llvm-6502/lib/Target/PowerPC/PPCScheduleE5500.td

310 lines
18 KiB
TableGen
Raw Normal View History

//===-- PPCScheduleE500mc.td - e5500 Scheduling Defs -------*- 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 Freescale e5500 64-bit
// Power processor.
//
// All information is derived from the "e5500 Core Reference Manual",
// Freescale Document Number e5500RM, Rev. 1, 03/2012.
//
//===----------------------------------------------------------------------===//
// Relevant functional units in the Freescale e5500 core
// (These are the same as for the e500mc)
//
// * Decode & Dispatch
// Can dispatch up to 2 instructions per clock cycle to either the GPR Issue
// queues (GIQx), FP Issue Queue (FIQ), or Branch issue queue (BIQ).
// def DIS0 : FuncUnit;
// def DIS1 : FuncUnit;
// * Execute
// 6 pipelined execution units: SFX0, SFX1, BU, FPU, LSU, CFX.
// The CFX has a bypass path, allowing non-divide instructions to execute
// while a divide instruction is being executed.
// def SFX0 : FuncUnit; // Simple unit 0
// def SFX1 : FuncUnit; // Simple unit 1
// def BU : FuncUnit; // Branch unit
// def CFX_DivBypass
// : FuncUnit; // CFX divide bypass path
// def CFX_0 : FuncUnit; // CFX pipeline stage 0
def CFX_1 : FuncUnit; // CFX pipeline stage 1
// def LSU_0 : FuncUnit; // LSU pipeline
// def FPU_0 : FuncUnit; // FPU pipeline
def PPCE5500Itineraries : ProcessorItineraries<
[DIS0, DIS1, SFX0, SFX1, BU, CFX_DivBypass, CFX_0, CFX_1,
LSU_0, FPU_0],
[CR_Bypass, GPR_Bypass, FPR_Bypass], [
InstrItinData<IntSimple , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1]>],
[5, 2, 2], // Latency = 1
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntGeneral , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1]>],
[5, 2, 2], // Latency = 1
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntCompare , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1]>],
[6, 2, 2], // Latency = 1 or 2
[CR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntDivD , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [CFX_0], 0>,
InstrStage<26, [CFX_DivBypass]>],
[30, 2, 2], // Latency= 4..26, Repeat rate= 4..26
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntDivW , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [CFX_0], 0>,
InstrStage<16, [CFX_DivBypass]>],
[20, 2, 2], // Latency= 4..16, Repeat rate= 4..16
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntMFFS , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [FPU_0]>],
[11], // Latency = 7, Repeat rate = 1
[FPR_Bypass]>,
InstrItinData<IntMTFSB0 , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<7, [FPU_0]>],
[11, 2, 2], // Latency = 7, Repeat rate = 7
[NoBypass, NoBypass, NoBypass]>,
InstrItinData<IntMulHD , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [CFX_0], 0>,
InstrStage<2, [CFX_1]>],
[9, 2, 2], // Latency = 4..7, Repeat rate = 2..4
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntMulHW , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [CFX_0], 0>,
InstrStage<1, [CFX_1]>],
[8, 2, 2], // Latency = 4, Repeat rate = 1
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntMulHWU , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [CFX_0], 0>,
InstrStage<1, [CFX_1]>],
[8, 2, 2], // Latency = 4, Repeat rate = 1
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntMulLI , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [CFX_0], 0>,
InstrStage<2, [CFX_1]>],
[8, 2, 2], // Latency = 4 or 5, Repeat = 2
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntRotate , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1]>],
[5, 2, 2], // Latency = 1
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntRotateD , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<2, [SFX0, SFX1]>],
[6, 2, 2], // Latency = 2, Repeat rate = 2
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntRotateDI , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1]>],
[5, 2, 2], // Latency = 1, Repeat rate = 1
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntShift , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<2, [SFX0, SFX1]>],
[6, 2, 2], // Latency = 2, Repeat rate = 2
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<IntTrapW , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<2, [SFX0]>],
[6, 2], // Latency = 2, Repeat rate = 2
[GPR_Bypass, GPR_Bypass]>,
InstrItinData<BrB , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [BU]>],
[5, 2], // Latency = 1
[NoBypass, GPR_Bypass]>,
InstrItinData<BrCR , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [BU]>],
[5, 2, 2], // Latency = 1
[CR_Bypass, CR_Bypass, CR_Bypass]>,
InstrItinData<BrMCR , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [BU]>],
[5, 2], // Latency = 1
[CR_Bypass, CR_Bypass]>,
InstrItinData<BrMCRX , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [CFX_0]>],
[5, 2, 2], // Latency = 1
[CR_Bypass, GPR_Bypass]>,
InstrItinData<LdStDCBA , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[GPR_Bypass, GPR_Bypass]>,
InstrItinData<LdStDCBF , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[GPR_Bypass, GPR_Bypass]>,
InstrItinData<LdStDCBI , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[GPR_Bypass, GPR_Bypass]>,
InstrItinData<LdStLoad , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3
[GPR_Bypass, GPR_Bypass]>,
InstrItinData<LdStLoadUpd , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[GPR_Bypass, GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<LdStLD , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[GPR_Bypass, GPR_Bypass]>,
InstrItinData<LdStLDARX , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<3, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 3
[GPR_Bypass, GPR_Bypass]>,
InstrItinData<LdStLDU , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[GPR_Bypass, GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<LdStStore , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[NoBypass, GPR_Bypass]>,
InstrItinData<LdStStoreUpd, [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[NoBypass, GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<LdStICBI , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[NoBypass, GPR_Bypass]>,
InstrItinData<LdStSTFD , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2, 2], // Latency = 3, Repeat rate = 1
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<LdStSTFDU , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2, 2], // Latency = 3, Repeat rate = 1
[GPR_Bypass, GPR_Bypass, GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<LdStLFD , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[8, 2, 2], // Latency = 4, Repeat rate = 1
[FPR_Bypass, GPR_Bypass, GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<LdStLFDU , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1], 0>,
InstrStage<1, [LSU_0]>],
[8, 2, 2], // Latency = 4, Repeat rate = 1
[FPR_Bypass, GPR_Bypass, GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<LdStLHA , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3
[GPR_Bypass, GPR_Bypass]>,
InstrItinData<LdStLHAU , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[GPR_Bypass, GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<LdStLMW , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<4, [LSU_0]>],
[8, 2], // Latency = r+3, Repeat rate = r+3
[NoBypass, GPR_Bypass]>,
InstrItinData<LdStLWARX , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<3, [LSU_0]>],
[7, 2, 2], // Latency = 3, Repeat rate = 3
[GPR_Bypass, GPR_Bypass, GPR_Bypass]>,
InstrItinData<LdStSTD , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[NoBypass, GPR_Bypass]>,
InstrItinData<LdStSTDCX , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[NoBypass, GPR_Bypass]>,
InstrItinData<LdStSTDU , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[NoBypass, GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<LdStSTWCX , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>],
[7, 2], // Latency = 3, Repeat rate = 1
[NoBypass, GPR_Bypass]>,
InstrItinData<LdStSync , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0]>]>,
InstrItinData<SprMTMSR , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<2, [CFX_0]>],
[6, 2], // Latency = 2, Repeat rate = 4
[GPR_Bypass, GPR_Bypass]>,
InstrItinData<SprTLBSYNC , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [LSU_0], 0>]>,
InstrItinData<SprMFCR , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<5, [CFX_0]>],
[9, 2], // Latency = 5, Repeat rate = 5
[GPR_Bypass, CR_Bypass]>,
InstrItinData<SprMFMSR , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<4, [SFX0]>],
[8, 2], // Latency = 4, Repeat rate = 4
[GPR_Bypass, GPR_Bypass]>,
InstrItinData<SprMFSPR , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [CFX_0]>],
[5], // Latency = 1, Repeat rate = 1
[GPR_Bypass]>,
InstrItinData<SprMFTB , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<4, [CFX_0]>],
[8, 2], // Latency = 4, Repeat rate = 4
[NoBypass, GPR_Bypass]>,
InstrItinData<SprMTSPR , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [SFX0, SFX1]>],
[5], // Latency = 1, Repeat rate = 1
[GPR_Bypass]>,
InstrItinData<FPGeneral , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [FPU_0]>],
[11, 2, 2], // Latency = 7, Repeat rate = 1
[FPR_Bypass, FPR_Bypass, FPR_Bypass]>,
InstrItinData<FPAddSub , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [FPU_0]>],
[11, 2, 2], // Latency = 7, Repeat rate = 1
[FPR_Bypass, FPR_Bypass, FPR_Bypass]>,
InstrItinData<FPCompare , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [FPU_0]>],
[11, 2, 2], // Latency = 7, Repeat rate = 1
[CR_Bypass, FPR_Bypass, FPR_Bypass]>,
InstrItinData<FPDivD , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<31, [FPU_0]>],
[39, 2, 2], // Latency = 35, Repeat rate = 31
[FPR_Bypass, FPR_Bypass, FPR_Bypass]>,
InstrItinData<FPDivS , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<16, [FPU_0]>],
[24, 2, 2], // Latency = 20, Repeat rate = 16
[FPR_Bypass, FPR_Bypass, FPR_Bypass]>,
InstrItinData<FPFused , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<1, [FPU_0]>],
[11, 2, 2, 2], // Latency = 7, Repeat rate = 1
[FPR_Bypass, FPR_Bypass, FPR_Bypass, FPR_Bypass]>,
InstrItinData<FPRes , [InstrStage<1, [DIS0, DIS1], 0>,
InstrStage<2, [FPU_0]>],
[12, 2], // Latency = 8, Repeat rate = 2
[FPR_Bypass, FPR_Bypass]>
]>;
// ===---------------------------------------------------------------------===//
// e5500 machine model for scheduling and other instruction cost heuristics.
def PPCE5500Model : SchedMachineModel {
let IssueWidth = 2; // 2 micro-ops are dispatched per cycle.
let MinLatency = -1; // OperandCycles are interpreted as MinLatency.
let LoadLatency = 6; // Optimistic load latency assuming bypass.
// This is overriden by OperandCycles if the
// Itineraries are queried instead.
let Itineraries = PPCE5500Itineraries;
}