1/*************************************************************************************** 2* Copyright (c) 2020-2021 Institute of Computing Technology, Chinese Academy of Sciences 3* Copyright (c) 2020-2021 Peng Cheng Laboratory 4* 5* XiangShan is licensed under Mulan PSL v2. 6* You can use this software according to the terms and conditions of the Mulan PSL v2. 7* You may obtain a copy of Mulan PSL v2 at: 8* http://license.coscl.org.cn/MulanPSL2 9* 10* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, 11* EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, 12* MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE. 13* 14* See the Mulan PSL v2 for more details. 15***************************************************************************************/ 16 17package xiangshan.backend.rename 18 19import chipsalliance.rocketchip.config.Parameters 20import chisel3._ 21import chisel3.util._ 22import xiangshan._ 23import utils._ 24import xiangshan.backend.decode.{FusionDecodeInfo, Imm_I, Imm_LUI_LOAD, Imm_U} 25import xiangshan.backend.rob.RobPtr 26import xiangshan.backend.rename.freelist._ 27import xiangshan.mem.mdp._ 28 29class Rename(implicit p: Parameters) extends XSModule with HasCircularQueuePtrHelper with HasPerfEvents { 30 val io = IO(new Bundle() { 31 val redirect = Flipped(ValidIO(new Redirect)) 32 val robCommits = Input(new RobCommitIO) 33 // from decode 34 val in = Vec(RenameWidth, Flipped(DecoupledIO(new CfCtrl))) 35 val fusionInfo = Vec(DecodeWidth - 1, Flipped(new FusionDecodeInfo)) 36 // ssit read result 37 val ssit = Flipped(Vec(RenameWidth, Output(new SSITEntry))) 38 // waittable read result 39 val waittable = Flipped(Vec(RenameWidth, Output(Bool()))) 40 // to rename table 41 val intReadPorts = Vec(RenameWidth, Vec(3, Input(UInt(PhyRegIdxWidth.W)))) 42 val fpReadPorts = Vec(RenameWidth, Vec(4, Input(UInt(PhyRegIdxWidth.W)))) 43 val vecReadPorts = Vec(RenameWidth, Vec(4, Input(UInt(PhyRegIdxWidth.W)))) 44 val intRenamePorts = Vec(RenameWidth, Output(new RatWritePort)) 45 val fpRenamePorts = Vec(RenameWidth, Output(new RatWritePort)) 46 val vecRenamePorts = Vec(RenameWidth, Output(new RatWritePort)) 47 // to dispatch1 48 val out = Vec(RenameWidth, DecoupledIO(new MicroOp)) 49 // debug arch ports 50 val debug_int_rat = Vec(32, Input(UInt(PhyRegIdxWidth.W))) 51 val debug_fp_rat = Vec(32, Input(UInt(PhyRegIdxWidth.W))) 52 }) 53 54 // create free list and rat 55 val intFreeList = Module(new MEFreeList(NRPhyRegs)) 56 val intRefCounter = Module(new RefCounter(NRPhyRegs)) 57 val fpFreeList = Module(new StdFreeList(NRPhyRegs - 32)) 58 59 intRefCounter.io.commit <> io.robCommits 60 intRefCounter.io.redirect := io.redirect.valid 61 intRefCounter.io.debug_int_rat <> io.debug_int_rat 62 intFreeList.io.commit <> io.robCommits 63 intFreeList.io.debug_rat <> io.debug_int_rat 64 fpFreeList.io.commit <> io.robCommits 65 fpFreeList.io.debug_rat <> io.debug_fp_rat 66 67 object RegType extends Enumeration { val Reg_I, Reg_F, Reg_V = Value } 68 import RegType._ 69 type RegType = RegType.Value 70 71 // decide if given instruction needs allocating a new physical register (CfCtrl: from decode; RobCommitInfo: from rob) 72 // fp and vec share `fpFreeList` 73 def needDestReg[T <: CfCtrl](int: Boolean, x: T): Bool = { 74 if (int) x.ctrl.rfWen && x.ctrl.ldest =/= 0.U else x.ctrl.fpWen || x.ctrl.vecWen 75 } 76 def needDestReg[T <: CfCtrl](reg_t: RegType, x: T): Bool = reg_t match { 77 case Reg_I => x.ctrl.rfWen && x.ctrl.ldest =/= 0.U 78 case Reg_F => x.ctrl.fpWen 79 case Reg_V => x.ctrl.vecWen 80 } 81 def needDestRegCommit[T <: RobCommitInfo](int: Boolean, x: T): Bool = { 82 if (int) x.rfWen else x.fpWen || x.vecWen 83 } 84 def needDestRegWalk[T <: RobCommitInfo](int: Boolean, x: T): Bool = { 85 if(int) x.rfWen && x.ldest =/= 0.U else x.fpWen || x.vecWen 86 } 87 88 // connect [redirect + walk] ports for __float point__ & __integer__ free list 89 Seq(fpFreeList, intFreeList).foreach { case fl => 90 fl.io.redirect := io.redirect.valid 91 fl.io.walk := io.robCommits.isWalk 92 } 93 // only when both fp and int free list and dispatch1 has enough space can we do allocation 94 // when isWalk, freelist can definitely allocate 95 intFreeList.io.doAllocate := fpFreeList.io.canAllocate && io.out(0).ready || io.robCommits.isWalk 96 fpFreeList.io.doAllocate := intFreeList.io.canAllocate && io.out(0).ready || io.robCommits.isWalk 97 98 // dispatch1 ready ++ float point free list ready ++ int free list ready ++ not walk 99 val canOut = io.out(0).ready && fpFreeList.io.canAllocate && intFreeList.io.canAllocate && !io.robCommits.isWalk 100 101 102 // speculatively assign the instruction with an robIdx 103 val validCount = PopCount(io.in.map(_.valid)) // number of instructions waiting to enter rob (from decode) 104 val robIdxHead = RegInit(0.U.asTypeOf(new RobPtr)) 105 val lastCycleMisprediction = RegNext(io.redirect.valid && !io.redirect.bits.flushItself()) 106 val robIdxHeadNext = Mux(io.redirect.valid, io.redirect.bits.robIdx, // redirect: move ptr to given rob index 107 Mux(lastCycleMisprediction, robIdxHead + 1.U, // mis-predict: not flush robIdx itself 108 Mux(canOut, robIdxHead + validCount, // instructions successfully entered next stage: increase robIdx 109 /* default */ robIdxHead))) // no instructions passed by this cycle: stick to old value 110 robIdxHead := robIdxHeadNext 111 112 /** 113 * Rename: allocate free physical register and update rename table 114 */ 115 val uops = Wire(Vec(RenameWidth, new MicroOp)) 116 uops.foreach( uop => { 117 uop.srcState(0) := DontCare 118 uop.srcState(1) := DontCare 119 uop.srcState(2) := DontCare 120 uop.robIdx := DontCare 121 uop.debugInfo := DontCare 122 uop.lqIdx := DontCare 123 uop.sqIdx := DontCare 124 }) 125 126 require(RenameWidth >= CommitWidth) 127 val needVecDest = Wire(Vec(RenameWidth, Bool())) 128 val needFpDest = Wire(Vec(RenameWidth, Bool())) 129 val needIntDest = Wire(Vec(RenameWidth, Bool())) 130 val needNotIntDest = Wire(Vec(RenameWidth, Bool())) 131 val hasValid = Cat(io.in.map(_.valid)).orR 132 133 val isMove = io.in.map(_.bits.ctrl.isMove) 134 135 val walkNeedNotIntDest = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B))) 136 val walkNeedIntDest = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B))) 137 val walkIsMove = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B))) 138 139 val intSpecWen = Wire(Vec(RenameWidth, Bool())) 140 val fpSpecWen = Wire(Vec(RenameWidth, Bool())) 141 val vecSpecWen = Wire(Vec(RenameWidth, Bool())) 142 143 val walkIntSpecWen = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B))) 144 145 val walkPdest = Wire(Vec(RenameWidth, UInt(PhyRegIdxWidth.W))) 146 147 // uop calculation 148 for (i <- 0 until RenameWidth) { 149 uops(i).cf := io.in(i).bits.cf 150 uops(i).ctrl := io.in(i).bits.ctrl 151 152 // update cf according to ssit result 153 uops(i).cf.storeSetHit := io.ssit(i).valid 154 uops(i).cf.loadWaitStrict := io.ssit(i).strict && io.ssit(i).valid 155 uops(i).cf.ssid := io.ssit(i).ssid 156 157 // update cf according to waittable result 158 uops(i).cf.loadWaitBit := io.waittable(i) 159 160 // alloc a new phy reg, fp and vec share the `fpFreeList` 161 needVecDest (i) := io.in(i).valid && needDestReg(Reg_V, io.in(i).bits) 162 needFpDest (i) := io.in(i).valid && needDestReg(Reg_F, io.in(i).bits) 163 needIntDest (i) := io.in(i).valid && needDestReg(Reg_I, io.in(i).bits) 164 needNotIntDest(i) := io.in(i).valid && needDestReg(int = false, io.in(i).bits) 165 if (i < CommitWidth) { 166 walkNeedNotIntDest(i) := io.robCommits.walkValid(i) && needDestRegWalk(int = false, io.robCommits.info(i)) 167 walkNeedIntDest(i) := io.robCommits.walkValid(i) && needDestRegWalk(int = true, io.robCommits.info(i)) 168 walkIsMove(i) := io.robCommits.info(i).isMove 169 } 170 fpFreeList.io.allocateReq(i) := Mux(io.robCommits.isWalk, walkNeedNotIntDest(i), needNotIntDest(i)) 171 intFreeList.io.allocateReq(i) := Mux(io.robCommits.isWalk, walkNeedIntDest(i) && !walkIsMove(i), needIntDest(i) && !isMove(i)) 172 173 // no valid instruction from decode stage || all resources (dispatch1 + both free lists) ready 174 io.in(i).ready := !hasValid || canOut 175 176 uops(i).robIdx := robIdxHead + PopCount(io.in.take(i).map(_.valid)) 177 178 uops(i).psrc(0) := Mux1H(uops(i).ctrl.srcType(0), Seq(io.intReadPorts(i)(0), io.fpReadPorts(i)(0), io.vecReadPorts(i)(0))) 179 uops(i).psrc(1) := Mux1H(uops(i).ctrl.srcType(1), Seq(io.intReadPorts(i)(1), io.fpReadPorts(i)(1), io.vecReadPorts(i)(1))) 180 // int psrc2 should be bypassed from next instruction if it is fused 181 if (i < RenameWidth - 1) { 182 when (io.fusionInfo(i).rs2FromRs2 || io.fusionInfo(i).rs2FromRs1) { 183 uops(i).psrc(1) := Mux(io.fusionInfo(i).rs2FromRs2, io.intReadPorts(i + 1)(1), io.intReadPorts(i + 1)(0)) 184 }.elsewhen(io.fusionInfo(i).rs2FromZero) { 185 uops(i).psrc(1) := 0.U 186 } 187 } 188 uops(i).psrc(2) := Mux1H(uops(i).ctrl.srcType(2)(2, 1), Seq(io.fpReadPorts(i)(2), io.vecReadPorts(i)(2))) 189 uops(i).old_pdest := Mux1H(Seq( 190 uops(i).ctrl.rfWen -> io.intReadPorts(i).last, 191 uops(i).ctrl.fpWen -> io.fpReadPorts (i).last, 192 uops(i).ctrl.vecWen -> io.vecReadPorts(i).last 193 )) 194 uops(i).eliminatedMove := isMove(i) 195 196 // update pdest 197 uops(i).pdest := Mux(needIntDest(i), intFreeList.io.allocatePhyReg(i), // normal int inst 198 // normal fp inst 199 Mux(needNotIntDest(i), fpFreeList.io.allocatePhyReg(i), 200 /* default */0.U)) 201 202 // Assign performance counters 203 uops(i).debugInfo.renameTime := GTimer() 204 205 io.out(i).valid := io.in(i).valid && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && !io.robCommits.isWalk 206 io.out(i).bits := uops(i) 207 // dirty code for fence. The lsrc is passed by imm. 208 when (io.out(i).bits.ctrl.fuType === FuType.fence) { 209 io.out(i).bits.ctrl.imm := Cat(io.in(i).bits.ctrl.lsrc(1), io.in(i).bits.ctrl.lsrc(0)) 210 } 211 // dirty code for SoftPrefetch (prefetch.r/prefetch.w) 212 when (io.in(i).bits.ctrl.isSoftPrefetch) { 213 io.out(i).bits.ctrl.fuType := FuType.ldu 214 io.out(i).bits.ctrl.fuOpType := Mux(io.in(i).bits.ctrl.lsrc(1) === 1.U, LSUOpType.prefetch_r, LSUOpType.prefetch_w) 215 io.out(i).bits.ctrl.selImm := SelImm.IMM_S 216 io.out(i).bits.ctrl.imm := Cat(io.in(i).bits.ctrl.imm(io.in(i).bits.ctrl.imm.getWidth - 1, 5), 0.U(5.W)) 217 } 218 219 // write speculative rename table 220 // we update rat later inside commit code 221 intSpecWen(i) := needIntDest(i) && intFreeList.io.canAllocate && intFreeList.io.doAllocate && !io.robCommits.isWalk && !io.redirect.valid 222 fpSpecWen(i) := needFpDest(i) && fpFreeList.io.canAllocate && fpFreeList.io.doAllocate && !io.robCommits.isWalk && !io.redirect.valid 223 vecSpecWen(i) := needVecDest(i) && fpFreeList.io.canAllocate && fpFreeList.io.doAllocate && !io.robCommits.isWalk && !io.redirect.valid 224 225 if (i < CommitWidth) { 226 walkIntSpecWen(i) := walkNeedIntDest(i) && !io.redirect.valid 227 walkPdest(i) := io.robCommits.info(i).pdest 228 } else { 229 walkPdest(i) := io.out(i).bits.pdest 230 } 231 232 intRefCounter.io.allocate(i).valid := Mux(io.robCommits.isWalk, walkIntSpecWen(i), intSpecWen(i)) 233 intRefCounter.io.allocate(i).bits := Mux(io.robCommits.isWalk, walkPdest(i), io.out(i).bits.pdest) 234 } 235 236 /** 237 * How to set psrc: 238 * - bypass the pdest to psrc if previous instructions write to the same ldest as lsrc 239 * - default: psrc from RAT 240 * How to set pdest: 241 * - Mux(isMove, psrc, pdest_from_freelist). 242 * 243 * The critical path of rename lies here: 244 * When move elimination is enabled, we need to update the rat with psrc. 245 * However, psrc maybe comes from previous instructions' pdest, which comes from freelist. 246 * 247 * If we expand these logic for pdest(N): 248 * pdest(N) = Mux(isMove(N), psrc(N), freelist_out(N)) 249 * = Mux(isMove(N), Mux(bypass(N, N - 1), pdest(N - 1), 250 * Mux(bypass(N, N - 2), pdest(N - 2), 251 * ... 252 * Mux(bypass(N, 0), pdest(0), 253 * rat_out(N))...)), 254 * freelist_out(N)) 255 */ 256 // a simple functional model for now 257 io.out(0).bits.pdest := Mux(isMove(0), uops(0).psrc.head, uops(0).pdest) 258 val bypassCond = Wire(Vec(4, MixedVec(List.tabulate(RenameWidth-1)(i => UInt((i+1).W))))) 259 for (i <- 1 until RenameWidth) { 260 val vecCond = io.in(i).bits.ctrl.srcType.map(_ === SrcType.vp) :+ needVecDest(i) 261 val fpCond = io.in(i).bits.ctrl.srcType.map(_ === SrcType.fp) :+ needFpDest(i) 262 val intCond = io.in(i).bits.ctrl.srcType.map(_ === SrcType.reg) :+ needIntDest(i) 263 val target = io.in(i).bits.ctrl.lsrc :+ io.in(i).bits.ctrl.ldest 264 for (((((cond1, cond2), cond3), t), j) <- vecCond.zip(fpCond).zip(intCond).zip(target).zipWithIndex) { 265 val destToSrc = io.in.take(i).zipWithIndex.map { case (in, j) => 266 val indexMatch = in.bits.ctrl.ldest === t 267 val writeMatch = cond3 && needIntDest(j) || cond2 && needFpDest(j) || cond1 && needVecDest(j) 268 indexMatch && writeMatch 269 } 270 bypassCond(j)(i - 1) := VecInit(destToSrc).asUInt 271 } 272 io.out(i).bits.psrc(0) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(0)(i-1).asBools).foldLeft(uops(i).psrc(0)) { 273 (z, next) => Mux(next._2, next._1, z) 274 } 275 io.out(i).bits.psrc(1) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(1)(i-1).asBools).foldLeft(uops(i).psrc(1)) { 276 (z, next) => Mux(next._2, next._1, z) 277 } 278 io.out(i).bits.psrc(2) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(2)(i-1).asBools).foldLeft(uops(i).psrc(2)) { 279 (z, next) => Mux(next._2, next._1, z) 280 } 281 io.out(i).bits.old_pdest := io.out.take(i).map(_.bits.pdest).zip(bypassCond(3)(i-1).asBools).foldLeft(uops(i).old_pdest) { 282 (z, next) => Mux(next._2, next._1, z) 283 } 284 io.out(i).bits.pdest := Mux(isMove(i), io.out(i).bits.psrc(0), uops(i).pdest) 285 286 // For fused-lui-load, load.src(0) is replaced by the imm. 287 val last_is_lui = io.in(i - 1).bits.ctrl.selImm === SelImm.IMM_U && io.in(i - 1).bits.ctrl.srcType(0) =/= SrcType.pc 288 val this_is_load = io.in(i).bits.ctrl.fuType === FuType.ldu 289 val lui_to_load = io.in(i - 1).valid && io.in(i - 1).bits.ctrl.ldest === io.in(i).bits.ctrl.lsrc(0) 290 val fused_lui_load = last_is_lui && this_is_load && lui_to_load 291 when (fused_lui_load) { 292 // The first LOAD operand (base address) is replaced by LUI-imm and stored in {psrc, imm} 293 val lui_imm = io.in(i - 1).bits.ctrl.imm 294 val ld_imm = io.in(i).bits.ctrl.imm 295 io.out(i).bits.ctrl.srcType(0) := SrcType.imm 296 io.out(i).bits.ctrl.imm := Imm_LUI_LOAD().immFromLuiLoad(lui_imm, ld_imm) 297 val psrcWidth = uops(i).psrc.head.getWidth 298 val lui_imm_in_imm = uops(i).ctrl.imm.getWidth - Imm_I().len 299 val left_lui_imm = Imm_U().len - lui_imm_in_imm 300 require(2 * psrcWidth >= left_lui_imm, "cannot fused lui and load with psrc") 301 io.out(i).bits.psrc(0) := lui_imm(lui_imm_in_imm + psrcWidth - 1, lui_imm_in_imm) 302 io.out(i).bits.psrc(1) := lui_imm(lui_imm.getWidth - 1, lui_imm_in_imm + psrcWidth) 303 } 304 305 } 306 307 /** 308 * Instructions commit: update freelist and rename table 309 */ 310 for (i <- 0 until CommitWidth) { 311 val commitValid = io.robCommits.isCommit && io.robCommits.commitValid(i) 312 val walkValid = io.robCommits.isWalk && io.robCommits.walkValid(i) 313 314 // I. RAT Update 315 // When redirect happens (mis-prediction), don't update the rename table 316 io.intRenamePorts(i).wen := intSpecWen(i) 317 io.intRenamePorts(i).addr := uops(i).ctrl.ldest 318 io.intRenamePorts(i).data := io.out(i).bits.pdest 319 320 io.fpRenamePorts(i).wen := fpSpecWen(i) 321 io.fpRenamePorts(i).addr := uops(i).ctrl.ldest 322 io.fpRenamePorts(i).data := fpFreeList.io.allocatePhyReg(i) 323 324 io.vecRenamePorts(i).wen := vecSpecWen(i) 325 io.vecRenamePorts(i).addr := uops(i).ctrl.ldest 326 io.vecRenamePorts(i).data := fpFreeList.io.allocatePhyReg(i) 327 328 // II. Free List Update 329 intFreeList.io.freeReq(i) := intRefCounter.io.freeRegs(i).valid 330 intFreeList.io.freePhyReg(i) := intRefCounter.io.freeRegs(i).bits 331 fpFreeList.io.freeReq(i) := commitValid && needDestRegCommit(int = false, io.robCommits.info(i)) 332 fpFreeList.io.freePhyReg(i) := io.robCommits.info(i).old_pdest 333 334 intRefCounter.io.deallocate(i).valid := commitValid && needDestRegCommit(int = true, io.robCommits.info(i)) && !io.robCommits.isWalk 335 intRefCounter.io.deallocate(i).bits := io.robCommits.info(i).old_pdest 336 } 337 338 when(io.robCommits.isWalk) { 339 (intFreeList.io.allocateReq zip intFreeList.io.allocatePhyReg).take(CommitWidth) zip io.robCommits.info foreach { 340 case ((reqValid, allocReg), commitInfo) => when(reqValid) { 341 XSError(allocReg =/= commitInfo.pdest, "walk alloc reg =/= rob reg\n") 342 } 343 } 344 (fpFreeList.io.allocateReq zip fpFreeList.io.allocatePhyReg).take(CommitWidth) zip io.robCommits.info foreach { 345 case ((reqValid, allocReg), commitInfo) => when(reqValid) { 346 XSError(allocReg =/= commitInfo.pdest, "walk alloc reg =/= rob reg\n") 347 } 348 } 349 } 350 351 /* 352 Debug and performance counters 353 */ 354 def printRenameInfo(in: DecoupledIO[CfCtrl], out: DecoupledIO[MicroOp]) = { 355 XSInfo(out.fire, p"pc:${Hexadecimal(in.bits.cf.pc)} in(${in.valid},${in.ready}) " + 356 p"lsrc(0):${in.bits.ctrl.lsrc(0)} -> psrc(0):${out.bits.psrc(0)} " + 357 p"lsrc(1):${in.bits.ctrl.lsrc(1)} -> psrc(1):${out.bits.psrc(1)} " + 358 p"lsrc(2):${in.bits.ctrl.lsrc(2)} -> psrc(2):${out.bits.psrc(2)} " + 359 p"ldest:${in.bits.ctrl.ldest} -> pdest:${out.bits.pdest} " + 360 p"old_pdest:${out.bits.old_pdest}\n" 361 ) 362 } 363 364 for ((x,y) <- io.in.zip(io.out)) { 365 printRenameInfo(x, y) 366 } 367 368 XSDebug(io.robCommits.isWalk, p"Walk Recovery Enabled\n") 369 XSDebug(io.robCommits.isWalk, p"validVec:${Binary(io.robCommits.walkValid.asUInt)}\n") 370 for (i <- 0 until CommitWidth) { 371 val info = io.robCommits.info(i) 372 XSDebug(io.robCommits.isWalk && io.robCommits.walkValid(i), p"[#$i walk info] pc:${Hexadecimal(info.pc)} " + 373 p"ldest:${info.ldest} rfWen:${info.rfWen} fpWen:${info.fpWen} vecWen:${info.vecWen}" + 374 p"pdest:${info.pdest} old_pdest:${info.old_pdest}\n") 375 } 376 377 XSDebug(p"inValidVec: ${Binary(Cat(io.in.map(_.valid)))}\n") 378 379 XSPerfAccumulate("in", Mux(RegNext(io.in(0).ready), PopCount(io.in.map(_.valid)), 0.U)) 380 XSPerfAccumulate("utilization", PopCount(io.in.map(_.valid))) 381 XSPerfAccumulate("waitInstr", PopCount((0 until RenameWidth).map(i => io.in(i).valid && !io.in(i).ready))) 382 XSPerfAccumulate("stall_cycle_dispatch", hasValid && !io.out(0).ready && fpFreeList.io.canAllocate && intFreeList.io.canAllocate && !io.robCommits.isWalk) 383 XSPerfAccumulate("stall_cycle_fp", hasValid && io.out(0).ready && !fpFreeList.io.canAllocate && intFreeList.io.canAllocate && !io.robCommits.isWalk) 384 XSPerfAccumulate("stall_cycle_int", hasValid && io.out(0).ready && fpFreeList.io.canAllocate && !intFreeList.io.canAllocate && !io.robCommits.isWalk) 385 XSPerfAccumulate("stall_cycle_walk", hasValid && io.out(0).ready && fpFreeList.io.canAllocate && intFreeList.io.canAllocate && io.robCommits.isWalk) 386 XSPerfAccumulate("recovery_bubbles", PopCount(io.in.map(_.valid && io.out(0).ready && fpFreeList.io.canAllocate && intFreeList.io.canAllocate && io.robCommits.isWalk))) 387 388 XSPerfAccumulate("move_instr_count", PopCount(io.out.map(out => out.fire && out.bits.ctrl.isMove))) 389 val is_fused_lui_load = io.out.map(o => o.fire && o.bits.ctrl.fuType === FuType.ldu && o.bits.ctrl.srcType(0) === SrcType.imm) 390 XSPerfAccumulate("fused_lui_load_instr_count", PopCount(is_fused_lui_load)) 391 392 val renamePerf = Seq( 393 ("rename_in ", PopCount(io.in.map(_.valid & io.in(0).ready )) ), 394 ("rename_waitinstr ", PopCount((0 until RenameWidth).map(i => io.in(i).valid && !io.in(i).ready)) ), 395 ("rename_stall_cycle_dispatch", hasValid && !io.out(0).ready && fpFreeList.io.canAllocate && intFreeList.io.canAllocate && !io.robCommits.isWalk), 396 ("rename_stall_cycle_fp ", hasValid && io.out(0).ready && !fpFreeList.io.canAllocate && intFreeList.io.canAllocate && !io.robCommits.isWalk), 397 ("rename_stall_cycle_int ", hasValid && io.out(0).ready && fpFreeList.io.canAllocate && !intFreeList.io.canAllocate && !io.robCommits.isWalk), 398 ("rename_stall_cycle_walk ", hasValid && io.out(0).ready && fpFreeList.io.canAllocate && intFreeList.io.canAllocate && io.robCommits.isWalk) 399 ) 400 val intFlPerf = intFreeList.getPerfEvents 401 val fpFlPerf = fpFreeList.getPerfEvents 402 val perfEvents = renamePerf ++ intFlPerf ++ fpFlPerf 403 generatePerfEvent() 404} 405