2 * Copyright (c) 2013-2014 ARM Limited
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.
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions are
16 * met: redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer;
18 * redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution;
21 * neither the name of the copyright holders nor the names of its
22 * contributors may be used to endorse or promote products derived from
23 * this software without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
26 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
27 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
28 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
29 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
30 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
31 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
32 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
33 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
34 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
35 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37 * Authors: Andrew Bardsley
42 #include "arch/decoder.hh"
43 #include "arch/utility.hh"
44 #include "cpu/minor/fetch2.hh"
45 #include "cpu/minor/pipeline.hh"
46 #include "cpu/pred/bpred_unit.hh"
47 #include "debug/Branch.hh"
48 #include "debug/Fetch.hh"
49 #include "debug/MinorTrace.hh"
54 Fetch2::Fetch2(const std::string
&name
,
56 MinorCPUParams
¶ms
,
57 Latch
<ForwardLineData
>::Output inp_
,
58 Latch
<BranchData
>::Output branchInp_
,
59 Latch
<BranchData
>::Input predictionOut_
,
60 Latch
<ForwardInstData
>::Input out_
,
61 std::vector
<InputBuffer
<ForwardInstData
>> &next_stage_input_buffer
) :
65 branchInp(branchInp_
),
66 predictionOut(predictionOut_
),
68 nextStageReserve(next_stage_input_buffer
),
69 outputWidth(params
.decodeInputWidth
),
70 processMoreThanOneInput(params
.fetch2CycleInput
),
71 branchPredictor(*params
.branchPred
),
72 fetchInfo(params
.numThreads
),
76 fatal("%s: decodeInputWidth must be >= 1 (%d)\n", name
, outputWidth
);
78 if (params
.fetch2InputBufferSize
< 1) {
79 fatal("%s: fetch2InputBufferSize must be >= 1 (%d)\n", name
,
80 params
.fetch2InputBufferSize
);
83 /* Per-thread input buffers */
84 for (ThreadID tid
= 0; tid
< params
.numThreads
; tid
++) {
85 inputBuffer
.push_back(
86 InputBuffer
<ForwardLineData
>(
87 name
+ ".inputBuffer" + std::to_string(tid
), "lines",
88 params
.fetch2InputBufferSize
));
92 const ForwardLineData
*
93 Fetch2::getInput(ThreadID tid
)
95 /* Get a line from the inputBuffer to work with */
96 if (!inputBuffer
[tid
].empty()) {
97 return &(inputBuffer
[tid
].front());
104 Fetch2::popInput(ThreadID tid
)
106 if (!inputBuffer
[tid
].empty()) {
107 inputBuffer
[tid
].front().freeLine();
108 inputBuffer
[tid
].pop();
111 fetchInfo
[tid
].inputIndex
= 0;
115 Fetch2::dumpAllInput(ThreadID tid
)
117 DPRINTF(Fetch
, "Dumping whole input buffer\n");
118 while (!inputBuffer
[tid
].empty())
121 fetchInfo
[tid
].inputIndex
= 0;
125 Fetch2::updateBranchPrediction(const BranchData
&branch
)
127 MinorDynInstPtr inst
= branch
.inst
;
129 /* Don't even consider instructions we didn't try to predict or faults */
130 if (inst
->isFault() || !inst
->triedToPredict
)
133 switch (branch
.reason
) {
134 case BranchData::NoBranch
:
135 /* No data to update */
137 case BranchData::Interrupt
:
138 /* Never try to predict interrupts */
140 case BranchData::SuspendThread
:
141 /* Don't need to act on suspends */
143 case BranchData::HaltFetch
:
144 /* Don't need to act on fetch wakeup */
146 case BranchData::BranchPrediction
:
147 /* Shouldn't happen. Fetch2 is the only source of
148 * BranchPredictions */
150 case BranchData::UnpredictedBranch
:
151 /* Unpredicted branch or barrier */
152 DPRINTF(Branch
, "Unpredicted branch seen inst: %s\n", *inst
);
153 branchPredictor
.squash(inst
->id
.fetchSeqNum
,
154 branch
.target
, true, inst
->id
.threadId
);
156 case BranchData::CorrectlyPredictedBranch
:
157 /* Predicted taken, was taken */
158 DPRINTF(Branch
, "Branch predicted correctly inst: %s\n", *inst
);
159 branchPredictor
.update(inst
->id
.fetchSeqNum
,
162 case BranchData::BadlyPredictedBranch
:
163 /* Predicted taken, not taken */
164 DPRINTF(Branch
, "Branch mis-predicted inst: %s\n", *inst
);
165 branchPredictor
.squash(inst
->id
.fetchSeqNum
,
166 branch
.target
/* Not used */, false, inst
->id
.threadId
);
168 case BranchData::BadlyPredictedBranchTarget
:
169 /* Predicted taken, was taken but to a different target */
170 DPRINTF(Branch
, "Branch mis-predicted target inst: %s target: %s\n",
171 *inst
, branch
.target
);
172 branchPredictor
.squash(inst
->id
.fetchSeqNum
,
173 branch
.target
, true, inst
->id
.threadId
);
179 Fetch2::predictBranch(MinorDynInstPtr inst
, BranchData
&branch
)
181 Fetch2ThreadInfo
&thread
= fetchInfo
[inst
->id
.threadId
];
182 TheISA::PCState inst_pc
= inst
->pc
;
184 assert(!inst
->predictedTaken
);
186 /* Skip non-control/sys call instructions */
187 if (inst
->staticInst
->isControl() ||
188 inst
->staticInst
->isSyscall())
190 /* Tried to predict */
191 inst
->triedToPredict
= true;
193 DPRINTF(Branch
, "Trying to predict for inst: %s\n", *inst
);
195 if (branchPredictor
.predict(inst
->staticInst
,
196 inst
->id
.fetchSeqNum
, inst_pc
,
199 inst
->predictedTaken
= true;
200 inst
->predictedTarget
= inst_pc
;
201 branch
.target
= inst_pc
;
204 DPRINTF(Branch
, "Not attempting prediction for inst: %s\n", *inst
);
207 /* If we predict taken, set branch and update sequence numbers */
208 if (inst
->predictedTaken
) {
209 /* Update the predictionSeqNum and remember the streamSeqNum that it
210 * was associated with */
211 thread
.expectedStreamSeqNum
= inst
->id
.streamSeqNum
;
213 BranchData new_branch
= BranchData(BranchData::BranchPrediction
,
215 inst
->id
.streamSeqNum
, thread
.predictionSeqNum
+ 1,
216 inst
->predictedTarget
, inst
);
218 /* Mark with a new prediction number by the stream number of the
219 * instruction causing the prediction */
220 thread
.predictionSeqNum
++;
223 DPRINTF(Branch
, "Branch predicted taken inst: %s target: %s"
224 " new predictionSeqNum: %d\n",
225 *inst
, inst
->predictedTarget
, thread
.predictionSeqNum
);
232 /* Push input onto appropriate input buffer */
233 if (!inp
.outputWire
->isBubble())
234 inputBuffer
[inp
.outputWire
->id
.threadId
].setTail(*inp
.outputWire
);
236 ForwardInstData
&insts_out
= *out
.inputWire
;
237 BranchData prediction
;
238 BranchData
&branch_inp
= *branchInp
.outputWire
;
240 assert(insts_out
.isBubble());
242 /* React to branches from Execute to update local branch prediction
244 updateBranchPrediction(branch_inp
);
246 /* If a branch arrives, don't try and do anything about it. Only
247 * react to your own predictions */
248 if (branch_inp
.isStreamChange()) {
249 DPRINTF(Fetch
, "Dumping all input as a stream changing branch"
251 dumpAllInput(branch_inp
.threadId
);
252 fetchInfo
[branch_inp
.threadId
].havePC
= false;
255 assert(insts_out
.isBubble());
256 /* Even when blocked, clear out input lines with the wrong
257 * prediction sequence number */
258 for (ThreadID tid
= 0; tid
< cpu
.numThreads
; tid
++) {
259 Fetch2ThreadInfo
&thread
= fetchInfo
[tid
];
261 thread
.blocked
= !nextStageReserve
[tid
].canReserve();
263 const ForwardLineData
*line_in
= getInput(tid
);
266 thread
.expectedStreamSeqNum
== line_in
->id
.streamSeqNum
&&
267 thread
.predictionSeqNum
!= line_in
->id
.predictionSeqNum
)
269 DPRINTF(Fetch
, "Discarding line %s"
270 " due to predictionSeqNum mismatch (expected: %d)\n",
271 line_in
->id
, thread
.predictionSeqNum
);
274 fetchInfo
[tid
].havePC
= false;
276 if (processMoreThanOneInput
) {
277 DPRINTF(Fetch
, "Wrapping\n");
278 line_in
= getInput(tid
);
285 ThreadID tid
= getScheduledThread();
286 DPRINTF(Fetch
, "Scheduled Thread: %d\n", tid
);
288 assert(insts_out
.isBubble());
289 if (tid
!= InvalidThreadID
) {
290 Fetch2ThreadInfo
&fetch_info
= fetchInfo
[tid
];
292 const ForwardLineData
*line_in
= getInput(tid
);
294 unsigned int output_index
= 0;
296 /* Pack instructions into the output while we can. This may involve
297 * using more than one input line. Note that lineWidth will be 0
298 * for faulting lines */
300 (line_in
->isFault() ||
301 fetch_info
.inputIndex
< line_in
->lineWidth
) && /* More input */
302 output_index
< outputWidth
&& /* More output to fill */
303 prediction
.isBubble() /* No predicted branch */)
305 ThreadContext
*thread
= cpu
.getContext(line_in
->id
.threadId
);
306 TheISA::Decoder
*decoder
= thread
->getDecoderPtr();
308 /* Discard line due to prediction sequence number being wrong but
309 * without the streamSeqNum number having changed */
311 fetch_info
.expectedStreamSeqNum
== line_in
->id
.streamSeqNum
&&
312 fetch_info
.predictionSeqNum
!= line_in
->id
.predictionSeqNum
;
314 /* Set the PC if the stream changes. Setting havePC to false in
315 * a previous cycle handles all other change of flow of control
317 bool set_pc
= fetch_info
.lastStreamSeqNum
!= line_in
->id
.streamSeqNum
;
319 if (!discard_line
&& (!fetch_info
.havePC
|| set_pc
)) {
320 /* Set the inputIndex to be the MachInst-aligned offset
321 * from lineBaseAddr of the new PC value */
322 fetch_info
.inputIndex
=
323 (line_in
->pc
.instAddr() & BaseCPU::PCMask
) -
324 line_in
->lineBaseAddr
;
325 DPRINTF(Fetch
, "Setting new PC value: %s inputIndex: 0x%x"
326 " lineBaseAddr: 0x%x lineWidth: 0x%x\n",
327 line_in
->pc
, fetch_info
.inputIndex
, line_in
->lineBaseAddr
,
329 fetch_info
.pc
= line_in
->pc
;
330 fetch_info
.havePC
= true;
334 /* The generated instruction. Leave as NULL if no instruction
335 * is to be packed into the output */
336 MinorDynInstPtr dyn_inst
= NULL
;
339 /* Rest of line was from an older prediction in the same
341 DPRINTF(Fetch
, "Discarding line %s (from inputIndex: %d)"
342 " due to predictionSeqNum mismatch (expected: %d)\n",
343 line_in
->id
, fetch_info
.inputIndex
,
344 fetch_info
.predictionSeqNum
);
345 } else if (line_in
->isFault()) {
346 /* Pack a fault as a MinorDynInst with ->fault set */
348 /* Make a new instruction and pick up the line, stream,
349 * prediction, thread ids from the incoming line */
350 dyn_inst
= new MinorDynInst(line_in
->id
);
352 /* Fetch and prediction sequence numbers originate here */
353 dyn_inst
->id
.fetchSeqNum
= fetch_info
.fetchSeqNum
;
354 dyn_inst
->id
.predictionSeqNum
= fetch_info
.predictionSeqNum
;
355 /* To complete the set, test that exec sequence number has
357 assert(dyn_inst
->id
.execSeqNum
== 0);
359 dyn_inst
->pc
= fetch_info
.pc
;
361 /* Pack a faulting instruction but allow other
362 * instructions to be generated. (Fetch2 makes no
363 * immediate judgement about streamSeqNum) */
364 dyn_inst
->fault
= line_in
->fault
;
365 DPRINTF(Fetch
, "Fault being passed output_index: "
366 "%d: %s\n", output_index
, dyn_inst
->fault
->name());
368 uint8_t *line
= line_in
->line
;
370 TheISA::MachInst inst_word
;
371 /* The instruction is wholly in the line, can just
373 inst_word
= TheISA::gtoh(
374 *(reinterpret_cast<TheISA::MachInst
*>
375 (line
+ fetch_info
.inputIndex
)));
377 if (!decoder
->instReady()) {
378 decoder
->moreBytes(fetch_info
.pc
,
379 line_in
->lineBaseAddr
+ fetch_info
.inputIndex
,
381 DPRINTF(Fetch
, "Offering MachInst to decoder addr: 0x%x\n",
382 line_in
->lineBaseAddr
+ fetch_info
.inputIndex
);
385 /* Maybe make the above a loop to accomodate ISAs with
386 * instructions longer than sizeof(MachInst) */
388 if (decoder
->instReady()) {
389 /* Make a new instruction and pick up the line, stream,
390 * prediction, thread ids from the incoming line */
391 dyn_inst
= new MinorDynInst(line_in
->id
);
393 /* Fetch and prediction sequence numbers originate here */
394 dyn_inst
->id
.fetchSeqNum
= fetch_info
.fetchSeqNum
;
395 dyn_inst
->id
.predictionSeqNum
= fetch_info
.predictionSeqNum
;
396 /* To complete the set, test that exec sequence number
397 * has not been set */
398 assert(dyn_inst
->id
.execSeqNum
== 0);
400 /* Note that the decoder can update the given PC.
401 * Remember not to assign it until *after* calling
403 StaticInstPtr decoded_inst
= decoder
->decode(fetch_info
.pc
);
404 dyn_inst
->staticInst
= decoded_inst
;
406 dyn_inst
->pc
= fetch_info
.pc
;
407 DPRINTF(Fetch
, "decoder inst %s\n", *dyn_inst
);
410 DPRINTF(Fetch
, "Instruction extracted from line %s"
411 " lineWidth: %d output_index: %d inputIndex: %d"
412 " pc: %s inst: %s\n",
414 line_in
->lineWidth
, output_index
, fetch_info
.inputIndex
,
415 fetch_info
.pc
, *dyn_inst
);
417 #if THE_ISA == X86_ISA || THE_ISA == ARM_ISA
418 /* In SE mode, it's possible to branch to a microop when
419 * replaying faults such as page faults (or simply
420 * intra-microcode branches in X86). Unfortunately,
421 * as Minor has micro-op decomposition in a separate
422 * pipeline stage from instruction decomposition, the
423 * following advancePC (which may follow a branch with
424 * microPC() != 0) *must* see a fresh macroop. This
425 * kludge should be improved with an addition to PCState
426 * but I offer it in this form for the moment
428 * X86 can branch within microops so we need to deal with
429 * the case that, after a branch, the first un-advanced PC
430 * may be pointing to a microop other than 0. Once
431 * advanced, however, the microop number *must* be 0 */
432 fetch_info
.pc
.upc(0);
433 fetch_info
.pc
.nupc(1);
436 /* Advance PC for the next instruction */
437 TheISA::advancePC(fetch_info
.pc
, decoded_inst
);
439 /* Predict any branches and issue a branch if
441 predictBranch(dyn_inst
, prediction
);
443 DPRINTF(Fetch
, "Inst not ready yet\n");
446 /* Step on the pointer into the line if there's no
447 * complete instruction waiting */
448 if (decoder
->needMoreBytes()) {
449 fetch_info
.inputIndex
+= sizeof(TheISA::MachInst
);
451 DPRINTF(Fetch
, "Updated inputIndex value PC: %s"
452 " inputIndex: 0x%x lineBaseAddr: 0x%x lineWidth: 0x%x\n",
453 line_in
->pc
, fetch_info
.inputIndex
, line_in
->lineBaseAddr
,
459 /* Step to next sequence number */
460 fetch_info
.fetchSeqNum
++;
462 /* Correctly size the output before writing */
463 if (output_index
== 0) {
464 insts_out
.resize(outputWidth
);
466 /* Pack the generated dynamic instruction into the output */
467 insts_out
.insts
[output_index
] = dyn_inst
;
470 /* Output MinorTrace instruction info for
471 * pre-microop decomposition macroops */
472 if (DTRACE(MinorTrace
) && !dyn_inst
->isFault() &&
473 dyn_inst
->staticInst
->isMacroop())
475 dyn_inst
->minorTraceInst(*this);
479 /* Remember the streamSeqNum of this line so we can tell when
480 * we change stream */
481 fetch_info
.lastStreamSeqNum
= line_in
->id
.streamSeqNum
;
483 /* Asked to discard line or there was a branch or fault */
484 if (!prediction
.isBubble() || /* The remains of a
485 line with a prediction in it */
486 line_in
->isFault() /* A line which is just a fault */)
488 DPRINTF(Fetch
, "Discarding all input on branch/fault\n");
490 fetch_info
.havePC
= false;
492 } else if (discard_line
) {
493 /* Just discard one line, one's behind it may have new
494 * stream sequence numbers. There's a DPRINTF above
497 fetch_info
.havePC
= false;
499 } else if (fetch_info
.inputIndex
== line_in
->lineWidth
) {
500 /* Got to end of a line, pop the line but keep PC
501 * in case this is a line-wrapping inst. */
506 if (!line_in
&& processMoreThanOneInput
) {
507 DPRINTF(Fetch
, "Wrapping\n");
508 line_in
= getInput(tid
);
512 /* The rest of the output (if any) should already have been packed
513 * with bubble instructions by insts_out's initialisation */
515 if (tid
== InvalidThreadID
) {
516 assert(insts_out
.isBubble());
518 /** Reserve a slot in the next stage and output data */
519 *predictionOut
.inputWire
= prediction
;
521 /* If we generated output, reserve space for the result in the next stage
522 * and mark the stage as being active this cycle */
523 if (!insts_out
.isBubble()) {
524 /* Note activity of following buffer */
525 cpu
.activityRecorder
->activity();
526 insts_out
.threadId
= tid
;
527 nextStageReserve
[tid
].reserve();
530 /* If we still have input to process and somewhere to put it,
531 * mark stage as active */
532 for (ThreadID i
= 0; i
< cpu
.numThreads
; i
++)
534 if (getInput(i
) && nextStageReserve
[i
].canReserve()) {
535 cpu
.activityRecorder
->activateStage(Pipeline::Fetch2StageId
);
540 /* Make sure the input (if any left) is pushed */
541 if (!inp
.outputWire
->isBubble())
542 inputBuffer
[inp
.outputWire
->id
.threadId
].pushTail();
546 Fetch2::getScheduledThread()
548 /* Select thread via policy. */
549 std::vector
<ThreadID
> priority_list
;
551 switch (cpu
.threadPolicy
) {
552 case Enums::SingleThreaded
:
553 priority_list
.push_back(0);
555 case Enums::RoundRobin
:
556 priority_list
= cpu
.roundRobinPriority(threadPriority
);
559 priority_list
= cpu
.randomPriority();
562 panic("Unknown fetch policy");
565 for (auto tid
: priority_list
) {
566 if (getInput(tid
) && !fetchInfo
[tid
].blocked
) {
567 threadPriority
= tid
;
572 return InvalidThreadID
;
578 for (const auto &buffer
: inputBuffer
) {
583 return (*inp
.outputWire
).isBubble() &&
584 (*predictionOut
.inputWire
).isBubble();
588 Fetch2::minorTrace() const
590 std::ostringstream data
;
592 if (fetchInfo
[0].blocked
)
595 (*out
.inputWire
).reportData(data
);
597 MINORTRACE("inputIndex=%d havePC=%d predictionSeqNum=%d insts=%s\n",
598 fetchInfo
[0].inputIndex
, fetchInfo
[0].havePC
, fetchInfo
[0].predictionSeqNum
, data
.str());
599 inputBuffer
[0].minorTrace();