mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-09-01 04:29:57 +00:00
d24ba9ff6e
As Andy pointed out to me a long time ago, there are no structural hazards in the later pipeline stages of the A2, and so modeling them is useless. Also, modeling the top pre-dispatch stages is deceiving because, when multiple hardware threads are active, those resources are shared among the threads. The bypass definitions were mostly wrong, and so those have been removed. The resulting itinerary is much simpler, and more accurate. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@190562 91177308-0d34-0410-b5e6-96231b3b80d8
156 lines
6.6 KiB
TableGen
156 lines
6.6 KiB
TableGen
//===- PPCScheduleA2.td - PPC A2 Scheduling Definitions --*- tablegen -*-===//
|
|
//
|
|
// The LLVM Compiler Infrastructure
|
|
//
|
|
// This file is distributed under the University of Illinois Open Source
|
|
// License. See LICENSE.TXT for details.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// Primary reference:
|
|
// A2 Processor User's Manual.
|
|
// IBM (as updated in) 2010.
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Functional units on the PowerPC A2 chip sets
|
|
//
|
|
def XU : FuncUnit; // XU pipeline
|
|
def FU : FuncUnit; // FI pipeline
|
|
|
|
//
|
|
// This file defines the itinerary class data for the PPC A2 processor.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
def PPCA2Itineraries : ProcessorItineraries<
|
|
[XU, FU], [], [
|
|
InstrItinData<IntSimple , [InstrStage<1, [XU]>],
|
|
[1, 1, 1]>,
|
|
InstrItinData<IntGeneral , [InstrStage<1, [XU]>],
|
|
[2, 1, 1]>,
|
|
InstrItinData<IntCompare , [InstrStage<1, [XU]>],
|
|
[2, 1, 1]>,
|
|
InstrItinData<IntDivW , [InstrStage<1, [XU]>],
|
|
[39, 1, 1]>,
|
|
InstrItinData<IntDivD , [InstrStage<1, [XU]>],
|
|
[71, 1, 1]>,
|
|
InstrItinData<IntMulHW , [InstrStage<1, [XU]>],
|
|
[5, 1, 1]>,
|
|
InstrItinData<IntMulHWU , [InstrStage<1, [XU]>],
|
|
[5, 1, 1]>,
|
|
InstrItinData<IntMulLI , [InstrStage<1, [XU]>],
|
|
[6, 1, 1]>,
|
|
InstrItinData<IntRotate , [InstrStage<1, [XU]>],
|
|
[2, 1, 1]>,
|
|
InstrItinData<IntRotateD , [InstrStage<1, [XU]>],
|
|
[2, 1, 1]>,
|
|
InstrItinData<IntRotateDI , [InstrStage<1, [XU]>],
|
|
[2, 1, 1]>,
|
|
InstrItinData<IntShift , [InstrStage<1, [XU]>],
|
|
[2, 1, 1]>,
|
|
InstrItinData<IntTrapW , [InstrStage<1, [XU]>],
|
|
[2, 1]>,
|
|
InstrItinData<IntTrapD , [InstrStage<1, [XU]>],
|
|
[2, 1]>,
|
|
InstrItinData<BrB , [InstrStage<1, [XU]>],
|
|
[6, 1, 1]>,
|
|
InstrItinData<BrCR , [InstrStage<1, [XU]>],
|
|
[1, 1, 1]>,
|
|
InstrItinData<BrMCR , [InstrStage<1, [XU]>],
|
|
[5, 1, 1]>,
|
|
InstrItinData<BrMCRX , [InstrStage<1, [XU]>],
|
|
[1, 1, 1]>,
|
|
InstrItinData<LdStDCBA , [InstrStage<1, [XU]>],
|
|
[1, 1, 1]>,
|
|
InstrItinData<LdStDCBF , [InstrStage<1, [XU]>],
|
|
[1, 1, 1]>,
|
|
InstrItinData<LdStDCBI , [InstrStage<1, [XU]>],
|
|
[1, 1, 1]>,
|
|
InstrItinData<LdStLoad , [InstrStage<1, [XU]>],
|
|
[6, 1, 1]>,
|
|
InstrItinData<LdStLoadUpd , [InstrStage<1, [XU]>],
|
|
[6, 2, 1, 1]>,
|
|
InstrItinData<LdStLDU , [InstrStage<1, [XU]>],
|
|
[6, 1, 1]>,
|
|
InstrItinData<LdStStore , [InstrStage<1, [XU]>],
|
|
[1, 1, 1]>,
|
|
InstrItinData<LdStStoreUpd, [InstrStage<1, [XU]>],
|
|
[2, 1, 1, 1]>,
|
|
InstrItinData<LdStICBI, [InstrStage<1, [XU]>],
|
|
[16, 1, 1]>,
|
|
InstrItinData<LdStSTFD , [InstrStage<1, [XU]>],
|
|
[1, 1, 1]>,
|
|
InstrItinData<LdStSTFDU , [InstrStage<1, [XU]>],
|
|
[2, 1, 1, 1]>,
|
|
InstrItinData<LdStLFD , [InstrStage<1, [XU]>],
|
|
[7, 1, 1]>,
|
|
InstrItinData<LdStLFDU , [InstrStage<1, [XU]>],
|
|
[7, 2, 1, 1]>,
|
|
InstrItinData<LdStLHA , [InstrStage<1, [XU]>],
|
|
[6, 1, 1]>,
|
|
InstrItinData<LdStLHAU , [InstrStage<1, [XU]>],
|
|
[6, 2, 1, 1]>,
|
|
InstrItinData<LdStLWARX , [InstrStage<1, [XU]>],
|
|
[82, 1, 1]>, // L2 latency
|
|
InstrItinData<LdStSTD , [InstrStage<1, [XU]>],
|
|
[1, 1, 1]>,
|
|
InstrItinData<LdStSTDU , [InstrStage<1, [XU]>],
|
|
[2, 1, 1, 1]>,
|
|
InstrItinData<LdStSTDCX , [InstrStage<1, [XU]>],
|
|
[82, 1, 1]>, // L2 latency
|
|
InstrItinData<LdStSTWCX , [InstrStage<1, [XU]>],
|
|
[82, 1, 1]>, // L2 latency
|
|
InstrItinData<LdStSync , [InstrStage<1, [XU]>],
|
|
[6]>,
|
|
InstrItinData<SprISYNC , [InstrStage<1, [XU]>],
|
|
[16]>,
|
|
InstrItinData<SprMTMSR , [InstrStage<1, [XU]>],
|
|
[16, 1]>,
|
|
InstrItinData<SprMFCR , [InstrStage<1, [XU]>],
|
|
[6, 1]>,
|
|
InstrItinData<SprMFMSR , [InstrStage<1, [XU]>],
|
|
[4, 1]>,
|
|
InstrItinData<SprMFSPR , [InstrStage<1, [XU]>],
|
|
[6, 1]>,
|
|
InstrItinData<SprMFTB , [InstrStage<1, [XU]>],
|
|
[4, 1]>,
|
|
InstrItinData<SprMTSPR , [InstrStage<1, [XU]>],
|
|
[6, 1]>,
|
|
InstrItinData<SprRFI , [InstrStage<1, [XU]>],
|
|
[16]>,
|
|
InstrItinData<SprSC , [InstrStage<1, [XU]>],
|
|
[16]>,
|
|
InstrItinData<FPGeneral , [InstrStage<1, [FU]>],
|
|
[6, 1, 1]>,
|
|
InstrItinData<FPAddSub , [InstrStage<1, [FU]>],
|
|
[6, 1, 1]>,
|
|
InstrItinData<FPCompare , [InstrStage<1, [FU]>],
|
|
[5, 1, 1]>,
|
|
InstrItinData<FPDivD , [InstrStage<1, [FU]>],
|
|
[72, 1, 1]>,
|
|
InstrItinData<FPDivS , [InstrStage<1, [FU]>],
|
|
[59, 1, 1]>,
|
|
InstrItinData<FPSqrt , [InstrStage<1, [FU]>],
|
|
[69, 1, 1]>,
|
|
InstrItinData<FPFused , [InstrStage<1, [FU]>],
|
|
[6, 1, 1, 1]>,
|
|
InstrItinData<FPRes , [InstrStage<1, [FU]>],
|
|
[6, 1]>
|
|
]>;
|
|
|
|
// ===---------------------------------------------------------------------===//
|
|
// A2 machine model for scheduling and other instruction cost heuristics.
|
|
|
|
def PPCA2Model : SchedMachineModel {
|
|
let IssueWidth = 1; // 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 MispredictPenalty = 13;
|
|
|
|
let Itineraries = PPCA2Itineraries;
|
|
}
|
|
|