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 35 with HasDCacheParameters 36 with HasVLSUParameters 37 { 38 val io = IO(new Bundle() { 39 val redirect = Flipped(ValidIO(new Redirect)) 40 val stin = Flipped(Decoupled(new MemExuInput)) 41 val issue = Valid(new MemExuInput) 42 val tlb = new TlbRequestIO() 43 val dcache = new DCacheStoreIO 44 val pmp = Flipped(new PMPRespBundle()) 45 val lsq = ValidIO(new LsPipelineBundle) 46 val lsq_replenish = Output(new LsPipelineBundle()) 47 val feedback_slow = ValidIO(new RSFeedback) 48 val prefetch_req = Flipped(DecoupledIO(new StorePrefetchReq)) 49 // provide prefetch info to sms 50 val prefetch_train = ValidIO(new StPrefetchTrainBundle()) 51 // speculative for gated control 52 val s1_prefetch_spec = Output(Bool()) 53 val s2_prefetch_spec = Output(Bool()) 54 val stld_nuke_query = Valid(new StoreNukeQueryIO) 55 val stout = DecoupledIO(new MemExuOutput) // writeback store 56 val vecstout = DecoupledIO(new VecPipelineFeedbackIO(isVStore = true)) 57 // store mask, send to sq in store_s0 58 val st_mask_out = Valid(new StoreMaskBundle) 59 val debug_ls = Output(new DebugLsInfoBundle) 60 // vector 61 val vecstin = Flipped(Decoupled(new VecPipeBundle(isVStore = true))) 62 val vec_isFirstIssue = Input(Bool()) 63 }) 64 65 val s1_ready, s2_ready, s3_ready = WireInit(false.B) 66 67 // Pipeline 68 // -------------------------------------------------------------------------------- 69 // stage 0 70 // -------------------------------------------------------------------------------- 71 // generate addr, use addr to query DCache and DTLB 72 val s0_iss_valid = io.stin.valid 73 val s0_prf_valid = io.prefetch_req.valid && io.dcache.req.ready 74 val s0_vec_valid = io.vecstin.valid 75 val s0_valid = s0_iss_valid || s0_prf_valid || s0_vec_valid 76 val s0_use_flow_vec = s0_vec_valid 77 val s0_use_flow_rs = s0_iss_valid && !s0_vec_valid 78 val s0_use_flow_prf = !s0_iss_valid && !s0_vec_valid && s0_prf_valid 79 val s0_stin = Mux(s0_use_flow_rs, io.stin.bits, 0.U.asTypeOf(io.stin.bits)) 80 val s0_vecstin = Mux(s0_use_flow_vec, io.vecstin.bits, 0.U.asTypeOf(io.vecstin.bits)) 81 val s0_uop = Mux(s0_use_flow_rs, s0_stin.uop, s0_vecstin.uop) 82 val s0_isFirstIssue = s0_use_flow_rs && io.stin.bits.isFirstIssue || s0_use_flow_vec && io.vec_isFirstIssue 83 val s0_iqIdx = Mux(s0_use_flow_rs, io.stin.bits.iqIdx, 0.U) 84 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 85 val s0_mem_idx = Mux(s0_use_flow_rs || s0_use_flow_vec, s0_uop.sqIdx.value, 0.U) 86 val s0_rob_idx = Mux(s0_use_flow_rs || s0_use_flow_vec, s0_uop.robIdx, 0.U.asTypeOf(s0_uop.robIdx)) 87 val s0_pc = Mux(s0_use_flow_rs || s0_use_flow_vec, s0_uop.pc, 0.U) 88 val s0_instr_type = Mux(s0_use_flow_rs || s0_use_flow_vec, STORE_SOURCE.U, DCACHE_PREFETCH_SOURCE.U) 89 val s0_wlineflag = Mux(s0_use_flow_rs, s0_uop.fuOpType === LSUOpType.cbo_zero, false.B) 90 val s0_out = Wire(new LsPipelineBundle) 91 val s0_kill = s0_uop.robIdx.needFlush(io.redirect) 92 val s0_can_go = s1_ready 93 val s0_fire = s0_valid && !s0_kill && s0_can_go 94 val s0_is128bit = is128Bit(s0_vecstin.alignedType) 95 // vector 96 val s0_vecActive = !s0_use_flow_vec || s0_vecstin.vecActive 97 // val s0_flowPtr = s0_vecstin.flowPtr 98 // val s0_isLastElem = s0_vecstin.isLastElem 99 val s0_secondInv = s0_vecstin.usSecondInv 100 val s0_elemIdx = s0_vecstin.elemIdx 101 val s0_alignedType = s0_vecstin.alignedType 102 val s0_mBIndex = s0_vecstin.mBIndex 103 104 // generate addr 105 // val saddr = s0_in.bits.src(0) + SignExt(s0_in.bits.uop.imm(11,0), VAddrBits) 106 val imm12 = WireInit(s0_uop.imm(11,0)) 107 val saddr_lo = s0_stin.src(0)(11,0) + Cat(0.U(1.W), imm12) 108 val saddr_hi = Mux(saddr_lo(12), 109 Mux(imm12(11), s0_stin.src(0)(VAddrBits-1, 12), s0_stin.src(0)(VAddrBits-1, 12)+1.U), 110 Mux(imm12(11), s0_stin.src(0)(VAddrBits-1, 12)+SignExt(1.U, VAddrBits-12), s0_stin.src(0)(VAddrBits-1, 12)), 111 ) 112 val s0_saddr = Cat(saddr_hi, saddr_lo(11,0)) 113 val s0_vaddr = Mux( 114 s0_use_flow_rs, 115 s0_saddr, 116 Mux( 117 s0_use_flow_vec, 118 s0_vecstin.vaddr, 119 io.prefetch_req.bits.vaddr 120 ) 121 ) 122 val s0_mask = Mux( 123 s0_use_flow_rs, 124 genVWmask128(s0_saddr, s0_uop.fuOpType(2,0)), 125 Mux( 126 s0_use_flow_vec, 127 s0_vecstin.mask, 128 // -1.asSInt.asUInt 129 Fill(VLEN/8, 1.U(1.W)) 130 ) 131 ) 132 133 io.tlb.req.valid := s0_valid 134 io.tlb.req.bits.vaddr := s0_vaddr 135 io.tlb.req.bits.cmd := TlbCmd.write 136 io.tlb.req.bits.size := s0_size 137 io.tlb.req.bits.kill := false.B 138 io.tlb.req.bits.memidx.is_ld := false.B 139 io.tlb.req.bits.memidx.is_st := true.B 140 io.tlb.req.bits.memidx.idx := s0_mem_idx 141 io.tlb.req.bits.debug.robIdx := s0_rob_idx 142 io.tlb.req.bits.no_translate := false.B 143 io.tlb.req.bits.debug.pc := s0_pc 144 io.tlb.req.bits.debug.isFirstIssue := s0_isFirstIssue 145 io.tlb.req_kill := false.B 146 io.tlb.req.bits.hyperinst := LSUOpType.isHsv(s0_uop.fuOpType) 147 io.tlb.req.bits.hlvx := false.B 148 149 // Dcache access here: not **real** dcache write 150 // just read meta and tag in dcache, to find out the store will hit or miss 151 152 // NOTE: The store request does not wait for the dcache to be ready. 153 // If the dcache is not ready at this time, the dcache is not queried. 154 // But, store prefetch request will always wait for dcache to be ready to make progress. 155 io.dcache.req.valid := s0_fire 156 io.dcache.req.bits.cmd := MemoryOpConstants.M_PFW 157 io.dcache.req.bits.vaddr := s0_vaddr 158 io.dcache.req.bits.instrtype := s0_instr_type 159 160 s0_out := DontCare 161 s0_out.vaddr := s0_vaddr 162 // Now data use its own io 163 // s1_out.data := genWdata(s1_in.src(1), s1_in.uop.fuOpType(1,0)) 164 s0_out.data := s0_stin.src(1) 165 s0_out.uop := s0_uop 166 s0_out.miss := false.B 167 s0_out.mask := s0_mask 168 s0_out.isFirstIssue := s0_isFirstIssue 169 s0_out.isHWPrefetch := s0_use_flow_prf 170 s0_out.wlineflag := s0_wlineflag 171 s0_out.isvec := s0_use_flow_vec 172 s0_out.is128bit := s0_is128bit 173 s0_out.vecActive := s0_vecActive 174 s0_out.usSecondInv := s0_secondInv 175 s0_out.elemIdx := s0_elemIdx 176 s0_out.alignedType := s0_alignedType 177 s0_out.mbIndex := s0_mBIndex 178 when(s0_valid && s0_isFirstIssue) { 179 s0_out.uop.debugInfo.tlbFirstReqTime := GTimer() 180 } 181 182 // exception check 183 val s0_addr_aligned = LookupTree(Mux(s0_use_flow_vec, s0_vecstin.alignedType(1,0), s0_uop.fuOpType(1, 0)), List( 184 "b00".U -> true.B, //b 185 "b01".U -> (s0_out.vaddr(0) === 0.U), //h 186 "b10".U -> (s0_out.vaddr(1,0) === 0.U), //w 187 "b11".U -> (s0_out.vaddr(2,0) === 0.U) //d 188 )) 189 // if vector store sends 128-bit requests, its address must be 128-aligned 190 XSError(s0_use_flow_vec && s0_out.vaddr(3, 0) =/= 0.U && s0_vecstin.alignedType(2), "unit stride 128 bit element is not aligned!") 191 s0_out.uop.exceptionVec(storeAddrMisaligned) := Mux(s0_use_flow_rs || s0_use_flow_vec, !s0_addr_aligned, false.B) 192 193 io.st_mask_out.valid := s0_use_flow_rs || s0_use_flow_vec 194 io.st_mask_out.bits.mask := s0_out.mask 195 io.st_mask_out.bits.sqIdx := s0_out.uop.sqIdx 196 197 io.stin.ready := s1_ready && s0_use_flow_rs 198 io.vecstin.ready := s1_ready && s0_use_flow_vec 199 io.prefetch_req.ready := s1_ready && io.dcache.req.ready && !s0_iss_valid && !s0_vec_valid 200 201 // Pipeline 202 // -------------------------------------------------------------------------------- 203 // stage 1 204 // -------------------------------------------------------------------------------- 205 // TLB resp (send paddr to dcache) 206 val s1_valid = RegInit(false.B) 207 val s1_in = RegEnable(s0_out, s0_fire) 208 val s1_out = Wire(new LsPipelineBundle) 209 val s1_kill = Wire(Bool()) 210 val s1_can_go = s2_ready 211 val s1_fire = s1_valid && !s1_kill && s1_can_go 212 val s1_vecActive = RegEnable(s0_out.vecActive, true.B, s0_fire) 213 214 // mmio cbo decoder 215 val s1_mmio_cbo = s1_in.uop.fuOpType === LSUOpType.cbo_clean || 216 s1_in.uop.fuOpType === LSUOpType.cbo_flush || 217 s1_in.uop.fuOpType === LSUOpType.cbo_inval 218 val s1_paddr = io.tlb.resp.bits.paddr(0) 219 val s1_gpaddr = io.tlb.resp.bits.gpaddr(0) 220 val s1_tlb_miss = io.tlb.resp.bits.miss 221 val s1_mmio = s1_mmio_cbo 222 val s1_exception = ExceptionNO.selectByFu(s1_out.uop.exceptionVec, StaCfg).asUInt.orR 223 val s1_isvec = RegEnable(s0_out.isvec, false.B, s0_fire) 224 // val s1_isLastElem = RegEnable(s0_isLastElem, false.B, s0_fire) 225 s1_kill := s1_in.uop.robIdx.needFlush(io.redirect) || (s1_tlb_miss && !s1_isvec) 226 227 s1_ready := !s1_valid || s1_kill || s2_ready 228 io.tlb.resp.ready := true.B // TODO: why dtlbResp needs a ready? 229 when (s0_fire) { s1_valid := true.B } 230 .elsewhen (s1_fire) { s1_valid := false.B } 231 .elsewhen (s1_kill) { s1_valid := false.B } 232 233 // st-ld violation dectect request. 234 io.stld_nuke_query.valid := s1_valid && !s1_tlb_miss && !s1_in.isHWPrefetch 235 io.stld_nuke_query.bits.robIdx := s1_in.uop.robIdx 236 io.stld_nuke_query.bits.paddr := s1_paddr 237 io.stld_nuke_query.bits.mask := s1_in.mask 238 io.stld_nuke_query.bits.matchLine := s1_in.isvec && s1_in.is128bit 239 240 // issue 241 io.issue.valid := s1_valid && !s1_tlb_miss && !s1_in.isHWPrefetch && !s1_isvec 242 io.issue.bits := RegEnable(s0_stin, s0_valid) 243 244 245 // Send TLB feedback to store issue queue 246 // Store feedback is generated in store_s1, sent to RS in store_s2 247 val s1_feedback = Wire(Valid(new RSFeedback)) 248 s1_feedback.valid := s1_valid & !s1_in.isHWPrefetch 249 s1_feedback.bits.hit := !s1_tlb_miss 250 s1_feedback.bits.flushState := io.tlb.resp.bits.ptwBack 251 s1_feedback.bits.robIdx := s1_out.uop.robIdx 252 s1_feedback.bits.sourceType := RSFeedbackType.tlbMiss 253 s1_feedback.bits.dataInvalidSqIdx := DontCare 254 s1_feedback.bits.sqIdx := s1_out.uop.sqIdx 255 256 XSDebug(s1_feedback.valid, 257 "S1 Store: tlbHit: %d robIdx: %d\n", 258 s1_feedback.bits.hit, 259 s1_feedback.bits.robIdx.value 260 ) 261 262 // io.feedback_slow := s1_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.gpaddr := s1_gpaddr 269 s1_out.miss := false.B 270 s1_out.mmio := s1_mmio 271 s1_out.tlbMiss := s1_tlb_miss 272 s1_out.atomic := s1_mmio 273 s1_out.uop.exceptionVec(storePageFault) := io.tlb.resp.bits.excp(0).pf.st && s1_vecActive 274 s1_out.uop.exceptionVec(storeAccessFault) := io.tlb.resp.bits.excp(0).af.st && s1_vecActive 275 s1_out.uop.exceptionVec(storeGuestPageFault) := io.tlb.resp.bits.excp(0).gpf.st && s1_vecActive 276 277 // scalar store and scalar load nuke check, and also other purposes 278 io.lsq.valid := s1_valid && !s1_in.isHWPrefetch 279 io.lsq.bits := s1_out 280 io.lsq.bits.miss := s1_tlb_miss 281 282 // kill dcache write intent request when tlb miss or exception 283 io.dcache.s1_kill := (s1_tlb_miss || s1_exception || s1_mmio || s1_in.uop.robIdx.needFlush(io.redirect)) 284 io.dcache.s1_paddr := s1_paddr 285 286 // write below io.out.bits assign sentence to prevent overwriting values 287 val s1_tlb_memidx = io.tlb.resp.bits.memidx 288 when(s1_tlb_memidx.is_st && io.tlb.resp.valid && !s1_tlb_miss && s1_tlb_memidx.idx === s1_out.uop.sqIdx.value) { 289 // printf("Store idx = %d\n", s1_tlb_memidx.idx) 290 s1_out.uop.debugInfo.tlbRespTime := GTimer() 291 } 292 293 // Pipeline 294 // -------------------------------------------------------------------------------- 295 // stage 2 296 // -------------------------------------------------------------------------------- 297 // mmio check 298 val s2_valid = RegInit(false.B) 299 val s2_in = RegEnable(s1_out, s1_fire) 300 val s2_out = Wire(new LsPipelineBundle) 301 val s2_kill = Wire(Bool()) 302 val s2_can_go = s3_ready 303 val s2_fire = s2_valid && !s2_kill && s2_can_go 304 val s2_vecActive = RegEnable(s1_out.vecActive, true.B, s1_fire) 305 306 s2_ready := !s2_valid || s2_kill || s3_ready 307 when (s1_fire) { s2_valid := true.B } 308 .elsewhen (s2_fire) { s2_valid := false.B } 309 .elsewhen (s2_kill) { s2_valid := false.B } 310 311 val s2_pmp = WireInit(io.pmp) 312 313 val s2_exception = (ExceptionNO.selectByFu(s2_out.uop.exceptionVec, StaCfg).asUInt.orR) && RegNext(s1_feedback.bits.hit) 314 val s2_mmio = (s2_in.mmio || s2_pmp.mmio) && RegNext(s1_feedback.bits.hit) 315 s2_kill := ((s2_mmio && !s2_exception) && !s2_in.isvec) || s2_in.uop.robIdx.needFlush(io.redirect) 316 317 s2_out := s2_in 318 s2_out.af := s2_pmp.st && !s2_in.isvec 319 s2_out.mmio := s2_mmio && !s2_exception 320 s2_out.atomic := s2_in.atomic || s2_pmp.atomic 321 s2_out.uop.exceptionVec(storeAccessFault) := (s2_in.uop.exceptionVec(storeAccessFault) || 322 s2_pmp.st || 323 (s2_in.isvec && s2_pmp.mmio && RegNext(s1_feedback.bits.hit)) 324 ) && s2_vecActive 325 326 // kill dcache write intent request when mmio or exception 327 io.dcache.s2_kill := (s2_mmio || s2_exception || s2_in.uop.robIdx.needFlush(io.redirect)) 328 io.dcache.s2_pc := s2_out.uop.pc 329 // TODO: dcache resp 330 io.dcache.resp.ready := true.B 331 332 // feedback tlb miss to RS in store_s2 333 val feedback_slow_valid = WireInit(false.B) 334 feedback_slow_valid := s1_feedback.valid && !s1_out.uop.robIdx.needFlush(io.redirect) && !s1_out.isvec 335 io.feedback_slow.valid := GatedValidRegNext(feedback_slow_valid) 336 io.feedback_slow.bits := RegEnable(s1_feedback.bits, feedback_slow_valid) 337 338 val s2_vecFeedback = RegNext(!s1_out.uop.robIdx.needFlush(io.redirect) && s1_feedback.bits.hit) && s2_in.isvec 339 340 // mmio and exception 341 io.lsq_replenish := s2_out 342 io.lsq_replenish.af := s2_out.af && !s2_kill 343 344 // prefetch related 345 io.lsq_replenish.miss := io.dcache.resp.fire && io.dcache.resp.bits.miss // miss info 346 347 // RegNext prefetch train for better timing 348 // ** Now, prefetch train is valid at store s3 ** 349 val s2_prefetch_train_valid = WireInit(false.B) 350 s2_prefetch_train_valid := s2_valid && io.dcache.resp.fire && !s2_out.mmio && !s2_in.tlbMiss && !s2_in.isHWPrefetch 351 if(EnableStorePrefetchSMS) { 352 io.s1_prefetch_spec := s1_fire 353 io.s2_prefetch_spec := s2_prefetch_train_valid 354 io.prefetch_train.valid := RegNext(s2_prefetch_train_valid) 355 io.prefetch_train.bits.fromLsPipelineBundle(s2_in, latch = true, enable = s2_prefetch_train_valid) 356 }else { 357 io.s1_prefetch_spec := false.B 358 io.s2_prefetch_spec := false.B 359 io.prefetch_train.valid := false.B 360 io.prefetch_train.bits.fromLsPipelineBundle(s2_in, latch = true, enable = false.B) 361 } 362 // override miss bit 363 io.prefetch_train.bits.miss := RegEnable(io.dcache.resp.bits.miss, s2_prefetch_train_valid) 364 // TODO: add prefetch and access bit 365 io.prefetch_train.bits.meta_prefetch := false.B 366 io.prefetch_train.bits.meta_access := false.B 367 368 // Pipeline 369 // -------------------------------------------------------------------------------- 370 // stage 3 371 // -------------------------------------------------------------------------------- 372 // store write back 373 val s3_valid = RegInit(false.B) 374 val s3_in = RegEnable(s2_out, s2_fire) 375 val s3_out = Wire(new MemExuOutput(isVector = true)) 376 val s3_kill = s3_in.uop.robIdx.needFlush(io.redirect) 377 val s3_can_go = s3_ready 378 val s3_fire = s3_valid && !s3_kill && s3_can_go 379 val s3_vecFeedback = RegEnable(s2_vecFeedback, s2_fire) 380 381 when (s2_fire) { s3_valid := (!s2_mmio || s2_exception) && !s2_out.isHWPrefetch } 382 .elsewhen (s3_fire) { s3_valid := false.B } 383 .elsewhen (s3_kill) { s3_valid := false.B } 384 385 // wb: writeback 386 val SelectGroupSize = RollbackGroupSize 387 val lgSelectGroupSize = log2Ceil(SelectGroupSize) 388 val TotalSelectCycles = scala.math.ceil(log2Ceil(LoadQueueRAWSize).toFloat / lgSelectGroupSize).toInt + 1 389 390 s3_out := DontCare 391 s3_out.uop := s3_in.uop 392 s3_out.data := DontCare 393 s3_out.debug.isMMIO := s3_in.mmio 394 s3_out.debug.paddr := s3_in.paddr 395 s3_out.debug.vaddr := s3_in.vaddr 396 s3_out.debug.isPerfCnt := false.B 397 398 // Pipeline 399 // -------------------------------------------------------------------------------- 400 // stage x 401 // -------------------------------------------------------------------------------- 402 // delay TotalSelectCycles - 2 cycle(s) 403 val TotalDelayCycles = TotalSelectCycles - 2 404 val sx_valid = Wire(Vec(TotalDelayCycles + 1, Bool())) 405 val sx_ready = Wire(Vec(TotalDelayCycles + 1, Bool())) 406 val sx_in = Wire(Vec(TotalDelayCycles + 1, new VecMemExuOutput(isVector = true))) 407 408 // backward ready signal 409 s3_ready := sx_ready.head 410 for (i <- 0 until TotalDelayCycles + 1) { 411 if (i == 0) { 412 sx_valid(i) := s3_valid 413 sx_in(i).output := s3_out 414 sx_in(i).vecFeedback := s3_vecFeedback 415 sx_in(i).mmio := s3_in.mmio 416 sx_in(i).usSecondInv := s3_in.usSecondInv 417 sx_in(i).elemIdx := s3_in.elemIdx 418 sx_in(i).alignedType := s3_in.alignedType 419 sx_in(i).mbIndex := s3_in.mbIndex 420 sx_in(i).mask := s3_in.mask 421 sx_in(i).vaddr := s3_in.vaddr 422 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)) 423 } else { 424 val cur_kill = sx_in(i).output.uop.robIdx.needFlush(io.redirect) 425 val cur_can_go = (if (i == TotalDelayCycles) io.stout.ready else sx_ready(i+1)) 426 val cur_fire = sx_valid(i) && !cur_kill && cur_can_go 427 val prev_fire = sx_valid(i-1) && !sx_in(i-1).output.uop.robIdx.needFlush(io.redirect) && sx_ready(i) 428 429 sx_ready(i) := !sx_valid(i) || cur_kill || (if (i == TotalDelayCycles) io.stout.ready else sx_ready(i+1)) 430 val sx_valid_can_go = prev_fire || cur_fire || cur_kill 431 sx_valid(i) := RegEnable(Mux(prev_fire, true.B, false.B), false.B, sx_valid_can_go) 432 sx_in(i) := RegEnable(sx_in(i-1), prev_fire) 433 } 434 } 435 val sx_last_valid = sx_valid.takeRight(1).head 436 val sx_last_ready = sx_ready.takeRight(1).head 437 val sx_last_in = sx_in.takeRight(1).head 438 sx_last_ready := !sx_last_valid || sx_last_in.output.uop.robIdx.needFlush(io.redirect) || io.stout.ready 439 440 io.stout.valid := sx_last_valid && !sx_last_in.output.uop.robIdx.needFlush(io.redirect) && isStore(sx_last_in.output.uop.fuType) 441 io.stout.bits := sx_last_in.output 442 443 io.vecstout.valid := sx_last_valid && !sx_last_in.output.uop.robIdx.needFlush(io.redirect) && isVStore(sx_last_in.output.uop.fuType) 444 // TODO: implement it! 445 io.vecstout.bits.mBIndex := sx_last_in.mbIndex 446 io.vecstout.bits.hit := sx_last_in.vecFeedback 447 io.vecstout.bits.isvec := true.B 448 io.vecstout.bits.sourceType := RSFeedbackType.tlbMiss 449 io.vecstout.bits.flushState := DontCare 450 io.vecstout.bits.mmio := sx_last_in.mmio 451 io.vecstout.bits.exceptionVec := sx_last_in.output.uop.exceptionVec 452 io.vecstout.bits.usSecondInv := sx_last_in.usSecondInv 453 io.vecstout.bits.vecFeedback := sx_last_in.vecFeedback 454 io.vecstout.bits.elemIdx := sx_last_in.elemIdx 455 io.vecstout.bits.alignedType := sx_last_in.alignedType 456 io.vecstout.bits.mask := sx_last_in.mask 457 io.vecstout.bits.vaddr := sx_last_in.vaddr 458 // io.vecstout.bits.reg_offset.map(_ := DontCare) 459 // io.vecstout.bits.elemIdx.map(_ := sx_last_in.elemIdx) 460 // io.vecstout.bits.elemIdxInsideVd.map(_ := DontCare) 461 // io.vecstout.bits.vecdata.map(_ := DontCare) 462 // io.vecstout.bits.mask.map(_ := DontCare) 463 // io.vecstout.bits.alignedType.map(_ := sx_last_in.alignedType) 464 465 io.debug_ls := DontCare 466 io.debug_ls.s1_robIdx := s1_in.uop.robIdx.value 467 io.debug_ls.s1_isTlbFirstMiss := io.tlb.resp.valid && io.tlb.resp.bits.miss && io.tlb.resp.bits.debug.isFirstIssue && !s1_in.isHWPrefetch 468 469 private def printPipeLine(pipeline: LsPipelineBundle, cond: Bool, name: String): Unit = { 470 XSDebug(cond, 471 p"$name" + p" pc ${Hexadecimal(pipeline.uop.pc)} " + 472 p"addr ${Hexadecimal(pipeline.vaddr)} -> ${Hexadecimal(pipeline.paddr)} " + 473 p"op ${Binary(pipeline.uop.fuOpType)} " + 474 p"data ${Hexadecimal(pipeline.data)} " + 475 p"mask ${Hexadecimal(pipeline.mask)}\n" 476 ) 477 } 478 479 printPipeLine(s0_out, s0_valid, "S0") 480 printPipeLine(s1_out, s1_valid, "S1") 481 482 // perf cnt 483 XSPerfAccumulate("s0_in_valid", s0_valid) 484 XSPerfAccumulate("s0_in_fire", s0_fire) 485 XSPerfAccumulate("s0_vecin_fire", s0_fire && s0_use_flow_vec) 486 XSPerfAccumulate("s0_in_fire_first_issue", s0_fire && s0_isFirstIssue) 487 XSPerfAccumulate("s0_addr_spec_success", s0_fire && !s0_use_flow_vec && s0_saddr(VAddrBits-1, 12) === s0_stin.src(0)(VAddrBits-1, 12)) 488 XSPerfAccumulate("s0_addr_spec_failed", s0_fire && !s0_use_flow_vec && s0_saddr(VAddrBits-1, 12) =/= s0_stin.src(0)(VAddrBits-1, 12)) 489 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) 490 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) 491 492 XSPerfAccumulate("s1_in_valid", s1_valid) 493 XSPerfAccumulate("s1_in_fire", s1_fire) 494 XSPerfAccumulate("s1_in_fire_first_issue", s1_fire && s1_in.isFirstIssue) 495 XSPerfAccumulate("s1_tlb_miss", s1_fire && s1_tlb_miss) 496 XSPerfAccumulate("s1_tlb_miss_first_issue", s1_fire && s1_tlb_miss && s1_in.isFirstIssue) 497 // end 498}