xref: /XiangShan/src/main/scala/xiangshan/backend/rename/Rename.scala (revision 887862dbb8debde8ab099befc426493834a69ee7)
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 org.chipsalliance.cde.config.Parameters
20import chisel3._
21import chisel3.util._
22import utility._
23import utils._
24import xiangshan._
25import xiangshan.backend.Bundles.{DecodedInst, DynInst}
26import xiangshan.backend.decode.{FusionDecodeInfo, ImmUnion, Imm_I, Imm_LUI_LOAD, Imm_U}
27import xiangshan.backend.fu.FuType
28import xiangshan.backend.rename.freelist._
29import xiangshan.backend.rob.{RobEnqIO, RobPtr}
30import xiangshan.mem.mdp._
31import xiangshan.ExceptionNO._
32import xiangshan.backend.fu.FuType._
33import xiangshan.mem.{EewLog2, GenUSWholeEmul}
34import xiangshan.mem.GenRealFlowNum
35import xiangshan.backend.trace._
36
37class Rename(implicit p: Parameters) extends XSModule with HasCircularQueuePtrHelper with HasPerfEvents {
38
39  // params alias
40  private val numRegSrc = backendParams.numRegSrc
41  private val numVecRegSrc = backendParams.numVecRegSrc
42  private val numVecRatPorts = numVecRegSrc
43
44  println(s"[Rename] numRegSrc: $numRegSrc")
45
46  val io = IO(new Bundle() {
47    val redirect = Flipped(ValidIO(new Redirect))
48    val rabCommits = Input(new RabCommitIO)
49    // from csr
50    val singleStep = Input(Bool())
51    // from decode
52    val in = Vec(RenameWidth, Flipped(DecoupledIO(new DecodedInst)))
53    val fusionInfo = Vec(DecodeWidth - 1, Flipped(new FusionDecodeInfo))
54    // ssit read result
55    val ssit = Flipped(Vec(RenameWidth, Output(new SSITEntry)))
56    // waittable read result
57    val waittable = Flipped(Vec(RenameWidth, Output(Bool())))
58    // to rename table
59    val intReadPorts = Vec(RenameWidth, Vec(2, Input(UInt(PhyRegIdxWidth.W))))
60    val fpReadPorts = Vec(RenameWidth, Vec(3, Input(UInt(PhyRegIdxWidth.W))))
61    val vecReadPorts = Vec(RenameWidth, Vec(numVecRatPorts, Input(UInt(PhyRegIdxWidth.W))))
62    val v0ReadPorts = Vec(RenameWidth, Vec(1, Input(UInt(PhyRegIdxWidth.W))))
63    val vlReadPorts = Vec(RenameWidth, Vec(1, Input(UInt(PhyRegIdxWidth.W))))
64    val intRenamePorts = Vec(RenameWidth, Output(new RatWritePort(log2Ceil(IntLogicRegs))))
65    val fpRenamePorts = Vec(RenameWidth, Output(new RatWritePort(log2Ceil(FpLogicRegs))))
66    val vecRenamePorts = Vec(RenameWidth, Output(new RatWritePort(log2Ceil(VecLogicRegs))))
67    val v0RenamePorts = Vec(RenameWidth, Output(new RatWritePort(log2Ceil(V0LogicRegs))))
68    val vlRenamePorts = Vec(RenameWidth, Output(new RatWritePort(log2Ceil(VlLogicRegs))))
69    // from rename table
70    val int_old_pdest = Vec(RabCommitWidth, Input(UInt(PhyRegIdxWidth.W)))
71    val fp_old_pdest = Vec(RabCommitWidth, Input(UInt(PhyRegIdxWidth.W)))
72    val vec_old_pdest = Vec(RabCommitWidth, Input(UInt(PhyRegIdxWidth.W)))
73    val v0_old_pdest = Vec(RabCommitWidth, Input(UInt(PhyRegIdxWidth.W)))
74    val vl_old_pdest = Vec(RabCommitWidth, Input(UInt(PhyRegIdxWidth.W)))
75    val int_need_free = Vec(RabCommitWidth, Input(Bool()))
76    // to dispatch1
77    val out = Vec(RenameWidth, DecoupledIO(new DynInst))
78    // for snapshots
79    val snpt = Input(new SnapshotPort)
80    val snptLastEnq = Flipped(ValidIO(new RobPtr))
81    val snptIsFull= Input(Bool())
82    // debug arch ports
83    val debug_int_rat = if (backendParams.debugEn) Some(Vec(32, Input(UInt(PhyRegIdxWidth.W)))) else None
84    val debug_fp_rat = if (backendParams.debugEn) Some(Vec(32, Input(UInt(PhyRegIdxWidth.W)))) else None
85    val debug_vec_rat = if (backendParams.debugEn) Some(Vec(31, Input(UInt(PhyRegIdxWidth.W)))) else None
86    val debug_v0_rat = if (backendParams.debugEn) Some(Vec(1, Input(UInt(PhyRegIdxWidth.W)))) else None
87    val debug_vl_rat = if (backendParams.debugEn) Some(Vec(1, Input(UInt(PhyRegIdxWidth.W)))) else None
88    // perf only
89    val stallReason = new Bundle {
90      val in = Flipped(new StallReasonIO(RenameWidth))
91      val out = new StallReasonIO(RenameWidth)
92    }
93  })
94
95  // io alias
96  private val dispatchCanAcc = io.out.head.ready
97
98  val compressUnit = Module(new CompressUnit())
99  // create free list and rat
100  val intFreeList = Module(new MEFreeList(IntPhyRegs))
101  val fpFreeList = Module(new StdFreeList(FpPhyRegs - FpLogicRegs, FpLogicRegs, Reg_F))
102  val vecFreeList = Module(new StdFreeList(VfPhyRegs - VecLogicRegs, VecLogicRegs, Reg_V, 31))
103  val v0FreeList = Module(new StdFreeList(V0PhyRegs - V0LogicRegs, V0LogicRegs, Reg_V0, 1))
104  val vlFreeList = Module(new StdFreeList(VlPhyRegs - VlLogicRegs, VlLogicRegs, Reg_Vl, 1))
105
106
107  intFreeList.io.commit    <> io.rabCommits
108  intFreeList.io.debug_rat.foreach(_ <> io.debug_int_rat.get)
109  fpFreeList.io.commit     <> io.rabCommits
110  fpFreeList.io.debug_rat.foreach(_ <> io.debug_fp_rat.get)
111  vecFreeList.io.commit    <> io.rabCommits
112  vecFreeList.io.debug_rat.foreach(_ <> io.debug_vec_rat.get)
113  v0FreeList.io.commit <> io.rabCommits
114  v0FreeList.io.debug_rat.foreach(_ <> io.debug_v0_rat.get)
115  vlFreeList.io.commit <> io.rabCommits
116  vlFreeList.io.debug_rat.foreach(_ <> io.debug_vl_rat.get)
117
118  // decide if given instruction needs allocating a new physical register (CfCtrl: from decode; RobCommitInfo: from rob)
119  def needDestReg[T <: DecodedInst](reg_t: RegType, x: T): Bool = reg_t match {
120    case Reg_I => x.rfWen && x.ldest =/= 0.U
121    case Reg_F => x.fpWen
122    case Reg_V => x.vecWen
123    case Reg_V0 => x.v0Wen
124    case Reg_Vl => x.vlWen
125  }
126  def needDestRegCommit[T <: RabCommitInfo](reg_t: RegType, x: T): Bool = {
127    reg_t match {
128      case Reg_I => x.rfWen
129      case Reg_F => x.fpWen
130      case Reg_V => x.vecWen
131      case Reg_V0 => x.v0Wen
132      case Reg_Vl => x.vlWen
133    }
134  }
135  def needDestRegWalk[T <: RabCommitInfo](reg_t: RegType, x: T): Bool = {
136    reg_t match {
137      case Reg_I => x.rfWen && x.ldest =/= 0.U
138      case Reg_F => x.fpWen
139      case Reg_V => x.vecWen
140      case Reg_V0 => x.v0Wen
141      case Reg_Vl => x.vlWen
142    }
143  }
144
145  // connect [redirect + walk] ports for fp & vec & int free list
146  Seq(fpFreeList, vecFreeList, intFreeList, v0FreeList, vlFreeList).foreach { case fl =>
147    fl.io.redirect := io.redirect.valid
148    fl.io.walk := io.rabCommits.isWalk
149  }
150  // only when all free list and dispatch1 has enough space can we do allocation
151  // when isWalk, freelist can definitely allocate
152  intFreeList.io.doAllocate := fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && dispatchCanAcc || io.rabCommits.isWalk
153  fpFreeList.io.doAllocate := intFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && dispatchCanAcc || io.rabCommits.isWalk
154  vecFreeList.io.doAllocate := intFreeList.io.canAllocate && fpFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && dispatchCanAcc || io.rabCommits.isWalk
155  v0FreeList.io.doAllocate := intFreeList.io.canAllocate && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && vlFreeList.io.canAllocate && dispatchCanAcc || io.rabCommits.isWalk
156  vlFreeList.io.doAllocate := intFreeList.io.canAllocate && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && dispatchCanAcc || io.rabCommits.isWalk
157
158  //           dispatch1 ready ++ float point free list ready ++ int free list ready ++ vec free list ready     ++ not walk
159  val canOut = dispatchCanAcc && fpFreeList.io.canAllocate && intFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !io.rabCommits.isWalk
160
161  compressUnit.io.in.zip(io.in).foreach{ case(sink, source) =>
162    sink.valid := source.valid && !io.singleStep
163    sink.bits := source.bits
164  }
165  val needRobFlags = compressUnit.io.out.needRobFlags
166  val instrSizesVec = compressUnit.io.out.instrSizes
167  val compressMasksVec = compressUnit.io.out.masks
168
169  // speculatively assign the instruction with an robIdx
170  val validCount = PopCount(io.in.zip(needRobFlags).map{ case(in, needRobFlag) => in.valid && in.bits.lastUop && needRobFlag}) // number of instructions waiting to enter rob (from decode)
171  val robIdxHead = RegInit(0.U.asTypeOf(new RobPtr))
172  val lastCycleMisprediction = GatedValidRegNext(io.redirect.valid && !io.redirect.bits.flushItself())
173  val robIdxHeadNext = Mux(io.redirect.valid, io.redirect.bits.robIdx, // redirect: move ptr to given rob index
174         Mux(lastCycleMisprediction, robIdxHead + 1.U, // mis-predict: not flush robIdx itself
175           Mux(canOut, robIdxHead + validCount, // instructions successfully entered next stage: increase robIdx
176                      /* default */  robIdxHead))) // no instructions passed by this cycle: stick to old value
177  robIdxHead := robIdxHeadNext
178
179  /**
180    * Rename: allocate free physical register and update rename table
181    */
182  val uops = Wire(Vec(RenameWidth, new DynInst))
183  uops.foreach( uop => {
184    uop.srcState      := DontCare
185    uop.debugInfo     := DontCare
186    uop.lqIdx         := DontCare
187    uop.sqIdx         := DontCare
188    uop.waitForRobIdx := DontCare
189    uop.singleStep    := DontCare
190    uop.snapshot      := DontCare
191    uop.srcLoadDependency := DontCare
192    uop.numLsElem       :=  DontCare
193    uop.hasException  :=  DontCare
194    uop.useRegCache   := DontCare
195    uop.regCacheIdx   := DontCare
196    uop.traceBlockInPipe := DontCare
197  })
198  private val fuType       = uops.map(_.fuType)
199  private val fuOpType     = uops.map(_.fuOpType)
200  private val vtype        = uops.map(_.vpu.vtype)
201  private val sew          = vtype.map(_.vsew)
202  private val lmul         = vtype.map(_.vlmul)
203  private val eew          = uops.map(_.vpu.veew)
204  private val mop          = fuOpType.map(fuOpTypeItem => LSUOpType.getVecLSMop(fuOpTypeItem))
205  private val isVlsType    = fuType.map(fuTypeItem => isVls(fuTypeItem))
206  private val isSegment    = fuType.map(fuTypeItem => isVsegls(fuTypeItem))
207  private val isUnitStride = fuOpType.map(fuOpTypeItem => LSUOpType.isAllUS(fuOpTypeItem))
208  private val nf           = fuOpType.zip(uops.map(_.vpu.nf)).map { case (fuOpTypeItem, nfItem) => Mux(LSUOpType.isWhole(fuOpTypeItem), 0.U, nfItem) }
209  private val mulBits      = 3 // dirty code
210  private val emul         = fuOpType.zipWithIndex.map { case (fuOpTypeItem, index) =>
211    Mux(
212      LSUOpType.isWhole(fuOpTypeItem),
213      GenUSWholeEmul(nf(index)),
214      Mux(
215        LSUOpType.isMasked(fuOpTypeItem),
216        0.U(mulBits.W),
217        EewLog2(eew(index)) - sew(index) + lmul(index)
218      )
219    )
220  }
221  private val isVecUnitType = isVlsType.zip(isUnitStride).map { case (isVlsTypeItme, isUnitStrideItem) =>
222    isVlsTypeItme && isUnitStrideItem
223  }
224  private val instType = isSegment.zip(mop).map { case (isSegementItem, mopItem) => Cat(isSegementItem, mopItem) }
225  // There is no way to calculate the 'flow' for 'unit-stride' exactly:
226  //  Whether 'unit-stride' needs to be split can only be known after obtaining the address.
227  // For scalar instructions, this is not handled here, and different assignments are done later according to the situation.
228  private val numLsElem = instType.zipWithIndex.map { case (instTypeItem, index) =>
229    Mux(
230      isVecUnitType(index),
231      VecMemUnitStrideMaxFlowNum.U,
232      GenRealFlowNum(instTypeItem, emul(index), lmul(index), eew(index), sew(index))
233    )
234  }
235  uops.zipWithIndex.map { case(u, i) =>
236    u.numLsElem := Mux(io.in(i).valid & isVlsType(i), numLsElem(i), 0.U)
237  }
238
239  val needVecDest    = Wire(Vec(RenameWidth, Bool()))
240  val needFpDest     = Wire(Vec(RenameWidth, Bool()))
241  val needIntDest    = Wire(Vec(RenameWidth, Bool()))
242  val needV0Dest     = Wire(Vec(RenameWidth, Bool()))
243  val needVlDest     = Wire(Vec(RenameWidth, Bool()))
244  private val inHeadValid = io.in.head.valid
245
246  val isMove = Wire(Vec(RenameWidth, Bool()))
247  isMove zip io.in.map(_.bits) foreach {
248    case (move, in) => move := Mux(in.exceptionVec.asUInt.orR, false.B, in.isMove)
249  }
250
251  val walkNeedIntDest = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
252  val walkNeedFpDest = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
253  val walkNeedVecDest = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
254  val walkNeedV0Dest = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
255  val walkNeedVlDest = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
256  val walkIsMove = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
257
258  val intSpecWen = Wire(Vec(RenameWidth, Bool()))
259  val fpSpecWen  = Wire(Vec(RenameWidth, Bool()))
260  val vecSpecWen = Wire(Vec(RenameWidth, Bool()))
261  val v0SpecWen = Wire(Vec(RenameWidth, Bool()))
262  val vlSpecWen = Wire(Vec(RenameWidth, Bool()))
263
264  val walkIntSpecWen = WireDefault(VecInit(Seq.fill(RenameWidth)(false.B)))
265
266  val walkPdest = Wire(Vec(RenameWidth, UInt(PhyRegIdxWidth.W)))
267
268  // uop calculation
269  for (i <- 0 until RenameWidth) {
270    (uops(i): Data).waiveAll :<= (io.in(i).bits: Data).waiveAll
271
272    // update cf according to ssit result
273    uops(i).storeSetHit := io.ssit(i).valid
274    uops(i).loadWaitStrict := io.ssit(i).strict && io.ssit(i).valid
275    uops(i).ssid := io.ssit(i).ssid
276
277    // update cf according to waittable result
278    uops(i).loadWaitBit := io.waittable(i)
279
280    uops(i).replayInst := false.B // set by IQ or MemQ
281    // alloc a new phy reg
282    needV0Dest(i) := io.in(i).valid && needDestReg(Reg_V0, io.in(i).bits)
283    needVlDest(i) := io.in(i).valid && needDestReg(Reg_Vl, io.in(i).bits)
284    needVecDest(i) := io.in(i).valid && needDestReg(Reg_V, io.in(i).bits)
285    needFpDest(i) := io.in(i).valid && needDestReg(Reg_F, io.in(i).bits)
286    needIntDest(i) := io.in(i).valid && needDestReg(Reg_I, io.in(i).bits)
287    if (i < RabCommitWidth) {
288      walkNeedIntDest(i) := io.rabCommits.walkValid(i) && needDestRegWalk(Reg_I, io.rabCommits.info(i))
289      walkNeedFpDest(i) := io.rabCommits.walkValid(i) && needDestRegWalk(Reg_F, io.rabCommits.info(i))
290      walkNeedVecDest(i) := io.rabCommits.walkValid(i) && needDestRegWalk(Reg_V, io.rabCommits.info(i))
291      walkNeedV0Dest(i) := io.rabCommits.walkValid(i) && needDestRegWalk(Reg_V0, io.rabCommits.info(i))
292      walkNeedVlDest(i) := io.rabCommits.walkValid(i) && needDestRegWalk(Reg_Vl, io.rabCommits.info(i))
293      walkIsMove(i) := io.rabCommits.info(i).isMove
294    }
295    fpFreeList.io.allocateReq(i) := needFpDest(i)
296    fpFreeList.io.walkReq(i) := walkNeedFpDest(i)
297    vecFreeList.io.allocateReq(i) := needVecDest(i)
298    vecFreeList.io.walkReq(i) := walkNeedVecDest(i)
299    v0FreeList.io.allocateReq(i) := needV0Dest(i)
300    v0FreeList.io.walkReq(i) := walkNeedV0Dest(i)
301    vlFreeList.io.allocateReq(i) := needVlDest(i)
302    vlFreeList.io.walkReq(i) := walkNeedVlDest(i)
303    intFreeList.io.allocateReq(i) := needIntDest(i) && !isMove(i)
304    intFreeList.io.walkReq(i) := walkNeedIntDest(i) && !walkIsMove(i)
305
306    // no valid instruction from decode stage || all resources (dispatch1 + both free lists) ready
307    io.in(i).ready := !io.in(0).valid || canOut
308
309    uops(i).robIdx := robIdxHead + PopCount(io.in.zip(needRobFlags).take(i).map{ case(in, needRobFlag) => in.valid && in.bits.lastUop && needRobFlag})
310    uops(i).instrSize := instrSizesVec(i)
311    val hasExceptionExceptFlushPipe = Cat(selectFrontend(uops(i).exceptionVec) :+ uops(i).exceptionVec(illegalInstr) :+ uops(i).exceptionVec(virtualInstr)).orR || TriggerAction.isDmode(uops(i).trigger)
312    when(isMove(i) || hasExceptionExceptFlushPipe) {
313      uops(i).numUops := 0.U
314      uops(i).numWB := 0.U
315    }
316    if (i > 0) {
317      when(!needRobFlags(i - 1)) {
318        uops(i).firstUop := false.B
319        uops(i).ftqPtr := uops(i - 1).ftqPtr
320        uops(i).ftqOffset := uops(i - 1).ftqOffset
321        uops(i).numUops := instrSizesVec(i) - PopCount(compressMasksVec(i) & Cat(isMove.reverse))
322        uops(i).numWB := instrSizesVec(i) - PopCount(compressMasksVec(i) & Cat(isMove.reverse))
323      }
324    }
325    when(!needRobFlags(i)) {
326      uops(i).lastUop := false.B
327      uops(i).numUops := instrSizesVec(i) - PopCount(compressMasksVec(i) & Cat(isMove.reverse))
328      uops(i).numWB := instrSizesVec(i) - PopCount(compressMasksVec(i) & Cat(isMove.reverse))
329    }
330    uops(i).wfflags := (compressMasksVec(i) & Cat(io.in.map(_.bits.wfflags).reverse)).orR
331    uops(i).dirtyFs := (compressMasksVec(i) & Cat(io.in.map(_.bits.fpWen).reverse)).orR
332    // vector instructions' uopSplitType cannot be UopSplitType.SCA_SIM
333    uops(i).dirtyVs := (compressMasksVec(i) & Cat(io.in.map(_.bits.uopSplitType =/= UopSplitType.SCA_SIM).reverse)).orR
334    // psrc0,psrc1,psrc2 don't require v0ReadPorts because their srcType can distinguish whether they are V0 or not
335    uops(i).psrc(0) := Mux1H(uops(i).srcType(0)(2, 0), Seq(io.intReadPorts(i)(0), io.fpReadPorts(i)(0), io.vecReadPorts(i)(0)))
336    uops(i).psrc(1) := Mux1H(uops(i).srcType(1)(2, 0), Seq(io.intReadPorts(i)(1), io.fpReadPorts(i)(1), io.vecReadPorts(i)(1)))
337    uops(i).psrc(2) := Mux1H(uops(i).srcType(2)(2, 1), Seq(io.fpReadPorts(i)(2), io.vecReadPorts(i)(2)))
338    uops(i).psrc(3) := io.v0ReadPorts(i)(0)
339    uops(i).psrc(4) := io.vlReadPorts(i)(0)
340
341    // int psrc2 should be bypassed from next instruction if it is fused
342    if (i < RenameWidth - 1) {
343      when (io.fusionInfo(i).rs2FromRs2 || io.fusionInfo(i).rs2FromRs1) {
344        uops(i).psrc(1) := Mux(io.fusionInfo(i).rs2FromRs2, io.intReadPorts(i + 1)(1), io.intReadPorts(i + 1)(0))
345      }.elsewhen(io.fusionInfo(i).rs2FromZero) {
346        uops(i).psrc(1) := 0.U
347      }
348    }
349    uops(i).eliminatedMove := isMove(i)
350
351    // update pdest
352    uops(i).pdest := MuxCase(0.U, Seq(
353      needIntDest(i)    ->  intFreeList.io.allocatePhyReg(i),
354      needFpDest(i)     ->  fpFreeList.io.allocatePhyReg(i),
355      needVecDest(i)    ->  vecFreeList.io.allocatePhyReg(i),
356      needV0Dest(i)    ->  v0FreeList.io.allocatePhyReg(i),
357      needVlDest(i)    ->  vlFreeList.io.allocatePhyReg(i),
358    ))
359
360    // Assign performance counters
361    uops(i).debugInfo.renameTime := GTimer()
362
363    io.out(i).valid := io.in(i).valid && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !io.rabCommits.isWalk
364    io.out(i).bits := uops(i)
365    // Todo: move these shit in decode stage
366    // dirty code for fence. The lsrc is passed by imm.
367    when (io.out(i).bits.fuType === FuType.fence.U) {
368      io.out(i).bits.imm := Cat(io.in(i).bits.lsrc(1), io.in(i).bits.lsrc(0))
369    }
370
371    // dirty code for SoftPrefetch (prefetch.r/prefetch.w)
372//    when (io.in(i).bits.isSoftPrefetch) {
373//      io.out(i).bits.fuType := FuType.ldu.U
374//      io.out(i).bits.fuOpType := Mux(io.in(i).bits.lsrc(1) === 1.U, LSUOpType.prefetch_r, LSUOpType.prefetch_w)
375//      io.out(i).bits.selImm := SelImm.IMM_S
376//      io.out(i).bits.imm := Cat(io.in(i).bits.imm(io.in(i).bits.imm.getWidth - 1, 5), 0.U(5.W))
377//    }
378
379    // dirty code for lui+addi(w) fusion
380    if (i < RenameWidth - 1) {
381      val fused_lui32 = io.in(i).bits.selImm === SelImm.IMM_LUI32 && io.in(i).bits.fuType === FuType.alu.U
382      when (fused_lui32) {
383        val lui_imm = io.in(i).bits.imm(19, 0)
384        val add_imm = io.in(i + 1).bits.imm(11, 0)
385        require(io.out(i).bits.imm.getWidth >= lui_imm.getWidth + add_imm.getWidth)
386        io.out(i).bits.imm := Cat(lui_imm, add_imm)
387      }
388    }
389
390    // write speculative rename table
391    // we update rat later inside commit code
392    intSpecWen(i) := needIntDest(i) && intFreeList.io.canAllocate && intFreeList.io.doAllocate && !io.rabCommits.isWalk && !io.redirect.valid
393    fpSpecWen(i)  := needFpDest(i)  && fpFreeList.io.canAllocate  && fpFreeList.io.doAllocate  && !io.rabCommits.isWalk && !io.redirect.valid
394    vecSpecWen(i) := needVecDest(i) && vecFreeList.io.canAllocate && vecFreeList.io.doAllocate && !io.rabCommits.isWalk && !io.redirect.valid
395    v0SpecWen(i) := needV0Dest(i) && v0FreeList.io.canAllocate && v0FreeList.io.doAllocate && !io.rabCommits.isWalk && !io.redirect.valid
396    vlSpecWen(i) := needVlDest(i) && vlFreeList.io.canAllocate && vlFreeList.io.doAllocate && !io.rabCommits.isWalk && !io.redirect.valid
397
398
399    if (i < RabCommitWidth) {
400      walkIntSpecWen(i) := walkNeedIntDest(i) && !io.redirect.valid
401      walkPdest(i) := io.rabCommits.info(i).pdest
402    } else {
403      walkPdest(i) := io.out(i).bits.pdest
404    }
405  }
406
407  /**
408   * trace begin
409   */
410  val inVec = io.in.map(_.bits)
411  val canRobCompressVec = inVec.map(_.canRobCompress)
412  val isRVCVec = inVec.map(_.preDecodeInfo.isRVC)
413  val halfWordNumVec = (0 until RenameWidth).map{
414    i => compressMasksVec(i).asBools.zip(isRVCVec).map{
415      case (mask, isRVC) => Mux(mask, Mux(isRVC, 1.U, 2.U), 0.U)
416    }
417  }
418
419  for (i <- 0 until RenameWidth) {
420    // iretire
421    uops(i).traceBlockInPipe.iretire := Mux(canRobCompressVec(i),
422      halfWordNumVec(i).reduce(_ +& _),
423      Mux(isRVCVec(i), 1.U, 2.U)
424    )
425
426    // ilastsize
427    val j = i
428    val lastIsRVC = WireInit(false.B)
429    (j until RenameWidth).map { j =>
430      when(compressMasksVec(i)(j)) {
431        lastIsRVC := io.in(j).bits.preDecodeInfo.isRVC
432      }
433    }
434
435    uops(i).traceBlockInPipe.ilastsize := Mux(canRobCompressVec(i),
436      Mux(lastIsRVC, Ilastsize.HalfWord, Ilastsize.Word),
437      Mux(isRVCVec(i), Ilastsize.HalfWord, Ilastsize.Word)
438    )
439
440    // itype
441    uops(i).traceBlockInPipe.itype := Itype.jumpTypeGen(inVec(i).preDecodeInfo.brType, inVec(i).ldest.asTypeOf(new OpRegType), inVec(i).lsrc(0).asTypeOf((new OpRegType)))
442  }
443  /**
444   * trace end
445   */
446
447  /**
448    * How to set psrc:
449    * - bypass the pdest to psrc if previous instructions write to the same ldest as lsrc
450    * - default: psrc from RAT
451    * How to set pdest:
452    * - Mux(isMove, psrc, pdest_from_freelist).
453    *
454    * The critical path of rename lies here:
455    * When move elimination is enabled, we need to update the rat with psrc.
456    * However, psrc maybe comes from previous instructions' pdest, which comes from freelist.
457    *
458    * If we expand these logic for pdest(N):
459    * pdest(N) = Mux(isMove(N), psrc(N), freelist_out(N))
460    *          = Mux(isMove(N), Mux(bypass(N, N - 1), pdest(N - 1),
461    *                           Mux(bypass(N, N - 2), pdest(N - 2),
462    *                           ...
463    *                           Mux(bypass(N, 0),     pdest(0),
464    *                                                 rat_out(N))...)),
465    *                           freelist_out(N))
466    */
467  // a simple functional model for now
468  io.out(0).bits.pdest := Mux(isMove(0), uops(0).psrc.head, uops(0).pdest)
469
470  // psrc(n) + pdest(1)
471  val bypassCond: Vec[MixedVec[UInt]] = Wire(Vec(numRegSrc + 1, MixedVec(List.tabulate(RenameWidth-1)(i => UInt((i+1).W)))))
472  require(io.in(0).bits.srcType.size == io.in(0).bits.numSrc)
473  private val pdestLoc = io.in.head.bits.srcType.size // 2 vector src: v0, vl&vtype
474  println(s"[Rename] idx of pdest in bypassCond $pdestLoc")
475  for (i <- 1 until RenameWidth) {
476    val v0Cond = io.in(i).bits.srcType.zipWithIndex.map{ case (s, i) =>
477      if (i == 3) (s === SrcType.vp) || (s === SrcType.v0)
478      else false.B
479    } :+ needV0Dest(i)
480    val vlCond = io.in(i).bits.srcType.zipWithIndex.map{ case (s, i) =>
481      if (i == 4) s === SrcType.vp
482      else false.B
483    } :+ needVlDest(i)
484    val vecCond = io.in(i).bits.srcType.map(_ === SrcType.vp) :+ needVecDest(i)
485    val fpCond  = io.in(i).bits.srcType.map(_ === SrcType.fp) :+ needFpDest(i)
486    val intCond = io.in(i).bits.srcType.map(_ === SrcType.xp) :+ needIntDest(i)
487    val target = io.in(i).bits.lsrc :+ io.in(i).bits.ldest
488    for ((((((cond1, (condV0, condVl)), cond2), cond3), t), j) <- vecCond.zip(v0Cond.zip(vlCond)).zip(fpCond).zip(intCond).zip(target).zipWithIndex) {
489      val destToSrc = io.in.take(i).zipWithIndex.map { case (in, j) =>
490        val indexMatch = in.bits.ldest === t
491        val writeMatch =  cond3 && needIntDest(j) || cond2 && needFpDest(j) || cond1 && needVecDest(j)
492        val v0vlMatch = condV0 && needV0Dest(j) || condVl && needVlDest(j)
493        indexMatch && writeMatch || v0vlMatch
494      }
495      bypassCond(j)(i - 1) := VecInit(destToSrc).asUInt
496    }
497    io.out(i).bits.psrc(0) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(0)(i-1).asBools).foldLeft(uops(i).psrc(0)) {
498      (z, next) => Mux(next._2, next._1, z)
499    }
500    io.out(i).bits.psrc(1) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(1)(i-1).asBools).foldLeft(uops(i).psrc(1)) {
501      (z, next) => Mux(next._2, next._1, z)
502    }
503    io.out(i).bits.psrc(2) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(2)(i-1).asBools).foldLeft(uops(i).psrc(2)) {
504      (z, next) => Mux(next._2, next._1, z)
505    }
506    io.out(i).bits.psrc(3) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(3)(i-1).asBools).foldLeft(uops(i).psrc(3)) {
507      (z, next) => Mux(next._2, next._1, z)
508    }
509    io.out(i).bits.psrc(4) := io.out.take(i).map(_.bits.pdest).zip(bypassCond(4)(i-1).asBools).foldLeft(uops(i).psrc(4)) {
510      (z, next) => Mux(next._2, next._1, z)
511    }
512    io.out(i).bits.pdest := Mux(isMove(i), io.out(i).bits.psrc(0), uops(i).pdest)
513
514    // Todo: better implementation for fields reuse
515    // For fused-lui-load, load.src(0) is replaced by the imm.
516    val last_is_lui = io.in(i - 1).bits.selImm === SelImm.IMM_U && io.in(i - 1).bits.srcType(0) =/= SrcType.pc
517    val this_is_load = io.in(i).bits.fuType === FuType.ldu.U
518    val lui_to_load = io.in(i - 1).valid && io.in(i - 1).bits.ldest === io.in(i).bits.lsrc(0)
519    val fused_lui_load = last_is_lui && this_is_load && lui_to_load
520    when (fused_lui_load) {
521      // The first LOAD operand (base address) is replaced by LUI-imm and stored in imm
522      val lui_imm = io.in(i - 1).bits.imm(ImmUnion.U.len - 1, 0)
523      val ld_imm = io.in(i).bits.imm(ImmUnion.I.len - 1, 0)
524      require(io.out(i).bits.imm.getWidth >= lui_imm.getWidth + ld_imm.getWidth)
525      io.out(i).bits.srcType(0) := SrcType.imm
526      io.out(i).bits.imm := Cat(lui_imm, ld_imm)
527    }
528
529  }
530
531  val genSnapshot = Cat(io.out.map(out => out.fire && out.bits.snapshot)).orR
532  val lastCycleCreateSnpt = RegInit(false.B)
533  lastCycleCreateSnpt := genSnapshot && !io.snptIsFull
534  val sameSnptDistance = (RobCommitWidth * 4).U
535  // notInSameSnpt: 1.robidxHead - snapLastEnq >= sameSnptDistance 2.no snap
536  val notInSameSnpt = GatedValidRegNext(distanceBetween(robIdxHeadNext, io.snptLastEnq.bits) >= sameSnptDistance || !io.snptLastEnq.valid)
537  val allowSnpt = if (EnableRenameSnapshot) notInSameSnpt && !lastCycleCreateSnpt && io.in.head.bits.firstUop else false.B
538  io.out.zip(io.in).foreach{ case (out, in) => out.bits.snapshot := allowSnpt && (!in.bits.preDecodeInfo.notCFI || FuType.isJump(in.bits.fuType)) && in.fire }
539  io.out.map{ x =>
540    x.bits.hasException := Cat(selectFrontend(x.bits.exceptionVec) :+ x.bits.exceptionVec(illegalInstr) :+ x.bits.exceptionVec(virtualInstr)).orR || TriggerAction.isDmode(x.bits.trigger)
541  }
542  if(backendParams.debugEn){
543    dontTouch(robIdxHeadNext)
544    dontTouch(notInSameSnpt)
545    dontTouch(genSnapshot)
546  }
547  intFreeList.io.snpt := io.snpt
548  fpFreeList.io.snpt := io.snpt
549  vecFreeList.io.snpt := io.snpt
550  v0FreeList.io.snpt := io.snpt
551  vlFreeList.io.snpt := io.snpt
552  intFreeList.io.snpt.snptEnq := genSnapshot
553  fpFreeList.io.snpt.snptEnq := genSnapshot
554  vecFreeList.io.snpt.snptEnq := genSnapshot
555  v0FreeList.io.snpt.snptEnq := genSnapshot
556  vlFreeList.io.snpt.snptEnq := genSnapshot
557
558  /**
559    * Instructions commit: update freelist and rename table
560    */
561  for (i <- 0 until RabCommitWidth) {
562    val commitValid = io.rabCommits.isCommit && io.rabCommits.commitValid(i)
563    val walkValid = io.rabCommits.isWalk && io.rabCommits.walkValid(i)
564
565    // I. RAT Update
566    // When redirect happens (mis-prediction), don't update the rename table
567    io.intRenamePorts(i).wen  := intSpecWen(i)
568    io.intRenamePorts(i).addr := uops(i).ldest(log2Ceil(IntLogicRegs) - 1, 0)
569    io.intRenamePorts(i).data := io.out(i).bits.pdest
570
571    io.fpRenamePorts(i).wen  := fpSpecWen(i)
572    io.fpRenamePorts(i).addr := uops(i).ldest(log2Ceil(FpLogicRegs) - 1, 0)
573    io.fpRenamePorts(i).data := fpFreeList.io.allocatePhyReg(i)
574
575    io.vecRenamePorts(i).wen := vecSpecWen(i)
576    io.vecRenamePorts(i).addr := uops(i).ldest(log2Ceil(VecLogicRegs) - 1, 0)
577    io.vecRenamePorts(i).data := vecFreeList.io.allocatePhyReg(i)
578
579    io.v0RenamePorts(i).wen := v0SpecWen(i)
580    io.v0RenamePorts(i).addr := uops(i).ldest(log2Ceil(V0LogicRegs) - 1, 0)
581    io.v0RenamePorts(i).data := v0FreeList.io.allocatePhyReg(i)
582
583    io.vlRenamePorts(i).wen := vlSpecWen(i)
584    io.vlRenamePorts(i).addr := uops(i).ldest(log2Ceil(VlLogicRegs) - 1, 0)
585    io.vlRenamePorts(i).data := vlFreeList.io.allocatePhyReg(i)
586
587    // II. Free List Update
588    intFreeList.io.freeReq(i) := io.int_need_free(i)
589    intFreeList.io.freePhyReg(i) := RegNext(io.int_old_pdest(i))
590    fpFreeList.io.freeReq(i)  := GatedValidRegNext(commitValid && needDestRegCommit(Reg_F, io.rabCommits.info(i)))
591    fpFreeList.io.freePhyReg(i) := io.fp_old_pdest(i)
592    vecFreeList.io.freeReq(i)  := GatedValidRegNext(commitValid && needDestRegCommit(Reg_V, io.rabCommits.info(i)))
593    vecFreeList.io.freePhyReg(i) := io.vec_old_pdest(i)
594    v0FreeList.io.freeReq(i) := GatedValidRegNext(commitValid && needDestRegCommit(Reg_V0, io.rabCommits.info(i)))
595    v0FreeList.io.freePhyReg(i) := io.v0_old_pdest(i)
596    vlFreeList.io.freeReq(i) := GatedValidRegNext(commitValid && needDestRegCommit(Reg_Vl, io.rabCommits.info(i)))
597    vlFreeList.io.freePhyReg(i) := io.vl_old_pdest(i)
598  }
599
600  /*
601  Debug and performance counters
602   */
603  def printRenameInfo(in: DecoupledIO[DecodedInst], out: DecoupledIO[DynInst]) = {
604    XSInfo(out.fire, p"pc:${Hexadecimal(in.bits.pc)} in(${in.valid},${in.ready}) " +
605      p"lsrc(0):${in.bits.lsrc(0)} -> psrc(0):${out.bits.psrc(0)} " +
606      p"lsrc(1):${in.bits.lsrc(1)} -> psrc(1):${out.bits.psrc(1)} " +
607      p"lsrc(2):${in.bits.lsrc(2)} -> psrc(2):${out.bits.psrc(2)} " +
608      p"ldest:${in.bits.ldest} -> pdest:${out.bits.pdest}\n"
609    )
610  }
611
612  for ((x,y) <- io.in.zip(io.out)) {
613    printRenameInfo(x, y)
614  }
615
616  io.out.map { case x =>
617    when(x.valid && x.bits.rfWen){
618      assert(x.bits.ldest =/= 0.U, "rfWen cannot be 1 when Int regfile ldest is 0")
619    }
620  }
621  val debugRedirect = RegEnable(io.redirect.bits, io.redirect.valid)
622  // bad speculation
623  val recStall = io.redirect.valid || io.rabCommits.isWalk
624  val ctrlRecStall = Mux(io.redirect.valid, io.redirect.bits.debugIsCtrl, io.rabCommits.isWalk && debugRedirect.debugIsCtrl)
625  val mvioRecStall = Mux(io.redirect.valid, io.redirect.bits.debugIsMemVio, io.rabCommits.isWalk && debugRedirect.debugIsMemVio)
626  val otherRecStall = recStall && !(ctrlRecStall || mvioRecStall)
627  XSPerfAccumulate("recovery_stall", recStall)
628  XSPerfAccumulate("control_recovery_stall", ctrlRecStall)
629  XSPerfAccumulate("mem_violation_recovery_stall", mvioRecStall)
630  XSPerfAccumulate("other_recovery_stall", otherRecStall)
631  // freelist stall
632  val notRecStall = !io.out.head.valid && !recStall
633  val intFlStall = notRecStall && inHeadValid && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !intFreeList.io.canAllocate
634  val fpFlStall = notRecStall && inHeadValid && intFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !fpFreeList.io.canAllocate
635  val vecFlStall = notRecStall && inHeadValid && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !vecFreeList.io.canAllocate
636  val v0FlStall = notRecStall && inHeadValid && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && vlFreeList.io.canAllocate && !v0FreeList.io.canAllocate
637  val vlFlStall = notRecStall && inHeadValid && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && !vlFreeList.io.canAllocate
638  val multiFlStall = notRecStall && inHeadValid && (PopCount(Cat(
639    !intFreeList.io.canAllocate,
640    !fpFreeList.io.canAllocate,
641    !vecFreeList.io.canAllocate,
642    !v0FreeList.io.canAllocate,
643    !vlFreeList.io.canAllocate,
644  )) > 1.U)
645  // other stall
646  val otherStall = notRecStall && !intFlStall && !fpFlStall && !vecFlStall && !v0FlStall && !vlFlStall && !multiFlStall
647
648  io.stallReason.in.backReason.valid := io.stallReason.out.backReason.valid || !io.in.head.ready
649  io.stallReason.in.backReason.bits := Mux(io.stallReason.out.backReason.valid, io.stallReason.out.backReason.bits,
650    MuxCase(TopDownCounters.OtherCoreStall.id.U, Seq(
651      ctrlRecStall  -> TopDownCounters.ControlRecoveryStall.id.U,
652      mvioRecStall  -> TopDownCounters.MemVioRecoveryStall.id.U,
653      otherRecStall -> TopDownCounters.OtherRecoveryStall.id.U,
654      intFlStall    -> TopDownCounters.IntFlStall.id.U,
655      fpFlStall     -> TopDownCounters.FpFlStall.id.U,
656      vecFlStall    -> TopDownCounters.VecFlStall.id.U,
657      v0FlStall     -> TopDownCounters.V0FlStall.id.U,
658      vlFlStall     -> TopDownCounters.VlFlStall.id.U,
659      multiFlStall  -> TopDownCounters.MultiFlStall.id.U,
660    )
661  ))
662  io.stallReason.out.reason.zip(io.stallReason.in.reason).zip(io.in.map(_.valid)).foreach { case ((out, in), valid) =>
663    out := Mux(io.stallReason.in.backReason.valid, io.stallReason.in.backReason.bits, in)
664  }
665
666  XSDebug(io.rabCommits.isWalk, p"Walk Recovery Enabled\n")
667  XSDebug(io.rabCommits.isWalk, p"validVec:${Binary(io.rabCommits.walkValid.asUInt)}\n")
668  for (i <- 0 until RabCommitWidth) {
669    val info = io.rabCommits.info(i)
670    XSDebug(io.rabCommits.isWalk && io.rabCommits.walkValid(i), p"[#$i walk info] " +
671      p"ldest:${info.ldest} rfWen:${info.rfWen} fpWen:${info.fpWen} vecWen:${info.vecWen} v0Wen:${info.v0Wen} vlWen:${info.vlWen}")
672  }
673
674  XSDebug(p"inValidVec: ${Binary(Cat(io.in.map(_.valid)))}\n")
675
676  XSPerfAccumulate("in_valid_count", PopCount(io.in.map(_.valid)))
677  XSPerfAccumulate("in_fire_count", PopCount(io.in.map(_.fire)))
678  XSPerfAccumulate("in_valid_not_ready_count", PopCount(io.in.map(x => x.valid && !x.ready)))
679  XSPerfAccumulate("wait_cycle", !io.in.head.valid && dispatchCanAcc)
680
681  // These stall reasons could overlap each other, but we configure the priority as fellows.
682  // walk stall > dispatch stall > int freelist stall > fp freelist stall
683  private val inHeadStall = io.in.head match { case x => x.valid && !x.ready }
684  private val stallForWalk      = inHeadValid &&  io.rabCommits.isWalk
685  private val stallForDispatch  = inHeadValid && !io.rabCommits.isWalk && !dispatchCanAcc
686  private val stallForIntFL     = inHeadValid && !io.rabCommits.isWalk && dispatchCanAcc && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !intFreeList.io.canAllocate
687  private val stallForFpFL      = inHeadValid && !io.rabCommits.isWalk && dispatchCanAcc && intFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !fpFreeList.io.canAllocate
688  private val stallForVecFL     = inHeadValid && !io.rabCommits.isWalk && dispatchCanAcc && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !vecFreeList.io.canAllocate
689  private val stallForV0FL      = inHeadValid && !io.rabCommits.isWalk && dispatchCanAcc && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && vlFreeList.io.canAllocate && !v0FreeList.io.canAllocate
690  private val stallForVlFL      = inHeadValid && !io.rabCommits.isWalk && dispatchCanAcc && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && !vlFreeList.io.canAllocate
691  XSPerfAccumulate("stall_cycle",          inHeadStall)
692  XSPerfAccumulate("stall_cycle_walk",     stallForWalk)
693  XSPerfAccumulate("stall_cycle_dispatch", stallForDispatch)
694  XSPerfAccumulate("stall_cycle_int",      stallForIntFL)
695  XSPerfAccumulate("stall_cycle_fp",       stallForFpFL)
696  XSPerfAccumulate("stall_cycle_vec",      stallForVecFL)
697  XSPerfAccumulate("stall_cycle_vec",      stallForV0FL)
698  XSPerfAccumulate("stall_cycle_vec",      stallForVlFL)
699
700  XSPerfHistogram("in_valid_range",  PopCount(io.in.map(_.valid)),  true.B, 0, DecodeWidth + 1, 1)
701  XSPerfHistogram("in_fire_range",   PopCount(io.in.map(_.fire)),   true.B, 0, DecodeWidth + 1, 1)
702  XSPerfHistogram("out_valid_range", PopCount(io.out.map(_.valid)), true.B, 0, DecodeWidth + 1, 1)
703  XSPerfHistogram("out_fire_range",  PopCount(io.out.map(_.fire)),  true.B, 0, DecodeWidth + 1, 1)
704
705  XSPerfAccumulate("move_instr_count", PopCount(io.out.map(out => out.fire && out.bits.isMove)))
706  val is_fused_lui_load = io.out.map(o => o.fire && o.bits.fuType === FuType.ldu.U && o.bits.srcType(0) === SrcType.imm)
707  XSPerfAccumulate("fused_lui_load_instr_count", PopCount(is_fused_lui_load))
708
709  val renamePerf = Seq(
710    ("rename_in                  ", PopCount(io.in.map(_.valid & io.in(0).ready ))                                                               ),
711    ("rename_waitinstr           ", PopCount((0 until RenameWidth).map(i => io.in(i).valid && !io.in(i).ready))                                  ),
712    ("rename_stall               ", inHeadStall),
713    ("rename_stall_cycle_walk    ", inHeadValid &&  io.rabCommits.isWalk),
714    ("rename_stall_cycle_dispatch", inHeadValid && !io.rabCommits.isWalk && !dispatchCanAcc),
715    ("rename_stall_cycle_int     ", inHeadValid && !io.rabCommits.isWalk && dispatchCanAcc && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !intFreeList.io.canAllocate),
716    ("rename_stall_cycle_fp      ", inHeadValid && !io.rabCommits.isWalk && dispatchCanAcc && intFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !fpFreeList.io.canAllocate),
717    ("rename_stall_cycle_vec     ", inHeadValid && !io.rabCommits.isWalk && dispatchCanAcc && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && v0FreeList.io.canAllocate && vlFreeList.io.canAllocate && !vecFreeList.io.canAllocate),
718    ("rename_stall_cycle_v0      ", inHeadValid && !io.rabCommits.isWalk && dispatchCanAcc && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && vlFreeList.io.canAllocate && !v0FreeList.io.canAllocate),
719    ("rename_stall_cycle_vl      ", inHeadValid && !io.rabCommits.isWalk && dispatchCanAcc && intFreeList.io.canAllocate && fpFreeList.io.canAllocate && vecFreeList.io.canAllocate && v0FreeList.io.canAllocate && !vlFreeList.io.canAllocate),
720  )
721  val intFlPerf = intFreeList.getPerfEvents
722  val fpFlPerf = fpFreeList.getPerfEvents
723  val vecFlPerf = vecFreeList.getPerfEvents
724  val v0FlPerf = v0FreeList.getPerfEvents
725  val vlFlPerf = vlFreeList.getPerfEvents
726  val perfEvents = renamePerf ++ intFlPerf ++ fpFlPerf ++ vecFlPerf ++ v0FlPerf ++ vlFlPerf
727  generatePerfEvent()
728}
729