From 997fc505a8b2ed843f2c87bd2371e6510f731324 Mon Sep 17 00:00:00 2001 From: Steve Reinhardt Date: Wed, 7 Feb 2007 10:53:37 -0800 Subject: Make memory commands dense again to avoid cache stat table explosion. Created MemCmd class to wrap enum and provide handy methods to check attributes, convert to string/int, etc. --HG-- extra : convert_revision : 57f147ad893443e3a2040c6d5b4cdb1a8033930b --- src/cpu/memtest/memtest.cc | 10 +- src/cpu/o3/fetch_impl.hh | 2 +- src/cpu/o3/lsq_unit.hh | 5 +- src/cpu/o3/lsq_unit_impl.hh | 3 +- src/cpu/simple/atomic.cc | 6 +- src/cpu/simple/timing.cc | 6 +- src/dev/io_device.hh | 7 +- src/mem/cache/base_cache.cc | 59 ++++--- src/mem/cache/base_cache.hh | 16 +- src/mem/cache/cache_impl.hh | 38 ++--- src/mem/cache/coherence/coherence_protocol.cc | 146 ++++++++--------- src/mem/cache/coherence/coherence_protocol.hh | 10 +- src/mem/cache/coherence/simple_coherence.hh | 7 +- src/mem/cache/coherence/uni_coherence.cc | 8 +- src/mem/cache/coherence/uni_coherence.hh | 12 +- src/mem/cache/miss/blocking_buffer.cc | 4 +- src/mem/cache/miss/blocking_buffer.hh | 2 +- src/mem/cache/miss/miss_buffer.hh | 2 +- src/mem/cache/miss/miss_queue.cc | 111 +++++++------ src/mem/cache/miss/miss_queue.hh | 20 +-- src/mem/cache/miss/mshr.cc | 4 +- src/mem/cache/miss/mshr.hh | 4 +- src/mem/cache/miss/mshr_queue.cc | 8 +- src/mem/cache/miss/mshr_queue.hh | 2 +- src/mem/cache/prefetch/base_prefetcher.cc | 2 +- src/mem/cache/tags/iic.cc | 3 +- src/mem/packet.cc | 126 +++++++-------- src/mem/packet.hh | 217 +++++++++++++++----------- src/mem/port.cc | 8 +- src/mem/port.hh | 2 +- src/mem/tport.cc | 2 +- 31 files changed, 448 insertions(+), 404 deletions(-) diff --git a/src/cpu/memtest/memtest.cc b/src/cpu/memtest/memtest.cc index 180f41541..8b3e9a11e 100644 --- a/src/cpu/memtest/memtest.cc +++ b/src/cpu/memtest/memtest.cc @@ -226,8 +226,8 @@ MemTest::completeRequest(PacketPtr pkt) assert(removeAddr != outstandingAddrs.end()); outstandingAddrs.erase(removeAddr); - switch (pkt->cmd) { - case Packet::ReadResp: + switch (pkt->cmd.toInt()) { + case MemCmd::ReadResp: if (memcmp(pkt_data, data, pkt->getSize()) != 0) { cerr << name() << ": on read of 0x" << hex << req->getPaddr() @@ -254,7 +254,7 @@ MemTest::completeRequest(PacketPtr pkt) exitSimLoop("Maximum number of loads reached!"); break; - case Packet::WriteResp: + case MemCmd::WriteResp: numWritesStat++; break; /* @@ -389,7 +389,7 @@ MemTest::tick() << dec << curTick << endl; } - PacketPtr pkt = new Packet(req, Packet::ReadReq, Packet::Broadcast); + PacketPtr pkt = new Packet(req, MemCmd::ReadReq, Packet::Broadcast); pkt->dataDynamicArray(new uint8_t[req->getSize()]); MemTestSenderState *state = new MemTestSenderState(result); pkt->senderState = state; @@ -429,7 +429,7 @@ MemTest::tick() << dec << curTick << endl; } */ - PacketPtr pkt = new Packet(req, Packet::WriteReq, Packet::Broadcast); + PacketPtr pkt = new Packet(req, MemCmd::WriteReq, Packet::Broadcast); uint8_t *pkt_data = new uint8_t[req->getSize()]; pkt->dataDynamicArray(pkt_data); memcpy(pkt_data, &data, req->getSize()); diff --git a/src/cpu/o3/fetch_impl.hh b/src/cpu/o3/fetch_impl.hh index b80fc72e1..e6a779823 100644 --- a/src/cpu/o3/fetch_impl.hh +++ b/src/cpu/o3/fetch_impl.hh @@ -602,7 +602,7 @@ DefaultFetch::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid // Build packet here. PacketPtr data_pkt = new Packet(mem_req, - Packet::ReadReq, Packet::Broadcast); + MemCmd::ReadReq, Packet::Broadcast); data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]); cacheDataPC[tid] = block_PC; diff --git a/src/cpu/o3/lsq_unit.hh b/src/cpu/o3/lsq_unit.hh index 9c7eb7780..2419afe29 100644 --- a/src/cpu/o3/lsq_unit.hh +++ b/src/cpu/o3/lsq_unit.hh @@ -574,7 +574,8 @@ LSQUnit::read(Request *req, T &data, int load_idx) "addr %#x, data %#x\n", store_idx, req->getVaddr(), data); - PacketPtr data_pkt = new Packet(req, Packet::ReadReq, Packet::Broadcast); + PacketPtr data_pkt = new Packet(req, MemCmd::ReadReq, + Packet::Broadcast); data_pkt->dataStatic(load_inst->memData); WritebackEvent *wb = new WritebackEvent(load_inst, data_pkt, this); @@ -638,7 +639,7 @@ LSQUnit::read(Request *req, T &data, int load_idx) // if we the cache is not blocked, do cache access if (!lsq->cacheBlocked()) { PacketPtr data_pkt = - new Packet(req, Packet::ReadReq, Packet::Broadcast); + new Packet(req, MemCmd::ReadReq, Packet::Broadcast); data_pkt->dataStatic(load_inst->memData); LSQSenderState *state = new LSQSenderState; diff --git a/src/cpu/o3/lsq_unit_impl.hh b/src/cpu/o3/lsq_unit_impl.hh index ebd9301f6..3ba22a530 100644 --- a/src/cpu/o3/lsq_unit_impl.hh +++ b/src/cpu/o3/lsq_unit_impl.hh @@ -613,7 +613,8 @@ LSQUnit::writebackStores() (sizeof(TheISA::IntReg) - req->getSize()) : 0), req->getSize()); - PacketPtr data_pkt = new Packet(req, Packet::WriteReq, Packet::Broadcast); + PacketPtr data_pkt = new Packet(req, MemCmd::WriteReq, + Packet::Broadcast); data_pkt->dataStatic(inst->memData); LSQSenderState *state = new LSQSenderState; diff --git a/src/cpu/simple/atomic.cc b/src/cpu/simple/atomic.cc index 3b3536e44..fa47b0eee 100644 --- a/src/cpu/simple/atomic.cc +++ b/src/cpu/simple/atomic.cc @@ -138,18 +138,18 @@ AtomicSimpleCPU::AtomicSimpleCPU(Params *p) ifetch_req = new Request(); ifetch_req->setThreadContext(p->cpu_id, 0); // Add thread ID if we add MT - ifetch_pkt = new Packet(ifetch_req, Packet::ReadReq, Packet::Broadcast); + ifetch_pkt = new Packet(ifetch_req, MemCmd::ReadReq, Packet::Broadcast); ifetch_pkt->dataStatic(&inst); data_read_req = new Request(); data_read_req->setThreadContext(p->cpu_id, 0); // Add thread ID here too - data_read_pkt = new Packet(data_read_req, Packet::ReadReq, + data_read_pkt = new Packet(data_read_req, MemCmd::ReadReq, Packet::Broadcast); data_read_pkt->dataStatic(&dataReg); data_write_req = new Request(); data_write_req->setThreadContext(p->cpu_id, 0); // Add thread ID here too - data_write_pkt = new Packet(data_write_req, Packet::WriteReq, + data_write_pkt = new Packet(data_write_req, MemCmd::WriteReq, Packet::Broadcast); } diff --git a/src/cpu/simple/timing.cc b/src/cpu/simple/timing.cc index dfffb0b1f..e4748c966 100644 --- a/src/cpu/simple/timing.cc +++ b/src/cpu/simple/timing.cc @@ -286,7 +286,7 @@ TimingSimpleCPU::read(Addr addr, T &data, unsigned flags) // Now do the access. if (fault == NoFault) { PacketPtr pkt = - new Packet(req, Packet::ReadReq, Packet::Broadcast); + new Packet(req, MemCmd::ReadReq, Packet::Broadcast); pkt->dataDynamic(new T); if (!dcachePort.sendTiming(pkt)) { @@ -365,7 +365,7 @@ TimingSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) // Now do the access. if (fault == NoFault) { assert(dcache_pkt == NULL); - dcache_pkt = new Packet(req, Packet::WriteReq, Packet::Broadcast); + dcache_pkt = new Packet(req, MemCmd::WriteReq, Packet::Broadcast); dcache_pkt->allocate(); dcache_pkt->set(data); @@ -454,7 +454,7 @@ TimingSimpleCPU::fetch() ifetch_req->setThreadContext(cpu_id, /* thread ID */ 0); Fault fault = setupFetchRequest(ifetch_req); - ifetch_pkt = new Packet(ifetch_req, Packet::ReadReq, Packet::Broadcast); + ifetch_pkt = new Packet(ifetch_req, MemCmd::ReadReq, Packet::Broadcast); ifetch_pkt->dataStatic(&inst); if (fault == NoFault) { diff --git a/src/dev/io_device.hh b/src/dev/io_device.hh index c56eba267..902cde909 100644 --- a/src/dev/io_device.hh +++ b/src/dev/io_device.hh @@ -256,10 +256,13 @@ class DmaDevice : public PioDevice virtual ~DmaDevice(); void dmaWrite(Addr addr, int size, Event *event, uint8_t *data) - { dmaPort->dmaAction(Packet::WriteInvalidateReq, addr, size, event, data) ; } + { + dmaPort->dmaAction(MemCmd::WriteInvalidateReq, + addr, size, event, data); + } void dmaRead(Addr addr, int size, Event *event, uint8_t *data = NULL) - { dmaPort->dmaAction(Packet::ReadReq, addr, size, event, data); } + { dmaPort->dmaAction(MemCmd::ReadReq, addr, size, event, data); } bool dmaPending() { return dmaPort->dmaPending(); } diff --git a/src/mem/cache/base_cache.cc b/src/mem/cache/base_cache.cc index ace46d752..ed665dafb 100644 --- a/src/mem/cache/base_cache.cc +++ b/src/mem/cache/base_cache.cc @@ -370,17 +370,12 @@ BaseCache::init() void BaseCache::regStats() { - Request temp_req((Addr) NULL, 4, 0); - Packet::Command temp_cmd = Packet::ReadReq; - Packet temp_pkt(&temp_req, temp_cmd, 0); //@todo FIx command strings so this isn't neccessary - temp_pkt.allocate(); //Temp allocate, all need data - using namespace Stats; // Hit statistics - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); hits[access_idx] .init(maxThreadsPerCPU) @@ -395,20 +390,20 @@ BaseCache::regStats() .desc("number of demand (read+write) hits") .flags(total) ; - demandHits = hits[Packet::ReadReq] + hits[Packet::WriteReq]; + demandHits = hits[MemCmd::ReadReq] + hits[MemCmd::WriteReq]; overallHits .name(name() + ".overall_hits") .desc("number of overall hits") .flags(total) ; - overallHits = demandHits + hits[Packet::SoftPFReq] + hits[Packet::HardPFReq] - + hits[Packet::Writeback]; + overallHits = demandHits + hits[MemCmd::SoftPFReq] + hits[MemCmd::HardPFReq] + + hits[MemCmd::Writeback]; // Miss statistics - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); misses[access_idx] .init(maxThreadsPerCPU) @@ -423,20 +418,20 @@ BaseCache::regStats() .desc("number of demand (read+write) misses") .flags(total) ; - demandMisses = misses[Packet::ReadReq] + misses[Packet::WriteReq]; + demandMisses = misses[MemCmd::ReadReq] + misses[MemCmd::WriteReq]; overallMisses .name(name() + ".overall_misses") .desc("number of overall misses") .flags(total) ; - overallMisses = demandMisses + misses[Packet::SoftPFReq] + - misses[Packet::HardPFReq] + misses[Packet::Writeback]; + overallMisses = demandMisses + misses[MemCmd::SoftPFReq] + + misses[MemCmd::HardPFReq] + misses[MemCmd::Writeback]; // Miss latency statistics - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); missLatency[access_idx] .init(maxThreadsPerCPU) @@ -451,20 +446,20 @@ BaseCache::regStats() .desc("number of demand (read+write) miss cycles") .flags(total) ; - demandMissLatency = missLatency[Packet::ReadReq] + missLatency[Packet::WriteReq]; + demandMissLatency = missLatency[MemCmd::ReadReq] + missLatency[MemCmd::WriteReq]; overallMissLatency .name(name() + ".overall_miss_latency") .desc("number of overall miss cycles") .flags(total) ; - overallMissLatency = demandMissLatency + missLatency[Packet::SoftPFReq] + - missLatency[Packet::HardPFReq]; + overallMissLatency = demandMissLatency + missLatency[MemCmd::SoftPFReq] + + missLatency[MemCmd::HardPFReq]; // access formulas - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); accesses[access_idx] .name(name() + "." + cstr + "_accesses") @@ -490,9 +485,9 @@ BaseCache::regStats() overallAccesses = overallHits + overallMisses; // miss rate formulas - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); missRate[access_idx] .name(name() + "." + cstr + "_miss_rate") @@ -518,9 +513,9 @@ BaseCache::regStats() overallMissRate = overallMisses / overallAccesses; // miss latency formulas - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); avgMissLatency[access_idx] .name(name() + "." + cstr + "_avg_miss_latency") diff --git a/src/mem/cache/base_cache.hh b/src/mem/cache/base_cache.hh index c10d98e8e..ee871c1c4 100644 --- a/src/mem/cache/base_cache.hh +++ b/src/mem/cache/base_cache.hh @@ -200,14 +200,14 @@ class BaseCache : public MemObject */ /** Number of hits per thread for each type of command. @sa Packet::Command */ - Stats::Vector<> hits[NUM_MEM_CMDS]; + Stats::Vector<> hits[MemCmd::NUM_MEM_CMDS]; /** Number of hits for demand accesses. */ Stats::Formula demandHits; /** Number of hit for all accesses. */ Stats::Formula overallHits; /** Number of misses per thread for each type of command. @sa Packet::Command */ - Stats::Vector<> misses[NUM_MEM_CMDS]; + Stats::Vector<> misses[MemCmd::NUM_MEM_CMDS]; /** Number of misses for demand accesses. */ Stats::Formula demandMisses; /** Number of misses for all accesses. */ @@ -217,28 +217,28 @@ class BaseCache : public MemObject * Total number of cycles per thread/command spent waiting for a miss. * Used to calculate the average miss latency. */ - Stats::Vector<> missLatency[NUM_MEM_CMDS]; + Stats::Vector<> missLatency[MemCmd::NUM_MEM_CMDS]; /** Total number of cycles spent waiting for demand misses. */ Stats::Formula demandMissLatency; /** Total number of cycles spent waiting for all misses. */ Stats::Formula overallMissLatency; /** The number of accesses per command and thread. */ - Stats::Formula accesses[NUM_MEM_CMDS]; + Stats::Formula accesses[MemCmd::NUM_MEM_CMDS]; /** The number of demand accesses. */ Stats::Formula demandAccesses; /** The number of overall accesses. */ Stats::Formula overallAccesses; /** The miss rate per command and thread. */ - Stats::Formula missRate[NUM_MEM_CMDS]; + Stats::Formula missRate[MemCmd::NUM_MEM_CMDS]; /** The miss rate of all demand accesses. */ Stats::Formula demandMissRate; /** The miss rate for all accesses. */ Stats::Formula overallMissRate; /** The average miss latency per command and thread. */ - Stats::Formula avgMissLatency[NUM_MEM_CMDS]; + Stats::Formula avgMissLatency[MemCmd::NUM_MEM_CMDS]; /** The average miss latency for demand misses. */ Stats::Formula demandAvgMissLatency; /** The average miss latency for all misses. */ @@ -535,7 +535,7 @@ class BaseCache : public MemObject } } else { - if (pkt->cmd != Packet::UpgradeReq) + if (pkt->cmd != MemCmd::UpgradeReq) { delete pkt->req; delete pkt; @@ -594,7 +594,7 @@ class BaseCache : public MemObject } } else { - if (pkt->cmd != Packet::UpgradeReq) + if (pkt->cmd != MemCmd::UpgradeReq) { delete pkt->req; delete pkt; diff --git a/src/mem/cache/cache_impl.hh b/src/mem/cache/cache_impl.hh index b8c896498..57ebff322 100644 --- a/src/mem/cache/cache_impl.hh +++ b/src/mem/cache/cache_impl.hh @@ -90,7 +90,7 @@ Cache(const std::string &_name, coherence->setCache(this); prefetcher->setCache(this); invalidateReq = new Request((Addr) NULL, blkSize, 0); - invalidatePkt = new Packet(invalidateReq, Packet::InvalidateReq, 0); + invalidatePkt = new Packet(invalidateReq, MemCmd::InvalidateReq, 0); } template @@ -239,7 +239,7 @@ Cache::handleFill(BlkType *blk, PacketPtr &pkt, target->flags |= SATISFIED; - if (target->cmd == Packet::InvalidateReq) { + if (target->cmd == MemCmd::InvalidateReq) { tags->invalidateBlk(blk); blk = NULL; } @@ -316,7 +316,7 @@ Cache::handleFill(BlkType *blk, MSHR * mshr, Tick completion_time = tags->getHitLatency() + transfer_offset ? pkt->finishTime : pkt->firstWordTime; - if (target->cmd == Packet::InvalidateReq) { + if (target->cmd == MemCmd::InvalidateReq) { //Mark the blk as invalid now, if it hasn't been already if (blk) { tags->invalidateBlk(blk); @@ -430,7 +430,7 @@ Cache::writebackBlk(BlkType *blk) Request *writebackReq = new Request(tags->regenerateBlkAddr(blk->tag, blk->set), blkSize, 0); - PacketPtr writeback = new Packet(writebackReq, Packet::Writeback, -1); + PacketPtr writeback = new Packet(writebackReq, MemCmd::Writeback, -1); writeback->allocate(); std::memcpy(writeback->getPtr(),blk->data,blkSize); @@ -555,11 +555,11 @@ Cache::access(PacketPtr &pkt) /** @todo make the fast write alloc (wh64) work with coherence. */ /** @todo Do we want to do fast writes for writebacks as well? */ if (!blk && pkt->getSize() >= blkSize && coherence->allowFastWrites() && - (pkt->cmd == Packet::WriteReq - || pkt->cmd == Packet::WriteInvalidateReq) ) { + (pkt->cmd == MemCmd::WriteReq + || pkt->cmd == MemCmd::WriteInvalidateReq) ) { // not outstanding misses, can do this MSHR* outstanding_miss = missQueue->findMSHR(pkt->getAddr()); - if (pkt->cmd == Packet::WriteInvalidateReq || !outstanding_miss) { + if (pkt->cmd == MemCmd::WriteInvalidateReq || !outstanding_miss) { if (outstanding_miss) { warn("WriteInv doing a fastallocate" "with an outstanding miss to the same address\n"); @@ -583,7 +583,7 @@ Cache::access(PacketPtr &pkt) // clear dirty bit if write through if (pkt->needsResponse()) respond(pkt, curTick+lat); - if (pkt->cmd == Packet::Writeback) { + if (pkt->cmd == MemCmd::Writeback) { //Signal that you can kill the pkt/req pkt->flags |= SATISFIED; } @@ -610,7 +610,7 @@ Cache::access(PacketPtr &pkt) missQueue->handleMiss(pkt, size, curTick + hitLatency); } - if (pkt->cmd == Packet::Writeback) { + if (pkt->cmd == MemCmd::Writeback) { //Need to clean up the packet on a writeback miss, but leave the request delete pkt; } @@ -627,11 +627,11 @@ Cache::getPacket() PacketPtr pkt = missQueue->getPacket(); if (pkt) { if (!pkt->req->isUncacheable()) { - if (pkt->cmd == Packet::HardPFReq) - misses[Packet::HardPFReq][0/*pkt->req->getThreadNum()*/]++; + if (pkt->cmd == MemCmd::HardPFReq) + misses[MemCmd::HardPFReq][0/*pkt->req->getThreadNum()*/]++; BlkType *blk = tags->findBlock(pkt->getAddr()); - Packet::Command cmd = coherence->getBusCmd(pkt->cmd, - (blk)? blk->status : 0); + MemCmd cmd = + coherence->getBusCmd(pkt->cmd, (blk) ? blk->status : 0); missQueue->setBusCmd(pkt, cmd); } } @@ -650,7 +650,7 @@ Cache::sendResult(PacketPtr &pkt, MSHR* mshr, if (success && !(SIGNAL_NACK_HACK)) { //Remember if it was an upgrade because writeback MSHR's are removed //in Mark in Service - bool upgrade = (mshr->pkt && mshr->pkt->cmd == Packet::UpgradeReq); + bool upgrade = (mshr->pkt && mshr->pkt->cmd == MemCmd::UpgradeReq); missQueue->markInService(mshr->pkt, mshr); @@ -775,8 +775,8 @@ Cache::snoop(PacketPtr &pkt) if (mshr) { if (mshr->inService) { if ((mshr->pkt->isInvalidate() || !mshr->pkt->isCacheFill()) - && (pkt->cmd != Packet::InvalidateReq - && pkt->cmd != Packet::WriteInvalidateReq)) { + && (pkt->cmd != MemCmd::InvalidateReq + && pkt->cmd != MemCmd::WriteInvalidateReq)) { //If the outstanding request was an invalidate //(upgrade,readex,..) Then we need to ACK the request //until we get the data Also NACK if the outstanding @@ -982,11 +982,11 @@ Cache::probe(PacketPtr &pkt, bool update, panic("Atomic access ran into outstanding MSHR's or WB's!"); } if (!pkt->req->isUncacheable() /*Uncacheables just go through*/ - && (pkt->cmd != Packet::Writeback)/*Writebacks on miss fall through*/) { + && (pkt->cmd != MemCmd::Writeback)/*Writebacks on miss fall through*/) { // Fetch the cache block to fill BlkType *blk = tags->findBlock(pkt->getAddr()); - Packet::Command temp_cmd = coherence->getBusCmd(pkt->cmd, - (blk)? blk->status : 0); + MemCmd temp_cmd = + coherence->getBusCmd(pkt->cmd, (blk) ? blk->status : 0); PacketPtr busPkt = new Packet(pkt->req,temp_cmd, -1, blkSize); diff --git a/src/mem/cache/coherence/coherence_protocol.cc b/src/mem/cache/coherence/coherence_protocol.cc index 3d7721805..e8520401d 100644 --- a/src/mem/cache/coherence/coherence_protocol.cc +++ b/src/mem/cache/coherence/coherence_protocol.cc @@ -47,7 +47,7 @@ using namespace std; CoherenceProtocol::StateTransition::StateTransition() - : busCmd(Packet::InvalidCmd), newState(-1), snoopFunc(invalidTransition) + : busCmd(MemCmd::InvalidCmd), newState(-1), snoopFunc(invalidTransition) { } @@ -59,132 +59,132 @@ CoherenceProtocol::regStats() // requestCount and snoopCount arrays, most of these are invalid, // so we just select the interesting ones to print here. - requestCount[Invalid][Packet::ReadReq] + requestCount[Invalid][MemCmd::ReadReq] .name(name() + ".read_invalid") .desc("read misses to invalid blocks") ; - requestCount[Invalid][Packet::WriteReq] + requestCount[Invalid][MemCmd::WriteReq] .name(name() +".write_invalid") .desc("write misses to invalid blocks") ; - requestCount[Invalid][Packet::SoftPFReq] + requestCount[Invalid][MemCmd::SoftPFReq] .name(name() +".swpf_invalid") .desc("soft prefetch misses to invalid blocks") ; - requestCount[Invalid][Packet::HardPFReq] + requestCount[Invalid][MemCmd::HardPFReq] .name(name() +".hwpf_invalid") .desc("hard prefetch misses to invalid blocks") ; - requestCount[Shared][Packet::WriteReq] + requestCount[Shared][MemCmd::WriteReq] .name(name() + ".write_shared") .desc("write misses to shared blocks") ; - requestCount[Owned][Packet::WriteReq] + requestCount[Owned][MemCmd::WriteReq] .name(name() + ".write_owned") .desc("write misses to owned blocks") ; - snoopCount[Shared][Packet::ReadReq] + snoopCount[Shared][MemCmd::ReadReq] .name(name() + ".snoop_read_shared") .desc("read snoops on shared blocks") ; - snoopCount[Shared][Packet::ReadExReq] + snoopCount[Shared][MemCmd::ReadExReq] .name(name() + ".snoop_readex_shared") .desc("readEx snoops on shared blocks") ; - snoopCount[Shared][Packet::UpgradeReq] + snoopCount[Shared][MemCmd::UpgradeReq] .name(name() + ".snoop_upgrade_shared") .desc("upgradee snoops on shared blocks") ; - snoopCount[Modified][Packet::ReadReq] + snoopCount[Modified][MemCmd::ReadReq] .name(name() + ".snoop_read_modified") .desc("read snoops on modified blocks") ; - snoopCount[Modified][Packet::ReadExReq] + snoopCount[Modified][MemCmd::ReadExReq] .name(name() + ".snoop_readex_modified") .desc("readEx snoops on modified blocks") ; - snoopCount[Owned][Packet::ReadReq] + snoopCount[Owned][MemCmd::ReadReq] .name(name() + ".snoop_read_owned") .desc("read snoops on owned blocks") ; - snoopCount[Owned][Packet::ReadExReq] + snoopCount[Owned][MemCmd::ReadExReq] .name(name() + ".snoop_readex_owned") .desc("readEx snoops on owned blocks") ; - snoopCount[Owned][Packet::UpgradeReq] + snoopCount[Owned][MemCmd::UpgradeReq] .name(name() + ".snoop_upgrade_owned") .desc("upgrade snoops on owned blocks") ; - snoopCount[Exclusive][Packet::ReadReq] + snoopCount[Exclusive][MemCmd::ReadReq] .name(name() + ".snoop_read_exclusive") .desc("read snoops on exclusive blocks") ; - snoopCount[Exclusive][Packet::ReadExReq] + snoopCount[Exclusive][MemCmd::ReadExReq] .name(name() + ".snoop_readex_exclusive") .desc("readEx snoops on exclusive blocks") ; - snoopCount[Shared][Packet::InvalidateReq] + snoopCount[Shared][MemCmd::InvalidateReq] .name(name() + ".snoop_inv_shared") .desc("Invalidate snoops on shared blocks") ; - snoopCount[Owned][Packet::InvalidateReq] + snoopCount[Owned][MemCmd::InvalidateReq] .name(name() + ".snoop_inv_owned") .desc("Invalidate snoops on owned blocks") ; - snoopCount[Exclusive][Packet::InvalidateReq] + snoopCount[Exclusive][MemCmd::InvalidateReq] .name(name() + ".snoop_inv_exclusive") .desc("Invalidate snoops on exclusive blocks") ; - snoopCount[Modified][Packet::InvalidateReq] + snoopCount[Modified][MemCmd::InvalidateReq] .name(name() + ".snoop_inv_modified") .desc("Invalidate snoops on modified blocks") ; - snoopCount[Invalid][Packet::InvalidateReq] + snoopCount[Invalid][MemCmd::InvalidateReq] .name(name() + ".snoop_inv_invalid") .desc("Invalidate snoops on invalid blocks") ; - snoopCount[Shared][Packet::WriteInvalidateReq] + snoopCount[Shared][MemCmd::WriteInvalidateReq] .name(name() + ".snoop_writeinv_shared") .desc("WriteInvalidate snoops on shared blocks") ; - snoopCount[Owned][Packet::WriteInvalidateReq] + snoopCount[Owned][MemCmd::WriteInvalidateReq] .name(name() + ".snoop_writeinv_owned") .desc("WriteInvalidate snoops on owned blocks") ; - snoopCount[Exclusive][Packet::WriteInvalidateReq] + snoopCount[Exclusive][MemCmd::WriteInvalidateReq] .name(name() + ".snoop_writeinv_exclusive") .desc("WriteInvalidate snoops on exclusive blocks") ; - snoopCount[Modified][Packet::WriteInvalidateReq] + snoopCount[Modified][MemCmd::WriteInvalidateReq] .name(name() + ".snoop_writeinv_modified") .desc("WriteInvalidate snoops on modified blocks") ; - snoopCount[Invalid][Packet::WriteInvalidateReq] + snoopCount[Invalid][MemCmd::WriteInvalidateReq] .name(name() + ".snoop_writeinv_invalid") .desc("WriteInvalidate snoops on invalid blocks") ; @@ -278,11 +278,13 @@ CoherenceProtocol::CoherenceProtocol(const string &name, } // set up a few shortcuts to save typing & visual clutter - typedef Packet P; - StateTransition (&tt)[stateMax+1][NUM_MEM_CMDS] = transitionTable; + typedef MemCmd MC; + StateTransition (&tt)[stateMax+1][MC::NUM_MEM_CMDS] = transitionTable; - P::Command writeToSharedCmd = doUpgrades ? P::UpgradeReq : P::ReadExReq; - P::Command writeToSharedResp = doUpgrades ? P::UpgradeReq : P::ReadExResp; + MC::Command writeToSharedCmd = + doUpgrades ? MC::UpgradeReq : MC::ReadExReq; + MC::Command writeToSharedResp = + doUpgrades ? MC::UpgradeReq : MC::ReadExResp; // Note that all transitions by default cause a panic. // Override the valid transitions with the appropriate actions here. @@ -290,34 +292,34 @@ CoherenceProtocol::CoherenceProtocol(const string &name, // // ----- incoming requests: specify outgoing bus request ----- // - tt[Invalid][P::ReadReq].onRequest(P::ReadReq); + tt[Invalid][MC::ReadReq].onRequest(MC::ReadReq); // we only support write allocate right now - tt[Invalid][P::WriteReq].onRequest(P::ReadExReq); - tt[Shared][P::WriteReq].onRequest(writeToSharedCmd); + tt[Invalid][MC::WriteReq].onRequest(MC::ReadExReq); + tt[Shared][MC::WriteReq].onRequest(writeToSharedCmd); if (hasOwned) { - tt[Owned][P::WriteReq].onRequest(writeToSharedCmd); + tt[Owned][MC::WriteReq].onRequest(writeToSharedCmd); } // Prefetching causes a read - tt[Invalid][P::SoftPFReq].onRequest(P::ReadReq); - tt[Invalid][P::HardPFReq].onRequest(P::ReadReq); + tt[Invalid][MC::SoftPFReq].onRequest(MC::ReadReq); + tt[Invalid][MC::HardPFReq].onRequest(MC::ReadReq); // // ----- on response to given request: specify new state ----- // - tt[Invalid][P::ReadExResp].onResponse(Modified); + tt[Invalid][MC::ReadExResp].onResponse(Modified); tt[Shared][writeToSharedResp].onResponse(Modified); // Go to Exclusive state on read response if we have one (will // move into shared if the shared line is asserted in the // getNewState function) // // originally had this as: - // tt[Invalid][P::ReadResp].onResponse(hasExclusive ? Exclusive: Shared); + // tt[Invalid][MC::ReadResp].onResponse(hasExclusive ? Exclusive: Shared); // ...but for some reason that caused a link error... if (hasExclusive) { - tt[Invalid][P::ReadResp].onResponse(Exclusive); + tt[Invalid][MC::ReadResp].onResponse(Exclusive); } else { - tt[Invalid][P::ReadResp].onResponse(Shared); + tt[Invalid][MC::ReadResp].onResponse(Shared); } if (hasOwned) { tt[Owned][writeToSharedResp].onResponse(Modified); @@ -326,58 +328,58 @@ CoherenceProtocol::CoherenceProtocol(const string &name, // // ----- bus snoop transition functions ----- // - tt[Invalid][P::ReadReq].onSnoop(nullTransition); - tt[Invalid][P::ReadExReq].onSnoop(nullTransition); - tt[Invalid][P::InvalidateReq].onSnoop(invalidateTrans); - tt[Invalid][P::WriteInvalidateReq].onSnoop(invalidateTrans); - tt[Shared][P::ReadReq].onSnoop(hasExclusive + tt[Invalid][MC::ReadReq].onSnoop(nullTransition); + tt[Invalid][MC::ReadExReq].onSnoop(nullTransition); + tt[Invalid][MC::InvalidateReq].onSnoop(invalidateTrans); + tt[Invalid][MC::WriteInvalidateReq].onSnoop(invalidateTrans); + tt[Shared][MC::ReadReq].onSnoop(hasExclusive ? assertShared : nullTransition); - tt[Shared][P::ReadExReq].onSnoop(invalidateTrans); - tt[Shared][P::InvalidateReq].onSnoop(invalidateTrans); - tt[Shared][P::WriteInvalidateReq].onSnoop(invalidateTrans); + tt[Shared][MC::ReadExReq].onSnoop(invalidateTrans); + tt[Shared][MC::InvalidateReq].onSnoop(invalidateTrans); + tt[Shared][MC::WriteInvalidateReq].onSnoop(invalidateTrans); if (doUpgrades) { - tt[Invalid][P::UpgradeReq].onSnoop(nullTransition); - tt[Shared][P::UpgradeReq].onSnoop(invalidateTrans); + tt[Invalid][MC::UpgradeReq].onSnoop(nullTransition); + tt[Shared][MC::UpgradeReq].onSnoop(invalidateTrans); } - tt[Modified][P::ReadExReq].onSnoop(supplyAndInvalidateTrans); - tt[Modified][P::ReadReq].onSnoop(hasOwned + tt[Modified][MC::ReadExReq].onSnoop(supplyAndInvalidateTrans); + tt[Modified][MC::ReadReq].onSnoop(hasOwned ? supplyAndGotoOwnedTrans : supplyAndGotoSharedTrans); - tt[Modified][P::InvalidateReq].onSnoop(invalidateTrans); - tt[Modified][P::WriteInvalidateReq].onSnoop(invalidateTrans); + tt[Modified][MC::InvalidateReq].onSnoop(invalidateTrans); + tt[Modified][MC::WriteInvalidateReq].onSnoop(invalidateTrans); if (hasExclusive) { - tt[Exclusive][P::ReadReq].onSnoop(assertShared); - tt[Exclusive][P::ReadExReq].onSnoop(invalidateTrans); - tt[Exclusive][P::InvalidateReq].onSnoop(invalidateTrans); - tt[Exclusive][P::WriteInvalidateReq].onSnoop(invalidateTrans); + tt[Exclusive][MC::ReadReq].onSnoop(assertShared); + tt[Exclusive][MC::ReadExReq].onSnoop(invalidateTrans); + tt[Exclusive][MC::InvalidateReq].onSnoop(invalidateTrans); + tt[Exclusive][MC::WriteInvalidateReq].onSnoop(invalidateTrans); } if (hasOwned) { - tt[Owned][P::ReadReq].onSnoop(supplyAndGotoOwnedTrans); - tt[Owned][P::ReadExReq].onSnoop(supplyAndInvalidateTrans); - tt[Owned][P::UpgradeReq].onSnoop(invalidateTrans); - tt[Owned][P::InvalidateReq].onSnoop(invalidateTrans); - tt[Owned][P::WriteInvalidateReq].onSnoop(invalidateTrans); + tt[Owned][MC::ReadReq].onSnoop(supplyAndGotoOwnedTrans); + tt[Owned][MC::ReadExReq].onSnoop(supplyAndInvalidateTrans); + tt[Owned][MC::UpgradeReq].onSnoop(invalidateTrans); + tt[Owned][MC::InvalidateReq].onSnoop(invalidateTrans); + tt[Owned][MC::WriteInvalidateReq].onSnoop(invalidateTrans); } // @todo add in hardware prefetch to this list } -Packet::Command -CoherenceProtocol::getBusCmd(Packet::Command cmdIn, CacheBlk::State state, +MemCmd +CoherenceProtocol::getBusCmd(MemCmd cmdIn, CacheBlk::State state, MSHR *mshr) { state &= stateMask; - int cmd_idx = (int) cmdIn; + int cmd_idx = cmdIn.toInt(); assert(0 <= state && state <= stateMax); - assert(0 <= cmd_idx && cmd_idx < NUM_MEM_CMDS); + assert(0 <= cmd_idx && cmd_idx < MemCmd::NUM_MEM_CMDS); - Packet::Command cmdOut = transitionTable[state][cmd_idx].busCmd; + MemCmd::Command cmdOut = transitionTable[state][cmd_idx].busCmd; - assert(cmdOut != Packet::InvalidCmd); + assert(cmdOut != MemCmd::InvalidCmd); ++requestCount[state][cmd_idx]; @@ -392,7 +394,7 @@ CoherenceProtocol::getNewState(PacketPtr &pkt, CacheBlk::State oldState) int cmd_idx = pkt->cmdToIndex(); assert(0 <= state && state <= stateMax); - assert(0 <= cmd_idx && cmd_idx < NUM_MEM_CMDS); + assert(0 <= cmd_idx && cmd_idx < MemCmd::NUM_MEM_CMDS); CacheBlk::State newState = transitionTable[state][cmd_idx].newState; @@ -425,7 +427,7 @@ CoherenceProtocol::handleBusRequest(BaseCache *cache, PacketPtr &pkt, int cmd_idx = pkt->cmdToIndex(); assert(0 <= state && state <= stateMax); - assert(0 <= cmd_idx && cmd_idx < NUM_MEM_CMDS); + assert(0 <= cmd_idx && cmd_idx < MemCmd::NUM_MEM_CMDS); // assert(mshr == NULL); // can't currently handle outstanding requests //Check first if MSHR, and also insure, if there is one, that it is not in service diff --git a/src/mem/cache/coherence/coherence_protocol.hh b/src/mem/cache/coherence/coherence_protocol.hh index 481277523..775bc807a 100644 --- a/src/mem/cache/coherence/coherence_protocol.hh +++ b/src/mem/cache/coherence/coherence_protocol.hh @@ -80,8 +80,8 @@ class CoherenceProtocol : public SimObject * @param mshr The MSHR matching the request. * @return The proper bus command, as determined by the protocol. */ - Packet::Command getBusCmd(Packet::Command cmd, CacheBlk::State status, - MSHR *mshr = NULL); + MemCmd getBusCmd(MemCmd cmd, CacheBlk::State status, + MSHR *mshr = NULL); /** * Return the proper state given the current state and the bus response. @@ -235,7 +235,7 @@ class CoherenceProtocol : public SimObject * The table of all possible transitions, organized by starting state and * request command. */ - StateTransition transitionTable[stateMax+1][NUM_MEM_CMDS]; + StateTransition transitionTable[stateMax+1][MemCmd::NUM_MEM_CMDS]; /** * @addtogroup CoherenceStatistics @@ -244,11 +244,11 @@ class CoherenceProtocol : public SimObject /** * State accesses from parent cache. */ - Stats::Scalar<> requestCount[stateMax+1][NUM_MEM_CMDS]; + Stats::Scalar<> requestCount[stateMax+1][MemCmd::NUM_MEM_CMDS]; /** * State accesses from snooped requests. */ - Stats::Scalar<> snoopCount[stateMax+1][NUM_MEM_CMDS]; + Stats::Scalar<> snoopCount[stateMax+1][MemCmd::NUM_MEM_CMDS]; /** * @} */ diff --git a/src/mem/cache/coherence/simple_coherence.hh b/src/mem/cache/coherence/simple_coherence.hh index a1fd33080..1c89c703a 100644 --- a/src/mem/cache/coherence/simple_coherence.hh +++ b/src/mem/cache/coherence/simple_coherence.hh @@ -131,7 +131,7 @@ class SimpleCoherence //Got rid of, there could be an MSHR, but it can't be in service if (blk != NULL) { - if (pkt->cmd != Packet::Writeback) { + if (pkt->cmd != MemCmd::Writeback) { return protocol->handleBusRequest(cache, pkt, blk, mshr, new_state); } @@ -148,9 +148,10 @@ class SimpleCoherence * @param state The current state of the cache block. * @return The proper bus command, as determined by the protocol. */ - Packet::Command getBusCmd(Packet::Command &cmd, CacheBlk::State state) + MemCmd getBusCmd(MemCmd cmd, + CacheBlk::State state) { - if (cmd == Packet::Writeback) return Packet::Writeback; + if (cmd == MemCmd::Writeback) return MemCmd::Writeback; return protocol->getBusCmd(cmd, state); } diff --git a/src/mem/cache/coherence/uni_coherence.cc b/src/mem/cache/coherence/uni_coherence.cc index ea615d70a..6061c89c3 100644 --- a/src/mem/cache/coherence/uni_coherence.cc +++ b/src/mem/cache/coherence/uni_coherence.cc @@ -99,19 +99,19 @@ UniCoherence::propogateInvalidate(PacketPtr pkt, bool isTiming) if (isTiming) { // Forward to other caches Request* req = new Request(pkt->req->getPaddr(), pkt->getSize(), 0); - PacketPtr tmp = new Packet(req, Packet::InvalidateReq, -1); + PacketPtr tmp = new Packet(req, MemCmd::InvalidateReq, -1); cshrs.allocate(tmp); cache->setSlaveRequest(Request_Coherence, curTick); if (cshrs.isFull()) cache->setBlockedForSnoop(Blocked_Coherence); } else { - PacketPtr tmp = new Packet(pkt->req, Packet::InvalidateReq, -1); + PacketPtr tmp = new Packet(pkt->req, MemCmd::InvalidateReq, -1); cache->cpuSidePort->sendAtomic(tmp); delete tmp; } /**/ -/* PacketPtr tmp = new Packet(pkt->req, Packet::InvalidateReq, -1); +/* PacketPtr tmp = new Packet(pkt->req, MemCmd::InvalidateReq, -1); cache->cpuSidePort->sendFunctional(tmp); delete tmp; */ @@ -119,7 +119,7 @@ UniCoherence::propogateInvalidate(PacketPtr pkt, bool isTiming) if (pkt->isRead()) { /*For now we will see if someone above us has the data by doing a functional access on reads. Fix this later */ - PacketPtr tmp = new Packet(pkt->req, Packet::ReadReq, -1); + PacketPtr tmp = new Packet(pkt->req, MemCmd::ReadReq, -1); tmp->allocate(); cache->cpuSidePort->sendFunctional(tmp); bool hit = (tmp->result == Packet::Success); diff --git a/src/mem/cache/coherence/uni_coherence.hh b/src/mem/cache/coherence/uni_coherence.hh index 9a4aacdec..9efb4e192 100644 --- a/src/mem/cache/coherence/uni_coherence.hh +++ b/src/mem/cache/coherence/uni_coherence.hh @@ -77,13 +77,13 @@ class UniCoherence * @return The proper bus command, as determined by the protocol. * @todo Make changes so writebacks don't get here. */ - Packet::Command getBusCmd(Packet::Command &cmd, CacheBlk::State state) + MemCmd getBusCmd(MemCmd cmd, CacheBlk::State state) { - if (cmd == Packet::HardPFReq && state) + if (cmd == MemCmd::HardPFReq && state) warn("Trying to issue a prefetch to a block we already have\n"); - if (cmd == Packet::Writeback) - return Packet::Writeback; - return Packet::ReadReq; + if (cmd == MemCmd::Writeback) + return MemCmd::Writeback; + return MemCmd::ReadReq; } /** @@ -96,7 +96,7 @@ class UniCoherence { if (pkt->senderState) //Blocking Buffers don't get mshrs { - if (((MSHR *)(pkt->senderState))->originalCmd == Packet::HardPFReq) { + if (((MSHR *)(pkt->senderState))->originalCmd == MemCmd::HardPFReq) { DPRINTF(HWPrefetch, "Marking a hardware prefetch as such in the state\n"); return BlkHWPrefetched | BlkValid | BlkWritable; } diff --git a/src/mem/cache/miss/blocking_buffer.cc b/src/mem/cache/miss/blocking_buffer.cc index a1af88341..e8ff26880 100644 --- a/src/mem/cache/miss/blocking_buffer.cc +++ b/src/mem/cache/miss/blocking_buffer.cc @@ -90,7 +90,7 @@ BlockingBuffer::getPacket() } void -BlockingBuffer::setBusCmd(PacketPtr &pkt, Packet::Command cmd) +BlockingBuffer::setBusCmd(PacketPtr &pkt, MemCmd cmd) { MSHR *mshr = (MSHR*) pkt->senderState; mshr->originalCmd = pkt->cmd; @@ -189,7 +189,7 @@ BlockingBuffer::doWriteback(Addr addr, { // Generate request Request * req = new Request(addr, size, 0); - PacketPtr pkt = new Packet(req, Packet::Writeback, -1); + PacketPtr pkt = new Packet(req, MemCmd::Writeback, -1); pkt->allocate(); if (data) { std::memcpy(pkt->getPtr(), data, size); diff --git a/src/mem/cache/miss/blocking_buffer.hh b/src/mem/cache/miss/blocking_buffer.hh index 24386a249..86b24d539 100644 --- a/src/mem/cache/miss/blocking_buffer.hh +++ b/src/mem/cache/miss/blocking_buffer.hh @@ -104,7 +104,7 @@ public: * @param pkt The request to update. * @param cmd The bus command to use. */ - void setBusCmd(PacketPtr &pkt, Packet::Command cmd); + void setBusCmd(PacketPtr &pkt, MemCmd cmd); /** * Restore the original command in case of a bus transmission error. diff --git a/src/mem/cache/miss/miss_buffer.hh b/src/mem/cache/miss/miss_buffer.hh index 3e3080578..9a86db304 100644 --- a/src/mem/cache/miss/miss_buffer.hh +++ b/src/mem/cache/miss/miss_buffer.hh @@ -123,7 +123,7 @@ class MissBuffer * @param pkt The request to update. * @param cmd The bus command to use. */ - virtual void setBusCmd(PacketPtr &pkt, Packet::Command cmd) = 0; + virtual void setBusCmd(PacketPtr &pkt, MemCmd cmd) = 0; /** * Restore the original command in case of a bus transmission error. diff --git a/src/mem/cache/miss/miss_queue.cc b/src/mem/cache/miss/miss_queue.cc index 1d3e22326..25b8fcbeb 100644 --- a/src/mem/cache/miss/miss_queue.cc +++ b/src/mem/cache/miss/miss_queue.cc @@ -67,17 +67,12 @@ MissQueue::regStats(const string &name) { MissBuffer::regStats(name); - Request temp_req((Addr) NULL, 4, 0); - Packet::Command temp_cmd = Packet::ReadReq; - Packet temp_pkt(&temp_req, temp_cmd, 0); //@todo FIx command strings so this isn't neccessary - temp_pkt.allocate(); - using namespace Stats; // MSHR hit statistics - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); mshr_hits[access_idx] .init(maxThreadsPerCPU) @@ -92,20 +87,20 @@ MissQueue::regStats(const string &name) .desc("number of demand (read+write) MSHR hits") .flags(total) ; - demandMshrHits = mshr_hits[Packet::ReadReq] + mshr_hits[Packet::WriteReq]; + demandMshrHits = mshr_hits[MemCmd::ReadReq] + mshr_hits[MemCmd::WriteReq]; overallMshrHits .name(name + ".overall_mshr_hits") .desc("number of overall MSHR hits") .flags(total) ; - overallMshrHits = demandMshrHits + mshr_hits[Packet::SoftPFReq] + - mshr_hits[Packet::HardPFReq]; + overallMshrHits = demandMshrHits + mshr_hits[MemCmd::SoftPFReq] + + mshr_hits[MemCmd::HardPFReq]; // MSHR miss statistics - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); mshr_misses[access_idx] .init(maxThreadsPerCPU) @@ -120,20 +115,20 @@ MissQueue::regStats(const string &name) .desc("number of demand (read+write) MSHR misses") .flags(total) ; - demandMshrMisses = mshr_misses[Packet::ReadReq] + mshr_misses[Packet::WriteReq]; + demandMshrMisses = mshr_misses[MemCmd::ReadReq] + mshr_misses[MemCmd::WriteReq]; overallMshrMisses .name(name + ".overall_mshr_misses") .desc("number of overall MSHR misses") .flags(total) ; - overallMshrMisses = demandMshrMisses + mshr_misses[Packet::SoftPFReq] + - mshr_misses[Packet::HardPFReq]; + overallMshrMisses = demandMshrMisses + mshr_misses[MemCmd::SoftPFReq] + + mshr_misses[MemCmd::HardPFReq]; // MSHR miss latency statistics - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); mshr_miss_latency[access_idx] .init(maxThreadsPerCPU) @@ -148,8 +143,8 @@ MissQueue::regStats(const string &name) .desc("number of demand (read+write) MSHR miss cycles") .flags(total) ; - demandMshrMissLatency = mshr_miss_latency[Packet::ReadReq] - + mshr_miss_latency[Packet::WriteReq]; + demandMshrMissLatency = mshr_miss_latency[MemCmd::ReadReq] + + mshr_miss_latency[MemCmd::WriteReq]; overallMshrMissLatency .name(name + ".overall_mshr_miss_latency") @@ -157,12 +152,12 @@ MissQueue::regStats(const string &name) .flags(total) ; overallMshrMissLatency = demandMshrMissLatency + - mshr_miss_latency[Packet::SoftPFReq] + mshr_miss_latency[Packet::HardPFReq]; + mshr_miss_latency[MemCmd::SoftPFReq] + mshr_miss_latency[MemCmd::HardPFReq]; // MSHR uncacheable statistics - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); mshr_uncacheable[access_idx] .init(maxThreadsPerCPU) @@ -177,14 +172,14 @@ MissQueue::regStats(const string &name) .desc("number of overall MSHR uncacheable misses") .flags(total) ; - overallMshrUncacheable = mshr_uncacheable[Packet::ReadReq] - + mshr_uncacheable[Packet::WriteReq] + mshr_uncacheable[Packet::SoftPFReq] - + mshr_uncacheable[Packet::HardPFReq]; + overallMshrUncacheable = mshr_uncacheable[MemCmd::ReadReq] + + mshr_uncacheable[MemCmd::WriteReq] + mshr_uncacheable[MemCmd::SoftPFReq] + + mshr_uncacheable[MemCmd::HardPFReq]; // MSHR miss latency statistics - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); mshr_uncacheable_lat[access_idx] .init(maxThreadsPerCPU) @@ -199,16 +194,16 @@ MissQueue::regStats(const string &name) .desc("number of overall MSHR uncacheable cycles") .flags(total) ; - overallMshrUncacheableLatency = mshr_uncacheable_lat[Packet::ReadReq] - + mshr_uncacheable_lat[Packet::WriteReq] - + mshr_uncacheable_lat[Packet::SoftPFReq] - + mshr_uncacheable_lat[Packet::HardPFReq]; + overallMshrUncacheableLatency = mshr_uncacheable_lat[MemCmd::ReadReq] + + mshr_uncacheable_lat[MemCmd::WriteReq] + + mshr_uncacheable_lat[MemCmd::SoftPFReq] + + mshr_uncacheable_lat[MemCmd::HardPFReq]; #if 0 // MSHR access formulas - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); mshrAccesses[access_idx] .name(name + "." + cstr + "_mshr_accesses") @@ -237,9 +232,9 @@ MissQueue::regStats(const string &name) #endif // MSHR miss rate formulas - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); mshrMissRate[access_idx] .name(name + "." + cstr + "_mshr_miss_rate") @@ -266,9 +261,9 @@ MissQueue::regStats(const string &name) overallMshrMissRate = overallMshrMisses / cache->overallAccesses; // mshrMiss latency formulas - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); avgMshrMissLatency[access_idx] .name(name + "." + cstr + "_avg_mshr_miss_latency") @@ -295,9 +290,9 @@ MissQueue::regStats(const string &name) overallAvgMshrMissLatency = overallMshrMissLatency / overallMshrMisses; // mshrUncacheable latency formulas - for (int access_idx = 0; access_idx < NUM_MEM_CMDS; ++access_idx) { - Packet::Command cmd = (Packet::Command)access_idx; - const string &cstr = temp_pkt.cmdIdxToString(cmd); + for (int access_idx = 0; access_idx < MemCmd::NUM_MEM_CMDS; ++access_idx) { + MemCmd cmd(access_idx); + const string &cstr = cmd.toString(); avgMshrUncacheableLatency[access_idx] .name(name + "." + cstr + "_avg_mshr_uncacheable_latency") @@ -351,7 +346,7 @@ MissQueue::allocateMiss(PacketPtr &pkt, int size, Tick time) if (mq.isFull()) { cache->setBlocked(Blocked_NoMSHRs); } - if (pkt->cmd != Packet::HardPFReq) { + if (pkt->cmd != MemCmd::HardPFReq) { //If we need to request the bus (not on HW prefetch), do so cache->setMasterRequest(Request_MSHR, time); } @@ -500,17 +495,17 @@ MissQueue::getPacket() } void -MissQueue::setBusCmd(PacketPtr &pkt, Packet::Command cmd) +MissQueue::setBusCmd(PacketPtr &pkt, MemCmd cmd) { assert(pkt->senderState != 0); MSHR * mshr = (MSHR*)pkt->senderState; mshr->originalCmd = pkt->cmd; - if (cmd == Packet::UpgradeReq || cmd == Packet::InvalidateReq) { + if (cmd == MemCmd::UpgradeReq || cmd == MemCmd::InvalidateReq) { pkt->flags |= NO_ALLOCATE; pkt->flags &= ~CACHE_LINE_FILL; } else if (!pkt->req->isUncacheable() && !pkt->isNoAllocate() && - (cmd & (1 << 6)/*NeedsResponse*/)) { + cmd.needsResponse()) { pkt->flags |= CACHE_LINE_FILL; } if (pkt->isCacheFill() || pkt->isNoAllocate()) @@ -552,7 +547,7 @@ MissQueue::markInService(PacketPtr &pkt, MSHR* mshr) if (!mq.havePending()){ cache->clearMasterRequest(Request_MSHR); } - if (mshr->originalCmd == Packet::HardPFReq) { + if (mshr->originalCmd == MemCmd::HardPFReq) { DPRINTF(HWPrefetch, "%s:Marking a HW_PF in service\n", cache->name()); //Also clear pending if need be @@ -576,7 +571,7 @@ void MissQueue::handleResponse(PacketPtr &pkt, Tick time) { MSHR* mshr = (MSHR*)pkt->senderState; - if (((MSHR*)(pkt->senderState))->originalCmd == Packet::HardPFReq) { + if (((MSHR*)(pkt->senderState))->originalCmd == MemCmd::HardPFReq) { DPRINTF(HWPrefetch, "%s:Handling the response to a HW_PF\n", cache->name()); } @@ -589,7 +584,7 @@ MissQueue::handleResponse(PacketPtr &pkt, Tick time) BlockedCause cause = NUM_BLOCKED_CAUSES; if (pkt->isCacheFill() && !pkt->isNoAllocate()) { - mshr_miss_latency[mshr->originalCmd][0/*pkt->req->getThreadNum()*/] += + mshr_miss_latency[mshr->originalCmd.toInt()][0/*pkt->req->getThreadNum()*/] += curTick - pkt->time; // targets were handled in the cache tags if (mshr == noTargetMSHR) { @@ -601,7 +596,7 @@ MissQueue::handleResponse(PacketPtr &pkt, Tick time) if (mshr->hasTargets()) { // Didn't satisfy all the targets, need to resend - Packet::Command cmd = mshr->getTarget()->cmd; + MemCmd cmd = mshr->getTarget()->cmd; mshr->pkt->setDest(Packet::Broadcast); mshr->pkt->result = Packet::Unknown; mq.markPending(mshr, cmd); @@ -618,7 +613,7 @@ MissQueue::handleResponse(PacketPtr &pkt, Tick time) } } else { if (pkt->req->isUncacheable()) { - mshr_uncacheable_lat[pkt->cmd][0/*pkt->req->getThreadNum()*/] += + mshr_uncacheable_lat[pkt->cmd.toInt()][0/*pkt->req->getThreadNum()*/] += curTick - pkt->time; } if (mshr->hasTargets() && pkt->req->isUncacheable()) { @@ -713,7 +708,7 @@ MissQueue::doWriteback(Addr addr, { // Generate request Request * req = new Request(addr, size, 0); - PacketPtr pkt = new Packet(req, Packet::Writeback, -1); + PacketPtr pkt = new Packet(req, MemCmd::Writeback, -1); pkt->allocate(); if (data) { memcpy(pkt->getPtr(), data, size); diff --git a/src/mem/cache/miss/miss_queue.hh b/src/mem/cache/miss/miss_queue.hh index 1f9bb1e0c..d3560ff36 100644 --- a/src/mem/cache/miss/miss_queue.hh +++ b/src/mem/cache/miss/miss_queue.hh @@ -77,59 +77,59 @@ class MissQueue : public MissBuffer * @{ */ /** Number of misses that hit in the MSHRs per command and thread. */ - Stats::Vector<> mshr_hits[NUM_MEM_CMDS]; + Stats::Vector<> mshr_hits[MemCmd::NUM_MEM_CMDS]; /** Demand misses that hit in the MSHRs. */ Stats::Formula demandMshrHits; /** Total number of misses that hit in the MSHRs. */ Stats::Formula overallMshrHits; /** Number of misses that miss in the MSHRs, per command and thread. */ - Stats::Vector<> mshr_misses[NUM_MEM_CMDS]; + Stats::Vector<> mshr_misses[MemCmd::NUM_MEM_CMDS]; /** Demand misses that miss in the MSHRs. */ Stats::Formula demandMshrMisses; /** Total number of misses that miss in the MSHRs. */ Stats::Formula overallMshrMisses; /** Number of misses that miss in the MSHRs, per command and thread. */ - Stats::Vector<> mshr_uncacheable[NUM_MEM_CMDS]; + Stats::Vector<> mshr_uncacheable[MemCmd::NUM_MEM_CMDS]; /** Total number of misses that miss in the MSHRs. */ Stats::Formula overallMshrUncacheable; /** Total cycle latency of each MSHR miss, per command and thread. */ - Stats::Vector<> mshr_miss_latency[NUM_MEM_CMDS]; + Stats::Vector<> mshr_miss_latency[MemCmd::NUM_MEM_CMDS]; /** Total cycle latency of demand MSHR misses. */ Stats::Formula demandMshrMissLatency; /** Total cycle latency of overall MSHR misses. */ Stats::Formula overallMshrMissLatency; /** Total cycle latency of each MSHR miss, per command and thread. */ - Stats::Vector<> mshr_uncacheable_lat[NUM_MEM_CMDS]; + Stats::Vector<> mshr_uncacheable_lat[MemCmd::NUM_MEM_CMDS]; /** Total cycle latency of overall MSHR misses. */ Stats::Formula overallMshrUncacheableLatency; /** The total number of MSHR accesses per command and thread. */ - Stats::Formula mshrAccesses[NUM_MEM_CMDS]; + Stats::Formula mshrAccesses[MemCmd::NUM_MEM_CMDS]; /** The total number of demand MSHR accesses. */ Stats::Formula demandMshrAccesses; /** The total number of MSHR accesses. */ Stats::Formula overallMshrAccesses; /** The miss rate in the MSHRs pre command and thread. */ - Stats::Formula mshrMissRate[NUM_MEM_CMDS]; + Stats::Formula mshrMissRate[MemCmd::NUM_MEM_CMDS]; /** The demand miss rate in the MSHRs. */ Stats::Formula demandMshrMissRate; /** The overall miss rate in the MSHRs. */ Stats::Formula overallMshrMissRate; /** The average latency of an MSHR miss, per command and thread. */ - Stats::Formula avgMshrMissLatency[NUM_MEM_CMDS]; + Stats::Formula avgMshrMissLatency[MemCmd::NUM_MEM_CMDS]; /** The average latency of a demand MSHR miss. */ Stats::Formula demandAvgMshrMissLatency; /** The average overall latency of an MSHR miss. */ Stats::Formula overallAvgMshrMissLatency; /** The average latency of an MSHR miss, per command and thread. */ - Stats::Formula avgMshrUncacheableLatency[NUM_MEM_CMDS]; + Stats::Formula avgMshrUncacheableLatency[MemCmd::NUM_MEM_CMDS]; /** The average overall latency of an MSHR miss. */ Stats::Formula overallAvgMshrUncacheableLatency; @@ -220,7 +220,7 @@ class MissQueue : public MissBuffer * @param pkt The request to update. * @param cmd The bus command to use. */ - void setBusCmd(PacketPtr &pkt, Packet::Command cmd); + void setBusCmd(PacketPtr &pkt, MemCmd cmd); /** * Restore the original command in case of a bus transmission error. diff --git a/src/mem/cache/miss/mshr.cc b/src/mem/cache/miss/mshr.cc index fc520b4b4..352d1ec6f 100644 --- a/src/mem/cache/miss/mshr.cc +++ b/src/mem/cache/miss/mshr.cc @@ -54,7 +54,7 @@ MSHR::MSHR() } void -MSHR::allocate(Packet::Command cmd, Addr _addr, int size, +MSHR::allocate(MemCmd cmd, Addr _addr, int size, PacketPtr &target) { addr = _addr; @@ -148,7 +148,7 @@ MSHR::allocateTarget(PacketPtr &target) */ if (!inService && target->isWrite()) { - pkt->cmd = Packet::WriteReq; + pkt->cmd = MemCmd::WriteReq; } } diff --git a/src/mem/cache/miss/mshr.hh b/src/mem/cache/miss/mshr.hh index 281ea9d49..d0410acda 100644 --- a/src/mem/cache/miss/mshr.hh +++ b/src/mem/cache/miss/mshr.hh @@ -72,7 +72,7 @@ class MSHR { /** The number of currently allocated targets. */ short ntargets; /** The original requesting command. */ - Packet::Command originalCmd; + MemCmd originalCmd; /** Order number of assigned by the miss queue. */ uint64_t order; @@ -100,7 +100,7 @@ public: * @param size The number of bytes to request. * @param pkt The original miss. */ - void allocate(Packet::Command cmd, Addr addr, int size, + void allocate(MemCmd cmd, Addr addr, int size, PacketPtr &pkt); /** diff --git a/src/mem/cache/miss/mshr_queue.cc b/src/mem/cache/miss/mshr_queue.cc index 6cb62429d..add11dfe7 100644 --- a/src/mem/cache/miss/mshr_queue.cc +++ b/src/mem/cache/miss/mshr_queue.cc @@ -136,7 +136,7 @@ MSHRQueue::allocateFetch(Addr addr, int size, PacketPtr &target) MSHR *mshr = freeList.front(); assert(mshr->getNumTargets() == 0); freeList.pop_front(); - mshr->allocate(Packet::ReadReq, addr, size, target); + mshr->allocate(MemCmd::ReadReq, addr, size, target); mshr->allocIter = allocatedList.insert(allocatedList.end(), mshr); mshr->readyIter = pendingList.insert(pendingList.end(), mshr); @@ -151,7 +151,7 @@ MSHRQueue::allocateTargetList(Addr addr, int size) assert(mshr->getNumTargets() == 0); freeList.pop_front(); PacketPtr dummy; - mshr->allocate(Packet::ReadReq, addr, size, dummy); + mshr->allocate(MemCmd::ReadReq, addr, size, dummy); mshr->allocIter = allocatedList.insert(allocatedList.end(), mshr); mshr->inService = true; ++inServiceMSHRs; @@ -196,7 +196,7 @@ void MSHRQueue::markInService(MSHR* mshr) { //assert(mshr == pendingList.front()); - if (!mshr->pkt->needsResponse() && !(mshr->pkt->cmd == Packet::UpgradeReq)) { + if (!mshr->pkt->needsResponse() && !(mshr->pkt->cmd == MemCmd::UpgradeReq)) { assert(mshr->getNumTargets() == 0); deallocate(mshr); return; @@ -209,7 +209,7 @@ MSHRQueue::markInService(MSHR* mshr) } void -MSHRQueue::markPending(MSHR* mshr, Packet::Command cmd) +MSHRQueue::markPending(MSHR* mshr, MemCmd cmd) { //assert(mshr->readyIter == NULL); mshr->pkt->cmd = cmd; diff --git a/src/mem/cache/miss/mshr_queue.hh b/src/mem/cache/miss/mshr_queue.hh index ec2ddae8a..5069db661 100644 --- a/src/mem/cache/miss/mshr_queue.hh +++ b/src/mem/cache/miss/mshr_queue.hh @@ -185,7 +185,7 @@ class MSHRQueue { * @param mshr The MSHR to resend. * @param cmd The command to resend. */ - void markPending(MSHR* mshr, Packet::Command cmd); + void markPending(MSHR* mshr, MemCmd cmd); /** * Squash outstanding requests with the given thread number. If a request diff --git a/src/mem/cache/prefetch/base_prefetcher.cc b/src/mem/cache/prefetch/base_prefetcher.cc index 4254800b1..44daf75e1 100644 --- a/src/mem/cache/prefetch/base_prefetcher.cc +++ b/src/mem/cache/prefetch/base_prefetcher.cc @@ -200,7 +200,7 @@ BasePrefetcher::handleMiss(PacketPtr &pkt, Tick time) //create a prefetch memreq Request * prefetchReq = new Request(*addr, blkSize, 0); PacketPtr prefetch; - prefetch = new Packet(prefetchReq, Packet::HardPFReq, -1); + prefetch = new Packet(prefetchReq, MemCmd::HardPFReq, -1); prefetch->allocate(); prefetch->req->setThreadContext(pkt->req->getCpuNum(), pkt->req->getThreadNum()); diff --git a/src/mem/cache/tags/iic.cc b/src/mem/cache/tags/iic.cc index e547e112e..20e2ef0ac 100644 --- a/src/mem/cache/tags/iic.cc +++ b/src/mem/cache/tags/iic.cc @@ -372,7 +372,8 @@ IIC::freeReplacementBlock(PacketList & writebacks) */ Request *writebackReq = new Request(regenerateBlkAddr(tag_ptr->tag, 0), blkSize, 0); - PacketPtr writeback = new Packet(writebackReq, Packet::Writeback, -1); + PacketPtr writeback = new Packet(writebackReq, MemCmd::Writeback, + -1); writeback->allocate(); memcpy(writeback->getPtr(), tag_ptr->data, blkSize); diff --git a/src/mem/packet.cc b/src/mem/packet.cc index 1e12ccc20..dde6c00d5 100644 --- a/src/mem/packet.cc +++ b/src/mem/packet.cc @@ -41,68 +41,62 @@ #include "base/trace.hh" #include "mem/packet.hh" -static const std::string ReadReqString("ReadReq"); -static const std::string WriteReqString("WriteReq"); -static const std::string WriteReqNoAckString("WriteReqNoAck|Writeback"); -static const std::string ReadRespString("ReadResp"); -static const std::string WriteRespString("WriteResp"); -static const std::string SoftPFReqString("SoftPFReq"); -static const std::string SoftPFRespString("SoftPFResp"); -static const std::string HardPFReqString("HardPFReq"); -static const std::string HardPFRespString("HardPFResp"); -static const std::string InvalidateReqString("InvalidateReq"); -static const std::string WriteInvalidateReqString("WriteInvalidateReq"); -static const std::string WriteInvalidateRespString("WriteInvalidateResp"); -static const std::string UpgradeReqString("UpgradeReq"); -static const std::string ReadExReqString("ReadExReq"); -static const std::string ReadExRespString("ReadExResp"); -static const std::string OtherCmdString(""); - -const std::string & -Packet::cmdString() const +// The one downside to bitsets is that static initializers can get ugly. +#define SET1(a1) (1 << (a1)) +#define SET2(a1, a2) (SET1(a1) | SET1(a2)) +#define SET3(a1, a2, a3) (SET2(a1, a2) | SET1(a3)) +#define SET4(a1, a2, a3, a4) (SET3(a1, a2, a3) | SET1(a4)) +#define SET5(a1, a2, a3, a4, a5) (SET4(a1, a2, a3, a4) | SET1(a5)) +#define SET6(a1, a2, a3, a4, a5, a6) (SET5(a1, a2, a3, a4, a5) | SET1(a6)) + +const MemCmd::CommandInfo +MemCmd::commandInfo[] = { - switch (cmd) { - case ReadReq: return ReadReqString; - case WriteReq: return WriteReqString; - case WriteReqNoAck: return WriteReqNoAckString; - case ReadResp: return ReadRespString; - case WriteResp: return WriteRespString; - case SoftPFReq: return SoftPFReqString; - case SoftPFResp: return SoftPFRespString; - case HardPFReq: return HardPFReqString; - case HardPFResp: return HardPFRespString; - case InvalidateReq: return InvalidateReqString; - case WriteInvalidateReq:return WriteInvalidateReqString; - case WriteInvalidateResp:return WriteInvalidateRespString; - case UpgradeReq: return UpgradeReqString; - case ReadExReq: return ReadExReqString; - case ReadExResp: return ReadExRespString; - default: return OtherCmdString; - } -} + /* InvalidCmd */ + { 0, InvalidCmd, "InvalidCmd" }, + /* ReadReq */ + { SET3(IsRead, IsRequest, NeedsResponse), ReadResp, "ReadReq" }, + /* WriteReq */ + { SET4(IsWrite, IsRequest, NeedsResponse, HasData), + WriteResp, "WriteReq" }, + /* WriteReqNoAck */ + { SET3(IsWrite, IsRequest, HasData), InvalidCmd, "WriteReqNoAck" }, + /* ReadResp */ + { SET3(IsRead, IsResponse, HasData), InvalidCmd, "ReadResp" }, + /* WriteResp */ + { SET2(IsWrite, IsResponse), InvalidCmd, "WriteResp" }, + /* Writeback */ + { SET3(IsWrite, IsRequest, HasData), InvalidCmd, "Writeback" }, + /* SoftPFReq */ + { SET4(IsRead, IsRequest, IsSWPrefetch, NeedsResponse), + SoftPFResp, "SoftPFReq" }, + /* HardPFReq */ + { SET4(IsRead, IsRequest, IsHWPrefetch, NeedsResponse), + HardPFResp, "HardPFReq" }, + /* SoftPFResp */ + { SET4(IsRead, IsResponse, IsSWPrefetch, HasData), + InvalidCmd, "SoftPFResp" }, + /* HardPFResp */ + { SET4(IsRead, IsResponse, IsHWPrefetch, HasData), + InvalidCmd, "HardPFResp" }, + /* InvalidateReq */ + { SET2(IsInvalidate, IsRequest), InvalidCmd, "InvalidateReq" }, + /* WriteInvalidateReq */ + { SET5(IsWrite, IsInvalidate, IsRequest, HasData, NeedsResponse), + WriteInvalidateResp, "WriteInvalidateReq" }, + /* WriteInvalidateResp */ + { SET5(IsWrite, IsInvalidate, IsRequest, NeedsResponse, IsResponse), + InvalidCmd, "WriteInvalidateResp" }, + /* UpgradeReq */ + { SET3(IsInvalidate, IsRequest, IsUpgrade), InvalidCmd, "UpgradeReq" }, + /* ReadExReq */ + { SET4(IsRead, IsInvalidate, IsRequest, NeedsResponse), + ReadExResp, "ReadExReq" }, + /* ReadExResp */ + { SET4(IsRead, IsInvalidate, IsResponse, HasData), + InvalidCmd, "ReadExResp" } +}; -const std::string & -Packet::cmdIdxToString(Packet::Command idx) -{ - switch (idx) { - case ReadReq: return ReadReqString; - case WriteReq: return WriteReqString; - case WriteReqNoAck: return WriteReqNoAckString; - case ReadResp: return ReadRespString; - case WriteResp: return WriteRespString; - case SoftPFReq: return SoftPFReqString; - case SoftPFResp: return SoftPFRespString; - case HardPFReq: return HardPFReqString; - case HardPFResp: return HardPFRespString; - case InvalidateReq: return InvalidateReqString; - case WriteInvalidateReq:return WriteInvalidateReqString; - case WriteInvalidateResp:return WriteInvalidateRespString; - case UpgradeReq: return UpgradeReqString; - case ReadExReq: return ReadExReqString; - case ReadExResp: return ReadExRespString; - default: return OtherCmdString; - } -} /** delete the data pointed to in the data pointer. Ok to call to matter how * data was allocted. */ @@ -149,9 +143,17 @@ fixDelayedResponsePacket(PacketPtr func, PacketPtr timing) bool result; if (timing->isRead() || timing->isWrite()) { - timing->toggleData(); + // Ugly hack to deal with the fact that we queue the requests + // and don't convert them to responses until we issue them on + // the bus. I tried to avoid this by converting packets to + // responses right away, but this breaks during snoops where a + // responder may do the conversion before other caches have + // done the snoop. Would work if we copied the packet instead + // of just hanging on to a pointer. + MemCmd oldCmd = timing->cmd; + timing->cmd = timing->cmd.responseCommand(); result = fixPacket(func, timing); - timing->toggleData(); + timing->cmd = oldCmd; } else { //Don't toggle if it isn't a read/write response diff --git a/src/mem/packet.hh b/src/mem/packet.hh index 15f605ca7..60f6e9ae0 100644 --- a/src/mem/packet.hh +++ b/src/mem/packet.hh @@ -40,6 +40,7 @@ #include #include +#include #include "base/misc.hh" #include "mem/request.hh" @@ -61,8 +62,114 @@ typedef std::list PacketList; #define NO_ALLOCATE (1 << 5) #define SNOOP_COMMIT (1 << 6) -//for now. @todo fix later -#define NUM_MEM_CMDS (1 << 11) + +class MemCmd +{ + public: + + /** List of all commands associated with a packet. */ + enum Command + { + InvalidCmd, + ReadReq, + WriteReq, + WriteReqNoAck, + ReadResp, + WriteResp, + Writeback, + SoftPFReq, + HardPFReq, + SoftPFResp, + HardPFResp, + InvalidateReq, + WriteInvalidateReq, + WriteInvalidateResp, + UpgradeReq, + ReadExReq, + ReadExResp, + NUM_MEM_CMDS + }; + + private: + /** List of command attributes. */ + enum Attribute + { + IsRead, + IsWrite, + IsPrefetch, + IsInvalidate, + IsRequest, + IsResponse, + NeedsResponse, + IsSWPrefetch, + IsHWPrefetch, + IsUpgrade, + HasData, + NUM_COMMAND_ATTRIBUTES + }; + + /** Structure that defines attributes and other data associated + * with a Command. */ + struct CommandInfo { + /** Set of attribute flags. */ + const std::bitset attributes; + /** Corresponding response for requests; InvalidCmd if no + * response is applicable. */ + const Command response; + /** String representation (for printing) */ + const std::string str; + }; + + /** Array to map Command enum to associated info. */ + static const CommandInfo commandInfo[]; + + private: + + Command cmd; + + bool testCmdAttrib(MemCmd::Attribute attrib) const { + return commandInfo[cmd].attributes[attrib] != 0; + } + + public: + + bool isRead() const { return testCmdAttrib(IsRead); } + bool isWrite() const { return testCmdAttrib(IsWrite); } + bool isRequest() const { return testCmdAttrib(IsRequest); } + bool isResponse() const { return testCmdAttrib(IsResponse); } + bool needsResponse() const { return testCmdAttrib(NeedsResponse); } + bool isInvalidate() const { return testCmdAttrib(IsInvalidate); } + bool hasData() const { return testCmdAttrib(HasData); } + + const Command responseCommand() const { + return commandInfo[cmd].response; + } + + /** Return the string to a cmd given by idx. */ + const std::string &toString() const { + return commandInfo[cmd].str; + } + + int toInt() const { return (int)cmd; } + + MemCmd(Command _cmd) + : cmd(_cmd) + { } + + MemCmd(int _cmd) + : cmd((Command)_cmd) + { } + + MemCmd() + : cmd(InvalidCmd) + { } + + bool operator==(MemCmd c2) { return (cmd == c2.cmd); } + bool operator!=(MemCmd c2) { return (cmd != c2.cmd); } + + friend class Packet; +}; + /** * A Packet is used to encapsulate a transfer between two objects in * the memory system (e.g., the L1 and L2 cache). (In contrast, a @@ -73,6 +180,9 @@ typedef std::list PacketList; class Packet { public: + + typedef MemCmd::Command Command; + /** Temporary FLAGS field until cache gets working, this should be in coherence/sender state. */ uint64_t flags; @@ -168,73 +278,27 @@ class Packet * to cast to the state appropriate to the sender. */ SenderState *senderState; - private: - /** List of command attributes. */ - // If you add a new CommandAttribute, make sure to increase NUM_MEM_CMDS - // as well. - enum CommandAttribute - { - IsRead = 1 << 0, - IsWrite = 1 << 1, - IsPrefetch = 1 << 2, - IsInvalidate = 1 << 3, - IsRequest = 1 << 4, - IsResponse = 1 << 5, - NeedsResponse = 1 << 6, - IsSWPrefetch = 1 << 7, - IsHWPrefetch = 1 << 8, - IsUpgrade = 1 << 9, - HasData = 1 << 10 - }; - public: - /** List of all commands associated with a packet. */ - enum Command - { - InvalidCmd = 0, - ReadReq = IsRead | IsRequest | NeedsResponse, - WriteReq = IsWrite | IsRequest | NeedsResponse | HasData, - WriteReqNoAck = IsWrite | IsRequest | HasData, - ReadResp = IsRead | IsResponse | NeedsResponse | HasData, - WriteResp = IsWrite | IsResponse | NeedsResponse, - Writeback = IsWrite | IsRequest | HasData, - SoftPFReq = IsRead | IsRequest | IsSWPrefetch | NeedsResponse, - HardPFReq = IsRead | IsRequest | IsHWPrefetch | NeedsResponse, - SoftPFResp = IsRead | IsResponse | IsSWPrefetch - | NeedsResponse | HasData, - HardPFResp = IsRead | IsResponse | IsHWPrefetch - | NeedsResponse | HasData, - InvalidateReq = IsInvalidate | IsRequest, - WriteInvalidateReq = IsWrite | IsInvalidate | IsRequest - | HasData | NeedsResponse, - WriteInvalidateResp = IsWrite | IsInvalidate | IsRequest - | NeedsResponse | IsResponse, - UpgradeReq = IsInvalidate | IsRequest | IsUpgrade, - ReadExReq = IsRead | IsInvalidate | IsRequest | NeedsResponse, - ReadExResp = IsRead | IsInvalidate | IsResponse - | NeedsResponse | HasData - }; + + /** The command field of the packet. */ + MemCmd cmd; /** Return the string name of the cmd field (for debugging and * tracing). */ - const std::string &cmdString() const; - - /** Reutrn the string to a cmd given by idx. */ - const std::string &cmdIdxToString(Command idx); + const std::string &cmdString() const { return cmd.toString(); } /** Return the index of this command. */ - inline int cmdToIndex() const { return (int) cmd; } + inline int cmdToIndex() const { return cmd.toInt(); } - /** The command field of the packet. */ - Command cmd; + public: - bool isRead() const { return (cmd & IsRead) != 0; } - bool isWrite() const { return (cmd & IsWrite) != 0; } - bool isRequest() const { return (cmd & IsRequest) != 0; } - bool isResponse() const { return (cmd & IsResponse) != 0; } - bool needsResponse() const { return (cmd & NeedsResponse) != 0; } - bool isInvalidate() const { return (cmd & IsInvalidate) != 0; } - bool hasData() const { return (cmd & HasData) != 0; } + bool isRead() const { return cmd.isRead(); } + bool isWrite() const { return cmd.isWrite(); } + bool isRequest() const { return cmd.isRequest(); } + bool isResponse() const { return cmd.isResponse(); } + bool needsResponse() const { return cmd.needsResponse(); } + bool isInvalidate() const { return cmd.isInvalidate(); } + bool hasData() const { return cmd.hasData(); } bool isCacheFill() const { return (flags & CACHE_LINE_FILL) != 0; } bool isNoAllocate() const { return (flags & NO_ALLOCATE) != 0; } @@ -268,13 +332,13 @@ class Packet Addr getOffset(int blkSize) const { return addr & (Addr)(blkSize - 1); } void addrOverride(Addr newAddr) { assert(addrSizeValid); addr = newAddr; } - void cmdOverride(Command newCmd) { cmd = newCmd; } + void cmdOverride(MemCmd newCmd) { cmd = newCmd; } /** Constructor. Note that a Request object must be constructed * first, but the Requests's physical address and size fields * need not be valid. The command and destination addresses * must be supplied. */ - Packet(Request *_req, Command _cmd, short _dest) + Packet(Request *_req, MemCmd _cmd, short _dest) : data(NULL), staticData(false), dynamicData(false), arrayData(false), addr(_req->paddr), size(_req->size), dest(_dest), addrSizeValid(_req->validPaddr), @@ -289,7 +353,7 @@ class Packet /** Alternate constructor if you are trying to create a packet with * a request that is for a whole block, not the address from the req. * this allows for overriding the size/addr of the req.*/ - Packet(Request *_req, Command _cmd, short _dest, int _blkSize) + Packet(Request *_req, MemCmd _cmd, short _dest, int _blkSize) : data(NULL), staticData(false), dynamicData(false), arrayData(false), addr(_req->paddr & ~(_blkSize - 1)), size(_blkSize), dest(_dest), @@ -334,25 +398,11 @@ class Packet void makeTimingResponse() { assert(needsResponse()); assert(isRequest()); - int icmd = (int)cmd; - icmd &= ~(IsRequest); - icmd |= IsResponse; - if (isRead()) - icmd |= HasData; - if (isWrite()) - icmd &= ~HasData; - cmd = (Command)icmd; + cmd = cmd.responseCommand(); dest = src; srcValid = false; } - - void toggleData() { - int icmd = (int)cmd; - icmd ^= HasData; - cmd = (Command)icmd; - } - /** * Take a request packet and modify it in place to be suitable for * returning as a response to that request. @@ -361,14 +411,7 @@ class Packet { assert(needsResponse()); assert(isRequest()); - int icmd = (int)cmd; - icmd &= ~(IsRequest); - icmd |= IsResponse; - if (isRead()) - icmd |= HasData; - if (isWrite()) - icmd &= ~HasData; - cmd = (Command)icmd; + cmd = cmd.responseCommand(); } /** diff --git a/src/mem/port.cc b/src/mem/port.cc index da719bbd9..048d7cf6d 100644 --- a/src/mem/port.cc +++ b/src/mem/port.cc @@ -46,7 +46,7 @@ Port::setPeer(Port *port) } void -Port::blobHelper(Addr addr, uint8_t *p, int size, Packet::Command cmd) +Port::blobHelper(Addr addr, uint8_t *p, int size, MemCmd cmd) { Request req; Packet pkt(&req, cmd, Packet::Broadcast); @@ -64,13 +64,13 @@ Port::blobHelper(Addr addr, uint8_t *p, int size, Packet::Command cmd) void Port::writeBlob(Addr addr, uint8_t *p, int size) { - blobHelper(addr, p, size, Packet::WriteReq); + blobHelper(addr, p, size, MemCmd::WriteReq); } void Port::readBlob(Addr addr, uint8_t *p, int size) { - blobHelper(addr, p, size, Packet::ReadReq); + blobHelper(addr, p, size, MemCmd::ReadReq); } void @@ -80,7 +80,7 @@ Port::memsetBlob(Addr addr, uint8_t val, int size) uint8_t *buf = new uint8_t[size]; std::memset(buf, val, size); - blobHelper(addr, buf, size, Packet::WriteReq); + blobHelper(addr, buf, size, MemCmd::WriteReq); delete [] buf; } diff --git a/src/mem/port.hh b/src/mem/port.hh index 5e55225bf..fdb5bfab4 100644 --- a/src/mem/port.hh +++ b/src/mem/port.hh @@ -245,7 +245,7 @@ class Port /** Internal helper function for read/writeBlob(). */ - void blobHelper(Addr addr, uint8_t *p, int size, Packet::Command cmd); + void blobHelper(Addr addr, uint8_t *p, int size, MemCmd cmd); }; /** A simple functional port that is only meant for one way communication to diff --git a/src/mem/tport.cc b/src/mem/tport.cc index c43c9aac0..b384a0444 100644 --- a/src/mem/tport.cc +++ b/src/mem/tport.cc @@ -68,7 +68,7 @@ SimpleTimingPort::recvTiming(PacketPtr pkt) sendTiming(pkt, latency); } else { - if (pkt->cmd != Packet::UpgradeReq) + if (pkt->cmd != MemCmd::UpgradeReq) { delete pkt->req; delete pkt; -- cgit v1.2.3