/*
* Copyright (c) 2010-2011 ARM Limited
* All rights reserved.
*
* The license below extends only to copyright in the software and shall
* not be construed as granting a license to any other intellectual
* property including but not limited to intellectual property relating
* to a hardware implementation of the functionality of the software
* licensed hereunder. You may use the software subject to the license
* terms below provided that you ensure that this notice is replicated
* unmodified and in its entirety in all distributions of the software,
* modified or unmodified, in source code or in binary form.
*
* Copyright (c) 2004-2006 The Regents of The University of Michigan
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met: redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer;
* redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution;
* neither the name of the copyright holders nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* Authors: Kevin Lim
* Korey Sewell
*/
#include <algorithm>
#include <cstring>
#include <list>
#include <map>
#include <queue>
#include "arch/isa_traits.hh"
#include "arch/tlb.hh"
#include "arch/utility.hh"
#include "arch/vtophys.hh"
#include "base/types.hh"
#include "config/the_isa.hh"
#include "config/use_checker.hh"
#include "cpu/base.hh"
#include "cpu/o3/fetch.hh"
#include "cpu/exetrace.hh"
#include "debug/Activity.hh"
#include "debug/Fetch.hh"
#include "mem/packet.hh"
#include "mem/request.hh"
#include "params/DerivO3CPU.hh"
#include "sim/byteswap.hh"
#include "sim/core.hh"
#include "sim/eventq.hh"
#include "sim/full_system.hh"
#include "sim/system.hh"
#if USE_CHECKER
#include "cpu/checker/cpu.hh"
#endif // USE_CHECKER
using namespace std;
template<class Impl>
DefaultFetch<Impl>::DefaultFetch(O3CPU *_cpu, DerivO3CPUParams *params)
: cpu(_cpu),
branchPred(params),
predecoder(NULL),
numInst(0),
decodeToFetchDelay(params->decodeToFetchDelay),
renameToFetchDelay(params->renameToFetchDelay),
iewToFetchDelay(params->iewToFetchDelay),
commitToFetchDelay(params->commitToFetchDelay),
fetchWidth(params->fetchWidth),
cacheBlocked(false),
retryPkt(NULL),
retryTid(InvalidThreadID),
numThreads(params->numThreads),
numFetchingThreads(params->smtNumFetchingThreads),
interruptPending(false),
drainPending(false),
switchedOut(false),
finishTranslationEvent(this)
{
if (numThreads > Impl::MaxThreads)
fatal("numThreads (%d) is larger than compiled limit (%d),\n"
"\tincrease MaxThreads in src/cpu/o3/impl.hh\n",
numThreads, static_cast<int>(Impl::MaxThreads));
// Set fetch stage's status to inactive.
_status = Inactive;
std::string policy = params->smtFetchPolicy;
// Convert string to lowercase
std::transform(policy.begin(), policy.end(), policy.begin(),
(int(*)(int)) tolower);
// Figure out fetch policy
if (policy == "singlethread") {
fetchPolicy = SingleThread;
if (numThreads > 1)
panic("Invalid Fetch Policy for a SMT workload.");
} else if (policy == "roundrobin") {
fetchPolicy = RoundRobin;
DPRINTF(Fetch, "Fetch policy set to Round Robin\n");
} else if (policy == "branch") {
fetchPolicy = Branch;
DPRINTF(Fetch, "Fetch policy set to Branch Count\n");
} else if (policy == "iqcount") {
fetchPolicy = IQ;
DPRINTF(Fetch, "Fetch policy set to IQ count\n");
} else if (policy == "lsqcount") {
fetchPolicy = LSQ;
DPRINTF(Fetch, "Fetch policy set to LSQ count\n");
} else {
fatal("Invalid Fetch Policy. Options Are: {SingleThread,"
" RoundRobin,LSQcount,IQcount}\n");
}
// Get the size of an instruction.
instSize = sizeof(TheISA::MachInst);
}
template <class Impl>
std::string
DefaultFetch<Impl>::name() const
{
return cpu->name() + ".fetch";
}
template <class Impl>
void
DefaultFetch<Impl>::regStats()
{
icacheStallCycles
.name(name() + ".icacheStallCycles")
.desc("Number of cycles fetch is stalled on an Icache miss")
.prereq(icacheStallCycles);
fetchedInsts
.name(name() + ".Insts")
.desc("Number of instructions fetch has processed")
.prereq(fetchedInsts);
fetchedBranches
.name(name() + ".Branches")
.desc("Number of branches that fetch encountered")
.prereq(fetchedBranches);
predictedBranches
.name(name() + ".predictedBranches")
.desc("Number of branches that fetch has predicted taken")
.prereq(predictedBranches);
fetchCycles
.name(name() + ".Cycles")
.desc("Number of cycles fetch has run and was not squashing or"
" blocked")
.prereq(fetchCycles);
fetchSquashCycles
.name(name() + ".SquashCycles")
.desc("Number of cycles fetch has spent squashing")
.prereq(fetchSquashCycles);
fetchTlbCycles
.name(name() + ".TlbCycles")
.desc("Number of cycles fetch has spent waiting for tlb")
.prereq(fetchTlbCycles);
fetchIdleCycles
.name(name() + ".IdleCycles")
.desc("Number of cycles fetch was idle")
.prereq(fetchIdleCycles);
fetchBlockedCycles
.name(name() + ".BlockedCycles")
.desc("Number of cycles fetch has spent blocked")
.prereq(fetchBlockedCycles);
fetchedCacheLines
.name(name() + ".CacheLines")
.desc("Number of cache lines fetched")
.prereq(fetchedCacheLines);
fetchMiscStallCycles
.name(name() + ".MiscStallCycles")
.desc("Number of cycles fetch has spent waiting on interrupts, or "
"bad addresses, or out of MSHRs")
.prereq(fetchMiscStallCycles);
fetchPendingDrainCycles
.name(name() + ".PendingDrainCycles")
.desc("Number of cycles fetch has spent waiting on pipes to drain")
.prereq(fetchPendingDrainCycles);
fetchNoActiveThreadStallCycles
.name(name() + ".NoActiveThreadStallCycles")
.desc("Number of stall cycles due to no active thread to fetch from")
.prereq(fetchNoActiveThreadStallCycles);
fetchPendingTrapStallCycles
.name(name() + ".PendingTrapStallCycles")
.desc("Number of stall cycles due to pending traps")
.prereq(fetchPendingTrapStallCycles);
fetchPendingQuiesceStallCycles
.name(name() + ".PendingQuiesceStallCycles")
.desc("Number of stall cycles due to pending quiesce instructions")
.prereq(fetchPendingQuiesceStallCycles);
fetchIcacheWaitRetryStallCycles
.name(name() + ".IcacheWaitRetryStallCycles")
.desc("Number of stall cycles due to full MSHR")
.prereq(fetchIcacheWaitRetryStallCycles);
fetchIcacheSquashes
.name(name() + ".IcacheSquashes")
.desc("Number of outstanding Icache misses that were squashed")
.prereq(fetchIcacheSquashes);
fetchTlbSquashes
.name(name() + ".ItlbSquashes")
.desc("Number of outstanding ITLB misses that were squashed")
.prereq(fetchTlbSquashes);
fetchNisnDist
.init(/* base value */ 0,
/* last value */ fetchWidth,
/* bucket size */ 1)
.name(name() + ".rateDist")
.desc("Number of instructions fetched each cycle (Total)")
.flags(Stats::pdf);
idleRate
.name(name() + ".idleRate")
.desc("Percent of cycles fetch was idle")
.prereq(idleRate);
idleRate = fetchIdleCycles * 100 / cpu->numCycles;
branchRate
.name(name() + ".branchRate")
.desc("Number of branch fetches per cycle")
.flags(Stats::total);
branchRate = fetchedBranches / cpu->numCycles;
fetchRate
.name(name() + ".rate")
.desc("Number of inst fetches per cycle")
.flags(Stats::total);
fetchRate = fetchedInsts / cpu->numCycles;
branchPred.regStats();
}
template<class Impl>
void
DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer)
{
timeBuffer = time_buffer;
// Create wires to get information from proper places in time buffer.
fromDecode = timeBuffer->getWire(-decodeToFetchDelay);
fromRename = timeBuffer->getWire(-renameToFetchDelay);
fromIEW = timeBuffer->getWire(-iewToFetchDelay);
fromCommit = timeBuffer->getWire(-commitToFetchDelay);
}
template<class Impl>
void
DefaultFetch<Impl>::setActiveThreads(std::list<ThreadID> *at_ptr)
{
activeThreads = at_ptr;
}
template<class Impl>
void
DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr)
{
fetchQueue = fq_ptr;
// Create wire to write information to proper place in fetch queue.
toDecode = fetchQueue->getWire(0);
}
template<class Impl>
void
DefaultFetch<Impl>::initStage()
{
// Setup PC and nextPC with initial state.
for (ThreadID tid = 0; tid < numThreads; tid++) {
pc[tid] = cpu->pcState(tid);
fetchOffset[tid] = 0;
macroop[tid] = NULL;
delayedCommit[tid] = false;
}
for (ThreadID tid = 0; tid < numThreads; tid++) {
fetchStatus[tid] = Running;
priorityList.push_back(tid);
memReq[tid] = NULL;
stalls[tid].decode = false;
stalls[tid].rename = false;
stalls[tid].iew = false;
stalls[tid].commit = false;
}
// Schedule fetch to get the correct PC from the CPU
// scheduleFetchStartupEvent(1);
// Fetch needs to start fetching instructions at the very beginning,
// so it must start up in active state.
switchToActive();
}
template<class Impl>
void
DefaultFetch<Impl>::setIcache()
{
assert(cpu->getIcachePort()->isConnected());
// Size of cache block.
cacheBlkSize = cpu->getIcachePort()->peerBlockSize();
// Create mask to get rid of offset bits.
cacheBlkMask = (cacheBlkSize - 1);
for (ThreadID tid = 0; tid < numThreads; tid++) {
// Create space to store a cache line.
cacheData[tid] = new uint8_t[cacheBlkSize];
cacheDataPC[tid] = 0;
cacheDataValid[tid] = false;
}
}
template<class Impl>
void
DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt)
{
ThreadID tid = pkt->req->threadId();
DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n", tid);
assert(!pkt->wasNacked());
// Only change the status if it's still waiting on the icache access
// to return.
if (fetchStatus[tid] != IcacheWaitResponse ||
pkt->req != memReq[tid] ||
isSwitchedOut()) {
++fetchIcacheSquashes;
delete pkt->req;
delete pkt;
return;
}
memcpy(cacheData[tid], pkt->getPtr<uint8_t>(), cacheBlkSize);
cacheDataValid[tid] = true;
if (!drainPending) {
// Wake up the CPU (if it went to sleep and was waiting on
// this completion event).
cpu->wakeCPU();
DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n",
tid);
switchToActive();
}
// Only switch to IcacheAccessComplete if we're not stalled as well.
if (checkStall(tid)) {
fetchStatus[tid] = Blocked;
} else {
fetchStatus[tid] = IcacheAccessComplete;
}
// Reset the mem req to NULL.
delete pkt->req;
delete pkt;
memReq[tid] = NULL;
}
template <class Impl>
bool
DefaultFetch<Impl>::drain()
{
// Fetch is ready to drain at any time.
cpu->signalDrained();
drainPending = true;
return true;
}
template <class Impl>
void
DefaultFetch<Impl>::resume()
{
drainPending = false;
}
template <class Impl>
void
DefaultFetch<Impl>::switchOut()
{
switchedOut = true;
// Branch predictor needs to have its state cleared.
branchPred.switchOut();
}
template <class Impl>
void
DefaultFetch<Impl>::takeOverFrom()
{
// the instruction port is now connected so we can get the block
// size
setIcache();
// Reset all state
for (ThreadID i = 0; i < Impl::MaxThreads; ++i) {
stalls[i].decode = 0;
stalls[i].rename = 0;
stalls[i].iew = 0;
stalls[i].commit = 0;
pc[i] = cpu->pcState(i);
fetchStatus[i] = Running;
}
numInst = 0;
wroteToTimeBuffer = false;
_status = Inactive;
switchedOut = false;
interruptPending = false;
branchPred.takeOverFrom();
}
template <class Impl>
void
DefaultFetch<Impl>::wakeFromQuiesce()
{
DPRINTF(Fetch, "Waking up from quiesce\n");
// Hopefully this is safe
// @todo: Allow other threads to wake from quiesce.
fetchStatus[0] = Running;
}
template <class Impl>
inline void
DefaultFetch<Impl>::switchToActive()
{
if (_status == Inactive) {
DPRINTF(Activity, "Activating stage.\n");
cpu->activateStage(O3CPU::FetchIdx);
_status = Active;
}
}
template <class Impl>
inline void
DefaultFetch<Impl>::switchToInactive()
{
if (_status == Active) {
DPRINTF(Activity, "Deactivating stage.\n");
cpu->deactivateStage(O3CPU::FetchIdx);
_status = Inactive;
}
}
template <class Impl>
bool
DefaultFetch<Impl>::lookupAndUpdateNextPC(
DynInstPtr &inst, TheISA::PCState &nextPC)
{
// Do branch prediction check here.
// A bit of a misnomer...next_PC is actually the current PC until
// this function updates it.
bool predict_taken;
if (!inst->isControl()) {
TheISA::advancePC(nextPC, inst->staticInst);
inst->setPredTarg(nextPC);
inst->setPredTaken(false);
return false;
}
ThreadID tid = inst->threadNumber;
predict_taken = branchPred.predict(inst, nextPC, tid);
if (predict_taken) {
DPRINTF(Fetch, "[tid:%i]: [sn:%i]: Branch predicted to be taken to %s.\n",
tid, inst->seqNum, nextPC);
} else {
DPRINTF(Fetch, "[tid:%i]: [sn:%i]:Branch predicted to be not taken.\n",
tid, inst->seqNum);
}
DPRINTF(Fetch, "[tid:%i]: [sn:%i] Branch predicted to go to %s.\n",
tid, inst->seqNum, nextPC);
inst->setPredTarg(nextPC);
inst->setPredTaken(predict_taken);
++fetchedBranches;
if (predict_taken) {
++predictedBranches;
}
return predict_taken;
}
template <class Impl>
bool
DefaultFetch<Impl>::fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc)
{
Fault fault = NoFault;
// @todo: not sure if these should block translation.
//AlphaDep
if (cacheBlocked) {
DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, cache blocked\n",
tid);
return false;
} else if (isSwitchedOut()) {
DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, switched out\n",
tid);
return false;
} else if (checkInterrupt(pc) && !delayedCommit[tid]) {
// Hold off fetch from getting new instructions when:
// Cache is blocked, or
// while an interrupt is pending and we're not in PAL mode, or
// fetch is switched out.
DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, interrupt pending\n",
tid);
return false;
}
// Align the fetch address so it's at the start of a cache block.
Addr block_PC = icacheBlockAlignPC(vaddr);
DPRINTF(Fetch, "[tid:%i] Fetching cache line %#x for addr %#x\n",
tid, block_PC, vaddr);
// Setup the memReq to do a read of the first instruction's address.
// Set the appropriate read size and flags as well.
// Build request here.
RequestPtr mem_req =
new Request(tid, block_PC, cacheBlkSize, Request::INST_FETCH,
cpu->instMasterId(), pc, cpu->thread[tid]->contextId(), tid);
memReq[tid] = mem_req;
// Initiate translation of the icache block
fetchStatus[tid] = ItlbWait;
FetchTranslation *trans = new FetchTranslation(this);
cpu->itb->translateTiming(mem_req, cpu->thread[tid]->getTC(),
trans, BaseTLB::Execute);
return true;
}
template <class Impl>
void
DefaultFetch<Impl>::finishTranslation(Fault fault, RequestPtr mem_req)
{
ThreadID tid = mem_req->threadId();
Addr block_PC = mem_req->getVaddr();
// Wake up CPU if it was idle
cpu->wakeCPU();
if (fetchStatus[tid] != ItlbWait || mem_req != memReq[tid] ||
mem_req->getVaddr() != memReq[tid]->getVaddr() || isSwitchedOut()) {
DPRINTF(Fetch, "[tid:%i] Ignoring itlb completed after squash\n",
tid);
++fetchTlbSquashes;
delete mem_req;
return;
}
// If translation was successful, attempt to read the icache block.
if (fault == NoFault) {
// Check that we're not going off into random memory
// If we have, just wait around for commit to squash something and put
// us on the right track
if (!cpu->system->isMemory(mem_req->getPaddr())) {
warn("Address %#x is outside of physical memory, stopping fetch\n",
mem_req->getPaddr());
fetchStatus[tid] = NoGoodAddr;
delete mem_req;
memReq[tid] = NULL;
return;
}
// Build packet here.
PacketPtr data_pkt = new Packet(mem_req,
MemCmd::ReadReq, Packet::Broadcast);
data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]);
cacheDataPC[tid] = block_PC;
cacheDataValid[tid] = false;
DPRINTF(Fetch, "Fetch: Doing instruction read.\n");
fetchedCacheLines++;
// Access the cache.
if (!cpu->getIcachePort()->sendTiming(data_pkt)) {
assert(retryPkt == NULL);
assert(retryTid == InvalidThreadID);
DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid);
fetchStatus[tid] = IcacheWaitRetry;
retryPkt = data_pkt;
retryTid = tid;
cacheBlocked = true;
} else {
DPRINTF(Fetch, "[tid:%i]: Doing Icache access.\n", tid);
DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache "
"response.\n", tid);
lastIcacheStall[tid] = curTick();
fetchStatus[tid] = IcacheWaitResponse;
}
} else {
if (!(numInst < fetchWidth)) {
assert(!finishTranslationEvent.scheduled());
finishTranslationEvent.setFault(fault);
finishTranslationEvent.setReq(mem_req);
cpu->schedule(finishTranslationEvent, cpu->nextCycle(curTick() + cpu->ticks(1)));
return;
}
DPRINTF(Fetch, "[tid:%i] Got back req with addr %#x but expected %#x\n",
tid, mem_req->getVaddr(), memReq[tid]->getVaddr());
// Translation faulted, icache request won't be sent.
delete mem_req;
memReq[tid] = NULL;
// Send the fault to commit. This thread will not do anything
// until commit handles the fault. The only other way it can
// wake up is if a squash comes along and changes the PC.
TheISA::PCState fetchPC = pc[tid];
DPRINTF(Fetch, "[tid:%i]: Translation faulted, building noop.\n", tid);
// We will use a nop in ordier to carry the fault.
DynInstPtr instruction = buildInst(tid,
decoder.decode(TheISA::NoopMachInst, fetchPC.instAddr()),
NULL, fetchPC, fetchPC, false);
instruction->setPredTarg(fetchPC);
instruction->fault = fault;
wroteToTimeBuffer = true;
DPRINTF(Activity, "Activity this cycle.\n");
cpu->activityThisCycle();
fetchStatus[tid] = TrapPending;
DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n", tid);
DPRINTF(Fetch, "[tid:%i]: fault (%s) detected @ PC %s.\n",
tid, fault->name(), pc[tid]);
}
_status = updateFetchStatus();
}
template <class Impl>
inline void
DefaultFetch<Impl>::doSquash(const TheISA::PCState &newPC,
const DynInstPtr squashInst, ThreadID tid)
{
DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %s.\n",
tid, newPC);
pc[tid] = newPC;
fetchOffset[tid] = 0;
if (squashInst && squashInst->pcState().instAddr() == newPC.instAddr())
macroop[tid] = squashInst->macroop;
else
macroop[tid] = NULL;
predecoder.reset();
// Clear the icache miss if it's outstanding.
if (fetchStatus[tid] == IcacheWaitResponse) {
DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n",
tid);
memReq[tid] = NULL;
} else if (fetchStatus[tid] == ItlbWait) {
DPRINTF(Fetch, "[tid:%i]: Squashing outstanding ITLB miss.\n",
tid);
memReq[tid] = NULL;
}
// Get rid of the retrying packet if it was from this thread.
if (retryTid == tid) {
assert(cacheBlocked);
if (retryPkt) {
delete retryPkt->req;
delete retryPkt;
}
retryPkt = NULL;
retryTid = InvalidThreadID;
}
fetchStatus[tid] = Squashing;
// microops are being squashed, it is not known wheather the
// youngest non-squashed microop was marked delayed commit
// or not. Setting the flag to true ensures that the
// interrupts are not handled when they cannot be, though
// some opportunities to handle interrupts may be missed.
delayedCommit[tid] = true;
++fetchSquashCycles;
}
template<class Impl>
void
DefaultFetch<Impl>::squashFromDecode(const TheISA::PCState &newPC,
const DynInstPtr squashInst,
const InstSeqNum seq_num, ThreadID tid)
{
DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n", tid);
doSquash(newPC, squashInst, tid);
// Tell the CPU to remove any instructions that are in flight between
// fetch and decode.
cpu->removeInstsUntil(seq_num, tid);
}
template<class Impl>
bool
DefaultFetch<Impl>::checkStall(ThreadID tid) const
{
bool ret_val = false;
if (cpu->contextSwitch) {
DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid);
ret_val = true;
} else if (stalls[tid].decode) {
DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid);
ret_val = true;
} else if (stalls[tid].rename) {
DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid);
ret_val = true;
} else if (stalls[tid].iew) {
DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid);
ret_val = true;
} else if (stalls[tid].commit) {
DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid);
ret_val = true;
}
return ret_val;
}
template<class Impl>
typename DefaultFetch<Impl>::FetchStatus
DefaultFetch<Impl>::updateFetchStatus()
{
//Check Running
list<ThreadID>::iterator threads = activeThreads->begin();
list<ThreadID>::iterator end = activeThreads->end();
while (threads != end) {
ThreadID tid = *threads++;
if (fetchStatus[tid] == Running ||
fetchStatus[tid] == Squashing ||
fetchStatus[tid] == IcacheAccessComplete) {
if (_status == Inactive) {
DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid);
if (fetchStatus[tid] == IcacheAccessComplete) {
DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache"
"completion\n",tid);
}
cpu->activateStage(O3CPU::FetchIdx);
}
return Active;
}
}
// Stage is switching from active to inactive, notify CPU of it.
if (_status == Active) {
DPRINTF(Activity, "Deactivating stage.\n");
cpu->deactivateStage(O3CPU::FetchIdx);
}
return Inactive;
}
template <class Impl>
void
DefaultFetch<Impl>::squash(const TheISA::PCState &newPC,
const InstSeqNum seq_num, DynInstPtr squashInst,
ThreadID tid)
{
DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n", tid);
doSquash(newPC, squashInst, tid);
// Tell the CPU to remove any instructions that are not in the ROB.
cpu->removeInstsNotInROB(tid);
}
template <class Impl>
void
DefaultFetch<Impl>::tick()
{
list<ThreadID>::iterator threads = activeThreads->begin();
list<ThreadID>::iterator end = activeThreads->end();
bool status_change = false;
wroteToTimeBuffer = false;
for (ThreadID i = 0; i < Impl::MaxThreads; ++i) {
issuePipelinedIfetch[i] = false;
}
while (threads != end) {
ThreadID tid = *threads++;
// Check the signals for each thread to determine the proper status
// for each thread.
bool updated_status = checkSignalsAndUpdate(tid);
status_change = status_change || updated_status;
}
DPRINTF(Fetch, "Running stage.\n");
if (FullSystem) {
if (fromCommit->commitInfo[0].interruptPending) {
interruptPending = true;
}
if (fromCommit->commitInfo[0].clearInterrupt) {
interruptPending = false;
}
}
for (threadFetched = 0; threadFetched < numFetchingThreads;
threadFetched++) {
// Fetch each of the actively fetching threads.
fetch(status_change);
}
// Record number of instructions fetched this cycle for distribution.
fetchNisnDist.sample(numInst);
if (status_change) {
// Change the fetch stage status if there was a status change.
_status = updateFetchStatus();
}
// If there was activity this cycle, inform the CPU of it.
if (wroteToTimeBuffer || cpu->contextSwitch) {
DPRINTF(Activity, "Activity this cycle.\n");
cpu->activityThisCycle();
}
// Issue the next I-cache request if possible.
for (ThreadID i = 0; i < Impl::MaxThreads; ++i) {
if (issuePipelinedIfetch[i]) {
pipelineIcacheAccesses(i);
}
}
// Reset the number of the instruction we've fetched.
numInst = 0;
}
template <class Impl>
bool
DefaultFetch<Impl>::checkSignalsAndUpdate(ThreadID tid)
{
// Update the per thread stall statuses.
if (fromDecode->decodeBlock[tid]) {
stalls[tid].decode = true;
}
if (fromDecode->decodeUnblock[tid]) {
assert(stalls[tid].decode);
assert(!fromDecode->decodeBlock[tid]);
stalls[tid].decode = false;
}
if (fromRename->renameBlock[tid]) {
stalls[tid].rename = true;
}
if (fromRename->renameUnblock[tid]) {
assert(stalls[tid].rename);
assert(!fromRename->renameBlock[tid]);
stalls[tid].rename = false;
}
if (fromIEW->iewBlock[tid]) {
stalls[tid].iew = true;
}
if (fromIEW->iewUnblock[tid]) {
assert(stalls[tid].iew);
assert(!fromIEW->iewBlock[tid]);
stalls[tid].iew = false;
}
if (fromCommit->commitBlock[tid]) {
stalls[tid].commit = true;
}
if (fromCommit->commitUnblock[tid]) {
assert(stalls[tid].commit);
assert(!fromCommit->commitBlock[tid]);
stalls[tid].commit = false;
}
// Check squash signals from commit.
if (fromCommit->commitInfo[tid].squash) {
DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash "
"from commit.\n",tid);
// In any case, squash.
squash(fromCommit->commitInfo[tid].pc,
fromCommit->commitInfo[tid].doneSeqNum,
fromCommit->commitInfo[tid].squashInst, tid);
// If it was a branch mispredict on a control instruction, update the
// branch predictor with that instruction, otherwise just kill the
// invalid state we generated in after sequence number
if (fromCommit->commitInfo[tid].mispredictInst &&
fromCommit->commitInfo[tid].mispredictInst->isControl()) {
branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum,
fromCommit->commitInfo[tid].pc,
fromCommit->commitInfo[tid].branchTaken,
tid);
} else {
branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum,
tid);
}
return true;
} else if (fromCommit->commitInfo[tid].doneSeqNum) {
// Update the branch predictor if it wasn't a squashed instruction
// that was broadcasted.
branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid);
}
// Check ROB squash signals from commit.
if (fromCommit->commitInfo[tid].robSquashing) {
DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid);
// Continue to squash.
fetchStatus[tid] = Squashing;
return true;
}
// Check squash signals from decode.
if (fromDecode->decodeInfo[tid].squash) {
DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash "
"from decode.\n",tid);
// Update the branch predictor.
if (fromDecode->decodeInfo[tid].branchMispredict) {
branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum,
fromDecode->decodeInfo[tid].nextPC,
fromDecode->decodeInfo[tid].branchTaken,
tid);
} else {
branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum,
tid);
}
if (fetchStatus[tid] != Squashing) {
DPRINTF(Fetch, "Squashing from decode with PC = %s\n",
fromDecode->decodeInfo[tid].nextPC);
// Squash unless we're already squashing
squashFromDecode(fromDecode->decodeInfo[tid].nextPC,
fromDecode->decodeInfo[tid].squashInst,
fromDecode->decodeInfo[tid].doneSeqNum,
tid);
return true;
}
}
if (checkStall(tid) &&
fetchStatus[tid] != IcacheWaitResponse &&
fetchStatus[tid] != IcacheWaitRetry) {
DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid);
fetchStatus[tid] = Blocked;
return true;
}
if (fetchStatus[tid] == Blocked ||
fetchStatus[tid] == Squashing) {
// Switch status to running if fetch isn't being told to block or
// squash this cycle.
DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n",
tid);
fetchStatus[tid] = Running;
return true;
}
// If we've reached this point, we have not gotten any signals that
// cause fetch to change its status. Fetch remains the same as before.
return false;
}
template<class Impl>
typename Impl::DynInstPtr
DefaultFetch<Impl>::buildInst(ThreadID tid, StaticInstPtr staticInst,
StaticInstPtr curMacroop, TheISA::PCState thisPC,
TheISA::PCState nextPC, bool trace)
{
// Get a sequence number.
InstSeqNum seq = cpu->getAndIncrementInstSeq();
// Create a new DynInst from the instruction fetched.
DynInstPtr instruction =
new DynInst(staticInst, curMacroop, thisPC, nextPC, seq, cpu);
instruction->setTid(tid);
instruction->setASID(tid);
instruction->setThreadState(cpu->thread[tid]);
DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x (%d) created "
"[sn:%lli].\n", tid, thisPC.instAddr(),
thisPC.microPC(), seq);
DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", tid,
instruction->staticInst->
disassemble(thisPC.instAddr()));
#if TRACING_ON
if (trace) {
instruction->traceData =
cpu->getTracer()->getInstRecord(curTick(), cpu->tcBase(tid),
instruction->staticInst, thisPC, curMacroop);
}
#else
instruction->traceData = NULL;
#endif
// Add instruction to the CPU's list of instructions.
instruction->setInstListIt(cpu->addInst(instruction));
// Write the instruction to the first slot in the queue
// that heads to decode.
assert(numInst < fetchWidth);
toDecode->insts[toDecode->size++] = instruction;
// Keep track of if we can take an interrupt at this boundary
delayedCommit[tid] = instruction->isDelayedCommit();
return instruction;
}
template<class Impl>
void
DefaultFetch<Impl>::fetch(bool &status_change)
{
//////////////////////////////////////////
// Start actual fetch
//////////////////////////////////////////
ThreadID tid = getFetchingThread(fetchPolicy);
if (tid == InvalidThreadID || drainPending) {
// Breaks looping condition in tick()
threadFetched = numFetchingThreads;
if (numThreads == 1) { // @todo Per-thread stats
profileStall(0);
}
return;
}
DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid);
// The current PC.
TheISA::PCState thisPC = pc[tid];
Addr pcOffset = fetchOffset[tid];
Addr fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask;
bool inRom = isRomMicroPC(thisPC.microPC());
// If returning from the delay of a cache miss, then update the status
// to running, otherwise do the cache access. Possibly move this up
// to tick() function.
if (fetchStatus[tid] == IcacheAccessComplete) {
DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", tid);
fetchStatus[tid] = Running;
status_change = true;
} else if (fetchStatus[tid] == Running) {
// Align the fetch PC so its at the start of a cache block.
Addr block_PC = icacheBlockAlignPC(fetchAddr);
// If buffer is no longer valid or fetchAddr has moved to point
// to the next cache block, AND we have no remaining ucode
// from a macro-op, then start fetch from icache.
if (!(cacheDataValid[tid] && block_PC == cacheDataPC[tid])
&& !inRom && !macroop[tid]) {
DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read "
"instruction, starting at PC %s.\n", tid, thisPC);
fetchCacheLine(fetchAddr, tid, thisPC.instAddr());
if (fetchStatus[tid] == IcacheWaitResponse)
++icacheStallCycles;
else if (fetchStatus[tid] == ItlbWait)
++fetchTlbCycles;
else
++fetchMiscStallCycles;
return;
} else if ((checkInterrupt(thisPC.instAddr()) && !delayedCommit[tid])
|| isSwitchedOut()) {
// Stall CPU if an interrupt is posted and we're not issuing
// an delayed commit micro-op currently (delayed commit instructions
// are not interruptable by interrupts, only faults)
++fetchMiscStallCycles;
DPRINTF(Fetch, "[tid:%i]: Fetch is stalled!\n", tid);
return;
}
} else {
if (fetchStatus[tid] == Idle) {
++fetchIdleCycles;
DPRINTF(Fetch, "[tid:%i]: Fetch is idle!\n", tid);
}
// Status is Idle, so fetch should do nothing.
return;
}
++fetchCycles;
TheISA::PCState nextPC = thisPC;
StaticInstPtr staticInst = NULL;
StaticInstPtr curMacroop = macroop[tid];
// If the read of the first instruction was successful, then grab the
// instructions from the rest of the cache line and put them into the
// queue heading to decode.
DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to "
"decode.\n", tid);
// Need to keep track of whether or not a predicted branch
// ended this fetch block.
bool predictedBranch = false;
TheISA::MachInst *cacheInsts =
reinterpret_cast<TheISA::MachInst *>(cacheData[tid]);
const unsigned numInsts = cacheBlkSize / instSize;
unsigned blkOffset = (fetchAddr - cacheDataPC[tid]) / instSize;
// Loop through instruction memory from the cache.
// Keep issuing while fetchWidth is available and branch is not
// predicted taken
while (numInst < fetchWidth && !predictedBranch) {
// We need to process more memory if we aren't going to get a
// StaticInst from the rom, the current macroop, or what's already
// in the predecoder.
bool needMem = !inRom && !curMacroop && !predecoder.extMachInstReady();
fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask;
Addr block_PC = icacheBlockAlignPC(fetchAddr);
if (needMem) {
// If buffer is no longer valid or fetchAddr has moved to point
// to the next cache block then start fetch from icache.
if (!cacheDataValid[tid] || block_PC != cacheDataPC[tid])
break;
if (blkOffset >= numInsts) {
// We need to process more memory, but we've run out of the
// current block.
break;
}
if (ISA_HAS_DELAY_SLOT && pcOffset == 0) {
// Walk past any annulled delay slot instructions.
Addr pcAddr = thisPC.instAddr() & BaseCPU::PCMask;
while (fetchAddr != pcAddr && blkOffset < numInsts) {
blkOffset++;
fetchAddr += instSize;
}
if (blkOffset >= numInsts)
break;
}
MachInst inst = TheISA::gtoh(cacheInsts[blkOffset]);
predecoder.setTC(cpu->thread[tid]->getTC());
predecoder.moreBytes(thisPC, fetchAddr, inst);
if (predecoder.needMoreBytes()) {
blkOffset++;
fetchAddr += instSize;
pcOffset += instSize;
}
}
// Extract as many instructions and/or microops as we can from
// the memory we've processed so far.
do {
if (!(curMacroop || inRom)) {
if (predecoder.extMachInstReady()) {
ExtMachInst extMachInst =
predecoder.getExtMachInst(thisPC);
staticInst =
decoder.decode(extMachInst, thisPC.instAddr());
// Increment stat of fetched instructions.
++fetchedInsts;
if (staticInst->isMacroop()) {
curMacroop = staticInst;
} else {
pcOffset = 0;
}
} else {
// We need more bytes for this instruction so blkOffset and
// pcOffset will be updated
break;
}
}
// Whether we're moving to a new macroop because we're at the
// end of the current one, or the branch predictor incorrectly
// thinks we are...
bool newMacro = false;
if (curMacroop || inRom) {
if (inRom) {
staticInst = cpu->microcodeRom.fetchMicroop(
thisPC.microPC(), curMacroop);
} else {
staticInst = curMacroop->fetchMicroop(thisPC.microPC());
}
newMacro |= staticInst->isLastMicroop();
}
DynInstPtr instruction =
buildInst(tid, staticInst, curMacroop,
thisPC, nextPC, true);
numInst++;
#if TRACING_ON
instruction->fetchTick = curTick();
#endif
nextPC = thisPC;
// If we're branching after this instruction, quite fetching
// from the same block then.
predictedBranch |= thisPC.branching();
predictedBranch |=
lookupAndUpdateNextPC(instruction, nextPC);
if (predictedBranch) {
DPRINTF(Fetch, "Branch detected with PC = %s\n", thisPC);
}
newMacro |= thisPC.instAddr() != nextPC.instAddr();
// Move to the next instruction, unless we have a branch.
thisPC = nextPC;
inRom = isRomMicroPC(thisPC.microPC());
if (newMacro) {
fetchAddr = thisPC.instAddr() & BaseCPU::PCMask;
blkOffset = (fetchAddr - cacheDataPC[tid]) / instSize;
pcOffset = 0;
curMacroop = NULL;
}
if (instruction->isQuiesce()) {
DPRINTF(Fetch,
"Quiesce instruction encountered, halting fetch!");
fetchStatus[tid] = QuiescePending;
status_change = true;
break;
}
} while ((curMacroop || predecoder.extMachInstReady()) &&
numInst < fetchWidth);
}
if (predictedBranch) {
DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch "
"instruction encountered.\n", tid);
} else if (numInst >= fetchWidth) {
DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth "
"for this cycle.\n", tid);
} else if (blkOffset >= cacheBlkSize) {
DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache "
"block.\n", tid);
}
macroop[tid] = curMacroop;
fetchOffset[tid] = pcOffset;
if (numInst > 0) {
wroteToTimeBuffer = true;
}
pc[tid] = thisPC;
// pipeline a fetch if we're crossing a cache boundary and not in
// a state that would preclude fetching
fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask;
Addr block_PC = icacheBlockAlignPC(fetchAddr);
issuePipelinedIfetch[tid] = block_PC != cacheDataPC[tid] &&
fetchStatus[tid] != IcacheWaitResponse &&
fetchStatus[tid] != ItlbWait &&
fetchStatus[tid] != IcacheWaitRetry &&
fetchStatus[tid] != QuiescePending &&
!curMacroop;
}
template<class Impl>
void
DefaultFetch<Impl>::recvRetry()
{
if (retryPkt != NULL) {
assert(cacheBlocked);
assert(retryTid != InvalidThreadID);
assert(fetchStatus[retryTid] == IcacheWaitRetry);
if (cpu->getIcachePort()->sendTiming(retryPkt)) {
fetchStatus[retryTid] = IcacheWaitResponse;
retryPkt = NULL;
retryTid = InvalidThreadID;
cacheBlocked = false;
}
} else {
assert(retryTid == InvalidThreadID);
// Access has been squashed since it was sent out. Just clear
// the cache being blocked.
cacheBlocked = false;
}
}
///////////////////////////////////////
// //
// SMT FETCH POLICY MAINTAINED HERE //
// //
///////////////////////////////////////
template<class Impl>
ThreadID
DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority)
{
if (numThreads > 1) {
switch (fetch_priority) {
case SingleThread:
return 0;
case RoundRobin:
return roundRobin();
case IQ:
return iqCount();
case LSQ:
return lsqCount();
case Branch:
return branchCount();
default:
return InvalidThreadID;
}
} else {
list<ThreadID>::iterator thread = activeThreads->begin();
if (thread == activeThreads->end()) {
return InvalidThreadID;
}
ThreadID tid = *thread;
if (fetchStatus[tid] == Running ||
fetchStatus[tid] == IcacheAccessComplete ||
fetchStatus[tid] == Idle) {
return tid;
} else {
return InvalidThreadID;
}
}
}
template<class Impl>
ThreadID
DefaultFetch<Impl>::roundRobin()
{
list<ThreadID>::iterator pri_iter = priorityList.begin();
list<ThreadID>::iterator end = priorityList.end();
ThreadID high_pri;
while (pri_iter != end) {
high_pri = *pri_iter;
assert(high_pri <= numThreads);
if (fetchStatus[high_pri] == Running ||
fetchStatus[high_pri] == IcacheAccessComplete ||
fetchStatus[high_pri] == Idle) {
priorityList.erase(pri_iter);
priorityList.push_back(high_pri);
return high_pri;
}
pri_iter++;
}
return InvalidThreadID;
}
template<class Impl>
ThreadID
DefaultFetch<Impl>::iqCount()
{
std::priority_queue<unsigned> PQ;
std::map<unsigned, ThreadID> threadMap;
list<ThreadID>::iterator threads = activeThreads->begin();
list<ThreadID>::iterator end = activeThreads->end();
while (threads != end) {
ThreadID tid = *threads++;
unsigned iqCount = fromIEW->iewInfo[tid].iqCount;
PQ.push(iqCount);
threadMap[iqCount] = tid;
}
while (!PQ.empty()) {
ThreadID high_pri = threadMap[PQ.top()];
if (fetchStatus[high_pri] == Running ||
fetchStatus[high_pri] == IcacheAccessComplete ||
fetchStatus[high_pri] == Idle)
return high_pri;
else
PQ.pop();
}
return InvalidThreadID;
}
template<class Impl>
ThreadID
DefaultFetch<Impl>::lsqCount()
{
std::priority_queue<unsigned> PQ;
std::map<unsigned, ThreadID> threadMap;
list<ThreadID>::iterator threads = activeThreads->begin();
list<ThreadID>::iterator end = activeThreads->end();
while (threads != end) {
ThreadID tid = *threads++;
unsigned ldstqCount = fromIEW->iewInfo[tid].ldstqCount;
PQ.push(ldstqCount);
threadMap[ldstqCount] = tid;
}
while (!PQ.empty()) {
ThreadID high_pri = threadMap[PQ.top()];
if (fetchStatus[high_pri] == Running ||
fetchStatus[high_pri] == IcacheAccessComplete ||
fetchStatus[high_pri] == Idle)
return high_pri;
else
PQ.pop();
}
return InvalidThreadID;
}
template<class Impl>
ThreadID
DefaultFetch<Impl>::branchCount()
{
#if 0
list<ThreadID>::iterator thread = activeThreads->begin();
assert(thread != activeThreads->end());
ThreadID tid = *thread;
#endif
panic("Branch Count Fetch policy unimplemented\n");
return InvalidThreadID;
}
template<class Impl>
void
DefaultFetch<Impl>::pipelineIcacheAccesses(ThreadID tid)
{
if (!issuePipelinedIfetch[tid]) {
return;
}
// The next PC to access.
TheISA::PCState thisPC = pc[tid];
if (isRomMicroPC(thisPC.microPC())) {
return;
}
Addr pcOffset = fetchOffset[tid];
Addr fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask;
// Align the fetch PC so its at the start of a cache block.
Addr block_PC = icacheBlockAlignPC(fetchAddr);
// Unless buffer already got the block, fetch it from icache.
if (!(cacheDataValid[tid] && block_PC == cacheDataPC[tid])) {
DPRINTF(Fetch, "[tid:%i]: Issuing a pipelined I-cache access, "
"starting at PC %s.\n", tid, thisPC);
fetchCacheLine(fetchAddr, tid, thisPC.instAddr());
}
}
template<class Impl>
void
DefaultFetch<Impl>::profileStall(ThreadID tid) {
DPRINTF(Fetch,"There are no more threads available to fetch from.\n");
// @todo Per-thread stats
if (drainPending) {
++fetchPendingDrainCycles;
DPRINTF(Fetch, "Fetch is waiting for a drain!\n");
} else if (activeThreads->empty()) {
++fetchNoActiveThreadStallCycles;
DPRINTF(Fetch, "Fetch has no active thread!\n");
} else if (fetchStatus[tid] == Blocked) {
++fetchBlockedCycles;
DPRINTF(Fetch, "[tid:%i]: Fetch is blocked!\n", tid);
} else if (fetchStatus[tid] == Squashing) {
++fetchSquashCycles;
DPRINTF(Fetch, "[tid:%i]: Fetch is squashing!\n", tid);
} else if (fetchStatus[tid] == IcacheWaitResponse) {
++icacheStallCycles;
DPRINTF(Fetch, "[tid:%i]: Fetch is waiting cache response!\n",
tid);
} else if (fetchStatus[tid] == ItlbWait) {
++fetchTlbCycles;
DPRINTF(Fetch, "[tid:%i]: Fetch is waiting ITLB walk to "
"finish!\n", tid);
} else if (fetchStatus[tid] == TrapPending) {
++fetchPendingTrapStallCycles;
DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for a pending trap!\n",
tid);
} else if (fetchStatus[tid] == QuiescePending) {
++fetchPendingQuiesceStallCycles;
DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for a pending quiesce "
"instruction!\n", tid);
} else if (fetchStatus[tid] == IcacheWaitRetry) {
++fetchIcacheWaitRetryStallCycles;
DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for an I-cache retry!\n",
tid);
} else if (fetchStatus[tid] == NoGoodAddr) {
DPRINTF(Fetch, "[tid:%i]: Fetch predicted non-executable address\n",
tid);
} else {
DPRINTF(Fetch, "[tid:%i]: Unexpected fetch stall reason (Status: %i).\n",
tid, fetchStatus[tid]);
}
}