cpu.hh (13652:45d94ac03a27) | cpu.hh (13892:0182a0601f66) |
---|---|
1/* 2 * Copyright (c) 2011-2013, 2016-2019 ARM Limited 3 * Copyright (c) 2013 Advanced Micro Devices, Inc. 4 * All rights reserved 5 * 6 * The license below extends only to copyright in the software and shall 7 * not be construed as granting a license to any other intellectual 8 * property including but not limited to intellectual property relating 9 * to a hardware implementation of the functionality of the software 10 * licensed hereunder. You may use the software subject to the license 11 * terms below provided that you ensure that this notice is replicated 12 * unmodified and in its entirety in all distributions of the software, 13 * modified or unmodified, in source code or in binary form. 14 * 15 * Copyright (c) 2004-2005 The Regents of The University of Michigan 16 * Copyright (c) 2011 Regents of the University of California 17 * All rights reserved. 18 * 19 * Redistribution and use in source and binary forms, with or without 20 * modification, are permitted provided that the following conditions are 21 * met: redistributions of source code must retain the above copyright 22 * notice, this list of conditions and the following disclaimer; 23 * redistributions in binary form must reproduce the above copyright 24 * notice, this list of conditions and the following disclaimer in the 25 * documentation and/or other materials provided with the distribution; 26 * neither the name of the copyright holders nor the names of its 27 * contributors may be used to endorse or promote products derived from 28 * this software without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 31 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 32 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 33 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 34 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 35 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 36 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 37 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 38 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 39 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 40 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 41 * 42 * Authors: Kevin Lim 43 * Korey Sewell 44 * Rick Strong 45 */ 46 47#ifndef __CPU_O3_CPU_HH__ 48#define __CPU_O3_CPU_HH__ 49 50#include <iostream> 51#include <list> 52#include <queue> 53#include <set> 54#include <vector> 55 56#include "arch/generic/types.hh" 57#include "arch/types.hh" 58#include "base/statistics.hh" 59#include "config/the_isa.hh" 60#include "cpu/o3/comm.hh" 61#include "cpu/o3/cpu_policy.hh" 62#include "cpu/o3/scoreboard.hh" 63#include "cpu/o3/thread_state.hh" 64#include "cpu/activity.hh" 65#include "cpu/base.hh" 66#include "cpu/simple_thread.hh" 67#include "cpu/timebuf.hh" 68//#include "cpu/o3/thread_context.hh" 69#include "params/DerivO3CPU.hh" 70#include "sim/process.hh" 71 72template <class> 73class Checker; 74class ThreadContext; 75template <class> 76class O3ThreadContext; 77 78class Checkpoint; | 1/* 2 * Copyright (c) 2011-2013, 2016-2019 ARM Limited 3 * Copyright (c) 2013 Advanced Micro Devices, Inc. 4 * All rights reserved 5 * 6 * The license below extends only to copyright in the software and shall 7 * not be construed as granting a license to any other intellectual 8 * property including but not limited to intellectual property relating 9 * to a hardware implementation of the functionality of the software 10 * licensed hereunder. You may use the software subject to the license 11 * terms below provided that you ensure that this notice is replicated 12 * unmodified and in its entirety in all distributions of the software, 13 * modified or unmodified, in source code or in binary form. 14 * 15 * Copyright (c) 2004-2005 The Regents of The University of Michigan 16 * Copyright (c) 2011 Regents of the University of California 17 * All rights reserved. 18 * 19 * Redistribution and use in source and binary forms, with or without 20 * modification, are permitted provided that the following conditions are 21 * met: redistributions of source code must retain the above copyright 22 * notice, this list of conditions and the following disclaimer; 23 * redistributions in binary form must reproduce the above copyright 24 * notice, this list of conditions and the following disclaimer in the 25 * documentation and/or other materials provided with the distribution; 26 * neither the name of the copyright holders nor the names of its 27 * contributors may be used to endorse or promote products derived from 28 * this software without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 31 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 32 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 33 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 34 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 35 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 36 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 37 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 38 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 39 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 40 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 41 * 42 * Authors: Kevin Lim 43 * Korey Sewell 44 * Rick Strong 45 */ 46 47#ifndef __CPU_O3_CPU_HH__ 48#define __CPU_O3_CPU_HH__ 49 50#include <iostream> 51#include <list> 52#include <queue> 53#include <set> 54#include <vector> 55 56#include "arch/generic/types.hh" 57#include "arch/types.hh" 58#include "base/statistics.hh" 59#include "config/the_isa.hh" 60#include "cpu/o3/comm.hh" 61#include "cpu/o3/cpu_policy.hh" 62#include "cpu/o3/scoreboard.hh" 63#include "cpu/o3/thread_state.hh" 64#include "cpu/activity.hh" 65#include "cpu/base.hh" 66#include "cpu/simple_thread.hh" 67#include "cpu/timebuf.hh" 68//#include "cpu/o3/thread_context.hh" 69#include "params/DerivO3CPU.hh" 70#include "sim/process.hh" 71 72template <class> 73class Checker; 74class ThreadContext; 75template <class> 76class O3ThreadContext; 77 78class Checkpoint; |
79class MemObject; | |
80class Process; 81 82struct BaseCPUParams; 83 84class BaseO3CPU : public BaseCPU 85{ 86 //Stuff that's pretty ISA independent will go here. 87 public: 88 BaseO3CPU(BaseCPUParams *params); 89 90 void regStats(); 91}; 92 93/** 94 * FullO3CPU class, has each of the stages (fetch through commit) 95 * within it, as well as all of the time buffers between stages. The 96 * tick() function for the CPU is defined here. 97 */ 98template <class Impl> 99class FullO3CPU : public BaseO3CPU 100{ 101 public: 102 // Typedefs from the Impl here. 103 typedef typename Impl::CPUPol CPUPolicy; 104 typedef typename Impl::DynInstPtr DynInstPtr; 105 typedef typename Impl::O3CPU O3CPU; 106 107 using VecElem = TheISA::VecElem; 108 using VecRegContainer = TheISA::VecRegContainer; 109 110 using VecPredRegContainer = TheISA::VecPredRegContainer; 111 112 typedef O3ThreadState<Impl> ImplState; 113 typedef O3ThreadState<Impl> Thread; 114 115 typedef typename std::list<DynInstPtr>::iterator ListIt; 116 117 friend class O3ThreadContext<Impl>; 118 119 public: 120 enum Status { 121 Running, 122 Idle, 123 Halted, 124 Blocked, 125 SwitchedOut 126 }; 127 128 BaseTLB *itb; 129 BaseTLB *dtb; 130 using LSQRequest = typename LSQ<Impl>::LSQRequest; 131 132 /** Overall CPU status. */ 133 Status _status; 134 135 private: 136 137 /** 138 * IcachePort class for instruction fetch. 139 */ 140 class IcachePort : public MasterPort 141 { 142 protected: 143 /** Pointer to fetch. */ 144 DefaultFetch<Impl> *fetch; 145 146 public: 147 /** Default constructor. */ 148 IcachePort(DefaultFetch<Impl> *_fetch, FullO3CPU<Impl>* _cpu) 149 : MasterPort(_cpu->name() + ".icache_port", _cpu), fetch(_fetch) 150 { } 151 152 protected: 153 154 /** Timing version of receive. Handles setting fetch to the 155 * proper status to start fetching. */ 156 virtual bool recvTimingResp(PacketPtr pkt); 157 158 /** Handles doing a retry of a failed fetch. */ 159 virtual void recvReqRetry(); 160 }; 161 162 /** 163 * DcachePort class for the load/store queue. 164 */ 165 class DcachePort : public MasterPort 166 { 167 protected: 168 169 /** Pointer to LSQ. */ 170 LSQ<Impl> *lsq; 171 FullO3CPU<Impl> *cpu; 172 173 public: 174 /** Default constructor. */ 175 DcachePort(LSQ<Impl> *_lsq, FullO3CPU<Impl>* _cpu) 176 : MasterPort(_cpu->name() + ".dcache_port", _cpu), lsq(_lsq), 177 cpu(_cpu) 178 { } 179 180 protected: 181 182 /** Timing version of receive. Handles writing back and 183 * completing the load or store that has returned from 184 * memory. */ 185 virtual bool recvTimingResp(PacketPtr pkt); 186 virtual void recvTimingSnoopReq(PacketPtr pkt); 187 188 virtual void recvFunctionalSnoop(PacketPtr pkt) 189 { 190 // @todo: Is there a need for potential invalidation here? 191 } 192 193 /** Handles doing a retry of the previous send. */ 194 virtual void recvReqRetry(); 195 196 /** 197 * As this CPU requires snooping to maintain the load store queue 198 * change the behaviour from the base CPU port. 199 * 200 * @return true since we have to snoop 201 */ 202 virtual bool isSnooping() const { return true; } 203 }; 204 205 /** The tick event used for scheduling CPU ticks. */ 206 EventFunctionWrapper tickEvent; 207 208 /** The exit event used for terminating all ready-to-exit threads */ 209 EventFunctionWrapper threadExitEvent; 210 211 /** Schedule tick event, regardless of its current state. */ 212 void scheduleTickEvent(Cycles delay) 213 { 214 if (tickEvent.squashed()) 215 reschedule(tickEvent, clockEdge(delay)); 216 else if (!tickEvent.scheduled()) 217 schedule(tickEvent, clockEdge(delay)); 218 } 219 220 /** Unschedule tick event, regardless of its current state. */ 221 void unscheduleTickEvent() 222 { 223 if (tickEvent.scheduled()) 224 tickEvent.squash(); 225 } 226 227 /** 228 * Check if the pipeline has drained and signal drain done. 229 * 230 * This method checks if a drain has been requested and if the CPU 231 * has drained successfully (i.e., there are no instructions in 232 * the pipeline). If the CPU has drained, it deschedules the tick 233 * event and signals the drain manager. 234 * 235 * @return False if a drain hasn't been requested or the CPU 236 * hasn't drained, true otherwise. 237 */ 238 bool tryDrain(); 239 240 /** 241 * Perform sanity checks after a drain. 242 * 243 * This method is called from drain() when it has determined that 244 * the CPU is fully drained when gem5 is compiled with the NDEBUG 245 * macro undefined. The intention of this method is to do more 246 * extensive tests than the isDrained() method to weed out any 247 * draining bugs. 248 */ 249 void drainSanityCheck() const; 250 251 /** Check if a system is in a drained state. */ 252 bool isDrained() const; 253 254 public: 255 /** Constructs a CPU with the given parameters. */ 256 FullO3CPU(DerivO3CPUParams *params); 257 /** Destructor. */ 258 ~FullO3CPU(); 259 260 /** Registers statistics. */ 261 void regStats() override; 262 263 ProbePointArg<PacketPtr> *ppInstAccessComplete; 264 ProbePointArg<std::pair<DynInstPtr, PacketPtr> > *ppDataAccessComplete; 265 266 /** Register probe points. */ 267 void regProbePoints() override; 268 269 void demapPage(Addr vaddr, uint64_t asn) 270 { 271 this->itb->demapPage(vaddr, asn); 272 this->dtb->demapPage(vaddr, asn); 273 } 274 275 void demapInstPage(Addr vaddr, uint64_t asn) 276 { 277 this->itb->demapPage(vaddr, asn); 278 } 279 280 void demapDataPage(Addr vaddr, uint64_t asn) 281 { 282 this->dtb->demapPage(vaddr, asn); 283 } 284 285 /** Ticks CPU, calling tick() on each stage, and checking the overall 286 * activity to see if the CPU should deschedule itself. 287 */ 288 void tick(); 289 290 /** Initialize the CPU */ 291 void init() override; 292 293 void startup() override; 294 295 /** Returns the Number of Active Threads in the CPU */ 296 int numActiveThreads() 297 { return activeThreads.size(); } 298 299 /** Add Thread to Active Threads List */ 300 void activateThread(ThreadID tid); 301 302 /** Remove Thread from Active Threads List */ 303 void deactivateThread(ThreadID tid); 304 305 /** Setup CPU to insert a thread's context */ 306 void insertThread(ThreadID tid); 307 308 /** Remove all of a thread's context from CPU */ 309 void removeThread(ThreadID tid); 310 311 /** Count the Total Instructions Committed in the CPU. */ 312 Counter totalInsts() const override; 313 314 /** Count the Total Ops (including micro ops) committed in the CPU. */ 315 Counter totalOps() const override; 316 317 /** Add Thread to Active Threads List. */ 318 void activateContext(ThreadID tid) override; 319 320 /** Remove Thread from Active Threads List */ 321 void suspendContext(ThreadID tid) override; 322 323 /** Remove Thread from Active Threads List && 324 * Remove Thread Context from CPU. 325 */ 326 void haltContext(ThreadID tid) override; 327 328 /** Update The Order In Which We Process Threads. */ 329 void updateThreadPriority(); 330 331 /** Is the CPU draining? */ 332 bool isDraining() const { return drainState() == DrainState::Draining; } 333 334 void serializeThread(CheckpointOut &cp, ThreadID tid) const override; 335 void unserializeThread(CheckpointIn &cp, ThreadID tid) override; 336 337 /** Insert tid to the list of threads trying to exit */ 338 void addThreadToExitingList(ThreadID tid); 339 340 /** Is the thread trying to exit? */ 341 bool isThreadExiting(ThreadID tid) const; 342 343 /** 344 * If a thread is trying to exit and its corresponding trap event 345 * has been completed, schedule an event to terminate the thread. 346 */ 347 void scheduleThreadExitEvent(ThreadID tid); 348 349 /** Terminate all threads that are ready to exit */ 350 void exitThreads(); 351 352 public: 353 /** Executes a syscall. 354 * @todo: Determine if this needs to be virtual. 355 */ 356 void syscall(int64_t callnum, ThreadID tid, Fault *fault); 357 358 /** Starts draining the CPU's pipeline of all instructions in 359 * order to stop all memory accesses. */ 360 DrainState drain() override; 361 362 /** Resumes execution after a drain. */ 363 void drainResume() override; 364 365 /** 366 * Commit has reached a safe point to drain a thread. 367 * 368 * Commit calls this method to inform the pipeline that it has 369 * reached a point where it is not executed microcode and is about 370 * to squash uncommitted instructions to fully drain the pipeline. 371 */ 372 void commitDrained(ThreadID tid); 373 374 /** Switches out this CPU. */ 375 void switchOut() override; 376 377 /** Takes over from another CPU. */ 378 void takeOverFrom(BaseCPU *oldCPU) override; 379 380 void verifyMemoryMode() const override; 381 382 /** Get the current instruction sequence number, and increment it. */ 383 InstSeqNum getAndIncrementInstSeq() 384 { return globalSeqNum++; } 385 386 /** Traps to handle given fault. */ 387 void trap(const Fault &fault, ThreadID tid, const StaticInstPtr &inst); 388 389 /** HW return from error interrupt. */ 390 Fault hwrei(ThreadID tid); 391 392 bool simPalCheck(int palFunc, ThreadID tid); 393 394 /** Check if a change in renaming is needed for vector registers. 395 * The vecMode variable is updated and propagated to rename maps. 396 * 397 * @param tid ThreadID 398 * @param freelist list of free registers 399 */ 400 void switchRenameMode(ThreadID tid, UnifiedFreeList* freelist); 401 402 /** Returns the Fault for any valid interrupt. */ 403 Fault getInterrupts(); 404 405 /** Processes any an interrupt fault. */ 406 void processInterrupts(const Fault &interrupt); 407 408 /** Halts the CPU. */ 409 void halt() { panic("Halt not implemented!\n"); } 410 411 /** Register accessors. Index refers to the physical register index. */ 412 413 /** Reads a miscellaneous register. */ 414 RegVal readMiscRegNoEffect(int misc_reg, ThreadID tid) const; 415 416 /** Reads a misc. register, including any side effects the read 417 * might have as defined by the architecture. 418 */ 419 RegVal readMiscReg(int misc_reg, ThreadID tid); 420 421 /** Sets a miscellaneous register. */ 422 void setMiscRegNoEffect(int misc_reg, RegVal val, ThreadID tid); 423 424 /** Sets a misc. register, including any side effects the write 425 * might have as defined by the architecture. 426 */ 427 void setMiscReg(int misc_reg, RegVal val, ThreadID tid); 428 429 RegVal readIntReg(PhysRegIdPtr phys_reg); 430 431 RegVal readFloatReg(PhysRegIdPtr phys_reg); 432 433 const VecRegContainer& readVecReg(PhysRegIdPtr reg_idx) const; 434 435 /** 436 * Read physical vector register for modification. 437 */ 438 VecRegContainer& getWritableVecReg(PhysRegIdPtr reg_idx); 439 440 /** Returns current vector renaming mode */ 441 Enums::VecRegRenameMode vecRenameMode() const { return vecMode; } 442 443 /** Sets the current vector renaming mode */ 444 void vecRenameMode(Enums::VecRegRenameMode vec_mode) 445 { vecMode = vec_mode; } 446 447 /** 448 * Read physical vector register lane 449 */ 450 template<typename VecElem, int LaneIdx> 451 VecLaneT<VecElem, true> 452 readVecLane(PhysRegIdPtr phys_reg) const 453 { 454 vecRegfileReads++; 455 return regFile.readVecLane<VecElem, LaneIdx>(phys_reg); 456 } 457 458 /** 459 * Read physical vector register lane 460 */ 461 template<typename VecElem> 462 VecLaneT<VecElem, true> 463 readVecLane(PhysRegIdPtr phys_reg) const 464 { 465 vecRegfileReads++; 466 return regFile.readVecLane<VecElem>(phys_reg); 467 } 468 469 /** Write a lane of the destination vector register. */ 470 template<typename LD> 471 void 472 setVecLane(PhysRegIdPtr phys_reg, const LD& val) 473 { 474 vecRegfileWrites++; 475 return regFile.setVecLane(phys_reg, val); 476 } 477 478 const VecElem& readVecElem(PhysRegIdPtr reg_idx) const; 479 480 const VecPredRegContainer& readVecPredReg(PhysRegIdPtr reg_idx) const; 481 482 VecPredRegContainer& getWritableVecPredReg(PhysRegIdPtr reg_idx); 483 484 RegVal readCCReg(PhysRegIdPtr phys_reg); 485 486 void setIntReg(PhysRegIdPtr phys_reg, RegVal val); 487 488 void setFloatReg(PhysRegIdPtr phys_reg, RegVal val); 489 490 void setVecReg(PhysRegIdPtr reg_idx, const VecRegContainer& val); 491 492 void setVecElem(PhysRegIdPtr reg_idx, const VecElem& val); 493 494 void setVecPredReg(PhysRegIdPtr reg_idx, const VecPredRegContainer& val); 495 496 void setCCReg(PhysRegIdPtr phys_reg, RegVal val); 497 498 RegVal readArchIntReg(int reg_idx, ThreadID tid); 499 500 RegVal readArchFloatReg(int reg_idx, ThreadID tid); 501 502 const VecRegContainer& readArchVecReg(int reg_idx, ThreadID tid) const; 503 /** Read architectural vector register for modification. */ 504 VecRegContainer& getWritableArchVecReg(int reg_idx, ThreadID tid); 505 506 /** Read architectural vector register lane. */ 507 template<typename VecElem> 508 VecLaneT<VecElem, true> 509 readArchVecLane(int reg_idx, int lId, ThreadID tid) const 510 { 511 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 512 RegId(VecRegClass, reg_idx)); 513 return readVecLane<VecElem>(phys_reg); 514 } 515 516 517 /** Write a lane of the destination vector register. */ 518 template<typename LD> 519 void 520 setArchVecLane(int reg_idx, int lId, ThreadID tid, const LD& val) 521 { 522 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 523 RegId(VecRegClass, reg_idx)); 524 setVecLane(phys_reg, val); 525 } 526 527 const VecElem& readArchVecElem(const RegIndex& reg_idx, 528 const ElemIndex& ldx, ThreadID tid) const; 529 530 const VecPredRegContainer& readArchVecPredReg(int reg_idx, 531 ThreadID tid) const; 532 533 VecPredRegContainer& getWritableArchVecPredReg(int reg_idx, ThreadID tid); 534 535 RegVal readArchCCReg(int reg_idx, ThreadID tid); 536 537 /** Architectural register accessors. Looks up in the commit 538 * rename table to obtain the true physical index of the 539 * architected register first, then accesses that physical 540 * register. 541 */ 542 void setArchIntReg(int reg_idx, RegVal val, ThreadID tid); 543 544 void setArchFloatReg(int reg_idx, RegVal val, ThreadID tid); 545 546 void setArchVecPredReg(int reg_idx, const VecPredRegContainer& val, 547 ThreadID tid); 548 549 void setArchVecReg(int reg_idx, const VecRegContainer& val, ThreadID tid); 550 551 void setArchVecElem(const RegIndex& reg_idx, const ElemIndex& ldx, 552 const VecElem& val, ThreadID tid); 553 554 void setArchCCReg(int reg_idx, RegVal val, ThreadID tid); 555 556 /** Sets the commit PC state of a specific thread. */ 557 void pcState(const TheISA::PCState &newPCState, ThreadID tid); 558 559 /** Reads the commit PC state of a specific thread. */ 560 TheISA::PCState pcState(ThreadID tid); 561 562 /** Reads the commit PC of a specific thread. */ 563 Addr instAddr(ThreadID tid); 564 565 /** Reads the commit micro PC of a specific thread. */ 566 MicroPC microPC(ThreadID tid); 567 568 /** Reads the next PC of a specific thread. */ 569 Addr nextInstAddr(ThreadID tid); 570 571 /** Initiates a squash of all in-flight instructions for a given 572 * thread. The source of the squash is an external update of 573 * state through the TC. 574 */ 575 void squashFromTC(ThreadID tid); 576 577 /** Function to add instruction onto the head of the list of the 578 * instructions. Used when new instructions are fetched. 579 */ 580 ListIt addInst(const DynInstPtr &inst); 581 582 /** Function to tell the CPU that an instruction has completed. */ 583 void instDone(ThreadID tid, const DynInstPtr &inst); 584 585 /** Remove an instruction from the front end of the list. There's 586 * no restriction on location of the instruction. 587 */ 588 void removeFrontInst(const DynInstPtr &inst); 589 590 /** Remove all instructions that are not currently in the ROB. 591 * There's also an option to not squash delay slot instructions.*/ 592 void removeInstsNotInROB(ThreadID tid); 593 594 /** Remove all instructions younger than the given sequence number. */ 595 void removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid); 596 597 /** Removes the instruction pointed to by the iterator. */ 598 inline void squashInstIt(const ListIt &instIt, ThreadID tid); 599 600 /** Cleans up all instructions on the remove list. */ 601 void cleanUpRemovedInsts(); 602 603 /** Debug function to print all instructions on the list. */ 604 void dumpInsts(); 605 606 public: 607#ifndef NDEBUG 608 /** Count of total number of dynamic instructions in flight. */ 609 int instcount; 610#endif 611 612 /** List of all the instructions in flight. */ 613 std::list<DynInstPtr> instList; 614 615 /** List of all the instructions that will be removed at the end of this 616 * cycle. 617 */ 618 std::queue<ListIt> removeList; 619 620#ifdef DEBUG 621 /** Debug structure to keep track of the sequence numbers still in 622 * flight. 623 */ 624 std::set<InstSeqNum> snList; 625#endif 626 627 /** Records if instructions need to be removed this cycle due to 628 * being retired or squashed. 629 */ 630 bool removeInstsThisCycle; 631 632 protected: 633 /** The fetch stage. */ 634 typename CPUPolicy::Fetch fetch; 635 636 /** The decode stage. */ 637 typename CPUPolicy::Decode decode; 638 639 /** The dispatch stage. */ 640 typename CPUPolicy::Rename rename; 641 642 /** The issue/execute/writeback stages. */ 643 typename CPUPolicy::IEW iew; 644 645 /** The commit stage. */ 646 typename CPUPolicy::Commit commit; 647 648 /** The rename mode of the vector registers */ 649 Enums::VecRegRenameMode vecMode; 650 651 /** The register file. */ 652 PhysRegFile regFile; 653 654 /** The free list. */ 655 typename CPUPolicy::FreeList freeList; 656 657 /** The rename map. */ 658 typename CPUPolicy::RenameMap renameMap[Impl::MaxThreads]; 659 660 /** The commit rename map. */ 661 typename CPUPolicy::RenameMap commitRenameMap[Impl::MaxThreads]; 662 663 /** The re-order buffer. */ 664 typename CPUPolicy::ROB rob; 665 666 /** Active Threads List */ 667 std::list<ThreadID> activeThreads; 668 669 /** 670 * This is a list of threads that are trying to exit. Each thread id 671 * is mapped to a boolean value denoting whether the thread is ready 672 * to exit. 673 */ 674 std::unordered_map<ThreadID, bool> exitingThreads; 675 676 /** Integer Register Scoreboard */ 677 Scoreboard scoreboard; 678 679 std::vector<TheISA::ISA *> isa; 680 681 /** Instruction port. Note that it has to appear after the fetch stage. */ 682 IcachePort icachePort; 683 684 /** Data port. Note that it has to appear after the iew stages */ 685 DcachePort dcachePort; 686 687 public: 688 /** Enum to give each stage a specific index, so when calling 689 * activateStage() or deactivateStage(), they can specify which stage 690 * is being activated/deactivated. 691 */ 692 enum StageIdx { 693 FetchIdx, 694 DecodeIdx, 695 RenameIdx, 696 IEWIdx, 697 CommitIdx, 698 NumStages }; 699 700 /** Typedefs from the Impl to get the structs that each of the 701 * time buffers should use. 702 */ 703 typedef typename CPUPolicy::TimeStruct TimeStruct; 704 705 typedef typename CPUPolicy::FetchStruct FetchStruct; 706 707 typedef typename CPUPolicy::DecodeStruct DecodeStruct; 708 709 typedef typename CPUPolicy::RenameStruct RenameStruct; 710 711 typedef typename CPUPolicy::IEWStruct IEWStruct; 712 713 /** The main time buffer to do backwards communication. */ 714 TimeBuffer<TimeStruct> timeBuffer; 715 716 /** The fetch stage's instruction queue. */ 717 TimeBuffer<FetchStruct> fetchQueue; 718 719 /** The decode stage's instruction queue. */ 720 TimeBuffer<DecodeStruct> decodeQueue; 721 722 /** The rename stage's instruction queue. */ 723 TimeBuffer<RenameStruct> renameQueue; 724 725 /** The IEW stage's instruction queue. */ 726 TimeBuffer<IEWStruct> iewQueue; 727 728 private: 729 /** The activity recorder; used to tell if the CPU has any 730 * activity remaining or if it can go to idle and deschedule 731 * itself. 732 */ 733 ActivityRecorder activityRec; 734 735 public: 736 /** Records that there was time buffer activity this cycle. */ 737 void activityThisCycle() { activityRec.activity(); } 738 739 /** Changes a stage's status to active within the activity recorder. */ 740 void activateStage(const StageIdx idx) 741 { activityRec.activateStage(idx); } 742 743 /** Changes a stage's status to inactive within the activity recorder. */ 744 void deactivateStage(const StageIdx idx) 745 { activityRec.deactivateStage(idx); } 746 747 /** Wakes the CPU, rescheduling the CPU if it's not already active. */ 748 void wakeCPU(); 749 750 virtual void wakeup(ThreadID tid) override; 751 752 /** Gets a free thread id. Use if thread ids change across system. */ 753 ThreadID getFreeTid(); 754 755 public: 756 /** Returns a pointer to a thread context. */ 757 ThreadContext * 758 tcBase(ThreadID tid) 759 { 760 return thread[tid]->getTC(); 761 } 762 763 /** The global sequence number counter. */ 764 InstSeqNum globalSeqNum;//[Impl::MaxThreads]; 765 766 /** Pointer to the checker, which can dynamically verify 767 * instruction results at run time. This can be set to NULL if it 768 * is not being used. 769 */ 770 Checker<Impl> *checker; 771 772 /** Pointer to the system. */ 773 System *system; 774 775 /** Pointers to all of the threads in the CPU. */ 776 std::vector<Thread *> thread; 777 778 /** Threads Scheduled to Enter CPU */ 779 std::list<int> cpuWaitList; 780 781 /** The cycle that the CPU was last running, used for statistics. */ 782 Cycles lastRunningCycle; 783 784 /** The cycle that the CPU was last activated by a new thread*/ 785 Tick lastActivatedCycle; 786 787 /** Mapping for system thread id to cpu id */ 788 std::map<ThreadID, unsigned> threadMap; 789 790 /** Available thread ids in the cpu*/ 791 std::vector<ThreadID> tids; 792 793 /** CPU pushRequest function, forwards request to LSQ. */ 794 Fault pushRequest(const DynInstPtr& inst, bool isLoad, uint8_t *data, 795 unsigned int size, Addr addr, Request::Flags flags, 796 uint64_t *res, AtomicOpFunctor *amo_op = nullptr) 797 { 798 return iew.ldstQueue.pushRequest(inst, isLoad, data, size, addr, 799 flags, res, amo_op); 800 } 801 802 /** CPU read function, forwards read to LSQ. */ 803 Fault read(LSQRequest* req, int load_idx) 804 { 805 return this->iew.ldstQueue.read(req, load_idx); 806 } 807 808 /** CPU write function, forwards write to LSQ. */ 809 Fault write(LSQRequest* req, uint8_t *data, int store_idx) 810 { 811 return this->iew.ldstQueue.write(req, data, store_idx); 812 } 813 814 /** Used by the fetch unit to get a hold of the instruction port. */ 815 MasterPort &getInstPort() override { return icachePort; } 816 817 /** Get the dcache port (used to find block size for translations). */ 818 MasterPort &getDataPort() override { return dcachePort; } 819 820 /** Stat for total number of times the CPU is descheduled. */ 821 Stats::Scalar timesIdled; 822 /** Stat for total number of cycles the CPU spends descheduled. */ 823 Stats::Scalar idleCycles; 824 /** Stat for total number of cycles the CPU spends descheduled due to a 825 * quiesce operation or waiting for an interrupt. */ 826 Stats::Scalar quiesceCycles; 827 /** Stat for the number of committed instructions per thread. */ 828 Stats::Vector committedInsts; 829 /** Stat for the number of committed ops (including micro ops) per thread. */ 830 Stats::Vector committedOps; 831 /** Stat for the CPI per thread. */ 832 Stats::Formula cpi; 833 /** Stat for the total CPI. */ 834 Stats::Formula totalCpi; 835 /** Stat for the IPC per thread. */ 836 Stats::Formula ipc; 837 /** Stat for the total IPC. */ 838 Stats::Formula totalIpc; 839 840 //number of integer register file accesses 841 Stats::Scalar intRegfileReads; 842 Stats::Scalar intRegfileWrites; 843 //number of float register file accesses 844 Stats::Scalar fpRegfileReads; 845 Stats::Scalar fpRegfileWrites; 846 //number of vector register file accesses 847 mutable Stats::Scalar vecRegfileReads; 848 Stats::Scalar vecRegfileWrites; 849 //number of predicate register file accesses 850 mutable Stats::Scalar vecPredRegfileReads; 851 Stats::Scalar vecPredRegfileWrites; 852 //number of CC register file accesses 853 Stats::Scalar ccRegfileReads; 854 Stats::Scalar ccRegfileWrites; 855 //number of misc 856 Stats::Scalar miscRegfileReads; 857 Stats::Scalar miscRegfileWrites; 858}; 859 860#endif // __CPU_O3_CPU_HH__ | 79class Process; 80 81struct BaseCPUParams; 82 83class BaseO3CPU : public BaseCPU 84{ 85 //Stuff that's pretty ISA independent will go here. 86 public: 87 BaseO3CPU(BaseCPUParams *params); 88 89 void regStats(); 90}; 91 92/** 93 * FullO3CPU class, has each of the stages (fetch through commit) 94 * within it, as well as all of the time buffers between stages. The 95 * tick() function for the CPU is defined here. 96 */ 97template <class Impl> 98class FullO3CPU : public BaseO3CPU 99{ 100 public: 101 // Typedefs from the Impl here. 102 typedef typename Impl::CPUPol CPUPolicy; 103 typedef typename Impl::DynInstPtr DynInstPtr; 104 typedef typename Impl::O3CPU O3CPU; 105 106 using VecElem = TheISA::VecElem; 107 using VecRegContainer = TheISA::VecRegContainer; 108 109 using VecPredRegContainer = TheISA::VecPredRegContainer; 110 111 typedef O3ThreadState<Impl> ImplState; 112 typedef O3ThreadState<Impl> Thread; 113 114 typedef typename std::list<DynInstPtr>::iterator ListIt; 115 116 friend class O3ThreadContext<Impl>; 117 118 public: 119 enum Status { 120 Running, 121 Idle, 122 Halted, 123 Blocked, 124 SwitchedOut 125 }; 126 127 BaseTLB *itb; 128 BaseTLB *dtb; 129 using LSQRequest = typename LSQ<Impl>::LSQRequest; 130 131 /** Overall CPU status. */ 132 Status _status; 133 134 private: 135 136 /** 137 * IcachePort class for instruction fetch. 138 */ 139 class IcachePort : public MasterPort 140 { 141 protected: 142 /** Pointer to fetch. */ 143 DefaultFetch<Impl> *fetch; 144 145 public: 146 /** Default constructor. */ 147 IcachePort(DefaultFetch<Impl> *_fetch, FullO3CPU<Impl>* _cpu) 148 : MasterPort(_cpu->name() + ".icache_port", _cpu), fetch(_fetch) 149 { } 150 151 protected: 152 153 /** Timing version of receive. Handles setting fetch to the 154 * proper status to start fetching. */ 155 virtual bool recvTimingResp(PacketPtr pkt); 156 157 /** Handles doing a retry of a failed fetch. */ 158 virtual void recvReqRetry(); 159 }; 160 161 /** 162 * DcachePort class for the load/store queue. 163 */ 164 class DcachePort : public MasterPort 165 { 166 protected: 167 168 /** Pointer to LSQ. */ 169 LSQ<Impl> *lsq; 170 FullO3CPU<Impl> *cpu; 171 172 public: 173 /** Default constructor. */ 174 DcachePort(LSQ<Impl> *_lsq, FullO3CPU<Impl>* _cpu) 175 : MasterPort(_cpu->name() + ".dcache_port", _cpu), lsq(_lsq), 176 cpu(_cpu) 177 { } 178 179 protected: 180 181 /** Timing version of receive. Handles writing back and 182 * completing the load or store that has returned from 183 * memory. */ 184 virtual bool recvTimingResp(PacketPtr pkt); 185 virtual void recvTimingSnoopReq(PacketPtr pkt); 186 187 virtual void recvFunctionalSnoop(PacketPtr pkt) 188 { 189 // @todo: Is there a need for potential invalidation here? 190 } 191 192 /** Handles doing a retry of the previous send. */ 193 virtual void recvReqRetry(); 194 195 /** 196 * As this CPU requires snooping to maintain the load store queue 197 * change the behaviour from the base CPU port. 198 * 199 * @return true since we have to snoop 200 */ 201 virtual bool isSnooping() const { return true; } 202 }; 203 204 /** The tick event used for scheduling CPU ticks. */ 205 EventFunctionWrapper tickEvent; 206 207 /** The exit event used for terminating all ready-to-exit threads */ 208 EventFunctionWrapper threadExitEvent; 209 210 /** Schedule tick event, regardless of its current state. */ 211 void scheduleTickEvent(Cycles delay) 212 { 213 if (tickEvent.squashed()) 214 reschedule(tickEvent, clockEdge(delay)); 215 else if (!tickEvent.scheduled()) 216 schedule(tickEvent, clockEdge(delay)); 217 } 218 219 /** Unschedule tick event, regardless of its current state. */ 220 void unscheduleTickEvent() 221 { 222 if (tickEvent.scheduled()) 223 tickEvent.squash(); 224 } 225 226 /** 227 * Check if the pipeline has drained and signal drain done. 228 * 229 * This method checks if a drain has been requested and if the CPU 230 * has drained successfully (i.e., there are no instructions in 231 * the pipeline). If the CPU has drained, it deschedules the tick 232 * event and signals the drain manager. 233 * 234 * @return False if a drain hasn't been requested or the CPU 235 * hasn't drained, true otherwise. 236 */ 237 bool tryDrain(); 238 239 /** 240 * Perform sanity checks after a drain. 241 * 242 * This method is called from drain() when it has determined that 243 * the CPU is fully drained when gem5 is compiled with the NDEBUG 244 * macro undefined. The intention of this method is to do more 245 * extensive tests than the isDrained() method to weed out any 246 * draining bugs. 247 */ 248 void drainSanityCheck() const; 249 250 /** Check if a system is in a drained state. */ 251 bool isDrained() const; 252 253 public: 254 /** Constructs a CPU with the given parameters. */ 255 FullO3CPU(DerivO3CPUParams *params); 256 /** Destructor. */ 257 ~FullO3CPU(); 258 259 /** Registers statistics. */ 260 void regStats() override; 261 262 ProbePointArg<PacketPtr> *ppInstAccessComplete; 263 ProbePointArg<std::pair<DynInstPtr, PacketPtr> > *ppDataAccessComplete; 264 265 /** Register probe points. */ 266 void regProbePoints() override; 267 268 void demapPage(Addr vaddr, uint64_t asn) 269 { 270 this->itb->demapPage(vaddr, asn); 271 this->dtb->demapPage(vaddr, asn); 272 } 273 274 void demapInstPage(Addr vaddr, uint64_t asn) 275 { 276 this->itb->demapPage(vaddr, asn); 277 } 278 279 void demapDataPage(Addr vaddr, uint64_t asn) 280 { 281 this->dtb->demapPage(vaddr, asn); 282 } 283 284 /** Ticks CPU, calling tick() on each stage, and checking the overall 285 * activity to see if the CPU should deschedule itself. 286 */ 287 void tick(); 288 289 /** Initialize the CPU */ 290 void init() override; 291 292 void startup() override; 293 294 /** Returns the Number of Active Threads in the CPU */ 295 int numActiveThreads() 296 { return activeThreads.size(); } 297 298 /** Add Thread to Active Threads List */ 299 void activateThread(ThreadID tid); 300 301 /** Remove Thread from Active Threads List */ 302 void deactivateThread(ThreadID tid); 303 304 /** Setup CPU to insert a thread's context */ 305 void insertThread(ThreadID tid); 306 307 /** Remove all of a thread's context from CPU */ 308 void removeThread(ThreadID tid); 309 310 /** Count the Total Instructions Committed in the CPU. */ 311 Counter totalInsts() const override; 312 313 /** Count the Total Ops (including micro ops) committed in the CPU. */ 314 Counter totalOps() const override; 315 316 /** Add Thread to Active Threads List. */ 317 void activateContext(ThreadID tid) override; 318 319 /** Remove Thread from Active Threads List */ 320 void suspendContext(ThreadID tid) override; 321 322 /** Remove Thread from Active Threads List && 323 * Remove Thread Context from CPU. 324 */ 325 void haltContext(ThreadID tid) override; 326 327 /** Update The Order In Which We Process Threads. */ 328 void updateThreadPriority(); 329 330 /** Is the CPU draining? */ 331 bool isDraining() const { return drainState() == DrainState::Draining; } 332 333 void serializeThread(CheckpointOut &cp, ThreadID tid) const override; 334 void unserializeThread(CheckpointIn &cp, ThreadID tid) override; 335 336 /** Insert tid to the list of threads trying to exit */ 337 void addThreadToExitingList(ThreadID tid); 338 339 /** Is the thread trying to exit? */ 340 bool isThreadExiting(ThreadID tid) const; 341 342 /** 343 * If a thread is trying to exit and its corresponding trap event 344 * has been completed, schedule an event to terminate the thread. 345 */ 346 void scheduleThreadExitEvent(ThreadID tid); 347 348 /** Terminate all threads that are ready to exit */ 349 void exitThreads(); 350 351 public: 352 /** Executes a syscall. 353 * @todo: Determine if this needs to be virtual. 354 */ 355 void syscall(int64_t callnum, ThreadID tid, Fault *fault); 356 357 /** Starts draining the CPU's pipeline of all instructions in 358 * order to stop all memory accesses. */ 359 DrainState drain() override; 360 361 /** Resumes execution after a drain. */ 362 void drainResume() override; 363 364 /** 365 * Commit has reached a safe point to drain a thread. 366 * 367 * Commit calls this method to inform the pipeline that it has 368 * reached a point where it is not executed microcode and is about 369 * to squash uncommitted instructions to fully drain the pipeline. 370 */ 371 void commitDrained(ThreadID tid); 372 373 /** Switches out this CPU. */ 374 void switchOut() override; 375 376 /** Takes over from another CPU. */ 377 void takeOverFrom(BaseCPU *oldCPU) override; 378 379 void verifyMemoryMode() const override; 380 381 /** Get the current instruction sequence number, and increment it. */ 382 InstSeqNum getAndIncrementInstSeq() 383 { return globalSeqNum++; } 384 385 /** Traps to handle given fault. */ 386 void trap(const Fault &fault, ThreadID tid, const StaticInstPtr &inst); 387 388 /** HW return from error interrupt. */ 389 Fault hwrei(ThreadID tid); 390 391 bool simPalCheck(int palFunc, ThreadID tid); 392 393 /** Check if a change in renaming is needed for vector registers. 394 * The vecMode variable is updated and propagated to rename maps. 395 * 396 * @param tid ThreadID 397 * @param freelist list of free registers 398 */ 399 void switchRenameMode(ThreadID tid, UnifiedFreeList* freelist); 400 401 /** Returns the Fault for any valid interrupt. */ 402 Fault getInterrupts(); 403 404 /** Processes any an interrupt fault. */ 405 void processInterrupts(const Fault &interrupt); 406 407 /** Halts the CPU. */ 408 void halt() { panic("Halt not implemented!\n"); } 409 410 /** Register accessors. Index refers to the physical register index. */ 411 412 /** Reads a miscellaneous register. */ 413 RegVal readMiscRegNoEffect(int misc_reg, ThreadID tid) const; 414 415 /** Reads a misc. register, including any side effects the read 416 * might have as defined by the architecture. 417 */ 418 RegVal readMiscReg(int misc_reg, ThreadID tid); 419 420 /** Sets a miscellaneous register. */ 421 void setMiscRegNoEffect(int misc_reg, RegVal val, ThreadID tid); 422 423 /** Sets a misc. register, including any side effects the write 424 * might have as defined by the architecture. 425 */ 426 void setMiscReg(int misc_reg, RegVal val, ThreadID tid); 427 428 RegVal readIntReg(PhysRegIdPtr phys_reg); 429 430 RegVal readFloatReg(PhysRegIdPtr phys_reg); 431 432 const VecRegContainer& readVecReg(PhysRegIdPtr reg_idx) const; 433 434 /** 435 * Read physical vector register for modification. 436 */ 437 VecRegContainer& getWritableVecReg(PhysRegIdPtr reg_idx); 438 439 /** Returns current vector renaming mode */ 440 Enums::VecRegRenameMode vecRenameMode() const { return vecMode; } 441 442 /** Sets the current vector renaming mode */ 443 void vecRenameMode(Enums::VecRegRenameMode vec_mode) 444 { vecMode = vec_mode; } 445 446 /** 447 * Read physical vector register lane 448 */ 449 template<typename VecElem, int LaneIdx> 450 VecLaneT<VecElem, true> 451 readVecLane(PhysRegIdPtr phys_reg) const 452 { 453 vecRegfileReads++; 454 return regFile.readVecLane<VecElem, LaneIdx>(phys_reg); 455 } 456 457 /** 458 * Read physical vector register lane 459 */ 460 template<typename VecElem> 461 VecLaneT<VecElem, true> 462 readVecLane(PhysRegIdPtr phys_reg) const 463 { 464 vecRegfileReads++; 465 return regFile.readVecLane<VecElem>(phys_reg); 466 } 467 468 /** Write a lane of the destination vector register. */ 469 template<typename LD> 470 void 471 setVecLane(PhysRegIdPtr phys_reg, const LD& val) 472 { 473 vecRegfileWrites++; 474 return regFile.setVecLane(phys_reg, val); 475 } 476 477 const VecElem& readVecElem(PhysRegIdPtr reg_idx) const; 478 479 const VecPredRegContainer& readVecPredReg(PhysRegIdPtr reg_idx) const; 480 481 VecPredRegContainer& getWritableVecPredReg(PhysRegIdPtr reg_idx); 482 483 RegVal readCCReg(PhysRegIdPtr phys_reg); 484 485 void setIntReg(PhysRegIdPtr phys_reg, RegVal val); 486 487 void setFloatReg(PhysRegIdPtr phys_reg, RegVal val); 488 489 void setVecReg(PhysRegIdPtr reg_idx, const VecRegContainer& val); 490 491 void setVecElem(PhysRegIdPtr reg_idx, const VecElem& val); 492 493 void setVecPredReg(PhysRegIdPtr reg_idx, const VecPredRegContainer& val); 494 495 void setCCReg(PhysRegIdPtr phys_reg, RegVal val); 496 497 RegVal readArchIntReg(int reg_idx, ThreadID tid); 498 499 RegVal readArchFloatReg(int reg_idx, ThreadID tid); 500 501 const VecRegContainer& readArchVecReg(int reg_idx, ThreadID tid) const; 502 /** Read architectural vector register for modification. */ 503 VecRegContainer& getWritableArchVecReg(int reg_idx, ThreadID tid); 504 505 /** Read architectural vector register lane. */ 506 template<typename VecElem> 507 VecLaneT<VecElem, true> 508 readArchVecLane(int reg_idx, int lId, ThreadID tid) const 509 { 510 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 511 RegId(VecRegClass, reg_idx)); 512 return readVecLane<VecElem>(phys_reg); 513 } 514 515 516 /** Write a lane of the destination vector register. */ 517 template<typename LD> 518 void 519 setArchVecLane(int reg_idx, int lId, ThreadID tid, const LD& val) 520 { 521 PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup( 522 RegId(VecRegClass, reg_idx)); 523 setVecLane(phys_reg, val); 524 } 525 526 const VecElem& readArchVecElem(const RegIndex& reg_idx, 527 const ElemIndex& ldx, ThreadID tid) const; 528 529 const VecPredRegContainer& readArchVecPredReg(int reg_idx, 530 ThreadID tid) const; 531 532 VecPredRegContainer& getWritableArchVecPredReg(int reg_idx, ThreadID tid); 533 534 RegVal readArchCCReg(int reg_idx, ThreadID tid); 535 536 /** Architectural register accessors. Looks up in the commit 537 * rename table to obtain the true physical index of the 538 * architected register first, then accesses that physical 539 * register. 540 */ 541 void setArchIntReg(int reg_idx, RegVal val, ThreadID tid); 542 543 void setArchFloatReg(int reg_idx, RegVal val, ThreadID tid); 544 545 void setArchVecPredReg(int reg_idx, const VecPredRegContainer& val, 546 ThreadID tid); 547 548 void setArchVecReg(int reg_idx, const VecRegContainer& val, ThreadID tid); 549 550 void setArchVecElem(const RegIndex& reg_idx, const ElemIndex& ldx, 551 const VecElem& val, ThreadID tid); 552 553 void setArchCCReg(int reg_idx, RegVal val, ThreadID tid); 554 555 /** Sets the commit PC state of a specific thread. */ 556 void pcState(const TheISA::PCState &newPCState, ThreadID tid); 557 558 /** Reads the commit PC state of a specific thread. */ 559 TheISA::PCState pcState(ThreadID tid); 560 561 /** Reads the commit PC of a specific thread. */ 562 Addr instAddr(ThreadID tid); 563 564 /** Reads the commit micro PC of a specific thread. */ 565 MicroPC microPC(ThreadID tid); 566 567 /** Reads the next PC of a specific thread. */ 568 Addr nextInstAddr(ThreadID tid); 569 570 /** Initiates a squash of all in-flight instructions for a given 571 * thread. The source of the squash is an external update of 572 * state through the TC. 573 */ 574 void squashFromTC(ThreadID tid); 575 576 /** Function to add instruction onto the head of the list of the 577 * instructions. Used when new instructions are fetched. 578 */ 579 ListIt addInst(const DynInstPtr &inst); 580 581 /** Function to tell the CPU that an instruction has completed. */ 582 void instDone(ThreadID tid, const DynInstPtr &inst); 583 584 /** Remove an instruction from the front end of the list. There's 585 * no restriction on location of the instruction. 586 */ 587 void removeFrontInst(const DynInstPtr &inst); 588 589 /** Remove all instructions that are not currently in the ROB. 590 * There's also an option to not squash delay slot instructions.*/ 591 void removeInstsNotInROB(ThreadID tid); 592 593 /** Remove all instructions younger than the given sequence number. */ 594 void removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid); 595 596 /** Removes the instruction pointed to by the iterator. */ 597 inline void squashInstIt(const ListIt &instIt, ThreadID tid); 598 599 /** Cleans up all instructions on the remove list. */ 600 void cleanUpRemovedInsts(); 601 602 /** Debug function to print all instructions on the list. */ 603 void dumpInsts(); 604 605 public: 606#ifndef NDEBUG 607 /** Count of total number of dynamic instructions in flight. */ 608 int instcount; 609#endif 610 611 /** List of all the instructions in flight. */ 612 std::list<DynInstPtr> instList; 613 614 /** List of all the instructions that will be removed at the end of this 615 * cycle. 616 */ 617 std::queue<ListIt> removeList; 618 619#ifdef DEBUG 620 /** Debug structure to keep track of the sequence numbers still in 621 * flight. 622 */ 623 std::set<InstSeqNum> snList; 624#endif 625 626 /** Records if instructions need to be removed this cycle due to 627 * being retired or squashed. 628 */ 629 bool removeInstsThisCycle; 630 631 protected: 632 /** The fetch stage. */ 633 typename CPUPolicy::Fetch fetch; 634 635 /** The decode stage. */ 636 typename CPUPolicy::Decode decode; 637 638 /** The dispatch stage. */ 639 typename CPUPolicy::Rename rename; 640 641 /** The issue/execute/writeback stages. */ 642 typename CPUPolicy::IEW iew; 643 644 /** The commit stage. */ 645 typename CPUPolicy::Commit commit; 646 647 /** The rename mode of the vector registers */ 648 Enums::VecRegRenameMode vecMode; 649 650 /** The register file. */ 651 PhysRegFile regFile; 652 653 /** The free list. */ 654 typename CPUPolicy::FreeList freeList; 655 656 /** The rename map. */ 657 typename CPUPolicy::RenameMap renameMap[Impl::MaxThreads]; 658 659 /** The commit rename map. */ 660 typename CPUPolicy::RenameMap commitRenameMap[Impl::MaxThreads]; 661 662 /** The re-order buffer. */ 663 typename CPUPolicy::ROB rob; 664 665 /** Active Threads List */ 666 std::list<ThreadID> activeThreads; 667 668 /** 669 * This is a list of threads that are trying to exit. Each thread id 670 * is mapped to a boolean value denoting whether the thread is ready 671 * to exit. 672 */ 673 std::unordered_map<ThreadID, bool> exitingThreads; 674 675 /** Integer Register Scoreboard */ 676 Scoreboard scoreboard; 677 678 std::vector<TheISA::ISA *> isa; 679 680 /** Instruction port. Note that it has to appear after the fetch stage. */ 681 IcachePort icachePort; 682 683 /** Data port. Note that it has to appear after the iew stages */ 684 DcachePort dcachePort; 685 686 public: 687 /** Enum to give each stage a specific index, so when calling 688 * activateStage() or deactivateStage(), they can specify which stage 689 * is being activated/deactivated. 690 */ 691 enum StageIdx { 692 FetchIdx, 693 DecodeIdx, 694 RenameIdx, 695 IEWIdx, 696 CommitIdx, 697 NumStages }; 698 699 /** Typedefs from the Impl to get the structs that each of the 700 * time buffers should use. 701 */ 702 typedef typename CPUPolicy::TimeStruct TimeStruct; 703 704 typedef typename CPUPolicy::FetchStruct FetchStruct; 705 706 typedef typename CPUPolicy::DecodeStruct DecodeStruct; 707 708 typedef typename CPUPolicy::RenameStruct RenameStruct; 709 710 typedef typename CPUPolicy::IEWStruct IEWStruct; 711 712 /** The main time buffer to do backwards communication. */ 713 TimeBuffer<TimeStruct> timeBuffer; 714 715 /** The fetch stage's instruction queue. */ 716 TimeBuffer<FetchStruct> fetchQueue; 717 718 /** The decode stage's instruction queue. */ 719 TimeBuffer<DecodeStruct> decodeQueue; 720 721 /** The rename stage's instruction queue. */ 722 TimeBuffer<RenameStruct> renameQueue; 723 724 /** The IEW stage's instruction queue. */ 725 TimeBuffer<IEWStruct> iewQueue; 726 727 private: 728 /** The activity recorder; used to tell if the CPU has any 729 * activity remaining or if it can go to idle and deschedule 730 * itself. 731 */ 732 ActivityRecorder activityRec; 733 734 public: 735 /** Records that there was time buffer activity this cycle. */ 736 void activityThisCycle() { activityRec.activity(); } 737 738 /** Changes a stage's status to active within the activity recorder. */ 739 void activateStage(const StageIdx idx) 740 { activityRec.activateStage(idx); } 741 742 /** Changes a stage's status to inactive within the activity recorder. */ 743 void deactivateStage(const StageIdx idx) 744 { activityRec.deactivateStage(idx); } 745 746 /** Wakes the CPU, rescheduling the CPU if it's not already active. */ 747 void wakeCPU(); 748 749 virtual void wakeup(ThreadID tid) override; 750 751 /** Gets a free thread id. Use if thread ids change across system. */ 752 ThreadID getFreeTid(); 753 754 public: 755 /** Returns a pointer to a thread context. */ 756 ThreadContext * 757 tcBase(ThreadID tid) 758 { 759 return thread[tid]->getTC(); 760 } 761 762 /** The global sequence number counter. */ 763 InstSeqNum globalSeqNum;//[Impl::MaxThreads]; 764 765 /** Pointer to the checker, which can dynamically verify 766 * instruction results at run time. This can be set to NULL if it 767 * is not being used. 768 */ 769 Checker<Impl> *checker; 770 771 /** Pointer to the system. */ 772 System *system; 773 774 /** Pointers to all of the threads in the CPU. */ 775 std::vector<Thread *> thread; 776 777 /** Threads Scheduled to Enter CPU */ 778 std::list<int> cpuWaitList; 779 780 /** The cycle that the CPU was last running, used for statistics. */ 781 Cycles lastRunningCycle; 782 783 /** The cycle that the CPU was last activated by a new thread*/ 784 Tick lastActivatedCycle; 785 786 /** Mapping for system thread id to cpu id */ 787 std::map<ThreadID, unsigned> threadMap; 788 789 /** Available thread ids in the cpu*/ 790 std::vector<ThreadID> tids; 791 792 /** CPU pushRequest function, forwards request to LSQ. */ 793 Fault pushRequest(const DynInstPtr& inst, bool isLoad, uint8_t *data, 794 unsigned int size, Addr addr, Request::Flags flags, 795 uint64_t *res, AtomicOpFunctor *amo_op = nullptr) 796 { 797 return iew.ldstQueue.pushRequest(inst, isLoad, data, size, addr, 798 flags, res, amo_op); 799 } 800 801 /** CPU read function, forwards read to LSQ. */ 802 Fault read(LSQRequest* req, int load_idx) 803 { 804 return this->iew.ldstQueue.read(req, load_idx); 805 } 806 807 /** CPU write function, forwards write to LSQ. */ 808 Fault write(LSQRequest* req, uint8_t *data, int store_idx) 809 { 810 return this->iew.ldstQueue.write(req, data, store_idx); 811 } 812 813 /** Used by the fetch unit to get a hold of the instruction port. */ 814 MasterPort &getInstPort() override { return icachePort; } 815 816 /** Get the dcache port (used to find block size for translations). */ 817 MasterPort &getDataPort() override { return dcachePort; } 818 819 /** Stat for total number of times the CPU is descheduled. */ 820 Stats::Scalar timesIdled; 821 /** Stat for total number of cycles the CPU spends descheduled. */ 822 Stats::Scalar idleCycles; 823 /** Stat for total number of cycles the CPU spends descheduled due to a 824 * quiesce operation or waiting for an interrupt. */ 825 Stats::Scalar quiesceCycles; 826 /** Stat for the number of committed instructions per thread. */ 827 Stats::Vector committedInsts; 828 /** Stat for the number of committed ops (including micro ops) per thread. */ 829 Stats::Vector committedOps; 830 /** Stat for the CPI per thread. */ 831 Stats::Formula cpi; 832 /** Stat for the total CPI. */ 833 Stats::Formula totalCpi; 834 /** Stat for the IPC per thread. */ 835 Stats::Formula ipc; 836 /** Stat for the total IPC. */ 837 Stats::Formula totalIpc; 838 839 //number of integer register file accesses 840 Stats::Scalar intRegfileReads; 841 Stats::Scalar intRegfileWrites; 842 //number of float register file accesses 843 Stats::Scalar fpRegfileReads; 844 Stats::Scalar fpRegfileWrites; 845 //number of vector register file accesses 846 mutable Stats::Scalar vecRegfileReads; 847 Stats::Scalar vecRegfileWrites; 848 //number of predicate register file accesses 849 mutable Stats::Scalar vecPredRegfileReads; 850 Stats::Scalar vecPredRegfileWrites; 851 //number of CC register file accesses 852 Stats::Scalar ccRegfileReads; 853 Stats::Scalar ccRegfileWrites; 854 //number of misc 855 Stats::Scalar miscRegfileReads; 856 Stats::Scalar miscRegfileWrites; 857}; 858 859#endif // __CPU_O3_CPU_HH__ |