llvm/lib/Target/PowerPC/PPCScheduleE500mc.td
Hal Finkel bc0bdb26da Split some PPC itinerary classes
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
2013-11-30 20:41:13 +00:00

315 lines
19 KiB
TableGen

//===-- PPCScheduleE500mc.td - e500mc 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 e500mc 32-bit
// Power processor.
//
// All information is derived from the "e500mc Core Reference Manual",
// Freescale Document Number E500MCRM, Rev. 1, 03/2012.
//
//===----------------------------------------------------------------------===//
// Relevant functional units in the Freescale e500mc core:
//
// * 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 E500_DIS0 : FuncUnit; // Dispatch stage - insn 1
def E500_DIS1 : FuncUnit; // Dispatch stage - insn 2
// * Execute
// 6 pipelined execution units: SFX0, SFX1, BU, FPU, LSU, CFX.
// Some instructions can only execute in SFX0 but not SFX1.
// The CFX has a bypass path, allowing non-divide instructions to execute
// while a divide instruction is executed.
def E500_SFX0 : FuncUnit; // Simple unit 0
def E500_SFX1 : FuncUnit; // Simple unit 1
def E500_BU : FuncUnit; // Branch unit
def E500_CFX_DivBypass
: FuncUnit; // CFX divide bypass path
def E500_CFX_0 : FuncUnit; // CFX pipeline
def E500_LSU_0 : FuncUnit; // LSU pipeline
def E500_FPU_0 : FuncUnit; // FPU pipeline
def E500_GPR_Bypass : Bypass;
def E500_FPR_Bypass : Bypass;
def E500_CR_Bypass : Bypass;
def PPCE500mcItineraries : ProcessorItineraries<
[E500_DIS0, E500_DIS1, E500_SFX0, E500_SFX1, E500_BU, E500_CFX_DivBypass,
E500_CFX_0, E500_LSU_0, E500_FPU_0],
[E500_CR_Bypass, E500_GPR_Bypass, E500_FPR_Bypass], [
InstrItinData<IIC_IntSimple, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1]>],
[4, 1, 1], // Latency = 1
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_IntGeneral, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1]>],
[4, 1, 1], // Latency = 1
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_IntCompare, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1]>],
[5, 1, 1], // Latency = 1 or 2
[E500_CR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_IntDivW, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_CFX_0], 0>,
InstrStage<14, [E500_CFX_DivBypass]>],
[17, 1, 1], // Latency=4..35, Repeat= 4..35
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_IntMFFS, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<8, [E500_FPU_0]>],
[11], // Latency = 8
[E500_FPR_Bypass]>,
InstrItinData<IIC_IntMTFSB0, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<8, [E500_FPU_0]>],
[11, 1, 1], // Latency = 8
[NoBypass, NoBypass, NoBypass]>,
InstrItinData<IIC_IntMulHW, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_CFX_0]>],
[7, 1, 1], // Latency = 4, Repeat rate = 1
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_IntMulHWU, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_CFX_0]>],
[7, 1, 1], // Latency = 4, Repeat rate = 1
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_IntMulLI, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_CFX_0]>],
[7, 1, 1], // Latency = 4, Repeat rate = 1
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_IntRotate, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1]>],
[4, 1, 1], // Latency = 1
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_IntShift, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1]>],
[4, 1, 1], // Latency = 1
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_IntTrapW, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<2, [E500_SFX0]>],
[5, 1], // Latency = 2, Repeat rate = 2
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_BrB, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_BU]>],
[4, 1], // Latency = 1
[NoBypass, E500_GPR_Bypass]>,
InstrItinData<IIC_BrCR, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_BU]>],
[4, 1, 1], // Latency = 1
[E500_CR_Bypass,
E500_CR_Bypass, E500_CR_Bypass]>,
InstrItinData<IIC_BrMCR, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_BU]>],
[4, 1], // Latency = 1
[E500_CR_Bypass, E500_CR_Bypass]>,
InstrItinData<IIC_BrMCRX, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1]>],
[4, 1, 1], // Latency = 1
[E500_CR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStDCBA, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3, Repeat rate = 1
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStDCBF, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStDCBI, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStLoad, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStLoadUpd, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[E500_GPR_Bypass, E500_GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<IIC_LdStLoadUpdX,[InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[E500_GPR_Bypass, E500_GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<IIC_LdStStore, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[NoBypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStStoreUpd,[InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[NoBypass, E500_GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<IIC_LdStICBI, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[NoBypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStSTFD, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1, 1], // Latency = 3
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStSTFDU, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1, 1], // Latency = 3
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<IIC_LdStLFD, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[7, 1, 1], // Latency = 4
[E500_FPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStLFDU, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1], 0>,
InstrStage<1, [E500_LSU_0]>],
[7, 1, 1], // Latency = 4
[E500_FPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<IIC_LdStLFDUX, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1], 0>,
InstrStage<1, [E500_LSU_0]>],
[7, 1, 1], // Latency = 4
[E500_FPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass],
2>, // 2 micro-ops
InstrItinData<IIC_LdStLHA, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStLHAU, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStLHAUX, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStLMW, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[7, 1], // Latency = r+3
[NoBypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStLWARX, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<3, [E500_LSU_0]>],
[6, 1, 1], // Latency = 3, Repeat rate = 3
[E500_GPR_Bypass,
E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStSTWCX, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>],
[6, 1], // Latency = 3
[NoBypass, E500_GPR_Bypass]>,
InstrItinData<IIC_LdStSync, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0]>]>,
InstrItinData<IIC_SprMFSR, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<4, [E500_SFX0]>],
[7, 1],
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_SprMTMSR, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<2, [E500_SFX0, E500_SFX1]>],
[5, 1], // Latency = 2, Repeat rate = 4
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_SprMTSR, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0]>],
[5, 1],
[NoBypass, E500_GPR_Bypass]>,
InstrItinData<IIC_SprTLBSYNC, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_LSU_0], 0>]>,
InstrItinData<IIC_SprMFCR, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<5, [E500_SFX0]>],
[8, 1],
[E500_GPR_Bypass, E500_CR_Bypass]>,
InstrItinData<IIC_SprMFCRF, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<5, [E500_SFX0]>],
[8, 1],
[E500_GPR_Bypass, E500_CR_Bypass]>,
InstrItinData<IIC_SprMFMSR, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<4, [E500_SFX0]>],
[7, 1], // Latency = 4, Repeat rate = 4
[E500_GPR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_SprMFSPR, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1]>],
[4, 1], // Latency = 1, Repeat rate = 1
[E500_GPR_Bypass, E500_CR_Bypass]>,
InstrItinData<IIC_SprMFTB, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<4, [E500_SFX0]>],
[7, 1], // Latency = 4, Repeat rate = 4
[NoBypass, E500_GPR_Bypass]>,
InstrItinData<IIC_SprMTSPR, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0, E500_SFX1]>],
[4, 1], // Latency = 1, Repeat rate = 1
[E500_CR_Bypass, E500_GPR_Bypass]>,
InstrItinData<IIC_SprMTSRIN, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<1, [E500_SFX0]>],
[4, 1],
[NoBypass, E500_GPR_Bypass]>,
InstrItinData<IIC_FPGeneral, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<2, [E500_FPU_0]>],
[11, 1, 1], // Latency = 8, Repeat rate = 2
[E500_FPR_Bypass,
E500_FPR_Bypass, E500_FPR_Bypass]>,
InstrItinData<IIC_FPAddSub, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<4, [E500_FPU_0]>],
[13, 1, 1], // Latency = 10, Repeat rate = 4
[E500_FPR_Bypass,
E500_FPR_Bypass, E500_FPR_Bypass]>,
InstrItinData<IIC_FPCompare, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<2, [E500_FPU_0]>],
[11, 1, 1], // Latency = 8, Repeat rate = 2
[E500_CR_Bypass,
E500_FPR_Bypass, E500_FPR_Bypass]>,
InstrItinData<IIC_FPDivD, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<68, [E500_FPU_0]>],
[71, 1, 1], // Latency = 68, Repeat rate = 68
[E500_FPR_Bypass,
E500_FPR_Bypass, E500_FPR_Bypass]>,
InstrItinData<IIC_FPDivS, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<38, [E500_FPU_0]>],
[41, 1, 1], // Latency = 38, Repeat rate = 38
[E500_FPR_Bypass,
E500_FPR_Bypass, E500_FPR_Bypass]>,
InstrItinData<IIC_FPFused, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<4, [E500_FPU_0]>],
[13, 1, 1, 1], // Latency = 10, Repeat rate = 4
[E500_FPR_Bypass,
E500_FPR_Bypass, E500_FPR_Bypass,
E500_FPR_Bypass]>,
InstrItinData<IIC_FPRes, [InstrStage<1, [E500_DIS0, E500_DIS1], 0>,
InstrStage<38, [E500_FPU_0]>],
[41, 1], // Latency = 38, Repeat rate = 38
[E500_FPR_Bypass, E500_FPR_Bypass]>
]>;
// ===---------------------------------------------------------------------===//
// e500mc machine model for scheduling and other instruction cost heuristics.
def PPCE500mcModel : SchedMachineModel {
let IssueWidth = 2; // 2 micro-ops are dispatched per cycle.
let MinLatency = -1; // OperandCycles are interpreted as MinLatency.
let LoadLatency = 5; // Optimistic load latency assuming bypass.
// This is overriden by OperandCycles if the
// Itineraries are queried instead.
let Itineraries = PPCE500mcItineraries;
}