleaseAtCycles = [1,8,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroupVEX8], (instrs VGATHERDPSYrm, VPGATHERDDYrm)>;
|
|
|
|
def SKLWriteResGroup198 : SchedWriteRes<[SKLPort0,SKLPort4,SKLPort5,SKLPort23,SKLPort237,SKLPort06,SKLPort0156]> {
|
|
let Latency = 23;
|
|
let NumMicroOps = 19;
|
|
let ReleaseAtCycles = [2,1,4,1,1,4,6];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup198], (instrs CMPXCHG16B)>;
|
|
|
|
def SKLWriteResGroup202 : SchedWriteRes<[SKLPort0,SKLPort5,SKLPort23]> {
|
|
let Latency = 25;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup202], (instregex "DIV_FI(16|32)m")>;
|
|
|
|
def SKLWriteResGroup206 : SchedWriteRes<[SKLPort0,SKLPort23]> {
|
|
let Latency = 27;
|
|
let NumMicroOps = 2;
|
|
let ReleaseAtCycles = [1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup206], (instregex "DIVR_F(32|64)m")>;
|
|
|
|
def SKLWriteResGroup208 : SchedWriteRes<[SKLPort0,SKLPort5,SKLPort23]> {
|
|
let Latency = 30;
|
|
let NumMicroOps = 3;
|
|
let ReleaseAtCycles = [1,1,1];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup208], (instregex "DIVR_FI(16|32)m")>;
|
|
|
|
def SKLWriteResGroup209 : SchedWriteRes<[SKLPort5,SKLPort6,SKLPort23,SKLPort06,SKLPort0156]> {
|
|
let Latency = 35;
|
|
let NumMicroOps = 23;
|
|
let ReleaseAtCycles = [1,5,3,4,10];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup209], (instregex "IN(8|16|32)ri",
|
|
"IN(8|16|32)rr")>;
|
|
|
|
def SKLWriteResGroup210 : SchedWriteRes<[SKLPort5,SKLPort6,SKLPort23,SKLPort237,SKLPort06,SKLPort0156]> {
|
|
let Latency = 35;
|
|
let NumMicroOps = 23;
|
|
let ReleaseAtCycles = [1,5,2,1,4,10];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup210], (instregex "OUT(8|16|32)ir",
|
|
"OUT(8|16|32)rr")>;
|
|
|
|
def SKLWriteResGroup211 : SchedWriteRes<[SKLPort1,SKLPort6,SKLPort23,SKLPort0156]> {
|
|
let Latency = 37;
|
|
let NumMicroOps = 31;
|
|
let ReleaseAtCycles = [1,8,1,21];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup211], (instregex "XRSTOR(64)?")>;
|
|
|
|
def SKLWriteResGroup212 : SchedWriteRes<[SKLPort1,SKLPort4,SKLPort5,SKLPort6,SKLPort23,SKLPort237,SKLPort15,SKLPort0156]> {
|
|
let Latency = 40;
|
|
let NumMicroOps = 18;
|
|
let ReleaseAtCycles = [1,1,2,3,1,1,1,8];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup212], (instrs VMCLEARm)>;
|
|
|
|
def SKLWriteResGroup213 : SchedWriteRes<[SKLPort4,SKLPort6,SKLPort23,SKLPort237,SKLPort0156]> {
|
|
let Latency = 41;
|
|
let NumMicroOps = 39;
|
|
let ReleaseAtCycles = [1,10,1,1,26];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup213], (instrs XSAVE64)>;
|
|
|
|
def SKLWriteResGroup214 : SchedWriteRes<[SKLPort5,SKLPort0156]> {
|
|
let Latency = 42;
|
|
let NumMicroOps = 22;
|
|
let ReleaseAtCycles = [2,20];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup214], (instrs RDTSCP)>;
|
|
|
|
def SKLWriteResGroup215 : SchedWriteRes<[SKLPort4,SKLPort6,SKLPort23,SKLPort237,SKLPort0156]> {
|
|
let Latency = 42;
|
|
let NumMicroOps = 40;
|
|
let ReleaseAtCycles = [1,11,1,1,26];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup215], (instrs XSAVE)>;
|
|
def: InstRW<[SKLWriteResGroup215], (instregex "XSAVEC", "XSAVES")>;
|
|
|
|
def SKLWriteResGroup216 : SchedWriteRes<[SKLPort4,SKLPort6,SKLPort23,SKLPort237,SKLPort0156]> {
|
|
let Latency = 46;
|
|
let NumMicroOps = 44;
|
|
let ReleaseAtCycles = [1,11,1,1,30];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup216], (instregex "XSAVEOPT")>;
|
|
|
|
def SKLWriteResGroup217 : SchedWriteRes<[SKLPort0,SKLPort23,SKLPort05,SKLPort06,SKLPort0156]> {
|
|
let Latency = 62;
|
|
let NumMicroOps = 64;
|
|
let ReleaseAtCycles = [2,8,5,10,39];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup217], (instrs FLDENVm)>;
|
|
|
|
def SKLWriteResGroup218 : SchedWriteRes<[SKLPort0,SKLPort6,SKLPort23,SKLPort05,SKLPort06,SKLPort15,SKLPort0156]> {
|
|
let Latency = 63;
|
|
let NumMicroOps = 88;
|
|
let ReleaseAtCycles = [4,4,31,1,2,1,45];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup218], (instrs FXRSTOR64)>;
|
|
|
|
def SKLWriteResGroup219 : SchedWriteRes<[SKLPort0,SKLPort6,SKLPort23,SKLPort05,SKLPort06,SKLPort15,SKLPort0156]> {
|
|
let Latency = 63;
|
|
let NumMicroOps = 90;
|
|
let ReleaseAtCycles = [4,2,33,1,2,1,47];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup219], (instrs FXRSTOR)>;
|
|
|
|
def SKLWriteResGroup220 : SchedWriteRes<[SKLPort5,SKLPort05,SKLPort0156]> {
|
|
let Latency = 75;
|
|
let NumMicroOps = 15;
|
|
let ReleaseAtCycles = [6,3,6];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup220], (instrs FNINIT)>;
|
|
|
|
def SKLWriteResGroup223 : SchedWriteRes<[SKLPort0,SKLPort1,SKLPort4,SKLPort5,SKLPort6,SKLPort237,SKLPort06,SKLPort0156]> {
|
|
let Latency = 106;
|
|
let NumMicroOps = 100;
|
|
let ReleaseAtCycles = [9,1,11,16,1,11,21,30];
|
|
}
|
|
def: InstRW<[SKLWriteResGroup223], (instrs FSTENVm)>;
|
|
|
|
def: InstRW<[WriteZero], (instrs CLC)>;
|
|
|
|
|
|
// Instruction 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 "Skylake 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 SKLWriteZeroLatency : SchedWriteRes<[]> {
|
|
let Latency = 0;
|
|
}
|
|
|
|
def SKLWriteZeroIdiom : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [SKLWriteZeroLatency]>,
|
|
SchedVar<NoSchedPred, [WriteALU]>
|
|
]>;
|
|
def : InstRW<[SKLWriteZeroIdiom], (instrs SUB32rr, SUB64rr,
|
|
XOR32rr, XOR64rr)>;
|
|
|
|
def SKLWriteFZeroIdiom : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [SKLWriteZeroLatency]>,
|
|
SchedVar<NoSchedPred, [WriteFLogic]>
|
|
]>;
|
|
def : InstRW<[SKLWriteFZeroIdiom], (instrs XORPSrr, VXORPSrr, XORPDrr,
|
|
VXORPDrr)>;
|
|
|
|
def SKLWriteFZeroIdiomY : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [SKLWriteZeroLatency]>,
|
|
SchedVar<NoSchedPred, [WriteFLogicY]>
|
|
]>;
|
|
def : InstRW<[SKLWriteFZeroIdiomY], (instrs VXORPSYrr, VXORPDYrr)>;
|
|
|
|
def SKLWriteVZeroIdiomLogicX : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [SKLWriteZeroLatency]>,
|
|
SchedVar<NoSchedPred, [WriteVecLogicX]>
|
|
]>;
|
|
def : InstRW<[SKLWriteVZeroIdiomLogicX], (instrs PXORrr, VPXORrr)>;
|
|
|
|
def SKLWriteVZeroIdiomLogicY : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [SKLWriteZeroLatency]>,
|
|
SchedVar<NoSchedPred, [WriteVecLogicY]>
|
|
]>;
|
|
def : InstRW<[SKLWriteVZeroIdiomLogicY], (instrs VPXORYrr)>;
|
|
|
|
def SKLWriteVZeroIdiomALUX : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [SKLWriteZeroLatency]>,
|
|
SchedVar<NoSchedPred, [WriteVecALUX]>
|
|
]>;
|
|
def : InstRW<[SKLWriteVZeroIdiomALUX], (instrs PCMPGTBrr, VPCMPGTBrr,
|
|
PCMPGTDrr, VPCMPGTDrr,
|
|
PCMPGTWrr, VPCMPGTWrr)>;
|
|
|
|
def SKLWriteVZeroIdiomALUY : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [SKLWriteZeroLatency]>,
|
|
SchedVar<NoSchedPred, [WriteVecALUY]>
|
|
]>;
|
|
def : InstRW<[SKLWriteVZeroIdiomALUY], (instrs VPCMPGTBYrr,
|
|
VPCMPGTDYrr,
|
|
VPCMPGTWYrr)>;
|
|
|
|
def SKLWritePSUB : SchedWriteRes<[SKLPort015]> {
|
|
let Latency = 1;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
|
|
def SKLWriteVZeroIdiomPSUB : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [SKLWriteZeroLatency]>,
|
|
SchedVar<NoSchedPred, [SKLWritePSUB]>
|
|
]>;
|
|
def : InstRW<[SKLWriteVZeroIdiomPSUB], (instrs PSUBBrr, VPSUBBrr,
|
|
PSUBDrr, VPSUBDrr,
|
|
PSUBQrr, VPSUBQrr,
|
|
PSUBWrr, VPSUBWrr,
|
|
VPSUBBYrr,
|
|
VPSUBDYrr,
|
|
VPSUBQYrr,
|
|
VPSUBWYrr)>;
|
|
|
|
def SKLWritePCMPGTQ : SchedWriteRes<[SKLPort5]> {
|
|
let Latency = 3;
|
|
let NumMicroOps = 1;
|
|
let ReleaseAtCycles = [1];
|
|
}
|
|
|
|
def SKLWriteVZeroIdiomPCMPGTQ : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<ZeroIdiomPredicate>, [SKLWriteZeroLatency]>,
|
|
SchedVar<NoSchedPred, [SKLWritePCMPGTQ]>
|
|
]>;
|
|
def : InstRW<[SKLWriteVZeroIdiomPCMPGTQ], (instrs PCMPGTQrr, VPCMPGTQrr,
|
|
VPCMPGTQYrr)>;
|
|
|
|
|
|
// CMOVs that use both Z and C flag require an extra uop.
|
|
def SKLWriteCMOVA_CMOVBErr : SchedWriteRes<[SKLPort06]> {
|
|
let Latency = 2;
|
|
let ReleaseAtCycles = [2];
|
|
let NumMicroOps = 2;
|
|
}
|
|
|
|
def SKLWriteCMOVA_CMOVBErm : SchedWriteRes<[SKLPort23,SKLPort06]> {
|
|
let Latency = 7;
|
|
let ReleaseAtCycles = [1,2];
|
|
let NumMicroOps = 3;
|
|
}
|
|
|
|
def SKLCMOVA_CMOVBErr : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<IsCMOVArr_Or_CMOVBErr>, [SKLWriteCMOVA_CMOVBErr]>,
|
|
SchedVar<NoSchedPred, [WriteCMOV]>
|
|
]>;
|
|
|
|
def SKLCMOVA_CMOVBErm : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<IsCMOVArm_Or_CMOVBErm>, [SKLWriteCMOVA_CMOVBErm]>,
|
|
SchedVar<NoSchedPred, [WriteCMOV.Folded]>
|
|
]>;
|
|
|
|
def : InstRW<[SKLCMOVA_CMOVBErr], (instrs CMOV16rr, CMOV32rr, CMOV64rr)>;
|
|
def : InstRW<[SKLCMOVA_CMOVBErm], (instrs CMOV16rm, CMOV32rm, CMOV64rm)>;
|
|
|
|
// SETCCs that use both Z and C flag require an extra uop.
|
|
def SKLWriteSETA_SETBEr : SchedWriteRes<[SKLPort06]> {
|
|
let Latency = 2;
|
|
let ReleaseAtCycles = [2];
|
|
let NumMicroOps = 2;
|
|
}
|
|
|
|
def SKLWriteSETA_SETBEm : SchedWriteRes<[SKLPort4,SKLPort237,SKLPort06]> {
|
|
let Latency = 3;
|
|
let ReleaseAtCycles = [1,1,2];
|
|
let NumMicroOps = 4;
|
|
}
|
|
|
|
def SKLSETA_SETBErr : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<IsSETAr_Or_SETBEr>, [SKLWriteSETA_SETBEr]>,
|
|
SchedVar<NoSchedPred, [WriteSETCC]>
|
|
]>;
|
|
|
|
def SKLSETA_SETBErm : SchedWriteVariant<[
|
|
SchedVar<MCSchedPredicate<IsSETAm_Or_SETBEm>, [SKLWriteSETA_SETBEm]>,
|
|
SchedVar<NoSchedPred, [WriteSETCCStore]>
|
|
]>;
|
|
|
|
def : InstRW<[SKLSETA_SETBErr], (instrs SETCCr)>;
|
|
def : InstRW<[SKLSETA_SETBErm], (instrs SETCCm)>;
|
|
|
|
///////////////////////////////////////////////////////////////////////////////
|
|
// Dependency breaking instructions.
|
|
///////////////////////////////////////////////////////////////////////////////
|
|
|
|
def : IsZeroIdiomFunction<[
|
|
// GPR Zero-idioms.
|
|
DepBreakingClass<[ SUB32rr, SUB64rr, XOR32rr, XOR64rr ], ZeroIdiomPredicate>,
|
|
|
|
// SSE Zero-idioms.
|
|
DepBreakingClass<[
|
|
// fp variants.
|
|
XORPSrr, XORPDrr,
|
|
|
|
// int variants.
|
|
PXORrr,
|
|
PSUBBrr, PSUBWrr, PSUBDrr, PSUBQrr,
|
|
PCMPGTBrr, PCMPGTDrr, PCMPGTQrr, PCMPGTWrr
|
|
], ZeroIdiomPredicate>,
|
|
|
|
// AVX Zero-idioms.
|
|
DepBreakingClass<[
|
|
// xmm fp variants.
|
|
VXORPSrr, VXORPDrr,
|
|
|
|
// xmm int variants.
|
|
VPXORrr,
|
|
VPSUBBrr, VPSUBWrr, VPSUBDrr, VPSUBQrr,
|
|
VPCMPGTBrr, VPCMPGTWrr, VPCMPGTDrr, VPCMPGTQrr,
|
|
|
|
// ymm variants.
|
|
VXORPSYrr, VXORPDYrr, VPXORYrr,
|
|
VPSUBBYrr, VPSUBWYrr, VPSUBDYrr, VPSUBQYrr,
|
|
VPCMPGTBYrr, VPCMPGTWYrr, VPCMPGTDYrr, VPCMPGTQYrr
|
|
], ZeroIdiomPredicate>,
|
|
]>;
|
|
|
|
} // SchedModel
|