xref: /XiangShan/src/main/scala/xiangshan/mem/pipeline/StoreUnit.scala (revision e4d4d30585412eb8ac83b5c75599a348356342a2)
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.mem
18
19import org.chipsalliance.cde.config.Parameters
20import chisel3._
21import chisel3.util._
22import utils._
23import utility._
24import xiangshan.ExceptionNO._
25import xiangshan._
26import xiangshan.backend.Bundles.{MemExuInput, MemExuOutput}
27import xiangshan.backend.fu.PMPRespBundle
28import xiangshan.backend.fu.FuConfig._
29import xiangshan.backend.fu.FuType._
30import xiangshan.backend.ctrlblock.DebugLsInfoBundle
31import xiangshan.cache.mmu.{TlbCmd, TlbReq, TlbRequestIO, TlbResp}
32import xiangshan.cache.{DcacheStoreRequestIO, DCacheStoreIO, MemoryOpConstants, HasDCacheParameters, StorePrefetchReq}
33
34class StoreUnit(implicit p: Parameters) extends XSModule with HasDCacheParameters {
35  val io = IO(new Bundle() {
36    val redirect        = Flipped(ValidIO(new Redirect))
37    val stin            = Flipped(Decoupled(new MemExuInput))
38    val issue           = Valid(new MemExuInput)
39    val tlb             = new TlbRequestIO()
40    val dcache          = new DCacheStoreIO
41    val pmp             = Flipped(new PMPRespBundle())
42    val lsq             = ValidIO(new LsPipelineBundle)
43    val lsq_replenish   = Output(new LsPipelineBundle())
44    val feedback_slow   = ValidIO(new RSFeedback)
45    val prefetch_req    = Flipped(DecoupledIO(new StorePrefetchReq))
46    // provide prefetch info to sms
47    val prefetch_train  = ValidIO(new StPrefetchTrainBundle())
48    val stld_nuke_query = Valid(new StoreNukeQueryIO)
49    val stout           = DecoupledIO(new MemExuOutput) // writeback store
50    // store mask, send to sq in store_s0
51    val st_mask_out     = Valid(new StoreMaskBundle)
52    val debug_ls        = Output(new DebugLsInfoBundle)
53    // vector
54    val vecstin           = Flipped(Decoupled(new VecStorePipeBundle()))
55    val vec_isFirstIssue  = Input(Bool())
56    val lsq_vec           = ValidIO(new LsPipelineBundle) // nuke check between vector stores and scalar loads
57    val vec_feedback_slow = ValidIO(new VSFQFeedback)
58  })
59
60  val s1_ready, s2_ready, s3_ready = WireInit(false.B)
61
62  // Pipeline
63  // --------------------------------------------------------------------------------
64  // stage 0
65  // --------------------------------------------------------------------------------
66  // generate addr, use addr to query DCache and DTLB
67  val s0_iss_valid    = io.stin.valid
68  val s0_prf_valid    = io.prefetch_req.valid && io.dcache.req.ready
69  val s0_vec_valid    = io.vecstin.valid
70  val s0_valid        = s0_iss_valid || s0_prf_valid || s0_vec_valid
71  val s0_use_flow_vec = s0_vec_valid
72  val s0_use_flow_rs  = s0_iss_valid && !s0_vec_valid
73  val s0_use_flow_prf = !s0_iss_valid && !s0_vec_valid && s0_prf_valid
74  val s0_stin         = Mux(s0_use_flow_rs, io.stin.bits, 0.U.asTypeOf(io.stin.bits))
75  val s0_vecstin      = Mux(s0_use_flow_vec, io.vecstin.bits, 0.U.asTypeOf(io.vecstin.bits))
76  val s0_uop          = Mux(s0_use_flow_rs, s0_stin.uop, s0_vecstin.uop)
77  val s0_isFirstIssue = s0_use_flow_rs && io.stin.bits.isFirstIssue || s0_use_flow_vec && io.vec_isFirstIssue
78  val s0_rsIdx        = Mux(s0_use_flow_rs, io.stin.bits.iqIdx, 0.U)
79  val s0_size         = Mux(
80    s0_use_flow_rs,
81    LSUOpType.size(s0_uop.fuOpType),
82    Mux(
83      s0_use_flow_vec,
84      io.vecstin.bits.alignedType,
85      3.U
86    )
87  )
88  val s0_mem_idx      = Mux(s0_use_flow_rs || s0_use_flow_vec, s0_uop.sqIdx.value, 0.U)
89  val s0_rob_idx      = Mux(s0_use_flow_rs || s0_use_flow_vec, s0_uop.robIdx, 0.U.asTypeOf(s0_uop.robIdx))
90  val s0_pc           = Mux(s0_use_flow_rs || s0_use_flow_vec, s0_uop.pc, 0.U)
91  val s0_instr_type   = Mux(s0_use_flow_rs || s0_use_flow_vec, STORE_SOURCE.U, DCACHE_PREFETCH_SOURCE.U)
92  val s0_wlineflag    = Mux(s0_use_flow_rs, s0_uop.fuOpType === LSUOpType.cbo_zero, false.B)
93  val s0_out          = Wire(new LsPipelineBundle)
94  val s0_kill         = s0_uop.robIdx.needFlush(io.redirect)
95  val s0_can_go       = s1_ready
96  val s0_fire         = s0_valid && !s0_kill && s0_can_go
97  // vector
98  val s0_exp          = !s0_use_flow_vec || s0_vecstin.exp
99  val s0_flowPtr      = s0_vecstin.flowPtr
100
101  // generate addr
102  // val saddr = s0_in.bits.src(0) + SignExt(s0_in.bits.uop.imm(11,0), VAddrBits)
103  val imm12 = WireInit(s0_uop.imm(11,0))
104  val saddr_lo = s0_stin.src(0)(11,0) + Cat(0.U(1.W), imm12)
105  val saddr_hi = Mux(saddr_lo(12),
106    Mux(imm12(11), s0_stin.src(0)(VAddrBits-1, 12), s0_stin.src(0)(VAddrBits-1, 12)+1.U),
107    Mux(imm12(11), s0_stin.src(0)(VAddrBits-1, 12)+SignExt(1.U, VAddrBits-12), s0_stin.src(0)(VAddrBits-1, 12)),
108  )
109  val s0_saddr = Cat(saddr_hi, saddr_lo(11,0))
110  val s0_vaddr = Mux(
111    s0_use_flow_rs,
112    s0_saddr,
113    Mux(
114      s0_use_flow_vec,
115      s0_vecstin.vaddr,
116      io.prefetch_req.bits.vaddr
117    )
118  )
119  val s0_mask  = Mux(
120    s0_use_flow_rs,
121    genVWmask(s0_saddr, s0_uop.fuOpType(1,0)),
122    Mux(
123      s0_use_flow_vec,
124      s0_vecstin.mask,
125      // -1.asSInt.asUInt
126      Fill(VLEN/8, 1.U(1.W))
127    )
128  )
129
130  io.tlb.req.valid                   := s0_valid
131  io.tlb.req.bits.vaddr              := s0_vaddr
132  io.tlb.req.bits.cmd                := TlbCmd.write
133  io.tlb.req.bits.size               := s0_size
134  io.tlb.req.bits.kill               := false.B
135  io.tlb.req.bits.memidx.is_ld       := false.B
136  io.tlb.req.bits.memidx.is_st       := true.B
137  io.tlb.req.bits.memidx.idx         := s0_mem_idx
138  io.tlb.req.bits.debug.robIdx       := s0_rob_idx
139  io.tlb.req.bits.no_translate       := false.B
140  io.tlb.req.bits.debug.pc           := s0_pc
141  io.tlb.req.bits.debug.isFirstIssue := s0_isFirstIssue
142  io.tlb.req_kill                    := false.B
143
144  // Dcache access here: not **real** dcache write
145  // just read meta and tag in dcache, to find out the store will hit or miss
146
147  // NOTE: The store request does not wait for the dcache to be ready.
148  //       If the dcache is not ready at this time, the dcache is not queried.
149  //       But, store prefetch request will always wait for dcache to be ready to make progress.
150  io.dcache.req.valid              := s0_fire
151  io.dcache.req.bits.cmd           := MemoryOpConstants.M_PFW
152  io.dcache.req.bits.vaddr         := s0_vaddr
153  io.dcache.req.bits.instrtype     := s0_instr_type
154
155  s0_out              := DontCare
156  s0_out.vaddr        := s0_vaddr
157  // Now data use its own io
158  // s1_out.data := genWdata(s1_in.src(1), s1_in.uop.fuOpType(1,0))
159  s0_out.data         := Mux(s0_use_flow_rs, s0_stin.src(1), s0_vecstin.src_vs3) // FIXME: remove data from pipeline
160  s0_out.uop          := s0_uop
161  s0_out.miss         := false.B
162  s0_out.rsIdx        := s0_rsIdx
163  s0_out.mask         := s0_mask
164  s0_out.isFirstIssue := s0_isFirstIssue
165  s0_out.isHWPrefetch := s0_use_flow_prf
166  s0_out.wlineflag    := s0_wlineflag
167  s0_out.isvec        := s0_use_flow_vec
168  s0_out.is128bit     := false.B
169  s0_out.exp          := s0_exp
170  s0_out.sflowPtr      := s0_flowPtr
171  when(s0_valid && s0_isFirstIssue) {
172    s0_out.uop.debugInfo.tlbFirstReqTime := GTimer()
173  }
174
175  // exception check
176  val s0_addr_aligned = LookupTree(Mux(s0_use_flow_vec, s0_vecstin.alignedType, s0_uop.fuOpType(1, 0)), List(
177    "b00".U   -> true.B,              //b
178    "b01".U   -> (s0_out.vaddr(0) === 0.U),   //h
179    "b10".U   -> (s0_out.vaddr(1,0) === 0.U), //w
180    "b11".U   -> (s0_out.vaddr(2,0) === 0.U)  //d
181  ))
182  s0_out.uop.exceptionVec(storeAddrMisaligned) := Mux(s0_use_flow_rs || s0_use_flow_vec, !s0_addr_aligned, false.B)
183
184  io.st_mask_out.valid       := s0_use_flow_rs
185  io.st_mask_out.bits.mask   := s0_out.mask
186  io.st_mask_out.bits.sqIdx  := s0_out.uop.sqIdx
187
188  io.stin.ready := s1_ready && s0_use_flow_rs
189  io.vecstin.ready := s1_ready && s0_use_flow_vec
190  io.prefetch_req.ready := s1_ready && io.dcache.req.ready && !s0_iss_valid && !s0_vec_valid
191
192  // Pipeline
193  // --------------------------------------------------------------------------------
194  // stage 1
195  // --------------------------------------------------------------------------------
196  // TLB resp (send paddr to dcache)
197  val s1_valid  = RegInit(false.B)
198  val s1_in     = RegEnable(s0_out, s0_fire)
199  val s1_out    = Wire(new LsPipelineBundle)
200  val s1_kill   = Wire(Bool())
201  val s1_can_go = s2_ready
202  val s1_fire   = s1_valid && !s1_kill && s1_can_go
203  val s1_exp    = RegEnable(s0_out.exp, true.B, s0_fire)
204
205  // mmio cbo decoder
206  val s1_mmio_cbo  = s1_in.uop.fuOpType === LSUOpType.cbo_clean ||
207                     s1_in.uop.fuOpType === LSUOpType.cbo_flush ||
208                     s1_in.uop.fuOpType === LSUOpType.cbo_inval
209  val s1_paddr     = io.tlb.resp.bits.paddr(0)
210  val s1_tlb_miss  = io.tlb.resp.bits.miss
211  val s1_mmio      = s1_mmio_cbo
212  val s1_exception = ExceptionNO.selectByFu(s1_out.uop.exceptionVec, StaCfg).asUInt.orR
213  val s1_isvec     = RegEnable(s0_out.isvec, false.B, s0_fire)
214  s1_kill := s1_in.uop.robIdx.needFlush(io.redirect) || s1_tlb_miss
215
216  s1_ready := !s1_valid || s1_kill || s2_ready
217  io.tlb.resp.ready := true.B // TODO: why dtlbResp needs a ready?
218  when (s0_fire) { s1_valid := true.B }
219  .elsewhen (s1_fire) { s1_valid := false.B }
220  .elsewhen (s1_kill) { s1_valid := false.B }
221
222  // st-ld violation dectect request.
223  io.stld_nuke_query.valid       := s1_valid && !s1_tlb_miss && !s1_in.isHWPrefetch
224  io.stld_nuke_query.bits.robIdx := s1_in.uop.robIdx
225  io.stld_nuke_query.bits.paddr  := s1_paddr
226  io.stld_nuke_query.bits.mask   := s1_in.mask
227
228  // issue
229  io.issue.valid := s1_valid && !s1_tlb_miss && !s1_in.isHWPrefetch && !s1_isvec
230  io.issue.bits  := RegEnable(s0_stin, s0_valid)
231
232
233  // Send TLB feedback to store issue queue
234  // Store feedback is generated in store_s1, sent to RS in store_s2
235  val s1_feedback = Wire(Valid(new RSFeedback))
236  s1_feedback.valid                 := s1_valid & !s1_in.isHWPrefetch && !s1_isvec
237  s1_feedback.bits.hit              := !s1_tlb_miss
238  s1_feedback.bits.flushState       := io.tlb.resp.bits.ptwBack
239  s1_feedback.bits.robIdx           := s1_out.uop.robIdx
240  s1_feedback.bits.sourceType       := RSFeedbackType.tlbMiss
241  s1_feedback.bits.dataInvalidSqIdx := DontCare
242
243  XSDebug(s1_feedback.valid,
244    "S1 Store: tlbHit: %d robIdx: %d\n",
245    s1_feedback.bits.hit,
246    s1_feedback.bits.robIdx.value
247  )
248
249  val s1_vec_feedback = Wire(Valid(new VSFQFeedback))
250  s1_vec_feedback.valid                 := s1_valid && !s1_in.isHWPrefetch && s1_isvec
251  s1_vec_feedback.bits.flowPtr          := s1_out.sflowPtr
252  s1_vec_feedback.bits.hit              := !s1_tlb_miss
253  s1_vec_feedback.bits.sourceType       := RSFeedbackType.tlbMiss
254  s1_vec_feedback.bits.paddr            := s1_paddr
255  XSDebug(s1_vec_feedback.valid,
256    "Vector S1 Store: tlbHit: %d flowPtr: %d\n",
257    s1_vec_feedback.bits.hit,
258    s1_vec_feedback.bits.flowPtr.value
259  )
260
261  // io.feedback_slow := s1_feedback
262  // io.vec_feedback_slow := s1_vec_feedback
263
264  // get paddr from dtlb, check if rollback is needed
265  // writeback store inst to lsq
266  s1_out         := s1_in
267  s1_out.paddr   := s1_paddr
268  s1_out.miss    := false.B
269  s1_out.mmio    := s1_mmio
270  s1_out.tlbMiss := s1_tlb_miss
271  s1_out.atomic  := s1_mmio
272  s1_out.uop.exceptionVec(storePageFault)   := io.tlb.resp.bits.excp(0).pf.st && s1_exp
273  s1_out.uop.exceptionVec(storeAccessFault) := io.tlb.resp.bits.excp(0).af.st && s1_exp
274
275  // scalar store and scalar load nuke check, and also other purposes
276  io.lsq.valid     := s1_valid && !s1_in.isHWPrefetch && !s1_isvec
277  io.lsq.bits      := s1_out
278  io.lsq.bits.miss := s1_tlb_miss
279  // vector store and scalar load nuke check
280  io.lsq_vec.valid := s1_valid && !s1_in.isHWPrefetch && s1_isvec
281  io.lsq_vec.bits  := s1_out
282  io.lsq_vec.bits.miss := s1_tlb_miss
283
284  // kill dcache write intent request when tlb miss or exception
285  io.dcache.s1_kill  := (s1_tlb_miss || s1_exception || s1_mmio || s1_in.uop.robIdx.needFlush(io.redirect))
286  io.dcache.s1_paddr := s1_paddr
287
288  // write below io.out.bits assign sentence to prevent overwriting values
289  val s1_tlb_memidx = io.tlb.resp.bits.memidx
290  when(s1_tlb_memidx.is_st && io.tlb.resp.valid && !s1_tlb_miss && s1_tlb_memidx.idx === s1_out.uop.sqIdx.value) {
291    // printf("Store idx = %d\n", s1_tlb_memidx.idx)
292    s1_out.uop.debugInfo.tlbRespTime := GTimer()
293  }
294
295  // Pipeline
296  // --------------------------------------------------------------------------------
297  // stage 2
298  // --------------------------------------------------------------------------------
299  // mmio check
300  val s2_valid  = RegInit(false.B)
301  val s2_in     = RegEnable(s1_out, s1_fire)
302  val s2_out    = Wire(new LsPipelineBundle)
303  val s2_kill   = Wire(Bool())
304  val s2_can_go = s3_ready
305  val s2_fire   = s2_valid && !s2_kill && s2_can_go
306  val s2_exp    = RegEnable(s1_out.exp, true.B, s1_fire)
307
308  s2_ready := !s2_valid || s2_kill || s3_ready
309  when (s1_fire) { s2_valid := true.B }
310  .elsewhen (s2_fire) { s2_valid := false.B }
311  .elsewhen (s2_kill) { s2_valid := false.B }
312
313  val s2_pmp = WireInit(io.pmp)
314
315  val s2_exception = ExceptionNO.selectByFu(s2_out.uop.exceptionVec, StaCfg).asUInt.orR
316  val s2_mmio = s2_in.mmio || s2_pmp.mmio
317  s2_kill := (s2_mmio && !s2_exception) || s2_in.uop.robIdx.needFlush(io.redirect)
318
319  s2_out        := s2_in
320  s2_out.mmio   := s2_mmio && !s2_exception
321  s2_out.atomic := s2_in.atomic || s2_pmp.atomic
322  s2_out.uop.exceptionVec(storeAccessFault) := (s2_in.uop.exceptionVec(storeAccessFault) || s2_pmp.st) && s2_exp
323
324  // kill dcache write intent request when mmio or exception
325  io.dcache.s2_kill := (s2_mmio || s2_exception || s2_in.uop.robIdx.needFlush(io.redirect))
326  io.dcache.s2_pc   := s2_out.uop.pc
327  // TODO: dcache resp
328  io.dcache.resp.ready := true.B
329
330  // feedback tlb miss to RS in store_s2
331  io.feedback_slow.valid := RegNext(s1_feedback.valid && !s1_out.uop.robIdx.needFlush(io.redirect))
332  io.feedback_slow.bits  := RegNext(s1_feedback.bits)
333
334  // vector feedback
335  io.vec_feedback_slow.valid := RegNext(s1_vec_feedback.valid && !s1_out.uop.robIdx.needFlush(io.redirect))
336  io.vec_feedback_slow.bits  := RegNext(s1_vec_feedback.bits)
337
338  // mmio and exception
339  io.lsq_replenish := s2_out
340
341  // prefetch related
342  io.lsq_replenish.miss := io.dcache.resp.fire && io.dcache.resp.bits.miss // miss info
343
344  // RegNext prefetch train for better timing
345  // ** Now, prefetch train is valid at store s3 **
346  io.prefetch_train.bits.fromLsPipelineBundle(s2_in, latch = true)
347  // override miss bit
348  io.prefetch_train.bits.miss := RegNext(io.dcache.resp.bits.miss)
349  // TODO: add prefetch and access bit
350  io.prefetch_train.bits.meta_prefetch := false.B
351  io.prefetch_train.bits.meta_access := false.B
352  if(EnableStorePrefetchSMS) {
353    io.prefetch_train.valid := RegNext(s2_valid && io.dcache.resp.fire && !s2_out.mmio && !s2_in.tlbMiss && !s2_in.isHWPrefetch)
354  }else {
355    io.prefetch_train.valid := false.B
356  }
357
358  // Pipeline
359  // --------------------------------------------------------------------------------
360  // stage 3
361  // --------------------------------------------------------------------------------
362  // store write back
363  val s3_valid  = RegInit(false.B)
364  val s3_in     = RegEnable(s2_out, s2_fire)
365  val s3_out    = Wire(new MemExuOutput)
366  val s3_kill   = s3_in.uop.robIdx.needFlush(io.redirect)
367  val s3_can_go = s3_ready
368  val s3_fire   = s3_valid && !s3_kill && s3_can_go
369
370  when (s2_fire) { s3_valid := (!s2_mmio || s2_exception) && !s2_out.isHWPrefetch  }
371  .elsewhen (s3_fire) { s3_valid := false.B }
372  .elsewhen (s3_kill) { s3_valid := false.B }
373
374  // wb: writeback
375  val SelectGroupSize   = RollbackGroupSize
376  val lgSelectGroupSize = log2Ceil(SelectGroupSize)
377  val TotalSelectCycles = scala.math.ceil(log2Ceil(LoadQueueRAWSize).toFloat / lgSelectGroupSize).toInt + 1
378
379  s3_out                 := DontCare
380  s3_out.uop             := s3_in.uop
381  s3_out.data            := DontCare
382  s3_out.debug.isMMIO    := s3_in.mmio
383  s3_out.debug.paddr     := s3_in.paddr
384  s3_out.debug.vaddr     := s3_in.vaddr
385  s3_out.debug.isPerfCnt := false.B
386
387  // Pipeline
388  // --------------------------------------------------------------------------------
389  // stage x
390  // --------------------------------------------------------------------------------
391  // delay TotalSelectCycles - 2 cycle(s)
392  val TotalDelayCycles = TotalSelectCycles - 2
393  val sx_valid = Wire(Vec(TotalDelayCycles + 1, Bool()))
394  val sx_ready = Wire(Vec(TotalDelayCycles + 1, Bool()))
395  val sx_in    = Wire(Vec(TotalDelayCycles + 1, new MemExuOutput))
396
397  // backward ready signal
398  s3_ready := sx_ready.head
399  for (i <- 0 until TotalDelayCycles + 1) {
400    if (i == 0) {
401      sx_valid(i) := s3_valid
402      sx_in(i)    := s3_out
403      sx_ready(i) := !s3_valid(i) || sx_in(i).uop.robIdx.needFlush(io.redirect) || (if (TotalDelayCycles == 0) io.stout.ready else sx_ready(i+1))
404    } else {
405      val cur_kill   = sx_in(i).uop.robIdx.needFlush(io.redirect)
406      val cur_can_go = (if (i == TotalDelayCycles) io.stout.ready else sx_ready(i+1))
407      val cur_fire   = sx_valid(i) && !cur_kill && cur_can_go
408      val prev_fire  = sx_valid(i-1) && !sx_in(i-1).uop.robIdx.needFlush(io.redirect) && sx_ready(i)
409
410      sx_ready(i) := !sx_valid(i) || cur_kill || (if (i == TotalDelayCycles) io.stout.ready else sx_ready(i+1))
411      val sx_valid_can_go = prev_fire || cur_fire || cur_kill
412      sx_valid(i) := RegEnable(Mux(prev_fire, true.B, false.B), false.B, sx_valid_can_go)
413      sx_in(i) := RegEnable(sx_in(i-1), prev_fire)
414    }
415  }
416  val sx_last_valid = sx_valid.takeRight(1).head
417  val sx_last_ready = sx_ready.takeRight(1).head
418  val sx_last_in    = sx_in.takeRight(1).head
419  sx_last_ready := !sx_last_valid || sx_last_in.uop.robIdx.needFlush(io.redirect) || io.stout.ready
420
421  io.stout.valid := sx_last_valid && !sx_last_in.uop.robIdx.needFlush(io.redirect) && isStore(sx_last_in.uop.fuType)
422  io.stout.bits := sx_last_in
423
424  io.debug_ls := DontCare
425  io.debug_ls.s1.isTlbFirstMiss := io.tlb.resp.valid && io.tlb.resp.bits.miss && io.tlb.resp.bits.debug.isFirstIssue && !s1_in.isHWPrefetch
426  io.debug_ls.s1_robIdx := s1_in.uop.robIdx.value
427
428  private def printPipeLine(pipeline: LsPipelineBundle, cond: Bool, name: String): Unit = {
429    XSDebug(cond,
430      p"$name" + p" pc ${Hexadecimal(pipeline.uop.pc)} " +
431        p"addr ${Hexadecimal(pipeline.vaddr)} -> ${Hexadecimal(pipeline.paddr)} " +
432        p"op ${Binary(pipeline.uop.fuOpType)} " +
433        p"data ${Hexadecimal(pipeline.data)} " +
434        p"mask ${Hexadecimal(pipeline.mask)}\n"
435    )
436  }
437
438  printPipeLine(s0_out, s0_valid, "S0")
439  printPipeLine(s1_out, s1_valid, "S1")
440
441  // perf cnt
442  XSPerfAccumulate("s0_in_valid",                s0_valid)
443  XSPerfAccumulate("s0_in_fire",                 s0_fire)
444  XSPerfAccumulate("s0_in_fire_first_issue",     s0_fire && s0_isFirstIssue)
445  XSPerfAccumulate("s0_addr_spec_success",       s0_fire && !s0_use_flow_vec && s0_saddr(VAddrBits-1, 12) === s0_stin.src(0)(VAddrBits-1, 12))
446  XSPerfAccumulate("s0_addr_spec_failed",        s0_fire && !s0_use_flow_vec && s0_saddr(VAddrBits-1, 12) =/= s0_stin.src(0)(VAddrBits-1, 12))
447  XSPerfAccumulate("s0_addr_spec_success_once",  s0_fire && !s0_use_flow_vec && s0_saddr(VAddrBits-1, 12) === s0_stin.src(0)(VAddrBits-1, 12) && s0_isFirstIssue)
448  XSPerfAccumulate("s0_addr_spec_failed_once",   s0_fire && !s0_use_flow_vec && s0_saddr(VAddrBits-1, 12) =/= s0_stin.src(0)(VAddrBits-1, 12) && s0_isFirstIssue)
449
450  XSPerfAccumulate("s1_in_valid",                s1_valid)
451  XSPerfAccumulate("s1_in_fire",                 s1_fire)
452  XSPerfAccumulate("s1_in_fire_first_issue",     s1_fire && s1_in.isFirstIssue)
453  XSPerfAccumulate("s1_tlb_miss",                s1_fire && s1_tlb_miss)
454  XSPerfAccumulate("s1_tlb_miss_first_issue",    s1_fire && s1_tlb_miss && s1_in.isFirstIssue)
455  // end
456}