xref: /XiangShan/src/main/scala/xiangshan/backend/rename/Rename.scala (revision 9ab1568e215c540ca0554308577ff4d1813bfa8c)
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 utility._
25import xiangshan.backend.decode.{FusionDecodeInfo, Imm_I, Imm_LUI_LOAD, Imm_U}
26import xiangshan.backend.rob.RobPtr
27import xiangshan.backend.rename.freelist._
28import xiangshan.mem.mdp._
29
30class Rename(implicit p: Parameters) extends XSModule with HasCircularQueuePtrHelper with HasPerfEvents {
31  val io = IO(new Bundle() {
32    val redirect = Flipped(ValidIO(new Redirect))
33    val robCommits = Input(new RobCommitIO)
34    // from decode
35    val in = Vec(RenameWidth, Flipped(DecoupledIO(new CfCtrl)))
36    val fusionInfo = Vec(DecodeWidth - 1, Flipped(new FusionDecodeInfo))
37    // ssit read result
38    val ssit = Flipped(Vec(RenameWidth, Output(new SSITEntry)))
39    // waittable read result
40    val waittable = Flipped(Vec(RenameWidth, Output(Bool())))
41    // to rename table
42    val intReadPorts = Vec(RenameWidth, Vec(3, Input(UInt(PhyRegIdxWidth.W))))
43    val fpReadPorts = Vec(RenameWidth, Vec(4, Input(UInt(PhyRegIdxWidth.W))))
44    val vecReadPorts = Vec(RenameWidth, Vec(5, Input(UInt(PhyRegIdxWidth.W))))
45    val intRenamePorts = Vec(RenameWidth, Output(new RatWritePort))
46    val fpRenamePorts = Vec(RenameWidth, Output(new RatWritePort))
47    val vecRenamePorts = Vec(RenameWidth, Output(new RatWritePort))
48    // to dispatch1
49    val out = Vec(RenameWidth, DecoupledIO(new MicroOp))
50    // debug arch ports
51    val debug_int_rat = Vec(32, Input(UInt(PhyRegIdxWidth.W)))
52    val debug_fp_rat = Vec(32, Input(UInt(PhyRegIdxWidth.W)))
53  })
54
55  // create free list and rat
56  val intFreeList = Module(new MEFreeList(NRPhyRegs))
57  val intRefCounter = Module(new RefCounter(NRPhyRegs))
58  val fpFreeList = Module(new StdFreeList(NRPhyRegs - 64))
59
60  intRefCounter.io.commit        <> io.robCommits
61  intRefCounter.io.redirect      := io.redirect.valid
62  intRefCounter.io.debug_int_rat <> io.debug_int_rat
63  intFreeList.io.commit    <> io.robCommits
64  intFreeList.io.debug_rat <> io.debug_int_rat
65  fpFreeList.io.commit     <> io.robCommits
66  fpFreeList.io.debug_rat  <> io.debug_fp_rat
67
68  // decide if given instruction needs allocating a new physical register (CfCtrl: from decode; RobCommitInfo: from rob)
69  // fp and vec share `fpFreeList`
70  def needDestReg[T <: CfCtrl](int: Boolean, x: T): Bool = {
71    if (int) x.ctrl.rfWen && x.ctrl.ldest =/= 0.U else x.ctrl.fpWen || x.ctrl.vecWen
72  }
73  def needDestReg[T <: CfCtrl](reg_t: RegType, x: T): Bool = reg_t match {
74    case Reg_I => x.ctrl.rfWen && x.ctrl.ldest =/= 0.U
75    case Reg_F => x.ctrl.fpWen
76    case Reg_V => x.ctrl.vecWen
77  }
78  def needDestRegCommit[T <: RobCommitInfo](int: Boolean, x: T): Bool = {
79    if (int) x.rfWen else x.fpWen || x.vecWen
80  }
81  def needDestRegWalk[T <: RobCommitInfo](int: Boolean, x: T): Bool = {
82    if(int) x.rfWen && x.ldest =/= 0.U else x.fpWen || x.vecWen
83  }
84
85  // connect [redirect + walk] ports for __float point__ & __integer__ free list
86  Seq(fpFreeList, intFreeList).foreach { case fl =>
87    fl.io.redirect := io.redirect.valid
88    fl.io.walk := io.robCommits.isWalk
89  }
90  // only when both fp and int free list and dispatch1 has enough space can we do allocation
91  // when isWalk, freelist can definitely allocate
92  intFreeList.io.doAllocate := fpFreeList.io.canAllocate && io.out(0).ready || io.robCommits.isWalk
93  fpFreeList.io.doAllocate := intFreeList.io.canAllocate && io.out(0).ready || io.robCommits.isWalk
94
95  //           dispatch1 ready ++ float point free list ready ++ int free list ready      ++ not walk
96  val canOut = io.out(0).ready && fpFreeList.io.canAllocate && intFreeList.io.canAllocate && !io.robCommits.isWalk
97
98
99  // speculatively assign the instruction with an robIdx
100  val validCount = PopCount(io.in.map(_.valid)) // number of instructions waiting to enter rob (from decode)
101  val robIdxHead = RegInit(0.U.asTypeOf(new RobPtr))
102  val lastCycleMisprediction = RegNext(io.redirect.valid && !io.redirect.bits.flushItself())
103  val robIdxHeadNext = Mux(io.redirect.valid, io.redirect.bits.robIdx, // redirect: move ptr to given rob index
104         Mux(lastCycleMisprediction, robIdxHead + 1.U, // mis-predict: not flush robIdx itself
105                         Mux(canOut, robIdxHead + validCount, // instructions successfully entered next stage: increase robIdx
106                      /* default */  robIdxHead))) // no instructions passed by this cycle: stick to old value
107  robIdxHead := robIdxHeadNext
108
109  /**
110    * Rename: allocate free physical register and update rename table
111    */
112  val uops = Wire(Vec(RenameWidth, new MicroOp))
113  uops.foreach( uop => {
114    uop.srcState := DontCare
115    uop.robIdx := DontCare
116    uop.debugInfo := DontCare
117    uop.lqIdx := DontCare
118    uop.sqIdx := DontCare
119  })
120
121  require(RenameWidth >= CommitWidth)
122  val needVecDest    = Wire(Vec(RenameWidth, Bool()))
123  val needFpDest     = Wire(Vec(RenameWidth, Bool()))
124  val needIntDest    = Wire(Vec(RenameWidth, Bool()))
125  val needNotIntDest = Wire(Vec(RenameWidth, Bool()))
126  val hasValid = Cat(io.in.map(_.valid)).orR
127
128  val isMove = io.in.map(_.bits.ctrl.isMove)
129
130  val walkNeedNotIntDest = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
131  val walkNeedIntDest = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
132  val walkIsMove = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
133
134  val intSpecWen = Wire(Vec(RenameWidth, Bool()))
135  val fpSpecWen  = Wire(Vec(RenameWidth, Bool()))
136  val vecSpecWen = Wire(Vec(RenameWidth, Bool()))
137
138  val walkIntSpecWen = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
139
140  val walkPdest = Wire(Vec(RenameWidth, UInt(PhyRegIdxWidth.W)))
141
142  // uop calculation
143  for (i <- 0 until RenameWidth) {
144    uops(i).cf := io.in(i).bits.cf
145    uops(i).ctrl := io.in(i).bits.ctrl
146
147    // update cf according to ssit result
148    uops(i).cf.storeSetHit := io.ssit(i).valid
149    uops(i).cf.loadWaitStrict := io.ssit(i).strict && io.ssit(i).valid
150    uops(i).cf.ssid := io.ssit(i).ssid
151
152    // update cf according to waittable result
153    uops(i).cf.loadWaitBit := io.waittable(i)
154
155    // alloc a new phy reg, fp and vec share the `fpFreeList`
156    needVecDest   (i) := io.in(i).valid && needDestReg(Reg_V,       io.in(i).bits)
157    needFpDest    (i) := io.in(i).valid && needDestReg(Reg_F,       io.in(i).bits)
158    needIntDest   (i) := io.in(i).valid && needDestReg(Reg_I,       io.in(i).bits)
159    needNotIntDest(i) := io.in(i).valid && needDestReg(int = false, io.in(i).bits)
160    if (i < CommitWidth) {
161      walkNeedNotIntDest(i) := io.robCommits.walkValid(i) && needDestRegWalk(int = false, io.robCommits.info(i))
162      walkNeedIntDest(i) := io.robCommits.walkValid(i) && needDestRegWalk(int = true, io.robCommits.info(i))
163      walkIsMove(i) := io.robCommits.info(i).isMove
164    }
165    fpFreeList.io.allocateReq(i) := Mux(io.robCommits.isWalk, walkNeedNotIntDest(i), needNotIntDest(i))
166    intFreeList.io.allocateReq(i) := Mux(io.robCommits.isWalk, walkNeedIntDest(i) && !walkIsMove(i), needIntDest(i) && !isMove(i))
167
168    // no valid instruction from decode stage || all resources (dispatch1 + both free lists) ready
169    io.in(i).ready := !hasValid || canOut
170
171    uops(i).robIdx := robIdxHead + PopCount(io.in.take(i).map(_.valid))
172
173    uops(i).psrc(0) := Mux1H(uops(i).ctrl.srcType(0), Seq(io.intReadPorts(i)(0), io.fpReadPorts(i)(0), io.vecReadPorts(i)(0)))
174    uops(i).psrc(1) := Mux1H(uops(i).ctrl.srcType(1), Seq(io.intReadPorts(i)(1), io.fpReadPorts(i)(1), io.vecReadPorts(i)(1)))
175    // int psrc2 should be bypassed from next instruction if it is fused
176    if (i < RenameWidth - 1) {
177      when (io.fusionInfo(i).rs2FromRs2 || io.fusionInfo(i).rs2FromRs1) {
178        uops(i).psrc(1) := Mux(io.fusionInfo(i).rs2FromRs2, io.intReadPorts(i + 1)(1), io.intReadPorts(i + 1)(0))
179      }.elsewhen(io.fusionInfo(i).rs2FromZero) {
180        uops(i).psrc(1) := 0.U
181      }
182    }
183    uops(i).psrc(2) := Mux1H(uops(i).ctrl.srcType(2)(2, 1), Seq(io.fpReadPorts(i)(2), io.vecReadPorts(i)(2)))
184    uops(i).psrc(3) := io.vecReadPorts(i)(3)
185    uops(i).old_pdest := Mux1H(Seq(
186      uops(i).ctrl.rfWen  -> io.intReadPorts(i).last,
187      uops(i).ctrl.fpWen  -> io.fpReadPorts (i).last,
188      uops(i).ctrl.vecWen -> io.vecReadPorts(i).last
189    ))
190    uops(i).eliminatedMove := isMove(i)
191
192    // update pdest
193    uops(i).pdest := Mux(needIntDest(i), intFreeList.io.allocatePhyReg(i), // normal int inst
194      // normal fp inst
195      Mux(needNotIntDest(i), fpFreeList.io.allocatePhyReg(i),
196        /* default */0.U))
197
198    // Assign performance counters
199    uops(i).debugInfo.renameTime := GTimer()
200
201    io.out(i).valid := io.in(i).valid && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && !io.robCommits.isWalk
202    io.out(i).bits := uops(i)
203    // dirty code for fence. The lsrc is passed by imm.
204    when (io.out(i).bits.ctrl.fuType === FuType.fence) {
205      io.out(i).bits.ctrl.imm := Cat(io.in(i).bits.ctrl.lsrc(1), io.in(i).bits.ctrl.lsrc(0))
206    }
207    // dirty code for SoftPrefetch (prefetch.r/prefetch.w)
208    when (io.in(i).bits.ctrl.isSoftPrefetch) {
209      io.out(i).bits.ctrl.fuType := FuType.ldu
210      io.out(i).bits.ctrl.fuOpType := Mux(io.in(i).bits.ctrl.lsrc(1) === 1.U, LSUOpType.prefetch_r, LSUOpType.prefetch_w)
211      io.out(i).bits.ctrl.selImm := SelImm.IMM_S
212      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))
213    }
214
215    // write speculative rename table
216    // we update rat later inside commit code
217    intSpecWen(i) := needIntDest(i) && intFreeList.io.canAllocate && intFreeList.io.doAllocate && !io.robCommits.isWalk && !io.redirect.valid
218    fpSpecWen(i) := needFpDest(i) && fpFreeList.io.canAllocate && fpFreeList.io.doAllocate && !io.robCommits.isWalk && !io.redirect.valid
219    vecSpecWen(i) := needVecDest(i) && fpFreeList.io.canAllocate && fpFreeList.io.doAllocate && !io.robCommits.isWalk && !io.redirect.valid
220
221    if (i < CommitWidth) {
222      walkIntSpecWen(i) := walkNeedIntDest(i) && !io.redirect.valid
223      walkPdest(i) := io.robCommits.info(i).pdest
224    } else {
225      walkPdest(i) := io.out(i).bits.pdest
226    }
227
228    intRefCounter.io.allocate(i).valid := Mux(io.robCommits.isWalk, walkIntSpecWen(i), intSpecWen(i))
229    intRefCounter.io.allocate(i).bits := Mux(io.robCommits.isWalk, walkPdest(i), io.out(i).bits.pdest)
230  }
231
232  /**
233    * How to set psrc:
234    * - bypass the pdest to psrc if previous instructions write to the same ldest as lsrc
235    * - default: psrc from RAT
236    * How to set pdest:
237    * - Mux(isMove, psrc, pdest_from_freelist).
238    *
239    * The critical path of rename lies here:
240    * When move elimination is enabled, we need to update the rat with psrc.
241    * However, psrc maybe comes from previous instructions' pdest, which comes from freelist.
242    *
243    * If we expand these logic for pdest(N):
244    * pdest(N) = Mux(isMove(N), psrc(N), freelist_out(N))
245    *          = Mux(isMove(N), Mux(bypass(N, N - 1), pdest(N - 1),
246    *                           Mux(bypass(N, N - 2), pdest(N - 2),
247    *                           ...
248    *                           Mux(bypass(N, 0),     pdest(0),
249    *                                                 rat_out(N))...)),
250    *                           freelist_out(N))
251    */
252  // a simple functional model for now
253  io.out(0).bits.pdest := Mux(isMove(0), uops(0).psrc.head, uops(0).pdest)
254  val bypassCond = Wire(Vec(5, MixedVec(List.tabulate(RenameWidth-1)(i => UInt((i+1).W)))))
255  for (i <- 1 until RenameWidth) {
256    val vecCond = io.in(i).bits.ctrl.srcType.map(_ === SrcType.vp) :+ needVecDest(i)
257    val fpCond = io.in(i).bits.ctrl.srcType.map(_ === SrcType.fp) :+ needFpDest(i)
258    val intCond = io.in(i).bits.ctrl.srcType.map(_ === SrcType.reg) :+ needIntDest(i)
259    val target = io.in(i).bits.ctrl.lsrc :+ io.in(i).bits.ctrl.ldest
260    for (((((cond1, cond2), cond3), t), j) <- vecCond.zip(fpCond).zip(intCond).zip(target).zipWithIndex) {
261      val destToSrc = io.in.take(i).zipWithIndex.map { case (in, j) =>
262        val indexMatch = in.bits.ctrl.ldest === t
263        val writeMatch =  cond3 && needIntDest(j) || cond2 && needFpDest(j) || cond1 && needVecDest(j)
264        indexMatch && writeMatch
265      }
266      bypassCond(j)(i - 1) := VecInit(destToSrc).asUInt
267    }
268    io.out(i).bits.psrc(0) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(0)(i-1).asBools).foldLeft(uops(i).psrc(0)) {
269      (z, next) => Mux(next._2, next._1, z)
270    }
271    io.out(i).bits.psrc(1) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(1)(i-1).asBools).foldLeft(uops(i).psrc(1)) {
272      (z, next) => Mux(next._2, next._1, z)
273    }
274    io.out(i).bits.psrc(2) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(2)(i-1).asBools).foldLeft(uops(i).psrc(2)) {
275      (z, next) => Mux(next._2, next._1, z)
276    }
277    io.out(i).bits.psrc(3) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(3)(i-1).asBools).foldLeft(uops(i).psrc(3)) {
278      (z, next) => Mux(next._2, next._1, z)
279    }
280    io.out(i).bits.old_pdest := io.out.take(i).map(_.bits.pdest).zip(bypassCond(4)(i-1).asBools).foldLeft(uops(i).old_pdest) {
281      (z, next) => Mux(next._2, next._1, z)
282    }
283    io.out(i).bits.pdest := Mux(isMove(i), io.out(i).bits.psrc(0), uops(i).pdest)
284
285    // For fused-lui-load, load.src(0) is replaced by the imm.
286    val last_is_lui = io.in(i - 1).bits.ctrl.selImm === SelImm.IMM_U && io.in(i - 1).bits.ctrl.srcType(0) =/= SrcType.pc
287    val this_is_load = io.in(i).bits.ctrl.fuType === FuType.ldu
288    val lui_to_load = io.in(i - 1).valid && io.in(i - 1).bits.ctrl.ldest === io.in(i).bits.ctrl.lsrc(0)
289    val fused_lui_load = last_is_lui && this_is_load && lui_to_load
290    when (fused_lui_load) {
291      // The first LOAD operand (base address) is replaced by LUI-imm and stored in {psrc, imm}
292      val lui_imm = io.in(i - 1).bits.ctrl.imm
293      val ld_imm = io.in(i).bits.ctrl.imm
294      io.out(i).bits.ctrl.srcType(0) := SrcType.imm
295      io.out(i).bits.ctrl.imm := Imm_LUI_LOAD().immFromLuiLoad(lui_imm, ld_imm)
296      val psrcWidth = uops(i).psrc.head.getWidth
297      val lui_imm_in_imm = uops(i).ctrl.imm.getWidth - Imm_I().len
298      val left_lui_imm = Imm_U().len - lui_imm_in_imm
299      require(2 * psrcWidth >= left_lui_imm, "cannot fused lui and load with psrc")
300      io.out(i).bits.psrc(0) := lui_imm(lui_imm_in_imm + psrcWidth - 1, lui_imm_in_imm)
301      io.out(i).bits.psrc(1) := lui_imm(lui_imm.getWidth - 1, lui_imm_in_imm + psrcWidth)
302    }
303
304  }
305
306  /**
307    * Instructions commit: update freelist and rename table
308    */
309  for (i <- 0 until CommitWidth) {
310    val commitValid = io.robCommits.isCommit && io.robCommits.commitValid(i)
311    val walkValid = io.robCommits.isWalk && io.robCommits.walkValid(i)
312
313    // I. RAT Update
314    // When redirect happens (mis-prediction), don't update the rename table
315    io.intRenamePorts(i).wen  := intSpecWen(i)
316    io.intRenamePorts(i).addr := uops(i).ctrl.ldest
317    io.intRenamePorts(i).data := io.out(i).bits.pdest
318
319    io.fpRenamePorts(i).wen  := fpSpecWen(i)
320    io.fpRenamePorts(i).addr := uops(i).ctrl.ldest
321    io.fpRenamePorts(i).data := fpFreeList.io.allocatePhyReg(i)
322
323    io.vecRenamePorts(i).wen  := vecSpecWen(i)
324    io.vecRenamePorts(i).addr := uops(i).ctrl.ldest
325    io.vecRenamePorts(i).data := fpFreeList.io.allocatePhyReg(i)
326
327    // II. Free List Update
328    intFreeList.io.freeReq(i) := intRefCounter.io.freeRegs(i).valid
329    intFreeList.io.freePhyReg(i) := intRefCounter.io.freeRegs(i).bits
330    fpFreeList.io.freeReq(i)  := commitValid && needDestRegCommit(int = false, io.robCommits.info(i))
331    fpFreeList.io.freePhyReg(i) := io.robCommits.info(i).old_pdest
332
333    intRefCounter.io.deallocate(i).valid := commitValid && needDestRegCommit(int = true, io.robCommits.info(i)) && !io.robCommits.isWalk
334    intRefCounter.io.deallocate(i).bits := io.robCommits.info(i).old_pdest
335  }
336
337  when(io.robCommits.isWalk) {
338    (intFreeList.io.allocateReq zip intFreeList.io.allocatePhyReg).take(CommitWidth) zip io.robCommits.info foreach {
339      case ((reqValid, allocReg), commitInfo) => when(reqValid) {
340        XSError(allocReg =/= commitInfo.pdest, "walk alloc reg =/= rob reg\n")
341      }
342    }
343    (fpFreeList.io.allocateReq zip fpFreeList.io.allocatePhyReg).take(CommitWidth) zip io.robCommits.info foreach {
344      case ((reqValid, allocReg), commitInfo) => when(reqValid) {
345        XSError(allocReg =/= commitInfo.pdest, "walk alloc reg =/= rob reg\n")
346      }
347    }
348  }
349
350  /*
351  Debug and performance counters
352   */
353  def printRenameInfo(in: DecoupledIO[CfCtrl], out: DecoupledIO[MicroOp]) = {
354    XSInfo(out.fire, p"pc:${Hexadecimal(in.bits.cf.pc)} in(${in.valid},${in.ready}) " +
355      p"lsrc(0):${in.bits.ctrl.lsrc(0)} -> psrc(0):${out.bits.psrc(0)} " +
356      p"lsrc(1):${in.bits.ctrl.lsrc(1)} -> psrc(1):${out.bits.psrc(1)} " +
357      p"lsrc(2):${in.bits.ctrl.lsrc(2)} -> psrc(2):${out.bits.psrc(2)} " +
358      p"lsrc(3):${in.bits.ctrl.lsrc(3)} -> psrc(3):${out.bits.psrc(3)} " +
359      p"ldest:${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