1/*
2 * Copyright (c) 2010-2012 ARM Limited
3 * All rights reserved
4 *
5 * The license below extends only to copyright in the software and shall
6 * not be construed as granting a license to any other intellectual
7 * property including but not limited to intellectual property relating
8 * to a hardware implementation of the functionality of the software
9 * licensed hereunder. You may use the software subject to the license
10 * terms below provided that you ensure that this notice is replicated
11 * unmodified and in its entirety in all distributions of the software,
12 * modified or unmodified, in source code or in binary form.
13 *
14 * Copyright (c) 2004-2006 The Regents of The University of Michigan
15 * All rights reserved.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions are
19 * met: redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer;
21 * redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution;
24 * neither the name of the copyright holders nor the names of its
25 * contributors may be used to endorse or promote products derived from
26 * this software without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 *
40 * Authors: Kevin Lim
41 * Korey Sewell
42 */
43
44#ifndef __CPU_O3_FETCH_HH__
45#define __CPU_O3_FETCH_HH__
46
47#include "arch/decoder.hh"
48#include "arch/utility.hh"
49#include "base/statistics.hh"
50#include "config/the_isa.hh"
51#include "cpu/pc_event.hh"
52#include "cpu/pred/bpred_unit.hh"
53#include "cpu/timebuf.hh"
54#include "cpu/translation.hh"
55#include "mem/packet.hh"
56#include "mem/port.hh"
57#include "sim/eventq.hh"
| 1/*
2 * Copyright (c) 2010-2012 ARM Limited
3 * All rights reserved
4 *
5 * The license below extends only to copyright in the software and shall
6 * not be construed as granting a license to any other intellectual
7 * property including but not limited to intellectual property relating
8 * to a hardware implementation of the functionality of the software
9 * licensed hereunder. You may use the software subject to the license
10 * terms below provided that you ensure that this notice is replicated
11 * unmodified and in its entirety in all distributions of the software,
12 * modified or unmodified, in source code or in binary form.
13 *
14 * Copyright (c) 2004-2006 The Regents of The University of Michigan
15 * All rights reserved.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions are
19 * met: redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer;
21 * redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution;
24 * neither the name of the copyright holders nor the names of its
25 * contributors may be used to endorse or promote products derived from
26 * this software without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 *
40 * Authors: Kevin Lim
41 * Korey Sewell
42 */
43
44#ifndef __CPU_O3_FETCH_HH__
45#define __CPU_O3_FETCH_HH__
46
47#include "arch/decoder.hh"
48#include "arch/utility.hh"
49#include "base/statistics.hh"
50#include "config/the_isa.hh"
51#include "cpu/pc_event.hh"
52#include "cpu/pred/bpred_unit.hh"
53#include "cpu/timebuf.hh"
54#include "cpu/translation.hh"
55#include "mem/packet.hh"
56#include "mem/port.hh"
57#include "sim/eventq.hh"
|
| 58#include "sim/probe/probe.hh"
|
58
59struct DerivO3CPUParams;
60
61/**
62 * DefaultFetch class handles both single threaded and SMT fetch. Its
63 * width is specified by the parameters; each cycle it tries to fetch
64 * that many instructions. It supports using a branch predictor to
65 * predict direction and targets.
66 * It supports the idling functionality of the CPU by indicating to
67 * the CPU when it is active and inactive.
68 */
69template <class Impl>
70class DefaultFetch
71{
72 public:
73 /** Typedefs from Impl. */
74 typedef typename Impl::CPUPol CPUPol;
75 typedef typename Impl::DynInst DynInst;
76 typedef typename Impl::DynInstPtr DynInstPtr;
77 typedef typename Impl::O3CPU O3CPU;
78
79 /** Typedefs from the CPU policy. */
80 typedef typename CPUPol::FetchStruct FetchStruct;
81 typedef typename CPUPol::TimeStruct TimeStruct;
82
83 /** Typedefs from ISA. */
84 typedef TheISA::MachInst MachInst;
85 typedef TheISA::ExtMachInst ExtMachInst;
86
87 class FetchTranslation : public BaseTLB::Translation
88 {
89 protected:
90 DefaultFetch<Impl> *fetch;
91
92 public:
93 FetchTranslation(DefaultFetch<Impl> *_fetch)
94 : fetch(_fetch)
95 {}
96
97 void
98 markDelayed()
99 {}
100
101 void
102 finish(Fault fault, RequestPtr req, ThreadContext *tc,
103 BaseTLB::Mode mode)
104 {
105 assert(mode == BaseTLB::Execute);
106 fetch->finishTranslation(fault, req);
107 delete this;
108 }
109 };
110
111 private:
112 /* Event to delay delivery of a fetch translation result in case of
113 * a fault and the nop to carry the fault cannot be generated
114 * immediately */
115 class FinishTranslationEvent : public Event
116 {
117 private:
118 DefaultFetch<Impl> *fetch;
119 Fault fault;
120 RequestPtr req;
121
122 public:
123 FinishTranslationEvent(DefaultFetch<Impl> *_fetch)
124 : fetch(_fetch)
125 {}
126
127 void setFault(Fault _fault)
128 {
129 fault = _fault;
130 }
131
132 void setReq(RequestPtr _req)
133 {
134 req = _req;
135 }
136
137 /** Process the delayed finish translation */
138 void process()
139 {
140 assert(fetch->numInst < fetch->fetchWidth);
141 fetch->finishTranslation(fault, req);
142 }
143
144 const char *description() const
145 {
146 return "FullO3CPU FetchFinishTranslation";
147 }
148 };
149
150 public:
151 /** Overall fetch status. Used to determine if the CPU can
152 * deschedule itsef due to a lack of activity.
153 */
154 enum FetchStatus {
155 Active,
156 Inactive
157 };
158
159 /** Individual thread status. */
160 enum ThreadStatus {
161 Running,
162 Idle,
163 Squashing,
164 Blocked,
165 Fetching,
166 TrapPending,
167 QuiescePending,
168 ItlbWait,
169 IcacheWaitResponse,
170 IcacheWaitRetry,
171 IcacheAccessComplete,
172 NoGoodAddr
173 };
174
175 /** Fetching Policy, Add new policies here.*/
176 enum FetchPriority {
177 SingleThread,
178 RoundRobin,
179 Branch,
180 IQ,
181 LSQ
182 };
183
184 private:
185 /** Fetch status. */
186 FetchStatus _status;
187
188 /** Per-thread status. */
189 ThreadStatus fetchStatus[Impl::MaxThreads];
190
191 /** Fetch policy. */
192 FetchPriority fetchPolicy;
193
194 /** List that has the threads organized by priority. */
195 std::list<ThreadID> priorityList;
196
| 59
60struct DerivO3CPUParams;
61
62/**
63 * DefaultFetch class handles both single threaded and SMT fetch. Its
64 * width is specified by the parameters; each cycle it tries to fetch
65 * that many instructions. It supports using a branch predictor to
66 * predict direction and targets.
67 * It supports the idling functionality of the CPU by indicating to
68 * the CPU when it is active and inactive.
69 */
70template <class Impl>
71class DefaultFetch
72{
73 public:
74 /** Typedefs from Impl. */
75 typedef typename Impl::CPUPol CPUPol;
76 typedef typename Impl::DynInst DynInst;
77 typedef typename Impl::DynInstPtr DynInstPtr;
78 typedef typename Impl::O3CPU O3CPU;
79
80 /** Typedefs from the CPU policy. */
81 typedef typename CPUPol::FetchStruct FetchStruct;
82 typedef typename CPUPol::TimeStruct TimeStruct;
83
84 /** Typedefs from ISA. */
85 typedef TheISA::MachInst MachInst;
86 typedef TheISA::ExtMachInst ExtMachInst;
87
88 class FetchTranslation : public BaseTLB::Translation
89 {
90 protected:
91 DefaultFetch<Impl> *fetch;
92
93 public:
94 FetchTranslation(DefaultFetch<Impl> *_fetch)
95 : fetch(_fetch)
96 {}
97
98 void
99 markDelayed()
100 {}
101
102 void
103 finish(Fault fault, RequestPtr req, ThreadContext *tc,
104 BaseTLB::Mode mode)
105 {
106 assert(mode == BaseTLB::Execute);
107 fetch->finishTranslation(fault, req);
108 delete this;
109 }
110 };
111
112 private:
113 /* Event to delay delivery of a fetch translation result in case of
114 * a fault and the nop to carry the fault cannot be generated
115 * immediately */
116 class FinishTranslationEvent : public Event
117 {
118 private:
119 DefaultFetch<Impl> *fetch;
120 Fault fault;
121 RequestPtr req;
122
123 public:
124 FinishTranslationEvent(DefaultFetch<Impl> *_fetch)
125 : fetch(_fetch)
126 {}
127
128 void setFault(Fault _fault)
129 {
130 fault = _fault;
131 }
132
133 void setReq(RequestPtr _req)
134 {
135 req = _req;
136 }
137
138 /** Process the delayed finish translation */
139 void process()
140 {
141 assert(fetch->numInst < fetch->fetchWidth);
142 fetch->finishTranslation(fault, req);
143 }
144
145 const char *description() const
146 {
147 return "FullO3CPU FetchFinishTranslation";
148 }
149 };
150
151 public:
152 /** Overall fetch status. Used to determine if the CPU can
153 * deschedule itsef due to a lack of activity.
154 */
155 enum FetchStatus {
156 Active,
157 Inactive
158 };
159
160 /** Individual thread status. */
161 enum ThreadStatus {
162 Running,
163 Idle,
164 Squashing,
165 Blocked,
166 Fetching,
167 TrapPending,
168 QuiescePending,
169 ItlbWait,
170 IcacheWaitResponse,
171 IcacheWaitRetry,
172 IcacheAccessComplete,
173 NoGoodAddr
174 };
175
176 /** Fetching Policy, Add new policies here.*/
177 enum FetchPriority {
178 SingleThread,
179 RoundRobin,
180 Branch,
181 IQ,
182 LSQ
183 };
184
185 private:
186 /** Fetch status. */
187 FetchStatus _status;
188
189 /** Per-thread status. */
190 ThreadStatus fetchStatus[Impl::MaxThreads];
191
192 /** Fetch policy. */
193 FetchPriority fetchPolicy;
194
195 /** List that has the threads organized by priority. */
196 std::list<ThreadID> priorityList;
197
|
| 198 /** Probe points. */
199 ProbePointArg<DynInstPtr> *ppFetch;
200
|
197 public:
198 /** DefaultFetch constructor. */
199 DefaultFetch(O3CPU *_cpu, DerivO3CPUParams *params);
200
201 /** Returns the name of fetch. */
202 std::string name() const;
203
204 /** Registers statistics. */
205 void regStats();
206
| 201 public:
202 /** DefaultFetch constructor. */
203 DefaultFetch(O3CPU *_cpu, DerivO3CPUParams *params);
204
205 /** Returns the name of fetch. */
206 std::string name() const;
207
208 /** Registers statistics. */
209 void regStats();
210
|
| 211 /** Registers probes. */
212 void regProbePoints();
213
|
207 /** Sets the main backwards communication time buffer pointer. */
208 void setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer);
209
210 /** Sets pointer to list of active threads. */
211 void setActiveThreads(std::list<ThreadID> *at_ptr);
212
213 /** Sets pointer to time buffer used to communicate to the next stage. */
214 void setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr);
215
216 /** Initialize stage. */
217 void startupStage();
218
219 /** Handles retrying the fetch access. */
220 void recvRetry();
221
222 /** Processes cache completion event. */
223 void processCacheCompletion(PacketPtr pkt);
224
225 /** Resume after a drain. */
226 void drainResume();
227
228 /** Perform sanity checks after a drain. */
229 void drainSanityCheck() const;
230
231 /** Has the stage drained? */
232 bool isDrained() const;
233
234 /** Takes over from another CPU's thread. */
235 void takeOverFrom();
236
237 /**
238 * Stall the fetch stage after reaching a safe drain point.
239 *
240 * The CPU uses this method to stop fetching instructions from a
241 * thread that has been drained. The drain stall is different from
242 * all other stalls in that it is signaled instantly from the
243 * commit stage (without the normal communication delay) when it
244 * has reached a safe point to drain from.
245 */
246 void drainStall(ThreadID tid);
247
248 /** Tells fetch to wake up from a quiesce instruction. */
249 void wakeFromQuiesce();
250
251 private:
252 /** Reset this pipeline stage */
253 void resetStage();
254
255 /** Changes the status of this stage to active, and indicates this
256 * to the CPU.
257 */
258 inline void switchToActive();
259
260 /** Changes the status of this stage to inactive, and indicates
261 * this to the CPU.
262 */
263 inline void switchToInactive();
264
265 /**
266 * Looks up in the branch predictor to see if the next PC should be
267 * either next PC+=MachInst or a branch target.
268 * @param next_PC Next PC variable passed in by reference. It is
269 * expected to be set to the current PC; it will be updated with what
270 * the next PC will be.
271 * @param next_NPC Used for ISAs which use delay slots.
272 * @return Whether or not a branch was predicted as taken.
273 */
274 bool lookupAndUpdateNextPC(DynInstPtr &inst, TheISA::PCState &pc);
275
276 /**
277 * Fetches the cache line that contains the fetch PC. Returns any
278 * fault that happened. Puts the data into the class variable
279 * fetchBuffer, which may not hold the entire fetched cache line.
280 * @param vaddr The memory address that is being fetched from.
281 * @param ret_fault The fault reference that will be set to the result of
282 * the icache access.
283 * @param tid Thread id.
284 * @param pc The actual PC of the current instruction.
285 * @return Any fault that occured.
286 */
287 bool fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc);
288 void finishTranslation(Fault fault, RequestPtr mem_req);
289
290
291 /** Check if an interrupt is pending and that we need to handle
292 */
293 bool
294 checkInterrupt(Addr pc)
295 {
296 return (interruptPending && (THE_ISA != ALPHA_ISA || !(pc & 0x3)));
297 }
298
299 /** Squashes a specific thread and resets the PC. */
300 inline void doSquash(const TheISA::PCState &newPC,
301 const DynInstPtr squashInst, ThreadID tid);
302
303 /** Squashes a specific thread and resets the PC. Also tells the CPU to
304 * remove any instructions between fetch and decode that should be sqaushed.
305 */
306 void squashFromDecode(const TheISA::PCState &newPC,
307 const DynInstPtr squashInst,
308 const InstSeqNum seq_num, ThreadID tid);
309
310 /** Checks if a thread is stalled. */
311 bool checkStall(ThreadID tid) const;
312
313 /** Updates overall fetch stage status; to be called at the end of each
314 * cycle. */
315 FetchStatus updateFetchStatus();
316
317 public:
318 /** Squashes a specific thread and resets the PC. Also tells the CPU to
319 * remove any instructions that are not in the ROB. The source of this
320 * squash should be the commit stage.
321 */
322 void squash(const TheISA::PCState &newPC, const InstSeqNum seq_num,
323 DynInstPtr squashInst, ThreadID tid);
324
325 /** Ticks the fetch stage, processing all inputs signals and fetching
326 * as many instructions as possible.
327 */
328 void tick();
329
330 /** Checks all input signals and updates the status as necessary.
331 * @return: Returns if the status has changed due to input signals.
332 */
333 bool checkSignalsAndUpdate(ThreadID tid);
334
335 /** Does the actual fetching of instructions and passing them on to the
336 * next stage.
337 * @param status_change fetch() sets this variable if there was a status
338 * change (ie switching to IcacheMissStall).
339 */
340 void fetch(bool &status_change);
341
342 /** Align a PC to the start of a fetch buffer block. */
343 Addr fetchBufferAlignPC(Addr addr)
344 {
345 return (addr & ~(fetchBufferMask));
346 }
347
348 /** The decoder. */
349 TheISA::Decoder *decoder[Impl::MaxThreads];
350
351 private:
352 DynInstPtr buildInst(ThreadID tid, StaticInstPtr staticInst,
353 StaticInstPtr curMacroop, TheISA::PCState thisPC,
354 TheISA::PCState nextPC, bool trace);
355
356 /** Returns the appropriate thread to fetch, given the fetch policy. */
357 ThreadID getFetchingThread(FetchPriority &fetch_priority);
358
359 /** Returns the appropriate thread to fetch using a round robin policy. */
360 ThreadID roundRobin();
361
362 /** Returns the appropriate thread to fetch using the IQ count policy. */
363 ThreadID iqCount();
364
365 /** Returns the appropriate thread to fetch using the LSQ count policy. */
366 ThreadID lsqCount();
367
368 /** Returns the appropriate thread to fetch using the branch count
369 * policy. */
370 ThreadID branchCount();
371
372 /** Pipeline the next I-cache access to the current one. */
373 void pipelineIcacheAccesses(ThreadID tid);
374
375 /** Profile the reasons of fetch stall. */
376 void profileStall(ThreadID tid);
377
378 private:
379 /** Pointer to the O3CPU. */
380 O3CPU *cpu;
381
382 /** Time buffer interface. */
383 TimeBuffer<TimeStruct> *timeBuffer;
384
385 /** Wire to get decode's information from backwards time buffer. */
386 typename TimeBuffer<TimeStruct>::wire fromDecode;
387
388 /** Wire to get rename's information from backwards time buffer. */
389 typename TimeBuffer<TimeStruct>::wire fromRename;
390
391 /** Wire to get iew's information from backwards time buffer. */
392 typename TimeBuffer<TimeStruct>::wire fromIEW;
393
394 /** Wire to get commit's information from backwards time buffer. */
395 typename TimeBuffer<TimeStruct>::wire fromCommit;
396
397 /** Internal fetch instruction queue. */
398 TimeBuffer<FetchStruct> *fetchQueue;
399
400 //Might be annoying how this name is different than the queue.
401 /** Wire used to write any information heading to decode. */
402 typename TimeBuffer<FetchStruct>::wire toDecode;
403
404 /** BPredUnit. */
405 BPredUnit *branchPred;
406
407 TheISA::PCState pc[Impl::MaxThreads];
408
409 Addr fetchOffset[Impl::MaxThreads];
410
411 StaticInstPtr macroop[Impl::MaxThreads];
412
413 /** Can the fetch stage redirect from an interrupt on this instruction? */
414 bool delayedCommit[Impl::MaxThreads];
415
416 /** Memory request used to access cache. */
417 RequestPtr memReq[Impl::MaxThreads];
418
419 /** Variable that tracks if fetch has written to the time buffer this
420 * cycle. Used to tell CPU if there is activity this cycle.
421 */
422 bool wroteToTimeBuffer;
423
424 /** Tracks how many instructions has been fetched this cycle. */
425 int numInst;
426
427 /** Source of possible stalls. */
428 struct Stalls {
429 bool decode;
430 bool rename;
431 bool iew;
432 bool commit;
433 bool drain;
434 };
435
436 /** Tracks which stages are telling fetch to stall. */
437 Stalls stalls[Impl::MaxThreads];
438
439 /** Decode to fetch delay. */
440 Cycles decodeToFetchDelay;
441
442 /** Rename to fetch delay. */
443 Cycles renameToFetchDelay;
444
445 /** IEW to fetch delay. */
446 Cycles iewToFetchDelay;
447
448 /** Commit to fetch delay. */
449 Cycles commitToFetchDelay;
450
451 /** The width of fetch in instructions. */
452 unsigned fetchWidth;
453
454 /** Is the cache blocked? If so no threads can access it. */
455 bool cacheBlocked;
456
457 /** The packet that is waiting to be retried. */
458 PacketPtr retryPkt;
459
460 /** The thread that is waiting on the cache to tell fetch to retry. */
461 ThreadID retryTid;
462
463 /** Cache block size. */
464 unsigned int cacheBlkSize;
465
466 /** The size of the fetch buffer in bytes. The fetch buffer
467 * itself may be smaller than a cache line.
468 */
469 unsigned fetchBufferSize;
470
471 /** Mask to align a fetch address to a fetch buffer boundary. */
472 Addr fetchBufferMask;
473
474 /** The fetch data that is being fetched and buffered. */
475 uint8_t *fetchBuffer[Impl::MaxThreads];
476
477 /** The PC of the first instruction loaded into the fetch buffer. */
478 Addr fetchBufferPC[Impl::MaxThreads];
479
480 /** Whether or not the fetch buffer data is valid. */
481 bool fetchBufferValid[Impl::MaxThreads];
482
483 /** Size of instructions. */
484 int instSize;
485
486 /** Icache stall statistics. */
487 Counter lastIcacheStall[Impl::MaxThreads];
488
489 /** List of Active Threads */
490 std::list<ThreadID> *activeThreads;
491
492 /** Number of threads. */
493 ThreadID numThreads;
494
495 /** Number of threads that are actively fetching. */
496 ThreadID numFetchingThreads;
497
498 /** Thread ID being fetched. */
499 ThreadID threadFetched;
500
501 /** Checks if there is an interrupt pending. If there is, fetch
502 * must stop once it is not fetching PAL instructions.
503 */
504 bool interruptPending;
505
506 /** Set to true if a pipelined I-cache request should be issued. */
507 bool issuePipelinedIfetch[Impl::MaxThreads];
508
509 /** Event used to delay fault generation of translation faults */
510 FinishTranslationEvent finishTranslationEvent;
511
512 // @todo: Consider making these vectors and tracking on a per thread basis.
513 /** Stat for total number of cycles stalled due to an icache miss. */
514 Stats::Scalar icacheStallCycles;
515 /** Stat for total number of fetched instructions. */
516 Stats::Scalar fetchedInsts;
517 /** Total number of fetched branches. */
518 Stats::Scalar fetchedBranches;
519 /** Stat for total number of predicted branches. */
520 Stats::Scalar predictedBranches;
521 /** Stat for total number of cycles spent fetching. */
522 Stats::Scalar fetchCycles;
523 /** Stat for total number of cycles spent squashing. */
524 Stats::Scalar fetchSquashCycles;
525 /** Stat for total number of cycles spent waiting for translation */
526 Stats::Scalar fetchTlbCycles;
527 /** Stat for total number of cycles spent blocked due to other stages in
528 * the pipeline.
529 */
530 Stats::Scalar fetchIdleCycles;
531 /** Total number of cycles spent blocked. */
532 Stats::Scalar fetchBlockedCycles;
533 /** Total number of cycles spent in any other state. */
534 Stats::Scalar fetchMiscStallCycles;
535 /** Total number of cycles spent in waiting for drains. */
536 Stats::Scalar fetchPendingDrainCycles;
537 /** Total number of stall cycles caused by no active threads to run. */
538 Stats::Scalar fetchNoActiveThreadStallCycles;
539 /** Total number of stall cycles caused by pending traps. */
540 Stats::Scalar fetchPendingTrapStallCycles;
541 /** Total number of stall cycles caused by pending quiesce instructions. */
542 Stats::Scalar fetchPendingQuiesceStallCycles;
543 /** Total number of stall cycles caused by I-cache wait retrys. */
544 Stats::Scalar fetchIcacheWaitRetryStallCycles;
545 /** Stat for total number of fetched cache lines. */
546 Stats::Scalar fetchedCacheLines;
547 /** Total number of outstanding icache accesses that were dropped
548 * due to a squash.
549 */
550 Stats::Scalar fetchIcacheSquashes;
551 /** Total number of outstanding tlb accesses that were dropped
552 * due to a squash.
553 */
554 Stats::Scalar fetchTlbSquashes;
555 /** Distribution of number of instructions fetched each cycle. */
556 Stats::Distribution fetchNisnDist;
557 /** Rate of how often fetch was idle. */
558 Stats::Formula idleRate;
559 /** Number of branch fetches per cycle. */
560 Stats::Formula branchRate;
561 /** Number of instruction fetched per cycle. */
562 Stats::Formula fetchRate;
563};
564
565#endif //__CPU_O3_FETCH_HH__
| 214 /** Sets the main backwards communication time buffer pointer. */
215 void setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer);
216
217 /** Sets pointer to list of active threads. */
218 void setActiveThreads(std::list<ThreadID> *at_ptr);
219
220 /** Sets pointer to time buffer used to communicate to the next stage. */
221 void setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr);
222
223 /** Initialize stage. */
224 void startupStage();
225
226 /** Handles retrying the fetch access. */
227 void recvRetry();
228
229 /** Processes cache completion event. */
230 void processCacheCompletion(PacketPtr pkt);
231
232 /** Resume after a drain. */
233 void drainResume();
234
235 /** Perform sanity checks after a drain. */
236 void drainSanityCheck() const;
237
238 /** Has the stage drained? */
239 bool isDrained() const;
240
241 /** Takes over from another CPU's thread. */
242 void takeOverFrom();
243
244 /**
245 * Stall the fetch stage after reaching a safe drain point.
246 *
247 * The CPU uses this method to stop fetching instructions from a
248 * thread that has been drained. The drain stall is different from
249 * all other stalls in that it is signaled instantly from the
250 * commit stage (without the normal communication delay) when it
251 * has reached a safe point to drain from.
252 */
253 void drainStall(ThreadID tid);
254
255 /** Tells fetch to wake up from a quiesce instruction. */
256 void wakeFromQuiesce();
257
258 private:
259 /** Reset this pipeline stage */
260 void resetStage();
261
262 /** Changes the status of this stage to active, and indicates this
263 * to the CPU.
264 */
265 inline void switchToActive();
266
267 /** Changes the status of this stage to inactive, and indicates
268 * this to the CPU.
269 */
270 inline void switchToInactive();
271
272 /**
273 * Looks up in the branch predictor to see if the next PC should be
274 * either next PC+=MachInst or a branch target.
275 * @param next_PC Next PC variable passed in by reference. It is
276 * expected to be set to the current PC; it will be updated with what
277 * the next PC will be.
278 * @param next_NPC Used for ISAs which use delay slots.
279 * @return Whether or not a branch was predicted as taken.
280 */
281 bool lookupAndUpdateNextPC(DynInstPtr &inst, TheISA::PCState &pc);
282
283 /**
284 * Fetches the cache line that contains the fetch PC. Returns any
285 * fault that happened. Puts the data into the class variable
286 * fetchBuffer, which may not hold the entire fetched cache line.
287 * @param vaddr The memory address that is being fetched from.
288 * @param ret_fault The fault reference that will be set to the result of
289 * the icache access.
290 * @param tid Thread id.
291 * @param pc The actual PC of the current instruction.
292 * @return Any fault that occured.
293 */
294 bool fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc);
295 void finishTranslation(Fault fault, RequestPtr mem_req);
296
297
298 /** Check if an interrupt is pending and that we need to handle
299 */
300 bool
301 checkInterrupt(Addr pc)
302 {
303 return (interruptPending && (THE_ISA != ALPHA_ISA || !(pc & 0x3)));
304 }
305
306 /** Squashes a specific thread and resets the PC. */
307 inline void doSquash(const TheISA::PCState &newPC,
308 const DynInstPtr squashInst, ThreadID tid);
309
310 /** Squashes a specific thread and resets the PC. Also tells the CPU to
311 * remove any instructions between fetch and decode that should be sqaushed.
312 */
313 void squashFromDecode(const TheISA::PCState &newPC,
314 const DynInstPtr squashInst,
315 const InstSeqNum seq_num, ThreadID tid);
316
317 /** Checks if a thread is stalled. */
318 bool checkStall(ThreadID tid) const;
319
320 /** Updates overall fetch stage status; to be called at the end of each
321 * cycle. */
322 FetchStatus updateFetchStatus();
323
324 public:
325 /** Squashes a specific thread and resets the PC. Also tells the CPU to
326 * remove any instructions that are not in the ROB. The source of this
327 * squash should be the commit stage.
328 */
329 void squash(const TheISA::PCState &newPC, const InstSeqNum seq_num,
330 DynInstPtr squashInst, ThreadID tid);
331
332 /** Ticks the fetch stage, processing all inputs signals and fetching
333 * as many instructions as possible.
334 */
335 void tick();
336
337 /** Checks all input signals and updates the status as necessary.
338 * @return: Returns if the status has changed due to input signals.
339 */
340 bool checkSignalsAndUpdate(ThreadID tid);
341
342 /** Does the actual fetching of instructions and passing them on to the
343 * next stage.
344 * @param status_change fetch() sets this variable if there was a status
345 * change (ie switching to IcacheMissStall).
346 */
347 void fetch(bool &status_change);
348
349 /** Align a PC to the start of a fetch buffer block. */
350 Addr fetchBufferAlignPC(Addr addr)
351 {
352 return (addr & ~(fetchBufferMask));
353 }
354
355 /** The decoder. */
356 TheISA::Decoder *decoder[Impl::MaxThreads];
357
358 private:
359 DynInstPtr buildInst(ThreadID tid, StaticInstPtr staticInst,
360 StaticInstPtr curMacroop, TheISA::PCState thisPC,
361 TheISA::PCState nextPC, bool trace);
362
363 /** Returns the appropriate thread to fetch, given the fetch policy. */
364 ThreadID getFetchingThread(FetchPriority &fetch_priority);
365
366 /** Returns the appropriate thread to fetch using a round robin policy. */
367 ThreadID roundRobin();
368
369 /** Returns the appropriate thread to fetch using the IQ count policy. */
370 ThreadID iqCount();
371
372 /** Returns the appropriate thread to fetch using the LSQ count policy. */
373 ThreadID lsqCount();
374
375 /** Returns the appropriate thread to fetch using the branch count
376 * policy. */
377 ThreadID branchCount();
378
379 /** Pipeline the next I-cache access to the current one. */
380 void pipelineIcacheAccesses(ThreadID tid);
381
382 /** Profile the reasons of fetch stall. */
383 void profileStall(ThreadID tid);
384
385 private:
386 /** Pointer to the O3CPU. */
387 O3CPU *cpu;
388
389 /** Time buffer interface. */
390 TimeBuffer<TimeStruct> *timeBuffer;
391
392 /** Wire to get decode's information from backwards time buffer. */
393 typename TimeBuffer<TimeStruct>::wire fromDecode;
394
395 /** Wire to get rename's information from backwards time buffer. */
396 typename TimeBuffer<TimeStruct>::wire fromRename;
397
398 /** Wire to get iew's information from backwards time buffer. */
399 typename TimeBuffer<TimeStruct>::wire fromIEW;
400
401 /** Wire to get commit's information from backwards time buffer. */
402 typename TimeBuffer<TimeStruct>::wire fromCommit;
403
404 /** Internal fetch instruction queue. */
405 TimeBuffer<FetchStruct> *fetchQueue;
406
407 //Might be annoying how this name is different than the queue.
408 /** Wire used to write any information heading to decode. */
409 typename TimeBuffer<FetchStruct>::wire toDecode;
410
411 /** BPredUnit. */
412 BPredUnit *branchPred;
413
414 TheISA::PCState pc[Impl::MaxThreads];
415
416 Addr fetchOffset[Impl::MaxThreads];
417
418 StaticInstPtr macroop[Impl::MaxThreads];
419
420 /** Can the fetch stage redirect from an interrupt on this instruction? */
421 bool delayedCommit[Impl::MaxThreads];
422
423 /** Memory request used to access cache. */
424 RequestPtr memReq[Impl::MaxThreads];
425
426 /** Variable that tracks if fetch has written to the time buffer this
427 * cycle. Used to tell CPU if there is activity this cycle.
428 */
429 bool wroteToTimeBuffer;
430
431 /** Tracks how many instructions has been fetched this cycle. */
432 int numInst;
433
434 /** Source of possible stalls. */
435 struct Stalls {
436 bool decode;
437 bool rename;
438 bool iew;
439 bool commit;
440 bool drain;
441 };
442
443 /** Tracks which stages are telling fetch to stall. */
444 Stalls stalls[Impl::MaxThreads];
445
446 /** Decode to fetch delay. */
447 Cycles decodeToFetchDelay;
448
449 /** Rename to fetch delay. */
450 Cycles renameToFetchDelay;
451
452 /** IEW to fetch delay. */
453 Cycles iewToFetchDelay;
454
455 /** Commit to fetch delay. */
456 Cycles commitToFetchDelay;
457
458 /** The width of fetch in instructions. */
459 unsigned fetchWidth;
460
461 /** Is the cache blocked? If so no threads can access it. */
462 bool cacheBlocked;
463
464 /** The packet that is waiting to be retried. */
465 PacketPtr retryPkt;
466
467 /** The thread that is waiting on the cache to tell fetch to retry. */
468 ThreadID retryTid;
469
470 /** Cache block size. */
471 unsigned int cacheBlkSize;
472
473 /** The size of the fetch buffer in bytes. The fetch buffer
474 * itself may be smaller than a cache line.
475 */
476 unsigned fetchBufferSize;
477
478 /** Mask to align a fetch address to a fetch buffer boundary. */
479 Addr fetchBufferMask;
480
481 /** The fetch data that is being fetched and buffered. */
482 uint8_t *fetchBuffer[Impl::MaxThreads];
483
484 /** The PC of the first instruction loaded into the fetch buffer. */
485 Addr fetchBufferPC[Impl::MaxThreads];
486
487 /** Whether or not the fetch buffer data is valid. */
488 bool fetchBufferValid[Impl::MaxThreads];
489
490 /** Size of instructions. */
491 int instSize;
492
493 /** Icache stall statistics. */
494 Counter lastIcacheStall[Impl::MaxThreads];
495
496 /** List of Active Threads */
497 std::list<ThreadID> *activeThreads;
498
499 /** Number of threads. */
500 ThreadID numThreads;
501
502 /** Number of threads that are actively fetching. */
503 ThreadID numFetchingThreads;
504
505 /** Thread ID being fetched. */
506 ThreadID threadFetched;
507
508 /** Checks if there is an interrupt pending. If there is, fetch
509 * must stop once it is not fetching PAL instructions.
510 */
511 bool interruptPending;
512
513 /** Set to true if a pipelined I-cache request should be issued. */
514 bool issuePipelinedIfetch[Impl::MaxThreads];
515
516 /** Event used to delay fault generation of translation faults */
517 FinishTranslationEvent finishTranslationEvent;
518
519 // @todo: Consider making these vectors and tracking on a per thread basis.
520 /** Stat for total number of cycles stalled due to an icache miss. */
521 Stats::Scalar icacheStallCycles;
522 /** Stat for total number of fetched instructions. */
523 Stats::Scalar fetchedInsts;
524 /** Total number of fetched branches. */
525 Stats::Scalar fetchedBranches;
526 /** Stat for total number of predicted branches. */
527 Stats::Scalar predictedBranches;
528 /** Stat for total number of cycles spent fetching. */
529 Stats::Scalar fetchCycles;
530 /** Stat for total number of cycles spent squashing. */
531 Stats::Scalar fetchSquashCycles;
532 /** Stat for total number of cycles spent waiting for translation */
533 Stats::Scalar fetchTlbCycles;
534 /** Stat for total number of cycles spent blocked due to other stages in
535 * the pipeline.
536 */
537 Stats::Scalar fetchIdleCycles;
538 /** Total number of cycles spent blocked. */
539 Stats::Scalar fetchBlockedCycles;
540 /** Total number of cycles spent in any other state. */
541 Stats::Scalar fetchMiscStallCycles;
542 /** Total number of cycles spent in waiting for drains. */
543 Stats::Scalar fetchPendingDrainCycles;
544 /** Total number of stall cycles caused by no active threads to run. */
545 Stats::Scalar fetchNoActiveThreadStallCycles;
546 /** Total number of stall cycles caused by pending traps. */
547 Stats::Scalar fetchPendingTrapStallCycles;
548 /** Total number of stall cycles caused by pending quiesce instructions. */
549 Stats::Scalar fetchPendingQuiesceStallCycles;
550 /** Total number of stall cycles caused by I-cache wait retrys. */
551 Stats::Scalar fetchIcacheWaitRetryStallCycles;
552 /** Stat for total number of fetched cache lines. */
553 Stats::Scalar fetchedCacheLines;
554 /** Total number of outstanding icache accesses that were dropped
555 * due to a squash.
556 */
557 Stats::Scalar fetchIcacheSquashes;
558 /** Total number of outstanding tlb accesses that were dropped
559 * due to a squash.
560 */
561 Stats::Scalar fetchTlbSquashes;
562 /** Distribution of number of instructions fetched each cycle. */
563 Stats::Distribution fetchNisnDist;
564 /** Rate of how often fetch was idle. */
565 Stats::Formula idleRate;
566 /** Number of branch fetches per cycle. */
567 Stats::Formula branchRate;
568 /** Number of instruction fetched per cycle. */
569 Stats::Formula fetchRate;
570};
571
572#endif //__CPU_O3_FETCH_HH__
|