2 * Copyright (c) 2007 MIPS Technologies, Inc.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 * Authors: Korey Sewell
32 #include "config/the_isa.hh"
33 #include "cpu/inorder/resources/fetch_seq_unit.hh"
34 #include "cpu/inorder/resource_pool.hh"
37 using namespace TheISA
;
38 using namespace ThePipeline
;
40 FetchSeqUnit::FetchSeqUnit(std::string res_name
, int res_id
, int res_width
,
41 int res_latency
, InOrderCPU
*_cpu
, ThePipeline::Params
*params
)
42 : Resource(res_name
, res_id
, res_width
, res_latency
, _cpu
),
43 instSize(sizeof(MachInst
))
45 for (ThreadID tid
= 0; tid
< ThePipeline::MaxThreads
; tid
++) {
46 delaySlotInfo
[tid
].numInsts
= 0;
47 delaySlotInfo
[tid
].targetReady
= false;
50 pcBlockStage
[tid
] = 0;
52 squashSeqNum
[tid
] = (InstSeqNum
)-1;
53 lastSquashCycle
[tid
] = 0;
57 FetchSeqUnit::~FetchSeqUnit()
59 delete [] resourceEvent
;
65 resourceEvent
= new FetchSeqEvent
[width
];
71 FetchSeqUnit::execute(int slot_num
)
73 // After this is working, change this to a reinterpret cast
74 // for performance considerations
75 ResourceRequest
* fs_req
= reqMap
[slot_num
];
76 DynInstPtr inst
= fs_req
->inst
;
77 ThreadID tid
= inst
->readTid();
78 int stage_num
= fs_req
->getStageNum();
79 int seq_num
= inst
->seqNum
;
81 fs_req
->fault
= NoFault
;
89 if (delaySlotInfo
[tid
].targetReady
&&
90 delaySlotInfo
[tid
].numInsts
== 0) {
92 PC
[tid
] = delaySlotInfo
[tid
].targetAddr
; //next_PC
93 nextPC
[tid
] = PC
[tid
] + instSize
; //next_NPC
94 nextNPC
[tid
] = PC
[tid
] + (2 * instSize
);
96 delaySlotInfo
[tid
].targetReady
= false;
98 DPRINTF(InOrderFetchSeq
, "[tid:%i]: Setting PC to delay slot target\n",tid
);
101 inst
->setPC(PC
[tid
]);
102 inst
->setNextPC(PC
[tid
] + instSize
);
103 inst
->setNextNPC(PC
[tid
] + (instSize
* 2));
105 #if ISA_HAS_DELAY_SLOT
106 inst
->setPredTarg(inst
->readNextNPC());
108 inst
->setPredTarg(inst
->readNextPC());
110 inst
->setMemAddr(PC
[tid
]);
111 inst
->setSeqNum(cpu
->getAndIncrementInstSeq(tid
));
113 DPRINTF(InOrderFetchSeq
, "[tid:%i]: Assigning [sn:%i] to PC %08p, NPC %08p, NNPC %08p\n", tid
,
114 inst
->seqNum
, inst
->readPC(), inst
->readNextPC(), inst
->readNextNPC());
116 if (delaySlotInfo
[tid
].numInsts
> 0) {
117 --delaySlotInfo
[tid
].numInsts
;
119 // It's OK to set PC to target of branch
120 if (delaySlotInfo
[tid
].numInsts
== 0) {
121 delaySlotInfo
[tid
].targetReady
= true;
124 DPRINTF(InOrderFetchSeq
, "[tid:%i]: %i delay slot inst(s) left to"
125 " process.\n", tid
, delaySlotInfo
[tid
].numInsts
);
128 PC
[tid
] = nextPC
[tid
];
129 nextPC
[tid
] = nextNPC
[tid
];
130 nextNPC
[tid
] += instSize
;
134 DPRINTF(InOrderStall
, "STALL: [tid:%i]: NPC not valid\n", tid
);
135 fs_req
->setCompleted(false);
142 if (inst
->isControl()) {
143 // If it's a return, then we must wait for resolved address.
144 if (inst
->isReturn() && !inst
->predTaken()) {
145 cpu
->pipelineStage
[stage_num
]->toPrevStages
->stageBlock
[stage_num
][tid
] = true;
146 pcValid
[tid
] = false;
147 pcBlockStage
[tid
] = stage_num
;
148 } else if (inst
->isCondDelaySlot() && !inst
->predTaken()) {
149 // Not-Taken AND Conditional Control
150 DPRINTF(InOrderFetchSeq
, "[tid:%i]: [sn:%i]: [PC:%08p] Predicted Not-Taken Cond. "
151 "Delay inst. Skipping delay slot and Updating PC to %08p\n",
152 tid
, inst
->seqNum
, inst
->readPC(), inst
->readPredTarg());
154 DPRINTF(InOrderFetchSeq
, "[tid:%i] Setting up squash to start from stage %i, after [sn:%i].\n",
155 tid
, stage_num
, seq_num
);
157 inst
->bdelaySeqNum
= seq_num
;
158 inst
->squashingStage
= stage_num
;
160 squashAfterInst(inst
, stage_num
, tid
);
161 } else if (!inst
->isCondDelaySlot() && !inst
->predTaken()) {
163 DPRINTF(InOrderFetchSeq
, "[tid:%i]: [sn:%i]: Predicted Not-Taken Control "
164 "inst. updating PC to %08p\n", tid
, inst
->seqNum
,
166 #if ISA_HAS_DELAY_SLOT
167 ++delaySlotInfo
[tid
].numInsts
;
168 delaySlotInfo
[tid
].targetReady
= false;
169 delaySlotInfo
[tid
].targetAddr
= inst
->readNextNPC();
171 assert(delaySlotInfo
[tid
].numInsts
== 0);
173 } else if (inst
->predTaken()) {
175 #if ISA_HAS_DELAY_SLOT
176 ++delaySlotInfo
[tid
].numInsts
;
177 delaySlotInfo
[tid
].targetReady
= false;
178 delaySlotInfo
[tid
].targetAddr
= inst
->readPredTarg();
180 DPRINTF(InOrderFetchSeq
, "[tid:%i]: [sn:%i] Updating delay slot target "
181 "to PC %08p\n", tid
, inst
->seqNum
, inst
->readPredTarg());
182 inst
->bdelaySeqNum
= seq_num
+ 1;
184 inst
->bdelaySeqNum
= seq_num
;
185 assert(delaySlotInfo
[tid
].numInsts
== 0);
188 inst
->squashingStage
= stage_num
;
190 DPRINTF(InOrderFetchSeq
, "[tid:%i] Setting up squash to start from stage %i, after [sn:%i].\n",
191 tid
, stage_num
, inst
->bdelaySeqNum
);
194 squashAfterInst(inst
, stage_num
, tid
);
197 DPRINTF(InOrderFetchSeq
, "[tid:%i]: [sn:%i]: Ignoring branch target update "
198 "since then is not a control instruction.\n", tid
, inst
->seqNum
);
206 fatal("Unrecognized command to %s", resName
);
211 FetchSeqUnit::squashAfterInst(DynInstPtr inst
, int stage_num
, ThreadID tid
)
213 // Squash In Pipeline Stage
214 cpu
->pipelineStage
[stage_num
]->squashDueToBranch(inst
, tid
);
216 // Squash inside current resource, so if there needs to be fetching on same cycle
217 // the fetch information will be correct.
218 // squash(inst, stage_num, inst->bdelaySeqNum, tid);
220 // Schedule Squash Through-out Resource Pool
221 cpu
->resPool
->scheduleEvent((InOrderCPU::CPUEventType
)ResourcePool::SquashAll
, inst
, 0);
224 FetchSeqUnit::squash(DynInstPtr inst
, int squash_stage
,
225 InstSeqNum squash_seq_num
, ThreadID tid
)
227 DPRINTF(InOrderFetchSeq
, "[tid:%i]: Updating due to squash from stage %i.\n",
230 InstSeqNum done_seq_num
= inst
->bdelaySeqNum
;
232 // Handles the case where we are squashing because of something that is
233 // not a branch...like a memory stall
234 Addr new_PC
= (inst
->isControl()) ?
235 inst
->readPredTarg() : inst
->readPC() + instSize
;
237 if (squashSeqNum
[tid
] <= done_seq_num
&&
238 lastSquashCycle
[tid
] == curTick
) {
239 DPRINTF(InOrderFetchSeq
, "[tid:%i]: Ignoring squash from stage %i, since"
240 "there is an outstanding squash that is older.\n",
243 squashSeqNum
[tid
] = done_seq_num
;
244 lastSquashCycle
[tid
] = curTick
;
246 // If The very next instruction number is the done seq. num,
247 // then we haven't seen the delay slot yet ... if it isn't
248 // the last done_seq_num then this is the delay slot inst.
249 if (cpu
->nextInstSeqNum(tid
) != done_seq_num
&&
250 !inst
->procDelaySlotOnMispred
) {
251 delaySlotInfo
[tid
].numInsts
= 0;
252 delaySlotInfo
[tid
].targetReady
= false;
256 nextPC
[tid
] = new_PC
+ instSize
;
257 nextNPC
[tid
] = new_PC
+ (2 * instSize
);
259 DPRINTF(InOrderFetchSeq
, "[tid:%i]: Setting PC to %08p.\n",
262 #if !ISA_HAS_DELAY_SLOT
266 delaySlotInfo
[tid
].numInsts
= 1;
267 delaySlotInfo
[tid
].targetReady
= false;
268 delaySlotInfo
[tid
].targetAddr
= (inst
->procDelaySlotOnMispred
) ? inst
->branchTarget() : new_PC
;
270 // Reset PC to Delay Slot Instruction
271 if (inst
->procDelaySlotOnMispred
) {
273 nextPC
[tid
] = new_PC
+ instSize
;
274 nextNPC
[tid
] = new_PC
+ (2 * instSize
);
279 // Unblock Any Stages Waiting for this information to be updated ...
281 cpu
->pipelineStage
[pcBlockStage
[tid
]]->toPrevStages
->stageUnblock
[pcBlockStage
[tid
]][tid
] = true;
287 Resource::squash(inst
, squash_stage
, squash_seq_num
, tid
);
290 FetchSeqUnit::FetchSeqEvent::FetchSeqEvent()
295 FetchSeqUnit::FetchSeqEvent::process()
297 FetchSeqUnit
* fs_res
= dynamic_cast<FetchSeqUnit
*>(resource
);
300 for (int i
=0; i
< MaxThreads
; i
++) {
301 fs_res
->PC
[i
] = fs_res
->cpu
->readPC(i
);
302 fs_res
->nextPC
[i
] = fs_res
->cpu
->readNextPC(i
);
303 fs_res
->nextNPC
[i
] = fs_res
->cpu
->readNextNPC(i
);
304 DPRINTF(InOrderFetchSeq
, "[tid:%i]: Setting PC:%08p NPC:%08p NNPC:%08p.\n",
305 fs_res
->PC
[i
], fs_res
->nextPC
[i
], fs_res
->nextNPC
[i
]);
307 fs_res
->pcValid
[i
] = true;
310 //cpu->fetchPriorityList.push_back(tid);
315 FetchSeqUnit::activateThread(ThreadID tid
)
319 PC
[tid
] = cpu
->readPC(tid
);
320 nextPC
[tid
] = cpu
->readNextPC(tid
);
321 nextNPC
[tid
] = cpu
->readNextNPC(tid
);
323 cpu
->fetchPriorityList
.push_back(tid
);
325 DPRINTF(InOrderFetchSeq
, "[tid:%i]: Reading PC:%08p NPC:%08p NNPC:%08p.\n",
326 tid
, PC
[tid
], nextPC
[tid
], nextNPC
[tid
]);
330 FetchSeqUnit::deactivateThread(ThreadID tid
)
332 delaySlotInfo
[tid
].numInsts
= 0;
333 delaySlotInfo
[tid
].targetReady
= false;
335 pcValid
[tid
] = false;
336 pcBlockStage
[tid
] = 0;
338 squashSeqNum
[tid
] = (InstSeqNum
)-1;
339 lastSquashCycle
[tid
] = 0;
341 list
<ThreadID
>::iterator thread_it
= find(cpu
->fetchPriorityList
.begin(),
342 cpu
->fetchPriorityList
.end(),
345 if (thread_it
!= cpu
->fetchPriorityList
.end())
346 cpu
->fetchPriorityList
.erase(thread_it
);
350 FetchSeqUnit::suspendThread(ThreadID tid
)
352 deactivateThread(tid
);
356 FetchSeqUnit::updateAfterContextSwitch(DynInstPtr inst
, ThreadID tid
)
360 if (cpu
->thread
[tid
]->lastGradIsBranch
) {
361 /** This function assumes that the instruction causing the context
362 * switch was right after the branch. Thus, if it's not, then
363 * we are updating incorrectly here
365 assert(cpu
->thread
[tid
]->lastBranchNextPC
== inst
->readPC());
367 PC
[tid
] = cpu
->thread
[tid
]->lastBranchNextNPC
;
368 nextPC
[tid
] = PC
[tid
] + instSize
;
369 nextNPC
[tid
] = nextPC
[tid
] + instSize
;
371 PC
[tid
] = inst
->readNextPC();
372 nextPC
[tid
] = inst
->readNextNPC();
373 nextNPC
[tid
] = inst
->readNextNPC() + instSize
;
376 DPRINTF(InOrderFetchSeq
, "[tid:%i]: Updating PCs due to Context Switch."
377 "Assigning PC:%08p NPC:%08p NNPC:%08p.\n", tid
, PC
[tid
],
378 nextPC
[tid
], nextNPC
[tid
]);