xref: /XiangShan/src/main/scala/xiangshan/mem/pipeline/StoreUnit.scala (revision a4d1b2d1ae4c6149f55fbcac48749c08714bfe0c)
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    val vecstout        = DecoupledIO(new VecPipelineFeedbackIO(isVStore = true))
51    // store mask, send to sq in store_s0
52    val st_mask_out     = Valid(new StoreMaskBundle)
53    val debug_ls        = Output(new DebugLsInfoBundle)
54    // vector
55    val vecstin           = Flipped(Decoupled(new VecPipeBundle(isVStore = true)))
56    val vec_isFirstIssue  = Input(Bool())
57  })
58
59  val s1_ready, s2_ready, s3_ready = WireInit(false.B)
60
61  // Pipeline
62  // --------------------------------------------------------------------------------
63  // stage 0
64  // --------------------------------------------------------------------------------
65  // generate addr, use addr to query DCache and DTLB
66  val s0_iss_valid    = io.stin.valid
67  val s0_prf_valid    = io.prefetch_req.valid && io.dcache.req.ready
68  val s0_vec_valid    = io.vecstin.valid
69  val s0_valid        = s0_iss_valid || s0_prf_valid || s0_vec_valid
70  val s0_use_flow_vec = s0_vec_valid
71  val s0_use_flow_rs  = s0_iss_valid && !s0_vec_valid
72  val s0_use_flow_prf = !s0_iss_valid && !s0_vec_valid && s0_prf_valid
73  val s0_stin         = Mux(s0_use_flow_rs, io.stin.bits, 0.U.asTypeOf(io.stin.bits))
74  val s0_vecstin      = Mux(s0_use_flow_vec, io.vecstin.bits, 0.U.asTypeOf(io.vecstin.bits))
75  val s0_uop          = Mux(s0_use_flow_rs, s0_stin.uop, s0_vecstin.uop)
76  val s0_isFirstIssue = s0_use_flow_rs && io.stin.bits.isFirstIssue || s0_use_flow_vec && io.vec_isFirstIssue
77  val s0_rsIdx        = Mux(s0_use_flow_rs, io.stin.bits.iqIdx, 0.U)
78  val s0_size         = Mux(s0_use_flow_rs || s0_use_flow_vec, s0_uop.fuOpType(2,0), 0.U)// may broken if use it in feature
79  val s0_mem_idx      = Mux(s0_use_flow_rs || s0_use_flow_vec, s0_uop.sqIdx.value, 0.U)
80  val s0_rob_idx      = Mux(s0_use_flow_rs || s0_use_flow_vec, s0_uop.robIdx, 0.U.asTypeOf(s0_uop.robIdx))
81  val s0_pc           = Mux(s0_use_flow_rs || s0_use_flow_vec, s0_uop.pc, 0.U)
82  val s0_instr_type   = Mux(s0_use_flow_rs || s0_use_flow_vec, STORE_SOURCE.U, DCACHE_PREFETCH_SOURCE.U)
83  val s0_wlineflag    = Mux(s0_use_flow_rs, s0_uop.fuOpType === LSUOpType.cbo_zero, false.B)
84  val s0_out          = Wire(new LsPipelineBundle)
85  val s0_kill         = s0_uop.robIdx.needFlush(io.redirect)
86  val s0_can_go       = s1_ready
87  val s0_fire         = s0_valid && !s0_kill && s0_can_go
88  // vector
89  val s0_vecActive    = !s0_use_flow_vec || s0_vecstin.vecActive
90  // val s0_flowPtr      = s0_vecstin.flowPtr
91  // val s0_isLastElem   = s0_vecstin.isLastElem
92  val s0_secondInv    = s0_vecstin.usSecondInv
93  val s0_elemIdx      = s0_vecstin.elemIdx
94  val s0_alignedType  = s0_vecstin.alignedType
95  val s0_mBIndex      = s0_vecstin.mBIndex
96
97  // generate addr
98  // val saddr = s0_in.bits.src(0) + SignExt(s0_in.bits.uop.imm(11,0), VAddrBits)
99  val imm12 = WireInit(s0_uop.imm(11,0))
100  val saddr_lo = s0_stin.src(0)(11,0) + Cat(0.U(1.W), imm12)
101  val saddr_hi = Mux(saddr_lo(12),
102    Mux(imm12(11), s0_stin.src(0)(VAddrBits-1, 12), s0_stin.src(0)(VAddrBits-1, 12)+1.U),
103    Mux(imm12(11), s0_stin.src(0)(VAddrBits-1, 12)+SignExt(1.U, VAddrBits-12), s0_stin.src(0)(VAddrBits-1, 12)),
104  )
105  val s0_saddr = Cat(saddr_hi, saddr_lo(11,0))
106  val s0_vaddr = Mux(
107    s0_use_flow_rs,
108    s0_saddr,
109    Mux(
110      s0_use_flow_vec,
111      s0_vecstin.vaddr,
112      io.prefetch_req.bits.vaddr
113    )
114  )
115  val s0_mask  = Mux(
116    s0_use_flow_rs,
117    genVWmask128(s0_saddr, s0_uop.fuOpType(2,0)),
118    Mux(
119      s0_use_flow_vec,
120      s0_vecstin.mask,
121      // -1.asSInt.asUInt
122      Fill(VLEN/8, 1.U(1.W))
123    )
124  )
125
126  io.tlb.req.valid                   := s0_valid
127  io.tlb.req.bits.vaddr              := s0_vaddr
128  io.tlb.req.bits.cmd                := TlbCmd.write
129  io.tlb.req.bits.size               := s0_size
130  io.tlb.req.bits.kill               := false.B
131  io.tlb.req.bits.memidx.is_ld       := false.B
132  io.tlb.req.bits.memidx.is_st       := true.B
133  io.tlb.req.bits.memidx.idx         := s0_mem_idx
134  io.tlb.req.bits.debug.robIdx       := s0_rob_idx
135  io.tlb.req.bits.no_translate       := false.B
136  io.tlb.req.bits.debug.pc           := s0_pc
137  io.tlb.req.bits.debug.isFirstIssue := s0_isFirstIssue
138  io.tlb.req_kill                    := false.B
139  io.tlb.req.bits.hyperinst          := LSUOpType.isHsv(s0_uop.fuOpType)
140  io.tlb.req.bits.hlvx               := false.B
141
142  // Dcache access here: not **real** dcache write
143  // just read meta and tag in dcache, to find out the store will hit or miss
144
145  // NOTE: The store request does not wait for the dcache to be ready.
146  //       If the dcache is not ready at this time, the dcache is not queried.
147  //       But, store prefetch request will always wait for dcache to be ready to make progress.
148  io.dcache.req.valid              := s0_fire
149  io.dcache.req.bits.cmd           := MemoryOpConstants.M_PFW
150  io.dcache.req.bits.vaddr         := s0_vaddr
151  io.dcache.req.bits.instrtype     := s0_instr_type
152
153  s0_out              := DontCare
154  s0_out.vaddr        := s0_vaddr
155  // Now data use its own io
156  // s1_out.data := genWdata(s1_in.src(1), s1_in.uop.fuOpType(1,0))
157  s0_out.data         := s0_stin.src(1)
158  s0_out.uop          := s0_uop
159  s0_out.miss         := false.B
160  s0_out.rsIdx        := s0_rsIdx
161  s0_out.mask         := s0_mask
162  s0_out.isFirstIssue := s0_isFirstIssue
163  s0_out.isHWPrefetch := s0_use_flow_prf
164  s0_out.wlineflag    := s0_wlineflag
165  s0_out.isvec        := s0_use_flow_vec
166  s0_out.is128bit     := false.B
167  s0_out.vecActive    := s0_vecActive
168  s0_out.usSecondInv  := s0_secondInv
169  s0_out.elemIdx      := s0_elemIdx
170  s0_out.alignedType  := s0_alignedType
171  s0_out.mbIndex      := s0_mBIndex
172  when(s0_valid && s0_isFirstIssue) {
173    s0_out.uop.debugInfo.tlbFirstReqTime := GTimer()
174  }
175
176  // exception check
177  val s0_addr_aligned = LookupTree(Mux(s0_use_flow_vec, s0_vecstin.alignedType(1,0), s0_uop.fuOpType(1, 0)), List(
178    "b00".U   -> true.B,              //b
179    "b01".U   -> (s0_out.vaddr(0) === 0.U),   //h
180    "b10".U   -> (s0_out.vaddr(1,0) === 0.U), //w
181    "b11".U   -> (s0_out.vaddr(2,0) === 0.U)  //d
182  ))
183  // if vector store sends 128-bit requests, its address must be 128-aligned
184  XSError(s0_use_flow_vec && s0_out.vaddr(3, 0) =/= 0.U && s0_vecstin.alignedType(2), "unit stride 128 bit element is not aligned!")
185  s0_out.uop.exceptionVec(storeAddrMisaligned) := Mux(s0_use_flow_rs || s0_use_flow_vec, !s0_addr_aligned, false.B)
186
187  io.st_mask_out.valid       := s0_use_flow_rs || s0_use_flow_vec
188  io.st_mask_out.bits.mask   := s0_out.mask
189  io.st_mask_out.bits.sqIdx  := s0_out.uop.sqIdx
190
191  io.stin.ready := s1_ready && s0_use_flow_rs
192  io.vecstin.ready := s1_ready && s0_use_flow_vec
193  io.prefetch_req.ready := s1_ready && io.dcache.req.ready && !s0_iss_valid && !s0_vec_valid
194
195  // Pipeline
196  // --------------------------------------------------------------------------------
197  // stage 1
198  // --------------------------------------------------------------------------------
199  // TLB resp (send paddr to dcache)
200  val s1_valid  = RegInit(false.B)
201  val s1_in     = RegEnable(s0_out, s0_fire)
202  val s1_out    = Wire(new LsPipelineBundle)
203  val s1_kill   = Wire(Bool())
204  val s1_can_go = s2_ready
205  val s1_fire   = s1_valid && !s1_kill && s1_can_go
206  val s1_vecActive    = RegEnable(s0_out.vecActive, true.B, s0_fire)
207
208  // mmio cbo decoder
209  val s1_mmio_cbo  = s1_in.uop.fuOpType === LSUOpType.cbo_clean ||
210                     s1_in.uop.fuOpType === LSUOpType.cbo_flush ||
211                     s1_in.uop.fuOpType === LSUOpType.cbo_inval
212  val s1_paddr     = io.tlb.resp.bits.paddr(0)
213  val s1_gpaddr    = io.tlb.resp.bits.gpaddr(0)
214  val s1_tlb_miss  = io.tlb.resp.bits.miss
215  val s1_mmio      = s1_mmio_cbo
216  val s1_exception = ExceptionNO.selectByFu(s1_out.uop.exceptionVec, StaCfg).asUInt.orR
217  val s1_isvec     = RegEnable(s0_out.isvec, false.B, s0_fire)
218  // val s1_isLastElem = RegEnable(s0_isLastElem, false.B, s0_fire)
219  s1_kill := s1_in.uop.robIdx.needFlush(io.redirect) || (s1_tlb_miss && !s1_isvec)
220
221  s1_ready := !s1_valid || s1_kill || s2_ready
222  io.tlb.resp.ready := true.B // TODO: why dtlbResp needs a ready?
223  when (s0_fire) { s1_valid := true.B }
224  .elsewhen (s1_fire) { s1_valid := false.B }
225  .elsewhen (s1_kill) { s1_valid := false.B }
226
227  // st-ld violation dectect request.
228  io.stld_nuke_query.valid       := s1_valid && !s1_tlb_miss && !s1_in.isHWPrefetch
229  io.stld_nuke_query.bits.robIdx := s1_in.uop.robIdx
230  io.stld_nuke_query.bits.paddr  := s1_paddr
231  io.stld_nuke_query.bits.mask   := s1_in.mask
232
233  // issue
234  io.issue.valid := s1_valid && !s1_tlb_miss && !s1_in.isHWPrefetch && !s1_isvec
235  io.issue.bits  := RegEnable(s0_stin, s0_valid)
236
237
238  // Send TLB feedback to store issue queue
239  // Store feedback is generated in store_s1, sent to RS in store_s2
240  val s1_feedback = Wire(Valid(new RSFeedback))
241  s1_feedback.valid                 := s1_valid & !s1_in.isHWPrefetch
242  s1_feedback.bits.hit              := !s1_tlb_miss
243  s1_feedback.bits.flushState       := io.tlb.resp.bits.ptwBack
244  s1_feedback.bits.robIdx           := s1_out.uop.robIdx
245  s1_feedback.bits.sourceType       := RSFeedbackType.tlbMiss
246  s1_feedback.bits.dataInvalidSqIdx := DontCare
247
248  XSDebug(s1_feedback.valid,
249    "S1 Store: tlbHit: %d robIdx: %d\n",
250    s1_feedback.bits.hit,
251    s1_feedback.bits.robIdx.value
252  )
253
254  // io.feedback_slow := s1_feedback
255
256  // get paddr from dtlb, check if rollback is needed
257  // writeback store inst to lsq
258  s1_out         := s1_in
259  s1_out.paddr   := s1_paddr
260  s1_out.gpaddr  := s1_gpaddr
261  s1_out.miss    := false.B
262  s1_out.mmio    := s1_mmio
263  s1_out.tlbMiss := s1_tlb_miss
264  s1_out.atomic  := s1_mmio
265  s1_out.uop.exceptionVec(storePageFault)      := io.tlb.resp.bits.excp(0).pf.st && s1_vecActive
266  s1_out.uop.exceptionVec(storeAccessFault)    := io.tlb.resp.bits.excp(0).af.st && s1_vecActive
267  s1_out.uop.exceptionVec(storeGuestPageFault) := io.tlb.resp.bits.excp(0).gpf.st && s1_vecActive
268
269  // scalar store and scalar load nuke check, and also other purposes
270  io.lsq.valid     := s1_valid && !s1_in.isHWPrefetch
271  io.lsq.bits      := s1_out
272  io.lsq.bits.miss := s1_tlb_miss
273
274  // kill dcache write intent request when tlb miss or exception
275  io.dcache.s1_kill  := (s1_tlb_miss || s1_exception || s1_mmio || s1_in.uop.robIdx.needFlush(io.redirect))
276  io.dcache.s1_paddr := s1_paddr
277
278  // write below io.out.bits assign sentence to prevent overwriting values
279  val s1_tlb_memidx = io.tlb.resp.bits.memidx
280  when(s1_tlb_memidx.is_st && io.tlb.resp.valid && !s1_tlb_miss && s1_tlb_memidx.idx === s1_out.uop.sqIdx.value) {
281    // printf("Store idx = %d\n", s1_tlb_memidx.idx)
282    s1_out.uop.debugInfo.tlbRespTime := GTimer()
283  }
284
285  // Pipeline
286  // --------------------------------------------------------------------------------
287  // stage 2
288  // --------------------------------------------------------------------------------
289  // mmio check
290  val s2_valid  = RegInit(false.B)
291  val s2_in     = RegEnable(s1_out, s1_fire)
292  val s2_out    = Wire(new LsPipelineBundle)
293  val s2_kill   = Wire(Bool())
294  val s2_can_go = s3_ready
295  val s2_fire   = s2_valid && !s2_kill && s2_can_go
296  val s2_vecActive    = RegEnable(s1_out.vecActive, true.B, s1_fire)
297
298  s2_ready := !s2_valid || s2_kill || s3_ready
299  when (s1_fire) { s2_valid := true.B }
300  .elsewhen (s2_fire) { s2_valid := false.B }
301  .elsewhen (s2_kill) { s2_valid := false.B }
302
303  val s2_pmp = WireInit(io.pmp)
304
305  val s2_exception = (ExceptionNO.selectByFu(s2_out.uop.exceptionVec, StaCfg).asUInt.orR) && RegNext(s1_feedback.bits.hit)
306  val s2_mmio = (s2_in.mmio || s2_pmp.mmio) && RegNext(s1_feedback.bits.hit)
307  s2_kill := ((s2_mmio && !s2_exception) && !s2_in.isvec) || s2_in.uop.robIdx.needFlush(io.redirect)
308
309  s2_out        := s2_in
310  s2_out.mmio   := s2_mmio && !s2_exception
311  s2_out.atomic := s2_in.atomic || s2_pmp.atomic
312  s2_out.uop.exceptionVec(storeAccessFault) := (s2_in.uop.exceptionVec(storeAccessFault) || s2_pmp.st) && s2_vecActive
313
314  // kill dcache write intent request when mmio or exception
315  io.dcache.s2_kill := (s2_mmio || s2_exception || s2_in.uop.robIdx.needFlush(io.redirect))
316  io.dcache.s2_pc   := s2_out.uop.pc
317  // TODO: dcache resp
318  io.dcache.resp.ready := true.B
319
320  // feedback tlb miss to RS in store_s2
321  io.feedback_slow.valid := RegNext(s1_feedback.valid && !s1_out.uop.robIdx.needFlush(io.redirect)) && !RegNext(s1_out.isvec)
322  io.feedback_slow.bits  := RegNext(s1_feedback.bits)
323
324  val s2_vecFeedback = RegNext(!s1_out.uop.robIdx.needFlush(io.redirect) && s1_feedback.bits.hit) && s2_in.isvec
325
326  // mmio and exception
327  io.lsq_replenish := s2_out
328
329  // prefetch related
330  io.lsq_replenish.miss := io.dcache.resp.fire && io.dcache.resp.bits.miss // miss info
331
332  // RegNext prefetch train for better timing
333  // ** Now, prefetch train is valid at store s3 **
334  io.prefetch_train.bits.fromLsPipelineBundle(s2_in, latch = true)
335  // override miss bit
336  io.prefetch_train.bits.miss := RegNext(io.dcache.resp.bits.miss)
337  // TODO: add prefetch and access bit
338  io.prefetch_train.bits.meta_prefetch := false.B
339  io.prefetch_train.bits.meta_access := false.B
340  if(EnableStorePrefetchSMS) {
341    io.prefetch_train.valid := RegNext(s2_valid && io.dcache.resp.fire && !s2_out.mmio && !s2_in.tlbMiss && !s2_in.isHWPrefetch)
342  }else {
343    io.prefetch_train.valid := false.B
344  }
345
346  // Pipeline
347  // --------------------------------------------------------------------------------
348  // stage 3
349  // --------------------------------------------------------------------------------
350  // store write back
351  val s3_valid  = RegInit(false.B)
352  val s3_in     = RegEnable(s2_out, s2_fire)
353  val s3_out    = Wire(new MemExuOutput(isVector = true))
354  val s3_kill   = s3_in.uop.robIdx.needFlush(io.redirect)
355  val s3_can_go = s3_ready
356  val s3_fire   = s3_valid && !s3_kill && s3_can_go
357  val s3_vecFeedback = RegEnable(s2_vecFeedback, s2_fire)
358
359  when (s2_fire) { s3_valid := (!s2_mmio || s2_exception) && !s2_out.isHWPrefetch  }
360  .elsewhen (s3_fire) { s3_valid := false.B }
361  .elsewhen (s3_kill) { s3_valid := false.B }
362
363  // wb: writeback
364  val SelectGroupSize   = RollbackGroupSize
365  val lgSelectGroupSize = log2Ceil(SelectGroupSize)
366  val TotalSelectCycles = scala.math.ceil(log2Ceil(LoadQueueRAWSize).toFloat / lgSelectGroupSize).toInt + 1
367
368  s3_out                 := DontCare
369  s3_out.uop             := s3_in.uop
370  s3_out.data            := DontCare
371  s3_out.debug.isMMIO    := s3_in.mmio
372  s3_out.debug.paddr     := s3_in.paddr
373  s3_out.debug.vaddr     := s3_in.vaddr
374  s3_out.debug.isPerfCnt := false.B
375
376  // Pipeline
377  // --------------------------------------------------------------------------------
378  // stage x
379  // --------------------------------------------------------------------------------
380  // delay TotalSelectCycles - 2 cycle(s)
381  val TotalDelayCycles = TotalSelectCycles - 2
382  val sx_valid = Wire(Vec(TotalDelayCycles + 1, Bool()))
383  val sx_ready = Wire(Vec(TotalDelayCycles + 1, Bool()))
384  val sx_in    = Wire(Vec(TotalDelayCycles + 1, new VecMemExuOutput(isVector = true)))
385
386  // backward ready signal
387  s3_ready := sx_ready.head
388  for (i <- 0 until TotalDelayCycles + 1) {
389    if (i == 0) {
390      sx_valid(i)          := s3_valid
391      sx_in(i).output      := s3_out
392      sx_in(i).vecFeedback := s3_vecFeedback
393      sx_in(i).mmio        := s3_in.mmio
394      sx_in(i).usSecondInv := s3_in.usSecondInv
395      sx_in(i).elemIdx     := s3_in.elemIdx
396      sx_in(i).alignedType := s3_in.alignedType
397      sx_in(i).mbIndex     := s3_in.mbIndex
398      sx_in(i).mask        := s3_in.mask
399      sx_in(i).vaddr       := s3_in.vaddr
400      sx_ready(i) := !s3_valid(i) || sx_in(i).output.uop.robIdx.needFlush(io.redirect) || (if (TotalDelayCycles == 0) io.stout.ready else sx_ready(i+1))
401    } else {
402      val cur_kill   = sx_in(i).output.uop.robIdx.needFlush(io.redirect)
403      val cur_can_go = (if (i == TotalDelayCycles) io.stout.ready else sx_ready(i+1))
404      val cur_fire   = sx_valid(i) && !cur_kill && cur_can_go
405      val prev_fire  = sx_valid(i-1) && !sx_in(i-1).output.uop.robIdx.needFlush(io.redirect) && sx_ready(i)
406
407      sx_ready(i) := !sx_valid(i) || cur_kill || (if (i == TotalDelayCycles) io.stout.ready else sx_ready(i+1))
408      val sx_valid_can_go = prev_fire || cur_fire || cur_kill
409      sx_valid(i) := RegEnable(Mux(prev_fire, true.B, false.B), false.B, sx_valid_can_go)
410      sx_in(i) := RegEnable(sx_in(i-1), prev_fire)
411    }
412  }
413  val sx_last_valid = sx_valid.takeRight(1).head
414  val sx_last_ready = sx_ready.takeRight(1).head
415  val sx_last_in    = sx_in.takeRight(1).head
416  sx_last_ready := !sx_last_valid || sx_last_in.output.uop.robIdx.needFlush(io.redirect) || io.stout.ready
417
418  io.stout.valid := sx_last_valid && !sx_last_in.output.uop.robIdx.needFlush(io.redirect) && isStore(sx_last_in.output.uop.fuType)
419  io.stout.bits := sx_last_in.output
420
421  io.vecstout.valid := sx_last_valid && !sx_last_in.output.uop.robIdx.needFlush(io.redirect) && isVStore(sx_last_in.output.uop.fuType)
422  // TODO: implement it!
423  io.vecstout.bits.mBIndex := sx_last_in.mbIndex
424  io.vecstout.bits.hit := sx_last_in.vecFeedback
425  io.vecstout.bits.isvec := true.B
426  io.vecstout.bits.sourceType := RSFeedbackType.tlbMiss
427  io.vecstout.bits.flushState := DontCare
428  io.vecstout.bits.mmio := sx_last_in.mmio
429  io.vecstout.bits.exceptionVec := sx_last_in.output.uop.exceptionVec
430  io.vecstout.bits.usSecondInv := sx_last_in.usSecondInv
431  io.vecstout.bits.vecFeedback := sx_last_in.vecFeedback
432  io.vecstout.bits.elemIdx     := sx_last_in.elemIdx
433  io.vecstout.bits.alignedType := sx_last_in.alignedType
434  io.vecstout.bits.mask        := sx_last_in.mask
435  io.vecstout.bits.vaddr       := sx_last_in.vaddr
436  // io.vecstout.bits.reg_offset.map(_ := DontCare)
437  // io.vecstout.bits.elemIdx.map(_ := sx_last_in.elemIdx)
438  // io.vecstout.bits.elemIdxInsideVd.map(_ := DontCare)
439  // io.vecstout.bits.vecdata.map(_ := DontCare)
440  // io.vecstout.bits.mask.map(_ := DontCare)
441  // io.vecstout.bits.alignedType.map(_ := sx_last_in.alignedType)
442
443  io.debug_ls := DontCare
444  io.debug_ls.s1_robIdx := s1_in.uop.robIdx.value
445  io.debug_ls.s1_isTlbFirstMiss := io.tlb.resp.valid && io.tlb.resp.bits.miss && io.tlb.resp.bits.debug.isFirstIssue && !s1_in.isHWPrefetch
446
447  private def printPipeLine(pipeline: LsPipelineBundle, cond: Bool, name: String): Unit = {
448    XSDebug(cond,
449      p"$name" + p" pc ${Hexadecimal(pipeline.uop.pc)} " +
450        p"addr ${Hexadecimal(pipeline.vaddr)} -> ${Hexadecimal(pipeline.paddr)} " +
451        p"op ${Binary(pipeline.uop.fuOpType)} " +
452        p"data ${Hexadecimal(pipeline.data)} " +
453        p"mask ${Hexadecimal(pipeline.mask)}\n"
454    )
455  }
456
457  printPipeLine(s0_out, s0_valid, "S0")
458  printPipeLine(s1_out, s1_valid, "S1")
459
460  // perf cnt
461  XSPerfAccumulate("s0_in_valid",                s0_valid)
462  XSPerfAccumulate("s0_in_fire",                 s0_fire)
463  XSPerfAccumulate("s0_vecin_fire",              s0_fire && s0_use_flow_vec)
464  XSPerfAccumulate("s0_in_fire_first_issue",     s0_fire && s0_isFirstIssue)
465  XSPerfAccumulate("s0_addr_spec_success",       s0_fire && !s0_use_flow_vec && s0_saddr(VAddrBits-1, 12) === s0_stin.src(0)(VAddrBits-1, 12))
466  XSPerfAccumulate("s0_addr_spec_failed",        s0_fire && !s0_use_flow_vec && s0_saddr(VAddrBits-1, 12) =/= s0_stin.src(0)(VAddrBits-1, 12))
467  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)
468  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)
469
470  XSPerfAccumulate("s1_in_valid",                s1_valid)
471  XSPerfAccumulate("s1_in_fire",                 s1_fire)
472  XSPerfAccumulate("s1_in_fire_first_issue",     s1_fire && s1_in.isFirstIssue)
473  XSPerfAccumulate("s1_tlb_miss",                s1_fire && s1_tlb_miss)
474  XSPerfAccumulate("s1_tlb_miss_first_issue",    s1_fire && s1_tlb_miss && s1_in.isFirstIssue)
475  // end
476}