1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 03:02:36 +01:00
llvm-mirror/lib/Target/ARM/ARMSchedule.td
Evgeny Leviant 900cb4bb83 Add support for branch forms of ALU instructions to Cortex-A57 model
Patch fixes scheduling of ALU instructions which modify pc register. Patch
also fixes computation of mutually exclusive predicates for sequences of
variants to be properly expanded

Differential revision: https://reviews.llvm.org/D91266
2020-11-24 11:43:51 +03:00

470 lines
16 KiB
TableGen

//===-- ARMSchedule.td - ARM Scheduling Definitions --------*- 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
//
//===----------------------------------------------------------------------===//
//===----------------------------------------------------------------------===//
// Instruction scheduling annotations for in-order and out-of-order CPUs.
// These annotations are independent of the itinerary class defined below.
// Here we define the subtarget independent read/write per-operand resources.
// The subtarget schedule definitions will then map these to the subtarget's
// resource usages.
// For example:
// The instruction cycle timings table might contain an entry for an operation
// like the following:
// Rd <- ADD Rn, Rm, <shift> Rs
// Uops | Latency from register | Uops - resource requirements - latency
// 2 | Rn: 1 Rm: 4 Rs: 4 | uop T0, Rm, Rs - P01 - 3
// | | uopc Rd, Rn, T0 - P01 - 1
// This is telling us that the result will be available in destination register
// Rd after a minimum of three cycles after the result in Rm and Rs is available
// and one cycle after the result in Rn is available. The micro-ops can execute
// on resource P01.
// To model this, we need to express that we need to dispatch two micro-ops,
// that the resource P01 is needed and that the latency to Rn is different than
// the latency to Rm and Rs. The scheduler can decrease Rn's producer latency by
// two.
// We will do this by assigning (abstract) resources to register defs/uses.
// ARMSchedule.td:
// def WriteALUsr : SchedWrite;
// def ReadAdvanceALUsr : ScheRead;
//
// ARMInstrInfo.td:
// def ADDrs : I<>, Sched<[WriteALUsr, ReadAdvanceALUsr, ReadDefault,
// ReadDefault]> { ...}
// ReadAdvance read resources allow us to define "pipeline by-passes" or
// shorter latencies to certain registers as needed in the example above.
// The "ReadDefault" can be omitted.
// Next, the subtarget td file assigns resources to the abstract resources
// defined here.
// ARMScheduleSubtarget.td:
// // Resources.
// def P01 : ProcResource<3>; // ALU unit (3 of it).
// ...
// // Resource usages.
// def : WriteRes<WriteALUsr, [P01, P01]> {
// Latency = 4; // Latency of 4.
// NumMicroOps = 2; // Dispatch 2 micro-ops.
// // The two instances of resource P01 are occupied for one cycle. It is one
// // cycle because these resources happen to be pipelined.
// ResourceCycles = [1, 1];
// }
// def : ReadAdvance<ReadAdvanceALUsr, 3>;
//===----------------------------------------------------------------------===//
// Sched definitions for integer pipeline instructions
//
// Basic ALU operation.
def WriteALU : SchedWrite;
def ReadALU : SchedRead;
// Basic ALU with shifts.
def WriteALUsi : SchedWrite; // Shift by immediate.
def WriteALUsr : SchedWrite; // Shift by register.
def WriteALUSsr : SchedWrite; // Shift by register (flag setting).
def ReadALUsr : SchedRead; // Some operands are read later.
// Compares.
def WriteCMP : SchedWrite;
def WriteCMPsi : SchedWrite;
def WriteCMPsr : SchedWrite;
// Multiplys.
def WriteMUL16 : SchedWrite; // 16-bit multiply.
def WriteMUL32 : SchedWrite; // 32-bit multiply.
def WriteMUL64Lo : SchedWrite; // 64-bit result. Low reg.
def WriteMUL64Hi : SchedWrite; // 64-bit result. High reg.
def ReadMUL : SchedRead;
// Multiply-accumulates.
def WriteMAC16 : SchedWrite; // 16-bit mac.
def WriteMAC32 : SchedWrite; // 32-bit mac.
def WriteMAC64Lo : SchedWrite; // 64-bit mac. Low reg.
def WriteMAC64Hi : SchedWrite; // 64-bit mac. High reg.
def ReadMAC : SchedRead;
// Divisions.
def WriteDIV : SchedWrite;
// Loads/Stores.
def WriteLd : SchedWrite;
def WritePreLd : SchedWrite;
def WriteST : SchedWrite;
// Branches.
def WriteBr : SchedWrite;
def WriteBrL : SchedWrite;
def WriteBrTbl : SchedWrite;
// Noop.
def WriteNoop : SchedWrite;
//===----------------------------------------------------------------------===//
// Sched definitions for floating-point and neon instructions
//
// Floating point conversions
def WriteFPCVT : SchedWrite;
def WriteFPMOV : SchedWrite; // FP -> GPR and vice-versa
// ALU operations (32/64-bit)
def WriteFPALU32 : SchedWrite;
def WriteFPALU64 : SchedWrite;
// Multiplication
def WriteFPMUL32 : SchedWrite;
def WriteFPMUL64 : SchedWrite;
def ReadFPMUL : SchedRead; // multiplier read
def ReadFPMAC : SchedRead; // accumulator read
// Multiply-accumulate
def WriteFPMAC32 : SchedWrite;
def WriteFPMAC64 : SchedWrite;
// Division
def WriteFPDIV32 : SchedWrite;
def WriteFPDIV64 : SchedWrite;
// Square-root
def WriteFPSQRT32 : SchedWrite;
def WriteFPSQRT64 : SchedWrite;
// Vector load and stores
def WriteVLD1 : SchedWrite;
def WriteVLD2 : SchedWrite;
def WriteVLD3 : SchedWrite;
def WriteVLD4 : SchedWrite;
def WriteVST1 : SchedWrite;
def WriteVST2 : SchedWrite;
def WriteVST3 : SchedWrite;
def WriteVST4 : SchedWrite;
// Define TII for use in SchedVariant Predicates.
def : PredicateProlog<[{
const ARMBaseInstrInfo *TII =
static_cast<const ARMBaseInstrInfo*>(SchedModel->getInstrInfo());
(void)TII;
const ARMSubtarget *STI =
static_cast<const ARMSubtarget*>(SchedModel->getSubtargetInfo());
(void)STI;
}]>;
def IsPredicated : CheckFunctionPredicateWithTII<
"ARM_MC::isPredicated",
"isPredicated"
>;
def IsPredicatedPred : MCSchedPredicate<IsPredicated>;
def IsCPSRDefined : CheckFunctionPredicateWithTII<
"ARM_MC::isCPSRDefined",
"ARMBaseInstrInfo::isCPSRDefined"
>;
def IsCPSRDefinedPred : MCSchedPredicate<IsCPSRDefined>;
let FunctionMapper = "ARM_AM::getAM2ShiftOpc" in {
class CheckAM2NoShift<int n> : CheckImmOperand_s<n, "ARM_AM::no_shift">;
class CheckAM2ShiftLSL<int n> : CheckImmOperand_s<n, "ARM_AM::lsl">;
}
let FunctionMapper = "ARM_AM::getAM2Op" in {
class CheckAM2OpAdd<int n> : CheckImmOperand_s<n, "ARM_AM::add"> {}
class CheckAM2OpSub<int n> : CheckImmOperand_s<n, "ARM_AM::sub"> {}
}
let FunctionMapper = "ARM_AM::getAM2Offset" in {
class CheckAM2Offset<int n, int of> : CheckImmOperand<n, of> {}
}
def IsLDMBaseRegInList : CheckFunctionPredicate<
"ARM_MC::isLDMBaseRegInList", "ARM_MC::isLDMBaseRegInList"
>;
let FunctionMapper = "ARM_AM::getAM3Op" in {
class CheckAM3OpSub<int n> : CheckImmOperand_s<n, "ARM_AM::sub"> {}
}
// LDM, base reg in list
def IsLDMBaseRegInListPred : MCSchedPredicate<IsLDMBaseRegInList>;
class IsRegPCPred<int n> : MCSchedPredicate<CheckRegOperand<n, PC>>;
class BranchWriteRes<int lat, int uops, list<ProcResourceKind> resl,
list<int> rcl, SchedWriteRes wr> :
SchedWriteRes<!listconcat(wr.ProcResources, resl)> {
let Latency = !add(wr.Latency, lat);
let ResourceCycles = !listconcat(wr.ResourceCycles, rcl);
let NumMicroOps = !add(wr.NumMicroOps, uops);
SchedWriteRes BaseWr = wr;
}
class CheckBranchForm<int n, BranchWriteRes br> :
SchedWriteVariant<[
SchedVar<IsRegPCPred<n>, [br]>,
SchedVar<NoSchedPred, [br.BaseWr]>
]>;
//===----------------------------------------------------------------------===//
// Instruction Itinerary classes used for ARM
//
def IIC_iALUx : InstrItinClass;
def IIC_iALUi : InstrItinClass;
def IIC_iALUr : InstrItinClass;
def IIC_iALUsi : InstrItinClass;
def IIC_iALUsir : InstrItinClass;
def IIC_iALUsr : InstrItinClass;
def IIC_iBITi : InstrItinClass;
def IIC_iBITr : InstrItinClass;
def IIC_iBITsi : InstrItinClass;
def IIC_iBITsr : InstrItinClass;
def IIC_iUNAr : InstrItinClass;
def IIC_iUNAsi : InstrItinClass;
def IIC_iEXTr : InstrItinClass;
def IIC_iEXTAr : InstrItinClass;
def IIC_iEXTAsr : InstrItinClass;
def IIC_iCMPi : InstrItinClass;
def IIC_iCMPr : InstrItinClass;
def IIC_iCMPsi : InstrItinClass;
def IIC_iCMPsr : InstrItinClass;
def IIC_iTSTi : InstrItinClass;
def IIC_iTSTr : InstrItinClass;
def IIC_iTSTsi : InstrItinClass;
def IIC_iTSTsr : InstrItinClass;
def IIC_iMOVi : InstrItinClass;
def IIC_iMOVr : InstrItinClass;
def IIC_iMOVsi : InstrItinClass;
def IIC_iMOVsr : InstrItinClass;
def IIC_iMOVix2 : InstrItinClass;
def IIC_iMOVix2addpc : InstrItinClass;
def IIC_iMOVix2ld : InstrItinClass;
def IIC_iMVNi : InstrItinClass;
def IIC_iMVNr : InstrItinClass;
def IIC_iMVNsi : InstrItinClass;
def IIC_iMVNsr : InstrItinClass;
def IIC_iCMOVi : InstrItinClass;
def IIC_iCMOVr : InstrItinClass;
def IIC_iCMOVsi : InstrItinClass;
def IIC_iCMOVsr : InstrItinClass;
def IIC_iCMOVix2 : InstrItinClass;
def IIC_iMUL16 : InstrItinClass;
def IIC_iMAC16 : InstrItinClass;
def IIC_iMUL32 : InstrItinClass;
def IIC_iMAC32 : InstrItinClass;
def IIC_iMUL64 : InstrItinClass;
def IIC_iMAC64 : InstrItinClass;
def IIC_iDIV : InstrItinClass;
def IIC_iLoad_i : InstrItinClass;
def IIC_iLoad_r : InstrItinClass;
def IIC_iLoad_si : InstrItinClass;
def IIC_iLoad_iu : InstrItinClass;
def IIC_iLoad_ru : InstrItinClass;
def IIC_iLoad_siu : InstrItinClass;
def IIC_iLoad_bh_i : InstrItinClass;
def IIC_iLoad_bh_r : InstrItinClass;
def IIC_iLoad_bh_si : InstrItinClass;
def IIC_iLoad_bh_iu : InstrItinClass;
def IIC_iLoad_bh_ru : InstrItinClass;
def IIC_iLoad_bh_siu : InstrItinClass;
def IIC_iLoad_d_i : InstrItinClass;
def IIC_iLoad_d_r : InstrItinClass;
def IIC_iLoad_d_ru : InstrItinClass;
def IIC_iLoad_m : InstrItinClass;
def IIC_iLoad_mu : InstrItinClass;
def IIC_iLoad_mBr : InstrItinClass;
def IIC_iPop : InstrItinClass;
def IIC_iPop_Br : InstrItinClass;
def IIC_iLoadiALU : InstrItinClass;
def IIC_iStore_i : InstrItinClass;
def IIC_iStore_r : InstrItinClass;
def IIC_iStore_si : InstrItinClass;
def IIC_iStore_iu : InstrItinClass;
def IIC_iStore_ru : InstrItinClass;
def IIC_iStore_siu : InstrItinClass;
def IIC_iStore_bh_i : InstrItinClass;
def IIC_iStore_bh_r : InstrItinClass;
def IIC_iStore_bh_si : InstrItinClass;
def IIC_iStore_bh_iu : InstrItinClass;
def IIC_iStore_bh_ru : InstrItinClass;
def IIC_iStore_bh_siu : InstrItinClass;
def IIC_iStore_d_i : InstrItinClass;
def IIC_iStore_d_r : InstrItinClass;
def IIC_iStore_d_ru : InstrItinClass;
def IIC_iStore_m : InstrItinClass;
def IIC_iStore_mu : InstrItinClass;
def IIC_Preload : InstrItinClass;
def IIC_Br : InstrItinClass;
def IIC_fpSTAT : InstrItinClass;
def IIC_fpUNA16 : InstrItinClass;
def IIC_fpUNA32 : InstrItinClass;
def IIC_fpUNA64 : InstrItinClass;
def IIC_fpCMP16 : InstrItinClass;
def IIC_fpCMP32 : InstrItinClass;
def IIC_fpCMP64 : InstrItinClass;
def IIC_fpCVTSD : InstrItinClass;
def IIC_fpCVTDS : InstrItinClass;
def IIC_fpCVTSH : InstrItinClass;
def IIC_fpCVTHS : InstrItinClass;
def IIC_fpCVTIH : InstrItinClass;
def IIC_fpCVTIS : InstrItinClass;
def IIC_fpCVTID : InstrItinClass;
def IIC_fpCVTHI : InstrItinClass;
def IIC_fpCVTSI : InstrItinClass;
def IIC_fpCVTDI : InstrItinClass;
def IIC_fpMOVIS : InstrItinClass;
def IIC_fpMOVID : InstrItinClass;
def IIC_fpMOVSI : InstrItinClass;
def IIC_fpMOVDI : InstrItinClass;
def IIC_fpALU16 : InstrItinClass;
def IIC_fpALU32 : InstrItinClass;
def IIC_fpALU64 : InstrItinClass;
def IIC_fpMUL16 : InstrItinClass;
def IIC_fpMUL32 : InstrItinClass;
def IIC_fpMUL64 : InstrItinClass;
def IIC_fpMAC16 : InstrItinClass;
def IIC_fpMAC32 : InstrItinClass;
def IIC_fpMAC64 : InstrItinClass;
def IIC_fpFMAC16 : InstrItinClass;
def IIC_fpFMAC32 : InstrItinClass;
def IIC_fpFMAC64 : InstrItinClass;
def IIC_fpDIV16 : InstrItinClass;
def IIC_fpDIV32 : InstrItinClass;
def IIC_fpDIV64 : InstrItinClass;
def IIC_fpSQRT16 : InstrItinClass;
def IIC_fpSQRT32 : InstrItinClass;
def IIC_fpSQRT64 : InstrItinClass;
def IIC_fpLoad16 : InstrItinClass;
def IIC_fpLoad32 : InstrItinClass;
def IIC_fpLoad64 : InstrItinClass;
def IIC_fpLoad_m : InstrItinClass;
def IIC_fpLoad_mu : InstrItinClass;
def IIC_fpStore16 : InstrItinClass;
def IIC_fpStore32 : InstrItinClass;
def IIC_fpStore64 : InstrItinClass;
def IIC_fpStore_m : InstrItinClass;
def IIC_fpStore_mu : InstrItinClass;
def IIC_VLD1 : InstrItinClass;
def IIC_VLD1x2 : InstrItinClass;
def IIC_VLD1x3 : InstrItinClass;
def IIC_VLD1x4 : InstrItinClass;
def IIC_VLD1u : InstrItinClass;
def IIC_VLD1x2u : InstrItinClass;
def IIC_VLD1x3u : InstrItinClass;
def IIC_VLD1x4u : InstrItinClass;
def IIC_VLD1ln : InstrItinClass;
def IIC_VLD1lnu : InstrItinClass;
def IIC_VLD1dup : InstrItinClass;
def IIC_VLD1dupu : InstrItinClass;
def IIC_VLD2 : InstrItinClass;
def IIC_VLD2x2 : InstrItinClass;
def IIC_VLD2u : InstrItinClass;
def IIC_VLD2x2u : InstrItinClass;
def IIC_VLD2ln : InstrItinClass;
def IIC_VLD2lnu : InstrItinClass;
def IIC_VLD2dup : InstrItinClass;
def IIC_VLD2dupu : InstrItinClass;
def IIC_VLD3 : InstrItinClass;
def IIC_VLD3ln : InstrItinClass;
def IIC_VLD3u : InstrItinClass;
def IIC_VLD3lnu : InstrItinClass;
def IIC_VLD3dup : InstrItinClass;
def IIC_VLD3dupu : InstrItinClass;
def IIC_VLD4 : InstrItinClass;
def IIC_VLD4ln : InstrItinClass;
def IIC_VLD4u : InstrItinClass;
def IIC_VLD4lnu : InstrItinClass;
def IIC_VLD4dup : InstrItinClass;
def IIC_VLD4dupu : InstrItinClass;
def IIC_VST1 : InstrItinClass;
def IIC_VST1x2 : InstrItinClass;
def IIC_VST1x3 : InstrItinClass;
def IIC_VST1x4 : InstrItinClass;
def IIC_VST1u : InstrItinClass;
def IIC_VST1x2u : InstrItinClass;
def IIC_VST1x3u : InstrItinClass;
def IIC_VST1x4u : InstrItinClass;
def IIC_VST1ln : InstrItinClass;
def IIC_VST1lnu : InstrItinClass;
def IIC_VST2 : InstrItinClass;
def IIC_VST2x2 : InstrItinClass;
def IIC_VST2u : InstrItinClass;
def IIC_VST2x2u : InstrItinClass;
def IIC_VST2ln : InstrItinClass;
def IIC_VST2lnu : InstrItinClass;
def IIC_VST3 : InstrItinClass;
def IIC_VST3u : InstrItinClass;
def IIC_VST3ln : InstrItinClass;
def IIC_VST3lnu : InstrItinClass;
def IIC_VST4 : InstrItinClass;
def IIC_VST4u : InstrItinClass;
def IIC_VST4ln : InstrItinClass;
def IIC_VST4lnu : InstrItinClass;
def IIC_VUNAD : InstrItinClass;
def IIC_VUNAQ : InstrItinClass;
def IIC_VBIND : InstrItinClass;
def IIC_VBINQ : InstrItinClass;
def IIC_VPBIND : InstrItinClass;
def IIC_VFMULD : InstrItinClass;
def IIC_VFMULQ : InstrItinClass;
def IIC_VMOV : InstrItinClass;
def IIC_VMOVImm : InstrItinClass;
def IIC_VMOVD : InstrItinClass;
def IIC_VMOVQ : InstrItinClass;
def IIC_VMOVIS : InstrItinClass;
def IIC_VMOVID : InstrItinClass;
def IIC_VMOVISL : InstrItinClass;
def IIC_VMOVSI : InstrItinClass;
def IIC_VMOVDI : InstrItinClass;
def IIC_VMOVN : InstrItinClass;
def IIC_VPERMD : InstrItinClass;
def IIC_VPERMQ : InstrItinClass;
def IIC_VPERMQ3 : InstrItinClass;
def IIC_VMACD : InstrItinClass;
def IIC_VMACQ : InstrItinClass;
def IIC_VFMACD : InstrItinClass;
def IIC_VFMACQ : InstrItinClass;
def IIC_VRECSD : InstrItinClass;
def IIC_VRECSQ : InstrItinClass;
def IIC_VCNTiD : InstrItinClass;
def IIC_VCNTiQ : InstrItinClass;
def IIC_VUNAiD : InstrItinClass;
def IIC_VUNAiQ : InstrItinClass;
def IIC_VQUNAiD : InstrItinClass;
def IIC_VQUNAiQ : InstrItinClass;
def IIC_VBINiD : InstrItinClass;
def IIC_VBINiQ : InstrItinClass;
def IIC_VSUBiD : InstrItinClass;
def IIC_VSUBiQ : InstrItinClass;
def IIC_VBINi4D : InstrItinClass;
def IIC_VBINi4Q : InstrItinClass;
def IIC_VSUBi4D : InstrItinClass;
def IIC_VSUBi4Q : InstrItinClass;
def IIC_VABAD : InstrItinClass;
def IIC_VABAQ : InstrItinClass;
def IIC_VSHLiD : InstrItinClass;
def IIC_VSHLiQ : InstrItinClass;
def IIC_VSHLi4D : InstrItinClass;
def IIC_VSHLi4Q : InstrItinClass;
def IIC_VPALiD : InstrItinClass;
def IIC_VPALiQ : InstrItinClass;
def IIC_VMULi16D : InstrItinClass;
def IIC_VMULi32D : InstrItinClass;
def IIC_VMULi16Q : InstrItinClass;
def IIC_VMULi32Q : InstrItinClass;
def IIC_VMACi16D : InstrItinClass;
def IIC_VMACi32D : InstrItinClass;
def IIC_VMACi16Q : InstrItinClass;
def IIC_VMACi32Q : InstrItinClass;
def IIC_VEXTD : InstrItinClass;
def IIC_VEXTQ : InstrItinClass;
def IIC_VTB1 : InstrItinClass;
def IIC_VTB2 : InstrItinClass;
def IIC_VTB3 : InstrItinClass;
def IIC_VTB4 : InstrItinClass;
def IIC_VTBX1 : InstrItinClass;
def IIC_VTBX2 : InstrItinClass;
def IIC_VTBX3 : InstrItinClass;
def IIC_VTBX4 : InstrItinClass;
def IIC_VDOTPROD : InstrItinClass;