mirror of
https://github.com/RPCS3/llvm.git
synced 2024-12-27 14:45:50 +00:00
bc0bdb26da
In preparation for adding scheduling definitions for the POWER7, split some PPC itinerary classes so that the P7's latencies and hazards can be better described. For the most part, this means differentiating indexed from non-index pre-increment loads and stores. Also, differentiate single from double-precision sqrt. No functionality change intended (except for a more-specific latency for single-precision sqrt on the A2). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195980 91177308-0d34-0410-b5e6-96231b3b80d8
130 lines
7.2 KiB
TableGen
130 lines
7.2 KiB
TableGen
//===-- PPCScheduleG5.td - PPC G5 Scheduling Definitions ---*- tablegen -*-===//
|
|
//
|
|
// The LLVM Compiler Infrastructure
|
|
//
|
|
// This file is distributed under the University of Illinois Open Source
|
|
// License. See LICENSE.TXT for details.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
//
|
|
// This file defines the itinerary class data for the G5 (970) processor.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
def G5_BPU : FuncUnit; // Branch unit
|
|
def G5_SLU : FuncUnit; // Store/load unit
|
|
def G5_SRU : FuncUnit; // special register unit
|
|
def G5_IU1 : FuncUnit; // integer unit 1 (simple)
|
|
def G5_IU2 : FuncUnit; // integer unit 2 (complex)
|
|
def G5_FPU1 : FuncUnit; // floating point unit 1
|
|
def G5_FPU2 : FuncUnit; // floating point unit 2
|
|
def G5_VPU : FuncUnit; // vector permutation unit
|
|
def G5_VIU1 : FuncUnit; // vector integer unit 1 (simple)
|
|
def G5_VIU2 : FuncUnit; // vector integer unit 2 (complex)
|
|
def G5_VFPU : FuncUnit; // vector floating point unit
|
|
|
|
def G5Itineraries : ProcessorItineraries<
|
|
[G5_IU1, G5_IU2, G5_SLU, G5_BPU, G5_FPU1, G5_FPU2,
|
|
G5_VFPU, G5_VIU1, G5_VIU2, G5_VPU], [], [
|
|
InstrItinData<IIC_IntSimple , [InstrStage<2, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntGeneral , [InstrStage<2, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntCompare , [InstrStage<3, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntDivD , [InstrStage<68, [G5_IU1]>]>,
|
|
InstrItinData<IIC_IntDivW , [InstrStage<36, [G5_IU1]>]>,
|
|
InstrItinData<IIC_IntMFFS , [InstrStage<6, [G5_IU2]>]>,
|
|
InstrItinData<IIC_IntMFVSCR , [InstrStage<1, [G5_VFPU]>]>,
|
|
InstrItinData<IIC_IntMTFSB0 , [InstrStage<6, [G5_FPU1, G5_FPU2]>]>,
|
|
InstrItinData<IIC_IntMulHD , [InstrStage<7, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntMulHW , [InstrStage<5, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntMulHWU , [InstrStage<5, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntMulLI , [InstrStage<4, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntRFID , [InstrStage<1, [G5_IU2]>]>,
|
|
InstrItinData<IIC_IntRotateD , [InstrStage<2, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntRotateDI , [InstrStage<2, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntRotate , [InstrStage<4, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntShift , [InstrStage<2, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntTrapD , [InstrStage<1, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_IntTrapW , [InstrStage<1, [G5_IU1, G5_IU2]>]>,
|
|
InstrItinData<IIC_BrB , [InstrStage<1, [G5_BPU]>]>,
|
|
InstrItinData<IIC_BrCR , [InstrStage<4, [G5_BPU]>]>,
|
|
InstrItinData<IIC_BrMCR , [InstrStage<2, [G5_BPU]>]>,
|
|
InstrItinData<IIC_BrMCRX , [InstrStage<3, [G5_BPU]>]>,
|
|
InstrItinData<IIC_LdStDCBF , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLoad , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLoadUpd , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLoadUpdX, [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStStore , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStStoreUpd, [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStDSS , [InstrStage<10, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStICBI , [InstrStage<40, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStSTFD , [InstrStage<4, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStSTFDU , [InstrStage<4, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLD , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLDU , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLDUX , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLDARX , [InstrStage<11, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLFD , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLFDU , [InstrStage<5, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLFDUX , [InstrStage<5, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLHA , [InstrStage<5, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLHAU , [InstrStage<5, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLHAUX , [InstrStage<5, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLMW , [InstrStage<64, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLVecX , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLWA , [InstrStage<5, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStLWARX , [InstrStage<11, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStSLBIA , [InstrStage<40, [G5_SLU]>]>, // needs work
|
|
InstrItinData<IIC_LdStSLBIE , [InstrStage<2, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStSTD , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStSTDU , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStSTDUX , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStSTDCX , [InstrStage<11, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStSTVEBX , [InstrStage<5, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStSTWCX , [InstrStage<11, [G5_SLU]>]>,
|
|
InstrItinData<IIC_LdStSync , [InstrStage<35, [G5_SLU]>]>,
|
|
InstrItinData<IIC_SprISYNC , [InstrStage<40, [G5_SLU]>]>, // needs work
|
|
InstrItinData<IIC_SprMFSR , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_SprMTMSR , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_SprMTSR , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_SprTLBSYNC , [InstrStage<3, [G5_SLU]>]>,
|
|
InstrItinData<IIC_SprMFCR , [InstrStage<2, [G5_IU2]>]>,
|
|
InstrItinData<IIC_SprMFCRF , [InstrStage<2, [G5_IU2]>]>,
|
|
InstrItinData<IIC_SprMFMSR , [InstrStage<3, [G5_IU2]>]>,
|
|
InstrItinData<IIC_SprMFSPR , [InstrStage<3, [G5_IU2]>]>,
|
|
InstrItinData<IIC_SprMFTB , [InstrStage<10, [G5_IU2]>]>,
|
|
InstrItinData<IIC_SprMTSPR , [InstrStage<8, [G5_IU2]>]>,
|
|
InstrItinData<IIC_SprSC , [InstrStage<1, [G5_IU2]>]>,
|
|
InstrItinData<IIC_FPGeneral , [InstrStage<6, [G5_FPU1, G5_FPU2]>]>,
|
|
InstrItinData<IIC_FPAddSub , [InstrStage<6, [G5_FPU1, G5_FPU2]>]>,
|
|
InstrItinData<IIC_FPCompare , [InstrStage<8, [G5_FPU1, G5_FPU2]>]>,
|
|
InstrItinData<IIC_FPDivD , [InstrStage<33, [G5_FPU1, G5_FPU2]>]>,
|
|
InstrItinData<IIC_FPDivS , [InstrStage<33, [G5_FPU1, G5_FPU2]>]>,
|
|
InstrItinData<IIC_FPFused , [InstrStage<6, [G5_FPU1, G5_FPU2]>]>,
|
|
InstrItinData<IIC_FPRes , [InstrStage<6, [G5_FPU1, G5_FPU2]>]>,
|
|
InstrItinData<IIC_FPSqrtD , [InstrStage<40, [G5_FPU1, G5_FPU2]>]>,
|
|
InstrItinData<IIC_FPSqrtS , [InstrStage<40, [G5_FPU1, G5_FPU2]>]>,
|
|
InstrItinData<IIC_VecGeneral , [InstrStage<2, [G5_VIU1]>]>,
|
|
InstrItinData<IIC_VecFP , [InstrStage<8, [G5_VFPU]>]>,
|
|
InstrItinData<IIC_VecFPCompare, [InstrStage<2, [G5_VFPU]>]>,
|
|
InstrItinData<IIC_VecComplex , [InstrStage<5, [G5_VIU2]>]>,
|
|
InstrItinData<IIC_VecPerm , [InstrStage<3, [G5_VPU]>]>,
|
|
InstrItinData<IIC_VecFPRound , [InstrStage<8, [G5_VFPU]>]>,
|
|
InstrItinData<IIC_VecVSL , [InstrStage<2, [G5_VIU1]>]>,
|
|
InstrItinData<IIC_VecVSR , [InstrStage<3, [G5_VPU]>]>
|
|
]>;
|
|
|
|
// ===---------------------------------------------------------------------===//
|
|
// G5 machine model for scheduling and other instruction cost heuristics.
|
|
|
|
def G5Model : SchedMachineModel {
|
|
let IssueWidth = 4; // 4 (non-branch) instructions are dispatched per cycle.
|
|
let MinLatency = 0; // Out-of-order dispatch.
|
|
let LoadLatency = 3; // Optimistic load latency assuming bypass.
|
|
// This is overriden by OperandCycles if the
|
|
// Itineraries are queried instead.
|
|
let MispredictPenalty = 16;
|
|
|
|
let Itineraries = G5Itineraries;
|
|
}
|
|
|