1//==- AArch64SchedCortexA55.td - ARM Cortex-A55 Scheduling Definitions -*- tablegen -*-=// 2// 3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4// See https://llvm.org/LICENSE.txt for license information. 5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6// 7//===----------------------------------------------------------------------===// 8// 9// This file defines the machine model for the ARM Cortex-A55 processors. 10// 11//===----------------------------------------------------------------------===// 12 13// ===---------------------------------------------------------------------===// 14// The following definitions describe the per-operand machine model. 15// This works with MachineScheduler. See MCSchedModel.h for details. 16 17// Cortex-A55 machine model for scheduling and other instruction cost heuristics. 18def CortexA55Model : SchedMachineModel { 19 let MicroOpBufferSize = 0; // The Cortex-A55 is an in-order processor 20 let IssueWidth = 2; // It dual-issues under most circumstances 21 let LoadLatency = 4; // Cycles for loads to access the cache. The 22 // optimisation guide shows that most loads have 23 // a latency of 3, but some have a latency of 4 24 // or 5. Setting it 4 looked to be good trade-off. 25 let MispredictPenalty = 8; // A branch direction mispredict. 26 let PostRAScheduler = 1; // Enable PostRA scheduler pass. 27 let CompleteModel = 0; // Covers instructions applicable to Cortex-A55. 28 29 list<Predicate> UnsupportedFeatures = [HasSVE]; 30 31 // FIXME: Remove when all errors have been fixed. 32 let FullInstRWOverlapCheck = 0; 33} 34 35//===----------------------------------------------------------------------===// 36// Define each kind of processor resource and number available. 37 38// Modeling each pipeline as a ProcResource using the BufferSize = 0 since the 39// Cortex-A55 is in-order. 40 41def CortexA55UnitALU : ProcResource<2> { let BufferSize = 0; } // Int ALU 42def CortexA55UnitMAC : ProcResource<1> { let BufferSize = 0; } // Int MAC, 64-bi wide 43def CortexA55UnitDiv : ProcResource<1> { let BufferSize = 0; } // Int Division, not pipelined 44def CortexA55UnitLd : ProcResource<1> { let BufferSize = 0; } // Load pipe 45def CortexA55UnitSt : ProcResource<1> { let BufferSize = 0; } // Store pipe 46def CortexA55UnitB : ProcResource<1> { let BufferSize = 0; } // Branch 47 48// The FP DIV/SQRT instructions execute totally differently from the FP ALU 49// instructions, which can mostly be dual-issued; that's why for now we model 50// them with 2 resources. 51def CortexA55UnitFPALU : ProcResource<2> { let BufferSize = 0; } // FP ALU 52def CortexA55UnitFPMAC : ProcResource<2> { let BufferSize = 0; } // FP MAC 53def CortexA55UnitFPDIV : ProcResource<1> { let BufferSize = 0; } // FP Div/SQRT, 64/128 54 55//===----------------------------------------------------------------------===// 56// Subtarget-specific SchedWrite types 57 58let SchedModel = CortexA55Model in { 59 60// These latencies are modeled without taking into account forwarding paths 61// (the software optimisation guide lists latencies taking into account 62// typical forwarding paths). 63def : WriteRes<WriteImm, [CortexA55UnitALU]> { let Latency = 3; } // MOVN, MOVZ 64def : WriteRes<WriteI, [CortexA55UnitALU]> { let Latency = 3; } // ALU 65def : WriteRes<WriteISReg, [CortexA55UnitALU]> { let Latency = 3; } // ALU of Shifted-Reg 66def : WriteRes<WriteIEReg, [CortexA55UnitALU]> { let Latency = 3; } // ALU of Extended-Reg 67def : WriteRes<WriteExtr, [CortexA55UnitALU]> { let Latency = 3; } // EXTR from a reg pair 68def : WriteRes<WriteIS, [CortexA55UnitALU]> { let Latency = 3; } // Shift/Scale 69 70// MAC 71def : WriteRes<WriteIM32, [CortexA55UnitMAC]> { let Latency = 4; } // 32-bit Multiply 72def : WriteRes<WriteIM64, [CortexA55UnitMAC]> { let Latency = 4; } // 64-bit Multiply 73 74// Div 75def : WriteRes<WriteID32, [CortexA55UnitDiv]> { 76 let Latency = 8; let ResourceCycles = [8]; 77} 78def : WriteRes<WriteID64, [CortexA55UnitDiv]> { 79 let Latency = 8; let ResourceCycles = [8]; 80} 81 82// Load 83def : WriteRes<WriteLD, [CortexA55UnitLd]> { let Latency = 3; } 84def : WriteRes<WriteLDIdx, [CortexA55UnitLd]> { let Latency = 4; } 85def : WriteRes<WriteLDHi, [CortexA55UnitLd]> { let Latency = 5; } 86 87// Vector Load - Vector loads take 1-5 cycles to issue. For the WriteVecLd 88// below, choosing the median of 3 which makes the latency 6. 89// An extra cycle is needed to get the swizzling right. 90def : WriteRes<WriteVLD, [CortexA55UnitLd]> { let Latency = 6; 91 let ResourceCycles = [3]; } 92def CortexA55WriteVLD1 : SchedWriteRes<[CortexA55UnitLd]> { let Latency = 4; } 93def CortexA55WriteVLD2 : SchedWriteRes<[CortexA55UnitLd]> { let Latency = 5; 94 let ResourceCycles = [2]; } 95def CortexA55WriteVLD3 : SchedWriteRes<[CortexA55UnitLd]> { let Latency = 6; 96 let ResourceCycles = [3]; } 97def CortexA55WriteVLD4 : SchedWriteRes<[CortexA55UnitLd]> { let Latency = 7; 98 let ResourceCycles = [4]; } 99def CortexA55WriteVLD5 : SchedWriteRes<[CortexA55UnitLd]> { let Latency = 8; 100 let ResourceCycles = [5]; } 101def CortexA55WriteVLD6 : SchedWriteRes<[CortexA55UnitLd]> { let Latency = 9; 102 let ResourceCycles = [6]; } 103def CortexA55WriteVLD7 : SchedWriteRes<[CortexA55UnitLd]> { let Latency = 10; 104 let ResourceCycles = [7]; } 105def CortexA55WriteVLD8 : SchedWriteRes<[CortexA55UnitLd]> { let Latency = 11; 106 let ResourceCycles = [8]; } 107 108// Pre/Post Indexing - Performed as part of address generation 109def : WriteRes<WriteAdr, []> { let Latency = 0; } 110 111// Store 112def : WriteRes<WriteST, [CortexA55UnitSt]> { let Latency = 4; } 113def : WriteRes<WriteSTP, [CortexA55UnitSt]> { let Latency = 4; } 114def : WriteRes<WriteSTIdx, [CortexA55UnitSt]> { let Latency = 4; } 115def : WriteRes<WriteSTX, [CortexA55UnitSt]> { let Latency = 4; } 116 117// Vector Store - Similar to vector loads, can take 1-3 cycles to issue. 118def : WriteRes<WriteVST, [CortexA55UnitSt]> { let Latency = 5; 119 let ResourceCycles = [2];} 120def CortexA55WriteVST1 : SchedWriteRes<[CortexA55UnitSt]> { let Latency = 4; } 121def CortexA55WriteVST2 : SchedWriteRes<[CortexA55UnitSt]> { let Latency = 5; 122 let ResourceCycles = [2]; } 123def CortexA55WriteVST3 : SchedWriteRes<[CortexA55UnitSt]> { let Latency = 6; 124 let ResourceCycles = [3]; } 125def CortexA55WriteVST4 : SchedWriteRes<[CortexA55UnitSt]> { let Latency = 5; 126 let ResourceCycles = [4]; } 127 128def : WriteRes<WriteAtomic, []> { let Unsupported = 1; } 129 130// Branch 131def : WriteRes<WriteBr, [CortexA55UnitB]>; 132def : WriteRes<WriteBrReg, [CortexA55UnitB]>; 133def : WriteRes<WriteSys, [CortexA55UnitB]>; 134def : WriteRes<WriteBarrier, [CortexA55UnitB]>; 135def : WriteRes<WriteHint, [CortexA55UnitB]>; 136 137// FP ALU 138// As WriteF result is produced in F5 and it can be mostly forwarded 139// to consumer at F1, the effectively latency is set as 4. 140def : WriteRes<WriteF, [CortexA55UnitFPALU]> { let Latency = 4; } 141def : WriteRes<WriteFCmp, [CortexA55UnitFPALU]> { let Latency = 3; } 142def : WriteRes<WriteFCvt, [CortexA55UnitFPALU]> { let Latency = 4; } 143def : WriteRes<WriteFCopy, [CortexA55UnitFPALU]> { let Latency = 3; } 144def : WriteRes<WriteFImm, [CortexA55UnitFPALU]> { let Latency = 3; } 145def : WriteRes<WriteV, [CortexA55UnitFPALU]> { let Latency = 4; } 146 147// FP ALU specific new schedwrite definitions 148def CortexA55WriteFPALU_F3 : SchedWriteRes<[CortexA55UnitFPALU]> { let Latency = 3;} 149def CortexA55WriteFPALU_F4 : SchedWriteRes<[CortexA55UnitFPALU]> { let Latency = 4;} 150def CortexA55WriteFPALU_F5 : SchedWriteRes<[CortexA55UnitFPALU]> { let Latency = 5;} 151 152// FP Mul, Div, Sqrt. Div/Sqrt are not pipelined 153def : WriteRes<WriteFMul, [CortexA55UnitFPMAC]> { let Latency = 4; } 154def : WriteRes<WriteFDiv, [CortexA55UnitFPDIV]> { let Latency = 22; 155 let ResourceCycles = [29]; } 156def CortexA55WriteFMAC : SchedWriteRes<[CortexA55UnitFPMAC]> { let Latency = 4; } 157def CortexA55WriteFDivHP : SchedWriteRes<[CortexA55UnitFPDIV]> { let Latency = 8; 158 let ResourceCycles = [5]; } 159def CortexA55WriteFDivSP : SchedWriteRes<[CortexA55UnitFPDIV]> { let Latency = 13; 160 let ResourceCycles = [10]; } 161def CortexA55WriteFDivDP : SchedWriteRes<[CortexA55UnitFPDIV]> { let Latency = 22; 162 let ResourceCycles = [19]; } 163def CortexA55WriteFSqrtHP : SchedWriteRes<[CortexA55UnitFPDIV]> { let Latency = 8; 164 let ResourceCycles = [5]; } 165def CortexA55WriteFSqrtSP : SchedWriteRes<[CortexA55UnitFPDIV]> { let Latency = 12; 166 let ResourceCycles = [9]; } 167def CortexA55WriteFSqrtDP : SchedWriteRes<[CortexA55UnitFPDIV]> { let Latency = 22; 168 let ResourceCycles = [19]; } 169 170//===----------------------------------------------------------------------===// 171// Subtarget-specific SchedRead types. 172 173def : ReadAdvance<ReadVLD, 0>; 174def : ReadAdvance<ReadExtrHi, 1>; 175def : ReadAdvance<ReadAdrBase, 1>; 176 177// ALU - ALU input operands are generally needed in EX1. An operand produced in 178// in say EX2 can be forwarded for consumption to ALU in EX1, thereby 179// allowing back-to-back ALU operations such as add. If an operand requires 180// a shift, it will, however, be required in ISS stage. 181def : ReadAdvance<ReadI, 2, [WriteImm,WriteI, 182 WriteISReg, WriteIEReg,WriteIS, 183 WriteID32,WriteID64, 184 WriteIM32,WriteIM64]>; 185// Shifted operand 186def CortexA55ReadShifted : SchedReadAdvance<1, [WriteImm,WriteI, 187 WriteISReg, WriteIEReg,WriteIS, 188 WriteID32,WriteID64, 189 WriteIM32,WriteIM64]>; 190def CortexA55ReadNotShifted : SchedReadAdvance<2, [WriteImm,WriteI, 191 WriteISReg, WriteIEReg,WriteIS, 192 WriteID32,WriteID64, 193 WriteIM32,WriteIM64]>; 194def CortexA55ReadISReg : SchedReadVariant<[ 195 SchedVar<RegShiftedPred, [CortexA55ReadShifted]>, 196 SchedVar<NoSchedPred, [CortexA55ReadNotShifted]>]>; 197def : SchedAlias<ReadISReg, CortexA55ReadISReg>; 198 199def CortexA55ReadIEReg : SchedReadVariant<[ 200 SchedVar<RegExtendedPred, [CortexA55ReadShifted]>, 201 SchedVar<NoSchedPred, [CortexA55ReadNotShifted]>]>; 202def : SchedAlias<ReadIEReg, CortexA55ReadIEReg>; 203 204// MUL 205def : ReadAdvance<ReadIM, 1, [WriteImm,WriteI, 206 WriteISReg, WriteIEReg,WriteIS, 207 WriteID32,WriteID64, 208 WriteIM32,WriteIM64]>; 209def : ReadAdvance<ReadIMA, 2, [WriteImm,WriteI, 210 WriteISReg, WriteIEReg,WriteIS, 211 WriteID32,WriteID64, 212 WriteIM32,WriteIM64]>; 213 214// Div 215def : ReadAdvance<ReadID, 1, [WriteImm,WriteI, 216 WriteISReg, WriteIEReg,WriteIS, 217 WriteID32,WriteID64, 218 WriteIM32,WriteIM64]>; 219 220//===----------------------------------------------------------------------===// 221// Subtarget-specific InstRWs. 222 223//--- 224// Miscellaneous 225//--- 226def : InstRW<[CortexA55WriteVLD2,CortexA55WriteVLD1], (instregex "LDP.*")>; 227def : InstRW<[WriteI], (instrs COPY)>; 228//--- 229// Vector Loads - 64-bit per cycle 230//--- 231// 1-element structures 232def : InstRW<[CortexA55WriteVLD1], (instregex "LD1i(8|16|32|64)$")>; // single element 233def : InstRW<[CortexA55WriteVLD1], (instregex "LD1Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>; // replicate 234def : InstRW<[CortexA55WriteVLD1], (instregex "LD1Onev(8b|4h|2s|1d)$")>; 235def : InstRW<[CortexA55WriteVLD2], (instregex "LD1Onev(16b|8h|4s|2d)$")>; 236def : InstRW<[CortexA55WriteVLD2], (instregex "LD1Twov(8b|4h|2s|1d)$")>; // multiple structures 237def : InstRW<[CortexA55WriteVLD4], (instregex "LD1Twov(16b|8h|4s|2d)$")>; 238def : InstRW<[CortexA55WriteVLD3], (instregex "LD1Threev(8b|4h|2s|1d)$")>; 239def : InstRW<[CortexA55WriteVLD6], (instregex "LD1Threev(16b|8h|4s|2d)$")>; 240def : InstRW<[CortexA55WriteVLD4], (instregex "LD1Fourv(8b|4h|2s|1d)$")>; 241def : InstRW<[CortexA55WriteVLD8], (instregex "LD1Fourv(16b|8h|4s|2d)$")>; 242 243def : InstRW<[CortexA55WriteVLD1, WriteAdr], (instregex "LD1i(8|16|32|64)_POST$")>; 244def : InstRW<[CortexA55WriteVLD1, WriteAdr], (instregex "LD1Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>; 245def : InstRW<[CortexA55WriteVLD1, WriteAdr], (instregex "LD1Onev(8b|4h|2s|1d)_POST$")>; 246def : InstRW<[CortexA55WriteVLD2, WriteAdr], (instregex "LD1Onev(16b|8h|4s|2d)_POST$")>; 247def : InstRW<[CortexA55WriteVLD2, WriteAdr], (instregex "LD1Twov(8b|4h|2s|1d)_POST$")>; 248def : InstRW<[CortexA55WriteVLD4, WriteAdr], (instregex "LD1Twov(16b|8h|4s|2d)_POST$")>; 249def : InstRW<[CortexA55WriteVLD3, WriteAdr], (instregex "LD1Threev(8b|4h|2s|1d)_POST$")>; 250def : InstRW<[CortexA55WriteVLD6, WriteAdr], (instregex "LD1Threev(16b|8h|4s|2d)_POST$")>; 251def : InstRW<[CortexA55WriteVLD4, WriteAdr], (instregex "LD1Fourv(8b|4h|2s|1d)_POST$")>; 252def : InstRW<[CortexA55WriteVLD8, WriteAdr], (instregex "LD1Fourv(16b|8h|4s|2d)_POST$")>; 253 254// 2-element structures 255def : InstRW<[CortexA55WriteVLD2], (instregex "LD2i(8|16|32|64)$")>; 256def : InstRW<[CortexA55WriteVLD2], (instregex "LD2Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>; 257def : InstRW<[CortexA55WriteVLD2], (instregex "LD2Twov(8b|4h|2s)$")>; 258def : InstRW<[CortexA55WriteVLD4], (instregex "LD2Twov(16b|8h|4s|2d)$")>; 259 260def : InstRW<[CortexA55WriteVLD2, WriteAdr], (instregex "LD2i(8|16|32|64)(_POST)?$")>; 261def : InstRW<[CortexA55WriteVLD2, WriteAdr], (instregex "LD2Rv(8b|4h|2s|1d|16b|8h|4s|2d)(_POST)?$")>; 262def : InstRW<[CortexA55WriteVLD2, WriteAdr], (instregex "LD2Twov(8b|4h|2s)(_POST)?$")>; 263def : InstRW<[CortexA55WriteVLD4, WriteAdr], (instregex "LD2Twov(16b|8h|4s|2d)(_POST)?$")>; 264 265// 3-element structures 266def : InstRW<[CortexA55WriteVLD2], (instregex "LD3i(8|16|32|64)$")>; 267def : InstRW<[CortexA55WriteVLD2], (instregex "LD3Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>; 268def : InstRW<[CortexA55WriteVLD3], (instregex "LD3Threev(8b|4h|2s|1d)$")>; 269def : InstRW<[CortexA55WriteVLD6], (instregex "LD3Threev(16b|8h|4s|2d)$")>; 270 271def : InstRW<[CortexA55WriteVLD2, WriteAdr], (instregex "LD3i(8|16|32|64)_POST$")>; 272def : InstRW<[CortexA55WriteVLD2, WriteAdr], (instregex "LD3Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>; 273def : InstRW<[CortexA55WriteVLD3, WriteAdr], (instregex "LD3Threev(8b|4h|2s|1d)_POST$")>; 274def : InstRW<[CortexA55WriteVLD6, WriteAdr], (instregex "LD3Threev(16b|8h|4s|2d)_POST$")>; 275 276// 4-element structures 277def : InstRW<[CortexA55WriteVLD2], (instregex "LD4i(8|16|32|64)$")>; // load single 4-el structure to one lane of 4 regs. 278def : InstRW<[CortexA55WriteVLD2], (instregex "LD4Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>; // load single 4-el structure, replicate to all lanes of 4 regs. 279def : InstRW<[CortexA55WriteVLD4], (instregex "LD4Fourv(8b|4h|2s|1d)$")>; // load multiple 4-el structures to 4 regs. 280def : InstRW<[CortexA55WriteVLD8], (instregex "LD4Fourv(16b|8h|4s|2d)$")>; 281 282def : InstRW<[CortexA55WriteVLD2, WriteAdr], (instregex "LD4i(8|16|32|64)_POST$")>; 283def : InstRW<[CortexA55WriteVLD2, WriteAdr], (instregex "LD4Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>; 284def : InstRW<[CortexA55WriteVLD4, WriteAdr], (instregex "LD4Fourv(8b|4h|2s|1d)_POST$")>; 285def : InstRW<[CortexA55WriteVLD8, WriteAdr], (instregex "LD4Fourv(16b|8h|4s|2d)_POST$")>; 286 287//--- 288// Vector Stores 289//--- 290def : InstRW<[CortexA55WriteVST1], (instregex "ST1i(8|16|32|64)$")>; 291def : InstRW<[CortexA55WriteVST1], (instregex "ST1Onev(8b|4h|2s|1d|16b|8h|4s|2d)$")>; 292def : InstRW<[CortexA55WriteVST1], (instregex "ST1Twov(8b|4h|2s|1d|16b|8h|4s|2d)$")>; 293def : InstRW<[CortexA55WriteVST2], (instregex "ST1Threev(8b|4h|2s|1d|16b|8h|4s|2d)$")>; 294def : InstRW<[CortexA55WriteVST4], (instregex "ST1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)$")>; 295def : InstRW<[CortexA55WriteVST1, WriteAdr], (instregex "ST1i(8|16|32|64)_POST$")>; 296def : InstRW<[CortexA55WriteVST1, WriteAdr], (instregex "ST1Onev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>; 297def : InstRW<[CortexA55WriteVST1, WriteAdr], (instregex "ST1Twov(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>; 298def : InstRW<[CortexA55WriteVST2, WriteAdr], (instregex "ST1Threev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>; 299def : InstRW<[CortexA55WriteVST4, WriteAdr], (instregex "ST1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>; 300 301def : InstRW<[CortexA55WriteVST2], (instregex "ST2i(8|16|32|64)$")>; 302def : InstRW<[CortexA55WriteVST2], (instregex "ST2Twov(8b|4h|2s)$")>; 303def : InstRW<[CortexA55WriteVST4], (instregex "ST2Twov(16b|8h|4s|2d)$")>; 304def : InstRW<[CortexA55WriteVST2, WriteAdr], (instregex "ST2i(8|16|32|64)_POST$")>; 305def : InstRW<[CortexA55WriteVST2, WriteAdr], (instregex "ST2Twov(8b|4h|2s)_POST$")>; 306def : InstRW<[CortexA55WriteVST4, WriteAdr], (instregex "ST2Twov(16b|8h|4s|2d)_POST$")>; 307 308def : InstRW<[CortexA55WriteVST2], (instregex "ST3i(8|16|32|64)$")>; 309def : InstRW<[CortexA55WriteVST4], (instregex "ST3Threev(8b|4h|2s|1d|16b|8h|4s|2d)$")>; 310def : InstRW<[CortexA55WriteVST2, WriteAdr], (instregex "ST3i(8|16|32|64)_POST$")>; 311def : InstRW<[CortexA55WriteVST4, WriteAdr], (instregex "ST3Threev(8b|4h|2s|1d|2d|16b|8h|4s|4d)_POST$")>; 312 313def : InstRW<[CortexA55WriteVST2], (instregex "ST4i(8|16|32|64)$")>; 314def : InstRW<[CortexA55WriteVST4], (instregex "ST4Fourv(8b|4h|2s|1d|16b|8h|4s|2d)$")>; 315def : InstRW<[CortexA55WriteVST2, WriteAdr], (instregex "ST4i(8|16|32|64)_POST$")>; 316def : InstRW<[CortexA55WriteVST4, WriteAdr], (instregex "ST4Fourv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>; 317 318//--- 319// Floating Point Conversions, MAC, DIV, SQRT 320//--- 321def : InstRW<[CortexA55WriteFPALU_F3], (instregex "^FCVT[ALMNPZ][SU](S|U)?(W|X)")>; 322def : InstRW<[CortexA55WriteFPALU_F4], (instregex "^FCVT(X)?[ALMNPXZ](S|U|N)?v")>; 323 324def : InstRW<[CortexA55WriteFPALU_F4], (instregex "^(S|U)CVTF(S|U)(W|X)(H|S|D)")>; 325def : InstRW<[CortexA55WriteFPALU_F4], (instregex "^(S|U)CVTF(h|s|d)")>; 326def : InstRW<[CortexA55WriteFPALU_F4], (instregex "^(S|U)CVTFv")>; 327 328def : InstRW<[CortexA55WriteFMAC], (instregex "^FN?M(ADD|SUB).*")>; 329def : InstRW<[CortexA55WriteFMAC], (instregex "^FML(A|S).*")>; 330def : InstRW<[CortexA55WriteFDivHP], (instrs FDIVHrr)>; 331def : InstRW<[CortexA55WriteFDivSP], (instrs FDIVSrr)>; 332def : InstRW<[CortexA55WriteFDivDP], (instrs FDIVDrr)>; 333def : InstRW<[CortexA55WriteFDivHP], (instregex "^FDIVv.*16$")>; 334def : InstRW<[CortexA55WriteFDivSP], (instregex "^FDIVv.*32$")>; 335def : InstRW<[CortexA55WriteFDivDP], (instregex "^FDIVv.*64$")>; 336def : InstRW<[CortexA55WriteFSqrtHP], (instregex "^.*SQRT.*16$")>; 337def : InstRW<[CortexA55WriteFSqrtSP], (instregex "^.*SQRT.*32$")>; 338def : InstRW<[CortexA55WriteFSqrtDP], (instregex "^.*SQRT.*64$")>; 339} 340