1/*
| 1/*
|
2 * Copyright (c) 2010-2011 ARM Limited
| 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#include <algorithm> 45#include <cstring> 46#include <list> 47#include <map> 48#include <queue> 49 50#include "arch/isa_traits.hh" 51#include "arch/tlb.hh" 52#include "arch/utility.hh" 53#include "arch/vtophys.hh" 54#include "base/types.hh" 55#include "config/the_isa.hh" 56#include "cpu/base.hh" 57//#include "cpu/checker/cpu.hh" 58#include "cpu/o3/fetch.hh" 59#include "cpu/exetrace.hh" 60#include "debug/Activity.hh"
| 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#include <algorithm> 45#include <cstring> 46#include <list> 47#include <map> 48#include <queue> 49 50#include "arch/isa_traits.hh" 51#include "arch/tlb.hh" 52#include "arch/utility.hh" 53#include "arch/vtophys.hh" 54#include "base/types.hh" 55#include "config/the_isa.hh" 56#include "cpu/base.hh" 57//#include "cpu/checker/cpu.hh" 58#include "cpu/o3/fetch.hh" 59#include "cpu/exetrace.hh" 60#include "debug/Activity.hh"
|
| 61#include "debug/Drain.hh"
|
61#include "debug/Fetch.hh" 62#include "mem/packet.hh" 63#include "params/DerivO3CPU.hh" 64#include "sim/byteswap.hh" 65#include "sim/core.hh" 66#include "sim/eventq.hh" 67#include "sim/full_system.hh" 68#include "sim/system.hh" 69 70using namespace std; 71 72template<class Impl> 73DefaultFetch<Impl>::DefaultFetch(O3CPU *_cpu, DerivO3CPUParams *params) 74 : cpu(_cpu), 75 branchPred(params),
| 62#include "debug/Fetch.hh" 63#include "mem/packet.hh" 64#include "params/DerivO3CPU.hh" 65#include "sim/byteswap.hh" 66#include "sim/core.hh" 67#include "sim/eventq.hh" 68#include "sim/full_system.hh" 69#include "sim/system.hh" 70 71using namespace std; 72 73template<class Impl> 74DefaultFetch<Impl>::DefaultFetch(O3CPU *_cpu, DerivO3CPUParams *params) 75 : cpu(_cpu), 76 branchPred(params),
|
76 numInst(0),
| |
77 decodeToFetchDelay(params->decodeToFetchDelay), 78 renameToFetchDelay(params->renameToFetchDelay), 79 iewToFetchDelay(params->iewToFetchDelay), 80 commitToFetchDelay(params->commitToFetchDelay), 81 fetchWidth(params->fetchWidth),
| 77 decodeToFetchDelay(params->decodeToFetchDelay), 78 renameToFetchDelay(params->renameToFetchDelay), 79 iewToFetchDelay(params->iewToFetchDelay), 80 commitToFetchDelay(params->commitToFetchDelay), 81 fetchWidth(params->fetchWidth),
|
82 cacheBlocked(false),
| |
83 retryPkt(NULL), 84 retryTid(InvalidThreadID), 85 numThreads(params->numThreads), 86 numFetchingThreads(params->smtNumFetchingThreads),
| 82 retryPkt(NULL), 83 retryTid(InvalidThreadID), 84 numThreads(params->numThreads), 85 numFetchingThreads(params->smtNumFetchingThreads),
|
87 interruptPending(false), 88 drainPending(false), 89 switchedOut(false),
| |
90 finishTranslationEvent(this) 91{ 92 if (numThreads > Impl::MaxThreads) 93 fatal("numThreads (%d) is larger than compiled limit (%d),\n" 94 "\tincrease MaxThreads in src/cpu/o3/impl.hh\n", 95 numThreads, static_cast<int>(Impl::MaxThreads)); 96 if (fetchWidth > Impl::MaxWidth) 97 fatal("fetchWidth (%d) is larger than compiled limit (%d),\n" 98 "\tincrease MaxWidth in src/cpu/o3/impl.hh\n", 99 fetchWidth, static_cast<int>(Impl::MaxWidth)); 100
| 86 finishTranslationEvent(this) 87{ 88 if (numThreads > Impl::MaxThreads) 89 fatal("numThreads (%d) is larger than compiled limit (%d),\n" 90 "\tincrease MaxThreads in src/cpu/o3/impl.hh\n", 91 numThreads, static_cast<int>(Impl::MaxThreads)); 92 if (fetchWidth > Impl::MaxWidth) 93 fatal("fetchWidth (%d) is larger than compiled limit (%d),\n" 94 "\tincrease MaxWidth in src/cpu/o3/impl.hh\n", 95 fetchWidth, static_cast<int>(Impl::MaxWidth)); 96
|
101 // Set fetch stage's status to inactive. 102 _status = Inactive; 103
| |
104 std::string policy = params->smtFetchPolicy; 105 106 // Convert string to lowercase 107 std::transform(policy.begin(), policy.end(), policy.begin(), 108 (int(*)(int)) tolower); 109 110 // Figure out fetch policy 111 if (policy == "singlethread") { 112 fetchPolicy = SingleThread; 113 if (numThreads > 1) 114 panic("Invalid Fetch Policy for a SMT workload."); 115 } else if (policy == "roundrobin") { 116 fetchPolicy = RoundRobin; 117 DPRINTF(Fetch, "Fetch policy set to Round Robin\n"); 118 } else if (policy == "branch") { 119 fetchPolicy = Branch; 120 DPRINTF(Fetch, "Fetch policy set to Branch Count\n"); 121 } else if (policy == "iqcount") { 122 fetchPolicy = IQ; 123 DPRINTF(Fetch, "Fetch policy set to IQ count\n"); 124 } else if (policy == "lsqcount") { 125 fetchPolicy = LSQ; 126 DPRINTF(Fetch, "Fetch policy set to LSQ count\n"); 127 } else { 128 fatal("Invalid Fetch Policy. Options Are: {SingleThread," 129 " RoundRobin,LSQcount,IQcount}\n"); 130 } 131 132 // Get the size of an instruction. 133 instSize = sizeof(TheISA::MachInst); 134 135 for (int i = 0; i < Impl::MaxThreads; i++) { 136 cacheData[i] = NULL; 137 decoder[i] = new TheISA::Decoder; 138 } 139} 140 141template <class Impl> 142std::string 143DefaultFetch<Impl>::name() const 144{ 145 return cpu->name() + ".fetch"; 146} 147 148template <class Impl> 149void 150DefaultFetch<Impl>::regStats() 151{ 152 icacheStallCycles 153 .name(name() + ".icacheStallCycles") 154 .desc("Number of cycles fetch is stalled on an Icache miss") 155 .prereq(icacheStallCycles); 156 157 fetchedInsts 158 .name(name() + ".Insts") 159 .desc("Number of instructions fetch has processed") 160 .prereq(fetchedInsts); 161 162 fetchedBranches 163 .name(name() + ".Branches") 164 .desc("Number of branches that fetch encountered") 165 .prereq(fetchedBranches); 166 167 predictedBranches 168 .name(name() + ".predictedBranches") 169 .desc("Number of branches that fetch has predicted taken") 170 .prereq(predictedBranches); 171 172 fetchCycles 173 .name(name() + ".Cycles") 174 .desc("Number of cycles fetch has run and was not squashing or" 175 " blocked") 176 .prereq(fetchCycles); 177 178 fetchSquashCycles 179 .name(name() + ".SquashCycles") 180 .desc("Number of cycles fetch has spent squashing") 181 .prereq(fetchSquashCycles); 182 183 fetchTlbCycles 184 .name(name() + ".TlbCycles") 185 .desc("Number of cycles fetch has spent waiting for tlb") 186 .prereq(fetchTlbCycles); 187 188 fetchIdleCycles 189 .name(name() + ".IdleCycles") 190 .desc("Number of cycles fetch was idle") 191 .prereq(fetchIdleCycles); 192 193 fetchBlockedCycles 194 .name(name() + ".BlockedCycles") 195 .desc("Number of cycles fetch has spent blocked") 196 .prereq(fetchBlockedCycles); 197 198 fetchedCacheLines 199 .name(name() + ".CacheLines") 200 .desc("Number of cache lines fetched") 201 .prereq(fetchedCacheLines); 202 203 fetchMiscStallCycles 204 .name(name() + ".MiscStallCycles") 205 .desc("Number of cycles fetch has spent waiting on interrupts, or " 206 "bad addresses, or out of MSHRs") 207 .prereq(fetchMiscStallCycles); 208 209 fetchPendingDrainCycles 210 .name(name() + ".PendingDrainCycles") 211 .desc("Number of cycles fetch has spent waiting on pipes to drain") 212 .prereq(fetchPendingDrainCycles); 213 214 fetchNoActiveThreadStallCycles 215 .name(name() + ".NoActiveThreadStallCycles") 216 .desc("Number of stall cycles due to no active thread to fetch from") 217 .prereq(fetchNoActiveThreadStallCycles); 218 219 fetchPendingTrapStallCycles 220 .name(name() + ".PendingTrapStallCycles") 221 .desc("Number of stall cycles due to pending traps") 222 .prereq(fetchPendingTrapStallCycles); 223 224 fetchPendingQuiesceStallCycles 225 .name(name() + ".PendingQuiesceStallCycles") 226 .desc("Number of stall cycles due to pending quiesce instructions") 227 .prereq(fetchPendingQuiesceStallCycles); 228 229 fetchIcacheWaitRetryStallCycles 230 .name(name() + ".IcacheWaitRetryStallCycles") 231 .desc("Number of stall cycles due to full MSHR") 232 .prereq(fetchIcacheWaitRetryStallCycles); 233 234 fetchIcacheSquashes 235 .name(name() + ".IcacheSquashes") 236 .desc("Number of outstanding Icache misses that were squashed") 237 .prereq(fetchIcacheSquashes); 238 239 fetchTlbSquashes 240 .name(name() + ".ItlbSquashes") 241 .desc("Number of outstanding ITLB misses that were squashed") 242 .prereq(fetchTlbSquashes); 243 244 fetchNisnDist 245 .init(/* base value */ 0, 246 /* last value */ fetchWidth, 247 /* bucket size */ 1) 248 .name(name() + ".rateDist") 249 .desc("Number of instructions fetched each cycle (Total)") 250 .flags(Stats::pdf); 251 252 idleRate 253 .name(name() + ".idleRate") 254 .desc("Percent of cycles fetch was idle") 255 .prereq(idleRate); 256 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 257 258 branchRate 259 .name(name() + ".branchRate") 260 .desc("Number of branch fetches per cycle") 261 .flags(Stats::total); 262 branchRate = fetchedBranches / cpu->numCycles; 263 264 fetchRate 265 .name(name() + ".rate") 266 .desc("Number of inst fetches per cycle") 267 .flags(Stats::total); 268 fetchRate = fetchedInsts / cpu->numCycles; 269 270 branchPred.regStats(); 271} 272 273template<class Impl> 274void 275DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 276{ 277 timeBuffer = time_buffer; 278 279 // Create wires to get information from proper places in time buffer. 280 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 281 fromRename = timeBuffer->getWire(-renameToFetchDelay); 282 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 283 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 284} 285 286template<class Impl> 287void 288DefaultFetch<Impl>::setActiveThreads(std::list<ThreadID> *at_ptr) 289{ 290 activeThreads = at_ptr; 291} 292 293template<class Impl> 294void 295DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 296{ 297 fetchQueue = fq_ptr; 298 299 // Create wire to write information to proper place in fetch queue. 300 toDecode = fetchQueue->getWire(0); 301} 302 303template<class Impl> 304void 305DefaultFetch<Impl>::startupStage() 306{
| 97 std::string policy = params->smtFetchPolicy; 98 99 // Convert string to lowercase 100 std::transform(policy.begin(), policy.end(), policy.begin(), 101 (int(*)(int)) tolower); 102 103 // Figure out fetch policy 104 if (policy == "singlethread") { 105 fetchPolicy = SingleThread; 106 if (numThreads > 1) 107 panic("Invalid Fetch Policy for a SMT workload."); 108 } else if (policy == "roundrobin") { 109 fetchPolicy = RoundRobin; 110 DPRINTF(Fetch, "Fetch policy set to Round Robin\n"); 111 } else if (policy == "branch") { 112 fetchPolicy = Branch; 113 DPRINTF(Fetch, "Fetch policy set to Branch Count\n"); 114 } else if (policy == "iqcount") { 115 fetchPolicy = IQ; 116 DPRINTF(Fetch, "Fetch policy set to IQ count\n"); 117 } else if (policy == "lsqcount") { 118 fetchPolicy = LSQ; 119 DPRINTF(Fetch, "Fetch policy set to LSQ count\n"); 120 } else { 121 fatal("Invalid Fetch Policy. Options Are: {SingleThread," 122 " RoundRobin,LSQcount,IQcount}\n"); 123 } 124 125 // Get the size of an instruction. 126 instSize = sizeof(TheISA::MachInst); 127 128 for (int i = 0; i < Impl::MaxThreads; i++) { 129 cacheData[i] = NULL; 130 decoder[i] = new TheISA::Decoder; 131 } 132} 133 134template <class Impl> 135std::string 136DefaultFetch<Impl>::name() const 137{ 138 return cpu->name() + ".fetch"; 139} 140 141template <class Impl> 142void 143DefaultFetch<Impl>::regStats() 144{ 145 icacheStallCycles 146 .name(name() + ".icacheStallCycles") 147 .desc("Number of cycles fetch is stalled on an Icache miss") 148 .prereq(icacheStallCycles); 149 150 fetchedInsts 151 .name(name() + ".Insts") 152 .desc("Number of instructions fetch has processed") 153 .prereq(fetchedInsts); 154 155 fetchedBranches 156 .name(name() + ".Branches") 157 .desc("Number of branches that fetch encountered") 158 .prereq(fetchedBranches); 159 160 predictedBranches 161 .name(name() + ".predictedBranches") 162 .desc("Number of branches that fetch has predicted taken") 163 .prereq(predictedBranches); 164 165 fetchCycles 166 .name(name() + ".Cycles") 167 .desc("Number of cycles fetch has run and was not squashing or" 168 " blocked") 169 .prereq(fetchCycles); 170 171 fetchSquashCycles 172 .name(name() + ".SquashCycles") 173 .desc("Number of cycles fetch has spent squashing") 174 .prereq(fetchSquashCycles); 175 176 fetchTlbCycles 177 .name(name() + ".TlbCycles") 178 .desc("Number of cycles fetch has spent waiting for tlb") 179 .prereq(fetchTlbCycles); 180 181 fetchIdleCycles 182 .name(name() + ".IdleCycles") 183 .desc("Number of cycles fetch was idle") 184 .prereq(fetchIdleCycles); 185 186 fetchBlockedCycles 187 .name(name() + ".BlockedCycles") 188 .desc("Number of cycles fetch has spent blocked") 189 .prereq(fetchBlockedCycles); 190 191 fetchedCacheLines 192 .name(name() + ".CacheLines") 193 .desc("Number of cache lines fetched") 194 .prereq(fetchedCacheLines); 195 196 fetchMiscStallCycles 197 .name(name() + ".MiscStallCycles") 198 .desc("Number of cycles fetch has spent waiting on interrupts, or " 199 "bad addresses, or out of MSHRs") 200 .prereq(fetchMiscStallCycles); 201 202 fetchPendingDrainCycles 203 .name(name() + ".PendingDrainCycles") 204 .desc("Number of cycles fetch has spent waiting on pipes to drain") 205 .prereq(fetchPendingDrainCycles); 206 207 fetchNoActiveThreadStallCycles 208 .name(name() + ".NoActiveThreadStallCycles") 209 .desc("Number of stall cycles due to no active thread to fetch from") 210 .prereq(fetchNoActiveThreadStallCycles); 211 212 fetchPendingTrapStallCycles 213 .name(name() + ".PendingTrapStallCycles") 214 .desc("Number of stall cycles due to pending traps") 215 .prereq(fetchPendingTrapStallCycles); 216 217 fetchPendingQuiesceStallCycles 218 .name(name() + ".PendingQuiesceStallCycles") 219 .desc("Number of stall cycles due to pending quiesce instructions") 220 .prereq(fetchPendingQuiesceStallCycles); 221 222 fetchIcacheWaitRetryStallCycles 223 .name(name() + ".IcacheWaitRetryStallCycles") 224 .desc("Number of stall cycles due to full MSHR") 225 .prereq(fetchIcacheWaitRetryStallCycles); 226 227 fetchIcacheSquashes 228 .name(name() + ".IcacheSquashes") 229 .desc("Number of outstanding Icache misses that were squashed") 230 .prereq(fetchIcacheSquashes); 231 232 fetchTlbSquashes 233 .name(name() + ".ItlbSquashes") 234 .desc("Number of outstanding ITLB misses that were squashed") 235 .prereq(fetchTlbSquashes); 236 237 fetchNisnDist 238 .init(/* base value */ 0, 239 /* last value */ fetchWidth, 240 /* bucket size */ 1) 241 .name(name() + ".rateDist") 242 .desc("Number of instructions fetched each cycle (Total)") 243 .flags(Stats::pdf); 244 245 idleRate 246 .name(name() + ".idleRate") 247 .desc("Percent of cycles fetch was idle") 248 .prereq(idleRate); 249 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 250 251 branchRate 252 .name(name() + ".branchRate") 253 .desc("Number of branch fetches per cycle") 254 .flags(Stats::total); 255 branchRate = fetchedBranches / cpu->numCycles; 256 257 fetchRate 258 .name(name() + ".rate") 259 .desc("Number of inst fetches per cycle") 260 .flags(Stats::total); 261 fetchRate = fetchedInsts / cpu->numCycles; 262 263 branchPred.regStats(); 264} 265 266template<class Impl> 267void 268DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 269{ 270 timeBuffer = time_buffer; 271 272 // Create wires to get information from proper places in time buffer. 273 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 274 fromRename = timeBuffer->getWire(-renameToFetchDelay); 275 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 276 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 277} 278 279template<class Impl> 280void 281DefaultFetch<Impl>::setActiveThreads(std::list<ThreadID> *at_ptr) 282{ 283 activeThreads = at_ptr; 284} 285 286template<class Impl> 287void 288DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 289{ 290 fetchQueue = fq_ptr; 291 292 // Create wire to write information to proper place in fetch queue. 293 toDecode = fetchQueue->getWire(0); 294} 295 296template<class Impl> 297void 298DefaultFetch<Impl>::startupStage() 299{
|
| 300 assert(priorityList.empty()); 301 resetStage(); 302 303 // Fetch needs to start fetching instructions at the very beginning, 304 // so it must start up in active state. 305 switchToActive(); 306} 307 308template<class Impl> 309void 310DefaultFetch<Impl>::resetStage() 311{ 312 numInst = 0; 313 interruptPending = false; 314 cacheBlocked = false; 315 316 priorityList.clear(); 317
|
307 // Setup PC and nextPC with initial state. 308 for (ThreadID tid = 0; tid < numThreads; tid++) {
| 318 // Setup PC and nextPC with initial state. 319 for (ThreadID tid = 0; tid < numThreads; tid++) {
|
| 320 fetchStatus[tid] = Running;
|
309 pc[tid] = cpu->pcState(tid); 310 fetchOffset[tid] = 0; 311 macroop[tid] = NULL;
| 321 pc[tid] = cpu->pcState(tid); 322 fetchOffset[tid] = 0; 323 macroop[tid] = NULL;
|
312 delayedCommit[tid] = false; 313 }
| |
314
| 324
|
315 for (ThreadID tid = 0; tid < numThreads; tid++) { 316 317 fetchStatus[tid] = Running; 318 319 priorityList.push_back(tid); 320
| 325 delayedCommit[tid] = false;
|
321 memReq[tid] = NULL; 322 323 stalls[tid].decode = false; 324 stalls[tid].rename = false; 325 stalls[tid].iew = false; 326 stalls[tid].commit = false;
| 326 memReq[tid] = NULL; 327 328 stalls[tid].decode = false; 329 stalls[tid].rename = false; 330 stalls[tid].iew = false; 331 stalls[tid].commit = false;
|
| 332 stalls[tid].drain = false; 333 334 priorityList.push_back(tid);
|
327 } 328
| 335 } 336
|
329 // Schedule fetch to get the correct PC from the CPU 330 // scheduleFetchStartupEvent(1);
| 337 wroteToTimeBuffer = false; 338 _status = Inactive;
|
331
| 339
|
332 // Fetch needs to start fetching instructions at the very beginning, 333 // so it must start up in active state. 334 switchToActive();
| 340 // this CPU could still be unconnected if we are restoring from a 341 // checkpoint and this CPU is to be switched in, thus we can only 342 // do this here if the instruction port is actually connected, if 343 // not we have to do it as part of takeOverFrom. 344 if (cpu->getInstPort().isConnected()) 345 setIcache();
|
335} 336 337template<class Impl> 338void 339DefaultFetch<Impl>::setIcache() 340{ 341 assert(cpu->getInstPort().isConnected()); 342 343 // Size of cache block. 344 cacheBlkSize = cpu->getInstPort().peerBlockSize(); 345 346 // Create mask to get rid of offset bits. 347 cacheBlkMask = (cacheBlkSize - 1); 348 349 for (ThreadID tid = 0; tid < numThreads; tid++) { 350 // Create space to store a cache line. 351 if (!cacheData[tid]) 352 cacheData[tid] = new uint8_t[cacheBlkSize]; 353 cacheDataPC[tid] = 0; 354 cacheDataValid[tid] = false; 355 } 356} 357 358template<class Impl> 359void 360DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 361{ 362 ThreadID tid = pkt->req->threadId(); 363 364 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n", tid);
| 346} 347 348template<class Impl> 349void 350DefaultFetch<Impl>::setIcache() 351{ 352 assert(cpu->getInstPort().isConnected()); 353 354 // Size of cache block. 355 cacheBlkSize = cpu->getInstPort().peerBlockSize(); 356 357 // Create mask to get rid of offset bits. 358 cacheBlkMask = (cacheBlkSize - 1); 359 360 for (ThreadID tid = 0; tid < numThreads; tid++) { 361 // Create space to store a cache line. 362 if (!cacheData[tid]) 363 cacheData[tid] = new uint8_t[cacheBlkSize]; 364 cacheDataPC[tid] = 0; 365 cacheDataValid[tid] = false; 366 } 367} 368 369template<class Impl> 370void 371DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 372{ 373 ThreadID tid = pkt->req->threadId(); 374 375 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n", tid);
|
| 376 assert(!cpu->switchedOut());
|
365 366 // Only change the status if it's still waiting on the icache access 367 // to return. 368 if (fetchStatus[tid] != IcacheWaitResponse ||
| 377 378 // Only change the status if it's still waiting on the icache access 379 // to return. 380 if (fetchStatus[tid] != IcacheWaitResponse ||
|
369 pkt->req != memReq[tid] || 370 isSwitchedOut()) {
| 381 pkt->req != memReq[tid]) {
|
371 ++fetchIcacheSquashes; 372 delete pkt->req; 373 delete pkt; 374 return; 375 } 376 377 memcpy(cacheData[tid], pkt->getPtr<uint8_t>(), cacheBlkSize); 378 cacheDataValid[tid] = true; 379
| 382 ++fetchIcacheSquashes; 383 delete pkt->req; 384 delete pkt; 385 return; 386 } 387 388 memcpy(cacheData[tid], pkt->getPtr<uint8_t>(), cacheBlkSize); 389 cacheDataValid[tid] = true; 390
|
380 if (!drainPending) { 381 // Wake up the CPU (if it went to sleep and was waiting on 382 // this completion event). 383 cpu->wakeCPU();
| 391 // Wake up the CPU (if it went to sleep and was waiting on 392 // this completion event). 393 cpu->wakeCPU();
|
384
| 394
|
385 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 386 tid);
| 395 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 396 tid);
|
387
| 397
|
388 switchToActive(); 389 }
| 398 switchToActive();
|
390 391 // Only switch to IcacheAccessComplete if we're not stalled as well. 392 if (checkStall(tid)) { 393 fetchStatus[tid] = Blocked; 394 } else { 395 fetchStatus[tid] = IcacheAccessComplete; 396 } 397 398 // Reset the mem req to NULL. 399 delete pkt->req; 400 delete pkt; 401 memReq[tid] = NULL; 402} 403 404template <class Impl>
| 399 400 // Only switch to IcacheAccessComplete if we're not stalled as well. 401 if (checkStall(tid)) { 402 fetchStatus[tid] = Blocked; 403 } else { 404 fetchStatus[tid] = IcacheAccessComplete; 405 } 406 407 // Reset the mem req to NULL. 408 delete pkt->req; 409 delete pkt; 410 memReq[tid] = NULL; 411} 412 413template <class Impl>
|
405bool 406DefaultFetch::drain()
| 414void 415DefaultFetch<Impl>::drainResume()
|
407{
| 416{
|
408 // Fetch is ready to drain at any time. 409 cpu->signalDrained(); 410 drainPending = true; 411 return true;
| 417 for (ThreadID i = 0; i < Impl::MaxThreads; ++i) 418 stalls[i].drain = false;
|
412} 413 414template <class Impl> 415void
| 419} 420 421template <class Impl> 422void
|
416DefaultFetch<Impl>::resume()
| 423DefaultFetch<Impl>::drainSanityCheck() const
|
417{
| 424{
|
418 drainPending = false;
| 425 assert(isDrained()); 426 assert(retryPkt == NULL); 427 assert(retryTid == InvalidThreadID); 428 assert(cacheBlocked == false); 429 assert(interruptPending == false); 430 431 for (ThreadID i = 0; i < numThreads; ++i) { 432 assert(!memReq[i]); 433 assert(!stalls[i].decode); 434 assert(!stalls[i].rename); 435 assert(!stalls[i].iew); 436 assert(!stalls[i].commit); 437 assert(fetchStatus[i] == Idle || stalls[i].drain); 438 } 439 440 branchPred.drainSanityCheck();
|
419} 420 421template <class Impl>
| 441} 442 443template <class Impl>
|
422void 423DefaultFetch<Impl>::switchOut()
| 444bool 445DefaultFetch<Impl>::isDrained() const
|
424{
| 446{
|
425 switchedOut = true; 426 // Branch predictor needs to have its state cleared. 427 branchPred.switchOut();
| 447 /* Make sure that threads are either idle of that the commit stage 448 * has signaled that draining has completed by setting the drain 449 * stall flag. This effectively forces the pipeline to be disabled 450 * until the whole system is drained (simulation may continue to 451 * drain other components). 452 */ 453 for (ThreadID i = 0; i < numThreads; ++i) { 454 if (!(fetchStatus[i] == Idle || 455 (fetchStatus[i] == Blocked && stalls[i].drain))) 456 return false; 457 } 458 459 /* The pipeline might start up again in the middle of the drain 460 * cycle if the finish translation event is scheduled, so make 461 * sure that's not the case. 462 */ 463 return !finishTranslationEvent.scheduled();
|
428} 429 430template <class Impl> 431void 432DefaultFetch<Impl>::takeOverFrom() 433{
| 464} 465 466template <class Impl> 467void 468DefaultFetch<Impl>::takeOverFrom() 469{
|
434 // the instruction port is now connected so we can get the block 435 // size 436 setIcache();
| 470 assert(cpu->getInstPort().isConnected()); 471 resetStage();
|
437
| 472
|
438 // Reset all state 439 for (ThreadID i = 0; i < Impl::MaxThreads; ++i) { 440 stalls[i].decode = 0; 441 stalls[i].rename = 0; 442 stalls[i].iew = 0; 443 stalls[i].commit = 0; 444 pc[i] = cpu->pcState(i); 445 fetchStatus[i] = Running; 446 } 447 numInst = 0; 448 wroteToTimeBuffer = false; 449 _status = Inactive; 450 switchedOut = false; 451 interruptPending = false;
| |
452 branchPred.takeOverFrom(); 453} 454 455template <class Impl> 456void
| 473 branchPred.takeOverFrom(); 474} 475 476template <class Impl> 477void
|
| 478DefaultFetch<Impl>::drainStall(ThreadID tid) 479{ 480 assert(cpu->isDraining()); 481 assert(!stalls[tid].drain); 482 DPRINTF(Drain, "%i: Thread drained.\n", tid); 483 stalls[tid].drain = true; 484} 485 486template <class Impl> 487void
|
457DefaultFetch<Impl>::wakeFromQuiesce() 458{ 459 DPRINTF(Fetch, "Waking up from quiesce\n"); 460 // Hopefully this is safe 461 // @todo: Allow other threads to wake from quiesce. 462 fetchStatus[0] = Running; 463} 464 465template <class Impl> 466inline void 467DefaultFetch<Impl>::switchToActive() 468{ 469 if (_status == Inactive) { 470 DPRINTF(Activity, "Activating stage.\n"); 471 472 cpu->activateStage(O3CPU::FetchIdx); 473 474 _status = Active; 475 } 476} 477 478template <class Impl> 479inline void 480DefaultFetch<Impl>::switchToInactive() 481{ 482 if (_status == Active) { 483 DPRINTF(Activity, "Deactivating stage.\n"); 484 485 cpu->deactivateStage(O3CPU::FetchIdx); 486 487 _status = Inactive; 488 } 489} 490 491template <class Impl> 492bool 493DefaultFetch<Impl>::lookupAndUpdateNextPC( 494 DynInstPtr &inst, TheISA::PCState &nextPC) 495{ 496 // Do branch prediction check here. 497 // A bit of a misnomer...next_PC is actually the current PC until 498 // this function updates it. 499 bool predict_taken; 500 501 if (!inst->isControl()) { 502 TheISA::advancePC(nextPC, inst->staticInst); 503 inst->setPredTarg(nextPC); 504 inst->setPredTaken(false); 505 return false; 506 } 507 508 ThreadID tid = inst->threadNumber; 509 predict_taken = branchPred.predict(inst, nextPC, tid); 510 511 if (predict_taken) { 512 DPRINTF(Fetch, "[tid:%i]: [sn:%i]: Branch predicted to be taken to %s.\n", 513 tid, inst->seqNum, nextPC); 514 } else { 515 DPRINTF(Fetch, "[tid:%i]: [sn:%i]:Branch predicted to be not taken.\n", 516 tid, inst->seqNum); 517 } 518 519 DPRINTF(Fetch, "[tid:%i]: [sn:%i] Branch predicted to go to %s.\n", 520 tid, inst->seqNum, nextPC); 521 inst->setPredTarg(nextPC); 522 inst->setPredTaken(predict_taken); 523 524 ++fetchedBranches; 525 526 if (predict_taken) { 527 ++predictedBranches; 528 } 529 530 return predict_taken; 531} 532 533template <class Impl> 534bool 535DefaultFetch<Impl>::fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc) 536{ 537 Fault fault = NoFault; 538
| 488DefaultFetch<Impl>::wakeFromQuiesce() 489{ 490 DPRINTF(Fetch, "Waking up from quiesce\n"); 491 // Hopefully this is safe 492 // @todo: Allow other threads to wake from quiesce. 493 fetchStatus[0] = Running; 494} 495 496template <class Impl> 497inline void 498DefaultFetch<Impl>::switchToActive() 499{ 500 if (_status == Inactive) { 501 DPRINTF(Activity, "Activating stage.\n"); 502 503 cpu->activateStage(O3CPU::FetchIdx); 504 505 _status = Active; 506 } 507} 508 509template <class Impl> 510inline void 511DefaultFetch<Impl>::switchToInactive() 512{ 513 if (_status == Active) { 514 DPRINTF(Activity, "Deactivating stage.\n"); 515 516 cpu->deactivateStage(O3CPU::FetchIdx); 517 518 _status = Inactive; 519 } 520} 521 522template <class Impl> 523bool 524DefaultFetch<Impl>::lookupAndUpdateNextPC( 525 DynInstPtr &inst, TheISA::PCState &nextPC) 526{ 527 // Do branch prediction check here. 528 // A bit of a misnomer...next_PC is actually the current PC until 529 // this function updates it. 530 bool predict_taken; 531 532 if (!inst->isControl()) { 533 TheISA::advancePC(nextPC, inst->staticInst); 534 inst->setPredTarg(nextPC); 535 inst->setPredTaken(false); 536 return false; 537 } 538 539 ThreadID tid = inst->threadNumber; 540 predict_taken = branchPred.predict(inst, nextPC, tid); 541 542 if (predict_taken) { 543 DPRINTF(Fetch, "[tid:%i]: [sn:%i]: Branch predicted to be taken to %s.\n", 544 tid, inst->seqNum, nextPC); 545 } else { 546 DPRINTF(Fetch, "[tid:%i]: [sn:%i]:Branch predicted to be not taken.\n", 547 tid, inst->seqNum); 548 } 549 550 DPRINTF(Fetch, "[tid:%i]: [sn:%i] Branch predicted to go to %s.\n", 551 tid, inst->seqNum, nextPC); 552 inst->setPredTarg(nextPC); 553 inst->setPredTaken(predict_taken); 554 555 ++fetchedBranches; 556 557 if (predict_taken) { 558 ++predictedBranches; 559 } 560 561 return predict_taken; 562} 563 564template <class Impl> 565bool 566DefaultFetch<Impl>::fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc) 567{ 568 Fault fault = NoFault; 569
|
| 570 assert(!cpu->switchedOut()); 571
|
539 // @todo: not sure if these should block translation. 540 //AlphaDep 541 if (cacheBlocked) { 542 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, cache blocked\n", 543 tid); 544 return false;
| 572 // @todo: not sure if these should block translation. 573 //AlphaDep 574 if (cacheBlocked) { 575 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, cache blocked\n", 576 tid); 577 return false;
|
545 } else if (isSwitchedOut()) { 546 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, switched out\n", 547 tid); 548 return false;
| |
549 } else if (checkInterrupt(pc) && !delayedCommit[tid]) { 550 // Hold off fetch from getting new instructions when: 551 // Cache is blocked, or 552 // while an interrupt is pending and we're not in PAL mode, or 553 // fetch is switched out. 554 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, interrupt pending\n", 555 tid); 556 return false; 557 } 558 559 // Align the fetch address so it's at the start of a cache block. 560 Addr block_PC = icacheBlockAlignPC(vaddr); 561 562 DPRINTF(Fetch, "[tid:%i] Fetching cache line %#x for addr %#x\n", 563 tid, block_PC, vaddr); 564 565 // Setup the memReq to do a read of the first instruction's address. 566 // Set the appropriate read size and flags as well. 567 // Build request here. 568 RequestPtr mem_req = 569 new Request(tid, block_PC, cacheBlkSize, Request::INST_FETCH, 570 cpu->instMasterId(), pc, cpu->thread[tid]->contextId(), tid); 571 572 memReq[tid] = mem_req; 573 574 // Initiate translation of the icache block 575 fetchStatus[tid] = ItlbWait; 576 FetchTranslation *trans = new FetchTranslation(this); 577 cpu->itb->translateTiming(mem_req, cpu->thread[tid]->getTC(), 578 trans, BaseTLB::Execute); 579 return true; 580} 581 582template <class Impl> 583void 584DefaultFetch<Impl>::finishTranslation(Fault fault, RequestPtr mem_req) 585{ 586 ThreadID tid = mem_req->threadId(); 587 Addr block_PC = mem_req->getVaddr(); 588
| 578 } else if (checkInterrupt(pc) && !delayedCommit[tid]) { 579 // Hold off fetch from getting new instructions when: 580 // Cache is blocked, or 581 // while an interrupt is pending and we're not in PAL mode, or 582 // fetch is switched out. 583 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, interrupt pending\n", 584 tid); 585 return false; 586 } 587 588 // Align the fetch address so it's at the start of a cache block. 589 Addr block_PC = icacheBlockAlignPC(vaddr); 590 591 DPRINTF(Fetch, "[tid:%i] Fetching cache line %#x for addr %#x\n", 592 tid, block_PC, vaddr); 593 594 // Setup the memReq to do a read of the first instruction's address. 595 // Set the appropriate read size and flags as well. 596 // Build request here. 597 RequestPtr mem_req = 598 new Request(tid, block_PC, cacheBlkSize, Request::INST_FETCH, 599 cpu->instMasterId(), pc, cpu->thread[tid]->contextId(), tid); 600 601 memReq[tid] = mem_req; 602 603 // Initiate translation of the icache block 604 fetchStatus[tid] = ItlbWait; 605 FetchTranslation *trans = new FetchTranslation(this); 606 cpu->itb->translateTiming(mem_req, cpu->thread[tid]->getTC(), 607 trans, BaseTLB::Execute); 608 return true; 609} 610 611template <class Impl> 612void 613DefaultFetch<Impl>::finishTranslation(Fault fault, RequestPtr mem_req) 614{ 615 ThreadID tid = mem_req->threadId(); 616 Addr block_PC = mem_req->getVaddr(); 617
|
| 618 assert(!cpu->switchedOut()); 619
|
589 // Wake up CPU if it was idle 590 cpu->wakeCPU(); 591 592 if (fetchStatus[tid] != ItlbWait || mem_req != memReq[tid] ||
| 620 // Wake up CPU if it was idle 621 cpu->wakeCPU(); 622 623 if (fetchStatus[tid] != ItlbWait || mem_req != memReq[tid] ||
|
593 mem_req->getVaddr() != memReq[tid]->getVaddr() || isSwitchedOut()) {
| 624 mem_req->getVaddr() != memReq[tid]->getVaddr()) {
|
594 DPRINTF(Fetch, "[tid:%i] Ignoring itlb completed after squash\n", 595 tid); 596 ++fetchTlbSquashes; 597 delete mem_req; 598 return; 599 } 600 601 602 // If translation was successful, attempt to read the icache block. 603 if (fault == NoFault) { 604 // Check that we're not going off into random memory 605 // If we have, just wait around for commit to squash something and put 606 // us on the right track 607 if (!cpu->system->isMemAddr(mem_req->getPaddr())) { 608 warn("Address %#x is outside of physical memory, stopping fetch\n", 609 mem_req->getPaddr()); 610 fetchStatus[tid] = NoGoodAddr; 611 delete mem_req; 612 memReq[tid] = NULL; 613 return; 614 } 615 616 // Build packet here. 617 PacketPtr data_pkt = new Packet(mem_req, MemCmd::ReadReq); 618 data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]); 619 620 cacheDataPC[tid] = block_PC; 621 cacheDataValid[tid] = false; 622 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 623 624 fetchedCacheLines++; 625 626 // Access the cache. 627 if (!cpu->getInstPort().sendTimingReq(data_pkt)) { 628 assert(retryPkt == NULL); 629 assert(retryTid == InvalidThreadID); 630 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 631 632 fetchStatus[tid] = IcacheWaitRetry; 633 retryPkt = data_pkt; 634 retryTid = tid; 635 cacheBlocked = true; 636 } else { 637 DPRINTF(Fetch, "[tid:%i]: Doing Icache access.\n", tid); 638 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 639 "response.\n", tid); 640 641 lastIcacheStall[tid] = curTick(); 642 fetchStatus[tid] = IcacheWaitResponse; 643 } 644 } else { 645 if (!(numInst < fetchWidth)) { 646 assert(!finishTranslationEvent.scheduled()); 647 finishTranslationEvent.setFault(fault); 648 finishTranslationEvent.setReq(mem_req); 649 cpu->schedule(finishTranslationEvent, 650 cpu->clockEdge(Cycles(1))); 651 return; 652 } 653 DPRINTF(Fetch, "[tid:%i] Got back req with addr %#x but expected %#x\n", 654 tid, mem_req->getVaddr(), memReq[tid]->getVaddr()); 655 // Translation faulted, icache request won't be sent. 656 delete mem_req; 657 memReq[tid] = NULL; 658 659 // Send the fault to commit. This thread will not do anything 660 // until commit handles the fault. The only other way it can 661 // wake up is if a squash comes along and changes the PC. 662 TheISA::PCState fetchPC = pc[tid]; 663 664 DPRINTF(Fetch, "[tid:%i]: Translation faulted, building noop.\n", tid); 665 // We will use a nop in ordier to carry the fault. 666 DynInstPtr instruction = buildInst(tid, 667 decoder[tid]->decode(TheISA::NoopMachInst, fetchPC.instAddr()), 668 NULL, fetchPC, fetchPC, false); 669 670 instruction->setPredTarg(fetchPC); 671 instruction->fault = fault; 672 wroteToTimeBuffer = true; 673 674 DPRINTF(Activity, "Activity this cycle.\n"); 675 cpu->activityThisCycle(); 676 677 fetchStatus[tid] = TrapPending; 678 679 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n", tid); 680 DPRINTF(Fetch, "[tid:%i]: fault (%s) detected @ PC %s.\n", 681 tid, fault->name(), pc[tid]); 682 } 683 _status = updateFetchStatus(); 684} 685 686template <class Impl> 687inline void 688DefaultFetch<Impl>::doSquash(const TheISA::PCState &newPC, 689 const DynInstPtr squashInst, ThreadID tid) 690{ 691 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %s.\n", 692 tid, newPC); 693 694 pc[tid] = newPC; 695 fetchOffset[tid] = 0; 696 if (squashInst && squashInst->pcState().instAddr() == newPC.instAddr()) 697 macroop[tid] = squashInst->macroop; 698 else 699 macroop[tid] = NULL; 700 decoder[tid]->reset(); 701 702 // Clear the icache miss if it's outstanding. 703 if (fetchStatus[tid] == IcacheWaitResponse) { 704 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 705 tid); 706 memReq[tid] = NULL; 707 } else if (fetchStatus[tid] == ItlbWait) { 708 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding ITLB miss.\n", 709 tid); 710 memReq[tid] = NULL; 711 } 712 713 // Get rid of the retrying packet if it was from this thread. 714 if (retryTid == tid) { 715 assert(cacheBlocked); 716 if (retryPkt) { 717 delete retryPkt->req; 718 delete retryPkt; 719 } 720 retryPkt = NULL; 721 retryTid = InvalidThreadID; 722 } 723 724 fetchStatus[tid] = Squashing; 725 726 // microops are being squashed, it is not known wheather the 727 // youngest non-squashed microop was marked delayed commit 728 // or not. Setting the flag to true ensures that the 729 // interrupts are not handled when they cannot be, though 730 // some opportunities to handle interrupts may be missed. 731 delayedCommit[tid] = true; 732 733 ++fetchSquashCycles; 734} 735 736template<class Impl> 737void 738DefaultFetch<Impl>::squashFromDecode(const TheISA::PCState &newPC, 739 const DynInstPtr squashInst, 740 const InstSeqNum seq_num, ThreadID tid) 741{ 742 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n", tid); 743 744 doSquash(newPC, squashInst, tid); 745 746 // Tell the CPU to remove any instructions that are in flight between 747 // fetch and decode. 748 cpu->removeInstsUntil(seq_num, tid); 749} 750 751template<class Impl> 752bool 753DefaultFetch<Impl>::checkStall(ThreadID tid) const 754{ 755 bool ret_val = false; 756 757 if (cpu->contextSwitch) { 758 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 759 ret_val = true;
| 625 DPRINTF(Fetch, "[tid:%i] Ignoring itlb completed after squash\n", 626 tid); 627 ++fetchTlbSquashes; 628 delete mem_req; 629 return; 630 } 631 632 633 // If translation was successful, attempt to read the icache block. 634 if (fault == NoFault) { 635 // Check that we're not going off into random memory 636 // If we have, just wait around for commit to squash something and put 637 // us on the right track 638 if (!cpu->system->isMemAddr(mem_req->getPaddr())) { 639 warn("Address %#x is outside of physical memory, stopping fetch\n", 640 mem_req->getPaddr()); 641 fetchStatus[tid] = NoGoodAddr; 642 delete mem_req; 643 memReq[tid] = NULL; 644 return; 645 } 646 647 // Build packet here. 648 PacketPtr data_pkt = new Packet(mem_req, MemCmd::ReadReq); 649 data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]); 650 651 cacheDataPC[tid] = block_PC; 652 cacheDataValid[tid] = false; 653 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 654 655 fetchedCacheLines++; 656 657 // Access the cache. 658 if (!cpu->getInstPort().sendTimingReq(data_pkt)) { 659 assert(retryPkt == NULL); 660 assert(retryTid == InvalidThreadID); 661 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 662 663 fetchStatus[tid] = IcacheWaitRetry; 664 retryPkt = data_pkt; 665 retryTid = tid; 666 cacheBlocked = true; 667 } else { 668 DPRINTF(Fetch, "[tid:%i]: Doing Icache access.\n", tid); 669 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 670 "response.\n", tid); 671 672 lastIcacheStall[tid] = curTick(); 673 fetchStatus[tid] = IcacheWaitResponse; 674 } 675 } else { 676 if (!(numInst < fetchWidth)) { 677 assert(!finishTranslationEvent.scheduled()); 678 finishTranslationEvent.setFault(fault); 679 finishTranslationEvent.setReq(mem_req); 680 cpu->schedule(finishTranslationEvent, 681 cpu->clockEdge(Cycles(1))); 682 return; 683 } 684 DPRINTF(Fetch, "[tid:%i] Got back req with addr %#x but expected %#x\n", 685 tid, mem_req->getVaddr(), memReq[tid]->getVaddr()); 686 // Translation faulted, icache request won't be sent. 687 delete mem_req; 688 memReq[tid] = NULL; 689 690 // Send the fault to commit. This thread will not do anything 691 // until commit handles the fault. The only other way it can 692 // wake up is if a squash comes along and changes the PC. 693 TheISA::PCState fetchPC = pc[tid]; 694 695 DPRINTF(Fetch, "[tid:%i]: Translation faulted, building noop.\n", tid); 696 // We will use a nop in ordier to carry the fault. 697 DynInstPtr instruction = buildInst(tid, 698 decoder[tid]->decode(TheISA::NoopMachInst, fetchPC.instAddr()), 699 NULL, fetchPC, fetchPC, false); 700 701 instruction->setPredTarg(fetchPC); 702 instruction->fault = fault; 703 wroteToTimeBuffer = true; 704 705 DPRINTF(Activity, "Activity this cycle.\n"); 706 cpu->activityThisCycle(); 707 708 fetchStatus[tid] = TrapPending; 709 710 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n", tid); 711 DPRINTF(Fetch, "[tid:%i]: fault (%s) detected @ PC %s.\n", 712 tid, fault->name(), pc[tid]); 713 } 714 _status = updateFetchStatus(); 715} 716 717template <class Impl> 718inline void 719DefaultFetch<Impl>::doSquash(const TheISA::PCState &newPC, 720 const DynInstPtr squashInst, ThreadID tid) 721{ 722 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %s.\n", 723 tid, newPC); 724 725 pc[tid] = newPC; 726 fetchOffset[tid] = 0; 727 if (squashInst && squashInst->pcState().instAddr() == newPC.instAddr()) 728 macroop[tid] = squashInst->macroop; 729 else 730 macroop[tid] = NULL; 731 decoder[tid]->reset(); 732 733 // Clear the icache miss if it's outstanding. 734 if (fetchStatus[tid] == IcacheWaitResponse) { 735 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 736 tid); 737 memReq[tid] = NULL; 738 } else if (fetchStatus[tid] == ItlbWait) { 739 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding ITLB miss.\n", 740 tid); 741 memReq[tid] = NULL; 742 } 743 744 // Get rid of the retrying packet if it was from this thread. 745 if (retryTid == tid) { 746 assert(cacheBlocked); 747 if (retryPkt) { 748 delete retryPkt->req; 749 delete retryPkt; 750 } 751 retryPkt = NULL; 752 retryTid = InvalidThreadID; 753 } 754 755 fetchStatus[tid] = Squashing; 756 757 // microops are being squashed, it is not known wheather the 758 // youngest non-squashed microop was marked delayed commit 759 // or not. Setting the flag to true ensures that the 760 // interrupts are not handled when they cannot be, though 761 // some opportunities to handle interrupts may be missed. 762 delayedCommit[tid] = true; 763 764 ++fetchSquashCycles; 765} 766 767template<class Impl> 768void 769DefaultFetch<Impl>::squashFromDecode(const TheISA::PCState &newPC, 770 const DynInstPtr squashInst, 771 const InstSeqNum seq_num, ThreadID tid) 772{ 773 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n", tid); 774 775 doSquash(newPC, squashInst, tid); 776 777 // Tell the CPU to remove any instructions that are in flight between 778 // fetch and decode. 779 cpu->removeInstsUntil(seq_num, tid); 780} 781 782template<class Impl> 783bool 784DefaultFetch<Impl>::checkStall(ThreadID tid) const 785{ 786 bool ret_val = false; 787 788 if (cpu->contextSwitch) { 789 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 790 ret_val = true;
|
| 791 } else if (stalls[tid].drain) { 792 assert(cpu->isDraining()); 793 DPRINTF(Fetch,"[tid:%i]: Drain stall detected.\n",tid); 794 ret_val = true;
|
760 } else if (stalls[tid].decode) { 761 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 762 ret_val = true; 763 } else if (stalls[tid].rename) { 764 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 765 ret_val = true; 766 } else if (stalls[tid].iew) { 767 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 768 ret_val = true; 769 } else if (stalls[tid].commit) { 770 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 771 ret_val = true; 772 } 773 774 return ret_val; 775} 776 777template<class Impl> 778typename DefaultFetch<Impl>::FetchStatus 779DefaultFetch<Impl>::updateFetchStatus() 780{ 781 //Check Running 782 list<ThreadID>::iterator threads = activeThreads->begin(); 783 list<ThreadID>::iterator end = activeThreads->end(); 784 785 while (threads != end) { 786 ThreadID tid = *threads++; 787 788 if (fetchStatus[tid] == Running || 789 fetchStatus[tid] == Squashing || 790 fetchStatus[tid] == IcacheAccessComplete) { 791 792 if (_status == Inactive) { 793 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 794 795 if (fetchStatus[tid] == IcacheAccessComplete) { 796 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 797 "completion\n",tid); 798 } 799 800 cpu->activateStage(O3CPU::FetchIdx); 801 } 802 803 return Active; 804 } 805 } 806 807 // Stage is switching from active to inactive, notify CPU of it. 808 if (_status == Active) { 809 DPRINTF(Activity, "Deactivating stage.\n"); 810 811 cpu->deactivateStage(O3CPU::FetchIdx); 812 } 813 814 return Inactive; 815} 816 817template <class Impl> 818void 819DefaultFetch<Impl>::squash(const TheISA::PCState &newPC, 820 const InstSeqNum seq_num, DynInstPtr squashInst, 821 ThreadID tid) 822{ 823 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n", tid); 824 825 doSquash(newPC, squashInst, tid); 826 827 // Tell the CPU to remove any instructions that are not in the ROB. 828 cpu->removeInstsNotInROB(tid); 829} 830 831template <class Impl> 832void 833DefaultFetch<Impl>::tick() 834{ 835 list<ThreadID>::iterator threads = activeThreads->begin(); 836 list<ThreadID>::iterator end = activeThreads->end(); 837 bool status_change = false; 838 839 wroteToTimeBuffer = false; 840 841 for (ThreadID i = 0; i < Impl::MaxThreads; ++i) { 842 issuePipelinedIfetch[i] = false; 843 } 844 845 while (threads != end) { 846 ThreadID tid = *threads++; 847 848 // Check the signals for each thread to determine the proper status 849 // for each thread. 850 bool updated_status = checkSignalsAndUpdate(tid); 851 status_change = status_change || updated_status; 852 } 853 854 DPRINTF(Fetch, "Running stage.\n"); 855 856 if (FullSystem) { 857 if (fromCommit->commitInfo[0].interruptPending) { 858 interruptPending = true; 859 } 860 861 if (fromCommit->commitInfo[0].clearInterrupt) { 862 interruptPending = false; 863 } 864 } 865 866 for (threadFetched = 0; threadFetched < numFetchingThreads; 867 threadFetched++) { 868 // Fetch each of the actively fetching threads. 869 fetch(status_change); 870 } 871 872 // Record number of instructions fetched this cycle for distribution. 873 fetchNisnDist.sample(numInst); 874 875 if (status_change) { 876 // Change the fetch stage status if there was a status change. 877 _status = updateFetchStatus(); 878 } 879 880 // If there was activity this cycle, inform the CPU of it. 881 if (wroteToTimeBuffer || cpu->contextSwitch) { 882 DPRINTF(Activity, "Activity this cycle.\n"); 883 884 cpu->activityThisCycle(); 885 } 886 887 // Issue the next I-cache request if possible. 888 for (ThreadID i = 0; i < Impl::MaxThreads; ++i) { 889 if (issuePipelinedIfetch[i]) { 890 pipelineIcacheAccesses(i); 891 } 892 } 893 894 // Reset the number of the instruction we've fetched. 895 numInst = 0; 896} 897 898template <class Impl> 899bool 900DefaultFetch<Impl>::checkSignalsAndUpdate(ThreadID tid) 901{ 902 // Update the per thread stall statuses. 903 if (fromDecode->decodeBlock[tid]) { 904 stalls[tid].decode = true; 905 } 906 907 if (fromDecode->decodeUnblock[tid]) { 908 assert(stalls[tid].decode); 909 assert(!fromDecode->decodeBlock[tid]); 910 stalls[tid].decode = false; 911 } 912 913 if (fromRename->renameBlock[tid]) { 914 stalls[tid].rename = true; 915 } 916 917 if (fromRename->renameUnblock[tid]) { 918 assert(stalls[tid].rename); 919 assert(!fromRename->renameBlock[tid]); 920 stalls[tid].rename = false; 921 } 922 923 if (fromIEW->iewBlock[tid]) { 924 stalls[tid].iew = true; 925 } 926 927 if (fromIEW->iewUnblock[tid]) { 928 assert(stalls[tid].iew); 929 assert(!fromIEW->iewBlock[tid]); 930 stalls[tid].iew = false; 931 } 932 933 if (fromCommit->commitBlock[tid]) { 934 stalls[tid].commit = true; 935 } 936 937 if (fromCommit->commitUnblock[tid]) { 938 assert(stalls[tid].commit); 939 assert(!fromCommit->commitBlock[tid]); 940 stalls[tid].commit = false; 941 } 942 943 // Check squash signals from commit. 944 if (fromCommit->commitInfo[tid].squash) { 945 946 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 947 "from commit.\n",tid); 948 // In any case, squash. 949 squash(fromCommit->commitInfo[tid].pc, 950 fromCommit->commitInfo[tid].doneSeqNum, 951 fromCommit->commitInfo[tid].squashInst, tid); 952 953 // If it was a branch mispredict on a control instruction, update the 954 // branch predictor with that instruction, otherwise just kill the 955 // invalid state we generated in after sequence number 956 if (fromCommit->commitInfo[tid].mispredictInst && 957 fromCommit->commitInfo[tid].mispredictInst->isControl()) { 958 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 959 fromCommit->commitInfo[tid].pc, 960 fromCommit->commitInfo[tid].branchTaken, 961 tid); 962 } else { 963 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 964 tid); 965 } 966 967 return true; 968 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 969 // Update the branch predictor if it wasn't a squashed instruction 970 // that was broadcasted. 971 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 972 } 973 974 // Check ROB squash signals from commit. 975 if (fromCommit->commitInfo[tid].robSquashing) { 976 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 977 978 // Continue to squash. 979 fetchStatus[tid] = Squashing; 980 981 return true; 982 } 983 984 // Check squash signals from decode. 985 if (fromDecode->decodeInfo[tid].squash) { 986 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 987 "from decode.\n",tid); 988 989 // Update the branch predictor. 990 if (fromDecode->decodeInfo[tid].branchMispredict) { 991 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 992 fromDecode->decodeInfo[tid].nextPC, 993 fromDecode->decodeInfo[tid].branchTaken, 994 tid); 995 } else { 996 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 997 tid); 998 } 999 1000 if (fetchStatus[tid] != Squashing) { 1001 1002 DPRINTF(Fetch, "Squashing from decode with PC = %s\n", 1003 fromDecode->decodeInfo[tid].nextPC); 1004 // Squash unless we're already squashing 1005 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 1006 fromDecode->decodeInfo[tid].squashInst, 1007 fromDecode->decodeInfo[tid].doneSeqNum, 1008 tid); 1009 1010 return true; 1011 } 1012 } 1013 1014 if (checkStall(tid) && 1015 fetchStatus[tid] != IcacheWaitResponse && 1016 fetchStatus[tid] != IcacheWaitRetry) { 1017 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 1018 1019 fetchStatus[tid] = Blocked; 1020 1021 return true; 1022 } 1023 1024 if (fetchStatus[tid] == Blocked || 1025 fetchStatus[tid] == Squashing) { 1026 // Switch status to running if fetch isn't being told to block or 1027 // squash this cycle. 1028 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 1029 tid); 1030 1031 fetchStatus[tid] = Running; 1032 1033 return true; 1034 } 1035 1036 // If we've reached this point, we have not gotten any signals that 1037 // cause fetch to change its status. Fetch remains the same as before. 1038 return false; 1039} 1040 1041template<class Impl> 1042typename Impl::DynInstPtr 1043DefaultFetch<Impl>::buildInst(ThreadID tid, StaticInstPtr staticInst, 1044 StaticInstPtr curMacroop, TheISA::PCState thisPC, 1045 TheISA::PCState nextPC, bool trace) 1046{ 1047 // Get a sequence number. 1048 InstSeqNum seq = cpu->getAndIncrementInstSeq(); 1049 1050 // Create a new DynInst from the instruction fetched. 1051 DynInstPtr instruction = 1052 new DynInst(staticInst, curMacroop, thisPC, nextPC, seq, cpu); 1053 instruction->setTid(tid); 1054 1055 instruction->setASID(tid); 1056 1057 instruction->setThreadState(cpu->thread[tid]); 1058 1059 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x (%d) created " 1060 "[sn:%lli].\n", tid, thisPC.instAddr(), 1061 thisPC.microPC(), seq); 1062 1063 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", tid, 1064 instruction->staticInst-> 1065 disassemble(thisPC.instAddr())); 1066 1067#if TRACING_ON 1068 if (trace) { 1069 instruction->traceData = 1070 cpu->getTracer()->getInstRecord(curTick(), cpu->tcBase(tid), 1071 instruction->staticInst, thisPC, curMacroop); 1072 } 1073#else 1074 instruction->traceData = NULL; 1075#endif 1076 1077 // Add instruction to the CPU's list of instructions. 1078 instruction->setInstListIt(cpu->addInst(instruction)); 1079 1080 // Write the instruction to the first slot in the queue 1081 // that heads to decode. 1082 assert(numInst < fetchWidth); 1083 toDecode->insts[toDecode->size++] = instruction; 1084 1085 // Keep track of if we can take an interrupt at this boundary 1086 delayedCommit[tid] = instruction->isDelayedCommit(); 1087 1088 return instruction; 1089} 1090 1091template<class Impl> 1092void 1093DefaultFetch<Impl>::fetch(bool &status_change) 1094{ 1095 ////////////////////////////////////////// 1096 // Start actual fetch 1097 ////////////////////////////////////////// 1098 ThreadID tid = getFetchingThread(fetchPolicy); 1099
| 795 } else if (stalls[tid].decode) { 796 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 797 ret_val = true; 798 } else if (stalls[tid].rename) { 799 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 800 ret_val = true; 801 } else if (stalls[tid].iew) { 802 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 803 ret_val = true; 804 } else if (stalls[tid].commit) { 805 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 806 ret_val = true; 807 } 808 809 return ret_val; 810} 811 812template<class Impl> 813typename DefaultFetch<Impl>::FetchStatus 814DefaultFetch<Impl>::updateFetchStatus() 815{ 816 //Check Running 817 list<ThreadID>::iterator threads = activeThreads->begin(); 818 list<ThreadID>::iterator end = activeThreads->end(); 819 820 while (threads != end) { 821 ThreadID tid = *threads++; 822 823 if (fetchStatus[tid] == Running || 824 fetchStatus[tid] == Squashing || 825 fetchStatus[tid] == IcacheAccessComplete) { 826 827 if (_status == Inactive) { 828 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 829 830 if (fetchStatus[tid] == IcacheAccessComplete) { 831 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 832 "completion\n",tid); 833 } 834 835 cpu->activateStage(O3CPU::FetchIdx); 836 } 837 838 return Active; 839 } 840 } 841 842 // Stage is switching from active to inactive, notify CPU of it. 843 if (_status == Active) { 844 DPRINTF(Activity, "Deactivating stage.\n"); 845 846 cpu->deactivateStage(O3CPU::FetchIdx); 847 } 848 849 return Inactive; 850} 851 852template <class Impl> 853void 854DefaultFetch<Impl>::squash(const TheISA::PCState &newPC, 855 const InstSeqNum seq_num, DynInstPtr squashInst, 856 ThreadID tid) 857{ 858 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n", tid); 859 860 doSquash(newPC, squashInst, tid); 861 862 // Tell the CPU to remove any instructions that are not in the ROB. 863 cpu->removeInstsNotInROB(tid); 864} 865 866template <class Impl> 867void 868DefaultFetch<Impl>::tick() 869{ 870 list<ThreadID>::iterator threads = activeThreads->begin(); 871 list<ThreadID>::iterator end = activeThreads->end(); 872 bool status_change = false; 873 874 wroteToTimeBuffer = false; 875 876 for (ThreadID i = 0; i < Impl::MaxThreads; ++i) { 877 issuePipelinedIfetch[i] = false; 878 } 879 880 while (threads != end) { 881 ThreadID tid = *threads++; 882 883 // Check the signals for each thread to determine the proper status 884 // for each thread. 885 bool updated_status = checkSignalsAndUpdate(tid); 886 status_change = status_change || updated_status; 887 } 888 889 DPRINTF(Fetch, "Running stage.\n"); 890 891 if (FullSystem) { 892 if (fromCommit->commitInfo[0].interruptPending) { 893 interruptPending = true; 894 } 895 896 if (fromCommit->commitInfo[0].clearInterrupt) { 897 interruptPending = false; 898 } 899 } 900 901 for (threadFetched = 0; threadFetched < numFetchingThreads; 902 threadFetched++) { 903 // Fetch each of the actively fetching threads. 904 fetch(status_change); 905 } 906 907 // Record number of instructions fetched this cycle for distribution. 908 fetchNisnDist.sample(numInst); 909 910 if (status_change) { 911 // Change the fetch stage status if there was a status change. 912 _status = updateFetchStatus(); 913 } 914 915 // If there was activity this cycle, inform the CPU of it. 916 if (wroteToTimeBuffer || cpu->contextSwitch) { 917 DPRINTF(Activity, "Activity this cycle.\n"); 918 919 cpu->activityThisCycle(); 920 } 921 922 // Issue the next I-cache request if possible. 923 for (ThreadID i = 0; i < Impl::MaxThreads; ++i) { 924 if (issuePipelinedIfetch[i]) { 925 pipelineIcacheAccesses(i); 926 } 927 } 928 929 // Reset the number of the instruction we've fetched. 930 numInst = 0; 931} 932 933template <class Impl> 934bool 935DefaultFetch<Impl>::checkSignalsAndUpdate(ThreadID tid) 936{ 937 // Update the per thread stall statuses. 938 if (fromDecode->decodeBlock[tid]) { 939 stalls[tid].decode = true; 940 } 941 942 if (fromDecode->decodeUnblock[tid]) { 943 assert(stalls[tid].decode); 944 assert(!fromDecode->decodeBlock[tid]); 945 stalls[tid].decode = false; 946 } 947 948 if (fromRename->renameBlock[tid]) { 949 stalls[tid].rename = true; 950 } 951 952 if (fromRename->renameUnblock[tid]) { 953 assert(stalls[tid].rename); 954 assert(!fromRename->renameBlock[tid]); 955 stalls[tid].rename = false; 956 } 957 958 if (fromIEW->iewBlock[tid]) { 959 stalls[tid].iew = true; 960 } 961 962 if (fromIEW->iewUnblock[tid]) { 963 assert(stalls[tid].iew); 964 assert(!fromIEW->iewBlock[tid]); 965 stalls[tid].iew = false; 966 } 967 968 if (fromCommit->commitBlock[tid]) { 969 stalls[tid].commit = true; 970 } 971 972 if (fromCommit->commitUnblock[tid]) { 973 assert(stalls[tid].commit); 974 assert(!fromCommit->commitBlock[tid]); 975 stalls[tid].commit = false; 976 } 977 978 // Check squash signals from commit. 979 if (fromCommit->commitInfo[tid].squash) { 980 981 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 982 "from commit.\n",tid); 983 // In any case, squash. 984 squash(fromCommit->commitInfo[tid].pc, 985 fromCommit->commitInfo[tid].doneSeqNum, 986 fromCommit->commitInfo[tid].squashInst, tid); 987 988 // If it was a branch mispredict on a control instruction, update the 989 // branch predictor with that instruction, otherwise just kill the 990 // invalid state we generated in after sequence number 991 if (fromCommit->commitInfo[tid].mispredictInst && 992 fromCommit->commitInfo[tid].mispredictInst->isControl()) { 993 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 994 fromCommit->commitInfo[tid].pc, 995 fromCommit->commitInfo[tid].branchTaken, 996 tid); 997 } else { 998 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 999 tid); 1000 } 1001 1002 return true; 1003 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 1004 // Update the branch predictor if it wasn't a squashed instruction 1005 // that was broadcasted. 1006 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 1007 } 1008 1009 // Check ROB squash signals from commit. 1010 if (fromCommit->commitInfo[tid].robSquashing) { 1011 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 1012 1013 // Continue to squash. 1014 fetchStatus[tid] = Squashing; 1015 1016 return true; 1017 } 1018 1019 // Check squash signals from decode. 1020 if (fromDecode->decodeInfo[tid].squash) { 1021 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 1022 "from decode.\n",tid); 1023 1024 // Update the branch predictor. 1025 if (fromDecode->decodeInfo[tid].branchMispredict) { 1026 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 1027 fromDecode->decodeInfo[tid].nextPC, 1028 fromDecode->decodeInfo[tid].branchTaken, 1029 tid); 1030 } else { 1031 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 1032 tid); 1033 } 1034 1035 if (fetchStatus[tid] != Squashing) { 1036 1037 DPRINTF(Fetch, "Squashing from decode with PC = %s\n", 1038 fromDecode->decodeInfo[tid].nextPC); 1039 // Squash unless we're already squashing 1040 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 1041 fromDecode->decodeInfo[tid].squashInst, 1042 fromDecode->decodeInfo[tid].doneSeqNum, 1043 tid); 1044 1045 return true; 1046 } 1047 } 1048 1049 if (checkStall(tid) && 1050 fetchStatus[tid] != IcacheWaitResponse && 1051 fetchStatus[tid] != IcacheWaitRetry) { 1052 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 1053 1054 fetchStatus[tid] = Blocked; 1055 1056 return true; 1057 } 1058 1059 if (fetchStatus[tid] == Blocked || 1060 fetchStatus[tid] == Squashing) { 1061 // Switch status to running if fetch isn't being told to block or 1062 // squash this cycle. 1063 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 1064 tid); 1065 1066 fetchStatus[tid] = Running; 1067 1068 return true; 1069 } 1070 1071 // If we've reached this point, we have not gotten any signals that 1072 // cause fetch to change its status. Fetch remains the same as before. 1073 return false; 1074} 1075 1076template<class Impl> 1077typename Impl::DynInstPtr 1078DefaultFetch<Impl>::buildInst(ThreadID tid, StaticInstPtr staticInst, 1079 StaticInstPtr curMacroop, TheISA::PCState thisPC, 1080 TheISA::PCState nextPC, bool trace) 1081{ 1082 // Get a sequence number. 1083 InstSeqNum seq = cpu->getAndIncrementInstSeq(); 1084 1085 // Create a new DynInst from the instruction fetched. 1086 DynInstPtr instruction = 1087 new DynInst(staticInst, curMacroop, thisPC, nextPC, seq, cpu); 1088 instruction->setTid(tid); 1089 1090 instruction->setASID(tid); 1091 1092 instruction->setThreadState(cpu->thread[tid]); 1093 1094 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x (%d) created " 1095 "[sn:%lli].\n", tid, thisPC.instAddr(), 1096 thisPC.microPC(), seq); 1097 1098 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", tid, 1099 instruction->staticInst-> 1100 disassemble(thisPC.instAddr())); 1101 1102#if TRACING_ON 1103 if (trace) { 1104 instruction->traceData = 1105 cpu->getTracer()->getInstRecord(curTick(), cpu->tcBase(tid), 1106 instruction->staticInst, thisPC, curMacroop); 1107 } 1108#else 1109 instruction->traceData = NULL; 1110#endif 1111 1112 // Add instruction to the CPU's list of instructions. 1113 instruction->setInstListIt(cpu->addInst(instruction)); 1114 1115 // Write the instruction to the first slot in the queue 1116 // that heads to decode. 1117 assert(numInst < fetchWidth); 1118 toDecode->insts[toDecode->size++] = instruction; 1119 1120 // Keep track of if we can take an interrupt at this boundary 1121 delayedCommit[tid] = instruction->isDelayedCommit(); 1122 1123 return instruction; 1124} 1125 1126template<class Impl> 1127void 1128DefaultFetch<Impl>::fetch(bool &status_change) 1129{ 1130 ////////////////////////////////////////// 1131 // Start actual fetch 1132 ////////////////////////////////////////// 1133 ThreadID tid = getFetchingThread(fetchPolicy); 1134
|
1100 if (tid == InvalidThreadID || drainPending) {
| 1135 assert(!cpu->switchedOut()); 1136 1137 if (tid == InvalidThreadID) {
|
1101 // Breaks looping condition in tick() 1102 threadFetched = numFetchingThreads; 1103 1104 if (numThreads == 1) { // @todo Per-thread stats 1105 profileStall(0); 1106 } 1107 1108 return; 1109 } 1110 1111 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 1112 1113 // The current PC. 1114 TheISA::PCState thisPC = pc[tid]; 1115 1116 Addr pcOffset = fetchOffset[tid]; 1117 Addr fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1118 1119 bool inRom = isRomMicroPC(thisPC.microPC()); 1120 1121 // If returning from the delay of a cache miss, then update the status 1122 // to running, otherwise do the cache access. Possibly move this up 1123 // to tick() function. 1124 if (fetchStatus[tid] == IcacheAccessComplete) { 1125 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", tid); 1126 1127 fetchStatus[tid] = Running; 1128 status_change = true; 1129 } else if (fetchStatus[tid] == Running) { 1130 // Align the fetch PC so its at the start of a cache block. 1131 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1132 1133 // If buffer is no longer valid or fetchAddr has moved to point 1134 // to the next cache block, AND we have no remaining ucode 1135 // from a macro-op, then start fetch from icache. 1136 if (!(cacheDataValid[tid] && block_PC == cacheDataPC[tid]) 1137 && !inRom && !macroop[tid]) { 1138 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 1139 "instruction, starting at PC %s.\n", tid, thisPC); 1140 1141 fetchCacheLine(fetchAddr, tid, thisPC.instAddr()); 1142 1143 if (fetchStatus[tid] == IcacheWaitResponse) 1144 ++icacheStallCycles; 1145 else if (fetchStatus[tid] == ItlbWait) 1146 ++fetchTlbCycles; 1147 else 1148 ++fetchMiscStallCycles; 1149 return;
| 1138 // Breaks looping condition in tick() 1139 threadFetched = numFetchingThreads; 1140 1141 if (numThreads == 1) { // @todo Per-thread stats 1142 profileStall(0); 1143 } 1144 1145 return; 1146 } 1147 1148 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 1149 1150 // The current PC. 1151 TheISA::PCState thisPC = pc[tid]; 1152 1153 Addr pcOffset = fetchOffset[tid]; 1154 Addr fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1155 1156 bool inRom = isRomMicroPC(thisPC.microPC()); 1157 1158 // If returning from the delay of a cache miss, then update the status 1159 // to running, otherwise do the cache access. Possibly move this up 1160 // to tick() function. 1161 if (fetchStatus[tid] == IcacheAccessComplete) { 1162 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", tid); 1163 1164 fetchStatus[tid] = Running; 1165 status_change = true; 1166 } else if (fetchStatus[tid] == Running) { 1167 // Align the fetch PC so its at the start of a cache block. 1168 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1169 1170 // If buffer is no longer valid or fetchAddr has moved to point 1171 // to the next cache block, AND we have no remaining ucode 1172 // from a macro-op, then start fetch from icache. 1173 if (!(cacheDataValid[tid] && block_PC == cacheDataPC[tid]) 1174 && !inRom && !macroop[tid]) { 1175 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 1176 "instruction, starting at PC %s.\n", tid, thisPC); 1177 1178 fetchCacheLine(fetchAddr, tid, thisPC.instAddr()); 1179 1180 if (fetchStatus[tid] == IcacheWaitResponse) 1181 ++icacheStallCycles; 1182 else if (fetchStatus[tid] == ItlbWait) 1183 ++fetchTlbCycles; 1184 else 1185 ++fetchMiscStallCycles; 1186 return;
|
1150 } else if ((checkInterrupt(thisPC.instAddr()) && !delayedCommit[tid]) 1151 || isSwitchedOut()) {
| 1187 } else if ((checkInterrupt(thisPC.instAddr()) && !delayedCommit[tid])) {
|
1152 // Stall CPU if an interrupt is posted and we're not issuing 1153 // an delayed commit micro-op currently (delayed commit instructions 1154 // are not interruptable by interrupts, only faults) 1155 ++fetchMiscStallCycles; 1156 DPRINTF(Fetch, "[tid:%i]: Fetch is stalled!\n", tid); 1157 return; 1158 } 1159 } else { 1160 if (fetchStatus[tid] == Idle) { 1161 ++fetchIdleCycles; 1162 DPRINTF(Fetch, "[tid:%i]: Fetch is idle!\n", tid); 1163 } 1164 1165 // Status is Idle, so fetch should do nothing. 1166 return; 1167 } 1168 1169 ++fetchCycles; 1170 1171 TheISA::PCState nextPC = thisPC; 1172 1173 StaticInstPtr staticInst = NULL; 1174 StaticInstPtr curMacroop = macroop[tid]; 1175 1176 // If the read of the first instruction was successful, then grab the 1177 // instructions from the rest of the cache line and put them into the 1178 // queue heading to decode. 1179 1180 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 1181 "decode.\n", tid); 1182 1183 // Need to keep track of whether or not a predicted branch 1184 // ended this fetch block. 1185 bool predictedBranch = false; 1186 1187 TheISA::MachInst *cacheInsts = 1188 reinterpret_cast<TheISA::MachInst *>(cacheData[tid]); 1189 1190 const unsigned numInsts = cacheBlkSize / instSize; 1191 unsigned blkOffset = (fetchAddr - cacheDataPC[tid]) / instSize; 1192 1193 // Loop through instruction memory from the cache. 1194 // Keep issuing while fetchWidth is available and branch is not 1195 // predicted taken 1196 while (numInst < fetchWidth && !predictedBranch) { 1197 1198 // We need to process more memory if we aren't going to get a 1199 // StaticInst from the rom, the current macroop, or what's already 1200 // in the decoder. 1201 bool needMem = !inRom && !curMacroop && 1202 !decoder[tid]->instReady(); 1203 fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1204 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1205 1206 if (needMem) { 1207 // If buffer is no longer valid or fetchAddr has moved to point 1208 // to the next cache block then start fetch from icache. 1209 if (!cacheDataValid[tid] || block_PC != cacheDataPC[tid]) 1210 break; 1211 1212 if (blkOffset >= numInsts) { 1213 // We need to process more memory, but we've run out of the 1214 // current block. 1215 break; 1216 } 1217 1218 if (ISA_HAS_DELAY_SLOT && pcOffset == 0) { 1219 // Walk past any annulled delay slot instructions. 1220 Addr pcAddr = thisPC.instAddr() & BaseCPU::PCMask; 1221 while (fetchAddr != pcAddr && blkOffset < numInsts) { 1222 blkOffset++; 1223 fetchAddr += instSize; 1224 } 1225 if (blkOffset >= numInsts) 1226 break; 1227 } 1228 1229 MachInst inst = TheISA::gtoh(cacheInsts[blkOffset]); 1230 decoder[tid]->moreBytes(thisPC, fetchAddr, inst); 1231 1232 if (decoder[tid]->needMoreBytes()) { 1233 blkOffset++; 1234 fetchAddr += instSize; 1235 pcOffset += instSize; 1236 } 1237 } 1238 1239 // Extract as many instructions and/or microops as we can from 1240 // the memory we've processed so far. 1241 do { 1242 if (!(curMacroop || inRom)) { 1243 if (decoder[tid]->instReady()) { 1244 staticInst = decoder[tid]->decode(thisPC); 1245 1246 // Increment stat of fetched instructions. 1247 ++fetchedInsts; 1248 1249 if (staticInst->isMacroop()) { 1250 curMacroop = staticInst; 1251 } else { 1252 pcOffset = 0; 1253 } 1254 } else { 1255 // We need more bytes for this instruction so blkOffset and 1256 // pcOffset will be updated 1257 break; 1258 } 1259 } 1260 // Whether we're moving to a new macroop because we're at the 1261 // end of the current one, or the branch predictor incorrectly 1262 // thinks we are... 1263 bool newMacro = false; 1264 if (curMacroop || inRom) { 1265 if (inRom) { 1266 staticInst = cpu->microcodeRom.fetchMicroop( 1267 thisPC.microPC(), curMacroop); 1268 } else { 1269 staticInst = curMacroop->fetchMicroop(thisPC.microPC()); 1270 } 1271 newMacro |= staticInst->isLastMicroop(); 1272 } 1273 1274 DynInstPtr instruction = 1275 buildInst(tid, staticInst, curMacroop, 1276 thisPC, nextPC, true); 1277 1278 numInst++; 1279 1280#if TRACING_ON 1281 instruction->fetchTick = curTick(); 1282#endif 1283 1284 nextPC = thisPC; 1285 1286 // If we're branching after this instruction, quite fetching 1287 // from the same block then. 1288 predictedBranch |= thisPC.branching(); 1289 predictedBranch |= 1290 lookupAndUpdateNextPC(instruction, nextPC); 1291 if (predictedBranch) { 1292 DPRINTF(Fetch, "Branch detected with PC = %s\n", thisPC); 1293 } 1294 1295 newMacro |= thisPC.instAddr() != nextPC.instAddr(); 1296 1297 // Move to the next instruction, unless we have a branch. 1298 thisPC = nextPC; 1299 inRom = isRomMicroPC(thisPC.microPC()); 1300 1301 if (newMacro) { 1302 fetchAddr = thisPC.instAddr() & BaseCPU::PCMask; 1303 blkOffset = (fetchAddr - cacheDataPC[tid]) / instSize; 1304 pcOffset = 0; 1305 curMacroop = NULL; 1306 } 1307 1308 if (instruction->isQuiesce()) { 1309 DPRINTF(Fetch, 1310 "Quiesce instruction encountered, halting fetch!"); 1311 fetchStatus[tid] = QuiescePending; 1312 status_change = true; 1313 break; 1314 } 1315 } while ((curMacroop || decoder[tid]->instReady()) && 1316 numInst < fetchWidth); 1317 } 1318 1319 if (predictedBranch) { 1320 DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch " 1321 "instruction encountered.\n", tid); 1322 } else if (numInst >= fetchWidth) { 1323 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth " 1324 "for this cycle.\n", tid); 1325 } else if (blkOffset >= cacheBlkSize) { 1326 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache " 1327 "block.\n", tid); 1328 } 1329 1330 macroop[tid] = curMacroop; 1331 fetchOffset[tid] = pcOffset; 1332 1333 if (numInst > 0) { 1334 wroteToTimeBuffer = true; 1335 } 1336 1337 pc[tid] = thisPC; 1338 1339 // pipeline a fetch if we're crossing a cache boundary and not in 1340 // a state that would preclude fetching 1341 fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1342 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1343 issuePipelinedIfetch[tid] = block_PC != cacheDataPC[tid] && 1344 fetchStatus[tid] != IcacheWaitResponse && 1345 fetchStatus[tid] != ItlbWait && 1346 fetchStatus[tid] != IcacheWaitRetry && 1347 fetchStatus[tid] != QuiescePending && 1348 !curMacroop; 1349} 1350 1351template<class Impl> 1352void 1353DefaultFetch<Impl>::recvRetry() 1354{ 1355 if (retryPkt != NULL) { 1356 assert(cacheBlocked); 1357 assert(retryTid != InvalidThreadID); 1358 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1359 1360 if (cpu->getInstPort().sendTimingReq(retryPkt)) { 1361 fetchStatus[retryTid] = IcacheWaitResponse; 1362 retryPkt = NULL; 1363 retryTid = InvalidThreadID; 1364 cacheBlocked = false; 1365 } 1366 } else { 1367 assert(retryTid == InvalidThreadID); 1368 // Access has been squashed since it was sent out. Just clear 1369 // the cache being blocked. 1370 cacheBlocked = false; 1371 } 1372} 1373 1374/////////////////////////////////////// 1375// // 1376// SMT FETCH POLICY MAINTAINED HERE // 1377// // 1378/////////////////////////////////////// 1379template<class Impl> 1380ThreadID 1381DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1382{ 1383 if (numThreads > 1) { 1384 switch (fetch_priority) { 1385 1386 case SingleThread: 1387 return 0; 1388 1389 case RoundRobin: 1390 return roundRobin(); 1391 1392 case IQ: 1393 return iqCount(); 1394 1395 case LSQ: 1396 return lsqCount(); 1397 1398 case Branch: 1399 return branchCount(); 1400 1401 default: 1402 return InvalidThreadID; 1403 } 1404 } else { 1405 list<ThreadID>::iterator thread = activeThreads->begin(); 1406 if (thread == activeThreads->end()) { 1407 return InvalidThreadID; 1408 } 1409 1410 ThreadID tid = *thread; 1411 1412 if (fetchStatus[tid] == Running || 1413 fetchStatus[tid] == IcacheAccessComplete || 1414 fetchStatus[tid] == Idle) { 1415 return tid; 1416 } else { 1417 return InvalidThreadID; 1418 } 1419 } 1420} 1421 1422 1423template<class Impl> 1424ThreadID 1425DefaultFetch<Impl>::roundRobin() 1426{ 1427 list<ThreadID>::iterator pri_iter = priorityList.begin(); 1428 list<ThreadID>::iterator end = priorityList.end(); 1429 1430 ThreadID high_pri; 1431 1432 while (pri_iter != end) { 1433 high_pri = *pri_iter; 1434 1435 assert(high_pri <= numThreads); 1436 1437 if (fetchStatus[high_pri] == Running || 1438 fetchStatus[high_pri] == IcacheAccessComplete || 1439 fetchStatus[high_pri] == Idle) { 1440 1441 priorityList.erase(pri_iter); 1442 priorityList.push_back(high_pri); 1443 1444 return high_pri; 1445 } 1446 1447 pri_iter++; 1448 } 1449 1450 return InvalidThreadID; 1451} 1452 1453template<class Impl> 1454ThreadID 1455DefaultFetch<Impl>::iqCount() 1456{ 1457 std::priority_queue<unsigned> PQ; 1458 std::map<unsigned, ThreadID> threadMap; 1459 1460 list<ThreadID>::iterator threads = activeThreads->begin(); 1461 list<ThreadID>::iterator end = activeThreads->end(); 1462 1463 while (threads != end) { 1464 ThreadID tid = *threads++; 1465 unsigned iqCount = fromIEW->iewInfo[tid].iqCount; 1466 1467 PQ.push(iqCount); 1468 threadMap[iqCount] = tid; 1469 } 1470 1471 while (!PQ.empty()) { 1472 ThreadID high_pri = threadMap[PQ.top()]; 1473 1474 if (fetchStatus[high_pri] == Running || 1475 fetchStatus[high_pri] == IcacheAccessComplete || 1476 fetchStatus[high_pri] == Idle) 1477 return high_pri; 1478 else 1479 PQ.pop(); 1480 1481 } 1482 1483 return InvalidThreadID; 1484} 1485 1486template<class Impl> 1487ThreadID 1488DefaultFetch<Impl>::lsqCount() 1489{ 1490 std::priority_queue<unsigned> PQ; 1491 std::map<unsigned, ThreadID> threadMap; 1492 1493 list<ThreadID>::iterator threads = activeThreads->begin(); 1494 list<ThreadID>::iterator end = activeThreads->end(); 1495 1496 while (threads != end) { 1497 ThreadID tid = *threads++; 1498 unsigned ldstqCount = fromIEW->iewInfo[tid].ldstqCount; 1499 1500 PQ.push(ldstqCount); 1501 threadMap[ldstqCount] = tid; 1502 } 1503 1504 while (!PQ.empty()) { 1505 ThreadID high_pri = threadMap[PQ.top()]; 1506 1507 if (fetchStatus[high_pri] == Running || 1508 fetchStatus[high_pri] == IcacheAccessComplete || 1509 fetchStatus[high_pri] == Idle) 1510 return high_pri; 1511 else 1512 PQ.pop(); 1513 } 1514 1515 return InvalidThreadID; 1516} 1517 1518template<class Impl> 1519ThreadID 1520DefaultFetch<Impl>::branchCount() 1521{ 1522#if 0 1523 list<ThreadID>::iterator thread = activeThreads->begin(); 1524 assert(thread != activeThreads->end()); 1525 ThreadID tid = *thread; 1526#endif 1527 1528 panic("Branch Count Fetch policy unimplemented\n"); 1529 return InvalidThreadID; 1530} 1531 1532template<class Impl> 1533void 1534DefaultFetch<Impl>::pipelineIcacheAccesses(ThreadID tid) 1535{ 1536 if (!issuePipelinedIfetch[tid]) { 1537 return; 1538 } 1539 1540 // The next PC to access. 1541 TheISA::PCState thisPC = pc[tid]; 1542 1543 if (isRomMicroPC(thisPC.microPC())) { 1544 return; 1545 } 1546 1547 Addr pcOffset = fetchOffset[tid]; 1548 Addr fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1549 1550 // Align the fetch PC so its at the start of a cache block. 1551 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1552 1553 // Unless buffer already got the block, fetch it from icache. 1554 if (!(cacheDataValid[tid] && block_PC == cacheDataPC[tid])) { 1555 DPRINTF(Fetch, "[tid:%i]: Issuing a pipelined I-cache access, " 1556 "starting at PC %s.\n", tid, thisPC); 1557 1558 fetchCacheLine(fetchAddr, tid, thisPC.instAddr()); 1559 } 1560} 1561 1562template<class Impl> 1563void 1564DefaultFetch<Impl>::profileStall(ThreadID tid) { 1565 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 1566 1567 // @todo Per-thread stats 1568
| 1188 // Stall CPU if an interrupt is posted and we're not issuing 1189 // an delayed commit micro-op currently (delayed commit instructions 1190 // are not interruptable by interrupts, only faults) 1191 ++fetchMiscStallCycles; 1192 DPRINTF(Fetch, "[tid:%i]: Fetch is stalled!\n", tid); 1193 return; 1194 } 1195 } else { 1196 if (fetchStatus[tid] == Idle) { 1197 ++fetchIdleCycles; 1198 DPRINTF(Fetch, "[tid:%i]: Fetch is idle!\n", tid); 1199 } 1200 1201 // Status is Idle, so fetch should do nothing. 1202 return; 1203 } 1204 1205 ++fetchCycles; 1206 1207 TheISA::PCState nextPC = thisPC; 1208 1209 StaticInstPtr staticInst = NULL; 1210 StaticInstPtr curMacroop = macroop[tid]; 1211 1212 // If the read of the first instruction was successful, then grab the 1213 // instructions from the rest of the cache line and put them into the 1214 // queue heading to decode. 1215 1216 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 1217 "decode.\n", tid); 1218 1219 // Need to keep track of whether or not a predicted branch 1220 // ended this fetch block. 1221 bool predictedBranch = false; 1222 1223 TheISA::MachInst *cacheInsts = 1224 reinterpret_cast<TheISA::MachInst *>(cacheData[tid]); 1225 1226 const unsigned numInsts = cacheBlkSize / instSize; 1227 unsigned blkOffset = (fetchAddr - cacheDataPC[tid]) / instSize; 1228 1229 // Loop through instruction memory from the cache. 1230 // Keep issuing while fetchWidth is available and branch is not 1231 // predicted taken 1232 while (numInst < fetchWidth && !predictedBranch) { 1233 1234 // We need to process more memory if we aren't going to get a 1235 // StaticInst from the rom, the current macroop, or what's already 1236 // in the decoder. 1237 bool needMem = !inRom && !curMacroop && 1238 !decoder[tid]->instReady(); 1239 fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1240 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1241 1242 if (needMem) { 1243 // If buffer is no longer valid or fetchAddr has moved to point 1244 // to the next cache block then start fetch from icache. 1245 if (!cacheDataValid[tid] || block_PC != cacheDataPC[tid]) 1246 break; 1247 1248 if (blkOffset >= numInsts) { 1249 // We need to process more memory, but we've run out of the 1250 // current block. 1251 break; 1252 } 1253 1254 if (ISA_HAS_DELAY_SLOT && pcOffset == 0) { 1255 // Walk past any annulled delay slot instructions. 1256 Addr pcAddr = thisPC.instAddr() & BaseCPU::PCMask; 1257 while (fetchAddr != pcAddr && blkOffset < numInsts) { 1258 blkOffset++; 1259 fetchAddr += instSize; 1260 } 1261 if (blkOffset >= numInsts) 1262 break; 1263 } 1264 1265 MachInst inst = TheISA::gtoh(cacheInsts[blkOffset]); 1266 decoder[tid]->moreBytes(thisPC, fetchAddr, inst); 1267 1268 if (decoder[tid]->needMoreBytes()) { 1269 blkOffset++; 1270 fetchAddr += instSize; 1271 pcOffset += instSize; 1272 } 1273 } 1274 1275 // Extract as many instructions and/or microops as we can from 1276 // the memory we've processed so far. 1277 do { 1278 if (!(curMacroop || inRom)) { 1279 if (decoder[tid]->instReady()) { 1280 staticInst = decoder[tid]->decode(thisPC); 1281 1282 // Increment stat of fetched instructions. 1283 ++fetchedInsts; 1284 1285 if (staticInst->isMacroop()) { 1286 curMacroop = staticInst; 1287 } else { 1288 pcOffset = 0; 1289 } 1290 } else { 1291 // We need more bytes for this instruction so blkOffset and 1292 // pcOffset will be updated 1293 break; 1294 } 1295 } 1296 // Whether we're moving to a new macroop because we're at the 1297 // end of the current one, or the branch predictor incorrectly 1298 // thinks we are... 1299 bool newMacro = false; 1300 if (curMacroop || inRom) { 1301 if (inRom) { 1302 staticInst = cpu->microcodeRom.fetchMicroop( 1303 thisPC.microPC(), curMacroop); 1304 } else { 1305 staticInst = curMacroop->fetchMicroop(thisPC.microPC()); 1306 } 1307 newMacro |= staticInst->isLastMicroop(); 1308 } 1309 1310 DynInstPtr instruction = 1311 buildInst(tid, staticInst, curMacroop, 1312 thisPC, nextPC, true); 1313 1314 numInst++; 1315 1316#if TRACING_ON 1317 instruction->fetchTick = curTick(); 1318#endif 1319 1320 nextPC = thisPC; 1321 1322 // If we're branching after this instruction, quite fetching 1323 // from the same block then. 1324 predictedBranch |= thisPC.branching(); 1325 predictedBranch |= 1326 lookupAndUpdateNextPC(instruction, nextPC); 1327 if (predictedBranch) { 1328 DPRINTF(Fetch, "Branch detected with PC = %s\n", thisPC); 1329 } 1330 1331 newMacro |= thisPC.instAddr() != nextPC.instAddr(); 1332 1333 // Move to the next instruction, unless we have a branch. 1334 thisPC = nextPC; 1335 inRom = isRomMicroPC(thisPC.microPC()); 1336 1337 if (newMacro) { 1338 fetchAddr = thisPC.instAddr() & BaseCPU::PCMask; 1339 blkOffset = (fetchAddr - cacheDataPC[tid]) / instSize; 1340 pcOffset = 0; 1341 curMacroop = NULL; 1342 } 1343 1344 if (instruction->isQuiesce()) { 1345 DPRINTF(Fetch, 1346 "Quiesce instruction encountered, halting fetch!"); 1347 fetchStatus[tid] = QuiescePending; 1348 status_change = true; 1349 break; 1350 } 1351 } while ((curMacroop || decoder[tid]->instReady()) && 1352 numInst < fetchWidth); 1353 } 1354 1355 if (predictedBranch) { 1356 DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch " 1357 "instruction encountered.\n", tid); 1358 } else if (numInst >= fetchWidth) { 1359 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth " 1360 "for this cycle.\n", tid); 1361 } else if (blkOffset >= cacheBlkSize) { 1362 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache " 1363 "block.\n", tid); 1364 } 1365 1366 macroop[tid] = curMacroop; 1367 fetchOffset[tid] = pcOffset; 1368 1369 if (numInst > 0) { 1370 wroteToTimeBuffer = true; 1371 } 1372 1373 pc[tid] = thisPC; 1374 1375 // pipeline a fetch if we're crossing a cache boundary and not in 1376 // a state that would preclude fetching 1377 fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1378 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1379 issuePipelinedIfetch[tid] = block_PC != cacheDataPC[tid] && 1380 fetchStatus[tid] != IcacheWaitResponse && 1381 fetchStatus[tid] != ItlbWait && 1382 fetchStatus[tid] != IcacheWaitRetry && 1383 fetchStatus[tid] != QuiescePending && 1384 !curMacroop; 1385} 1386 1387template<class Impl> 1388void 1389DefaultFetch<Impl>::recvRetry() 1390{ 1391 if (retryPkt != NULL) { 1392 assert(cacheBlocked); 1393 assert(retryTid != InvalidThreadID); 1394 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1395 1396 if (cpu->getInstPort().sendTimingReq(retryPkt)) { 1397 fetchStatus[retryTid] = IcacheWaitResponse; 1398 retryPkt = NULL; 1399 retryTid = InvalidThreadID; 1400 cacheBlocked = false; 1401 } 1402 } else { 1403 assert(retryTid == InvalidThreadID); 1404 // Access has been squashed since it was sent out. Just clear 1405 // the cache being blocked. 1406 cacheBlocked = false; 1407 } 1408} 1409 1410/////////////////////////////////////// 1411// // 1412// SMT FETCH POLICY MAINTAINED HERE // 1413// // 1414/////////////////////////////////////// 1415template<class Impl> 1416ThreadID 1417DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1418{ 1419 if (numThreads > 1) { 1420 switch (fetch_priority) { 1421 1422 case SingleThread: 1423 return 0; 1424 1425 case RoundRobin: 1426 return roundRobin(); 1427 1428 case IQ: 1429 return iqCount(); 1430 1431 case LSQ: 1432 return lsqCount(); 1433 1434 case Branch: 1435 return branchCount(); 1436 1437 default: 1438 return InvalidThreadID; 1439 } 1440 } else { 1441 list<ThreadID>::iterator thread = activeThreads->begin(); 1442 if (thread == activeThreads->end()) { 1443 return InvalidThreadID; 1444 } 1445 1446 ThreadID tid = *thread; 1447 1448 if (fetchStatus[tid] == Running || 1449 fetchStatus[tid] == IcacheAccessComplete || 1450 fetchStatus[tid] == Idle) { 1451 return tid; 1452 } else { 1453 return InvalidThreadID; 1454 } 1455 } 1456} 1457 1458 1459template<class Impl> 1460ThreadID 1461DefaultFetch<Impl>::roundRobin() 1462{ 1463 list<ThreadID>::iterator pri_iter = priorityList.begin(); 1464 list<ThreadID>::iterator end = priorityList.end(); 1465 1466 ThreadID high_pri; 1467 1468 while (pri_iter != end) { 1469 high_pri = *pri_iter; 1470 1471 assert(high_pri <= numThreads); 1472 1473 if (fetchStatus[high_pri] == Running || 1474 fetchStatus[high_pri] == IcacheAccessComplete || 1475 fetchStatus[high_pri] == Idle) { 1476 1477 priorityList.erase(pri_iter); 1478 priorityList.push_back(high_pri); 1479 1480 return high_pri; 1481 } 1482 1483 pri_iter++; 1484 } 1485 1486 return InvalidThreadID; 1487} 1488 1489template<class Impl> 1490ThreadID 1491DefaultFetch<Impl>::iqCount() 1492{ 1493 std::priority_queue<unsigned> PQ; 1494 std::map<unsigned, ThreadID> threadMap; 1495 1496 list<ThreadID>::iterator threads = activeThreads->begin(); 1497 list<ThreadID>::iterator end = activeThreads->end(); 1498 1499 while (threads != end) { 1500 ThreadID tid = *threads++; 1501 unsigned iqCount = fromIEW->iewInfo[tid].iqCount; 1502 1503 PQ.push(iqCount); 1504 threadMap[iqCount] = tid; 1505 } 1506 1507 while (!PQ.empty()) { 1508 ThreadID high_pri = threadMap[PQ.top()]; 1509 1510 if (fetchStatus[high_pri] == Running || 1511 fetchStatus[high_pri] == IcacheAccessComplete || 1512 fetchStatus[high_pri] == Idle) 1513 return high_pri; 1514 else 1515 PQ.pop(); 1516 1517 } 1518 1519 return InvalidThreadID; 1520} 1521 1522template<class Impl> 1523ThreadID 1524DefaultFetch<Impl>::lsqCount() 1525{ 1526 std::priority_queue<unsigned> PQ; 1527 std::map<unsigned, ThreadID> threadMap; 1528 1529 list<ThreadID>::iterator threads = activeThreads->begin(); 1530 list<ThreadID>::iterator end = activeThreads->end(); 1531 1532 while (threads != end) { 1533 ThreadID tid = *threads++; 1534 unsigned ldstqCount = fromIEW->iewInfo[tid].ldstqCount; 1535 1536 PQ.push(ldstqCount); 1537 threadMap[ldstqCount] = tid; 1538 } 1539 1540 while (!PQ.empty()) { 1541 ThreadID high_pri = threadMap[PQ.top()]; 1542 1543 if (fetchStatus[high_pri] == Running || 1544 fetchStatus[high_pri] == IcacheAccessComplete || 1545 fetchStatus[high_pri] == Idle) 1546 return high_pri; 1547 else 1548 PQ.pop(); 1549 } 1550 1551 return InvalidThreadID; 1552} 1553 1554template<class Impl> 1555ThreadID 1556DefaultFetch<Impl>::branchCount() 1557{ 1558#if 0 1559 list<ThreadID>::iterator thread = activeThreads->begin(); 1560 assert(thread != activeThreads->end()); 1561 ThreadID tid = *thread; 1562#endif 1563 1564 panic("Branch Count Fetch policy unimplemented\n"); 1565 return InvalidThreadID; 1566} 1567 1568template<class Impl> 1569void 1570DefaultFetch<Impl>::pipelineIcacheAccesses(ThreadID tid) 1571{ 1572 if (!issuePipelinedIfetch[tid]) { 1573 return; 1574 } 1575 1576 // The next PC to access. 1577 TheISA::PCState thisPC = pc[tid]; 1578 1579 if (isRomMicroPC(thisPC.microPC())) { 1580 return; 1581 } 1582 1583 Addr pcOffset = fetchOffset[tid]; 1584 Addr fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1585 1586 // Align the fetch PC so its at the start of a cache block. 1587 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1588 1589 // Unless buffer already got the block, fetch it from icache. 1590 if (!(cacheDataValid[tid] && block_PC == cacheDataPC[tid])) { 1591 DPRINTF(Fetch, "[tid:%i]: Issuing a pipelined I-cache access, " 1592 "starting at PC %s.\n", tid, thisPC); 1593 1594 fetchCacheLine(fetchAddr, tid, thisPC.instAddr()); 1595 } 1596} 1597 1598template<class Impl> 1599void 1600DefaultFetch<Impl>::profileStall(ThreadID tid) { 1601 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 1602 1603 // @todo Per-thread stats 1604
|
1569 if (drainPending) {
| 1605 if (stalls[tid].drain) {
|
1570 ++fetchPendingDrainCycles; 1571 DPRINTF(Fetch, "Fetch is waiting for a drain!\n"); 1572 } else if (activeThreads->empty()) { 1573 ++fetchNoActiveThreadStallCycles; 1574 DPRINTF(Fetch, "Fetch has no active thread!\n"); 1575 } else if (fetchStatus[tid] == Blocked) { 1576 ++fetchBlockedCycles; 1577 DPRINTF(Fetch, "[tid:%i]: Fetch is blocked!\n", tid); 1578 } else if (fetchStatus[tid] == Squashing) { 1579 ++fetchSquashCycles; 1580 DPRINTF(Fetch, "[tid:%i]: Fetch is squashing!\n", tid); 1581 } else if (fetchStatus[tid] == IcacheWaitResponse) { 1582 ++icacheStallCycles; 1583 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting cache response!\n", 1584 tid); 1585 } else if (fetchStatus[tid] == ItlbWait) { 1586 ++fetchTlbCycles; 1587 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting ITLB walk to " 1588 "finish!\n", tid); 1589 } else if (fetchStatus[tid] == TrapPending) { 1590 ++fetchPendingTrapStallCycles; 1591 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for a pending trap!\n", 1592 tid); 1593 } else if (fetchStatus[tid] == QuiescePending) { 1594 ++fetchPendingQuiesceStallCycles; 1595 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for a pending quiesce " 1596 "instruction!\n", tid); 1597 } else if (fetchStatus[tid] == IcacheWaitRetry) { 1598 ++fetchIcacheWaitRetryStallCycles; 1599 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for an I-cache retry!\n", 1600 tid); 1601 } else if (fetchStatus[tid] == NoGoodAddr) { 1602 DPRINTF(Fetch, "[tid:%i]: Fetch predicted non-executable address\n", 1603 tid); 1604 } else { 1605 DPRINTF(Fetch, "[tid:%i]: Unexpected fetch stall reason (Status: %i).\n", 1606 tid, fetchStatus[tid]); 1607 } 1608}
| 1606 ++fetchPendingDrainCycles; 1607 DPRINTF(Fetch, "Fetch is waiting for a drain!\n"); 1608 } else if (activeThreads->empty()) { 1609 ++fetchNoActiveThreadStallCycles; 1610 DPRINTF(Fetch, "Fetch has no active thread!\n"); 1611 } else if (fetchStatus[tid] == Blocked) { 1612 ++fetchBlockedCycles; 1613 DPRINTF(Fetch, "[tid:%i]: Fetch is blocked!\n", tid); 1614 } else if (fetchStatus[tid] == Squashing) { 1615 ++fetchSquashCycles; 1616 DPRINTF(Fetch, "[tid:%i]: Fetch is squashing!\n", tid); 1617 } else if (fetchStatus[tid] == IcacheWaitResponse) { 1618 ++icacheStallCycles; 1619 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting cache response!\n", 1620 tid); 1621 } else if (fetchStatus[tid] == ItlbWait) { 1622 ++fetchTlbCycles; 1623 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting ITLB walk to " 1624 "finish!\n", tid); 1625 } else if (fetchStatus[tid] == TrapPending) { 1626 ++fetchPendingTrapStallCycles; 1627 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for a pending trap!\n", 1628 tid); 1629 } else if (fetchStatus[tid] == QuiescePending) { 1630 ++fetchPendingQuiesceStallCycles; 1631 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for a pending quiesce " 1632 "instruction!\n", tid); 1633 } else if (fetchStatus[tid] == IcacheWaitRetry) { 1634 ++fetchIcacheWaitRetryStallCycles; 1635 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for an I-cache retry!\n", 1636 tid); 1637 } else if (fetchStatus[tid] == NoGoodAddr) { 1638 DPRINTF(Fetch, "[tid:%i]: Fetch predicted non-executable address\n", 1639 tid); 1640 } else { 1641 DPRINTF(Fetch, "[tid:%i]: Unexpected fetch stall reason (Status: %i).\n", 1642 tid, fetchStatus[tid]); 1643 } 1644}
|