mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-22 02:33:06 +01:00
e25a1a8f41
First patch in a series adding MC layer support for the Arm Scalable Matrix Extension. This patch adds the following features: sme, sme-i64, sme-f64 The sme-i64 and sme-f64 flags are for the optional I16I64 and F64F64 features. If a target supports I16I64 then the following instructions are implemented: * 64-bit integer ADDHA and ADDVA variants (D105570). * SMOPA, SMOPS, SUMOPA, SUMOPS, UMOPA, UMOPS, USMOPA, and USMOPS instructions that accumulate 16-bit integer outer products into 64-bit integer tiles. If a target supports F64F64 then the FMOPA and FMOPS instructions that accumulate double-precision floating-point outer products into double-precision tiles are implemented. Outer products are implemented in D105571. The reference can be found here: https://developer.arm.com/documentation/ddi0602/2021-06 Reviewed By: CarolineConcatto Differential Revision: https://reviews.llvm.org/D105569
1015 lines
51 KiB
TableGen
1015 lines
51 KiB
TableGen
//=- AArch64SchedExynosM5.td - Samsung Exynos M5 Sched Defs --*- 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
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
//
|
|
// This file defines the machine model for the Samsung Exynos M5 to support
|
|
// instruction scheduling and other instruction cost heuristics.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// The Exynos-M5 is an advanced superscalar microprocessor with a 6-wide
|
|
// in-order stage for decode and dispatch and a wider issue stage.
|
|
// The execution units and loads and stores are out-of-order.
|
|
|
|
def ExynosM5Model : SchedMachineModel {
|
|
let IssueWidth = 6; // Up to 6 uops per cycle.
|
|
let MicroOpBufferSize = 228; // ROB size.
|
|
let LoopMicroOpBufferSize = 60; // Based on the instruction queue size.
|
|
let LoadLatency = 4; // Optimistic load cases.
|
|
let MispredictPenalty = 15; // Minimum branch misprediction penalty.
|
|
let CompleteModel = 1; // Use the default model otherwise.
|
|
|
|
list<Predicate> UnsupportedFeatures = !listconcat(SVEUnsupported.F,
|
|
PAUnsupported.F,
|
|
SMEUnsupported.F);
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Define each kind of processor resource and number available on the Exynos-M5.
|
|
|
|
let SchedModel = ExynosM5Model in {
|
|
|
|
def M5UnitA : ProcResource<2>; // Simple integer
|
|
def M5UnitC : ProcResource<2>; // Simple and complex integer
|
|
let Super = M5UnitC, BufferSize = 1 in
|
|
def M5UnitD : ProcResource<1>; // Integer division (inside C0, serialized)
|
|
def M5UnitE : ProcResource<2>; // Simple 32-bit integer
|
|
let Super = M5UnitC in
|
|
def M5UnitF : ProcResource<2>; // CRC (inside C)
|
|
def M5UnitB : ProcResource<1>; // Branch
|
|
def M5UnitL0 : ProcResource<1>; // Load
|
|
def M5UnitS0 : ProcResource<1>; // Store
|
|
def M5PipeLS : ProcResource<1>; // Load/Store
|
|
let Super = M5PipeLS in {
|
|
def M5UnitL1 : ProcResource<1>;
|
|
def M5UnitS1 : ProcResource<1>;
|
|
}
|
|
def M5PipeF0 : ProcResource<1>; // FP #0
|
|
let Super = M5PipeF0 in {
|
|
def M5UnitFMAC0 : ProcResource<1>; // FP multiplication
|
|
def M5UnitFADD0 : ProcResource<1>; // Simple FP
|
|
def M5UnitNALU0 : ProcResource<1>; // Simple vector
|
|
def M5UnitNDOT0 : ProcResource<1>; // Dot product vector
|
|
def M5UnitNHAD : ProcResource<1>; // Horizontal vector
|
|
def M5UnitNMSC : ProcResource<1>; // FP and vector miscellanea
|
|
def M5UnitNMUL0 : ProcResource<1>; // Vector multiplication
|
|
def M5UnitNSHT0 : ProcResource<1>; // Vector shifting
|
|
def M5UnitNSHF0 : ProcResource<1>; // Vector shuffling
|
|
def M5UnitNCRY0 : ProcResource<1>; // Cryptographic
|
|
}
|
|
def M5PipeF1 : ProcResource<1>; // FP #1
|
|
let Super = M5PipeF1 in {
|
|
def M5UnitFMAC1 : ProcResource<1>; // FP multiplication
|
|
def M5UnitFADD1 : ProcResource<1>; // Simple FP
|
|
def M5UnitFCVT0 : ProcResource<1>; // FP conversion
|
|
def M5UnitFDIV0 : ProcResource<2>; // FP division (serialized)
|
|
def M5UnitFSQR0 : ProcResource<2>; // FP square root (serialized)
|
|
def M5UnitFST0 : ProcResource<1>; // FP store
|
|
def M5UnitNALU1 : ProcResource<1>; // Simple vector
|
|
def M5UnitNDOT1 : ProcResource<1>; // Dot product vector
|
|
def M5UnitNSHT1 : ProcResource<1>; // Vector shifting
|
|
def M5UnitNSHF1 : ProcResource<1>; // Vector shuffling
|
|
}
|
|
def M5PipeF2 : ProcResource<1>; // FP #2
|
|
let Super = M5PipeF2 in {
|
|
def M5UnitFMAC2 : ProcResource<1>; // FP multiplication
|
|
def M5UnitFADD2 : ProcResource<1>; // Simple FP
|
|
def M5UnitFCVT1 : ProcResource<1>; // FP conversion
|
|
def M5UnitFDIV1 : ProcResource<2>; // FP division (serialized)
|
|
def M5UnitFSQR1 : ProcResource<2>; // FP square root (serialized)
|
|
def M5UnitFST1 : ProcResource<1>; // FP store
|
|
def M5UnitNALU2 : ProcResource<1>; // Simple vector
|
|
def M5UnitNDOT2 : ProcResource<1>; // Dot product vector
|
|
def M5UnitNMUL1 : ProcResource<1>; // Vector multiplication
|
|
def M5UnitNSHT2 : ProcResource<1>; // Vector shifting
|
|
def M5UnitNCRY1 : ProcResource<1>; // Cryptographic
|
|
}
|
|
|
|
def M5UnitAX : ProcResGroup<[M5UnitA,
|
|
M5UnitC]>;
|
|
def M5UnitAW : ProcResGroup<[M5UnitA,
|
|
M5UnitC,
|
|
M5UnitE]>;
|
|
def M5UnitL : ProcResGroup<[M5UnitL0,
|
|
M5UnitL1]>;
|
|
def M5UnitS : ProcResGroup<[M5UnitS0,
|
|
M5UnitS1]>;
|
|
def M5UnitFMAC : ProcResGroup<[M5UnitFMAC0,
|
|
M5UnitFMAC1,
|
|
M5UnitFMAC2]>;
|
|
def M5UnitFADD : ProcResGroup<[M5UnitFADD0,
|
|
M5UnitFADD1,
|
|
M5UnitFADD2]>;
|
|
def M5UnitFCVT : ProcResGroup<[M5UnitFCVT0,
|
|
M5UnitFCVT1]>;
|
|
def M5UnitFDIV : ProcResGroup<[M5UnitFDIV0,
|
|
M5UnitFDIV1]>;
|
|
def M5UnitFSQR : ProcResGroup<[M5UnitFSQR0,
|
|
M5UnitFSQR1]>;
|
|
def M5UnitFST : ProcResGroup<[M5UnitFST0,
|
|
M5UnitFST1]>;
|
|
def M5UnitNALU : ProcResGroup<[M5UnitNALU0,
|
|
M5UnitNALU1,
|
|
M5UnitNALU2]>;
|
|
def M5UnitNDOT : ProcResGroup<[M5UnitNDOT0,
|
|
M5UnitNDOT1,
|
|
M5UnitNDOT2]>;
|
|
def M5UnitNMUL : ProcResGroup<[M5UnitNMUL0,
|
|
M5UnitNMUL1]>;
|
|
def M5UnitNSHT : ProcResGroup<[M5UnitNSHT0,
|
|
M5UnitNSHT1,
|
|
M5UnitNSHT2]>;
|
|
def M5UnitNSHF : ProcResGroup<[M5UnitNSHF0,
|
|
M5UnitNSHF1]>;
|
|
def M5UnitNCRY : ProcResGroup<[M5UnitNCRY0,
|
|
M5UnitNCRY1]>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Resources details.
|
|
|
|
def M5WriteZ0 : SchedWriteRes<[]> { let Latency = 0; }
|
|
def M5WriteZ1 : SchedWriteRes<[]> { let Latency = 1;
|
|
let NumMicroOps = 0; }
|
|
def M5WriteZ4 : SchedWriteRes<[]> { let Latency = 4;
|
|
let NumMicroOps = 0; }
|
|
|
|
def M5WriteA1W : SchedWriteRes<[M5UnitAW]> { let Latency = 1; }
|
|
def M5WriteA1X : SchedWriteRes<[M5UnitAX]> { let Latency = 1; }
|
|
def M5WriteAAW : SchedWriteRes<[M5UnitAW]> { let Latency = 2;
|
|
let ResourceCycles = [2]; }
|
|
def M5WriteAAX : SchedWriteRes<[M5UnitAX]> { let Latency = 2;
|
|
let ResourceCycles = [2]; }
|
|
def M5WriteAB : SchedWriteRes<[M5UnitAX,
|
|
M5UnitC,
|
|
M5UnitE]> { let Latency = 2;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteAC : SchedWriteRes<[M5UnitAX,
|
|
M5UnitAX,
|
|
M5UnitC]> { let Latency = 3;
|
|
let NumMicroOps = 3; }
|
|
def M5WriteAD : SchedWriteRes<[M5UnitAW,
|
|
M5UnitC]> { let Latency = 2;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteAFW : SchedWriteRes<[M5UnitAW]> { let Latency = 2;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteAFX : SchedWriteRes<[M5UnitAX]> { let Latency = 2;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteAUW : SchedWriteVariant<[SchedVar<IsCopyIdiomPred, [M5WriteZ0]>,
|
|
SchedVar<ExynosArithPred, [M5WriteA1W]>,
|
|
SchedVar<ExynosLogicExPred, [M5WriteA1W]>,
|
|
SchedVar<NoSchedPred, [M5WriteAAW]>]>;
|
|
def M5WriteAUX : SchedWriteVariant<[SchedVar<IsCopyIdiomPred, [M5WriteZ0]>,
|
|
SchedVar<ExynosArithPred, [M5WriteA1X]>,
|
|
SchedVar<ExynosLogicExPred, [M5WriteA1X]>,
|
|
SchedVar<NoSchedPred, [M5WriteAAX]>]>;
|
|
def M5WriteAVW : SchedWriteVariant<[SchedVar<ExynosResetPred, [M5WriteZ0]>,
|
|
SchedVar<ExynosArithPred, [M5WriteA1W]>,
|
|
SchedVar<ExynosLogicExPred, [M5WriteA1W]>,
|
|
SchedVar<NoSchedPred, [M5WriteAAW]>]>;
|
|
def M5WriteAVX : SchedWriteVariant<[SchedVar<ExynosResetPred, [M5WriteZ0]>,
|
|
SchedVar<ExynosArithPred, [M5WriteA1X]>,
|
|
SchedVar<ExynosLogicExPred, [M5WriteA1X]>,
|
|
SchedVar<NoSchedPred, [M5WriteAAX]>]>;
|
|
def M5WriteAXW : SchedWriteVariant<[SchedVar<ExynosArithPred, [M5WriteA1W]>,
|
|
SchedVar<ExynosLogicExPred, [M5WriteA1W]>,
|
|
SchedVar<NoSchedPred, [M5WriteAAW]>]>;
|
|
def M5WriteAXX : SchedWriteVariant<[SchedVar<ExynosArithPred, [M5WriteA1X]>,
|
|
SchedVar<ExynosLogicExPred, [M5WriteA1X]>,
|
|
SchedVar<NoSchedPred, [M5WriteAAX]>]>;
|
|
def M5WriteAYW : SchedWriteVariant<[SchedVar<ExynosRotateRightImmPred, [M5WriteA1W]>,
|
|
SchedVar<NoSchedPred, [M5WriteAFW]>]>;
|
|
def M5WriteAYX : SchedWriteVariant<[SchedVar<ExynosRotateRightImmPred, [M5WriteA1X]>,
|
|
SchedVar<NoSchedPred, [M5WriteAFX]>]>;
|
|
|
|
def M5WriteB1 : SchedWriteRes<[M5UnitB]> { let Latency = 1; }
|
|
def M5WriteBX : SchedWriteVariant<[SchedVar<ExynosBranchLinkLRPred, [M5WriteAC]>,
|
|
SchedVar<NoSchedPred, [M5WriteAB]>]>;
|
|
|
|
def M5WriteC1 : SchedWriteRes<[M5UnitC]> { let Latency = 1; }
|
|
def M5WriteC2 : SchedWriteRes<[M5UnitC]> { let Latency = 2; }
|
|
def M5WriteCA : SchedWriteRes<[M5UnitC]> { let Latency = 3;
|
|
let ResourceCycles = [2]; }
|
|
|
|
def M5WriteD10 : SchedWriteRes<[M5UnitD]> { let Latency = 10;
|
|
let ResourceCycles = [10]; }
|
|
def M5WriteD16 : SchedWriteRes<[M5UnitD]> { let Latency = 16;
|
|
let ResourceCycles = [16]; }
|
|
|
|
def M5WriteF2 : SchedWriteRes<[M5UnitF]> { let Latency = 2; }
|
|
|
|
def M5WriteL4 : SchedWriteRes<[M5UnitL]> { let Latency = 4; }
|
|
def M5WriteL5 : SchedWriteRes<[M5UnitL]> { let Latency = 5; }
|
|
def M5WriteL6 : SchedWriteRes<[M5UnitL]> { let Latency = 6; }
|
|
def M5WriteLA : SchedWriteRes<[M5UnitL,
|
|
M5UnitL]> { let Latency = 6;
|
|
let NumMicroOps = 1; }
|
|
def M5WriteLB : SchedWriteRes<[M5UnitAX,
|
|
M5UnitL]> { let Latency = 6;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteLC : SchedWriteRes<[M5UnitAX,
|
|
M5UnitL,
|
|
M5UnitL]> { let Latency = 6;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteLD : SchedWriteRes<[M5UnitAX,
|
|
M5UnitL]> { let Latency = 4;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteLE : SchedWriteRes<[M5UnitAX,
|
|
M5UnitL]> { let Latency = 7;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteLFW : SchedWriteRes<[M5UnitAW,
|
|
M5UnitAW,
|
|
M5UnitAW,
|
|
M5UnitAW,
|
|
M5UnitL]> { let Latency = 15;
|
|
let NumMicroOps = 6;
|
|
let ResourceCycles = [1, 1, 1, 1, 15]; }
|
|
def M5WriteLFX : SchedWriteRes<[M5UnitAX,
|
|
M5UnitAX,
|
|
M5UnitAX,
|
|
M5UnitAX,
|
|
M5UnitL]> { let Latency = 15;
|
|
let NumMicroOps = 6;
|
|
let ResourceCycles = [1, 1, 1, 1, 15]; }
|
|
def M5WriteLGW : SchedWriteRes<[M5UnitAW,
|
|
M5UnitL]> { let Latency = 13;
|
|
let NumMicroOps = 1;
|
|
let ResourceCycles = [1, 13]; }
|
|
def M5WriteLGX : SchedWriteRes<[M5UnitAX,
|
|
M5UnitL]> { let Latency = 13;
|
|
let NumMicroOps = 1;
|
|
let ResourceCycles = [1, 13]; }
|
|
def M5WriteLH : SchedWriteRes<[]> { let Latency = 6;
|
|
let NumMicroOps = 0; }
|
|
def M5WriteLX : SchedWriteVariant<[SchedVar<ExynosScaledIdxPred, [M5WriteL5]>,
|
|
SchedVar<NoSchedPred, [M5WriteL4]>]>;
|
|
def M5WriteLY : SchedWriteVariant<[SchedVar<ExynosScaledIdxPred, [M5WriteLE]>,
|
|
SchedVar<NoSchedPred, [M5WriteL6]>]>;
|
|
|
|
def M5WriteS1 : SchedWriteRes<[M5UnitS]> { let Latency = 1; }
|
|
def M5WriteSA : SchedWriteRes<[M5UnitS0]> { let Latency = 4; }
|
|
def M5WriteSB : SchedWriteRes<[M5UnitAX,
|
|
M5UnitS]> { let Latency = 2;
|
|
let NumMicroOps = 1; }
|
|
def M5WriteSX : SchedWriteVariant<[SchedVar<ExynosScaledIdxPred, [M5WriteSB]>,
|
|
SchedVar<NoSchedPred, [M5WriteS1]>]>;
|
|
|
|
def M5ReadAdrBase : SchedReadVariant<[SchedVar<
|
|
MCSchedPredicate<
|
|
CheckAny<
|
|
[ScaledIdxFn,
|
|
ExynosScaledIdxFn]>>, [ReadDefault]>,
|
|
SchedVar<NoSchedPred, [ReadDefault]>]>;
|
|
|
|
def M5WriteNEONB : SchedWriteRes<[M5UnitNALU,
|
|
M5UnitS0]> { let Latency = 5;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteNEONH : SchedWriteRes<[M5UnitNALU,
|
|
M5UnitS0]> { let Latency = 2;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteNEONI : SchedWriteRes<[M5UnitS0,
|
|
M5UnitNSHF]> { let Latency = 6;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteNEONK : SchedWriteRes<[M5UnitNSHF,
|
|
M5UnitFCVT0,
|
|
M5UnitS0]> { let Latency = 5;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteNEONN : SchedWriteRes<[M5UnitNMSC,
|
|
M5UnitNMSC]> { let Latency = 5;
|
|
let NumMicroOps = 2;
|
|
let ResourceCycles = [7, 7]; }
|
|
def M5WriteNEONO : SchedWriteRes<[M5UnitNMSC,
|
|
M5UnitNMSC,
|
|
M5UnitNMSC]> { let Latency = 8;
|
|
let NumMicroOps = 3;
|
|
let ResourceCycles = [10, 10, 10]; }
|
|
def M5WriteNEONP : SchedWriteRes<[M5UnitNSHF,
|
|
M5UnitS0,
|
|
M5UnitFCVT]> { let Latency = 7;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteNEONQ : SchedWriteRes<[M5UnitNMSC,
|
|
M5UnitC]> { let Latency = 3;
|
|
let NumMicroOps = 1; }
|
|
def M5WriteNEONU : SchedWriteRes<[M5UnitFSQR,
|
|
M5UnitFSQR]> { let Latency = 7;
|
|
let ResourceCycles = [4, 4]; }
|
|
def M5WriteNEONV : SchedWriteRes<[M5UnitFDIV,
|
|
M5UnitFDIV]> { let Latency = 7;
|
|
let ResourceCycles = [6, 6]; }
|
|
def M5WriteNEONW : SchedWriteRes<[M5UnitFDIV,
|
|
M5UnitFDIV]> { let Latency = 12;
|
|
let ResourceCycles = [9, 9]; }
|
|
def M5WriteNEONX : SchedWriteRes<[M5UnitFSQR,
|
|
M5UnitFSQR]> { let Latency = 8;
|
|
let ResourceCycles = [5, 5]; }
|
|
def M5WriteNEONY : SchedWriteRes<[M5UnitFSQR,
|
|
M5UnitFSQR]> { let Latency = 12;
|
|
let ResourceCycles = [9, 9]; }
|
|
def M5WriteNEONZ : SchedWriteVariant<[SchedVar<ExynosQFormPred, [M5WriteNEONO]>,
|
|
SchedVar<NoSchedPred, [M5WriteNEONN]>]>;
|
|
|
|
def M5WriteFADD2 : SchedWriteRes<[M5UnitFADD]> { let Latency = 2; }
|
|
|
|
def M5WriteFCVT2 : SchedWriteRes<[M5UnitFCVT]> { let Latency = 2; }
|
|
def M5WriteFCVT2A : SchedWriteRes<[M5UnitFCVT0]> { let Latency = 2; }
|
|
def M5WriteFCVT3 : SchedWriteRes<[M5UnitFCVT]> { let Latency = 3; }
|
|
def M5WriteFCVT3A : SchedWriteRes<[M5UnitFCVT0]> { let Latency = 3; }
|
|
def M5WriteFCVTA : SchedWriteRes<[M5UnitFCVT0,
|
|
M5UnitS0]> { let Latency = 3;
|
|
let NumMicroOps = 1; }
|
|
def M5WriteFCVTB : SchedWriteRes<[M5UnitFCVT,
|
|
M5UnitS0]> { let Latency = 4;
|
|
let NumMicroOps = 1; }
|
|
def M5WriteFCVTC : SchedWriteRes<[M5UnitFCVT,
|
|
M5UnitS0]> { let Latency = 6;
|
|
let NumMicroOps = 1; }
|
|
|
|
def M5WriteFDIV5 : SchedWriteRes<[M5UnitFDIV]> { let Latency = 5;
|
|
let ResourceCycles = [2]; }
|
|
def M5WriteFDIV7 : SchedWriteRes<[M5UnitFDIV]> { let Latency = 7;
|
|
let ResourceCycles = [4]; }
|
|
def M5WriteFDIV12 : SchedWriteRes<[M5UnitFDIV]> { let Latency = 12;
|
|
let ResourceCycles = [9]; }
|
|
|
|
def M5WriteFMAC3 : SchedWriteRes<[M5UnitFMAC]> { let Latency = 3; }
|
|
def M5WriteFMAC4 : SchedWriteRes<[M5UnitFMAC]> { let Latency = 4; }
|
|
def M5WriteFMAC5 : SchedWriteRes<[M5UnitFMAC]> { let Latency = 5; }
|
|
|
|
def M5WriteFSQR5 : SchedWriteRes<[M5UnitFSQR]> { let Latency = 5;
|
|
let ResourceCycles = [2]; }
|
|
def M5WriteFSQR7 : SchedWriteRes<[M5UnitFSQR]> { let Latency = 7;
|
|
let ResourceCycles = [4]; }
|
|
def M5WriteFSQR8 : SchedWriteRes<[M5UnitFSQR]> { let Latency = 8;
|
|
let ResourceCycles = [5]; }
|
|
def M5WriteFSQR12 : SchedWriteRes<[M5UnitFSQR]> { let Latency = 12;
|
|
let ResourceCycles = [9]; }
|
|
|
|
def M5WriteNALU1 : SchedWriteRes<[M5UnitNALU]> { let Latency = 1; }
|
|
def M5WriteNALU2 : SchedWriteRes<[M5UnitNALU]> { let Latency = 2; }
|
|
|
|
def M5WriteNDOT2 : SchedWriteRes<[M5UnitNDOT]> { let Latency = 2; }
|
|
|
|
def M5WriteNCRY2 : SchedWriteRes<[M5UnitNCRY]> { let Latency = 2; }
|
|
def M5WriteNCRY1A : SchedWriteRes<[M5UnitNCRY0]> { let Latency = 1; }
|
|
def M5WriteNCRY2A : SchedWriteRes<[M5UnitNCRY0]> { let Latency = 2; }
|
|
def M5WriteNCRY3A : SchedWriteRes<[M5UnitNCRY0]> { let Latency = 3; }
|
|
def M5WriteNCRY5A : SchedWriteRes<[M5UnitNCRY]> { let Latency = 5; }
|
|
|
|
def M5WriteNHAD1 : SchedWriteRes<[M5UnitNHAD]> { let Latency = 1; }
|
|
def M5WriteNHAD3 : SchedWriteRes<[M5UnitNHAD]> { let Latency = 3; }
|
|
|
|
def M5WriteNMSC1 : SchedWriteRes<[M5UnitNMSC]> { let Latency = 1; }
|
|
def M5WriteNMSC2 : SchedWriteRes<[M5UnitNMSC]> { let Latency = 2; }
|
|
|
|
def M5WriteNMUL3 : SchedWriteRes<[M5UnitNMUL]> { let Latency = 3; }
|
|
|
|
def M5WriteNSHF1 : SchedWriteRes<[M5UnitNSHF]> { let Latency = 1; }
|
|
def M5WriteNSHF2 : SchedWriteRes<[M5UnitNSHF]> { let Latency = 2; }
|
|
def M5WriteNSHFA : SchedWriteRes<[M5UnitNSHF]> { let Latency = 2; }
|
|
def M5WriteNSHFB : SchedWriteRes<[M5UnitNSHF]> { let Latency = 4;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteNSHFC : SchedWriteRes<[M5UnitNSHF]> { let Latency = 6;
|
|
let NumMicroOps = 3; }
|
|
def M5WriteNSHFD : SchedWriteRes<[M5UnitNSHF]> { let Latency = 8;
|
|
let NumMicroOps = 4; }
|
|
|
|
def M5WriteNSHT2 : SchedWriteRes<[M5UnitNSHT]> { let Latency = 2; }
|
|
def M5WriteNSHT4A : SchedWriteRes<[M5UnitNSHT1]> { let Latency = 4; }
|
|
|
|
def M5WriteVLDA : SchedWriteRes<[M5UnitL,
|
|
M5UnitL]> { let Latency = 6;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteVLDB : SchedWriteRes<[M5UnitL,
|
|
M5UnitL,
|
|
M5UnitL]> { let Latency = 7;
|
|
let NumMicroOps = 3; }
|
|
def M5WriteVLDC : SchedWriteRes<[M5UnitL,
|
|
M5UnitL,
|
|
M5UnitL,
|
|
M5UnitL]> { let Latency = 7;
|
|
let NumMicroOps = 4; }
|
|
def M5WriteVLDD : SchedWriteRes<[M5UnitL,
|
|
M5UnitNSHF]> { let Latency = 7;
|
|
let NumMicroOps = 2;
|
|
let ResourceCycles = [2, 1]; }
|
|
def M5WriteVLDF : SchedWriteRes<[M5UnitL,
|
|
M5UnitL]> { let Latency = 11;
|
|
let NumMicroOps = 2;
|
|
let ResourceCycles = [6, 5]; }
|
|
def M5WriteVLDG : SchedWriteRes<[M5UnitL,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF]> { let Latency = 7;
|
|
let NumMicroOps = 3;
|
|
let ResourceCycles = [2, 1, 1]; }
|
|
def M5WriteVLDI : SchedWriteRes<[M5UnitL,
|
|
M5UnitL,
|
|
M5UnitL]> { let Latency = 13;
|
|
let NumMicroOps = 3; }
|
|
def M5WriteVLDJ : SchedWriteRes<[M5UnitL,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF]> { let Latency = 8;
|
|
let NumMicroOps = 4; }
|
|
def M5WriteVLDK : SchedWriteRes<[M5UnitL,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF]> { let Latency = 8;
|
|
let NumMicroOps = 5; }
|
|
def M5WriteVLDL : SchedWriteRes<[M5UnitL,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitL,
|
|
M5UnitNSHF]> { let Latency = 8;
|
|
let NumMicroOps = 5; }
|
|
def M5WriteVLDM : SchedWriteRes<[M5UnitL,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitL,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF]> { let Latency = 8;
|
|
let NumMicroOps = 6; }
|
|
def M5WriteVLDN : SchedWriteRes<[M5UnitL,
|
|
M5UnitL,
|
|
M5UnitL,
|
|
M5UnitL]> { let Latency = 15;
|
|
let NumMicroOps = 4;
|
|
let ResourceCycles = [2, 2, 2, 2]; }
|
|
|
|
def M5WriteVST1 : SchedWriteRes<[M5UnitS,
|
|
M5UnitFST]> { let Latency = 1;
|
|
let NumMicroOps = 1; }
|
|
def M5WriteVSTA : SchedWriteRes<[M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST]> { let Latency = 2;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteVSTB : SchedWriteRes<[M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST]> { let Latency = 3;
|
|
let NumMicroOps = 3; }
|
|
def M5WriteVSTC : SchedWriteRes<[M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST]> { let Latency = 4;
|
|
let NumMicroOps = 4; }
|
|
def M5WriteVSTD : SchedWriteRes<[M5UnitS,
|
|
M5UnitFST]> { let Latency = 2; }
|
|
def M5WriteVSTE : SchedWriteRes<[M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST]> { let Latency = 2;
|
|
let NumMicroOps = 1; }
|
|
def M5WriteVSTF : SchedWriteRes<[M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitS,
|
|
M5UnitFST]> { let Latency = 4;
|
|
let NumMicroOps = 3; }
|
|
def M5WriteVSTG : SchedWriteRes<[M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST]> { let Latency = 4;
|
|
let NumMicroOps = 5; }
|
|
def M5WriteVSTH : SchedWriteRes<[M5UnitS0,
|
|
M5UnitFST]> { let Latency = 1;
|
|
let NumMicroOps = 1; }
|
|
def M5WriteVSTI : SchedWriteRes<[M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST]> { let Latency = 8;
|
|
let NumMicroOps = 5;
|
|
let ResourceCycles = [1, 1, 1, 1, 2, 1, 2, 1, 2, 1, 2, 1]; }
|
|
def M5WriteVSTJ : SchedWriteRes<[M5UnitA,
|
|
M5UnitS0,
|
|
M5UnitFST]> { let Latency = 1;
|
|
let NumMicroOps = 1; }
|
|
def M5WriteVSTK : SchedWriteRes<[M5UnitAX,
|
|
M5UnitS,
|
|
M5UnitFST]> { let Latency = 3;
|
|
let NumMicroOps = 2; }
|
|
def M5WriteVSTL : SchedWriteRes<[M5UnitNSHF,
|
|
M5UnitNSHF,
|
|
M5UnitS,
|
|
M5UnitFST,
|
|
M5UnitS,
|
|
M5UnitFST]> { let Latency = 4;
|
|
let NumMicroOps = 4;
|
|
let ResourceCycles = [1, 1, 2, 1, 2, 1]; }
|
|
def M5WriteVSTY : SchedWriteVariant<[SchedVar<ExynosScaledIdxPred, [M5WriteVSTK]>,
|
|
SchedVar<NoSchedPred, [WriteVST]>]>;
|
|
|
|
// Special cases.
|
|
def M5WriteCOPY : SchedWriteVariant<[SchedVar<ExynosFPPred, [M5WriteNALU2]>,
|
|
SchedVar<NoSchedPred, [M5WriteZ0]>]>;
|
|
def M5WriteMOVI : SchedWriteVariant<[SchedVar<IsZeroFPIdiomPred, [M5WriteZ0]>,
|
|
SchedVar<NoSchedPred, [M5WriteNALU1]>]>;
|
|
|
|
// Fast forwarding.
|
|
def M5ReadFM1 : SchedReadAdvance<+1, [M5WriteF2]>;
|
|
def M5ReadAESM2 : SchedReadAdvance<+2, [M5WriteNCRY2]>;
|
|
def M5ReadFMACM1 : SchedReadAdvance<+1, [M5WriteFMAC4,
|
|
M5WriteFMAC5]>;
|
|
def M5ReadNMULM1 : SchedReadAdvance<+1, [M5WriteNMUL3]>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Coarse scheduling model.
|
|
|
|
// Branch instructions.
|
|
def : SchedAlias<WriteBr, M5WriteZ0>;
|
|
def : SchedAlias<WriteBrReg, M5WriteC1>;
|
|
|
|
// Arithmetic and logical integer instructions.
|
|
def : SchedAlias<WriteI, M5WriteA1W>;
|
|
def : SchedAlias<WriteIEReg, M5WriteA1W>; // FIXME: M5WriteAX crashes TableGen.
|
|
def : SchedAlias<WriteISReg, M5WriteA1W>; // FIXME: M5WriteAX crashes TableGen.
|
|
def : SchedAlias<WriteIS, M5WriteA1W>;
|
|
|
|
// Move instructions.
|
|
def : SchedAlias<WriteImm, M5WriteA1W>;
|
|
|
|
// Divide and multiply instructions.
|
|
def : SchedAlias<WriteID32, M5WriteD10>;
|
|
def : SchedAlias<WriteID64, M5WriteD16>;
|
|
def : SchedAlias<WriteIM32, M5WriteC2>;
|
|
def : SchedAlias<WriteIM64, M5WriteCA>;
|
|
|
|
// Miscellaneous instructions.
|
|
def : SchedAlias<WriteExtr, M5WriteAYW>;
|
|
|
|
// Addressing modes.
|
|
def : SchedAlias<WriteAdr, M5WriteZ1>;
|
|
def : SchedAlias<ReadAdrBase, M5ReadAdrBase>;
|
|
|
|
// Load instructions.
|
|
def : SchedAlias<WriteLD, M5WriteL4>;
|
|
def : SchedAlias<WriteLDHi, M5WriteZ4>;
|
|
def : SchedAlias<WriteLDIdx, M5WriteLX>;
|
|
|
|
// Store instructions.
|
|
def : SchedAlias<WriteST, M5WriteS1>;
|
|
def : SchedAlias<WriteSTP, M5WriteS1>;
|
|
def : SchedAlias<WriteSTX, M5WriteS1>;
|
|
def : SchedAlias<WriteSTIdx, M5WriteSX>;
|
|
|
|
// Atomic load and store instructions.
|
|
def : SchedAlias<WriteAtomic, M5WriteLGW>;
|
|
|
|
// FP data instructions.
|
|
def : SchedAlias<WriteF, M5WriteFADD2>;
|
|
def : SchedAlias<WriteFCmp, M5WriteNMSC2>;
|
|
def : SchedAlias<WriteFDiv, M5WriteFDIV12>;
|
|
def : SchedAlias<WriteFMul, M5WriteFMAC3>;
|
|
|
|
// FP miscellaneous instructions.
|
|
def : SchedAlias<WriteFCvt, M5WriteFCVT2>;
|
|
def : SchedAlias<WriteFImm, M5WriteNALU1>;
|
|
def : SchedAlias<WriteFCopy, M5WriteNALU2>;
|
|
|
|
// FP load instructions.
|
|
def : SchedAlias<WriteVLD, M5WriteL6>;
|
|
|
|
// FP store instructions.
|
|
def : SchedAlias<WriteVST, M5WriteVST1>;
|
|
|
|
// ASIMD FP instructions.
|
|
def : SchedAlias<WriteV, M5WriteNALU1>;
|
|
|
|
// Other miscellaneous instructions.
|
|
def : WriteRes<WriteBarrier, []> { let Latency = 1; }
|
|
def : WriteRes<WriteHint, []> { let Latency = 1; }
|
|
def : WriteRes<WriteSys, []> { let Latency = 1; }
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Generic fast forwarding.
|
|
|
|
// TODO: Add FP register forwarding rules.
|
|
|
|
def : ReadAdvance<ReadI, 0>;
|
|
def : ReadAdvance<ReadISReg, 0>;
|
|
def : ReadAdvance<ReadIEReg, 0>;
|
|
def : ReadAdvance<ReadIM, 0>;
|
|
// TODO: The forwarding for 32 bits actually saves 2 cycles.
|
|
def : ReadAdvance<ReadIMA, 3, [WriteIM32, WriteIM64]>;
|
|
def : ReadAdvance<ReadID, 0>;
|
|
def : ReadAdvance<ReadExtrHi, 0>;
|
|
def : ReadAdvance<ReadAdrBase, 0>;
|
|
def : ReadAdvance<ReadVLD, 0>;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Finer scheduling model.
|
|
|
|
// Branch instructions
|
|
def : InstRW<[M5WriteB1], (instrs Bcc)>;
|
|
def : InstRW<[M5WriteAFX], (instrs BL)>;
|
|
def : InstRW<[M5WriteBX], (instrs BLR)>;
|
|
def : InstRW<[M5WriteC1], (instregex "^CBN?Z[WX]")>;
|
|
def : InstRW<[M5WriteAD], (instregex "^TBN?ZW")>;
|
|
def : InstRW<[M5WriteAB], (instregex "^TBN?ZX")>;
|
|
|
|
// Arithmetic and logical integer instructions.
|
|
def : InstRW<[M5WriteA1W], (instregex "^(ADC|SBC)S?Wr$")>;
|
|
def : InstRW<[M5WriteA1X], (instregex "^(ADC|SBC)S?Xr$")>;
|
|
def : InstRW<[M5WriteAXW], (instregex "^(ADD|AND|BIC|EON|EOR|ORN|SUB)Wrs$")>;
|
|
def : InstRW<[M5WriteAXX], (instregex "^(ADD|AND|BIC|EON|EOR|ORN|SUB)Xrs$")>;
|
|
def : InstRW<[M5WriteAUW], (instrs ORRWrs)>;
|
|
def : InstRW<[M5WriteAUX], (instrs ORRXrs)>;
|
|
def : InstRW<[M5WriteAXW], (instregex "^(ADD|AND|BIC|SUB)SWrs$")>;
|
|
def : InstRW<[M5WriteAXX], (instregex "^(ADD|AND|BIC|SUB)SXrs$")>;
|
|
def : InstRW<[M5WriteAXW], (instregex "^(ADD|SUB)S?Wrx(64)?$")>;
|
|
def : InstRW<[M5WriteAXX], (instregex "^(ADD|SUB)S?Xrx(64)?$")>;
|
|
def : InstRW<[M5WriteAVW], (instrs ADDWri, ORRWri)>;
|
|
def : InstRW<[M5WriteAVX], (instrs ADDXri, ORRXri)>;
|
|
def : InstRW<[M5WriteA1W], (instregex "^CCM[NP]W[ir]$")>;
|
|
def : InstRW<[M5WriteA1X], (instregex "^CCM[NP]X[ir]$")>;
|
|
def : InstRW<[M5WriteA1W], (instrs CSELWr, CSINCWr, CSINVWr, CSNEGWr)>;
|
|
def : InstRW<[M5WriteA1X], (instrs CSELXr, CSINCXr, CSINVXr, CSNEGXr)>;
|
|
|
|
// Move instructions.
|
|
def : InstRW<[M5WriteCOPY], (instrs COPY)>;
|
|
def : InstRW<[M5WriteZ0], (instrs ADR, ADRP)>;
|
|
def : InstRW<[M5WriteZ0], (instregex "^MOV[NZ][WX]i$")>;
|
|
|
|
// Shift instructions.
|
|
def : InstRW<[M5WriteA1W], (instrs ASRVWr, LSLVWr, LSRVWr, RORVWr)>;
|
|
def : InstRW<[M5WriteA1X], (instrs ASRVXr, LSLVXr, LSRVXr, RORVXr)>;
|
|
|
|
// Miscellaneous instructions.
|
|
def : InstRW<[M5WriteAYW], (instrs EXTRWrri)>;
|
|
def : InstRW<[M5WriteAYX], (instrs EXTRXrri)>;
|
|
def : InstRW<[M5WriteA1W], (instrs BFMWri, SBFMWri, UBFMWri)>;
|
|
def : InstRW<[M5WriteA1X], (instrs BFMXri, SBFMXri, UBFMXri)>;
|
|
def : InstRW<[M5WriteA1W], (instrs CLSWr, CLZWr)>;
|
|
def : InstRW<[M5WriteA1X], (instrs CLSXr, CLZXr)>;
|
|
def : InstRW<[M5WriteA1W], (instrs RBITWr, REVWr, REV16Wr)>;
|
|
def : InstRW<[M5WriteA1X], (instrs RBITXr, REVXr, REV16Xr, REV32Xr)>;
|
|
|
|
// Load instructions.
|
|
def : InstRW<[M5WriteLD,
|
|
WriteLDHi,
|
|
WriteAdr], (instregex "^LDP(SW|W|X)(post|pre)")>;
|
|
def : InstRW<[M5WriteL5,
|
|
ReadAdrBase], (instregex "^LDR(BB|SBW|SBX|HH|SHW|SHX|SW|W|X)roW")>;
|
|
def : InstRW<[WriteLDIdx,
|
|
ReadAdrBase], (instregex "^LDR(BB|SBW|SBX|HH|SHW|SHX|SW|W|X)roX")>;
|
|
def : InstRW<[M5WriteL5,
|
|
ReadAdrBase], (instrs PRFMroW)>;
|
|
def : InstRW<[WriteLDIdx,
|
|
ReadAdrBase], (instrs PRFMroX)>;
|
|
|
|
// Store instructions.
|
|
def : InstRW<[M5WriteSB,
|
|
ReadAdrBase], (instregex "^STR(BB|HH|W|X)roW")>;
|
|
def : InstRW<[WriteST,
|
|
ReadAdrBase], (instregex "^STR(BB|HH|W|X)roX")>;
|
|
|
|
// Atomic load and store instructions.
|
|
def : InstRW<[M5WriteLGW], (instregex "^CAS(A|AL|L)?[BHW]$")>;
|
|
def : InstRW<[M5WriteLGX], (instregex "^CAS(A|AL|L)?X$")>;
|
|
def : InstRW<[M5WriteLFW], (instregex "^CASP(A|AL|L)?W$")>;
|
|
def : InstRW<[M5WriteLFX], (instregex "^CASP(A|AL|L)?X$")>;
|
|
def : InstRW<[M5WriteLGW], (instregex "^LD(ADD|CLR|EOR|SET|[SU]MAX|[SU]MIN)(A|AL|L)?[BHW]$")>;
|
|
def : InstRW<[M5WriteLGX], (instregex "^LD(ADD|CLR|EOR|SET|[SU]MAX|[SU]MIN)(A|AL|L)?X$")>;
|
|
def : InstRW<[M5WriteLGW], (instregex "^SWP(A|AL|L)?[BHW]$")>;
|
|
def : InstRW<[M5WriteLGX], (instregex "^SWP(A|AL|L)?X$")>;
|
|
|
|
// FP data instructions.
|
|
def : InstRW<[M5WriteNSHF1], (instrs FABSHr, FABSSr,FABSDr)>;
|
|
def : InstRW<[M5WriteFADD2], (instregex "^F(ADD|SUB)[HSD]rr")>;
|
|
def : InstRW<[M5WriteFADD2], (instregex "^FADDPv.i(16|32|64)")>;
|
|
def : InstRW<[M5WriteNEONQ], (instregex "^FCCMPE?[HSD]rr")>;
|
|
def : InstRW<[M5WriteNMSC2], (instregex "^FCMPE?[HSD]r[ir]")>;
|
|
def : InstRW<[M5WriteNMSC1], (instregex "^F(AC|CM)(EQ|GE|GT|LE|LT)(16|32|64|v1)")>;
|
|
def : InstRW<[M5WriteFDIV5], (instrs FDIVHrr)>;
|
|
def : InstRW<[M5WriteFDIV7], (instrs FDIVSrr)>;
|
|
def : InstRW<[M5WriteFDIV12], (instrs FDIVDrr)>;
|
|
def : InstRW<[M5WriteNMSC1], (instregex "^F(MAX|MIN)(NM)?[HSD]rr")>;
|
|
def : InstRW<[M5WriteFMAC3], (instregex "^FN?MUL[HSD]rr")>;
|
|
def : InstRW<[M5WriteFMAC3], (instrs FMULX16, FMULX32, FMULX64)>;
|
|
def : InstRW<[M5WriteFMAC4,
|
|
M5ReadFMACM1], (instregex "^FN?M(ADD|SUB)[HSD]rrr")>;
|
|
def : InstRW<[M5WriteNALU2], (instrs FNEGHr, FNEGSr, FNEGDr)>;
|
|
def : InstRW<[M5WriteFCVT3A], (instregex "^FRINT.+r")>;
|
|
def : InstRW<[M5WriteNEONH], (instregex "^FCSEL[HSD]rrr")>;
|
|
def : InstRW<[M5WriteFSQR5], (instrs FSQRTHr)>;
|
|
def : InstRW<[M5WriteFSQR8], (instrs FSQRTSr)>;
|
|
def : InstRW<[M5WriteFSQR12], (instrs FSQRTDr)>;
|
|
|
|
// FP miscellaneous instructions.
|
|
def : InstRW<[M5WriteFCVT2], (instregex "^FCVT[HSD][HSD]r")>;
|
|
def : InstRW<[M5WriteFCVTC], (instregex "^[SU]CVTF[SU][XW][HSD]ri")>;
|
|
def : InstRW<[M5WriteFCVTB], (instregex "^FCVT[AMNPZ][SU][SU][XW][HSD]r")>;
|
|
def : InstRW<[M5WriteNALU1], (instregex "^FMOV[HSD]i")>;
|
|
def : InstRW<[M5WriteNALU2], (instregex "^FMOV[HSD]r")>;
|
|
def : InstRW<[M5WriteSA], (instregex "^FMOV[WX][HSD]r")>;
|
|
def : InstRW<[M5WriteFCVTA], (instregex "^FMOV[HSD][WX]r")>;
|
|
def : InstRW<[M5WriteNEONI], (instregex "^FMOVXDHighr")>;
|
|
def : InstRW<[M5WriteNEONK], (instregex "^FMOVDXHighr")>;
|
|
def : InstRW<[M5WriteFCVT3], (instregex "^F(RECP|RSQRT)Ev1(f16|i32|i64)")>;
|
|
def : InstRW<[M5WriteNMSC1], (instregex "^FRECPXv1")>;
|
|
def : InstRW<[M5WriteFMAC4], (instregex "^F(RECP|RSQRT)S(16|32|64)")>;
|
|
|
|
// FP load instructions.
|
|
def : InstRW<[WriteVLD], (instregex "^LDR[SDQ]l")>;
|
|
def : InstRW<[WriteVLD], (instregex "^LDUR[BHSDQ]i")>;
|
|
def : InstRW<[WriteVLD,
|
|
WriteAdr], (instregex "^LDR[BHSDQ](post|pre)")>;
|
|
def : InstRW<[WriteVLD], (instregex "^LDR[BHSDQ]ui")>;
|
|
def : InstRW<[M5WriteLE,
|
|
ReadAdrBase], (instregex "^LDR[BHSDQ]roW")>;
|
|
def : InstRW<[WriteVLD,
|
|
ReadAdrBase], (instregex "^LDR[BHSD]roX")>;
|
|
def : InstRW<[M5WriteLY,
|
|
ReadAdrBase], (instrs LDRQroX)>;
|
|
def : InstRW<[WriteVLD,
|
|
M5WriteLH], (instregex "^LDN?P[SD]i")>;
|
|
def : InstRW<[M5WriteLA,
|
|
M5WriteLH], (instregex "^LDN?PQi")>;
|
|
def : InstRW<[M5WriteLB,
|
|
M5WriteLH,
|
|
WriteAdr], (instregex "^LDP[SD](post|pre)")>;
|
|
def : InstRW<[M5WriteLC,
|
|
M5WriteLH,
|
|
WriteAdr], (instregex "^LDPQ(post|pre)")>;
|
|
|
|
// FP store instructions.
|
|
def : InstRW<[WriteVST], (instregex "^STUR[BHSDQ]i")>;
|
|
def : InstRW<[WriteVST,
|
|
WriteAdr], (instregex "^STR[BHSDQ](post|pre)")>;
|
|
def : InstRW<[WriteVST], (instregex "^STR[BHSDQ]ui")>;
|
|
def : InstRW<[WriteVST,
|
|
ReadAdrBase], (instregex "^STR[BHSD]ro[WX]")>;
|
|
def : InstRW<[M5WriteVSTK,
|
|
ReadAdrBase], (instregex "^STRQroW")>;
|
|
def : InstRW<[M5WriteVSTY,
|
|
ReadAdrBase], (instregex "^STRQroX")>;
|
|
def : InstRW<[WriteVST], (instregex "^STN?P[SD]i")>;
|
|
def : InstRW<[M5WriteVSTH], (instregex "^STN?PQi")>;
|
|
def : InstRW<[WriteVST,
|
|
WriteAdr], (instregex "^STP[SD](post|pre)")>;
|
|
def : InstRW<[M5WriteVSTJ,
|
|
WriteAdr], (instregex "^STPQ(post|pre)")>;
|
|
|
|
// ASIMD instructions.
|
|
def : InstRW<[M5WriteNHAD1], (instregex "^[SU]ABDL?v")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^[SU]ABAL?v")>;
|
|
def : InstRW<[M5WriteNMSC1], (instregex "^ABSv")>;
|
|
def : InstRW<[M5WriteNALU2], (instregex "^(ADD|NEG|SUB)v")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^[SU]?ADDL?Pv")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^[SU]H(ADD|SUB)v")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^[SU](ADD|SUB)[LW]v")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^R?(ADD|SUB)HN2?v")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^[SU]Q(ADD|SUB)v")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^(SU|US)QADDv")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^[SU]RHADDv")>;
|
|
def : InstRW<[M5WriteNMSC1], (instregex "^SQ(ABS|NEG)v")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^[SU]?ADDL?Vv")>;
|
|
def : InstRW<[M5WriteNMSC1], (instregex "^CM(EQ|GE|GT|HI|HS|LE|LT)v")>;
|
|
def : InstRW<[M5WriteNALU2], (instregex "^CMTSTv")>;
|
|
def : InstRW<[M5WriteNALU2], (instregex "^(AND|BIC|EOR|NOT|ORN|ORR)v")>;
|
|
def : InstRW<[M5WriteNMSC1], (instregex "^[SU](MIN|MAX)v")>;
|
|
def : InstRW<[M5WriteNMSC2], (instregex "^[SU](MIN|MAX)Pv")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^[SU](MIN|MAX)Vv")>;
|
|
def : InstRW<[M5WriteNMUL3], (instregex "^(SQR?D)?MULH?v")>;
|
|
def : InstRW<[M5WriteNMUL3,
|
|
M5ReadNMULM1], (instregex "^ML[AS]v")>;
|
|
def : InstRW<[M5WriteNMUL3,
|
|
M5ReadNMULM1], (instregex "^SQRDML[AS]H")>;
|
|
def : InstRW<[M5WriteNMUL3], (instregex "^(S|U|SQD)ML[AS]L(v1(i32|i64)|v2i32|v4i16|v8i8)")>;
|
|
def : InstRW<[M5WriteNMUL3,
|
|
M5ReadNMULM1], (instregex "^(S|U|SQD)ML[AS]L(v4i32|v8i16|v16i8)")>;
|
|
def : InstRW<[M5WriteNMUL3,
|
|
M5ReadNMULM1], (instregex "^(S|U|SQD)MULL(v1(i32|i64)|v2i32|v4i16|v8i8)")>;
|
|
def : InstRW<[M5WriteNMUL3,
|
|
M5ReadNMULM1], (instregex "^(S|U|SQD)MULL(v4i32|v8i16|v16i8)")>;
|
|
def : InstRW<[M5WriteNDOT2], (instregex "^[SU]DOT(lane)?v")>;
|
|
def : InstRW<[M5WriteNHAD3], (instregex "^[SU]ADALPv")>;
|
|
def : InstRW<[M5WriteNSHT4A], (instregex "^[SU]R?SRA[dv]")>;
|
|
def : InstRW<[M5WriteNSHT2], (instregex "^SHL[dv]")>;
|
|
def : InstRW<[M5WriteNSHT2], (instregex "^S[LR]I[dv]")>;
|
|
def : InstRW<[M5WriteNSHT2], (instregex "^[SU]SH[LR][dv]")>;
|
|
def : InstRW<[M5WriteNSHT2], (instregex "^[SU]?SHLLv")>;
|
|
def : InstRW<[M5WriteNSHT4A], (instregex "^[SU]?Q?R?SHRU?N[bhsv]")>;
|
|
def : InstRW<[M5WriteNSHT4A], (instregex "^[SU]RSH[LR][dv]")>;
|
|
def : InstRW<[M5WriteNSHT4A], (instregex "^[SU]QR?SHLU?[bhsdv]")>;
|
|
|
|
// ASIMD FP instructions.
|
|
def : InstRW<[M5WriteNSHF2], (instregex "^FABSv.f(16|32|64)")>;
|
|
def : InstRW<[M5WriteFADD2], (instregex "^F(ABD|ADD|SUB)v.f(16|32|64)")>;
|
|
def : InstRW<[M5WriteFADD2], (instregex "^FADDPv.f(16|32|64)")>;
|
|
def : InstRW<[M5WriteNMSC1], (instregex "^F(AC|CM)(EQ|GE|GT|LE|LT)v[^1]")>;
|
|
def : InstRW<[M5WriteFCVT2], (instregex "^FCVT(L|N|XN)v")>;
|
|
def : InstRW<[M5WriteFCVT2A], (instregex "^FCVT[AMNPZ][SU]v")>;
|
|
def : InstRW<[M5WriteFCVT2], (instregex "^[SU]CVTFv.[fi](16|32|64)")>;
|
|
def : InstRW<[M5WriteFDIV7], (instrs FDIVv4f16)>;
|
|
def : InstRW<[M5WriteNEONV], (instrs FDIVv8f16)>;
|
|
def : InstRW<[M5WriteFDIV7], (instrs FDIVv2f32)>;
|
|
def : InstRW<[M5WriteNEONV], (instrs FDIVv4f32)>;
|
|
def : InstRW<[M5WriteNEONW], (instrs FDIVv2f64)>;
|
|
def : InstRW<[M5WriteNMSC1], (instregex "^F(MAX|MIN)(NM)?v")>;
|
|
def : InstRW<[M5WriteNMSC2], (instregex "^F(MAX|MIN)(NM)?Pv")>;
|
|
def : InstRW<[M5WriteNEONZ], (instregex "^F(MAX|MIN)(NM)?Vv")>;
|
|
def : InstRW<[M5WriteFMAC3], (instregex "^FMULX?v.[fi](16|32|64)")>;
|
|
def : InstRW<[M5WriteFMAC4,
|
|
M5ReadFMACM1], (instregex "^FML[AS]v.[fi](16|32|64)")>;
|
|
def : InstRW<[M5WriteNALU2], (instregex "^FNEGv.f(16|32|64)")>;
|
|
def : InstRW<[M5WriteFCVT3A], (instregex "^FRINT[AIMNPXZ]v")>;
|
|
def : InstRW<[M5WriteFSQR7], (instrs FSQRTv4f16)>;
|
|
def : InstRW<[M5WriteNEONU], (instrs FSQRTv8f16)>;
|
|
def : InstRW<[M5WriteFSQR8], (instrs FSQRTv2f32)>;
|
|
def : InstRW<[M5WriteNEONX], (instrs FSQRTv4f32)>;
|
|
def : InstRW<[M5WriteNEONY], (instrs FSQRTv2f64)>;
|
|
|
|
// ASIMD miscellaneous instructions.
|
|
def : InstRW<[M5WriteNALU2], (instregex "^RBITv")>;
|
|
def : InstRW<[M5WriteNALU2], (instregex "^(BIF|BIT|BSL|BSP)v")>;
|
|
def : InstRW<[M5WriteNALU2], (instregex "^CL[STZ]v")>;
|
|
def : InstRW<[M5WriteNEONB], (instregex "^DUPv.+gpr")>;
|
|
def : InstRW<[M5WriteNSHF2], (instregex "^CPY")>;
|
|
def : InstRW<[M5WriteNSHF2], (instregex "^DUPv.+lane")>;
|
|
def : InstRW<[M5WriteNSHF2], (instregex "^EXTv")>;
|
|
def : InstRW<[M5WriteNSHT4A], (instregex "^XTNv")>;
|
|
def : InstRW<[M5WriteNSHT4A], (instregex "^[SU]?QXTU?Nv")>;
|
|
def : InstRW<[M5WriteNEONB], (instregex "^INSv.+gpr")>;
|
|
def : InstRW<[M5WriteNSHF2], (instregex "^INSv.+lane")>;
|
|
def : InstRW<[M5WriteMOVI], (instregex "^(MOV|MVN)I")>;
|
|
def : InstRW<[M5WriteNALU1], (instregex "^FMOVv.f(16|32|64)")>;
|
|
def : InstRW<[M5WriteFCVT3], (instregex "^F(RECP|RSQRT)Ev[248]f(16|32|64)")>;
|
|
def : InstRW<[M5WriteFCVT3], (instregex "^U(RECP|RSQRT)Ev[24]i32")>;
|
|
def : InstRW<[M5WriteFMAC4], (instregex "^F(RECP|RSQRT)Sv.f(16|32|64)")>;
|
|
def : InstRW<[M5WriteNSHF2], (instregex "^REV(16|32|64)v")>;
|
|
def : InstRW<[M5WriteNSHFA], (instregex "^TB[LX]v(8|16)i8One")>;
|
|
def : InstRW<[M5WriteNSHFB], (instregex "^TB[LX]v(8|16)i8Two")>;
|
|
def : InstRW<[M5WriteNSHFC], (instregex "^TB[LX]v(8|16)i8Three")>;
|
|
def : InstRW<[M5WriteNSHFD], (instregex "^TB[LX]v(8|16)i8Four")>;
|
|
def : InstRW<[M5WriteNEONP], (instregex "^[SU]MOVv")>;
|
|
def : InstRW<[M5WriteNSHF2], (instregex "^(TRN|UZP|ZIP)[12]v")>;
|
|
|
|
// ASIMD load instructions.
|
|
def : InstRW<[WriteVLD], (instregex "LD1Onev(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[WriteVLD,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD1Onev(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVLDA], (instregex "LD1Twov(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[M5WriteVLDA,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD1Twov(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVLDB], (instregex "LD1Threev(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[M5WriteVLDB,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD1Threev(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVLDC], (instregex "LD1Fourv(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[M5WriteVLDC,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD1Fourv(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVLDD], (instregex "LD1i(8|16|32|64)$")>;
|
|
def : InstRW<[M5WriteVLDD,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD1i(8|16|32|64)_POST$")>;
|
|
def : InstRW<[WriteVLD], (instregex "LD1Rv(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[WriteVLD,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD1Rv(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVLDF], (instregex "LD2Twov(8b|16b|4h|8h|2s|4s|2d)$")>;
|
|
def : InstRW<[M5WriteVLDF,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD2Twov(8b|16b|4h|8h|2s|4s|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVLDG], (instregex "LD2i(8|16|32|64)$")>;
|
|
def : InstRW<[M5WriteVLDG,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD2i(8|16|32|64)_POST$")>;
|
|
def : InstRW<[M5WriteVLDA], (instregex "LD2Rv(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[M5WriteVLDA,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD2Rv(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVLDI], (instregex "LD3Threev(8b|16b|4h|8h|2s|4s|2d)$")>;
|
|
def : InstRW<[M5WriteVLDI,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD3Threev(8b|16b|4h|8h|2s|4s|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVLDJ], (instregex "LD3i(8|16|32)$")>;
|
|
def : InstRW<[M5WriteVLDJ,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD3i(8|16|32)_POST$")>;
|
|
def : InstRW<[M5WriteVLDL], (instregex "LD3i64$")>;
|
|
def : InstRW<[M5WriteVLDL,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD3i64_POST$")>;
|
|
def : InstRW<[M5WriteVLDB], (instregex "LD3Rv(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[M5WriteVLDB,
|
|
M5WriteA1X], (instregex "LD3Rv(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVLDN], (instregex "LD4Fourv(8b|16b|4h|8h|2s|4s|2d)$")>;
|
|
def : InstRW<[M5WriteVLDN,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD4Fourv(8b|16b|4h|8h|2s|4s|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVLDK], (instregex "LD4i(8|16|32)$")>;
|
|
def : InstRW<[M5WriteVLDK,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD4i(8|16|32)_POST$")>;
|
|
def : InstRW<[M5WriteVLDM], (instregex "LD4i64$")>;
|
|
def : InstRW<[M5WriteVLDM,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD4i64_POST$")>;
|
|
def : InstRW<[M5WriteVLDC], (instregex "LD4Rv(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[M5WriteVLDC,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "LD4Rv(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
|
|
// ASIMD store instructions.
|
|
def : InstRW<[WriteVST], (instregex "ST1Onev(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[WriteVST,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST1Onev(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVSTA], (instregex "ST1Twov(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[M5WriteVSTA,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST1Twov(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
|
|
def : InstRW<[M5WriteVSTB], (instregex "ST1Threev(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[M5WriteVSTB,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST1Threev(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVSTC], (instregex "ST1Fourv(8b|16b|4h|8h|2s|4s|1d|2d)$")>;
|
|
def : InstRW<[M5WriteVSTC,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST1Fourv(8b|16b|4h|8h|2s|4s|1d|2d)_POST$")>;
|
|
def : InstRW<[WriteVST], (instregex "ST1i(8|16|32|64)$")>;
|
|
def : InstRW<[WriteVST,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST1i(8|16|32|64)_POST$")>;
|
|
def : InstRW<[M5WriteVSTD], (instregex "ST2Twov(8b|4h|2s)$")>;
|
|
def : InstRW<[M5WriteVSTD,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST2Twov(8b|4h|2s)_POST$")>;
|
|
def : InstRW<[M5WriteVSTE], (instregex "ST2Twov(16b|8h|4s|2d)$")>;
|
|
def : InstRW<[M5WriteVSTE,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST2Twov(16b|8h|4s|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVSTD], (instregex "ST2i(8|16|32|64)$")>;
|
|
def : InstRW<[M5WriteVSTD,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST2i(8|16|32|64)_POST$")>;
|
|
def : InstRW<[M5WriteVSTF], (instregex "ST3Threev(8b|4h|2s)$")>;
|
|
def : InstRW<[M5WriteVSTF,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST3Threev(8b|4h|2s)_POST$")>;
|
|
def : InstRW<[M5WriteVSTG], (instregex "ST3Threev(16b|8h|4s|2d)$")>;
|
|
def : InstRW<[M5WriteVSTG,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST3Threev(16b|8h|4s|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVSTA], (instregex "ST3i(8|16|32|64)$")>;
|
|
def : InstRW<[M5WriteVSTA,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST3i(8|16|32|64)_POST$")>;
|
|
def : InstRW<[M5WriteVSTL], (instregex "ST4Fourv(8b|4h|2s)$")>;
|
|
def : InstRW<[M5WriteVSTL,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST4Fourv(8b|4h|2s)_POST$")>;
|
|
def : InstRW<[M5WriteVSTI], (instregex "ST4Fourv(16b|8h|4s|2d)$")>;
|
|
def : InstRW<[M5WriteVSTI,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST4Fourv(16b|8h|4s|2d)_POST$")>;
|
|
def : InstRW<[M5WriteVSTA], (instregex "ST4i(8|16|32|64)$")>;
|
|
def : InstRW<[M5WriteVSTA,
|
|
M5WriteA1X,
|
|
WriteAdr], (instregex "ST4i(8|16|32|64)_POST$")>;
|
|
|
|
// Cryptography instructions.
|
|
def : InstRW<[M5WriteNCRY2], (instregex "^AES[DE]")>;
|
|
def : InstRW<[M5WriteNCRY2,
|
|
M5ReadAESM2], (instregex "^AESI?MC")>;
|
|
def : InstRW<[M5WriteNCRY2A], (instregex "^PMULv")>;
|
|
def : InstRW<[M5WriteNCRY1A], (instregex "^PMULLv(1|8)i")>;
|
|
def : InstRW<[M5WriteNCRY3A], (instregex "^PMULLv(2|16)i")>;
|
|
def : InstRW<[M5WriteNCRY2A], (instregex "^SHA1(H|SU[01])")>;
|
|
def : InstRW<[M5WriteNCRY5A], (instregex "^SHA1[CMP]")>;
|
|
def : InstRW<[M5WriteNCRY2A], (instrs SHA256SU0rr)>;
|
|
def : InstRW<[M5WriteNCRY5A], (instrs SHA256SU1rrr)>;
|
|
def : InstRW<[M5WriteNCRY5A], (instregex "^SHA256H2?")>;
|
|
|
|
// CRC instructions.
|
|
def : InstRW<[M5WriteF2,
|
|
M5ReadFM1], (instregex "^CRC32C?[BHWX]")>;
|
|
|
|
} // SchedModel = ExynosM5Model
|