diff options
Diffstat (limited to 'src/mem')
51 files changed, 13220 insertions, 3 deletions
diff --git a/src/mem/cache/base_cache.cc b/src/mem/cache/base_cache.cc new file mode 100644 index 000000000..89e23ce31 --- /dev/null +++ b/src/mem/cache/base_cache.cc @@ -0,0 +1,334 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** + * @file + * Definition of BaseCache functions. + */ + +#include "mem/cache/base_cache.hh" +#include "cpu/smt.hh" +#include "cpu/base.hh" + +using namespace std; + +BaseCache::CachePort::CachePort(const std::string &_name, BaseCache *_cache, + bool _isCpuSide) + : Port(_name), cache(_cache), isCpuSide(_isCpuSide) +{ + blocked = false; + //Start ports at null if more than one is created we should panic + //cpuSidePort = NULL; + //memSidePort = NULL; +} + +void +BaseCache::CachePort::recvStatusChange(Port::Status status) +{ + cache->recvStatusChange(status, isCpuSide); +} + +void +BaseCache::CachePort::getDeviceAddressRanges(AddrRangeList &resp, + AddrRangeList &snoop) +{ + cache->getAddressRanges(resp, snoop); +} + +int +BaseCache::CachePort::deviceBlockSize() +{ + return cache->getBlockSize(); +} + +bool +BaseCache::CachePort::recvTiming(Packet *pkt) +{ + return cache->doTimingAccess(pkt, this, isCpuSide); +} + +Tick +BaseCache::CachePort::recvAtomic(Packet *pkt) +{ + return cache->doAtomicAccess(pkt, isCpuSide); +} + +void +BaseCache::CachePort::recvFunctional(Packet *pkt) +{ + cache->doFunctionalAccess(pkt, isCpuSide); +} + +void +BaseCache::CachePort::setBlocked() +{ + blocked = true; +} + +void +BaseCache::CachePort::clearBlocked() +{ + blocked = false; +} + +Port* +BaseCache::getPort(const std::string &if_name) +{ + if(if_name == "cpu_side") + { + if(cpuSidePort != NULL) + panic("Already have a cpu side for this cache\n"); + cpuSidePort = new CachePort(name() + "-cpu_side_port", this, true); + return cpuSidePort; + } + else if(if_name == "mem_side") + { + if(memSidePort != NULL) + panic("Already have a mem side for this cache\n"); + memSidePort = new CachePort(name() + "-mem_side_port", this, false); + return memSidePort; + } + else panic("Port name %s unrecognized\n", if_name); +} + +void +BaseCache::regStats() +{ + Request temp_req; + Packet::Command temp_cmd = Packet::ReadReq; + Packet temp_pkt(&temp_req, temp_cmd, 0); //@todo FIx command strings so this isn't neccessary + + 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); + + hits[access_idx] + .init(maxThreadsPerCPU) + .name(name() + "." + cstr + "_hits") + .desc("number of " + cstr + " hits") + .flags(total | nozero | nonan) + ; + } + + demandHits + .name(name() + ".demand_hits") + .desc("number of demand (read+write) hits") + .flags(total) + ; + demandHits = hits[Packet::ReadReq] + hits[Packet::WriteReq]; + + overallHits + .name(name() + ".overall_hits") + .desc("number of overall hits") + .flags(total) + ; + overallHits = demandHits + hits[Packet::SoftPFReq] + hits[Packet::HardPFReq] + + hits[Packet::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); + + misses[access_idx] + .init(maxThreadsPerCPU) + .name(name() + "." + cstr + "_misses") + .desc("number of " + cstr + " misses") + .flags(total | nozero | nonan) + ; + } + + demandMisses + .name(name() + ".demand_misses") + .desc("number of demand (read+write) misses") + .flags(total) + ; + demandMisses = misses[Packet::ReadReq] + misses[Packet::WriteReq]; + + overallMisses + .name(name() + ".overall_misses") + .desc("number of overall misses") + .flags(total) + ; + overallMisses = demandMisses + misses[Packet::SoftPFReq] + + misses[Packet::HardPFReq] + misses[Packet::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); + + missLatency[access_idx] + .init(maxThreadsPerCPU) + .name(name() + "." + cstr + "_miss_latency") + .desc("number of " + cstr + " miss cycles") + .flags(total | nozero | nonan) + ; + } + + demandMissLatency + .name(name() + ".demand_miss_latency") + .desc("number of demand (read+write) miss cycles") + .flags(total) + ; + demandMissLatency = missLatency[Packet::ReadReq] + missLatency[Packet::WriteReq]; + + overallMissLatency + .name(name() + ".overall_miss_latency") + .desc("number of overall miss cycles") + .flags(total) + ; + overallMissLatency = demandMissLatency + missLatency[Packet::SoftPFReq] + + missLatency[Packet::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); + + accesses[access_idx] + .name(name() + "." + cstr + "_accesses") + .desc("number of " + cstr + " accesses(hits+misses)") + .flags(total | nozero | nonan) + ; + + accesses[access_idx] = hits[access_idx] + misses[access_idx]; + } + + demandAccesses + .name(name() + ".demand_accesses") + .desc("number of demand (read+write) accesses") + .flags(total) + ; + demandAccesses = demandHits + demandMisses; + + overallAccesses + .name(name() + ".overall_accesses") + .desc("number of overall (read+write) accesses") + .flags(total) + ; + 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); + + missRate[access_idx] + .name(name() + "." + cstr + "_miss_rate") + .desc("miss rate for " + cstr + " accesses") + .flags(total | nozero | nonan) + ; + + missRate[access_idx] = misses[access_idx] / accesses[access_idx]; + } + + demandMissRate + .name(name() + ".demand_miss_rate") + .desc("miss rate for demand accesses") + .flags(total) + ; + demandMissRate = demandMisses / demandAccesses; + + overallMissRate + .name(name() + ".overall_miss_rate") + .desc("miss rate for overall accesses") + .flags(total) + ; + 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); + + avgMissLatency[access_idx] + .name(name() + "." + cstr + "_avg_miss_latency") + .desc("average " + cstr + " miss latency") + .flags(total | nozero | nonan) + ; + + avgMissLatency[access_idx] = + missLatency[access_idx] / misses[access_idx]; + } + + demandAvgMissLatency + .name(name() + ".demand_avg_miss_latency") + .desc("average overall miss latency") + .flags(total) + ; + demandAvgMissLatency = demandMissLatency / demandMisses; + + overallAvgMissLatency + .name(name() + ".overall_avg_miss_latency") + .desc("average overall miss latency") + .flags(total) + ; + overallAvgMissLatency = overallMissLatency / overallMisses; + + blocked_cycles.init(NUM_BLOCKED_CAUSES); + blocked_cycles + .name(name() + ".blocked_cycles") + .desc("number of cycles access was blocked") + .subname(Blocked_NoMSHRs, "no_mshrs") + .subname(Blocked_NoTargets, "no_targets") + ; + + + blocked_causes.init(NUM_BLOCKED_CAUSES); + blocked_causes + .name(name() + ".blocked") + .desc("number of cycles access was blocked") + .subname(Blocked_NoMSHRs, "no_mshrs") + .subname(Blocked_NoTargets, "no_targets") + ; + + avg_blocked + .name(name() + ".avg_blocked_cycles") + .desc("average number of cycles each access was blocked") + .subname(Blocked_NoMSHRs, "no_mshrs") + .subname(Blocked_NoTargets, "no_targets") + ; + + avg_blocked = blocked_cycles / blocked_causes; + + fastWrites + .name(name() + ".fast_writes") + .desc("number of fast writes performed") + ; + + cacheCopies + .name(name() + ".cache_copies") + .desc("number of cache copies performed") + ; +} diff --git a/src/mem/cache/base_cache.hh b/src/mem/cache/base_cache.hh new file mode 100644 index 000000000..977e0ae29 --- /dev/null +++ b/src/mem/cache/base_cache.hh @@ -0,0 +1,489 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** + * @file + * Declares a basic cache interface BaseCache. + */ + +#ifndef __BASE_CACHE_HH__ +#define __BASE_CACHE_HH__ + +#include <vector> +#include <string> +#include <list> +#include <inttypes.h> + +#include "base/statistics.hh" +#include "base/trace.hh" +#include "mem/mem_object.hh" +#include "mem/packet.hh" +#include "mem/port.hh" +#include "mem/request.hh" +#include "sim/eventq.hh" + +/** + * Reasons for Caches to be Blocked. + */ +enum BlockedCause{ + Blocked_NoMSHRs, + Blocked_NoTargets, + Blocked_NoWBBuffers, + Blocked_Coherence, + Blocked_Copy, + NUM_BLOCKED_CAUSES +}; + +/** + * Reasons for cache to request a bus. + */ +enum RequestCause{ + Request_MSHR, + Request_WB, + Request_Coherence, + Request_PF +}; + +/** + * A basic cache interface. Implements some common functions for speed. + */ +class BaseCache : public MemObject +{ + class CachePort : public Port + { + BaseCache *cache; + + public: + CachePort(const std::string &_name, BaseCache *_cache, bool _isCpuSide); + + protected: + virtual bool recvTiming(Packet *pkt); + + virtual Tick recvAtomic(Packet *pkt); + + virtual void recvFunctional(Packet *pkt); + + virtual void recvStatusChange(Status status); + + virtual void getDeviceAddressRanges(AddrRangeList &resp, + AddrRangeList &snoop); + + virtual int deviceBlockSize(); + + public: + void setBlocked(); + + void clearBlocked(); + + bool blocked; + + bool isCpuSide; + }; + + struct CacheEvent : public Event + { + Packet *pkt; + CachePort *cachePort; + + CacheEvent(Packet *pkt, CachePort *cachePort); + void process(); + const char *description(); + }; + + protected: + CachePort *cpuSidePort; + CachePort *memSidePort; + + public: + virtual Port *getPort(const std::string &if_name); + + private: + //To be defined in cache_impl.hh not in base class + virtual bool doTimingAccess(Packet *pkt, CachePort *cachePort, bool isCpuSide); + virtual Tick doAtomicAccess(Packet *pkt, bool isCpuSide); + virtual void doFunctionalAccess(Packet *pkt, bool isCpuSide); + virtual void recvStatusChange(Port::Status status, bool isCpuSide); + + /** + * Bit vector of the blocking reasons for the access path. + * @sa #BlockedCause + */ + uint8_t blocked; + + /** + * Bit vector for the blocking reasons for the snoop path. + * @sa #BlockedCause + */ + uint8_t blockedSnoop; + + /** + * Bit vector for the outstanding requests for the master interface. + */ + uint8_t masterRequests; + + /** + * Bit vector for the outstanding requests for the slave interface. + */ + uint8_t slaveRequests; + + protected: + + /** True if this cache is connected to the CPU. */ + bool topLevelCache; + + /** Stores time the cache blocked for statistics. */ + Tick blockedCycle; + + /** Block size of this cache */ + const int blkSize; + + /** The number of misses to trigger an exit event. */ + Counter missCount; + + public: + // Statistics + /** + * @addtogroup CacheStatistics + * @{ + */ + + /** Number of hits per thread for each type of command. @sa Packet::Command */ + Stats::Vector<> hits[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]; + /** Number of misses for demand accesses. */ + Stats::Formula demandMisses; + /** Number of misses for all accesses. */ + Stats::Formula overallMisses; + + /** + * 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]; + /** 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]; + /** 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]; + /** 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]; + /** The average miss latency for demand misses. */ + Stats::Formula demandAvgMissLatency; + /** The average miss latency for all misses. */ + Stats::Formula overallAvgMissLatency; + + /** The total number of cycles blocked for each blocked cause. */ + Stats::Vector<> blocked_cycles; + /** The number of times this cache blocked for each blocked cause. */ + Stats::Vector<> blocked_causes; + + /** The average number of cycles blocked for each blocked cause. */ + Stats::Formula avg_blocked; + + /** The number of fast writes (WH64) performed. */ + Stats::Scalar<> fastWrites; + + /** The number of cache copies performed. */ + Stats::Scalar<> cacheCopies; + + /** + * @} + */ + + /** + * Register stats for this object. + */ + virtual void regStats(); + + public: + + class Params + { + public: + /** List of address ranges of this cache. */ + std::vector<Range<Addr> > addrRange; + /** The hit latency for this cache. */ + int hitLatency; + /** The block size of this cache. */ + int blkSize; + /** + * The maximum number of misses this cache should handle before + * ending the simulation. + */ + Counter maxMisses; + + /** + * Construct an instance of this parameter class. + */ + Params(std::vector<Range<Addr> > addr_range, + int hit_latency, int _blkSize, Counter max_misses) + : addrRange(addr_range), hitLatency(hit_latency), blkSize(_blkSize), + maxMisses(max_misses) + { + } + }; + + /** + * Create and initialize a basic cache object. + * @param name The name of this cache. + * @param hier_params Pointer to the HierParams object for this hierarchy + * of this cache. + * @param params The parameter object for this BaseCache. + */ + BaseCache(const std::string &name, Params ¶ms) + : MemObject(name), blocked(0), blockedSnoop(0), masterRequests(0), + slaveRequests(0), topLevelCache(false), blkSize(params.blkSize), + missCount(params.maxMisses) + { + //Start ports at null if more than one is created we should panic + cpuSidePort = NULL; + memSidePort = NULL; + } + + /** + * Query block size of a cache. + * @return The block size + */ + int getBlockSize() const + { + return blkSize; + } + + /** + * Returns true if this cache is connect to the CPU. + * @return True if this is a L1 cache. + */ + bool isTopLevel() + { + return topLevelCache; + } + + /** + * Returns true if the cache is blocked for accesses. + */ + bool isBlocked() + { + return blocked != 0; + } + + /** + * Returns true if the cache is blocked for snoops. + */ + bool isBlockedForSnoop() + { + return blockedSnoop != 0; + } + + /** + * Marks the access path of the cache as blocked for the given cause. This + * also sets the blocked flag in the slave interface. + * @param cause The reason for the cache blocking. + */ + void setBlocked(BlockedCause cause) + { + uint8_t flag = 1 << cause; + if (blocked == 0) { + blocked_causes[cause]++; + blockedCycle = curTick; + } + blocked |= flag; + DPRINTF(Cache,"Blocking for cause %s\n", cause); + cpuSidePort->setBlocked(); + } + + /** + * Marks the snoop path of the cache as blocked for the given cause. This + * also sets the blocked flag in the master interface. + * @param cause The reason to block the snoop path. + */ + void setBlockedForSnoop(BlockedCause cause) + { + uint8_t flag = 1 << cause; + blockedSnoop |= flag; + memSidePort->setBlocked(); + } + + /** + * Marks the cache as unblocked for the given cause. This also clears the + * blocked flags in the appropriate interfaces. + * @param cause The newly unblocked cause. + * @warning Calling this function can cause a blocked request on the bus to + * access the cache. The cache must be in a state to handle that request. + */ + void clearBlocked(BlockedCause cause) + { + uint8_t flag = 1 << cause; + blocked &= ~flag; + blockedSnoop &= ~flag; + DPRINTF(Cache,"Unblocking for cause %s, causes left=%i\n", + cause, blocked); + if (!isBlocked()) { + blocked_cycles[cause] += curTick - blockedCycle; + DPRINTF(Cache,"Unblocking from all causes\n"); + cpuSidePort->clearBlocked(); + } + if (!isBlockedForSnoop()) { + memSidePort->clearBlocked(); + } + + } + + /** + * True if the master bus should be requested. + * @return True if there are outstanding requests for the master bus. + */ + bool doMasterRequest() + { + return masterRequests != 0; + } + + /** + * Request the master bus for the given cause and time. + * @param cause The reason for the request. + * @param time The time to make the request. + */ + void setMasterRequest(RequestCause cause, Tick time) + { + uint8_t flag = 1<<cause; + masterRequests |= flag; + assert("Implement\n" && 0); +// mi->pktuest(time); + } + + /** + * Clear the master bus request for the given cause. + * @param cause The request reason to clear. + */ + void clearMasterRequest(RequestCause cause) + { + uint8_t flag = 1<<cause; + masterRequests &= ~flag; + } + + /** + * Return true if the slave bus should be requested. + * @return True if there are outstanding requests for the slave bus. + */ + bool doSlaveRequest() + { + return slaveRequests != 0; + } + + /** + * Request the slave bus for the given reason and time. + * @param cause The reason for the request. + * @param time The time to make the request. + */ + void setSlaveRequest(RequestCause cause, Tick time) + { + uint8_t flag = 1<<cause; + slaveRequests |= flag; + assert("Implement\n" && 0); +// si->pktuest(time); + } + + /** + * Clear the slave bus request for the given reason. + * @param cause The request reason to clear. + */ + void clearSlaveRequest(RequestCause cause) + { + uint8_t flag = 1<<cause; + slaveRequests &= ~flag; + } + + /** + * Send a response to the slave interface. + * @param req The request being responded to. + * @param time The time the response is ready. + */ + void respond(Packet *pkt, Tick time) + { + assert("Implement\n" && 0); +// si->respond(pkt,time); + } + + /** + * Send a reponse to the slave interface and calculate miss latency. + * @param req The request to respond to. + * @param time The time the response is ready. + */ + void respondToMiss(Packet *pkt, Tick time) + { + if (!pkt->req->isUncacheable()) { + missLatency[pkt->cmdToIndex()][pkt->req->getThreadNum()] += time - pkt->time; + } + assert("Implement\n" && 0); +// si->respond(pkt,time); + } + + /** + * Suppliess the data if cache to cache transfers are enabled. + * @param req The bus transaction to fulfill. + */ + void respondToSnoop(Packet *pkt) + { + assert("Implement\n" && 0); +// mi->respond(pkt,curTick + hitLatency); + } + + /** + * Notification from master interface that a address range changed. Nothing + * to do for a cache. + */ + void rangeChange() {} + + void getAddressRanges(AddrRangeList &resp, AddrRangeList &snoop) + { + panic("Unimplimented\n"); + } +}; + +#endif //__BASE_CACHE_HH__ diff --git a/src/mem/cache/cache.cc b/src/mem/cache/cache.cc new file mode 100644 index 000000000..db66c096e --- /dev/null +++ b/src/mem/cache/cache.cc @@ -0,0 +1,146 @@ +/* + * Copyright (c) 2004-2005 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: Erik Hallnor + * Steve Reinhardt + * Lisa Hsu + * Kevin Lim + */ + +/** + * @file + * Cache template instantiations. + */ + +#include "mem/config/cache.hh" +#include "mem/config/compression.hh" + +#include "mem/cache/tags/cache_tags.hh" + +#if defined(USE_CACHE_LRU) +#include "mem/cache/tags/lru.hh" +#endif + +#if defined(USE_CACHE_FALRU) +#include "mem/cache/tags/fa_lru.hh" +#endif + +#if defined(USE_CACHE_IIC) +#include "mem/cache/tags/iic.hh" +#endif + +#if defined(USE_CACHE_SPLIT) +#include "mem/cache/tags/split.hh" +#endif + +#if defined(USE_CACHE_SPLIT_LIFO) +#include "mem/cache/tags/split_lifo.hh" +#endif + +#include "base/compression/null_compression.hh" +#if defined(USE_LZSS_COMPRESSION) +#include "base/compression/lzss_compression.hh" +#endif + +#include "mem/cache/miss/miss_queue.hh" +#include "mem/cache/miss/blocking_buffer.hh" + +#include "mem/cache/coherence/uni_coherence.hh" +#include "mem/cache/coherence/simple_coherence.hh" + +#include "mem/cache/cache_impl.hh" + +// Template Instantiations +#ifndef DOXYGEN_SHOULD_SKIP_THIS + + +#if defined(USE_CACHE_FALRU) +template class Cache<CacheTags<FALRU,NullCompression>, BlockingBuffer, SimpleCoherence>; +template class Cache<CacheTags<FALRU,NullCompression>, BlockingBuffer, UniCoherence>; +template class Cache<CacheTags<FALRU,NullCompression>, MissQueue, SimpleCoherence>; +template class Cache<CacheTags<FALRU,NullCompression>, MissQueue, UniCoherence>; +#if defined(USE_LZSS_COMPRESSION) +template class Cache<CacheTags<FALRU,LZSSCompression>, BlockingBuffer, SimpleCoherence>; +template class Cache<CacheTags<FALRU,LZSSCompression>, BlockingBuffer, UniCoherence>; +template class Cache<CacheTags<FALRU,LZSSCompression>, MissQueue, SimpleCoherence>; +template class Cache<CacheTags<FALRU,LZSSCompression>, MissQueue, UniCoherence>; +#endif +#endif + +#if defined(USE_CACHE_IIC) +template class Cache<CacheTags<IIC,NullCompression>, BlockingBuffer, SimpleCoherence>; +template class Cache<CacheTags<IIC,NullCompression>, BlockingBuffer, UniCoherence>; +template class Cache<CacheTags<IIC,NullCompression>, MissQueue, SimpleCoherence>; +template class Cache<CacheTags<IIC,NullCompression>, MissQueue, UniCoherence>; +#if defined(USE_LZSS_COMPRESSION) +template class Cache<CacheTags<IIC,LZSSCompression>, BlockingBuffer, SimpleCoherence>; +template class Cache<CacheTags<IIC,LZSSCompression>, BlockingBuffer, UniCoherence>; +template class Cache<CacheTags<IIC,LZSSCompression>, MissQueue, SimpleCoherence>; +template class Cache<CacheTags<IIC,LZSSCompression>, MissQueue, UniCoherence>; +#endif +#endif + +#if defined(USE_CACHE_LRU) +template class Cache<CacheTags<LRU,NullCompression>, BlockingBuffer, SimpleCoherence>; +template class Cache<CacheTags<LRU,NullCompression>, BlockingBuffer, UniCoherence>; +template class Cache<CacheTags<LRU,NullCompression>, MissQueue, SimpleCoherence>; +template class Cache<CacheTags<LRU,NullCompression>, MissQueue, UniCoherence>; +#if defined(USE_LZSS_COMPRESSION) +template class Cache<CacheTags<LRU,LZSSCompression>, BlockingBuffer, SimpleCoherence>; +template class Cache<CacheTags<LRU,LZSSCompression>, BlockingBuffer, UniCoherence>; +template class Cache<CacheTags<LRU,LZSSCompression>, MissQueue, SimpleCoherence>; +template class Cache<CacheTags<LRU,LZSSCompression>, MissQueue, UniCoherence>; +#endif +#endif + +#if defined(USE_CACHE_SPLIT) +template class Cache<CacheTags<Split,NullCompression>, BlockingBuffer, SimpleCoherence>; +template class Cache<CacheTags<Split,NullCompression>, BlockingBuffer, UniCoherence>; +template class Cache<CacheTags<Split,NullCompression>, MissQueue, SimpleCoherence>; +template class Cache<CacheTags<Split,NullCompression>, MissQueue, UniCoherence>; +#if defined(USE_LZSS_COMPRESSION) +template class Cache<CacheTags<Split,LZSSCompression>, BlockingBuffer, SimpleCoherence>; +template class Cache<CacheTags<Split,LZSSCompression>, BlockingBuffer, UniCoherence>; +template class Cache<CacheTags<Split,LZSSCompression>, MissQueue, SimpleCoherence>; +template class Cache<CacheTags<Split,LZSSCompression>, MissQueue, UniCoherence>; +#endif +#endif + +#if defined(USE_CACHE_SPLIT_LIFO) +template class Cache<CacheTags<SplitLIFO,NullCompression>, BlockingBuffer, SimpleCoherence>; +template class Cache<CacheTags<SplitLIFO,NullCompression>, BlockingBuffer, UniCoherence>; +template class Cache<CacheTags<SplitLIFO,NullCompression>, MissQueue, SimpleCoherence>; +template class Cache<CacheTags<SplitLIFO,NullCompression>, MissQueue, UniCoherence>; +#if defined(USE_LZSS_COMPRESSION) +template class Cache<CacheTags<SplitLIFO,LZSSCompression>, BlockingBuffer, SimpleCoherence>; +template class Cache<CacheTags<SplitLIFO,LZSSCompression>, BlockingBuffer, UniCoherence>; +template class Cache<CacheTags<SplitLIFO,LZSSCompression>, MissQueue, SimpleCoherence>; +template class Cache<CacheTags<SplitLIFO,LZSSCompression>, MissQueue, UniCoherence>; +#endif +#endif + +#endif //DOXYGEN_SHOULD_SKIP_THIS diff --git a/src/mem/cache/cache.hh b/src/mem/cache/cache.hh new file mode 100644 index 000000000..d2af1d8bf --- /dev/null +++ b/src/mem/cache/cache.hh @@ -0,0 +1,281 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + * Dave Greene + * Steve Reinhardt + */ + +/** + * @file + * Describes a cache based on template policies. + */ + +#ifndef __CACHE_HH__ +#define __CACHE_HH__ + +#include "base/misc.hh" // fatal, panic, and warn +#include "cpu/smt.hh" // SMT_MAX_THREADS + +#include "mem/cache/base_cache.hh" +#include "mem/cache/prefetch/prefetcher.hh" + +//Forward decleration +class MSHR; + + +/** + * A template-policy based cache. The behavior of the cache can be altered by + * supplying different template policies. TagStore handles all tag and data + * storage @sa TagStore. Buffering handles all misses and writes/writebacks + * @sa MissQueue. Coherence handles all coherence policy details @sa + * UniCoherence, SimpleMultiCoherence. + */ +template <class TagStore, class Buffering, class Coherence> +class Cache : public BaseCache +{ + public: + /** Define the type of cache block to use. */ + typedef typename TagStore::BlkType BlkType; + + bool prefetchAccess; + protected: + + /** Tag and data Storage */ + TagStore *tags; + /** Miss and Writeback handler */ + Buffering *missQueue; + /** Coherence protocol. */ + Coherence *coherence; + + /** Prefetcher */ + Prefetcher<TagStore, Buffering> *prefetcher; + + /** Do fast copies in this cache. */ + bool doCopy; + + /** Block on a delayed copy. */ + bool blockOnCopy; + + /** + * The clock ratio of the outgoing bus. + * Used for calculating critical word first. + */ + int busRatio; + + /** + * The bus width in bytes of the outgoing bus. + * Used for calculating critical word first. + */ + int busWidth; + + /** + * The latency of a hit in this device. + */ + int hitLatency; + + /** + * A permanent mem req to always be used to cause invalidations. + * Used to append to target list, to cause an invalidation. + */ + Packet * invalidatePkt; + + /** + * Temporarily move a block into a MSHR. + * @todo Remove this when LSQ/SB are fixed and implemented in memtest. + */ + void pseudoFill(Addr addr, int asid); + + /** + * Temporarily move a block into an existing MSHR. + * @todo Remove this when LSQ/SB are fixed and implemented in memtest. + */ + void pseudoFill(MSHR *mshr); + + public: + + class Params + { + public: + TagStore *tags; + Buffering *missQueue; + Coherence *coherence; + bool doCopy; + bool blockOnCopy; + BaseCache::Params baseParams; + Prefetcher<TagStore, Buffering> *prefetcher; + bool prefetchAccess; + int hitLatency; + + Params(TagStore *_tags, Buffering *mq, Coherence *coh, + bool do_copy, BaseCache::Params params, + Prefetcher<TagStore, Buffering> *_prefetcher, + bool prefetch_access, int hit_latency) + : tags(_tags), missQueue(mq), coherence(coh), doCopy(do_copy), + blockOnCopy(false), baseParams(params), + prefetcher(_prefetcher), prefetchAccess(prefetch_access), + hitLatency(hit_latency) + { + } + }; + + /** Instantiates a basic cache object. */ + Cache(const std::string &_name, Params ¶ms); + + bool doTimingAccess(Packet *pkt, CachePort *cachePort, + bool isCpuSide); + + Tick doAtomicAccess(Packet *pkt, CachePort *cachePort, + bool isCpuSide); + + void doFunctionalAccess(Packet *pkt, CachePort *cachePort, + bool isCpuSide); + + void recvStatusChange(Port::Status status, bool isCpuSide); + + void regStats(); + + /** + * Performs the access specified by the request. + * @param req The request to perform. + * @return The result of the access. + */ + bool access(Packet * &pkt); + + /** + * Selects a request to send on the bus. + * @return The memory request to service. + */ + Packet * getPacket(); + + /** + * Was the request was sent successfully? + * @param req The request. + * @param success True if the request was sent successfully. + */ + void sendResult(Packet * &pkt, bool success); + + /** + * Handles a response (cache line fill/write ack) from the bus. + * @param req The request being responded to. + */ + void handleResponse(Packet * &pkt); + + /** + * Start handling a copy transaction. + * @param req The copy request to perform. + */ + void startCopy(Packet * &pkt); + + /** + * Handle a delayed copy transaction. + * @param req The delayed copy request to continue. + * @param addr The address being responded to. + * @param blk The block of the current response. + * @param mshr The mshr being handled. + */ + void handleCopy(Packet * &pkt, Addr addr, BlkType *blk, MSHR *mshr); + + /** + * Selects a coherence message to forward to lower levels of the hierarchy. + * @return The coherence message to forward. + */ + Packet * getCoherenceReq(); + + /** + * Snoops bus transactions to maintain coherence. + * @param req The current bus transaction. + */ + void snoop(Packet * &pkt); + + void snoopResponse(Packet * &pkt); + + /** + * Invalidates the block containing address if found. + * @param addr The address to look for. + * @param asid The address space ID of the address. + * @todo Is this function necessary? + */ + void invalidateBlk(Addr addr, int asid); + + /** + * Aquash all requests associated with specified thread. + * intended for use by I-cache. + * @param req->getThreadNum()ber The thread to squash. + */ + void squash(int threadNum) + { + missQueue->squash(threadNum); + } + + /** + * Return the number of outstanding misses in a Cache. + * Default returns 0. + * + * @retval unsigned The number of missing still outstanding. + */ + unsigned outstandingMisses() const + { + return missQueue->getMisses(); + } + + /** + * Send a response to the slave interface. + * @param req The request being responded to. + * @param time The time the response is ready. + */ + void respond(Packet * &pkt, Tick time) + { + //si->respond(pkt,time); + cpuSidePort->sendAtomic(pkt); + } + + /** + * Perform the access specified in the request and return the estimated + * time of completion. This function can either update the hierarchy state + * or just perform the access wherever the data is found depending on the + * state of the update flag. + * @param req The memory request to satisfy + * @param update If true, update the hierarchy, otherwise just perform the + * request. + * @return The estimated completion time. + */ + Tick probe(Packet * &pkt, bool update); + + /** + * Snoop for the provided request in the cache and return the estimated + * time of completion. + * @todo Can a snoop probe not change state? + * @param req The memory request to satisfy + * @param update If true, update the hierarchy, otherwise just perform the + * request. + * @return The estimated completion time. + */ + Tick snoopProbe(Packet * &pkt, bool update); +}; + +#endif // __CACHE_HH__ diff --git a/src/mem/cache/cache_blk.hh b/src/mem/cache/cache_blk.hh new file mode 100644 index 000000000..67e65d25b --- /dev/null +++ b/src/mem/cache/cache_blk.hh @@ -0,0 +1,198 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** @file + * Definitions of a simple cache block class. + */ + +#ifndef __CACHE_BLK_HH__ +#define __CACHE_BLK_HH__ + +#include "sim/root.hh" // for Tick +#include "arch/isa_traits.hh" // for Addr + +/** + * Cache block status bit assignments + */ +enum CacheBlkStatusBits { + /** valid, readable */ + BlkValid = 0x01, + /** write permission */ + BlkWritable = 0x02, + /** dirty (modified) */ + BlkDirty = 0x04, + /** compressed */ + BlkCompressed = 0x08, + /** block was referenced */ + BlkReferenced = 0x10, + /** block was a hardware prefetch yet unaccessed*/ + BlkHWPrefetched = 0x20 +}; + +/** + * A Basic Cache block. + * Contains the tag, status, and a pointer to data. + */ +class CacheBlk +{ + public: + /** The address space ID of this block. */ + int asid; + /** Data block tag value. */ + Addr tag; + /** + * Contains a copy of the data in this block for easy access. This is used + * for efficient execution when the data could be actually stored in + * another format (COW, compressed, sub-blocked, etc). In all cases the + * data stored here should be kept consistant with the actual data + * referenced by this block. + */ + uint8_t *data; + /** the number of bytes stored in this block. */ + int size; + + /** block state: OR of CacheBlkStatusBit */ + typedef unsigned State; + + /** The current status of this block. @sa CacheBlockStatusBits */ + State status; + + /** Which curTick will this block be accessable */ + Tick whenReady; + + /** + * The set this block belongs to. + * @todo Move this into subclasses when we fix CacheTags to use them. + */ + int set; + + /** Number of references to this block since it was brought in. */ + int refCount; + + CacheBlk() + : asid(-1), tag(0), data(0) ,size(0), status(0), whenReady(0), + set(-1), refCount(0) + {} + + /** + * Copy the state of the given block into this one. + * @param rhs The block to copy. + * @return a const reference to this block. + */ + const CacheBlk& operator=(const CacheBlk& rhs) + { + asid = rhs.asid; + tag = rhs.tag; + data = rhs.data; + size = rhs.size; + status = rhs.status; + whenReady = rhs.whenReady; + set = rhs.set; + refCount = rhs.refCount; + return *this; + } + + /** + * Checks the write permissions of this block. + * @return True if the block is writable. + */ + bool isWritable() const + { + const int needed_bits = BlkWritable | BlkValid; + return (status & needed_bits) == needed_bits; + } + + /** + * Checks that a block is valid (readable). + * @return True if the block is valid. + */ + bool isValid() const + { + return (status & BlkValid) != 0; + } + + /** + * Check to see if a block has been written. + * @return True if the block is dirty. + */ + bool isModified() const + { + return (status & BlkDirty) != 0; + } + + /** + * Check to see if this block contains compressed data. + * @return True iF the block's data is compressed. + */ + bool isCompressed() const + { + return (status & BlkCompressed) != 0; + } + + /** + * Check if this block has been referenced. + * @return True if the block has been referenced. + */ + bool isReferenced() const + { + return (status & BlkReferenced) != 0; + } + + /** + * Check if this block was the result of a hardware prefetch, yet to + * be touched. + * @return True if the block was a hardware prefetch, unaccesed. + */ + bool isPrefetch() const + { + return (status & BlkHWPrefetched) != 0; + } + + +}; + +/** + * Output a CacheBlk to the given ostream. + * @param out The stream for the output. + * @param blk The cache block to print. + * + * @return The output stream. + */ +inline std::ostream & +operator<<(std::ostream &out, const CacheBlk &blk) +{ + out << std::hex << std::endl; + out << " Tag: " << blk.tag << std::endl; + out << " Status: " << blk.status << std::endl; + + return(out << std::dec); +} + +#endif //__CACHE_BLK_HH__ diff --git a/src/mem/cache/cache_builder.cc b/src/mem/cache/cache_builder.cc new file mode 100644 index 000000000..8758dc57a --- /dev/null +++ b/src/mem/cache/cache_builder.cc @@ -0,0 +1,480 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + * Nathan Binkert + */ + +/** + * @file + * Simobject instatiation of caches. + */ +#include <vector> + +// Must be included first to determine which caches we want +#include "mem/config/cache.hh" +#include "mem/config/compression.hh" +#include "mem/config/prefetch.hh" + +#include "mem/cache/base_cache.hh" +#include "mem/cache/cache.hh" +#include "mem/bus.hh" +#include "mem/cache/coherence/coherence_protocol.hh" +#include "sim/builder.hh" + +// Tag Templates +#if defined(USE_CACHE_LRU) +#include "mem/cache/tags/lru.hh" +#endif + +#if defined(USE_CACHE_FALRU) +#include "mem/cache/tags/fa_lru.hh" +#endif + +#if defined(USE_CACHE_IIC) +#include "mem/cache/tags/iic.hh" +#endif + +#if defined(USE_CACHE_SPLIT) +#include "mem/cache/tags/split.hh" +#endif + +#if defined(USE_CACHE_SPLIT_LIFO) +#include "mem/cache/tags/split_lifo.hh" +#endif + +// Compression Templates +#include "base/compression/null_compression.hh" +#if defined(USE_LZSS_COMPRESSION) +#include "base/compression/lzss_compression.hh" +#endif + +// CacheTags Templates +#include "mem/cache/tags/cache_tags.hh" + +// MissQueue Templates +#include "mem/cache/miss/miss_queue.hh" +#include "mem/cache/miss/blocking_buffer.hh" + +// Coherence Templates +#include "mem/cache/coherence/uni_coherence.hh" +#include "mem/cache/coherence/simple_coherence.hh" + +//Prefetcher Headers +#if defined(USE_GHB) +#include "mem/cache/prefetch/ghb_prefetcher.hh" +#endif +#if defined(USE_TAGGED) +#include "mem/cache/prefetch/tagged_prefetcher.hh" +#endif +#if defined(USE_STRIDED) +#include "mem/cache/prefetch/stride_prefetcher.hh" +#endif + + +using namespace std; +using namespace TheISA; + +#ifndef DOXYGEN_SHOULD_SKIP_THIS + +BEGIN_DECLARE_SIM_OBJECT_PARAMS(BaseCache) + + Param<int> size; + Param<int> assoc; + Param<int> block_size; + Param<int> latency; + Param<int> mshrs; + Param<int> tgts_per_mshr; + Param<int> write_buffers; + Param<bool> prioritizeRequests; +// SimObjectParam<Bus *> in_bus; +// SimObjectParam<Bus *> out_bus; + Param<bool> do_copy; + SimObjectParam<CoherenceProtocol *> protocol; + Param<Addr> trace_addr; + Param<int> hash_delay; +#if defined(USE_CACHE_IIC) + SimObjectParam<Repl *> repl; +#endif + Param<bool> compressed_bus; + Param<bool> store_compressed; + Param<bool> adaptive_compression; + Param<int> compression_latency; + Param<int> subblock_size; + Param<Counter> max_miss_count; +// SimObjectParam<HierParams *> hier; + VectorParam<Range<Addr> > addr_range; +// SimObjectParam<MemTraceWriter *> mem_trace; + Param<bool> split; + Param<int> split_size; + Param<bool> lifo; + Param<bool> two_queue; + Param<bool> prefetch_miss; + Param<bool> prefetch_access; + Param<int> prefetcher_size; + Param<bool> prefetch_past_page; + Param<bool> prefetch_serial_squash; + Param<Tick> prefetch_latency; + Param<int> prefetch_degree; + Param<string> prefetch_policy; + Param<bool> prefetch_cache_check_push; + Param<bool> prefetch_use_cpu_id; + Param<bool> prefetch_data_accesses_only; + Param<int> hit_latency; + +END_DECLARE_SIM_OBJECT_PARAMS(BaseCache) + + +BEGIN_INIT_SIM_OBJECT_PARAMS(BaseCache) + + INIT_PARAM(size, "capacity in bytes"), + INIT_PARAM(assoc, "associativity"), + INIT_PARAM(block_size, "block size in bytes"), + INIT_PARAM(latency, "hit latency in CPU cycles"), + INIT_PARAM(mshrs, "number of MSHRs (max outstanding requests)"), + INIT_PARAM(tgts_per_mshr, "max number of accesses per MSHR"), + INIT_PARAM_DFLT(write_buffers, "number of write buffers", 8), + INIT_PARAM_DFLT(prioritizeRequests, "always service demand misses first", + false), +/* INIT_PARAM_DFLT(in_bus, "incoming bus object", NULL), + INIT_PARAM(out_bus, "outgoing bus object"), +*/ + INIT_PARAM_DFLT(do_copy, "perform fast copies in the cache", false), + INIT_PARAM_DFLT(protocol, "coherence protocol to use in the cache", NULL), + INIT_PARAM_DFLT(trace_addr, "address to trace", 0), + + INIT_PARAM_DFLT(hash_delay, "time in cycles of hash access",1), +#if defined(USE_CACHE_IIC) + INIT_PARAM_DFLT(repl, "replacement policy",NULL), +#endif + INIT_PARAM_DFLT(compressed_bus, + "This cache connects to a compressed memory", + false), + INIT_PARAM_DFLT(store_compressed, "Store compressed data in the cache", + false), + INIT_PARAM_DFLT(adaptive_compression, "Use an adaptive compression scheme", + false), + INIT_PARAM_DFLT(compression_latency, + "Latency in cycles of compression algorithm", + 0), + INIT_PARAM_DFLT(subblock_size, + "Size of subblock in IIC used for compression", + 0), + INIT_PARAM_DFLT(max_miss_count, + "The number of misses to handle before calling exit", + 0), +/* INIT_PARAM_DFLT(hier, + "Hierarchy global variables", + &defaultHierParams), +*/ + INIT_PARAM_DFLT(addr_range, "The address range in bytes", + vector<Range<Addr> >(1,RangeIn((Addr)0, MaxAddr))), +// INIT_PARAM_DFLT(mem_trace, "Memory trace to write accesses to", NULL), + INIT_PARAM_DFLT(split, "Whether this is a partitioned cache", false), + INIT_PARAM_DFLT(split_size, "the number of \"ways\" belonging to the LRU partition", 0), + INIT_PARAM_DFLT(lifo, "whether you are using a LIFO repl. policy", false), + INIT_PARAM_DFLT(two_queue, "whether the lifo should have two queue replacement", false), + INIT_PARAM_DFLT(prefetch_miss, "wheter you are using the hardware prefetcher from Miss stream", false), + INIT_PARAM_DFLT(prefetch_access, "wheter you are using the hardware prefetcher from Access stream", false), + INIT_PARAM_DFLT(prefetcher_size, "Number of entries in the harware prefetch queue", 100), + INIT_PARAM_DFLT(prefetch_past_page, "Allow prefetches to cross virtual page boundaries", false), + INIT_PARAM_DFLT(prefetch_serial_squash, "Squash prefetches with a later time on a subsequent miss", false), + INIT_PARAM_DFLT(prefetch_latency, "Latency of the prefetcher", 10), + INIT_PARAM_DFLT(prefetch_degree, "Degree of the prefetch depth", 1), + INIT_PARAM_DFLT(prefetch_policy, "Type of prefetcher to use", "none"), + INIT_PARAM_DFLT(prefetch_cache_check_push, "Check if in cash on push or pop of prefetch queue", true), + INIT_PARAM_DFLT(prefetch_use_cpu_id, "Use the CPU ID to seperate calculations of prefetches", true), + INIT_PARAM_DFLT(prefetch_data_accesses_only, "Only prefetch on data not on instruction accesses", false), + INIT_PARAM_DFLT(hit_latency, "Hit Latecny for a succesful access", 1) +END_INIT_SIM_OBJECT_PARAMS(BaseCache) + + +#define BUILD_CACHE(t, comp, b, c) do { \ + Prefetcher<CacheTags<t, comp>, b> *pf; \ + if (pf_policy == "tagged") { \ + BUILD_TAGGED_PREFETCHER(t, comp, b); \ + } \ + else if (pf_policy == "stride") { \ + BUILD_STRIDED_PREFETCHER(t, comp, b); \ + } \ + else if (pf_policy == "ghb") { \ + BUILD_GHB_PREFETCHER(t, comp, b); \ + } \ + else { \ + BUILD_NULL_PREFETCHER(t, comp, b); \ + } \ + Cache<CacheTags<t, comp>, b, c>::Params params(tagStore, mq, coh, \ + do_copy, base_params, \ + /*in_bus, out_bus,*/ pf, \ + prefetch_access); \ + Cache<CacheTags<t, comp>, b, c> *retval = \ + new Cache<CacheTags<t, comp>, b, c>(getInstanceName(), /*hier,*/ \ + params); \ +/* if (in_bus == NULL) { \ + retval->setSlaveInterface(new MemoryInterface<Cache<CacheTags<t, comp>, b, c> >(getInstanceName(), hier, retval, mem_trace)); \ + } else { \ + retval->setSlaveInterface(new SlaveInterface<Cache<CacheTags<t, comp>, b, c>, Bus>(getInstanceName(), hier, retval, in_bus, mem_trace)); \ + } \ + retval->setMasterInterface(new MasterInterface<Cache<CacheTags<t, comp>, b, c>, Bus>(getInstanceName(), hier, retval, out_bus)); \ + out_bus->rangeChange(); \ + return retval; \ +*/return true; \ + } while (0) + +#define BUILD_CACHE_PANIC(x) do { \ + panic("%s not compiled into M5", x); \ + } while (0) + +#if defined(USE_LZSS_COMPRESSION) +#define BUILD_COMPRESSED_CACHE(TAGS, tags, b, c) do { \ + if (compressed_bus || store_compressed){ \ + CacheTags<TAGS, LZSSCompression> *tagStore = \ + new CacheTags<TAGS, LZSSCompression>(tags, \ + compression_latency, \ + true, store_compressed, \ + adaptive_compression, \ + prefetch_miss); \ + BUILD_CACHE(TAGS, LZSSCompression, b, c); \ + } else { \ + CacheTags<TAGS, NullCompression> *tagStore = \ + new CacheTags<TAGS, NullCompression>(tags, \ + compression_latency, \ + true, store_compressed, \ + adaptive_compression, \ + prefetch_miss); \ + BUILD_CACHE(TAGS, NullCompression, b, c); \ + } \ + } while (0) +#else +#define BUILD_COMPRESSED_CACHE(TAGS, tags, b, c) do { \ + if (compressed_bus || store_compressed){ \ + BUILD_CACHE_PANIC("compressed caches"); \ + } else { \ + CacheTags<TAGS, NullCompression> *tagStore = \ + new CacheTags<TAGS, NullCompression>(tags, \ + compression_latency, \ + true, store_compressed, \ + adaptive_compression \ + prefetch_miss); \ + BUILD_CACHE(TAGS, NullCompression, b, c); \ + } \ + } while (0) +#endif + +#if defined(USE_CACHE_FALRU) +#define BUILD_FALRU_CACHE(b,c) do { \ + FALRU *tags = new FALRU(block_size, size, latency); \ + BUILD_COMPRESSED_CACHE(FALRU, tags, b, c); \ + } while (0) +#else +#define BUILD_FALRU_CACHE(b, c) BUILD_CACHE_PANIC("falru cache") +#endif + +#if defined(USE_CACHE_LRU) +#define BUILD_LRU_CACHE(b, c) do { \ + LRU *tags = new LRU(numSets, block_size, assoc, latency); \ + BUILD_COMPRESSED_CACHE(LRU, tags, b, c); \ + } while (0) +#else +#define BUILD_LRU_CACHE(b, c) BUILD_CACHE_PANIC("lru cache") +#endif + +#if defined(USE_CACHE_SPLIT) +#define BUILD_SPLIT_CACHE(b, c) do { \ + Split *tags = new Split(numSets, block_size, assoc, split_size, lifo, \ + two_queue, latency); \ + BUILD_COMPRESSED_CACHE(Split, tags, b, c); \ + } while (0) +#else +#define BUILD_SPLIT_CACHE(b, c) BUILD_CACHE_PANIC("split cache") +#endif + +#if defined(USE_CACHE_SPLIT_LIFO) +#define BUILD_SPLIT_LIFO_CACHE(b, c) do { \ + SplitLIFO *tags = new SplitLIFO(block_size, size, assoc, \ + latency, two_queue, -1); \ + BUILD_COMPRESSED_CACHE(SplitLIFO, tags, b, c); \ + } while (0) +#else +#define BUILD_SPLIT_LIFO_CACHE(b, c) BUILD_CACHE_PANIC("lifo cache") +#endif + +#if defined(USE_CACHE_IIC) +#define BUILD_IIC_CACHE(b ,c) do { \ + IIC *tags = new IIC(iic_params); \ + BUILD_COMPRESSED_CACHE(IIC, tags, b, c); \ + } while (0) +#else +#define BUILD_IIC_CACHE(b, c) BUILD_CACHE_PANIC("iic") +#endif + +#define BUILD_CACHES(b, c) do { \ + if (repl == NULL) { \ + if (numSets == 1) { \ + BUILD_FALRU_CACHE(b, c); \ + } else { \ + if (split == true) { \ + BUILD_SPLIT_CACHE(b, c); \ + } else if (lifo == true) { \ + BUILD_SPLIT_LIFO_CACHE(b, c); \ + } else { \ + BUILD_LRU_CACHE(b, c); \ + } \ + } \ + } else { \ + BUILD_IIC_CACHE(b, c); \ + } \ + } while (0) + +#define BUILD_COHERENCE(b) do { \ + if (protocol == NULL) { \ + UniCoherence *coh = new UniCoherence(); \ + BUILD_CACHES(b, UniCoherence); \ + } else { \ + SimpleCoherence *coh = new SimpleCoherence(protocol); \ + BUILD_CACHES(b, SimpleCoherence); \ + } \ + } while (0) + +#if defined(USE_TAGGED) +#define BUILD_TAGGED_PREFETCHER(t, comp, b) pf = new \ + TaggedPrefetcher<CacheTags<t, comp>, b>(prefetcher_size, \ + !prefetch_past_page, \ + prefetch_serial_squash, \ + prefetch_cache_check_push, \ + prefetch_data_accesses_only, \ + prefetch_latency, \ + prefetch_degree) +#else +#define BUILD_TAGGED_PREFETCHER(t, comp, b) BUILD_CACHE_PANIC("Tagged Prefetcher") +#endif + +#if defined(USE_STRIDED) +#define BUILD_STRIDED_PREFETCHER(t, comp, b) pf = new \ + StridePrefetcher<CacheTags<t, comp>, b>(prefetcher_size, \ + !prefetch_past_page, \ + prefetch_serial_squash, \ + prefetch_cache_check_push, \ + prefetch_data_accesses_only, \ + prefetch_latency, \ + prefetch_degree, \ + prefetch_use_cpu_id) +#else +#define BUILD_STRIDED_PREFETCHER(t, comp, b) BUILD_CACHE_PANIC("Stride Prefetcher") +#endif + +#if defined(USE_GHB) +#define BUILD_GHB_PREFETCHER(t, comp, b) pf = new \ + GHBPrefetcher<CacheTags<t, comp>, b>(prefetcher_size, \ + !prefetch_past_page, \ + prefetch_serial_squash, \ + prefetch_cache_check_push, \ + prefetch_data_accesses_only, \ + prefetch_latency, \ + prefetch_degree, \ + prefetch_use_cpu_id) +#else +#define BUILD_GHB_PREFETCHER(t, comp, b) BUILD_CACHE_PANIC("GHB Prefetcher") +#endif + +#if defined(USE_TAGGED) +#define BUILD_NULL_PREFETCHER(t, comp, b) pf = new \ + TaggedPrefetcher<CacheTags<t, comp>, b>(prefetcher_size, \ + !prefetch_past_page, \ + prefetch_serial_squash, \ + prefetch_cache_check_push, \ + prefetch_data_accesses_only, \ + prefetch_latency, \ + prefetch_degree) +#else +#define BUILD_NULL_PREFETCHER(t, comp, b) BUILD_CACHE_PANIC("NULL Prefetcher (uses Tagged)") +#endif + +CREATE_SIM_OBJECT(BaseCache) +{ + string name = getInstanceName(); + int numSets = size / (assoc * block_size); + string pf_policy = prefetch_policy; + if (subblock_size == 0) { + subblock_size = block_size; + } + + // Build BaseCache param object + BaseCache::Params base_params(addr_range, latency, + block_size, max_miss_count); + + //Warnings about prefetcher policy + if (pf_policy == "none" && (prefetch_miss || prefetch_access)) { + panic("With no prefetcher, you shouldn't prefetch from" + " either miss or access stream\n"); + } + if ((pf_policy == "tagged" || pf_policy == "stride" || + pf_policy == "ghb") && !(prefetch_miss || prefetch_access)) { + warn("With this prefetcher you should chose a prefetch" + " stream (miss or access)\nNo Prefetching will occur\n"); + } + if ((pf_policy == "tagged" || pf_policy == "stride" || + pf_policy == "ghb") && prefetch_miss && prefetch_access) { + panic("Can't do prefetches from both miss and access" + " stream\n"); + } + if (pf_policy != "tagged" && pf_policy != "stride" && + pf_policy != "ghb" && pf_policy != "none") { + panic("Unrecognized form of a prefetcher: %s, try using" + "['none','stride','tagged','ghb']\n", pf_policy); + } + +#if defined(USE_CACHE_IIC) + // Build IIC params + IIC::Params iic_params; + iic_params.size = size; + iic_params.numSets = numSets; + iic_params.blkSize = block_size; + iic_params.assoc = assoc; + iic_params.hashDelay = hash_delay; + iic_params.hitLatency = latency; + iic_params.rp = repl; + iic_params.subblockSize = subblock_size; +#else + const void *repl = NULL; +#endif + + if (mshrs == 1 /*|| out_bus->doEvents() == false*/) { + BlockingBuffer *mq = new BlockingBuffer(true); + BUILD_COHERENCE(BlockingBuffer); + } else { + MissQueue *mq = new MissQueue(mshrs, tgts_per_mshr, write_buffers, + true, prefetch_miss); + BUILD_COHERENCE(MissQueue); + } + return NULL; +} + +REGISTER_SIM_OBJECT("BaseCache", BaseCache) + + +#endif //DOXYGEN_SHOULD_SKIP_THIS diff --git a/src/mem/cache/cache_impl.hh b/src/mem/cache/cache_impl.hh new file mode 100644 index 000000000..dbf2e49f1 --- /dev/null +++ b/src/mem/cache/cache_impl.hh @@ -0,0 +1,655 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + * Dave Greene + * Nathan Binkert + */ + +/** + * @file + * Cache definitions. + */ + +#include <assert.h> +#include <math.h> + +#include <cassert> +#include <iostream> +#include <string> + +#include "sim/host.hh" +#include "base/misc.hh" +#include "cpu/smt.hh" + +#include "mem/cache/cache.hh" +#include "mem/cache/cache_blk.hh" +#include "mem/cache/miss/mshr.hh" +#include "mem/cache/prefetch/prefetcher.hh" + +#include "sim/sim_events.hh" // for SimExitEvent + +using namespace std; + +template<class TagStore, class Buffering, class Coherence> +bool +Cache<TagStore,Buffering,Coherence>:: +doTimingAccess(Packet *pkt, CachePort *cachePort, bool isCpuSide) +{ + if (isCpuSide) + { + access(pkt); + } + else + { + if (pkt->isResponse()) + handleResponse(pkt); + else + snoop(pkt); + } + return true; //Deal with blocking.... +} + +template<class TagStore, class Buffering, class Coherence> +Tick +Cache<TagStore,Buffering,Coherence>:: +doAtomicAccess(Packet *pkt, CachePort *cachePort, bool isCpuSide) +{ + if (isCpuSide) + { + probe(pkt, true); + } + else + { + if (pkt->isResponse()) + handleResponse(pkt); + else + snoopProbe(pkt, true); + } + //Fix this timing info + return hitLatency; +} + +template<class TagStore, class Buffering, class Coherence> +void +Cache<TagStore,Buffering,Coherence>:: +doFunctionalAccess(Packet *pkt, CachePort *cachePort, bool isCpuSide) +{ + if (isCpuSide) + { + probe(pkt, false); + } + else + { + if (pkt->isResponse()) + handleResponse(pkt); + else + snoopProbe(pkt, false); + } +} + +template<class TagStore, class Buffering, class Coherence> +void +Cache<TagStore,Buffering,Coherence>:: +recvStatusChange(Port::Status status, bool isCpuSide) +{ + +} + + +template<class TagStore, class Buffering, class Coherence> +Cache<TagStore,Buffering,Coherence>:: +Cache(const std::string &_name, + Cache<TagStore,Buffering,Coherence>::Params ¶ms) + : BaseCache(_name, params.baseParams), + prefetchAccess(params.prefetchAccess), + tags(params.tags), missQueue(params.missQueue), + coherence(params.coherence), prefetcher(params.prefetcher), + doCopy(params.doCopy), blockOnCopy(params.blockOnCopy) +{ +//FIX BUS POINTERS +// if (params.in == NULL) { + topLevelCache = true; +// } +//PLEASE FIX THIS, BUS SIZES NOT BEING USED + tags->setCache(this, blkSize, 1/*params.out->width, params.out->clockRate*/); + tags->setPrefetcher(prefetcher); + missQueue->setCache(this); + missQueue->setPrefetcher(prefetcher); + coherence->setCache(this); + prefetcher->setCache(this); + prefetcher->setTags(tags); + prefetcher->setBuffer(missQueue); +#if 0 + invalidatePkt = new Packet; + invalidatePkt->cmd = Packet::InvalidateReq; +#endif +} + +template<class TagStore, class Buffering, class Coherence> +void +Cache<TagStore,Buffering,Coherence>::regStats() +{ + BaseCache::regStats(); + tags->regStats(name()); + missQueue->regStats(name()); + coherence->regStats(name()); + prefetcher->regStats(name()); +} + +template<class TagStore, class Buffering, class Coherence> +bool +Cache<TagStore,Buffering,Coherence>::access(PacketPtr &pkt) +{ +//@todo Add back in MemDebug Calls +// MemDebug::cacheAccess(pkt); + BlkType *blk = NULL; + PacketList writebacks; + int size = blkSize; + int lat = hitLatency; + if (prefetchAccess) { + //We are determining prefetches on access stream, call prefetcher + prefetcher->handleMiss(pkt, curTick); + } + if (!pkt->req->isUncacheable()) { + if (pkt->isInvalidate() && !pkt->isRead() + && !pkt->isWrite()) { + //Upgrade or Invalidate + //Look into what happens if two slave caches on bus + DPRINTF(Cache, "%s %d %x ? blk_addr: %x\n", pkt->cmdString(), + pkt->req->getAsid(), pkt->getAddr() & (((ULL(1))<<48)-1), + pkt->getAddr() & ~((Addr)blkSize - 1)); + + //@todo Should this return latency have the hit latency in it? +// respond(pkt,curTick+lat); + pkt->flags |= SATISFIED; +// return MA_HIT; //@todo, return values + return true; + } + blk = tags->handleAccess(pkt, lat, writebacks); + } else { + size = pkt->getSize(); + } + // If this is a block size write/hint (WH64) allocate the block here + // if the coherence protocol allows it. + /** @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) ) { + // not outstanding misses, can do this + MSHR* outstanding_miss = missQueue->findMSHR(pkt->getAddr(), pkt->req->getAsid()); + if (pkt->cmd == Packet::WriteInvalidateReq || !outstanding_miss) { + if (outstanding_miss) { + warn("WriteInv doing a fastallocate" + "with an outstanding miss to the same address\n"); + } + blk = tags->handleFill(NULL, pkt, BlkValid | BlkWritable, + writebacks); + ++fastWrites; + } + } + while (!writebacks.empty()) { + missQueue->doWriteback(writebacks.front()); + writebacks.pop_front(); + } + DPRINTF(Cache, "%s %d %x %s blk_addr: %x pc %x\n", pkt->cmdString(), + pkt->req->getAsid(), pkt->getAddr() & (((ULL(1))<<48)-1), (blk) ? "hit" : "miss", + pkt->getAddr() & ~((Addr)blkSize - 1), pkt->req->getPC()); + if (blk) { + // Hit + hits[pkt->cmdToIndex()][pkt->req->getThreadNum()]++; + // clear dirty bit if write through + if (pkt->needsResponse()) + respond(pkt, curTick+lat); +// return MA_HIT; + return true; + } + + // Miss + if (!pkt->req->isUncacheable()) { + misses[pkt->cmdToIndex()][pkt->req->getThreadNum()]++; + /** @todo Move miss count code into BaseCache */ + if (missCount) { + --missCount; + if (missCount == 0) + new SimLoopExitEvent(curTick, "A cache reached the maximum miss count"); + } + } + missQueue->handleMiss(pkt, size, curTick + hitLatency); +// return MA_CACHE_MISS; + return true; +} + + +template<class TagStore, class Buffering, class Coherence> +Packet * +Cache<TagStore,Buffering,Coherence>::getPacket() +{ + Packet * pkt = missQueue->getPacket(); + if (pkt) { + if (!pkt->req->isUncacheable()) { + if (pkt->cmd == Packet::HardPFReq) misses[Packet::HardPFReq][pkt->req->getThreadNum()]++; + BlkType *blk = tags->findBlock(pkt); + Packet::Command cmd = coherence->getBusCmd(pkt->cmd, + (blk)? blk->status : 0); + missQueue->setBusCmd(pkt, cmd); + } + } + + assert(!doMasterRequest() || missQueue->havePending()); + assert(!pkt || pkt->time <= curTick); + return pkt; +} + +template<class TagStore, class Buffering, class Coherence> +void +Cache<TagStore,Buffering,Coherence>::sendResult(PacketPtr &pkt, bool success) +{ + if (success) { + missQueue->markInService(pkt); + //Temp Hack for UPGRADES + if (pkt->cmd == Packet::UpgradeReq) { + handleResponse(pkt); + } + } else if (pkt && !pkt->req->isUncacheable()) { + missQueue->restoreOrigCmd(pkt); + } +} + +template<class TagStore, class Buffering, class Coherence> +void +Cache<TagStore,Buffering,Coherence>::handleResponse(Packet * &pkt) +{ + BlkType *blk = NULL; + if (pkt->senderState) { +// MemDebug::cacheResponse(pkt); + DPRINTF(Cache, "Handling reponse to %x, blk addr: %x\n",pkt->getAddr(), + pkt->getAddr() & (((ULL(1))<<48)-1)); + + if (pkt->isCacheFill() && !pkt->isNoAllocate()) { + blk = tags->findBlock(pkt); + CacheBlk::State old_state = (blk) ? blk->status : 0; + PacketList writebacks; + blk = tags->handleFill(blk, (MSHR*)pkt->senderState, + coherence->getNewState(pkt,old_state), + writebacks); + while (!writebacks.empty()) { + missQueue->doWriteback(writebacks.front()); + } + } + missQueue->handleResponse(pkt, curTick + hitLatency); + } +} + +template<class TagStore, class Buffering, class Coherence> +void +Cache<TagStore,Buffering,Coherence>::pseudoFill(Addr addr, int asid) +{ + // Need to temporarily move this blk into MSHRs + MSHR *mshr = missQueue->allocateTargetList(addr, asid); + int lat; + PacketList dummy; + // Read the data into the mshr + BlkType *blk = tags->handleAccess(mshr->pkt, lat, dummy, false); + assert(dummy.empty()); + assert(mshr->pkt->flags & SATISFIED); + // can overload order since it isn't used on non pending blocks + mshr->order = blk->status; + // temporarily remove the block from the cache. + tags->invalidateBlk(addr, asid); +} + +template<class TagStore, class Buffering, class Coherence> +void +Cache<TagStore,Buffering,Coherence>::pseudoFill(MSHR *mshr) +{ + // Need to temporarily move this blk into MSHRs + assert(mshr->pkt->cmd == Packet::ReadReq); + int lat; + PacketList dummy; + // Read the data into the mshr + BlkType *blk = tags->handleAccess(mshr->pkt, lat, dummy, false); + assert(dummy.empty()); + assert(mshr->pkt->flags & SATISFIED); + // can overload order since it isn't used on non pending blocks + mshr->order = blk->status; + // temporarily remove the block from the cache. + tags->invalidateBlk(mshr->pkt->getAddr(), mshr->pkt->req->getAsid()); +} + + +template<class TagStore, class Buffering, class Coherence> +Packet * +Cache<TagStore,Buffering,Coherence>::getCoherenceReq() +{ + return coherence->getPacket(); +} + + +template<class TagStore, class Buffering, class Coherence> +void +Cache<TagStore,Buffering,Coherence>::snoop(Packet * &pkt) +{ + + Addr blk_addr = pkt->getAddr() & ~(Addr(blkSize-1)); + BlkType *blk = tags->findBlock(pkt); + MSHR *mshr = missQueue->findMSHR(blk_addr, pkt->req->getAsid()); + if (isTopLevel() && coherence->hasProtocol()) { //@todo Move this into handle bus req + //If we find an mshr, and it is in service, we need to NACK or invalidate + if (mshr) { + if (mshr->inService) { + if ((mshr->pkt->isInvalidate() || !mshr->pkt->isCacheFill()) + && (pkt->cmd != Packet::InvalidateReq && pkt->cmd != Packet::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 request is not a cachefill (writeback) + pkt->flags |= NACKED_LINE; + return; + } + else { + //The supplier will be someone else, because we are waiting for + //the data. This should cause this cache to be forced to go to + //the shared state, not the exclusive even though the shared line + //won't be asserted. But for now we will just invlidate ourselves + //and allow the other cache to go into the exclusive state. + //@todo Make it so a read to a pending read doesn't invalidate. + //@todo Make it so that a read to a pending read can't be exclusive now. + + //Set the address so find match works + invalidatePkt->addrOverride(pkt->getAddr()); + + //Append the invalidate on + missQueue->addTarget(mshr,invalidatePkt); + DPRINTF(Cache, "Appending Invalidate to blk_addr: %x\n", pkt->getAddr() & (((ULL(1))<<48)-1)); + return; + } + } + } + //We also need to check the writeback buffers and handle those + std::vector<MSHR *> writebacks; + if (missQueue->findWrites(blk_addr, pkt->req->getAsid(), writebacks)) { + DPRINTF(Cache, "Snoop hit in writeback to blk_addr: %x\n", pkt->getAddr() & (((ULL(1))<<48)-1)); + + //Look through writebacks for any non-uncachable writes, use that + for (int i=0; i<writebacks.size(); i++) { + mshr = writebacks[i]; + + if (!mshr->pkt->req->isUncacheable()) { + if (pkt->isRead()) { + //Only Upgrades don't get here + //Supply the data + pkt->flags |= SATISFIED; + + //If we are in an exclusive protocol, make it ask again + //to get write permissions (upgrade), signal shared + pkt->flags |= SHARED_LINE; + + assert(pkt->isRead()); + Addr offset = pkt->getAddr() & ~(blkSize - 1); + assert(offset < blkSize); + assert(pkt->getSize() <= blkSize); + assert(offset + pkt->getSize() <=blkSize); + memcpy(pkt->getPtr<uint8_t>(), mshr->pkt->getPtr<uint8_t>() + offset, pkt->getSize()); + + respondToSnoop(pkt); + } + + if (pkt->isInvalidate()) { + //This must be an upgrade or other cache will take ownership + missQueue->markInService(mshr->pkt); + } + return; + } + } + } + } + CacheBlk::State new_state; + bool satisfy = coherence->handleBusRequest(pkt,blk,mshr, new_state); + if (satisfy) { + tags->handleSnoop(blk, new_state, pkt); + respondToSnoop(pkt); + return; + } + tags->handleSnoop(blk, new_state); +} + +template<class TagStore, class Buffering, class Coherence> +void +Cache<TagStore,Buffering,Coherence>::snoopResponse(Packet * &pkt) +{ + //Need to handle the response, if NACKED + if (pkt->flags & NACKED_LINE) { + //Need to mark it as not in service, and retry for bus + assert(0); //Yeah, we saw a NACK come through + + //For now this should never get called, we return false when we see a NACK + //instead, by doing this we allow the bus_blocked mechanism to handle the retry + //For now it retrys in just 2 cycles, need to figure out how to change that + //Eventually we will want to also have success come in as a parameter + //Need to make sure that we handle the functionality that happens on successufl + //return of the sendAddr function + } +} + +template<class TagStore, class Buffering, class Coherence> +void +Cache<TagStore,Buffering,Coherence>::invalidateBlk(Addr addr, int asid) +{ + tags->invalidateBlk(addr,asid); +} + + +/** + * @todo Fix to not assume write allocate + */ +template<class TagStore, class Buffering, class Coherence> +Tick +Cache<TagStore,Buffering,Coherence>::probe(Packet * &pkt, bool update) +{ +// MemDebug::cacheProbe(pkt); + if (!pkt->req->isUncacheable()) { + if (pkt->isInvalidate() && !pkt->isRead() + && !pkt->isWrite()) { + //Upgrade or Invalidate, satisfy it, don't forward + DPRINTF(Cache, "%s %d %x ? blk_addr: %x\n", pkt->cmdString(), + pkt->req->getAsid(), pkt->getAddr() & (((ULL(1))<<48)-1), + pkt->getAddr() & ~((Addr)blkSize - 1)); + pkt->flags |= SATISFIED; + return 0; + } + } + + PacketList writebacks; + int lat; + BlkType *blk = tags->handleAccess(pkt, lat, writebacks, update); + + if (!blk) { + // Need to check for outstanding misses and writes + Addr blk_addr = pkt->getAddr() & ~(blkSize - 1); + + // There can only be one matching outstanding miss. + MSHR* mshr = missQueue->findMSHR(blk_addr, pkt->req->getAsid()); + + // There can be many matching outstanding writes. + vector<MSHR*> writes; + missQueue->findWrites(blk_addr, pkt->req->getAsid(), writes); + + if (!update) { + memSidePort->sendFunctional(pkt); + // Check for data in MSHR and writebuffer. + if (mshr) { + warn("Found outstanding miss on an non-update probe"); + MSHR::TargetList *targets = mshr->getTargetList(); + MSHR::TargetList::iterator i = targets->begin(); + MSHR::TargetList::iterator end = targets->end(); + for (; i != end; ++i) { + Packet * target = *i; + // If the target contains data, and it overlaps the + // probed request, need to update data + if (target->isWrite() && target->intersect(pkt)) { + uint8_t* pkt_data; + uint8_t* write_data; + int data_size; + if (target->getAddr() < pkt->getAddr()) { + int offset = pkt->getAddr() - target->getAddr(); + pkt_data = pkt->getPtr<uint8_t>(); + write_data = target->getPtr<uint8_t>() + offset; + data_size = target->getSize() - offset; + assert(data_size > 0); + if (data_size > pkt->getSize()) + data_size = pkt->getSize(); + } else { + int offset = target->getAddr() - pkt->getAddr(); + pkt_data = pkt->getPtr<uint8_t>() + offset; + write_data = target->getPtr<uint8_t>(); + data_size = pkt->getSize() - offset; + assert(data_size > pkt->getSize()); + if (data_size > target->getSize()) + data_size = target->getSize(); + } + + if (pkt->isWrite()) { + memcpy(pkt_data, write_data, data_size); + } else { + memcpy(write_data, pkt_data, data_size); + } + } + } + } + for (int i = 0; i < writes.size(); ++i) { + Packet * write = writes[i]->pkt; + if (write->intersect(pkt)) { + warn("Found outstanding write on an non-update probe"); + uint8_t* pkt_data; + uint8_t* write_data; + int data_size; + if (write->getAddr() < pkt->getAddr()) { + int offset = pkt->getAddr() - write->getAddr(); + pkt_data = pkt->getPtr<uint8_t>(); + write_data = write->getPtr<uint8_t>() + offset; + data_size = write->getSize() - offset; + assert(data_size > 0); + if (data_size > pkt->getSize()) + data_size = pkt->getSize(); + } else { + int offset = write->getAddr() - pkt->getAddr(); + pkt_data = pkt->getPtr<uint8_t>() + offset; + write_data = write->getPtr<uint8_t>(); + data_size = pkt->getSize() - offset; + assert(data_size > pkt->getSize()); + if (data_size > write->getSize()) + data_size = write->getSize(); + } + + if (pkt->isWrite()) { + memcpy(pkt_data, write_data, data_size); + } else { + memcpy(write_data, pkt_data, data_size); + } + + } + } + return 0; + } else { + // update the cache state and statistics + if (mshr || !writes.empty()){ + // Can't handle it, return pktuest unsatisfied. + return 0; + } + if (!pkt->req->isUncacheable()) { + // Fetch the cache block to fill + BlkType *blk = tags->findBlock(pkt); + Packet::Command temp_cmd = coherence->getBusCmd(pkt->cmd, + (blk)? blk->status : 0); + + Packet * busPkt = new Packet(pkt->req,temp_cmd, -1, blkSize); + + uint8_t* temp_data = new uint8_t[blkSize]; + busPkt->dataDynamicArray<uint8_t>(temp_data); + + busPkt->time = curTick; + + lat = memSidePort->sendAtomic(busPkt); + + if (!(busPkt->flags & SATISFIED)) { + // blocked at a higher level, just return + return 0; + } + + misses[pkt->cmdToIndex()][pkt->req->getThreadNum()]++; + + CacheBlk::State old_state = (blk) ? blk->status : 0; + tags->handleFill(blk, busPkt, + coherence->getNewState(busPkt, old_state), + writebacks, pkt); + // Handle writebacks if needed + while (!writebacks.empty()){ + memSidePort->sendAtomic(writebacks.front()); + writebacks.pop_front(); + } + return lat + hitLatency; + } else { + return memSidePort->sendAtomic(pkt); + } + } + } else { + // There was a cache hit. + // Handle writebacks if needed + while (!writebacks.empty()){ + memSidePort->sendAtomic(writebacks.front()); + writebacks.pop_front(); + } + + if (update) { + hits[pkt->cmdToIndex()][pkt->req->getThreadNum()]++; + } else if (pkt->isWrite()) { + // Still need to change data in all locations. + return memSidePort->sendAtomic(pkt); + } + return curTick + lat; + } + fatal("Probe not handled.\n"); + return 0; +} + +template<class TagStore, class Buffering, class Coherence> +Tick +Cache<TagStore,Buffering,Coherence>::snoopProbe(PacketPtr &pkt, bool update) +{ + Addr blk_addr = pkt->getAddr() & ~(Addr(blkSize-1)); + BlkType *blk = tags->findBlock(pkt); + MSHR *mshr = missQueue->findMSHR(blk_addr, pkt->req->getAsid()); + CacheBlk::State new_state = 0; + bool satisfy = coherence->handleBusRequest(pkt,blk,mshr, new_state); + if (satisfy) { + tags->handleSnoop(blk, new_state, pkt); + return hitLatency; + } + tags->handleSnoop(blk, new_state); + return 0; +} + diff --git a/src/mem/cache/coherence/coherence_protocol.cc b/src/mem/cache/coherence/coherence_protocol.cc new file mode 100644 index 000000000..bcf3ce9c5 --- /dev/null +++ b/src/mem/cache/coherence/coherence_protocol.cc @@ -0,0 +1,567 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + * Steve Reinhardt + * Ron Dreslinski + */ + +/** + * @file + * Definitions of CoherenceProtocol. + */ + +#include <string> + +#include "base/misc.hh" +#include "mem/cache/miss/mshr.hh" +#include "mem/cache/cache.hh" +#include "mem/cache/coherence/coherence_protocol.hh" +#include "sim/builder.hh" + +using namespace std; + + +CoherenceProtocol::StateTransition::StateTransition() + : busCmd(Packet::InvalidCmd), newState(-1), snoopFunc(invalidTransition) +{ +} + + +void +CoherenceProtocol::regStats() +{ + // Even though we count all the possible transitions in the + // requestCount and snoopCount arrays, most of these are invalid, + // so we just select the interesting ones to print here. + + requestCount[Invalid][Packet::ReadReq] + .name(name() + ".read_invalid") + .desc("read misses to invalid blocks") + ; + + requestCount[Invalid][Packet::WriteReq] + .name(name() +".write_invalid") + .desc("write misses to invalid blocks") + ; + + requestCount[Invalid][Packet::SoftPFReq] + .name(name() +".swpf_invalid") + .desc("soft prefetch misses to invalid blocks") + ; + + requestCount[Invalid][Packet::HardPFReq] + .name(name() +".hwpf_invalid") + .desc("hard prefetch misses to invalid blocks") + ; + + requestCount[Shared][Packet::WriteReq] + .name(name() + ".write_shared") + .desc("write misses to shared blocks") + ; + + requestCount[Owned][Packet::WriteReq] + .name(name() + ".write_owned") + .desc("write misses to owned blocks") + ; + + snoopCount[Shared][Packet::ReadReq] + .name(name() + ".snoop_read_shared") + .desc("read snoops on shared blocks") + ; + + snoopCount[Shared][Packet::ReadExReq] + .name(name() + ".snoop_readex_shared") + .desc("readEx snoops on shared blocks") + ; + + snoopCount[Shared][Packet::UpgradeReq] + .name(name() + ".snoop_upgrade_shared") + .desc("upgradee snoops on shared blocks") + ; + + snoopCount[Modified][Packet::ReadReq] + .name(name() + ".snoop_read_modified") + .desc("read snoops on modified blocks") + ; + + snoopCount[Modified][Packet::ReadExReq] + .name(name() + ".snoop_readex_modified") + .desc("readEx snoops on modified blocks") + ; + + snoopCount[Owned][Packet::ReadReq] + .name(name() + ".snoop_read_owned") + .desc("read snoops on owned blocks") + ; + + snoopCount[Owned][Packet::ReadExReq] + .name(name() + ".snoop_readex_owned") + .desc("readEx snoops on owned blocks") + ; + + snoopCount[Owned][Packet::UpgradeReq] + .name(name() + ".snoop_upgrade_owned") + .desc("upgrade snoops on owned blocks") + ; + + snoopCount[Exclusive][Packet::ReadReq] + .name(name() + ".snoop_read_exclusive") + .desc("read snoops on exclusive blocks") + ; + + snoopCount[Exclusive][Packet::ReadExReq] + .name(name() + ".snoop_readex_exclusive") + .desc("readEx snoops on exclusive blocks") + ; + + snoopCount[Shared][Packet::InvalidateReq] + .name(name() + ".snoop_inv_shared") + .desc("Invalidate snoops on shared blocks") + ; + + snoopCount[Owned][Packet::InvalidateReq] + .name(name() + ".snoop_inv_owned") + .desc("Invalidate snoops on owned blocks") + ; + + snoopCount[Exclusive][Packet::InvalidateReq] + .name(name() + ".snoop_inv_exclusive") + .desc("Invalidate snoops on exclusive blocks") + ; + + snoopCount[Modified][Packet::InvalidateReq] + .name(name() + ".snoop_inv_modified") + .desc("Invalidate snoops on modified blocks") + ; + + snoopCount[Invalid][Packet::InvalidateReq] + .name(name() + ".snoop_inv_invalid") + .desc("Invalidate snoops on invalid blocks") + ; + + snoopCount[Shared][Packet::WriteInvalidateReq] + .name(name() + ".snoop_writeinv_shared") + .desc("WriteInvalidate snoops on shared blocks") + ; + + snoopCount[Owned][Packet::WriteInvalidateReq] + .name(name() + ".snoop_writeinv_owned") + .desc("WriteInvalidate snoops on owned blocks") + ; + + snoopCount[Exclusive][Packet::WriteInvalidateReq] + .name(name() + ".snoop_writeinv_exclusive") + .desc("WriteInvalidate snoops on exclusive blocks") + ; + + snoopCount[Modified][Packet::WriteInvalidateReq] + .name(name() + ".snoop_writeinv_modified") + .desc("WriteInvalidate snoops on modified blocks") + ; + + snoopCount[Invalid][Packet::WriteInvalidateReq] + .name(name() + ".snoop_writeinv_invalid") + .desc("WriteInvalidate snoops on invalid blocks") + ; +} + + +bool +CoherenceProtocol::invalidateTrans(BaseCache *cache, Packet * &pkt, + CacheBlk *blk, MSHR *mshr, + CacheBlk::State & new_state) +{ + // invalidate the block + new_state = (blk->status & ~stateMask) | Invalid; + return false; +} + + +bool +CoherenceProtocol::supplyTrans(BaseCache *cache, Packet * &pkt, + CacheBlk *blk, + MSHR *mshr, + CacheBlk::State & new_state + ) +{ + return true; +} + + +bool +CoherenceProtocol::supplyAndGotoSharedTrans(BaseCache *cache, Packet * &pkt, + CacheBlk *blk, + MSHR *mshr, + CacheBlk::State & new_state) +{ + new_state = (blk->status & ~stateMask) | Shared; + pkt->flags |= SHARED_LINE; + return supplyTrans(cache, pkt, blk, mshr, new_state); +} + + +bool +CoherenceProtocol::supplyAndGotoOwnedTrans(BaseCache *cache, Packet * &pkt, + CacheBlk *blk, + MSHR *mshr, + CacheBlk::State & new_state) +{ + new_state = (blk->status & ~stateMask) | Owned; + pkt->flags |= SHARED_LINE; + return supplyTrans(cache, pkt, blk, mshr, new_state); +} + + +bool +CoherenceProtocol::supplyAndInvalidateTrans(BaseCache *cache, Packet * &pkt, + CacheBlk *blk, + MSHR *mshr, + CacheBlk::State & new_state) +{ + new_state = (blk->status & ~stateMask) | Invalid; + return supplyTrans(cache, pkt, blk, mshr, new_state); +} + +bool +CoherenceProtocol::assertShared(BaseCache *cache, Packet * &pkt, + CacheBlk *blk, + MSHR *mshr, + CacheBlk::State & new_state) +{ + new_state = (blk->status & ~stateMask) | Shared; + pkt->flags |= SHARED_LINE; + return false; +} + +CoherenceProtocol::CoherenceProtocol(const string &name, + const string &protocol, + const bool doUpgrades) + : SimObject(name) +{ + if ((protocol == "mosi" || protocol == "moesi") && !doUpgrades) { + cerr << "CoherenceProtocol: ownership protocols require upgrade transactions" + << "(write miss on owned block generates ReadExcl, which will clobber dirty block)" + << endl; + fatal(""); + } + + Packet::Command writeToSharedCmd = doUpgrades ? Packet::UpgradeReq : Packet::ReadExReq; + Packet::Command writeToSharedResp = doUpgrades ? Packet::UpgradeResp : Packet::ReadExResp; + +//@todo add in hardware prefetch to this list + if (protocol == "msi") { + // incoming requests: specify outgoing bus request + transitionTable[Invalid][Packet::ReadReq].onRequest(Packet::ReadReq); + transitionTable[Invalid][Packet::WriteReq].onRequest(Packet::ReadExReq); + transitionTable[Shared][Packet::WriteReq].onRequest(writeToSharedCmd); + //Prefetching causes a read + transitionTable[Invalid][Packet::SoftPFReq].onRequest(Packet::ReadReq); + transitionTable[Invalid][Packet::HardPFReq].onRequest(Packet::ReadReq); + + // on response to given request: specify new state + transitionTable[Invalid][Packet::ReadResp].onResponse(Shared); + transitionTable[Invalid][Packet::ReadExResp].onResponse(Modified); + transitionTable[Shared][writeToSharedResp].onResponse(Modified); + + // bus snoop transition functions + transitionTable[Invalid][Packet::ReadReq].onSnoop(nullTransition); + transitionTable[Invalid][Packet::ReadExReq].onSnoop(nullTransition); + transitionTable[Shared][Packet::ReadReq].onSnoop(nullTransition); + transitionTable[Shared][Packet::ReadExReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::ReadExReq].onSnoop(supplyAndInvalidateTrans); + transitionTable[Modified][Packet::ReadReq].onSnoop(supplyAndGotoSharedTrans); + //Tansitions on seeing a DMA (writeInv(samelevel) or DMAInv) + transitionTable[Invalid][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Shared][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Invalid][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Shared][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + + if (doUpgrades) { + transitionTable[Invalid][Packet::UpgradeReq].onSnoop(nullTransition); + transitionTable[Shared][Packet::UpgradeReq].onSnoop(invalidateTrans); + } + } + + else if(protocol == "mesi") { + // incoming requests: specify outgoing bus request + transitionTable[Invalid][Packet::ReadReq].onRequest(Packet::ReadReq); + transitionTable[Invalid][Packet::WriteReq].onRequest(Packet::ReadExReq); + transitionTable[Shared][Packet::WriteReq].onRequest(writeToSharedCmd); + //Prefetching causes a read + transitionTable[Invalid][Packet::SoftPFReq].onRequest(Packet::ReadReq); + transitionTable[Invalid][Packet::HardPFReq].onRequest(Packet::ReadReq); + + // on response to given request: specify new state + transitionTable[Invalid][Packet::ReadResp].onResponse(Exclusive); + //It will move into shared if the shared line is asserted in the + //getNewState function + transitionTable[Invalid][Packet::ReadExResp].onResponse(Modified); + transitionTable[Shared][writeToSharedResp].onResponse(Modified); + + // bus snoop transition functions + transitionTable[Invalid][Packet::ReadReq].onSnoop(nullTransition); + transitionTable[Invalid][Packet::ReadExReq].onSnoop(nullTransition); + transitionTable[Shared][Packet::ReadReq].onSnoop(assertShared); + transitionTable[Shared][Packet::ReadExReq].onSnoop(invalidateTrans); + transitionTable[Exclusive][Packet::ReadReq].onSnoop(assertShared); + transitionTable[Exclusive][Packet::ReadExReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::ReadExReq].onSnoop(supplyAndInvalidateTrans); + transitionTable[Modified][Packet::ReadReq].onSnoop(supplyAndGotoSharedTrans); + //Tansitions on seeing a DMA (writeInv(samelevel) or DMAInv) + transitionTable[Invalid][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Shared][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Exclusive][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Invalid][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Shared][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Exclusive][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + + if (doUpgrades) { + transitionTable[Invalid][Packet::UpgradeReq].onSnoop(nullTransition); + transitionTable[Shared][Packet::UpgradeReq].onSnoop(invalidateTrans); + } + } + + else if(protocol == "mosi") { + // incoming requests: specify outgoing bus request + transitionTable[Invalid][Packet::ReadReq].onRequest(Packet::ReadReq); + transitionTable[Invalid][Packet::WriteReq].onRequest(Packet::ReadExReq); + transitionTable[Shared][Packet::WriteReq].onRequest(writeToSharedCmd); + transitionTable[Owned][Packet::WriteReq].onRequest(writeToSharedCmd); + //Prefetching causes a read + transitionTable[Invalid][Packet::SoftPFReq].onRequest(Packet::ReadReq); + transitionTable[Invalid][Packet::HardPFReq].onRequest(Packet::ReadReq); + + // on response to given request: specify new state + transitionTable[Invalid][Packet::ReadResp].onResponse(Shared); + transitionTable[Invalid][Packet::ReadExResp].onResponse(Modified); + transitionTable[Shared][writeToSharedResp].onResponse(Modified); + transitionTable[Owned][writeToSharedResp].onResponse(Modified); + + // bus snoop transition functions + transitionTable[Invalid][Packet::ReadReq].onSnoop(nullTransition); + transitionTable[Invalid][Packet::ReadExReq].onSnoop(nullTransition); + transitionTable[Invalid][Packet::UpgradeReq].onSnoop(nullTransition); + transitionTable[Shared][Packet::ReadReq].onSnoop(assertShared); + transitionTable[Shared][Packet::ReadExReq].onSnoop(invalidateTrans); + transitionTable[Shared][Packet::UpgradeReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::ReadExReq].onSnoop(supplyAndInvalidateTrans); + transitionTable[Modified][Packet::ReadReq].onSnoop(supplyAndGotoOwnedTrans); + transitionTable[Owned][Packet::ReadReq].onSnoop(supplyAndGotoOwnedTrans); + transitionTable[Owned][Packet::ReadExReq].onSnoop(supplyAndInvalidateTrans); + transitionTable[Owned][Packet::UpgradeReq].onSnoop(invalidateTrans); + //Tansitions on seeing a DMA (writeInv(samelevel) or DMAInv) + transitionTable[Invalid][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Shared][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Owned][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Invalid][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Shared][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Owned][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + } + + else if(protocol == "moesi") { + // incoming requests: specify outgoing bus request + transitionTable[Invalid][Packet::ReadReq].onRequest(Packet::ReadReq); + transitionTable[Invalid][Packet::WriteReq].onRequest(Packet::ReadExReq); + transitionTable[Shared][Packet::WriteReq].onRequest(writeToSharedCmd); + transitionTable[Owned][Packet::WriteReq].onRequest(writeToSharedCmd); + //Prefetching causes a read + transitionTable[Invalid][Packet::SoftPFReq].onRequest(Packet::ReadReq); + transitionTable[Invalid][Packet::HardPFReq].onRequest(Packet::ReadReq); + + // on response to given request: specify new state + transitionTable[Invalid][Packet::ReadResp].onResponse(Exclusive); + //It will move into shared if the shared line is asserted in the + //getNewState function + transitionTable[Invalid][Packet::ReadExResp].onResponse(Modified); + transitionTable[Shared][writeToSharedResp].onResponse(Modified); + transitionTable[Owned][writeToSharedResp].onResponse(Modified); + + // bus snoop transition functions + transitionTable[Invalid][Packet::ReadReq].onSnoop(nullTransition); + transitionTable[Invalid][Packet::ReadExReq].onSnoop(nullTransition); + transitionTable[Invalid][Packet::UpgradeReq].onSnoop(nullTransition); + transitionTable[Shared][Packet::ReadReq].onSnoop(assertShared); + transitionTable[Shared][Packet::ReadExReq].onSnoop(invalidateTrans); + transitionTable[Shared][Packet::UpgradeReq].onSnoop(invalidateTrans); + transitionTable[Exclusive][Packet::ReadReq].onSnoop(assertShared); + transitionTable[Exclusive][Packet::ReadExReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::ReadReq].onSnoop(supplyAndGotoOwnedTrans); + transitionTable[Modified][Packet::ReadExReq].onSnoop(supplyAndInvalidateTrans); + transitionTable[Owned][Packet::ReadReq].onSnoop(supplyAndGotoOwnedTrans); + transitionTable[Owned][Packet::ReadExReq].onSnoop(supplyAndInvalidateTrans); + transitionTable[Owned][Packet::UpgradeReq].onSnoop(invalidateTrans); + //Transitions on seeing a DMA (writeInv(samelevel) or DMAInv) + transitionTable[Invalid][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Shared][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Exclusive][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Owned][Packet::InvalidateReq].onSnoop(invalidateTrans); + transitionTable[Invalid][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Shared][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Exclusive][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Modified][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + transitionTable[Owned][Packet::WriteInvalidateReq].onSnoop(invalidateTrans); + } + + else { + cerr << "CoherenceProtocol: unrecognized protocol " << protocol + << endl; + fatal(""); + } +} + + +Packet::Command +CoherenceProtocol::getBusCmd(Packet::Command cmdIn, CacheBlk::State state, + MSHR *mshr) +{ + state &= stateMask; + int cmd_idx = (int) cmdIn; + + assert(0 <= state && state <= stateMax); + assert(0 <= cmd_idx && cmd_idx < NUM_MEM_CMDS); + + Packet::Command cmdOut = transitionTable[state][cmd_idx].busCmd; + + assert(cmdOut != Packet::InvalidCmd); + + ++requestCount[state][cmd_idx]; + + return cmdOut; +} + + +CacheBlk::State +CoherenceProtocol::getNewState(Packet * &pkt, CacheBlk::State oldState) +{ + CacheBlk::State state = oldState & stateMask; + int cmd_idx = pkt->cmdToIndex(); + + assert(0 <= state && state <= stateMax); + assert(0 <= cmd_idx && cmd_idx < NUM_MEM_CMDS); + + CacheBlk::State newState = transitionTable[state][cmd_idx].newState; + + //Check if it's exclusive and the shared line was asserted, + //then goto shared instead + if (newState == Exclusive && (pkt->flags & SHARED_LINE)) { + newState = Shared; + } + + assert(newState != -1); + + //Make sure not to loose any other state information + newState = (oldState & ~stateMask) | newState; + return newState; +} + + +bool +CoherenceProtocol::handleBusRequest(BaseCache *cache, Packet * &pkt, + CacheBlk *blk, + MSHR *mshr, + CacheBlk::State & new_state) +{ + if (blk == NULL) { + // nothing to do if we don't have a block + return false; + } + + CacheBlk::State state = blk->status & stateMask; + int cmd_idx = pkt->cmdToIndex(); + + assert(0 <= state && state <= stateMax); + assert(0 <= cmd_idx && cmd_idx < 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 + assert(!mshr || mshr->inService == 0); + ++snoopCount[state][cmd_idx]; + + bool ret = transitionTable[state][cmd_idx].snoopFunc(cache, pkt, blk, mshr, + new_state); + + + + return ret; +} + +bool +CoherenceProtocol::nullTransition(BaseCache *cache, Packet * &pkt, + CacheBlk *blk, MSHR *mshr, + CacheBlk::State & new_state) +{ + // do nothing + if (blk) + new_state = blk->status; + return false; +} + + +bool +CoherenceProtocol::invalidTransition(BaseCache *cache, Packet * &pkt, + CacheBlk *blk, MSHR *mshr, + CacheBlk::State & new_state) +{ + panic("Invalid transition"); + return false; +} + +#ifndef DOXYGEN_SHOULD_SKIP_THIS + +BEGIN_DECLARE_SIM_OBJECT_PARAMS(CoherenceProtocol) + + Param<string> protocol; + Param<bool> do_upgrades; + +END_DECLARE_SIM_OBJECT_PARAMS(CoherenceProtocol) + + +BEGIN_INIT_SIM_OBJECT_PARAMS(CoherenceProtocol) + + INIT_PARAM(protocol, "name of coherence protocol"), + INIT_PARAM_DFLT(do_upgrades, "use upgrade transactions?", true) + +END_INIT_SIM_OBJECT_PARAMS(CoherenceProtocol) + + +CREATE_SIM_OBJECT(CoherenceProtocol) +{ + return new CoherenceProtocol(getInstanceName(), protocol, + do_upgrades); +} + +REGISTER_SIM_OBJECT("CoherenceProtocol", CoherenceProtocol) + +#endif // DOXYGEN_SHOULD_SKIP_THIS diff --git a/src/mem/cache/coherence/coherence_protocol.hh b/src/mem/cache/coherence/coherence_protocol.hh new file mode 100644 index 000000000..21351ace4 --- /dev/null +++ b/src/mem/cache/coherence/coherence_protocol.hh @@ -0,0 +1,263 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + * Ron Dreslinski + * Steve Reinhardt + */ + +/** + * @file + * Declaration of CoherenceProcotol a basic coherence policy. + */ +#ifndef __COHERENCE_PROTOCOL_HH__ +#define __COHERENCE_PROTOCOL_HH__ + +#include <string> + +#include "sim/sim_object.hh" +#include "mem/packet.hh" +#include "mem/cache/cache_blk.hh" +#include "base/statistics.hh" + +class BaseCache; +class MSHR; + +/** + * A simple coherence policy for the memory hierarchy. Currently implements + * MSI, MESI, and MOESI protocols. + */ +class CoherenceProtocol : public SimObject +{ + public: + /** + * Contruct and initialize this policy. + * @param name The name of this policy. + * @param protocol The string representation of the protocol to use. + * @param doUpgrades True if bus upgrades should be used. + */ + CoherenceProtocol(const std::string &name, const std::string &protocol, + const bool doUpgrades); + + /** + * Destructor. + */ + virtual ~CoherenceProtocol() {}; + + /** + * Register statistics + */ + virtual void regStats(); + + /** + * Get the proper bus command for the given command and status. + * @param cmd The request's command. + * @param status The current state of the cache block. + * @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); + + /** + * Return the proper state given the current state and the bus response. + * @param req The bus response. + * @param oldState The current block state. + * @return The new state. + */ + CacheBlk::State getNewState(Packet * &pkt, + CacheBlk::State oldState); + + /** + * Handle snooped bus requests. + * @param cache The cache that snooped the request. + * @param req The snooped bus request. + * @param blk The cache block corresponding to the request, if any. + * @param mshr The MSHR corresponding to the request, if any. + * @param new_state The new coherence state of the block. + * @return True if the request should be satisfied locally. + */ + bool handleBusRequest(BaseCache *cache, Packet * &pkt, CacheBlk *blk, + MSHR *mshr, CacheBlk::State &new_state); + + protected: + /** Snoop function type. */ + typedef bool (*SnoopFuncType)(BaseCache *, Packet *&, CacheBlk *, + MSHR *, CacheBlk::State&); + + // + // Standard snoop transition functions + // + + /** + * Do nothing transition. + */ + static bool nullTransition(BaseCache *, Packet *&, CacheBlk *, + MSHR *, CacheBlk::State&); + + /** + * Invalid transition, basically panic. + */ + static bool invalidTransition(BaseCache *, Packet *&, CacheBlk *, + MSHR *, CacheBlk::State&); + + /** + * Invalidate block, move to Invalid state. + */ + static bool invalidateTrans(BaseCache *, Packet *&, CacheBlk *, + MSHR *, CacheBlk::State&); + + /** + * Supply data, no state transition. + */ + static bool supplyTrans(BaseCache *, Packet *&, CacheBlk *, + MSHR *, CacheBlk::State&); + + /** + * Supply data and go to Shared state. + */ + static bool supplyAndGotoSharedTrans(BaseCache *, Packet *&, CacheBlk *, + MSHR *, CacheBlk::State&); + + /** + * Supply data and go to Owned state. + */ + static bool supplyAndGotoOwnedTrans(BaseCache *, Packet *&, CacheBlk *, + MSHR *, CacheBlk::State&); + + /** + * Invalidate block, supply data, and go to Invalid state. + */ + static bool supplyAndInvalidateTrans(BaseCache *, Packet *&, CacheBlk *, + MSHR *, CacheBlk::State&); + + /** + * Assert the shared line for a block that is shared/exclusive. + */ + static bool assertShared(BaseCache *, Packet *&, CacheBlk *, + MSHR *, CacheBlk::State&); + + /** + * Definition of protocol state transitions. + */ + class StateTransition + { + friend class CoherenceProtocol; + + /** The bus command of this transition. */ + Packet::Command busCmd; + /** The state to transition to. */ + int newState; + /** The snoop function for this transition. */ + SnoopFuncType snoopFunc; + + /** + * Constructor, defaults to invalid transition. + */ + StateTransition(); + + /** + * Initialize bus command. + * @param cmd The bus command to use. + */ + void onRequest(Packet::Command cmd) + { + busCmd = cmd; + } + + /** + * Set the transition state. + * @param s The new state. + */ + void onResponse(CacheBlk::State s) + { + newState = s; + } + + /** + * Initialize the snoop function. + * @param f The new snoop function. + */ + void onSnoop(SnoopFuncType f) + { + snoopFunc = f; + } + }; + + friend class CoherenceProtocol::StateTransition; + + /** Mask to select status bits relevant to coherence protocol. */ + const static CacheBlk::State + stateMask = BlkValid | BlkWritable | BlkDirty; + + /** The Modified (M) state. */ + const static CacheBlk::State + Modified = BlkValid | BlkWritable | BlkDirty; + /** The Owned (O) state. */ + const static CacheBlk::State + Owned = BlkValid | BlkDirty; + /** The Exclusive (E) state. */ + const static CacheBlk::State + Exclusive = BlkValid | BlkWritable; + /** The Shared (S) state. */ + const static CacheBlk::State + Shared = BlkValid; + /** The Invalid (I) state. */ + const static CacheBlk::State + Invalid = 0; + + /** + * Maximum state encoding value (used to size transition lookup + * table). Could be more than number of states, depends on + * encoding of status bits. + */ + const static int stateMax = stateMask; + + /** + * The table of all possible transitions, organized by starting state and + * request command. + */ + StateTransition transitionTable[stateMax+1][NUM_MEM_CMDS]; + + /** + * @addtogroup CoherenceStatistics + * @{ + */ + /** + * State accesses from parent cache. + */ + Stats::Scalar<> requestCount[stateMax+1][NUM_MEM_CMDS]; + /** + * State accesses from snooped requests. + */ + Stats::Scalar<> snoopCount[stateMax+1][NUM_MEM_CMDS]; + /** + * @} + */ +}; + +#endif // __COHERENCE_PROTOCOL_HH__ diff --git a/src/mem/cache/coherence/simple_coherence.hh b/src/mem/cache/coherence/simple_coherence.hh new file mode 100644 index 000000000..ca9d18beb --- /dev/null +++ b/src/mem/cache/coherence/simple_coherence.hh @@ -0,0 +1,161 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + * Ron Dreslinski + */ + +/** + * @file + * Declaration of a simple coherence policy. + */ + +#ifndef __SIMPLE_COHERENCE_HH__ +#define __SIMPLE_COHERENCE_HH__ + +#include <string> + +#include "mem/packet.hh" +#include "mem/cache/cache_blk.hh" +#include "mem/cache/miss/mshr_queue.hh" +#include "mem/cache/coherence/coherence_protocol.hh" + +class BaseCache; + +/** + * A simple MP coherence policy. This policy assumes an atomic bus and only one + * level of cache. + */ +class SimpleCoherence +{ + protected: + /** Pointer to the parent cache. */ + BaseCache *cache; + /** Pointer to the coherence protocol. */ + CoherenceProtocol *protocol; + + public: + /** + * Construct and initialize this coherence policy. + * @param _protocol The coherence protocol to use. + */ + SimpleCoherence(CoherenceProtocol *_protocol) + : protocol(_protocol) + { + } + + /** + * Set the pointer to the parent cache. + * @param _cache The parent cache. + */ + void setCache(BaseCache *_cache) + { + cache = _cache; + } + + /** + * Register statistics. + * @param name The name to prepend to stat descriptions. + */ + void regStats(const std::string &name) + { + } + + /** + * This policy does not forward invalidates, return NULL. + * @return NULL. + */ + Packet * getPacket() + { + return NULL; + } + + /** + * Return the proper state given the current state and the bus response. + * @param req The bus response. + * @param current The current block state. + * @return The new state. + */ + CacheBlk::State getNewState(Packet * &pkt, CacheBlk::State current) + { + return protocol->getNewState(pkt, current); + } + + /** + * Handle snooped bus requests. + * @param req The snooped bus request. + * @param blk The cache block corresponding to the request, if any. + * @param mshr The MSHR corresponding to the request, if any. + * @param new_state Return the new state for the block. + */ + bool handleBusRequest(Packet * &pkt, CacheBlk *blk, MSHR *mshr, + CacheBlk::State &new_state) + { +// assert(mshr == NULL); +//Got rid of, there could be an MSHR, but it can't be in service + if (blk != NULL) + { + if (pkt->cmd != Packet::Writeback) { + return protocol->handleBusRequest(cache, pkt, blk, mshr, + new_state); + } + else { //It is a writeback, must be ownership protocol, just keep state + new_state = blk->status; + } + } + return false; + } + + /** + * Get the proper bus command for the given command and status. + * @param cmd The request's command. + * @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) + { + if (cmd == Packet::Writeback) return Packet::Writeback; + return protocol->getBusCmd(cmd, state); + } + + /** + * Return true if this coherence policy can handle fast cache writes. + */ + bool allowFastWrites() { return false; } + + bool hasProtocol() { return true; } +}; + +#endif //__SIMPLE_COHERENCE_HH__ + + + + + + + + diff --git a/src/mem/cache/coherence/uni_coherence.cc b/src/mem/cache/coherence/uni_coherence.cc new file mode 100644 index 000000000..5ab706269 --- /dev/null +++ b/src/mem/cache/coherence/uni_coherence.cc @@ -0,0 +1,86 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +#include "mem/cache/coherence/uni_coherence.hh" +#include "mem/cache/base_cache.hh" + +#include "base/trace.hh" + +using namespace std; + +UniCoherence::UniCoherence() + : cshrs(50) +{ +} + +Packet * +UniCoherence::getPacket() +{ + bool unblock = cshrs.isFull(); + Packet* pkt = cshrs.getReq(); + cshrs.markInService((MSHR*)pkt->senderState); + if (!cshrs.havePending()) { + cache->clearSlaveRequest(Request_Coherence); + } + if (unblock) { + //since CSHRs are always used as buffers, should always get rid of one + assert(!cshrs.isFull()); + cache->clearBlocked(Blocked_Coherence); + } + return pkt; +} + +/** + * @todo add support for returning slave requests, not doing them here. + */ +bool +UniCoherence::handleBusRequest(Packet * &pkt, CacheBlk *blk, MSHR *mshr, + CacheBlk::State &new_state) +{ + new_state = 0; + if (pkt->isInvalidate()) { + DPRINTF(Cache, "snoop inval on blk %x (blk ptr %x)\n", + pkt->getAddr(), blk); + if (!cache->isTopLevel()) { + // Forward to other caches + Packet * tmp = new Packet(pkt->req, Packet::InvalidateReq, -1); + cshrs.allocate(tmp); + cache->setSlaveRequest(Request_Coherence, curTick); + if (cshrs.isFull()) { + cache->setBlockedForSnoop(Blocked_Coherence); + } + } + } else { + if (blk) { + new_state = blk->status; + } + } + return false; +} diff --git a/src/mem/cache/coherence/uni_coherence.hh b/src/mem/cache/coherence/uni_coherence.hh new file mode 100644 index 000000000..764bf6276 --- /dev/null +++ b/src/mem/cache/coherence/uni_coherence.hh @@ -0,0 +1,136 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +#ifndef __UNI_COHERENCE_HH__ +#define __UNI_COHERENCE_HH__ + +#include "base/trace.hh" +#include "base/misc.hh" +#include "mem/cache/cache_blk.hh" +#include "mem/cache/miss/mshr_queue.hh" +#include "mem/packet.hh" + +class BaseCache; + +class UniCoherence +{ + protected: + /** Buffers to hold forwarded invalidates. */ + MSHRQueue cshrs; + /** Pointer to the parent cache. */ + BaseCache *cache; + + public: + /** + * Construct and initialize this coherence policy. + */ + UniCoherence(); + + /** + * Set the pointer to the parent cache. + * @param _cache The parent cache. + */ + void setCache(BaseCache *_cache) + { + cache = _cache; + } + + /** + * Register statistics. + * @param name The name to prepend to stat descriptions. + */ + void regStats(const std::string &name) + { + } + + /** + * Return Read. + * @param cmd The request's command. + * @param state The current state of the cache block. + * @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) + { + if (cmd == Packet::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; + } + + /** + * Just return readable and writeable. + * @param req The bus response. + * @param current The current block state. + * @return The new state. + */ + CacheBlk::State getNewState(Packet * &pkt, CacheBlk::State current) + { + if (pkt->senderState) //Blocking Buffers don't get mshrs + { + if (((MSHR *)(pkt->senderState))->originalCmd == Packet::HardPFReq) { + DPRINTF(HWPrefetch, "Marking a hardware prefetch as such in the state\n"); + return BlkHWPrefetched | BlkValid | BlkWritable; + } + else { + return BlkValid | BlkWritable; + } + } + //@todo What about prefetching with blocking buffers + else + return BlkValid | BlkWritable; + } + /** + * Return outstanding invalidate to forward. + * @return The next invalidate to forward to lower levels of cache. + */ + Packet * getPacket(); + + /** + * Handle snooped bus requests. + * @param req The snooped bus request. + * @param blk The cache block corresponding to the request, if any. + * @param mshr The MSHR corresponding to the request, if any. + * @param new_state The new coherence state of the block. + * @return True if the request should be satisfied locally. + */ + bool handleBusRequest(Packet * &pkt, CacheBlk *blk, MSHR *mshr, + CacheBlk::State &new_state); + + /** + * Return true if this coherence policy can handle fast cache writes. + */ + bool allowFastWrites() { return true; } + + bool hasProtocol() { return false; } +}; + +#endif //__UNI_COHERENCE_HH__ diff --git a/src/mem/cache/miss/blocking_buffer.cc b/src/mem/cache/miss/blocking_buffer.cc new file mode 100644 index 000000000..d745cb8c6 --- /dev/null +++ b/src/mem/cache/miss/blocking_buffer.cc @@ -0,0 +1,246 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** + * @file + * Definitions of a simple buffer for a blocking cache. + */ + +#include "cpu/smt.hh" //for maxThreadsPerCPU +#include "mem/cache/base_cache.hh" +#include "mem/cache/miss/blocking_buffer.hh" +#include "mem/cache/prefetch/base_prefetcher.hh" +#include "sim/eventq.hh" // for Event declaration. +#include "mem/request.hh" + +using namespace TheISA; + +/** + * @todo Move writebacks into shared BaseBuffer class. + */ +void +BlockingBuffer::regStats(const std::string &name) +{ + using namespace Stats; + writebacks + .init(maxThreadsPerCPU) + .name(name + ".writebacks") + .desc("number of writebacks") + .flags(total) + ; +} + +void +BlockingBuffer::setCache(BaseCache *_cache) +{ + cache = _cache; + blkSize = cache->getBlockSize(); +} + +void +BlockingBuffer::setPrefetcher(BasePrefetcher *_prefetcher) +{ + prefetcher = _prefetcher; +} +void +BlockingBuffer::handleMiss(Packet * &pkt, int blk_size, Tick time) +{ + Addr blk_addr = pkt->getAddr() & ~(Addr)(blk_size - 1); + if (pkt->isWrite() && (pkt->req->isUncacheable() || !writeAllocate || + !pkt->needsResponse())) { + if (!pkt->needsResponse()) { + wb.allocateAsBuffer(pkt); + } else { + wb.allocate(pkt->cmd, blk_addr, pkt->req->getAsid(), blk_size, pkt); + } + + memcpy(wb.pkt->getPtr<uint8_t>(), pkt->getPtr<uint8_t>(), blk_size); + + cache->setBlocked(Blocked_NoWBBuffers); + cache->setMasterRequest(Request_WB, time); + return; + } + + if (!pkt->needsResponse()) { + miss.allocateAsBuffer(pkt); + } else { + miss.allocate(pkt->cmd, blk_addr, pkt->req->getAsid(), blk_size, pkt); + } + if (!pkt->req->isUncacheable()) { + miss.pkt->flags |= CACHE_LINE_FILL; + } + cache->setBlocked(Blocked_NoMSHRs); + cache->setMasterRequest(Request_MSHR, time); +} + +Packet * +BlockingBuffer::getPacket() +{ + if (miss.pkt && !miss.inService) { + return miss.pkt; + } + return wb.pkt; +} + +void +BlockingBuffer::setBusCmd(Packet * &pkt, Packet::Command cmd) +{ + MSHR *mshr = (MSHR*) pkt->senderState; + mshr->originalCmd = pkt->cmd; + if (pkt->isCacheFill()) + pkt->cmdOverride(cmd); +} + +void +BlockingBuffer::restoreOrigCmd(Packet * &pkt) +{ + pkt->cmdOverride(((MSHR*)(pkt->senderState))->originalCmd); +} + +void +BlockingBuffer::markInService(Packet * &pkt) +{ + if (!pkt->isCacheFill() && pkt->isWrite()) { + // Forwarding a write/ writeback, don't need to change + // the command + assert((MSHR*)pkt->senderState == &wb); + cache->clearMasterRequest(Request_WB); + if (!pkt->needsResponse()) { + assert(wb.getNumTargets() == 0); + wb.deallocate(); + cache->clearBlocked(Blocked_NoWBBuffers); + } else { + wb.inService = true; + } + } else { + assert((MSHR*)pkt->senderState == &miss); + cache->clearMasterRequest(Request_MSHR); + if (!pkt->needsResponse()) { + assert(miss.getNumTargets() == 0); + miss.deallocate(); + cache->clearBlocked(Blocked_NoMSHRs); + } else { + //mark in service + miss.inService = true; + } + } +} + +void +BlockingBuffer::handleResponse(Packet * &pkt, Tick time) +{ + if (pkt->isCacheFill()) { + // targets were handled in the cache tags + assert((MSHR*)pkt->senderState == &miss); + miss.deallocate(); + cache->clearBlocked(Blocked_NoMSHRs); + } else { + if (((MSHR*)(pkt->senderState))->hasTargets()) { + // Should only have 1 target if we had any + assert(((MSHR*)(pkt->senderState))->getNumTargets() == 1); + Packet * target = ((MSHR*)(pkt->senderState))->getTarget(); + ((MSHR*)(pkt->senderState))->popTarget(); + if (pkt->isRead()) { + memcpy(target->getPtr<uint8_t>(), pkt->getPtr<uint8_t>(), target->getSize()); + } + cache->respond(target, time); + assert(!((MSHR*)(pkt->senderState))->hasTargets()); + } + + if (pkt->isWrite()) { + assert(((MSHR*)(pkt->senderState)) == &wb); + wb.deallocate(); + cache->clearBlocked(Blocked_NoWBBuffers); + } else { + miss.deallocate(); + cache->clearBlocked(Blocked_NoMSHRs); + } + } +} + +void +BlockingBuffer::squash(int threadNum) +{ + if (miss.threadNum == threadNum) { + Packet * target = miss.getTarget(); + miss.popTarget(); + assert(target->req->getThreadNum() == threadNum); + target = NULL; + assert(!miss.hasTargets()); + miss.ntargets=0; + if (!miss.inService) { + miss.deallocate(); + cache->clearBlocked(Blocked_NoMSHRs); + cache->clearMasterRequest(Request_MSHR); + } + } +} + +void +BlockingBuffer::doWriteback(Addr addr, int asid, + int size, uint8_t *data, bool compressed) +{ + // Generate request + Request * req = new Request(addr, size, 0); + Packet * pkt = new Packet(req, Packet::Writeback, -1); + uint8_t *new_data = new uint8_t[size]; + pkt->dataDynamicArray<uint8_t>(new_data); + if (data) { + memcpy(pkt->getPtr<uint8_t>(), data, size); + } + + if (compressed) { + pkt->flags |= COMPRESSED; + } + + ///All writebacks charged to same thread @todo figure this out + writebacks[pkt->req->getThreadNum()]++; + + wb.allocateAsBuffer(pkt); + cache->setMasterRequest(Request_WB, curTick); + cache->setBlocked(Blocked_NoWBBuffers); +} + + + +void +BlockingBuffer::doWriteback(Packet * &pkt) +{ + writebacks[pkt->req->getThreadNum()]++; + + wb.allocateAsBuffer(pkt); + + // Since allocate as buffer copies the request, + // need to copy data here. + memcpy(wb.pkt->getPtr<uint8_t>(), pkt->getPtr<uint8_t>(), pkt->getSize()); + + cache->setBlocked(Blocked_NoWBBuffers); + cache->setMasterRequest(Request_WB, curTick); +} diff --git a/src/mem/cache/miss/blocking_buffer.hh b/src/mem/cache/miss/blocking_buffer.hh new file mode 100644 index 000000000..39a06a377 --- /dev/null +++ b/src/mem/cache/miss/blocking_buffer.hh @@ -0,0 +1,256 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** + * @file + * Declaration of a simple buffer for a blocking cache. + */ + +#ifndef __BLOCKING_BUFFER_HH__ +#define __BLOCKING_BUFFER_HH__ + +#include <vector> + +#include "mem/cache/miss/mshr.hh" +#include "base/statistics.hh" + +class BaseCache; +class BasePrefetcher; + +/** + * Miss and writeback storage for a blocking cache. + */ +class BlockingBuffer +{ +protected: + /** Miss storage. */ + MSHR miss; + /** WB storage. */ + MSHR wb; + + //Params + + /** Allocate on write misses. */ + const bool writeAllocate; + + /** Pointer to the parent cache. */ + BaseCache* cache; + + BasePrefetcher* prefetcher; + + /** Block size of the parent cache. */ + int blkSize; + + // Statistics + /** + * @addtogroup CacheStatistics + * @{ + */ + /** Number of blocks written back per thread. */ + Stats::Vector<> writebacks; + + /** + * @} + */ + +public: + /** + * Builds and initializes this buffer. + * @param write_allocate If true, treat write misses the same as reads. + */ + BlockingBuffer(bool write_allocate) + : writeAllocate(write_allocate) + { + } + + /** + * Register statistics for this object. + * @param name The name of the parent cache. + */ + void regStats(const std::string &name); + + /** + * Called by the parent cache to set the back pointer. + * @param _cache A pointer to the parent cache. + */ + void setCache(BaseCache *_cache); + + void setPrefetcher(BasePrefetcher *_prefetcher); + + /** + * Handle a cache miss properly. Requests the bus and marks the cache as + * blocked. + * @param req The request that missed in the cache. + * @param blk_size The block size of the cache. + * @param time The time the miss is detected. + */ + void handleMiss(Packet * &pkt, int blk_size, Tick time); + + /** + * Fetch the block for the given address and buffer the given target. + * @param addr The address to fetch. + * @param asid The address space of the address. + * @param blk_size The block size of the cache. + * @param time The time the miss is detected. + * @param target The target for the fetch. + */ + MSHR* fetchBlock(Addr addr, int asid, int blk_size, Tick time, + Packet * &target) + { + fatal("Unimplemented"); + } + + /** + * Selects a outstanding request to service. + * @return The request to service, NULL if none found. + */ + Packet * getPacket(); + + /** + * Set the command to the given bus command. + * @param req The request to update. + * @param cmd The bus command to use. + */ + void setBusCmd(Packet * &pkt, Packet::Command cmd); + + /** + * Restore the original command in case of a bus transmission error. + * @param req The request to reset. + */ + void restoreOrigCmd(Packet * &pkt); + + /** + * Marks a request as in service (sent on the bus). This can have side + * effect since storage for no response commands is deallocated once they + * are successfully sent. + * @param req The request that was sent on the bus. + */ + void markInService(Packet * &pkt); + + /** + * Frees the resources of the request and unblock the cache. + * @param req The request that has been satisfied. + * @param time The time when the request is satisfied. + */ + void handleResponse(Packet * &pkt, Tick time); + + /** + * Removes all outstanding requests for a given thread number. If a request + * has been sent to the bus, this function removes all of its targets. + * @param req->getThreadNum()ber The thread number of the requests to squash. + */ + void squash(int threadNum); + + /** + * Return the current number of outstanding misses. + * @return the number of outstanding misses. + */ + int getMisses() + { + return miss.getNumTargets(); + } + + /** + * Searches for the supplied address in the miss "queue". + * @param addr The address to look for. + * @param asid The address space id. + * @return A pointer to miss if it matches. + */ + MSHR* findMSHR(Addr addr, int asid) + { + if (miss.addr == addr && miss.pkt) + return &miss; + return NULL; + } + + /** + * Searches for the supplied address in the write buffer. + * @param addr The address to look for. + * @param asid The address space id. + * @param writes List of pointers to the matching writes. + * @return True if there is a matching write. + */ + bool findWrites(Addr addr, int asid, std::vector<MSHR*>& writes) + { + if (wb.addr == addr && wb.pkt) { + writes.push_back(&wb); + return true; + } + return false; + } + + + + /** + * Perform a writeback of dirty data to the given address. + * @param addr The address to write to. + * @param asid The address space id. + * @param size The number of bytes to write. + * @param data The data to write, can be NULL. + * @param compressed True if the data is compressed. + */ + void doWriteback(Addr addr, int asid, + int size, uint8_t *data, bool compressed); + + /** + * Perform a writeback request. + * @param req The writeback request. + */ + void doWriteback(Packet * &pkt); + + /** + * Returns true if there are outstanding requests. + * @return True if there are outstanding requests. + */ + bool havePending() + { + return !miss.inService || !wb.inService; + } + + /** + * Add a target to the given MSHR. This assumes it is in the miss queue. + * @param mshr The mshr to add a target to. + * @param req The target to add. + */ + void addTarget(MSHR *mshr, Packet * &pkt) + { + fatal("Shouldn't call this on a blocking buffer."); + } + + /** + * Dummy implmentation. + */ + MSHR* allocateTargetList(Addr addr, int asid) + { + fatal("Unimplemented"); + } +}; + +#endif // __BLOCKING_BUFFER_HH__ diff --git a/src/mem/cache/miss/miss_queue.cc b/src/mem/cache/miss/miss_queue.cc new file mode 100644 index 000000000..34290351d --- /dev/null +++ b/src/mem/cache/miss/miss_queue.cc @@ -0,0 +1,757 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + * Ron Dreslinski + */ + +/** + * @file + * Miss and writeback queue definitions. + */ + +#include "cpu/smt.hh" //for maxThreadsPerCPU +#include "mem/cache/base_cache.hh" +#include "mem/cache/miss/miss_queue.hh" +#include "mem/cache/prefetch/base_prefetcher.hh" + +using namespace std; + +// simple constructor +/** + * @todo Remove the +16 from the write buffer constructor once we handle + * stalling on writebacks do to compression writes. + */ +MissQueue::MissQueue(int numMSHRs, int numTargets, int write_buffers, + bool write_allocate, bool prefetch_miss) + : mq(numMSHRs, 4), wb(write_buffers,numMSHRs+1000), numMSHR(numMSHRs), + numTarget(numTargets), writeBuffers(write_buffers), + writeAllocate(write_allocate), order(0), prefetchMiss(prefetch_miss) +{ + noTargetMSHR = NULL; +} + +void +MissQueue::regStats(const string &name) +{ + Request temp_req; + Packet::Command temp_cmd = Packet::ReadReq; + Packet temp_pkt(&temp_req, temp_cmd, 0); //@todo FIx command strings so this isn't neccessary + + using namespace Stats; + + writebacks + .init(maxThreadsPerCPU) + .name(name + ".writebacks") + .desc("number of writebacks") + .flags(total) + ; + + // 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); + + mshr_hits[access_idx] + .init(maxThreadsPerCPU) + .name(name + "." + cstr + "_mshr_hits") + .desc("number of " + cstr + " MSHR hits") + .flags(total | nozero | nonan) + ; + } + + demandMshrHits + .name(name + ".demand_mshr_hits") + .desc("number of demand (read+write) MSHR hits") + .flags(total) + ; + demandMshrHits = mshr_hits[Packet::ReadReq] + mshr_hits[Packet::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]; + + // 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); + + mshr_misses[access_idx] + .init(maxThreadsPerCPU) + .name(name + "." + cstr + "_mshr_misses") + .desc("number of " + cstr + " MSHR misses") + .flags(total | nozero | nonan) + ; + } + + demandMshrMisses + .name(name + ".demand_mshr_misses") + .desc("number of demand (read+write) MSHR misses") + .flags(total) + ; + demandMshrMisses = mshr_misses[Packet::ReadReq] + mshr_misses[Packet::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]; + + // 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); + + mshr_miss_latency[access_idx] + .init(maxThreadsPerCPU) + .name(name + "." + cstr + "_mshr_miss_latency") + .desc("number of " + cstr + " MSHR miss cycles") + .flags(total | nozero | nonan) + ; + } + + demandMshrMissLatency + .name(name + ".demand_mshr_miss_latency") + .desc("number of demand (read+write) MSHR miss cycles") + .flags(total) + ; + demandMshrMissLatency = mshr_miss_latency[Packet::ReadReq] + + mshr_miss_latency[Packet::WriteReq]; + + overallMshrMissLatency + .name(name + ".overall_mshr_miss_latency") + .desc("number of overall MSHR miss cycles") + .flags(total) + ; + overallMshrMissLatency = demandMshrMissLatency + + mshr_miss_latency[Packet::SoftPFReq] + mshr_miss_latency[Packet::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); + + mshr_uncacheable[access_idx] + .init(maxThreadsPerCPU) + .name(name + "." + cstr + "_mshr_uncacheable") + .desc("number of " + cstr + " MSHR uncacheable") + .flags(total | nozero | nonan) + ; + } + + overallMshrUncacheable + .name(name + ".overall_mshr_uncacheable_misses") + .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]; + + // 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); + + mshr_uncacheable_lat[access_idx] + .init(maxThreadsPerCPU) + .name(name + "." + cstr + "_mshr_uncacheable_latency") + .desc("number of " + cstr + " MSHR uncacheable cycles") + .flags(total | nozero | nonan) + ; + } + + overallMshrUncacheableLatency + .name(name + ".overall_mshr_uncacheable_latency") + .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]; + +#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); + + mshrAccesses[access_idx] + .name(name + "." + cstr + "_mshr_accesses") + .desc("number of " + cstr + " mshr accesses(hits+misses)") + .flags(total | nozero | nonan) + ; + mshrAccesses[access_idx] = + mshr_hits[access_idx] + mshr_misses[access_idx] + + mshr_uncacheable[access_idx]; + } + + demandMshrAccesses + .name(name + ".demand_mshr_accesses") + .desc("number of demand (read+write) mshr accesses") + .flags(total | nozero | nonan) + ; + demandMshrAccesses = demandMshrHits + demandMshrMisses; + + overallMshrAccesses + .name(name + ".overall_mshr_accesses") + .desc("number of overall (read+write) mshr accesses") + .flags(total | nozero | nonan) + ; + overallMshrAccesses = overallMshrHits + overallMshrMisses + + overallMshrUncacheable; +#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); + + mshrMissRate[access_idx] + .name(name + "." + cstr + "_mshr_miss_rate") + .desc("mshr miss rate for " + cstr + " accesses") + .flags(total | nozero | nonan) + ; + + mshrMissRate[access_idx] = + mshr_misses[access_idx] / cache->accesses[access_idx]; + } + + demandMshrMissRate + .name(name + ".demand_mshr_miss_rate") + .desc("mshr miss rate for demand accesses") + .flags(total) + ; + demandMshrMissRate = demandMshrMisses / cache->demandAccesses; + + overallMshrMissRate + .name(name + ".overall_mshr_miss_rate") + .desc("mshr miss rate for overall accesses") + .flags(total) + ; + 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); + + avgMshrMissLatency[access_idx] + .name(name + "." + cstr + "_avg_mshr_miss_latency") + .desc("average " + cstr + " mshr miss latency") + .flags(total | nozero | nonan) + ; + + avgMshrMissLatency[access_idx] = + mshr_miss_latency[access_idx] / mshr_misses[access_idx]; + } + + demandAvgMshrMissLatency + .name(name + ".demand_avg_mshr_miss_latency") + .desc("average overall mshr miss latency") + .flags(total) + ; + demandAvgMshrMissLatency = demandMshrMissLatency / demandMshrMisses; + + overallAvgMshrMissLatency + .name(name + ".overall_avg_mshr_miss_latency") + .desc("average overall mshr miss latency") + .flags(total) + ; + 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); + + avgMshrUncacheableLatency[access_idx] + .name(name + "." + cstr + "_avg_mshr_uncacheable_latency") + .desc("average " + cstr + " mshr uncacheable latency") + .flags(total | nozero | nonan) + ; + + avgMshrUncacheableLatency[access_idx] = + mshr_uncacheable_lat[access_idx] / mshr_uncacheable[access_idx]; + } + + overallAvgMshrUncacheableLatency + .name(name + ".overall_avg_mshr_uncacheable_latency") + .desc("average overall mshr uncacheable latency") + .flags(total) + ; + overallAvgMshrUncacheableLatency = overallMshrUncacheableLatency / overallMshrUncacheable; + + mshr_cap_events + .init(maxThreadsPerCPU) + .name(name + ".mshr_cap_events") + .desc("number of times MSHR cap was activated") + .flags(total) + ; + + //software prefetching stats + soft_prefetch_mshr_full + .init(maxThreadsPerCPU) + .name(name + ".soft_prefetch_mshr_full") + .desc("number of mshr full events for SW prefetching instrutions") + .flags(total) + ; + + mshr_no_allocate_misses + .name(name +".no_allocate_misses") + .desc("Number of misses that were no-allocate") + ; + +} + +void +MissQueue::setCache(BaseCache *_cache) +{ + cache = _cache; + blkSize = cache->getBlockSize(); +} + +void +MissQueue::setPrefetcher(BasePrefetcher *_prefetcher) +{ + prefetcher = _prefetcher; +} + +MSHR* +MissQueue::allocateMiss(Packet * &pkt, int size, Tick time) +{ + MSHR* mshr = mq.allocate(pkt, size); + mshr->order = order++; + if (!pkt->req->isUncacheable() ){//&& !pkt->isNoAllocate()) { + // Mark this as a cache line fill + mshr->pkt->flags |= CACHE_LINE_FILL; + } + if (mq.isFull()) { + cache->setBlocked(Blocked_NoMSHRs); + } + if (pkt->cmd != Packet::HardPFReq) { + //If we need to request the bus (not on HW prefetch), do so + cache->setMasterRequest(Request_MSHR, time); + } + return mshr; +} + + +MSHR* +MissQueue::allocateWrite(Packet * &pkt, int size, Tick time) +{ + MSHR* mshr = wb.allocate(pkt,pkt->getSize()); + mshr->order = order++; + +//REMOVING COMPRESSION FOR NOW +#if 0 + if (pkt->isCompressed()) { + mshr->pkt->deleteData(); + mshr->pkt->actualSize = pkt->actualSize; + mshr->pkt->data = new uint8_t[pkt->actualSize]; + memcpy(mshr->pkt->data, pkt->data, pkt->actualSize); + } else { +#endif + memcpy(mshr->pkt->getPtr<uint8_t>(), pkt->getPtr<uint8_t>(), pkt->getSize()); + //{ + + if (wb.isFull()) { + cache->setBlocked(Blocked_NoWBBuffers); + } + + cache->setMasterRequest(Request_WB, time); + + return mshr; +} + + +/** + * @todo Remove SW prefetches on mshr hits. + */ +void +MissQueue::handleMiss(Packet * &pkt, int blkSize, Tick time) +{ +// if (!cache->isTopLevel()) + if (prefetchMiss) prefetcher->handleMiss(pkt, time); + + int size = blkSize; + Addr blkAddr = pkt->getAddr() & ~(Addr)(blkSize-1); + MSHR* mshr = NULL; + if (!pkt->req->isUncacheable()) { + mshr = mq.findMatch(blkAddr, pkt->req->getAsid()); + if (mshr) { + //@todo remove hw_pf here + mshr_hits[pkt->cmdToIndex()][pkt->req->getThreadNum()]++; + if (mshr->threadNum != pkt->req->getThreadNum()) { + mshr->threadNum = -1; + } + mq.allocateTarget(mshr, pkt); + if (mshr->pkt->isNoAllocate() && !pkt->isNoAllocate()) { + //We are adding an allocate after a no-allocate + mshr->pkt->flags &= ~NO_ALLOCATE; + } + if (mshr->getNumTargets() == numTarget) { + noTargetMSHR = mshr; + cache->setBlocked(Blocked_NoTargets); + mq.moveToFront(mshr); + } + return; + } + if (pkt->isNoAllocate()) { + //Count no-allocate requests differently + mshr_no_allocate_misses++; + } + else { + mshr_misses[pkt->cmdToIndex()][pkt->req->getThreadNum()]++; + } + } else { + //Count uncacheable accesses + mshr_uncacheable[pkt->cmdToIndex()][pkt->req->getThreadNum()]++; + size = pkt->getSize(); + } + if (pkt->isWrite() && (pkt->req->isUncacheable() || !writeAllocate || + !pkt->needsResponse())) { + /** + * @todo Add write merging here. + */ + mshr = allocateWrite(pkt, pkt->getSize(), time); + return; + } + + mshr = allocateMiss(pkt, size, time); +} + +MSHR* +MissQueue::fetchBlock(Addr addr, int asid, int blk_size, Tick time, + Packet * &target) +{ + Addr blkAddr = addr & ~(Addr)(blk_size - 1); + assert(mq.findMatch(addr, asid) == NULL); + MSHR *mshr = mq.allocateFetch(blkAddr, asid, blk_size, target); + mshr->order = order++; + mshr->pkt->flags |= CACHE_LINE_FILL; + if (mq.isFull()) { + cache->setBlocked(Blocked_NoMSHRs); + } + cache->setMasterRequest(Request_MSHR, time); + return mshr; +} + +Packet * +MissQueue::getPacket() +{ + Packet * pkt = mq.getReq(); + if (((wb.isFull() && wb.inServiceMSHRs == 0) || !pkt || + pkt->time > curTick) && wb.havePending()) { + pkt = wb.getReq(); + // Need to search for earlier miss. + MSHR *mshr = mq.findPending(pkt); + if (mshr && mshr->order < ((MSHR*)(pkt->senderState))->order) { + // Service misses in order until conflict is cleared. + return mq.getReq(); + } + } + if (pkt) { + MSHR* mshr = wb.findPending(pkt); + if (mshr /*&& mshr->order < pkt->senderState->order*/) { + // The only way this happens is if we are + // doing a write and we didn't have permissions + // then subsequently saw a writeback(owned got evicted) + // We need to make sure to perform the writeback first + // To preserve the dirty data, then we can issue the write + return wb.getReq(); + } + } + else if (!mq.isFull()){ + //If we have a miss queue slot, we can try a prefetch + pkt = prefetcher->getPacket(); + if (pkt) { + //Update statistic on number of prefetches issued (hwpf_mshr_misses) + mshr_misses[pkt->cmdToIndex()][pkt->req->getThreadNum()]++; + //It will request the bus for the future, but should clear that immedieatley + allocateMiss(pkt, pkt->getSize(), curTick); + pkt = mq.getReq(); + assert(pkt); //We should get back a req b/c we just put one in + } + } + return pkt; +} + +void +MissQueue::setBusCmd(Packet * &pkt, Packet::Command cmd) +{ + assert(pkt->senderState != 0); + MSHR * mshr = (MSHR*)pkt->senderState; + mshr->originalCmd = pkt->cmd; + if (pkt->isCacheFill() || pkt->isNoAllocate()) + pkt->cmd = cmd; +} + +void +MissQueue::restoreOrigCmd(Packet * &pkt) +{ + pkt->cmd = ((MSHR*)(pkt->senderState))->originalCmd; +} + +void +MissQueue::markInService(Packet * &pkt) +{ + assert(pkt->senderState != 0); + bool unblock = false; + BlockedCause cause = NUM_BLOCKED_CAUSES; + + /** + * @todo Should include MSHRQueue pointer in MSHR to select the correct + * one. + */ + if ((!pkt->isCacheFill() && pkt->isWrite())) { + // Forwarding a write/ writeback, don't need to change + // the command + unblock = wb.isFull(); + wb.markInService((MSHR*)pkt->senderState); + if (!wb.havePending()){ + cache->clearMasterRequest(Request_WB); + } + if (unblock) { + // Do we really unblock? + unblock = !wb.isFull(); + cause = Blocked_NoWBBuffers; + } + } else { + unblock = mq.isFull(); + mq.markInService((MSHR*)pkt->senderState); + if (!mq.havePending()){ + cache->clearMasterRequest(Request_MSHR); + } + if (((MSHR*)(pkt->senderState))->originalCmd == Packet::HardPFReq) { + DPRINTF(HWPrefetch, "%s:Marking a HW_PF in service\n", + cache->name()); + //Also clear pending if need be + if (!prefetcher->havePending()) + { + cache->clearMasterRequest(Request_PF); + } + } + if (unblock) { + unblock = !mq.isFull(); + cause = Blocked_NoMSHRs; + } + } + if (unblock) { + cache->clearBlocked(cause); + } +} + + +void +MissQueue::handleResponse(Packet * &pkt, Tick time) +{ + MSHR* mshr = (MSHR*)pkt->senderState; + if (((MSHR*)(pkt->senderState))->originalCmd == Packet::HardPFReq) { + DPRINTF(HWPrefetch, "%s:Handling the response to a HW_PF\n", + cache->name()); + } +#ifndef NDEBUG + int num_targets = mshr->getNumTargets(); +#endif + + bool unblock = false; + bool unblock_target = false; + BlockedCause cause = NUM_BLOCKED_CAUSES; + + if (pkt->isCacheFill() && !pkt->isNoAllocate()) { + mshr_miss_latency[mshr->originalCmd][pkt->req->getThreadNum()] += + curTick - pkt->time; + // targets were handled in the cache tags + if (mshr == noTargetMSHR) { + // we always clear at least one target + unblock_target = true; + cause = Blocked_NoTargets; + noTargetMSHR = NULL; + } + + if (mshr->hasTargets()) { + // Didn't satisfy all the targets, need to resend + Packet::Command cmd = mshr->getTarget()->cmd; + mq.markPending(mshr, cmd); + mshr->order = order++; + cache->setMasterRequest(Request_MSHR, time); + } + else { + unblock = mq.isFull(); + mq.deallocate(mshr); + if (unblock) { + unblock = !mq.isFull(); + cause = Blocked_NoMSHRs; + } + } + } else { + if (pkt->req->isUncacheable()) { + mshr_uncacheable_lat[pkt->cmd][pkt->req->getThreadNum()] += + curTick - pkt->time; + } + if (mshr->hasTargets() && pkt->req->isUncacheable()) { + // Should only have 1 target if we had any + assert(num_targets == 1); + Packet * target = mshr->getTarget(); + mshr->popTarget(); + if (pkt->isRead()) { + memcpy(target->getPtr<uint8_t>(), pkt->getPtr<uint8_t>(), + target->getSize()); + } + cache->respond(target, time); + assert(!mshr->hasTargets()); + } + else if (mshr->hasTargets()) { + //Must be a no_allocate with possibly more than one target + assert(mshr->pkt->isNoAllocate()); + while (mshr->hasTargets()) { + Packet * target = mshr->getTarget(); + mshr->popTarget(); + if (pkt->isRead()) { + memcpy(target->getPtr<uint8_t>(), pkt->getPtr<uint8_t>(), + target->getSize()); + } + cache->respond(target, time); + } + } + + if (pkt->isWrite()) { + // If the wrtie buffer is full, we might unblock now + unblock = wb.isFull(); + wb.deallocate(mshr); + if (unblock) { + // Did we really unblock? + unblock = !wb.isFull(); + cause = Blocked_NoWBBuffers; + } + } else { + unblock = mq.isFull(); + mq.deallocate(mshr); + if (unblock) { + unblock = !mq.isFull(); + cause = Blocked_NoMSHRs; + } + } + } + if (unblock || unblock_target) { + cache->clearBlocked(cause); + } +} + +void +MissQueue::squash(int threadNum) +{ + bool unblock = false; + BlockedCause cause = NUM_BLOCKED_CAUSES; + + if (noTargetMSHR && noTargetMSHR->threadNum == threadNum) { + noTargetMSHR = NULL; + unblock = true; + cause = Blocked_NoTargets; + } + if (mq.isFull()) { + unblock = true; + cause = Blocked_NoMSHRs; + } + mq.squash(threadNum); + if (!mq.havePending()) { + cache->clearMasterRequest(Request_MSHR); + } + if (unblock && !mq.isFull()) { + cache->clearBlocked(cause); + } + +} + +MSHR* +MissQueue::findMSHR(Addr addr, int asid) const +{ + return mq.findMatch(addr,asid); +} + +bool +MissQueue::findWrites(Addr addr, int asid, vector<MSHR*> &writes) const +{ + return wb.findMatches(addr,asid,writes); +} + +void +MissQueue::doWriteback(Addr addr, int asid, + int size, uint8_t *data, bool compressed) +{ + // Generate request + Request * req = new Request(addr, size, 0); + Packet * pkt = new Packet(req, Packet::Writeback, -1); + uint8_t *new_data = new uint8_t[size]; + pkt->dataDynamicArray<uint8_t>(new_data); + if (data) { + memcpy(pkt->getPtr<uint8_t>(), data, size); + } + + if (compressed) { + pkt->flags |= COMPRESSED; + } + + ///All writebacks charged to same thread @todo figure this out + writebacks[pkt->req->getThreadNum()]++; + + allocateWrite(pkt, 0, curTick); +} + + +void +MissQueue::doWriteback(Packet * &pkt) +{ + writebacks[pkt->req->getThreadNum()]++; + allocateWrite(pkt, 0, curTick); +} + + +MSHR* +MissQueue::allocateTargetList(Addr addr, int asid) +{ + MSHR* mshr = mq.allocateTargetList(addr, asid, blkSize); + mshr->pkt->flags |= CACHE_LINE_FILL; + if (mq.isFull()) { + cache->setBlocked(Blocked_NoMSHRs); + } + return mshr; +} + +bool +MissQueue::havePending() +{ + return mq.havePending() || wb.havePending() || prefetcher->havePending(); +} diff --git a/src/mem/cache/miss/miss_queue.hh b/src/mem/cache/miss/miss_queue.hh new file mode 100644 index 000000000..b88b7038c --- /dev/null +++ b/src/mem/cache/miss/miss_queue.hh @@ -0,0 +1,349 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** + * @file + * Miss and writeback queue declarations. + */ + +#ifndef __MISS_QUEUE_HH__ +#define __MISS_QUEUE_HH__ + +#include <vector> + +#include "mem/cache/miss/mshr.hh" +#include "mem/cache/miss/mshr_queue.hh" +#include "base/statistics.hh" + +class BaseCache; +class BasePrefetcher; +/** + * Manages cache misses and writebacks. Contains MSHRs to store miss data + * and the writebuffer for writes/writebacks. + * @todo need to handle data on writes better (encapsulate). + * @todo need to make replacements/writebacks happen in Cache::access + */ +class MissQueue +{ + protected: + /** The MSHRs. */ + MSHRQueue mq; + /** Write Buffer. */ + MSHRQueue wb; + + // PARAMTERS + + /** The number of MSHRs in the miss queue. */ + const int numMSHR; + /** The number of targets for each MSHR. */ + const int numTarget; + /** The number of write buffers. */ + const int writeBuffers; + /** True if the cache should allocate on a write miss. */ + const bool writeAllocate; + /** Pointer to the parent cache. */ + BaseCache* cache; + + /** The Prefetcher */ + BasePrefetcher *prefetcher; + + /** The block size of the parent cache. */ + int blkSize; + + /** Increasing order number assigned to each incoming request. */ + uint64_t order; + + bool prefetchMiss; + + // Statistics + /** + * @addtogroup CacheStatistics + * @{ + */ + /** Number of blocks written back per thread. */ + Stats::Vector<> writebacks; + + /** Number of misses that hit in the MSHRs per command and thread. */ + Stats::Vector<> mshr_hits[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]; + /** 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]; + /** 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]; + /** 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]; + /** 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]; + /** 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]; + /** 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]; + /** 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]; + /** The average overall latency of an MSHR miss. */ + Stats::Formula overallAvgMshrUncacheableLatency; + + /** The number of times a thread hit its MSHR cap. */ + Stats::Vector<> mshr_cap_events; + /** The number of times software prefetches caused the MSHR to block. */ + Stats::Vector<> soft_prefetch_mshr_full; + + Stats::Scalar<> mshr_no_allocate_misses; + + /** + * @} + */ + + private: + /** Pointer to the MSHR that has no targets. */ + MSHR* noTargetMSHR; + + /** + * Allocate a new MSHR to handle the provided miss. + * @param req The miss to buffer. + * @param size The number of bytes to fetch. + * @param time The time the miss occurs. + * @return A pointer to the new MSHR. + */ + MSHR* allocateMiss(Packet * &pkt, int size, Tick time); + + /** + * Allocate a new WriteBuffer to handle the provided write. + * @param req The write to handle. + * @param size The number of bytes to write. + * @param time The time the write occurs. + * @return A pointer to the new write buffer. + */ + MSHR* allocateWrite(Packet * &pkt, int size, Tick time); + + public: + /** + * Simple Constructor. Initializes all needed internal storage and sets + * parameters. + * @param numMSHRs The number of outstanding misses to handle. + * @param numTargets The number of outstanding targets to each miss. + * @param write_buffers The number of outstanding writes to handle. + * @param write_allocate If true, treat write misses the same as reads. + */ + MissQueue(int numMSHRs, int numTargets, int write_buffers, + bool write_allocate, bool prefetch_miss); + + /** + * Deletes all allocated internal storage. + */ + ~MissQueue(); + + /** + * Register statistics for this object. + * @param name The name of the parent cache. + */ + void regStats(const std::string &name); + + /** + * Called by the parent cache to set the back pointer. + * @param _cache A pointer to the parent cache. + */ + void setCache(BaseCache *_cache); + + void setPrefetcher(BasePrefetcher *_prefetcher); + + /** + * Handle a cache miss properly. Either allocate an MSHR for the request, + * or forward it through the write buffer. + * @param req The request that missed in the cache. + * @param blk_size The block size of the cache. + * @param time The time the miss is detected. + */ + void handleMiss(Packet * &pkt, int blk_size, Tick time); + + /** + * Fetch the block for the given address and buffer the given target. + * @param addr The address to fetch. + * @param asid The address space of the address. + * @param blk_size The block size of the cache. + * @param time The time the miss is detected. + * @param target The target for the fetch. + */ + MSHR* fetchBlock(Addr addr, int asid, int blk_size, Tick time, + Packet * &target); + + /** + * Selects a outstanding request to service. + * @return The request to service, NULL if none found. + */ + Packet * getPacket(); + + /** + * Set the command to the given bus command. + * @param req The request to update. + * @param cmd The bus command to use. + */ + void setBusCmd(Packet * &pkt, Packet::Command cmd); + + /** + * Restore the original command in case of a bus transmission error. + * @param req The request to reset. + */ + void restoreOrigCmd(Packet * &pkt); + + /** + * Marks a request as in service (sent on the bus). This can have side + * effect since storage for no response commands is deallocated once they + * are successfully sent. + * @param req The request that was sent on the bus. + */ + void markInService(Packet * &pkt); + + /** + * Collect statistics and free resources of a satisfied request. + * @param req The request that has been satisfied. + * @param time The time when the request is satisfied. + */ + void handleResponse(Packet * &pkt, Tick time); + + /** + * Removes all outstanding requests for a given thread number. If a request + * has been sent to the bus, this function removes all of its targets. + * @param req->getThreadNum()ber The thread number of the requests to squash. + */ + void squash(int threadNum); + + /** + * Return the current number of outstanding misses. + * @return the number of outstanding misses. + */ + int getMisses() + { + return mq.getAllocatedTargets(); + } + + /** + * Searches for the supplied address in the miss queue. + * @param addr The address to look for. + * @param asid The address space id. + * @return The MSHR that contains the address, NULL if not found. + * @warning Currently only searches the miss queue. If non write allocate + * might need to search the write buffer for coherence. + */ + MSHR* findMSHR(Addr addr, int asid) const; + + /** + * Searches for the supplied address in the write buffer. + * @param addr The address to look for. + * @param asid The address space id. + * @param writes The list of writes that match the address. + * @return True if any writes are found + */ + bool findWrites(Addr addr, int asid, std::vector<MSHR*>& writes) const; + + /** + * Perform a writeback of dirty data to the given address. + * @param addr The address to write to. + * @param asid The address space id. + * @param xc The execution context of the address space. + * @param size The number of bytes to write. + * @param data The data to write, can be NULL. + * @param compressed True if the data is compressed. + */ + void doWriteback(Addr addr, int asid, + int size, uint8_t *data, bool compressed); + + /** + * Perform the given writeback request. + * @param req The writeback request. + */ + void doWriteback(Packet * &pkt); + + /** + * Returns true if there are outstanding requests. + * @return True if there are outstanding requests. + */ + bool havePending(); + + /** + * Add a target to the given MSHR. This assumes it is in the miss queue. + * @param mshr The mshr to add a target to. + * @param req The target to add. + */ + void addTarget(MSHR *mshr, Packet * &pkt) + { + mq.allocateTarget(mshr, pkt); + } + + /** + * Allocate a MSHR to hold a list of targets to a block involved in a copy. + * If the block is marked done then the MSHR already holds the data to + * fill the block. Otherwise the block needs to be fetched. + * @param addr The address to buffer. + * @param asid The address space ID. + * @return A pointer to the allocated MSHR. + */ + MSHR* allocateTargetList(Addr addr, int asid); + +}; + +#endif //__MISS_QUEUE_HH__ diff --git a/src/mem/cache/miss/mshr.cc b/src/mem/cache/miss/mshr.cc new file mode 100644 index 000000000..fe8cbeea4 --- /dev/null +++ b/src/mem/cache/miss/mshr.cc @@ -0,0 +1,182 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + * Dave Greene + */ + +/** + * @file + * Miss Status and Handling Register (MSHR) definitions. + */ + +#include <assert.h> +#include <string> +#include <vector> + +#include "mem/cache/miss/mshr.hh" +#include "sim/root.hh" // for curTick +#include "sim/host.hh" +#include "base/misc.hh" +#include "mem/cache/cache.hh" + +using namespace std; + +MSHR::MSHR() +{ + inService = false; + ntargets = 0; + threadNum = -1; +} + +void +MSHR::allocate(Packet::Command cmd, Addr _addr, int _asid, int size, + Packet * &target) +{ + assert("NEED TO FIX YET\n" && 0); +#if 0 + assert(targets.empty()); + addr = _addr; + asid = _asid; + + pkt = new Packet(); // allocate new memory request + pkt->addr = addr; //picked physical address for now + pkt->cmd = cmd; + pkt->size = size; + pkt->data = new uint8_t[size]; + pkt->senderState = this; + //Set the time here for latency calculations + pkt->time = curTick; + + if (target) { + pkt->req = target->req; + allocateTarget(target); + } +#endif +} + +// Since we aren't sure if data is being used, don't copy here. +/** + * @todo When we have a "global" data flag, might want to copy data here. + */ +void +MSHR::allocateAsBuffer(Packet * &target) +{ + addr = target->getAddr(); + asid = target->req->getAsid(); + threadNum = target->req->getThreadNum(); + pkt = new Packet(target->req, target->cmd, -1); + uint8_t *new_data = new uint8_t[target->getSize()]; + pkt->dataDynamicArray<uint8_t>(new_data); + pkt->senderState = (Packet::SenderState*)this; + pkt->time = curTick; +} + +void +MSHR::deallocate() +{ + assert(targets.empty()); + assert(ntargets == 0); + pkt = NULL; + inService = false; + allocIter = NULL; + readyIter = NULL; +} + +/* + * Adds a target to an MSHR + */ +void +MSHR::allocateTarget(Packet * &target) +{ + //If we append an invalidate and we issued a read to the bus, + //but now have some pending writes, we need to move + //the invalidate to before the first non-read + if (inService && pkt->isRead() && target->isInvalidate()) { + std::list<Packet *> temp; + + while (!targets.empty()) { + if (!targets.front()->isRead()) break; + //Place on top of temp stack + temp.push_front(targets.front()); + //Remove from targets + targets.pop_front(); + } + + //Now that we have all the reads off until first non-read, we can + //place the invalidate on + targets.push_front(target); + + //Now we pop off the temp_stack and put them back + while (!temp.empty()) { + targets.push_front(temp.front()); + temp.pop_front(); + } + } + else { + targets.push_back(target); + } + + ++ntargets; + assert(targets.size() == ntargets); + /** + * @todo really prioritize the target commands. + */ + + if (!inService && target->isWrite()) { + pkt->cmd = Packet::WriteReq; + } +} + + + +void +MSHR::dump() +{ + ccprintf(cerr, + "inService: %d thread: %d\n" + "Addr: %x asid: %d ntargets %d\n" + "Targets:\n", + inService, threadNum, addr, asid, ntargets); + + TargetListIterator tar_it = targets.begin(); + for (int i = 0; i < ntargets; i++) { + assert(tar_it != targets.end()); + + ccprintf(cerr, "\t%d: Addr: %x cmd: %d\n", + i, (*tar_it)->getAddr(), (*tar_it)->cmdToIndex()); + + tar_it++; + } + ccprintf(cerr, "\n"); +} + +MSHR::~MSHR() +{ + if (pkt) + pkt = NULL; +} diff --git a/src/mem/cache/miss/mshr.hh b/src/mem/cache/miss/mshr.hh new file mode 100644 index 000000000..167aa26cd --- /dev/null +++ b/src/mem/cache/miss/mshr.hh @@ -0,0 +1,179 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + */ + +/** + * @file + * Miss Status and Handling Register (MSHR) declaration. + */ + +#ifndef __MSHR_HH__ +#define __MSHR_HH__ + +#include "mem/packet.hh" +#include <list> +#include <deque> + +class MSHR; + +/** + * Miss Status and handling Register. This class keeps all the information + * needed to handle a cache miss including a list of target requests. + */ +class MSHR { + public: + /** Defines the Data structure of the MSHR targetlist. */ + typedef std::list<Packet *> TargetList; + /** Target list iterator. */ + typedef std::list<Packet *>::iterator TargetListIterator; + /** A list of MSHRs. */ + typedef std::list<MSHR *> List; + /** MSHR list iterator. */ + typedef List::iterator Iterator; + /** MSHR list const_iterator. */ + typedef List::const_iterator ConstIterator; + + /** Address of the miss. */ + Addr addr; + /** Adress space id of the miss. */ + short asid; + /** True if the request has been sent to the bus. */ + bool inService; + /** Thread number of the miss. */ + int threadNum; + /** The request that is forwarded to the next level of the hierarchy. */ + Packet * pkt; + /** The number of currently allocated targets. */ + short ntargets; + /** The original requesting command. */ + Packet::Command originalCmd; + /** Order number of assigned by the miss queue. */ + uint64_t order; + + /** + * Pointer to this MSHR on the ready list. + * @sa MissQueue, MSHRQueue::readyList + */ + Iterator readyIter; + /** + * Pointer to this MSHR on the allocated list. + * @sa MissQueue, MSHRQueue::allocatedList + */ + Iterator allocIter; + +private: + /** List of all requests that match the address */ + TargetList targets; + +public: + /** + * Allocate a miss to this MSHR. + * @param cmd The requesting command. + * @param addr The address of the miss. + * @param asid The address space id of the miss. + * @param size The number of bytes to request. + * @param req The original miss. + */ + void allocate(Packet::Command cmd, Addr addr, int asid, int size, + Packet * &pkt); + + /** + * Allocate this MSHR as a buffer for the given request. + * @param target The memory request to buffer. + */ + void allocateAsBuffer(Packet * &target); + + /** + * Mark this MSHR as free. + */ + void deallocate(); + + /** + * Add a request to the list of targets. + * @param target The target. + */ + void allocateTarget(Packet * &target); + + /** A simple constructor. */ + MSHR(); + /** A simple destructor. */ + ~MSHR(); + + /** + * Returns the current number of allocated targets. + * @return The current number of allocated targets. + */ + int getNumTargets() + { + return(ntargets); + } + + /** + * Returns a pointer to the target list. + * @return a pointer to the target list. + */ + TargetList* getTargetList() + { + return &targets; + } + + /** + * Returns a reference to the first target. + * @return A pointer to the first target. + */ + Packet * getTarget() + { + return targets.front(); + } + + /** + * Pop first target. + */ + void popTarget() + { + --ntargets; + targets.pop_front(); + } + + /** + * Returns true if there are targets left. + * @return true if there are targets + */ + bool hasTargets() + { + return !targets.empty(); + } + + /** + * Prints the contents of this MSHR to stderr. + */ + void dump(); +}; + +#endif //__MSHR_HH__ diff --git a/src/mem/cache/miss/mshr_queue.cc b/src/mem/cache/miss/mshr_queue.cc new file mode 100644 index 000000000..6516a99f8 --- /dev/null +++ b/src/mem/cache/miss/mshr_queue.cc @@ -0,0 +1,269 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** @file + * Definition of the MSHRQueue. + */ + +#include "mem/cache/miss/mshr_queue.hh" +#include "sim/eventq.hh" + +using namespace std; + +MSHRQueue::MSHRQueue(int num_mshrs, int reserve) + : numMSHRs(num_mshrs + reserve - 1), numReserve(reserve) +{ + allocated = 0; + inServiceMSHRs = 0; + allocatedTargets = 0; + registers = new MSHR[numMSHRs]; + for (int i = 0; i < numMSHRs; ++i) { + freeList.push_back(®isters[i]); + } +} + +MSHRQueue::~MSHRQueue() +{ + delete [] registers; +} + +MSHR* +MSHRQueue::findMatch(Addr addr, int asid) const +{ + MSHR::ConstIterator i = allocatedList.begin(); + MSHR::ConstIterator end = allocatedList.end(); + for (; i != end; ++i) { + MSHR *mshr = *i; + if (mshr->addr == addr) { + return mshr; + } + } + return NULL; +} + +bool +MSHRQueue::findMatches(Addr addr, int asid, vector<MSHR*>& matches) const +{ + // Need an empty vector + assert(matches.empty()); + bool retval = false; + MSHR::ConstIterator i = allocatedList.begin(); + MSHR::ConstIterator end = allocatedList.end(); + for (; i != end; ++i) { + MSHR *mshr = *i; + if (mshr->addr == addr) { + retval = true; + matches.push_back(mshr); + } + } + return retval; + +} + +MSHR* +MSHRQueue::findPending(Packet * &pkt) const +{ + MSHR::ConstIterator i = pendingList.begin(); + MSHR::ConstIterator end = pendingList.end(); + for (; i != end; ++i) { + MSHR *mshr = *i; + if (mshr->addr < pkt->getAddr()) { + if (mshr->addr + mshr->pkt->getSize() > pkt->getAddr()) { + return mshr; + } + } else { + if (pkt->getAddr() + pkt->getSize() > mshr->addr) { + return mshr; + } + } + + //need to check destination address for copies. + //TEMP NOT DOING COPIES +#if 0 + if (mshr->pkt->cmd == Copy) { + Addr dest = mshr->pkt->dest; + if (dest < pkt->addr) { + if (dest + mshr->pkt->size > pkt->addr) { + return mshr; + } + } else { + if (pkt->addr + pkt->size > dest) { + return mshr; + } + } + } +#endif + } + return NULL; +} + +MSHR* +MSHRQueue::allocate(Packet * &pkt, int size) +{ + Addr aligned_addr = pkt->getAddr() & ~((Addr)size - 1); + MSHR *mshr = freeList.front(); + assert(mshr->getNumTargets() == 0); + freeList.pop_front(); + + if (!pkt->needsResponse()) { + mshr->allocateAsBuffer(pkt); + } else { + assert(size !=0); + mshr->allocate(pkt->cmd, aligned_addr, pkt->req->getAsid(), size, pkt); + allocatedTargets += 1; + } + mshr->allocIter = allocatedList.insert(allocatedList.end(), mshr); + mshr->readyIter = pendingList.insert(pendingList.end(), mshr); + + allocated += 1; + return mshr; +} + +MSHR* +MSHRQueue::allocateFetch(Addr addr, int asid, int size, Packet * &target) +{ + MSHR *mshr = freeList.front(); + assert(mshr->getNumTargets() == 0); + freeList.pop_front(); + mshr->allocate(Packet::ReadReq, addr, asid, size, target); + mshr->allocIter = allocatedList.insert(allocatedList.end(), mshr); + mshr->readyIter = pendingList.insert(pendingList.end(), mshr); + + allocated += 1; + return mshr; +} + +MSHR* +MSHRQueue::allocateTargetList(Addr addr, int asid, int size) +{ + MSHR *mshr = freeList.front(); + assert(mshr->getNumTargets() == 0); + freeList.pop_front(); + Packet * dummy; + mshr->allocate(Packet::ReadReq, addr, asid, size, dummy); + mshr->allocIter = allocatedList.insert(allocatedList.end(), mshr); + mshr->inService = true; + ++inServiceMSHRs; + ++allocated; + return mshr; +} + + +void +MSHRQueue::deallocate(MSHR* mshr) +{ + deallocateOne(mshr); +} + +MSHR::Iterator +MSHRQueue::deallocateOne(MSHR* mshr) +{ + MSHR::Iterator retval = allocatedList.erase(mshr->allocIter); + freeList.push_front(mshr); + allocated--; + allocatedTargets -= mshr->getNumTargets(); + if (mshr->inService) { + inServiceMSHRs--; + } else { + pendingList.erase(mshr->readyIter); + } + mshr->deallocate(); + return retval; +} + +void +MSHRQueue::moveToFront(MSHR *mshr) +{ + if (!mshr->inService) { + assert(mshr == *(mshr->readyIter)); + pendingList.erase(mshr->readyIter); + mshr->readyIter = pendingList.insert(pendingList.begin(), mshr); + } +} + +void +MSHRQueue::markInService(MSHR* mshr) +{ + //assert(mshr == pendingList.front()); + if (!mshr->pkt->needsResponse()) { + assert(mshr->getNumTargets() == 0); + deallocate(mshr); + return; + } + mshr->inService = true; + pendingList.erase(mshr->readyIter); + mshr->readyIter = NULL; + inServiceMSHRs += 1; + //pendingList.pop_front(); +} + +void +MSHRQueue::markPending(MSHR* mshr, Packet::Command cmd) +{ + assert(mshr->readyIter == NULL); + mshr->pkt->cmd = cmd; + mshr->pkt->flags &= ~SATISFIED; + mshr->inService = false; + --inServiceMSHRs; + /** + * @ todo might want to add rerequests to front of pending list for + * performance. + */ + mshr->readyIter = pendingList.insert(pendingList.end(), mshr); +} + +void +MSHRQueue::squash(int threadNum) +{ + MSHR::Iterator i = allocatedList.begin(); + MSHR::Iterator end = allocatedList.end(); + for (; i != end;) { + MSHR *mshr = *i; + if (mshr->threadNum == threadNum) { + while (mshr->hasTargets()) { + Packet * target = mshr->getTarget(); + mshr->popTarget(); + + assert(target->req->getThreadNum() == threadNum); + target = NULL; + } + assert(!mshr->hasTargets()); + assert(mshr->ntargets==0); + if (!mshr->inService) { + i = deallocateOne(mshr); + } else { + //mshr->pkt->flags &= ~CACHE_LINE_FILL; + ++i; + } + } else { + ++i; + } + } +} diff --git a/src/mem/cache/miss/mshr_queue.hh b/src/mem/cache/miss/mshr_queue.hh new file mode 100644 index 000000000..a67f1b9a6 --- /dev/null +++ b/src/mem/cache/miss/mshr_queue.hh @@ -0,0 +1,239 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** @file + * Declaration of a structure to manage MSHRs. + */ + +#ifndef __MSHR_QUEUE_HH__ +#define __MSHR_QUEUE_HH__ + +#include <vector> +#include "mem/cache/miss/mshr.hh" + +/** + * A Class for maintaining a list of pending and allocated memory requests. + */ +class MSHRQueue { + private: + /** MSHR storage. */ + MSHR* registers; + /** Holds pointers to all allocated MSHRs. */ + MSHR::List allocatedList; + /** Holds pointers to MSHRs that haven't been sent to the bus. */ + MSHR::List pendingList; + /** Holds non allocated MSHRs. */ + MSHR::List freeList; + + // Parameters + /** + * The total number of MSHRs in this queue. This number is set as the + * number of MSHRs requested plus (numReserve - 1). This allows for + * the same number of effective MSHRs while still maintaining the reserve. + */ + const int numMSHRs; + + /** + * The number of MSHRs to hold in reserve. This is needed because copy + * operations can allocate upto 4 MSHRs at one time. + */ + const int numReserve; + + public: + /** The number of allocated MSHRs. */ + int allocated; + /** The number of MSHRs that have been forwarded to the bus. */ + int inServiceMSHRs; + /** The number of targets waiting for response. */ + int allocatedTargets; + + /** + * Create a queue with a given number of MSHRs. + * @param num_mshrs The number of MSHRs in this queue. + * @param reserve The minimum number of MSHRs needed to satisfy any access. + */ + MSHRQueue(int num_mshrs, int reserve = 1); + + /** Destructor */ + ~MSHRQueue(); + + /** + * Find the first MSHR that matches the provide address and asid. + * @param addr The address to find. + * @param asid The address space id. + * @return Pointer to the matching MSHR, null if not found. + */ + MSHR* findMatch(Addr addr, int asid) const; + + /** + * Find and return all the matching MSHRs in the provided vector. + * @param addr The address to find. + * @param asid The address space ID. + * @param matches The vector to return pointers to the matching MSHRs. + * @return True if any matches are found, false otherwise. + * @todo Typedef the vector?? + */ + bool findMatches(Addr addr, int asid, std::vector<MSHR*>& matches) const; + + /** + * Find any pending requests that overlap the given request. + * @param req The request to find. + * @return A pointer to the earliest matching MSHR. + */ + MSHR* findPending(Packet * &pkt) const; + + /** + * Allocates a new MSHR for the request and size. This places the request + * as the first target in the MSHR. + * @param req The request to handle. + * @param size The number in bytes to fetch from memory. + * @return The a pointer to the MSHR allocated. + * + * @pre There are free MSHRs. + */ + MSHR* allocate(Packet * &pkt, int size = 0); + + /** + * Allocate a read request for the given address, and places the given + * target on the target list. + * @param addr The address to fetch. + * @param asid The address space for the fetch. + * @param size The number of bytes to request. + * @param target The first target for the request. + * @return Pointer to the new MSHR. + */ + MSHR* allocateFetch(Addr addr, int asid, int size, Packet * &target); + + /** + * Allocate a target list for the given address. + * @param addr The address to fetch. + * @param asid The address space for the fetch. + * @param size The number of bytes to request. + * @return Pointer to the new MSHR. + */ + MSHR* allocateTargetList(Addr addr, int asid, int size); + + /** + * Removes the given MSHR from the queue. This places the MSHR on the + * free list. + * @param mshr + */ + void deallocate(MSHR* mshr); + + /** + * Allocates a target to the given MSHR. Used to keep track of the number + * of outstanding targets. + * @param mshr The MSHR to allocate the target to. + * @param req The target request. + */ + void allocateTarget(MSHR* mshr, Packet * &pkt) + { + mshr->allocateTarget(pkt); + allocatedTargets += 1; + } + + /** + * Remove a MSHR from the queue. Returns an iterator into the allocatedList + * for faster squash implementation. + * @param mshr The MSHR to remove. + * @return An iterator to the next entry in the allocatedList. + */ + MSHR::Iterator deallocateOne(MSHR* mshr); + + /** + * Moves the MSHR to the front of the pending list if it is not in service. + * @param mshr The mshr to move. + */ + void moveToFront(MSHR *mshr); + + /** + * Mark the given MSHR as in service. This removes the MSHR from the + * pendingList. Deallocates the MSHR if it does not expect a response. + * @param mshr The MSHR to mark in service. + */ + void markInService(MSHR* mshr); + + /** + * Mark an in service mshr as pending, used to resend a request. + * @param mshr The MSHR to resend. + * @param cmd The command to resend. + */ + void markPending(MSHR* mshr, Packet::Command cmd); + + /** + * Squash outstanding requests with the given thread number. If a request + * is in service, just squashes the targets. + * @param req->getThreadNum()ber The thread to squash. + */ + void squash(int threadNum); + + /** + * Returns true if the pending list is not empty. + * @return True if there are outstanding requests. + */ + bool havePending() const + { + return !pendingList.empty(); + } + + /** + * Returns true if there are no free MSHRs. + * @return True if this queue is full. + */ + bool isFull() const + { + return (allocated > numMSHRs - numReserve); + } + + /** + * Returns the request at the head of the pendingList. + * @return The next request to service. + */ + Packet * getReq() const + { + if (pendingList.empty()) { + return NULL; + } + MSHR* mshr = pendingList.front(); + return mshr->pkt; + } + + /** + * Returns the number of outstanding targets. + * @return the number of allocated targets. + */ + int getAllocatedTargets() const + { + return allocatedTargets; + } + +}; + +#endif //__MSHR_QUEUE_HH__ diff --git a/src/mem/cache/prefetch/base_prefetcher.cc b/src/mem/cache/prefetch/base_prefetcher.cc new file mode 100644 index 000000000..29da53746 --- /dev/null +++ b/src/mem/cache/prefetch/base_prefetcher.cc @@ -0,0 +1,250 @@ +/* + * Copyright (c) 2005 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: Ron Dreslinski + */ + +/** + * @file + * Hardware Prefetcher Definition. + */ + +#include "base/trace.hh" +#include "mem/cache/base_cache.hh" +#include "mem/cache/prefetch/base_prefetcher.hh" +#include "mem/request.hh" +#include <list> + +BasePrefetcher::BasePrefetcher(int size, bool pageStop, bool serialSquash, + bool cacheCheckPush, bool onlyData) + :size(size), pageStop(pageStop), serialSquash(serialSquash), + cacheCheckPush(cacheCheckPush), only_data(onlyData) +{ +} + +void +BasePrefetcher::setCache(BaseCache *_cache) +{ + cache = _cache; + blkSize = cache->getBlockSize(); +} + +void +BasePrefetcher::regStats(const std::string &name) +{ + pfIdentified + .name(name + ".prefetcher.num_hwpf_identified") + .desc("number of hwpf identified") + ; + + pfMSHRHit + .name(name + ".prefetcher.num_hwpf_already_in_mshr") + .desc("number of hwpf that were already in mshr") + ; + + pfCacheHit + .name(name + ".prefetcher.num_hwpf_already_in_cache") + .desc("number of hwpf that were already in the cache") + ; + + pfBufferHit + .name(name + ".prefetcher.num_hwpf_already_in_prefetcher") + .desc("number of hwpf that were already in the prefetch queue") + ; + + pfRemovedFull + .name(name + ".prefetcher.num_hwpf_evicted") + .desc("number of hwpf removed due to no buffer left") + ; + + pfRemovedMSHR + .name(name + ".prefetcher.num_hwpf_removed_MSHR_hit") + .desc("number of hwpf removed because MSHR allocated") + ; + + pfIssued + .name(name + ".prefetcher.num_hwpf_issued") + .desc("number of hwpf issued") + ; + + pfSpanPage + .name(name + ".prefetcher.num_hwpf_span_page") + .desc("number of hwpf spanning a virtual page") + ; + + pfSquashed + .name(name + ".prefetcher.num_hwpf_squashed_from_miss") + .desc("number of hwpf that got squashed due to a miss aborting calculation time") + ; +} + +Packet * +BasePrefetcher::getPacket() +{ + DPRINTF(HWPrefetch, "%s:Requesting a hw_pf to issue\n", cache->name()); + + if (pf.empty()) { + DPRINTF(HWPrefetch, "%s:No HW_PF found\n", cache->name()); + return NULL; + } + + Packet * pkt; + bool keepTrying = false; + do { + pkt = *pf.begin(); + pf.pop_front(); + if (!cacheCheckPush) { + keepTrying = inCache(pkt); + } + if (pf.empty()) { + cache->clearMasterRequest(Request_PF); + if (keepTrying) return NULL; //None left, all were in cache + } + } while (keepTrying); + + pfIssued++; + return pkt; +} + +void +BasePrefetcher::handleMiss(Packet * &pkt, Tick time) +{ + if (!pkt->req->isUncacheable() && !(pkt->req->isInstRead() && only_data)) + { + //Calculate the blk address + Addr blkAddr = pkt->getAddr() & ~(Addr)(blkSize-1); + + //Check if miss is in pfq, if so remove it + std::list<Packet *>::iterator iter = inPrefetch(blkAddr); + if (iter != pf.end()) { + DPRINTF(HWPrefetch, "%s:Saw a miss to a queued prefetch, removing it\n", cache->name()); + pfRemovedMSHR++; + pf.erase(iter); + if (pf.empty()) + cache->clearMasterRequest(Request_PF); + } + + //Remove anything in queue with delay older than time + //since everything is inserted in time order, start from end + //and work until pf.empty() or time is earlier + //This is done to emulate Aborting the previous work on a new miss + //Needed for serial calculators like GHB + if (serialSquash) { + iter = pf.end(); + iter--; + while (!pf.empty() && ((*iter)->time >= time)) { + pfSquashed++; + pf.pop_back(); + iter--; + } + if (pf.empty()) + cache->clearMasterRequest(Request_PF); + } + + + std::list<Addr> addresses; + std::list<Tick> delays; + calculatePrefetch(pkt, addresses, delays); + + std::list<Addr>::iterator addr = addresses.begin(); + std::list<Tick>::iterator delay = delays.begin(); + while (addr != addresses.end()) + { + DPRINTF(HWPrefetch, "%s:Found a pf canidate, inserting into prefetch queue\n", cache->name()); + //temp calc this here... + pfIdentified++; + //create a prefetch memreq + Request * prefetchReq = new Request(*addr, blkSize, 0); + Packet * prefetch; + prefetch = new Packet(prefetchReq, Packet::HardPFReq, -1); + uint8_t *new_data = new uint8_t[blkSize]; + prefetch->dataDynamicArray<uint8_t>(new_data); + prefetch->req->setThreadContext(pkt->req->getCpuNum(), + pkt->req->getThreadNum()); + + prefetch->time = time + (*delay); //@todo ADD LATENCY HERE + //... initialize + + //Check if it is already in the cache + if (cacheCheckPush) { + if (inCache(prefetch)) { + addr++; + delay++; + continue; + } + } + + //Check if it is already in the miss_queue + if (inMissQueue(prefetch->getAddr(), prefetch->req->getAsid())) { + addr++; + delay++; + continue; + } + + //Check if it is already in the pf buffer + if (inPrefetch(prefetch->getAddr()) != pf.end()) { + pfBufferHit++; + addr++; + delay++; + continue; + } + + //We just remove the head if we are full + if (pf.size() == size) + { + DPRINTF(HWPrefetch, "%s:Inserting into prefetch queue, it was full removing oldest\n", cache->name()); + pfRemovedFull++; + pf.pop_front(); + } + + pf.push_back(prefetch); + prefetch->flags |= CACHE_LINE_FILL; + + //Make sure to request the bus, with proper delay + cache->setMasterRequest(Request_PF, prefetch->time); + + //Increment through the list + addr++; + delay++; + } + } +} + +std::list<Packet *>::iterator +BasePrefetcher::inPrefetch(Addr address) +{ + //Guaranteed to only be one match, we always check before inserting + std::list<Packet *>::iterator iter; + for (iter=pf.begin(); iter != pf.end(); iter++) { + if (((*iter)->getAddr() & ~(Addr)(blkSize-1)) == address) { + return iter; + } + } + return pf.end(); +} + + diff --git a/src/mem/cache/prefetch/base_prefetcher.hh b/src/mem/cache/prefetch/base_prefetcher.hh new file mode 100644 index 000000000..3e4fc89d1 --- /dev/null +++ b/src/mem/cache/prefetch/base_prefetcher.hh @@ -0,0 +1,117 @@ +/* + * Copyright (c) 2005 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: Ron Dreslinski + */ + +/** + * @file + * Miss and writeback queue declarations. + */ + +#ifndef __MEM_CACHE_PREFETCH_BASE_PREFETCHER_HH__ +#define __MEM_CACHE_PREFETCH_BASE_PREFETCHER_HH__ + +#include "mem/packet.hh" +#include <list> + +class BaseCache; +class BasePrefetcher +{ + protected: + + /** The Prefetch Queue. */ + std::list<Packet *> pf; + + // PARAMETERS + + /** The number of MSHRs in the Prefetch Queue. */ + const int size; + + /** Pointr to the parent cache. */ + BaseCache* cache; + + /** The block size of the parent cache. */ + int blkSize; + + /** Do we prefetch across page boundaries. */ + bool pageStop; + + /** Do we remove prefetches with later times than a new miss.*/ + bool serialSquash; + + /** Do we check if it is in the cache when inserting into buffer, + or removing.*/ + bool cacheCheckPush; + + /** Do we prefetch on only data reads, or on inst reads as well. */ + bool only_data; + + public: + + Stats::Scalar<> pfIdentified; + Stats::Scalar<> pfMSHRHit; + Stats::Scalar<> pfCacheHit; + Stats::Scalar<> pfBufferHit; + Stats::Scalar<> pfRemovedFull; + Stats::Scalar<> pfRemovedMSHR; + Stats::Scalar<> pfIssued; + Stats::Scalar<> pfSpanPage; + Stats::Scalar<> pfSquashed; + + void regStats(const std::string &name); + + public: + BasePrefetcher(int numMSHRS, bool pageStop, bool serialSquash, + bool cacheCheckPush, bool onlyData); + + virtual ~BasePrefetcher() {} + + void setCache(BaseCache *_cache); + + void handleMiss(Packet * &pkt, Tick time); + + Packet * getPacket(); + + bool havePending() + { + return !pf.empty(); + } + + virtual void calculatePrefetch(Packet * &pkt, + std::list<Addr> &addresses, + std::list<Tick> &delays) = 0; + + virtual bool inCache(Packet * &pkt) = 0; + + virtual bool inMissQueue(Addr address, int asid) = 0; + + std::list<Packet *>::iterator inPrefetch(Addr address); +}; + + +#endif //__MEM_CACHE_PREFETCH_BASE_PREFETCHER_HH__ diff --git a/src/mem/cache/prefetch/ghb_prefetcher.cc b/src/mem/cache/prefetch/ghb_prefetcher.cc new file mode 100644 index 000000000..247ec6e8b --- /dev/null +++ b/src/mem/cache/prefetch/ghb_prefetcher.cc @@ -0,0 +1,54 @@ +/* + * Copyright (c) 2005 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: Ron Dreslinski + * Steve Reinhardt + */ + +/** + * @file + * GHB Prefetcher template instantiations. + */ + +#include "mem/cache/tags/cache_tags.hh" + +#include "mem/cache/tags/lru.hh" + +#include "base/compression/null_compression.hh" + +#include "mem/cache/miss/miss_queue.hh" +#include "mem/cache/miss/blocking_buffer.hh" + +#include "mem/cache/prefetch/ghb_prefetcher.hh" + +// Template Instantiations +#ifndef DOXYGEN_SHOULD_SKIP_THIS + +template class GHBPrefetcher<CacheTags<LRU,NullCompression>, MissQueue>; +template class GHBPrefetcher<CacheTags<LRU,NullCompression>, BlockingBuffer>; + +#endif //DOXYGEN_SHOULD_SKIP_THIS diff --git a/src/mem/cache/prefetch/ghb_prefetcher.hh b/src/mem/cache/prefetch/ghb_prefetcher.hh new file mode 100644 index 000000000..c22b763d1 --- /dev/null +++ b/src/mem/cache/prefetch/ghb_prefetcher.hh @@ -0,0 +1,114 @@ +/* + * Copyright (c) 2005 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: Ron Dreslinski + */ + +/** + * @file + * Describes a ghb prefetcher based on template policies. + */ + +#ifndef __MEM_CACHE_PREFETCH_GHB_PREFETCHER_HH__ +#define __MEM_CACHE_PREFETCH_GHB_PREFETCHER_HH__ + +#include "base/misc.hh" // fatal, panic, and warn + +#include "mem/cache/prefetch/prefetcher.hh" + +/** + * A template-policy based cache. The behavior of the cache can be altered by + * supplying different template policies. TagStore handles all tag and data + * storage @sa TagStore. Buffering handles all misses and writes/writebacks + * @sa MissQueue. Coherence handles all coherence policy details @sa + * UniCoherence, SimpleMultiCoherence. + */ +template <class TagStore, class Buffering> +class GHBPrefetcher : public Prefetcher<TagStore, Buffering> +{ + protected: + + Buffering* mq; + TagStore* tags; + + Addr second_last_miss_addr[64/*MAX_CPUS*/]; + Addr last_miss_addr[64/*MAX_CPUS*/]; + + Tick latency; + int degree; + bool useCPUId; + + public: + + GHBPrefetcher(int size, bool pageStop, bool serialSquash, + bool cacheCheckPush, bool onlyData, + Tick latency, int degree, bool useCPUId) + :Prefetcher<TagStore, Buffering>(size, pageStop, serialSquash, + cacheCheckPush, onlyData), + latency(latency), degree(degree), useCPUId(useCPUId) + { + } + + ~GHBPrefetcher() {} + + void calculatePrefetch(Packet * &pkt, std::list<Addr> &addresses, + std::list<Tick> &delays) + { + Addr blkAddr = pkt->getAddr() & ~(Addr)(this->blkSize-1); + int cpuID = pkt->req->getCpuNum(); + if (!useCPUId) cpuID = 0; + + + int new_stride = blkAddr - last_miss_addr[cpuID]; + int old_stride = last_miss_addr[cpuID] - + second_last_miss_addr[cpuID]; + + second_last_miss_addr[cpuID] = last_miss_addr[cpuID]; + last_miss_addr[cpuID] = blkAddr; + + if (new_stride == old_stride) { + for (int d=1; d <= degree; d++) { + Addr newAddr = blkAddr + d * new_stride; + if (this->pageStop && + (blkAddr & ~(TheISA::VMPageSize - 1)) != + (newAddr & ~(TheISA::VMPageSize - 1))) + { + //Spanned the page, so now stop + this->pfSpanPage += degree - d + 1; + return; + } + else + { + addresses.push_back(newAddr); + delays.push_back(latency); + } + } + } + } +}; + +#endif // __MEM_CACHE_PREFETCH_GHB_PREFETCHER_HH__ diff --git a/src/mem/cache/prefetch/stride_prefetcher.cc b/src/mem/cache/prefetch/stride_prefetcher.cc new file mode 100644 index 000000000..93a096468 --- /dev/null +++ b/src/mem/cache/prefetch/stride_prefetcher.cc @@ -0,0 +1,54 @@ +/* + * Copyright (c) 2005 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: Ron Dreslinski + * Steve Reinhardt + */ + +/** + * @file + * Stride Prefetcher template instantiations. + */ + +#include "mem/cache/tags/cache_tags.hh" + +#include "mem/cache/tags/lru.hh" + +#include "base/compression/null_compression.hh" + +#include "mem/cache/miss/miss_queue.hh" +#include "mem/cache/miss/blocking_buffer.hh" + +#include "mem/cache/prefetch/stride_prefetcher.hh" + +// Template Instantiations +#ifndef DOXYGEN_SHOULD_SKIP_THIS + +template class StridePrefetcher<CacheTags<LRU,NullCompression>, MissQueue>; +template class StridePrefetcher<CacheTags<LRU,NullCompression>, BlockingBuffer>; + +#endif //DOXYGEN_SHOULD_SKIP_THIS diff --git a/src/mem/cache/prefetch/stride_prefetcher.hh b/src/mem/cache/prefetch/stride_prefetcher.hh new file mode 100644 index 000000000..4a8ee7de4 --- /dev/null +++ b/src/mem/cache/prefetch/stride_prefetcher.hh @@ -0,0 +1,149 @@ +/* + * Copyright (c) 2005 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: Ron Dreslinski + */ + +/** + * @file + * Describes a strided prefetcher based on template policies. + */ + +#ifndef __MEM_CACHE_PREFETCH_STRIDE_PREFETCHER_HH__ +#define __MEM_CACHE_PREFETCH_STRIDE_PREFETCHER_HH__ + +#include "base/misc.hh" // fatal, panic, and warn + +#include "mem/cache/prefetch/prefetcher.hh" + +/** + * A template-policy based cache. The behavior of the cache can be altered by + * supplying different template policies. TagStore handles all tag and data + * storage @sa TagStore. Buffering handles all misses and writes/writebacks + * @sa MissQueue. Coherence handles all coherence policy details @sa + * UniCoherence, SimpleMultiCoherence. + */ +template <class TagStore, class Buffering> +class StridePrefetcher : public Prefetcher<TagStore, Buffering> +{ + protected: + + Buffering* mq; + TagStore* tags; + + class strideEntry + { + public: + Addr IAddr; + Addr MAddr; + int stride; + int64_t confidence; + +/* bool operator < (strideEntry a,strideEntry b) + { + if (a.confidence == b.confidence) { + return true; //?????? + } + else return a.confidence < b.confidence; + }*/ + }; + Addr* lastMissAddr[64/*MAX_CPUS*/]; + + std::list<strideEntry*> table[64/*MAX_CPUS*/]; + Tick latency; + int degree; + bool useCPUId; + + + public: + + StridePrefetcher(int size, bool pageStop, bool serialSquash, + bool cacheCheckPush, bool onlyData, + Tick latency, int degree, bool useCPUId) + :Prefetcher<TagStore, Buffering>(size, pageStop, serialSquash, + cacheCheckPush, onlyData), + latency(latency), degree(degree), useCPUId(useCPUId) + { + } + + ~StridePrefetcher() {} + + void calculatePrefetch(Packet * &pkt, std::list<Addr> &addresses, + std::list<Tick> &delays) + { +// Addr blkAddr = pkt->paddr & ~(Addr)(this->blkSize-1); + int cpuID = pkt->req->getCpuNum(); + if (!useCPUId) cpuID = 0; + + /* Scan Table for IAddr Match */ +/* std::list<strideEntry*>::iterator iter; + for (iter=table[cpuID].begin(); + iter !=table[cpuID].end(); + iter++) { + if ((*iter)->IAddr == pkt->pc) break; + } + + if (iter != table[cpuID].end()) { + //Hit in table + + int newStride = blkAddr - (*iter)->MAddr; + if (newStride == (*iter)->stride) { + (*iter)->confidence++; + } + else { + (*iter)->stride = newStride; + (*iter)->confidence--; + } + + (*iter)->MAddr = blkAddr; + + for (int d=1; d <= degree; d++) { + Addr newAddr = blkAddr + d * newStride; + if (this->pageStop && + (blkAddr & ~(TheISA::VMPageSize - 1)) != + (newAddr & ~(TheISA::VMPageSize - 1))) + { + //Spanned the page, so now stop + this->pfSpanPage += degree - d + 1; + return; + } + else + { + addresses.push_back(newAddr); + delays.push_back(latency); + } + } + } + else { + //Miss in table + //Find lowest confidence and replace + + } +*/ } +}; + +#endif // __MEM_CACHE_PREFETCH_STRIDE_PREFETCHER_HH__ diff --git a/src/mem/cache/prefetch/tagged_prefetcher.hh b/src/mem/cache/prefetch/tagged_prefetcher.hh new file mode 100644 index 000000000..17f500dd8 --- /dev/null +++ b/src/mem/cache/prefetch/tagged_prefetcher.hh @@ -0,0 +1,71 @@ +/* + * Copyright (c) 2005 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: Ron Dreslinski + */ + +/** + * @file + * Describes a tagged prefetcher based on template policies. + */ + +#ifndef __MEM_CACHE_PREFETCH_TAGGED_PREFETCHER_HH__ +#define __MEM_CACHE_PREFETCH_TAGGED_PREFETCHER_HH__ + +#include "mem/cache/prefetch/prefetcher.hh" + +/** + * A template-policy based cache. The behavior of the cache can be altered by + * supplying different template policies. TagStore handles all tag and data + * storage @sa TagStore. Buffering handles all misses and writes/writebacks + * @sa MissQueue. Coherence handles all coherence policy details @sa + * UniCoherence, SimpleMultiCoherence. + */ +template <class TagStore, class Buffering> +class TaggedPrefetcher : public Prefetcher<TagStore, Buffering> +{ + protected: + + Buffering* mq; + TagStore* tags; + + Tick latency; + int degree; + + public: + + TaggedPrefetcher(int size, bool pageStop, bool serialSquash, + bool cacheCheckPush, bool onlyData, + Tick latency, int degree); + + ~TaggedPrefetcher() {} + + void calculatePrefetch(Packet * &pkt, std::list<Addr> &addresses, + std::list<Tick> &delays); +}; + +#endif // __MEM_CACHE_PREFETCH_TAGGED_PREFETCHER_HH__ diff --git a/src/mem/cache/prefetch/tagged_prefetcher_impl.hh b/src/mem/cache/prefetch/tagged_prefetcher_impl.hh index 7bdabbe14..db5c94820 100644 --- a/src/mem/cache/prefetch/tagged_prefetcher_impl.hh +++ b/src/mem/cache/prefetch/tagged_prefetcher_impl.hh @@ -49,10 +49,10 @@ TaggedPrefetcher(int size, bool pageStop, bool serialSquash, template <class TagStore, class Buffering> void TaggedPrefetcher<TagStore, Buffering>:: -calculatePrefetch(MemReqPtr &req, std::list<Addr> &addresses, +calculatePrefetch(Packet * &pkt, std::list<Addr> &addresses, std::list<Tick> &delays) { - Addr blkAddr = req->paddr & ~(Addr)(this->blkSize-1); + Addr blkAddr = pkt->getAddr() & ~(Addr)(this->blkSize-1); for (int d=1; d <= degree; d++) { Addr newAddr = blkAddr + d*(this->blkSize); diff --git a/src/mem/cache/tags/base_tags.cc b/src/mem/cache/tags/base_tags.cc new file mode 100644 index 000000000..153737300 --- /dev/null +++ b/src/mem/cache/tags/base_tags.cc @@ -0,0 +1,91 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + * Ron Dreslinski + */ + +/** + * @file + * Definitions of BaseTags. + */ + +#include "mem/cache/tags/base_tags.hh" + +#include "mem/cache/base_cache.hh" +#include "cpu/smt.hh" //maxThreadsPerCPU +#include "sim/sim_exit.hh" + +using namespace std; + +void +BaseTags::setCache(BaseCache *_cache) +{ + cache = _cache; + objName = cache->name(); +} + +void +BaseTags::regStats(const string &name) +{ + using namespace Stats; + replacements + .init(maxThreadsPerCPU) + .name(name + ".replacements") + .desc("number of replacements") + .flags(total) + ; + + tagsInUse + .name(name + ".tagsinuse") + .desc("Cycle average of tags in use") + ; + + totalRefs + .name(name + ".total_refs") + .desc("Total number of references to valid blocks.") + ; + + sampledRefs + .name(name + ".sampled_refs") + .desc("Sample count of references to valid blocks.") + ; + + avgRefs + .name(name + ".avg_refs") + .desc("Average number of references to valid blocks.") + ; + + avgRefs = totalRefs/sampledRefs; + + warmupCycle + .name(name + ".warmup_cycle") + .desc("Cycle when the warmup percentage was hit.") + ; + + registerExitCallback(new BaseTagsCallback(this)); +} diff --git a/src/mem/cache/tags/base_tags.hh b/src/mem/cache/tags/base_tags.hh new file mode 100644 index 000000000..b7b0c7ef0 --- /dev/null +++ b/src/mem/cache/tags/base_tags.hh @@ -0,0 +1,143 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + * Ron Dreslinski + */ + +/** + * @file + * Declaration of a common base class for cache tagstore objects. + */ + +#ifndef __BASE_TAGS_HH__ +#define __BASE_TAGS_HH__ + +#include <string> +#include "base/statistics.hh" +#include "base/callback.hh" + +class BaseCache; + +/** + * A common base class of Cache tagstore objects. + */ +class BaseTags +{ + protected: + /** Pointer to the parent cache. */ + BaseCache *cache; + + /** Local copy of the parent cache name. Used for DPRINTF. */ + std::string objName; + + /** + * The number of tags that need to be touched to meet the warmup + * percentage. + */ + int warmupBound; + /** Marked true when the cache is warmed up. */ + bool warmedUp; + + // Statistics + /** + * @addtogroup CacheStatistics + * @{ + */ + + /** Number of replacements of valid blocks per thread. */ + Stats::Vector<> replacements; + /** Per cycle average of the number of tags that hold valid data. */ + Stats::Average<> tagsInUse; + + /** The total number of references to a block before it is replaced. */ + Stats::Scalar<> totalRefs; + + /** + * The number of reference counts sampled. This is different from + * replacements because we sample all the valid blocks when the simulator + * exits. + */ + Stats::Scalar<> sampledRefs; + + /** + * Average number of references to a block before is was replaced. + * @todo This should change to an average stat once we have them. + */ + Stats::Formula avgRefs; + + /** The cycle that the warmup percentage was hit. */ + Stats::Scalar<> warmupCycle; + /** + * @} + */ + + public: + + /** + * Destructor. + */ + virtual ~BaseTags() {} + + /** + * Set the parent cache back pointer. Also copies the cache name to + * objName. + * @param _cache Pointer to parent cache. + */ + void setCache(BaseCache *_cache); + + /** + * Return the parent cache name. + * @return the parent cache name. + */ + const std::string &name() const + { + return objName; + } + + /** + * Register local statistics. + * @param name The name to preceed each statistic name. + */ + void regStats(const std::string &name); + + /** + * Average in the reference count for valid blocks when the simulation + * exits. + */ + virtual void cleanupRefs() {} +}; + +class BaseTagsCallback : public Callback +{ + BaseTags *tags; + public: + BaseTagsCallback(BaseTags *t) : tags(t) {} + virtual void process() { tags->cleanupRefs(); }; +}; + +#endif //__BASE_TAGS_HH__ diff --git a/src/mem/cache/tags/fa_lru.cc b/src/mem/cache/tags/fa_lru.cc new file mode 100644 index 000000000..82d2c410d --- /dev/null +++ b/src/mem/cache/tags/fa_lru.cc @@ -0,0 +1,334 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** + * @file + * Definitions a fully associative LRU tagstore. + */ + +#include <sstream> + +#include <assert.h> + +#include "mem/cache/tags/fa_lru.hh" +#include "base/intmath.hh" +#include "base/misc.hh" + +using namespace std; + +FALRU::FALRU(int _blkSize, int _size, int hit_latency) + : blkSize(_blkSize), size(_size), + numBlks(size/blkSize), hitLatency(hit_latency) +{ + if (!isPowerOf2(blkSize)) + fatal("cache block size (in bytes) `%d' must be a power of two", + blkSize); + if (!(hitLatency > 0)) + fatal("Access latency in cycles must be at least one cycle"); + if (!isPowerOf2(size)) + fatal("Cache Size must be power of 2 for now"); + + // Track all cache sizes from 128K up by powers of 2 + numCaches = floorLog2(size) - 17; + if (numCaches >0){ + cacheBoundaries = new FALRUBlk *[numCaches]; + cacheMask = (1 << numCaches) - 1; + } else { + cacheMask = 0; + } + + warmedUp = false; + warmupBound = size/blkSize; + + blks = new FALRUBlk[numBlks]; + head = &(blks[0]); + tail = &(blks[numBlks-1]); + + head->prev = NULL; + head->next = &(blks[1]); + head->inCache = cacheMask; + + tail->prev = &(blks[numBlks-2]); + tail->next = NULL; + tail->inCache = 0; + + int index = (1 << 17) / blkSize; + int j = 0; + int flags = cacheMask; + for (int i = 1; i < numBlks-1; i++) { + blks[i].inCache = flags; + if (i == index - 1){ + cacheBoundaries[j] = &(blks[i]); + flags &= ~ (1<<j); + ++j; + index = index << 1; + } + blks[i].prev = &(blks[i-1]); + blks[i].next = &(blks[i+1]); + blks[i].isTouched = false; + } + assert(j == numCaches); + assert(index == numBlks); + //assert(check()); +} + +void +FALRU::regStats(const string &name) +{ + using namespace Stats; + BaseTags::regStats(name); + hits + .init(numCaches+1) + .name(name + ".falru_hits") + .desc("The number of hits in each cache size.") + ; + misses + .init(numCaches+1) + .name(name + ".falru_misses") + .desc("The number of misses in each cache size.") + ; + accesses + .name(name + ".falru_accesses") + .desc("The number of accesses to the FA LRU cache.") + ; + + for (int i = 0; i < numCaches+1; ++i) { + stringstream size_str; + if (i < 3){ + size_str << (1<<(i+7)) <<"K"; + } else { + size_str << (1<<(i-3)) <<"M"; + } + + hits.subname(i, size_str.str()); + hits.subdesc(i, "Hits in a " + size_str.str() +" cache"); + misses.subname(i, size_str.str()); + misses.subdesc(i, "Misses in a " + size_str.str() +" cache"); + } +} + +FALRUBlk * +FALRU::hashLookup(Addr addr) const +{ + tagIterator iter = tagHash.find(addr); + if (iter != tagHash.end()) { + return (*iter).second; + } + return NULL; +} + +bool +FALRU::probe(int asid, Addr addr) const +{ + Addr blkAddr = blkAlign(addr); + FALRUBlk* blk = hashLookup(blkAddr); + return blk && blk->tag == blkAddr && blk->isValid(); +} + +void +FALRU::invalidateBlk(int asid, Addr addr) +{ + Addr blkAddr = blkAlign(addr); + FALRUBlk* blk = (*tagHash.find(blkAddr)).second; + if (blk) { + assert(blk->tag == blkAddr); + blk->status = 0; + blk->isTouched = false; + tagsInUse--; + } +} + +FALRUBlk* +FALRU::findBlock(Addr addr, int asid, int &lat, int *inCache) +{ + accesses++; + int tmp_in_cache = 0; + Addr blkAddr = blkAlign(addr); + FALRUBlk* blk = hashLookup(blkAddr); + + if (blk && blk->isValid()) { + assert(blk->tag == blkAddr); + tmp_in_cache = blk->inCache; + for (int i = 0; i < numCaches; i++) { + if (1<<i & blk->inCache) { + hits[i]++; + } else { + misses[i]++; + } + } + hits[numCaches]++; + if (blk != head){ + moveToHead(blk); + } + } else { + blk = NULL; + for (int i = 0; i < numCaches+1; ++i) { + misses[i]++; + } + } + if (inCache) { + *inCache = tmp_in_cache; + } + + lat = hitLatency; + //assert(check()); + return blk; +} + +FALRUBlk* +FALRU::findBlock(Packet * &pkt, int &lat, int *inCache) +{ + Addr addr = pkt->getAddr(); + + accesses++; + int tmp_in_cache = 0; + Addr blkAddr = blkAlign(addr); + FALRUBlk* blk = hashLookup(blkAddr); + + if (blk && blk->isValid()) { + assert(blk->tag == blkAddr); + tmp_in_cache = blk->inCache; + for (int i = 0; i < numCaches; i++) { + if (1<<i & blk->inCache) { + hits[i]++; + } else { + misses[i]++; + } + } + hits[numCaches]++; + if (blk != head){ + moveToHead(blk); + } + } else { + blk = NULL; + for (int i = 0; i < numCaches+1; ++i) { + misses[i]++; + } + } + if (inCache) { + *inCache = tmp_in_cache; + } + + lat = hitLatency; + //assert(check()); + return blk; +} + +FALRUBlk* +FALRU::findBlock(Addr addr, int asid) const +{ + Addr blkAddr = blkAlign(addr); + FALRUBlk* blk = hashLookup(blkAddr); + + if (blk && blk->isValid()) { + assert(blk->tag == blkAddr); + } else { + blk = NULL; + } + return blk; +} + +FALRUBlk* +FALRU::findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks) +{ + FALRUBlk * blk = tail; + assert(blk->inCache == 0); + moveToHead(blk); + tagHash.erase(blk->tag); + tagHash[blkAlign(pkt->getAddr())] = blk; + if (blk->isValid()) { + replacements[0]++; + } else { + tagsInUse++; + blk->isTouched = true; + if (!warmedUp && tagsInUse.value() >= warmupBound) { + warmedUp = true; + warmupCycle = curTick; + } + } + //assert(check()); + return blk; +} + +void +FALRU::moveToHead(FALRUBlk *blk) +{ + int updateMask = blk->inCache ^ cacheMask; + for (int i = 0; i < numCaches; i++){ + if ((1<<i) & updateMask) { + cacheBoundaries[i]->inCache &= ~(1<<i); + cacheBoundaries[i] = cacheBoundaries[i]->prev; + } else if (cacheBoundaries[i] == blk) { + cacheBoundaries[i] = blk->prev; + } + } + blk->inCache = cacheMask; + if (blk != head) { + if (blk == tail){ + assert(blk->next == NULL); + tail = blk->prev; + tail->next = NULL; + } else { + blk->prev->next = blk->next; + blk->next->prev = blk->prev; + } + blk->next = head; + blk->prev = NULL; + head->prev = blk; + head = blk; + } +} + +bool +FALRU::check() +{ + FALRUBlk* blk = head; + int size = 0; + int boundary = 1<<17; + int j = 0; + int flags = cacheMask; + while (blk) { + size += blkSize; + if (blk->inCache != flags) { + return false; + } + if (size == boundary && blk != tail) { + if (cacheBoundaries[j] != blk) { + return false; + } + flags &=~(1 << j); + boundary = boundary<<1; + ++j; + } + blk = blk->next; + } + return true; +} diff --git a/src/mem/cache/tags/fa_lru.hh b/src/mem/cache/tags/fa_lru.hh new file mode 100644 index 000000000..566e36c27 --- /dev/null +++ b/src/mem/cache/tags/fa_lru.hh @@ -0,0 +1,346 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** + * @file + * Declaration of a fully associative LRU tag store. + */ + +#ifndef __FA_LRU_HH__ +#define __FA_LRU_HH__ + +#include <list> + +#include "mem/cache/cache_blk.hh" +#include "mem/packet.hh" +#include "base/hashmap.hh" +#include "mem/cache/tags/base_tags.hh" + +/** + * A fully associative cache block. + */ +class FALRUBlk : public CacheBlk +{ +public: + /** The previous block in LRU order. */ + FALRUBlk *prev; + /** The next block in LRU order. */ + FALRUBlk *next; + /** Has this block been touched? */ + bool isTouched; + + /** + * A bit mask of the sizes of cache that this block is resident in. + * Each bit represents a power of 2 in MB size cache. + * If bit 0 is set, this block is in a 1MB cache + * If bit 2 is set, this block is in a 4MB cache, etc. + * There is one bit for each cache smaller than the full size (default + * 16MB). + */ + int inCache; +}; + +/** + * A fully associative LRU cache. Keeps statistics for accesses to a number of + * cache sizes at once. + */ +class FALRU : public BaseTags +{ + public: + /** Typedef the block type used in this class. */ + typedef FALRUBlk BlkType; + /** Typedef a list of pointers to the local block type. */ + typedef std::list<FALRUBlk*> BlkList; + protected: + /** The block size of the cache. */ + const int blkSize; + /** The size of the cache. */ + const int size; + /** The number of blocks in the cache. */ + const int numBlks; // calculated internally + /** The hit latency of the cache. */ + const int hitLatency; + + /** Array of pointers to blocks at the cache size boundaries. */ + FALRUBlk **cacheBoundaries; + /** A mask for the FALRUBlk::inCache bits. */ + int cacheMask; + /** The number of different size caches being tracked. */ + int numCaches; + + /** The cache blocks. */ + FALRUBlk *blks; + + /** The MRU block. */ + FALRUBlk *head; + /** The LRU block. */ + FALRUBlk *tail; + + /** Hash table type mapping addresses to cache block pointers. */ + typedef m5::hash_map<Addr, FALRUBlk *, m5::hash<Addr> > hash_t; + /** Iterator into the address hash table. */ + typedef hash_t::const_iterator tagIterator; + + /** The address hash table. */ + hash_t tagHash; + + /** + * Find the cache block for the given address. + * @param addr The address to find. + * @return The cache block of the address, if any. + */ + FALRUBlk * hashLookup(Addr addr) const; + + /** + * Move a cache block to the MRU position. + * @param blk The block to promote. + */ + void moveToHead(FALRUBlk *blk); + + /** + * Check to make sure all the cache boundaries are still where they should + * be. Used for debugging. + * @return True if everything is correct. + */ + bool check(); + + /** + * @defgroup FALRUStats Fully Associative LRU specific statistics + * The FA lru stack lets us track multiple cache sizes at once. These + * statistics track the hits and misses for different cache sizes. + * @{ + */ + + /** Hits in each cache size >= 128K. */ + Stats::Vector<> hits; + /** Misses in each cache size >= 128K. */ + Stats::Vector<> misses; + /** Total number of accesses. */ + Stats::Scalar<> accesses; + + /** + * @} + */ + +public: + /** + * Construct and initialize this cache tagstore. + * @param blkSize The block size of the cache. + * @param size The size of the cache. + * @param hit_latency The hit latency of the cache. + */ + FALRU(int blkSize, int size, int hit_latency); + + /** + * Register the stats for this object. + * @param name The name to prepend to the stats name. + */ + void regStats(const std::string &name); + + /** + * Return true if the address is found in the cache. + * @param asid The address space ID. + * @param addr The address to look for. + * @return True if the address is in the cache. + */ + bool probe(int asid, Addr addr) const; + + /** + * Invalidate the cache block that contains the given addr. + * @param asid The address space ID. + * @param addr The address to invalidate. + */ + void invalidateBlk(int asid, Addr addr); + + /** + * Find the block in the cache and update the replacement data. Returns + * the access latency and the in cache flags as a side effect + * @param addr The address to look for. + * @param asid The address space ID. + * @param lat The latency of the access. + * @param inCache The FALRUBlk::inCache flags. + * @return Pointer to the cache block. + */ + FALRUBlk* findBlock(Addr addr, int asid, int &lat, int *inCache = 0); + + /** + * Find the block in the cache and update the replacement data. Returns + * the access latency and the in cache flags as a side effect + * @param req The req whose block to find + * @param lat The latency of the access. + * @param inCache The FALRUBlk::inCache flags. + * @return Pointer to the cache block. + */ + FALRUBlk* findBlock(Packet * &pkt, int &lat, int *inCache = 0); + + /** + * Find the block in the cache, do not update the replacement data. + * @param addr The address to look for. + * @param asid The address space ID. + * @return Pointer to the cache block. + */ + FALRUBlk* findBlock(Addr addr, int asid) const; + + /** + * Find a replacement block for the address provided. + * @param req The request to a find a replacement candidate for. + * @param writebacks List for any writebacks to be performed. + * @param compress_blocks List of blocks to compress, for adaptive comp. + * @return The block to place the replacement in. + */ + FALRUBlk* findReplacement(Packet * &pkt, PacketList & writebacks, + BlkList &compress_blocks); + + /** + * Return the hit latency of this cache. + * @return The hit latency. + */ + int getHitLatency() const + { + return hitLatency; + } + + /** + * Return the block size of this cache. + * @return The block size. + */ + int getBlockSize() + { + return blkSize; + } + + /** + * Return the subblock size of this cache, always the block size. + * @return The block size. + */ + int getSubBlockSize() + { + return blkSize; + } + + /** + * Align an address to the block size. + * @param addr the address to align. + * @return The aligned address. + */ + Addr blkAlign(Addr addr) const + { + return (addr & ~(Addr)(blkSize-1)); + } + + /** + * Generate the tag from the addres. For fully associative this is just the + * block address. + * @param addr The address to get the tag from. + * @param blk ignored here + * @return The tag. + */ + Addr extractTag(Addr addr, FALRUBlk *blk) const + { + return blkAlign(addr); + } + + /** + * Return the set of an address. Only one set in a fully associative cache. + * @param addr The address to get the set from. + * @return 0. + */ + int extractSet(Addr addr) const + { + return 0; + } + + /** + * Calculate the block offset of an address. + * @param addr the address to get the offset of. + * @return the block offset. + */ + int extractBlkOffset(Addr addr) const + { + return (addr & (Addr)(blkSize-1)); + } + + /** + * Regenerate the block address from the tag and the set. + * @param tag The tag of the block. + * @param set The set the block belongs to. + * @return the block address. + */ + Addr regenerateBlkAddr(Addr tag, int set) const + { + return (tag); + } + + /** + * Read the data out of the internal storage of a cache block. FALRU + * currently doesn't support data storage. + * @param blk The cache block to read. + * @param data The buffer to read the data into. + * @return The data from the cache block. + */ + void readData(FALRUBlk *blk, uint8_t *data) + { + } + + /** + * Write data into the internal storage of a cache block. FALRU + * currently doesn't support data storage. + * @param blk The cache block to be written. + * @param data The data to write. + * @param size The number of bytes to write. + * @param writebacks A list for any writebacks to be performed. May be + * needed when writing to a compressed block. + */ + void writeData(FALRUBlk *blk, uint8_t *data, int size, + PacketList &writebacks) + { + } + + /** + * Unimplemented. Perform a cache block copy from block aligned addresses. + * @param source The block aligned source address. + * @param dest The block aligned destination adddress. + * @param asid The address space ID. + * @param writebacks List for any generated writeback requests. + */ + void doCopy(Addr source, Addr dest, int asid, PacketList &writebacks) + { + } + + /** + * Unimplemented. + */ + void fixCopy(Packet * &pkt, PacketList &writebacks) + { + } + +}; + +#endif diff --git a/src/mem/cache/tags/iic.cc b/src/mem/cache/tags/iic.cc new file mode 100644 index 000000000..0071ca283 --- /dev/null +++ b/src/mem/cache/tags/iic.cc @@ -0,0 +1,879 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + */ + +/** + * @file + * Definitions of the Indirect Index Cache tagstore. + */ + +#include <algorithm> +#include <string> +#include <vector> + +#include <math.h> + +#include "mem/cache/base_cache.hh" +#include "mem/cache/tags/iic.hh" +#include "base/intmath.hh" +#include "sim/root.hh" // for curTick + +#include "base/trace.hh" // for DPRINTF + + +using namespace std; + +/** Track the number of accesses to each cache set. */ +#define PROFILE_IIC 1 + +IIC::IIC(IIC::Params ¶ms) : + hashSets(params.numSets), blkSize(params.blkSize), assoc(params.assoc), + hitLatency(params.hitLatency), subSize(params.subblockSize), + numSub(blkSize/subSize), + trivialSize((floorLog2(params.size/subSize)*numSub)/8), + tagShift(floorLog2(blkSize)), blkMask(blkSize - 1), + subShift(floorLog2(subSize)), subMask(numSub - 1), + hashDelay(params.hashDelay), + numBlocks(params.size/subSize), + numTags(hashSets * assoc + params.size/blkSize -1), + numSecondary(params.size/blkSize), + tagNull(numTags), + primaryBound(hashSets * assoc) +{ + int i; + + // Check parameters + if (blkSize < 4 || !isPowerOf2(blkSize)) { + fatal("Block size must be at least 4 and a power of 2"); + } + if (hashSets <= 0 || !isPowerOf2(hashSets)) { + fatal("# of hashsets must be non-zero and a power of 2"); + } + if (assoc <= 0) { + fatal("associativity must be greater than zero"); + } + if (hitLatency <= 0) { + fatal("access latency must be greater than zero"); + } + if (numSub*subSize != blkSize) { + fatal("blocksize must be evenly divisible by subblock size"); + } + + // debug stuff + freeSecond = numSecondary; + + warmedUp = false; + warmupBound = params.size/blkSize; + + // Replacement Policy Initialization + repl = params.rp; + repl->setIIC(this); + + //last_miss_time = 0 + + // allocate data reference counters + dataReferenceCount = new int[numBlocks]; + memset(dataReferenceCount, 0, numBlocks*sizeof(int)); + + // Allocate storage for both internal data and block fast access data. + // We allocate it as one large chunk to reduce overhead and to make + // deletion easier. + int data_index = 0; + dataStore = new uint8_t[(numBlocks + numTags) * blkSize]; + dataBlks = new uint8_t*[numBlocks]; + for (i = 0; i < numBlocks; ++i) { + dataBlks[i] = &dataStore[data_index]; + freeDataBlock(i); + data_index += subSize; + } + + assert(data_index == numBlocks * subSize); + + // allocate and init tag store + tagStore = new IICTag[numTags]; + + int blkIndex = 0; + // allocate and init sets + sets = new IICSet[hashSets]; + for (i = 0; i < hashSets; ++i) { + sets[i].assoc = assoc; + sets[i].tags = new IICTag*[assoc]; + sets[i].chain_ptr = tagNull; + + for (int j = 0; j < assoc; ++j) { + IICTag *tag = &tagStore[blkIndex++]; + tag->chain_ptr = tagNull; + tag->data_ptr.resize(numSub); + tag->size = blkSize; + tag->trivialData = new uint8_t[trivialSize]; + tag->numData = 0; + sets[i].tags[j] = tag; + tag->set = i; + tag->data = &dataStore[data_index]; + data_index += blkSize; + } + } + + assert(blkIndex == primaryBound); + + for (i = primaryBound; i < tagNull; i++) { + tagStore[i].chain_ptr = i+1; + //setup data ptrs to subblocks + tagStore[i].data_ptr.resize(numSub); + tagStore[i].size = blkSize; + tagStore[i].trivialData = new uint8_t[trivialSize]; + tagStore[i].numData = 0; + tagStore[i].set = 0; + tagStore[i].data = &dataStore[data_index]; + data_index += blkSize; + } + freelist = primaryBound; +} + +IIC::~IIC() +{ + delete [] dataReferenceCount; + delete [] dataStore; + delete [] tagStore; + delete [] sets; +} + +/* register cache stats */ +void +IIC::regStats(const string &name) +{ + using namespace Stats; + + BaseTags::regStats(name); + + hitHashDepth.init(0, 20, 1); + missHashDepth.init(0, 20, 1); + setAccess.init(0, hashSets, 1); + + /** IIC Statistics */ + hitHashDepth + .name(name + ".hit_hash_depth_dist") + .desc("Dist. of Hash lookup depths") + .flags(pdf) + ; + + missHashDepth + .name(name + ".miss_hash_depth_dist") + .desc("Dist. of Hash lookup depths") + .flags(pdf) + ; + + repl->regStats(name); + + if (PROFILE_IIC) + setAccess + .name(name + ".set_access_dist") + .desc("Dist. of Accesses across sets") + .flags(pdf) + ; + + missDepthTotal + .name(name + ".miss_depth_total") + .desc("Total of miss depths") + ; + + hashMiss + .name(name + ".hash_miss") + .desc("Total of misses in hash table") + ; + + hitDepthTotal + .name(name + ".hit_depth_total") + .desc("Total of hit depths") + ; + + hashHit + .name(name + ".hash_hit") + .desc("Total of hites in hash table") + ; +} + +// probe cache for presence of given block. +bool +IIC::probe(int asid, Addr addr) const +{ + return (findBlock(addr,asid) != NULL); +} + +IICTag* +IIC::findBlock(Addr addr, int asid, int &lat) +{ + Addr tag = extractTag(addr); + unsigned set = hash(addr); + int set_lat; + + unsigned long chain_ptr; + + if (PROFILE_IIC) + setAccess.sample(set); + + IICTag *tag_ptr = sets[set].findTag(asid, tag, chain_ptr); + set_lat = 1; + if (tag_ptr == NULL && chain_ptr != tagNull) { + int secondary_depth; + tag_ptr = secondaryChain(asid, tag, chain_ptr, &secondary_depth); + set_lat += secondary_depth; + // set depth for statistics fix this later!!! egh + sets[set].depth = set_lat; + + if (tag_ptr != NULL) { + /* need to move tag into primary table */ + // need to preserve chain: fix this egh + sets[set].tags[assoc-1]->chain_ptr = tag_ptr->chain_ptr; + tagSwap(tag_ptr - tagStore, sets[set].tags[assoc-1] - tagStore); + tag_ptr = sets[set].findTag(asid, tag, chain_ptr); + assert(tag_ptr!=NULL); + } + + } + set_lat = set_lat * hashDelay + hitLatency; + if (tag_ptr != NULL) { + // IIC replacement: if this is not the first element of + // list, reorder + sets[set].moveToHead(tag_ptr); + + hitHashDepth.sample(sets[set].depth); + hashHit++; + hitDepthTotal += sets[set].depth; + tag_ptr->status |= BlkReferenced; + lat = set_lat; + if (tag_ptr->whenReady > curTick && tag_ptr->whenReady - curTick > set_lat) { + lat = tag_ptr->whenReady - curTick; + } + + tag_ptr->refCount += 1; + } + else { + // fall through: cache block not found, not a hit... + missHashDepth.sample(sets[set].depth); + hashMiss++; + missDepthTotal += sets[set].depth; + lat = set_lat; + } + return tag_ptr; +} + +IICTag* +IIC::findBlock(Packet * &pkt, int &lat) +{ + Addr addr = pkt->getAddr(); + int asid = pkt->req->getAsid(); + + Addr tag = extractTag(addr); + unsigned set = hash(addr); + int set_lat; + + unsigned long chain_ptr; + + if (PROFILE_IIC) + setAccess.sample(set); + + IICTag *tag_ptr = sets[set].findTag(asid, tag, chain_ptr); + set_lat = 1; + if (tag_ptr == NULL && chain_ptr != tagNull) { + int secondary_depth; + tag_ptr = secondaryChain(asid, tag, chain_ptr, &secondary_depth); + set_lat += secondary_depth; + // set depth for statistics fix this later!!! egh + sets[set].depth = set_lat; + + if (tag_ptr != NULL) { + /* need to move tag into primary table */ + // need to preserve chain: fix this egh + sets[set].tags[assoc-1]->chain_ptr = tag_ptr->chain_ptr; + tagSwap(tag_ptr - tagStore, sets[set].tags[assoc-1] - tagStore); + tag_ptr = sets[set].findTag(asid, tag, chain_ptr); + assert(tag_ptr!=NULL); + } + + } + set_lat = set_lat * hashDelay + hitLatency; + if (tag_ptr != NULL) { + // IIC replacement: if this is not the first element of + // list, reorder + sets[set].moveToHead(tag_ptr); + + hitHashDepth.sample(sets[set].depth); + hashHit++; + hitDepthTotal += sets[set].depth; + tag_ptr->status |= BlkReferenced; + lat = set_lat; + if (tag_ptr->whenReady > curTick && tag_ptr->whenReady - curTick > set_lat) { + lat = tag_ptr->whenReady - curTick; + } + + tag_ptr->refCount += 1; + } + else { + // fall through: cache block not found, not a hit... + missHashDepth.sample(sets[set].depth); + hashMiss++; + missDepthTotal += sets[set].depth; + lat = set_lat; + } + return tag_ptr; +} + +IICTag* +IIC::findBlock(Addr addr, int asid) const +{ + Addr tag = extractTag(addr); + unsigned set = hash(addr); + + unsigned long chain_ptr; + + IICTag *tag_ptr = sets[set].findTag(asid, tag, chain_ptr); + if (tag_ptr == NULL && chain_ptr != tagNull) { + int secondary_depth; + tag_ptr = secondaryChain(asid, tag, chain_ptr, &secondary_depth); + } + return tag_ptr; +} + + +IICTag* +IIC::findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks) +{ + DPRINTF(IIC, "Finding Replacement for %x\n", pkt->getAddr()); + unsigned set = hash(pkt->getAddr()); + IICTag *tag_ptr; + unsigned long *tmp_data = new unsigned long[numSub]; + + // Get a enough subblocks for a full cache line + for (int i = 0; i < numSub; ++i){ + tmp_data[i] = getFreeDataBlock(writebacks); + assert(dataReferenceCount[tmp_data[i]]==0); + } + + tag_ptr = getFreeTag(set, writebacks); + + tag_ptr->set = set; + for (int i=0; i< numSub; ++i) { + tag_ptr->data_ptr[i] = tmp_data[i]; + dataReferenceCount[tag_ptr->data_ptr[i]]++; + } + tag_ptr->numData = numSub; + assert(tag_ptr - tagStore < primaryBound); // make sure it is in primary + tag_ptr->chain_ptr = tagNull; + sets[set].moveToHead(tag_ptr); + delete [] tmp_data; + + list<unsigned long> tag_indexes; + repl->doAdvance(tag_indexes); + while (!tag_indexes.empty()) { + if (!tagStore[tag_indexes.front()].isCompressed()) { + compress_blocks.push_back(&tagStore[tag_indexes.front()]); + } + tag_indexes.pop_front(); + } + + tag_ptr->re = (void*)repl->add(tag_ptr-tagStore); + + return tag_ptr; +} + +void +IIC::freeReplacementBlock(PacketList & writebacks) +{ + IICTag *tag_ptr; + unsigned long data_ptr; + /* consult replacement policy */ + tag_ptr = &tagStore[repl->getRepl()]; + assert(tag_ptr->isValid()); + + DPRINTF(Cache, "Replacing %x in IIC: %s\n", + regenerateBlkAddr(tag_ptr->tag,0), + tag_ptr->isModified() ? "writeback" : "clean"); + /* write back replaced block data */ + if (tag_ptr && (tag_ptr->isValid())) { + replacements[0]++; + totalRefs += tag_ptr->refCount; + ++sampledRefs; + tag_ptr->refCount = 0; + + if (tag_ptr->isModified()) { +/* Packet * writeback = + buildWritebackReq(regenerateBlkAddr(tag_ptr->tag, 0), + tag_ptr->req->asid, tag_ptr->xc, blkSize, + tag_ptr->data, + tag_ptr->size); +*/ + Request *writebackReq = new Request(regenerateBlkAddr(tag_ptr->tag, 0), + blkSize, 0); + Packet *writeback = new Packet(writebackReq, Packet::Writeback, -1); + writeback->dataDynamic<uint8_t>(tag_ptr->data); + + writebacks.push_back(writeback); + } + } + + // free the data blocks + for (int i = 0; i < tag_ptr->numData; ++i) { + data_ptr = tag_ptr->data_ptr[i]; + assert(dataReferenceCount[data_ptr]>0); + if (--dataReferenceCount[data_ptr] == 0) { + freeDataBlock(data_ptr); + } + } + freeTag(tag_ptr); +} + +unsigned long +IIC::getFreeDataBlock(PacketList & writebacks) +{ + struct IICTag *tag_ptr; + unsigned long data_ptr; + + tag_ptr = NULL; + /* find data block */ + while (blkFreelist.empty()) { + freeReplacementBlock(writebacks); + } + + data_ptr = blkFreelist.front(); + blkFreelist.pop_front(); + DPRINTF(IICMore,"Found free data at %d\n",data_ptr); + return data_ptr; +} + + + +IICTag* +IIC::getFreeTag(int set, PacketList & writebacks) +{ + unsigned long tag_index; + IICTag *tag_ptr; + // Add new tag + tag_ptr = sets[set].findFree(); + // if no free in primary, and secondary exists + if (!tag_ptr && numSecondary) { + // need to spill a tag into secondary storage + while (freelist == tagNull) { + // get replacements until one is in secondary + freeReplacementBlock(writebacks); + } + + tag_index = freelist; + freelist = tagStore[freelist].chain_ptr; + freeSecond--; + + assert(tag_index != tagNull); + tagSwap(tag_index, sets[set].tags[assoc-1] - tagStore); + tagStore[tag_index].chain_ptr = sets[set].chain_ptr; + sets[set].chain_ptr = tag_index; + + tag_ptr = sets[set].tags[assoc-1]; + } + DPRINTF(IICMore,"Found free tag at %d\n",tag_ptr - tagStore); + tagsInUse++; + if (!warmedUp && tagsInUse.value() >= warmupBound) { + warmedUp = true; + warmupCycle = curTick; + } + + return tag_ptr; +} + +void +IIC::freeTag(IICTag *tag_ptr) +{ + unsigned long tag_index, tmp_index; + // Fix tag_ptr + if (tag_ptr) { + // we have a tag to clear + DPRINTF(IICMore,"Freeing Tag for %x\n", + regenerateBlkAddr(tag_ptr->tag,0)); + tagsInUse--; + tag_ptr->status = 0; + tag_ptr->numData = 0; + tag_ptr->re = NULL; + tag_index = tag_ptr - tagStore; + if (tag_index >= primaryBound) { + // tag_ptr points to secondary store + assert(tag_index < tagNull); // remove this?? egh + if (tag_ptr->chain_ptr == tagNull) { + // need to fix chain list + unsigned tmp_set = hash(tag_ptr->tag << tagShift); + if (sets[tmp_set].chain_ptr == tag_index) { + sets[tmp_set].chain_ptr = tagNull; + } else { + tmp_index = sets[tmp_set].chain_ptr; + while (tmp_index != tagNull + && tagStore[tmp_index].chain_ptr != tag_index) { + tmp_index = tagStore[tmp_index].chain_ptr; + } + assert(tmp_index != tagNull); + tagStore[tmp_index].chain_ptr = tagNull; + } + tag_ptr->chain_ptr = freelist; + freelist = tag_index; + freeSecond++; + } else { + // copy next chained entry to this tag location + tmp_index = tag_ptr->chain_ptr; + tagSwap(tmp_index, tag_index); + tagStore[tmp_index].chain_ptr = freelist; + freelist = tmp_index; + freeSecond++; + } + } else { + // tag_ptr in primary hash table + assert(tag_index < primaryBound); + tag_ptr->status = 0; + unsigned tmp_set = hash(tag_ptr->tag << tagShift); + if (sets[tmp_set].chain_ptr != tagNull) { // collapse chain + tmp_index = sets[tmp_set].chain_ptr; + tagSwap(tag_index, tmp_index); + tagStore[tmp_index].chain_ptr = freelist; + freelist = tmp_index; + freeSecond++; + sets[tmp_set].chain_ptr = tag_ptr->chain_ptr; + sets[tmp_set].moveToTail(tag_ptr); + } + } + } +} + +void +IIC::freeDataBlock(unsigned long data_ptr) +{ + assert(dataReferenceCount[data_ptr] == 0); + DPRINTF(IICMore, "Freeing data at %d\n", data_ptr); + blkFreelist.push_front(data_ptr); +} + +/** Use a simple modulo hash. */ +#define SIMPLE_HASH 0 + +unsigned +IIC::hash(Addr addr) const { +#if SIMPLE_HASH + return extractTag(addr) % iic_hash_size; +#else + Addr tag, mask, x, y; + tag = extractTag(addr); + mask = hashSets-1; /* assumes iic_hash_size is a power of 2 */ + x = tag & mask; + y = (tag >> (int)(::log(hashSets)/::log(2))) & mask; + assert (x < hashSets && y < hashSets); + return x ^ y; +#endif +} + + +void +IICSet::moveToHead(IICTag *tag) +{ + if (tags[0] == tag) + return; + + // write 'next' block into blks[i], moving up from MRU toward LRU + // until we overwrite the block we moved to head. + + // start by setting up to write 'blk' into blks[0] + int i = 0; + IICTag *next = tag; + + do { + assert(i < assoc); + // swap blks[i] and next + IICTag *tmp = tags[i]; + tags[i] = next; + next = tmp; + ++i; + } while (next != tag); +} + +void +IICSet::moveToTail(IICTag *tag) +{ + if (tags[assoc-1] == tag) + return; + + // write 'next' block into blks[i], moving up from MRU toward LRU + // until we overwrite the block we moved to head. + + // start by setting up to write 'blk' into blks[0] + int i = assoc - 1; + IICTag *next = tag; + + do { + assert(i >= 0); + // swap blks[i] and next + IICTag *tmp = tags[i]; + tags[i] = next; + next = tmp; + --i; + } while (next != tag); +} + +void +IIC::tagSwap(unsigned long index1, unsigned long index2) +{ + DPRINTF(IIC,"Swapping tag[%d]=%x for tag[%d]=%x\n",index1, + tagStore[index1].tag<<tagShift, index2, + tagStore[index2].tag<<tagShift); + IICTag tmp_tag; + tmp_tag = tagStore[index1]; + tagStore[index1] = tagStore[index2]; + tagStore[index2] = tmp_tag; + if (tagStore[index1].isValid()) + repl->fixTag(tagStore[index1].re, index2, index1); + if (tagStore[index2].isValid()) + repl->fixTag(tagStore[index2].re, index1, index2); +} + + +IICTag * +IIC::secondaryChain(int asid, Addr tag, unsigned long chain_ptr, + int *_depth) const +{ + int depth = 0; + while (chain_ptr != tagNull) { + DPRINTF(IIC,"Searching secondary at %d for %x\n", chain_ptr, + tag<<tagShift); + if (tagStore[chain_ptr].tag == tag && + tagStore[chain_ptr].asid == asid && + (tagStore[chain_ptr].isValid())) { + *_depth = depth; + return &tagStore[chain_ptr]; + } + depth++; + chain_ptr = tagStore[chain_ptr].chain_ptr; + } + *_depth = depth; + return NULL; +} + +void +IIC::decompressBlock(unsigned long index) +{ + IICTag *tag_ptr = &tagStore[index]; + if (tag_ptr->isCompressed()) { + // decompress the data here. + } +} + +void +IIC::compressBlock(unsigned long index) +{ + IICTag *tag_ptr = &tagStore[index]; + if (!tag_ptr->isCompressed()) { + // Compress the data here. + } +} + +void +IIC::invalidateBlk(int asid, Addr addr) +{ + IICTag* tag_ptr = findBlock(addr, asid); + if (tag_ptr) { + for (int i = 0; i < tag_ptr->numData; ++i) { + dataReferenceCount[tag_ptr->data_ptr[i]]--; + if (dataReferenceCount[tag_ptr->data_ptr[i]] == 0) { + freeDataBlock(tag_ptr->data_ptr[i]); + } + } + repl->removeEntry(tag_ptr->re); + freeTag(tag_ptr); + } +} + +void +IIC::readData(IICTag *blk, uint8_t *data){ +// assert(cache->doData()); + assert(blk->size <= trivialSize || blk->numData > 0); + int data_size = blk->size; + if (data_size > trivialSize) { + for (int i = 0; i < blk->numData; ++i){ + memcpy(data+i*subSize, + &(dataBlks[blk->data_ptr[i]][0]), + (data_size>subSize)?subSize:data_size); + data_size -= subSize; + } + } else { + memcpy(data,blk->trivialData,data_size); + } +} + +void +IIC::writeData(IICTag *blk, uint8_t *write_data, int size, + PacketList & writebacks){ +// assert(cache->doData()); + assert(size < blkSize || !blk->isCompressed()); + DPRINTF(IIC, "Writing %d bytes to %x\n", size, + blk->tag<<tagShift); + // Find the number of subblocks needed, (round up) + int num_subs = (size + (subSize -1))/subSize; + if (size <= trivialSize) { + num_subs = 0; + } + assert(num_subs <= numSub); + if (num_subs > blk->numData) { + // need to allocate more data blocks + for (int i = blk->numData; i < num_subs; ++i){ + blk->data_ptr[i] = getFreeDataBlock(writebacks); + dataReferenceCount[blk->data_ptr[i]] += 1; + } + } else if (num_subs < blk->numData){ + // can free data blocks + for (int i=num_subs; i < blk->numData; ++i){ + // decrement reference count and compare to zero + /** + * @todo + * Make this work with copying. + */ + if (--dataReferenceCount[blk->data_ptr[i]] == 0) { + freeDataBlock(blk->data_ptr[i]); + } + } + } + + blk->numData = num_subs; + blk->size = size; + assert(size <= trivialSize || blk->numData > 0); + if (size > trivialSize){ + for (int i = 0; i < blk->numData; ++i){ + memcpy(&dataBlks[blk->data_ptr[i]][0], write_data + i*subSize, + (size>subSize)?subSize:size); + size -= subSize; + } + } else { + memcpy(blk->trivialData,write_data,size); + } +} + + +/** + * @todo This code can break if the src is evicted to get a tag for the dest. + */ +void +IIC::doCopy(Addr source, Addr dest, int asid, PacketList &writebacks) +{ +//Copy unsuported now +#if 0 + IICTag *dest_tag = findBlock(dest, asid); + + if (dest_tag) { + for (int i = 0; i < dest_tag->numData; ++i) { + if (--dataReferenceCount[dest_tag->data_ptr[i]] == 0) { + freeDataBlock(dest_tag->data_ptr[i]); + } + } + // Reset replacement entry + } else { + dest_tag = getFreeTag(hash(dest), writebacks); + dest_tag->re = (void*) repl->add(dest_tag - tagStore); + dest_tag->set = hash(dest); + dest_tag->tag = extractTag(dest); + dest_tag->asid = asid; + dest_tag->status = BlkValid | BlkWritable; + } + // Find the source tag here since it might move if we need to find a + // tag for the destination. + IICTag *src_tag = findBlock(source, asid); + assert(src_tag); + assert(!cache->doData() || src_tag->size <= trivialSize + || src_tag->numData > 0); + // point dest to source data and inc counter + for (int i = 0; i < src_tag->numData; ++i) { + dest_tag->data_ptr[i] = src_tag->data_ptr[i]; + ++dataReferenceCount[dest_tag->data_ptr[i]]; + } + + // Maintain fast access data. + memcpy(dest_tag->data, src_tag->data, blkSize); + + dest_tag->xc = src_tag->xc; + dest_tag->size = src_tag->size; + dest_tag->numData = src_tag->numData; + if (src_tag->numData == 0) { + // Data is stored in the trivial data, just copy it. + memcpy(dest_tag->trivialData, src_tag->trivialData, src_tag->size); + } + + dest_tag->status |= BlkDirty; + if (dest_tag->size < blkSize) { + dest_tag->status |= BlkCompressed; + } else { + dest_tag->status &= ~BlkCompressed; + } +#endif +} + +void +IIC::fixCopy(Packet * &pkt, PacketList &writebacks) +{ +#if 0 + // if reference counter is greater than 1, do copy + // else do write + Addr blk_addr = blkAlign(pkt->getAddr); + IICTag* blk = findBlock(blk_addr, pkt->req->getAsid()); + + if (blk->numData > 0 && dataReferenceCount[blk->data_ptr[0]] != 1) { + // copy the data + // Mark the block as referenced so it doesn't get replaced. + blk->status |= BlkReferenced; + for (int i = 0; i < blk->numData; ++i){ + unsigned long new_data = getFreeDataBlock(writebacks); + // Need to refresh pointer + /** + * @todo Remove this refetch once we change IIC to pointer based + */ + blk = findBlock(blk_addr, pkt->req->getAsid()); + assert(blk); + if (cache->doData()) { + memcpy(&(dataBlks[new_data][0]), + &(dataBlks[blk->data_ptr[i]][0]), + subSize); + } + dataReferenceCount[blk->data_ptr[i]]--; + dataReferenceCount[new_data]++; + blk->data_ptr[i] = new_data; + } + } +#endif +} + +void +IIC::cleanupRefs() +{ + for (int i = 0; i < numTags; ++i) { + if (tagStore[i].isValid()) { + totalRefs += tagStore[i].refCount; + ++sampledRefs; + } + } +} diff --git a/src/mem/cache/tags/iic.hh b/src/mem/cache/tags/iic.hh new file mode 100644 index 000000000..6628f7e7a --- /dev/null +++ b/src/mem/cache/tags/iic.hh @@ -0,0 +1,574 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + */ + +/** + * @file + * Declaration of the Indirect Index Cache (IIC) tags store. + */ + +#ifndef __IIC_HH__ +#define __IIC_HH__ + +#include <list> +#include <vector> + +#include "mem/cache/cache_blk.hh" +#include "mem/cache/tags/repl/repl.hh" +#include "mem/packet.hh" +#include "base/statistics.hh" +#include "mem/cache/tags/base_tags.hh" + +class BaseCache; // Forward declaration + +/** + * IIC cache blk. + */ +class IICTag : public CacheBlk +{ + public: + /** + * Copy the contents of the given IICTag into this one. + * @param rhs The tag to copy. + * @return const reference to this tag. + */ + const IICTag& operator=(const IICTag& rhs) + { + CacheBlk::operator=(rhs); + chain_ptr = rhs.chain_ptr; + re = rhs.re; + set = rhs.set; + trivialData = rhs.trivialData; + numData = rhs.numData; + data_ptr.clear(); + for (int i = 0; i < rhs.numData; ++i) { + data_ptr.push_back(rhs.data_ptr[i]); + } + return *this; + } + + /** Hash chain pointer into secondary store. */ + unsigned long chain_ptr; + /** Data array pointers for each subblock. */ + std::vector<unsigned long> data_ptr; + /** Replacement Entry pointer. */ + void *re; + /** + * An array to store small compressed data. Conceputally the same size + * as the unsused data array pointers. + */ + uint8_t *trivialData; + /** + * The number of allocated subblocks. + */ + int numData; +}; + +/** + * A hash set for the IIC primary lookup table. + */ +class IICSet{ + public: + /** The associativity of the primary table. */ + int assoc; + + /** The number of hash chains followed when finding the last block. */ + int depth; + /** The current number of blocks on the chain. */ + int size; + + /** Tag pointer into the secondary tag storage. */ + unsigned long chain_ptr; + + /** The LRU list of the primary table. MRU is at 0 index. */ + IICTag ** tags; + + /** + * Find the addr in this set, return the chain pointer to the secondary if + * it isn't found. + * @param asid The address space ID. + * @param tag The address to find. + * @param chain_ptr The chain pointer to start the search of the secondary + * @return Pointer to the tag, NULL if not found. + */ + IICTag* findTag(int asid, Addr tag, unsigned long &chain_ptr) + { + depth = 1; + for (int i = 0; i < assoc; ++i) { + if (tags[i]->tag == tag && tags[i]->isValid()) { + return tags[i]; + } + } + chain_ptr = this->chain_ptr; + return 0; + } + + /** + * Find an usused tag in this set. + * @return Pointer to the unused tag, NULL if none are free. + */ + IICTag* findFree() + { + for (int i = 0; i < assoc; ++i) { + if (!tags[i]->isValid()) { + return tags[i]; + } + } + return 0; + } + + /** + * Move a tag to the head of the LRU list + * @param tag The tag to move. + */ + void moveToHead(IICTag *tag); + + /** + * Move a tag to the tail (LRU) of the LRU list + * @param tag The tag to move. + */ + void moveToTail(IICTag *tag); +}; + +/** + * The IIC tag store. This is a hardware-realizable, fully-associative tag + * store that uses software replacement, e.g. Gen. + */ +class IIC : public BaseTags +{ + public: + /** Typedef of the block type used in this class. */ + typedef IICTag BlkType; + /** Typedef for list of pointers to the local block type. */ + typedef std::list<IICTag*> BlkList; + protected: + /** The number of set in the primary table. */ + const int hashSets; + /** The block size in bytes. */ + const int blkSize; + /** The associativity of the primary table. */ + const int assoc; + /** The base hit latency. */ + const int hitLatency; + /** The subblock size, used for compression. */ + const int subSize; + + /** The number of subblocks */ + const int numSub; + /** The number of bytes used by data pointers */ + const int trivialSize; + + /** The amount to shift address to get the tag. */ + const int tagShift; + /** The mask to get block offset bits. */ + const unsigned blkMask; + + /** The amount to shift to get the subblock number. */ + const int subShift; + /** The mask to get the correct subblock number. */ + const unsigned subMask; + + /** The latency of a hash lookup. */ + const int hashDelay; + /** The number of data blocks. */ + const int numBlocks; + /** The total number of tags in primary and secondary. */ + const int numTags; + /** The number of tags in the secondary tag store. */ + const int numSecondary; + + /** The Null tag pointer. */ + const int tagNull; + /** The last tag in the primary table. */ + const int primaryBound; + + /** All of the tags */ + IICTag *tagStore; + /** + * Pointer to the head of the secondary freelist (maintained with chain + * pointers. + */ + unsigned long freelist; + /** + * The data block freelist. + */ + std::list<unsigned long> blkFreelist; + + /** The primary table. */ + IICSet *sets; + + /** The replacement policy. */ + Repl *repl; + + /** An array of data reference counters. */ + int *dataReferenceCount; + + /** The data blocks. */ + uint8_t *dataStore; + + /** Storage for the fast access data of each cache block. */ + uint8_t **dataBlks; + + /** + * Count of the current number of free secondary tags. + * Used for debugging. + */ + int freeSecond; + + // IIC Statistics + /** + * @addtogroup IICStatistics IIC Statistics + * @{ + */ + + /** Hash hit depth of cache hits. */ + Stats::Distribution<> hitHashDepth; + /** Hash depth for cache misses. */ + Stats::Distribution<> missHashDepth; + /** Count of accesses to each hash set. */ + Stats::Distribution<> setAccess; + + /** The total hash depth for every miss. */ + Stats::Scalar<> missDepthTotal; + /** The total hash depth for all hits. */ + Stats::Scalar<> hitDepthTotal; + /** The number of hash misses. */ + Stats::Scalar<> hashMiss; + /** The number of hash hits. */ + Stats::Scalar<> hashHit; + /** @} */ + + public: + /** + * Collection of parameters for the IIC. + */ + class Params { + public: + /** The size in bytes of the cache. */ + int size; + /** The number of sets in the primary table. */ + int numSets; + /** The block size in bytes. */ + int blkSize; + /** The associativity of the primary table. */ + int assoc; + /** The number of cycles for each hash lookup. */ + int hashDelay; + /** The number of cycles to read the data. */ + int hitLatency; + /** The replacement policy. */ + Repl *rp; + /** The subblock size in bytes. */ + int subblockSize; + }; + + /** + * Construct and initialize this tag store. + * @param params The IIC parameters. + * @todo + * Should make a way to have less tags in the primary than blks in the + * cache. Also should be able to specify number of secondary blks. + */ + IIC(Params ¶ms); + + /** + * Destructor. + */ + virtual ~IIC(); + + /** + * Register the statistics. + * @param name The name to prepend to the statistic descriptions. + */ + void regStats(const std::string &name); + + /** + * Regenerate the block address from the tag. + * @param tag The tag of the block. + * @param set Not needed for the iic. + * @return The block address. + */ + Addr regenerateBlkAddr(Addr tag, int set) { + return (((Addr)tag << tagShift)); + } + + /** + * Return the block size. + * @return The block size. + */ + int getBlockSize() + { + return blkSize; + } + + /** + * Return the subblock size. + * @return The subblock size. + */ + int getSubBlockSize() + { + return subSize; + } + + /** + * Return the hit latency. + * @return the hit latency. + */ + int getHitLatency() const + { + return hitLatency; + } + + /** + * Generate the tag from the address. + * @param addr The address to a get a tag for. + * @param blk Ignored here. + * @return the tag. + */ + Addr extractTag(Addr addr, IICTag *blk) const + { + return (addr >> tagShift); + } + + /** + * Generate the tag from the address. + * @param addr The address to a get a tag for. + * @return the tag. + */ + Addr extractTag(Addr addr) const + { + return (addr >> tagShift); + } + + /** + * Return the set, always 0 for IIC. + * @return 0. + */ + int extractSet(Addr addr) const + { + return 0; + } + + /** + * Get the block offset of an address. + * @param addr The address to get the offset of. + * @return the block offset of the address. + */ + int extractBlkOffset(Addr addr) const + { + return (addr & blkMask); + } + + /** + * Align an address to the block size. + * @param addr the address to align. + * @return The block address. + */ + Addr blkAlign(Addr addr) const + { + return (addr & ~(Addr)blkMask); + } + + /** + * Check for the address in the tagstore. + * @param asid The address space ID. + * @param addr The address to find. + * @return true if it is found. + */ + bool probe(int asid, Addr addr) const; + + /** + * Swap the position of two tags. + * @param index1 The first tag location. + * @param index2 The second tag location. + */ + void tagSwap(unsigned long index1, unsigned long index2); + + /** + * Clear the reference bit of the tag and return its old value. + * @param index The pointer of the tag to manipulate. + * @return The previous state of the reference bit. + */ + bool clearRef(unsigned long index) + { + bool tmp = tagStore[index].isReferenced(); + tagStore[index].status &= ~BlkReferenced; + return tmp; + } + + /** + * Decompress a block if it is compressed. + * @param index The tag store index for the block to uncompress. + */ + void decompressBlock(unsigned long index); + + /** + * Try and compress a block if it is not already compressed. + * @param index The tag store index for the block to compress. + */ + void compressBlock(unsigned long index); + + /** + * Invalidate the block containing the address. + * @param asid The address space ID. + * @param addr The address to invalidate. + */ + void invalidateBlk(int asid, Addr addr); + + /** + * Find the block and update the replacement data. This call also returns + * the access latency as a side effect. + * @param addr The address to find. + * @param asid The address space ID. + * @param lat The access latency. + * @return A pointer to the block found, if any. + */ + IICTag* findBlock(Addr addr, int asid, int &lat); + + /** + * Find the block and update the replacement data. This call also returns + * the access latency as a side effect. + * @param req The req whose block to find + * @param lat The access latency. + * @return A pointer to the block found, if any. + */ + IICTag* findBlock(Packet * &pkt, int &lat); + + /** + * Find the block, do not update the replacement data. + * @param addr The address to find. + * @param asid The address space ID. + * @return A pointer to the block found, if any. + */ + IICTag* findBlock(Addr addr, int asid) const; + + /** + * Find a replacement block for the address provided. + * @param req The request to a find a replacement candidate for. + * @param writebacks List for any writebacks to be performed. + * @param compress_blocks List of blocks to compress, for adaptive comp. + * @return The block to place the replacement in. + */ + IICTag* findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks); + + /** + * Read the data from the internal storage of the given cache block. + * @param blk The block to read the data from. + * @param data The buffer to read the data into. + * @return The cache block's data. + */ + void readData(IICTag *blk, uint8_t *data); + + /** + * Write the data into the internal storage of the given cache block. + * @param blk The block to write to. + * @param data The data to write. + * @param size The number of bytes to write. + * @param writebacks A list for any writebacks to be performed. May be + * needed when writing to a compressed block. + */ + void writeData(IICTag *blk, uint8_t *data, int size, + PacketList & writebacks); + + /** + * Perform a block aligned copy from the source address to the destination. + * @param source The block-aligned source address. + * @param dest The block-aligned destination address. + * @param asid The address space DI. + * @param writebacks List for any generated writeback requests. + */ + void doCopy(Addr source, Addr dest, int asid, PacketList &writebacks); + + /** + * If a block is currently marked copy on write, copy it before writing. + * @param req The write request. + * @param writebacks List for any generated writeback requests. + */ + void fixCopy(Packet * &pkt, PacketList &writebacks); + + /** + * Called at end of simulation to complete average block reference stats. + */ + virtual void cleanupRefs(); +private: + /** + * Return the hash of the address. + * @param addr The address to hash. + * @return the hash of the address. + */ + unsigned hash(Addr addr) const; + + /** + * Search for a block in the secondary tag store. Returns the number of + * hash lookups as a side effect. + * @param asid The address space ID. + * @param tag The tag to match. + * @param chain_ptr The first entry to search. + * @param depth The number of hash lookups made while searching. + * @return A pointer to the block if found. + */ + IICTag *secondaryChain(int asid, Addr tag, unsigned long chain_ptr, + int *depth) const; + + /** + * Free the resources associated with the next replacement block. + * @param writebacks A list of any writebacks to perform. + */ + void freeReplacementBlock(PacketList & writebacks); + + /** + * Return the pointer to a free data block. + * @param writebacks A list of any writebacks to perform. + * @return A pointer to a free data block. + */ + unsigned long getFreeDataBlock(PacketList & writebacks); + + /** + * Get a free tag in the given hash set. + * @param set The hash set to search. + * @param writebacks A list of any writebacks to perform. + * @return a pointer to a free tag. + */ + IICTag* getFreeTag(int set, PacketList & writebacks); + + /** + * Free the resources associated with the given tag. + * @param tag_ptr The tag to free. + */ + void freeTag(IICTag *tag_ptr); + + /** + * Mark the given data block as being available. + * @param data_ptr The data block to free. + */ + void freeDataBlock(unsigned long data_ptr); +}; +#endif // __IIC_HH__ + diff --git a/src/mem/cache/tags/lru.cc b/src/mem/cache/tags/lru.cc new file mode 100644 index 000000000..81b84e11e --- /dev/null +++ b/src/mem/cache/tags/lru.cc @@ -0,0 +1,309 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** + * @file + * Definitions of LRU tag store. + */ + +#include <string> + +#include "mem/cache/base_cache.hh" +#include "base/intmath.hh" +#include "mem/cache/tags/lru.hh" +#include "sim/root.hh" + +using namespace std; + +LRUBlk* +CacheSet::findBlk(int asid, Addr tag) const +{ + for (int i = 0; i < assoc; ++i) { + if (blks[i]->tag == tag && blks[i]->isValid()) { + return blks[i]; + } + } + return 0; +} + + +void +CacheSet::moveToHead(LRUBlk *blk) +{ + // nothing to do if blk is already head + if (blks[0] == blk) + return; + + // write 'next' block into blks[i], moving up from MRU toward LRU + // until we overwrite the block we moved to head. + + // start by setting up to write 'blk' into blks[0] + int i = 0; + LRUBlk *next = blk; + + do { + assert(i < assoc); + // swap blks[i] and next + LRUBlk *tmp = blks[i]; + blks[i] = next; + next = tmp; + ++i; + } while (next != blk); +} + + +// create and initialize a LRU/MRU cache structure +LRU::LRU(int _numSets, int _blkSize, int _assoc, int _hit_latency) : + numSets(_numSets), blkSize(_blkSize), assoc(_assoc), hitLatency(_hit_latency) +{ + // Check parameters + if (blkSize < 4 || !isPowerOf2(blkSize)) { + fatal("Block size must be at least 4 and a power of 2"); + } + if (numSets <= 0 || !isPowerOf2(numSets)) { + fatal("# of sets must be non-zero and a power of 2"); + } + if (assoc <= 0) { + fatal("associativity must be greater than zero"); + } + if (hitLatency <= 0) { + fatal("access latency must be greater than zero"); + } + + LRUBlk *blk; + int i, j, blkIndex; + + blkMask = blkSize - 1; + setShift = floorLog2(blkSize); + setMask = numSets - 1; + tagShift = setShift + floorLog2(numSets); + warmedUp = false; + /** @todo Make warmup percentage a parameter. */ + warmupBound = numSets * assoc; + + sets = new CacheSet[numSets]; + blks = new LRUBlk[numSets * assoc]; + // allocate data storage in one big chunk + dataBlks = new uint8_t[numSets*assoc*blkSize]; + + blkIndex = 0; // index into blks array + for (i = 0; i < numSets; ++i) { + sets[i].assoc = assoc; + + sets[i].blks = new LRUBlk*[assoc]; + + // link in the data blocks + for (j = 0; j < assoc; ++j) { + // locate next cache block + blk = &blks[blkIndex]; + blk->data = &dataBlks[blkSize*blkIndex]; + ++blkIndex; + + // invalidate new cache block + blk->status = 0; + + //EGH Fix Me : do we need to initialize blk? + + // Setting the tag to j is just to prevent long chains in the hash + // table; won't matter because the block is invalid + blk->tag = j; + blk->whenReady = 0; + blk->asid = -1; + blk->isTouched = false; + blk->size = blkSize; + sets[i].blks[j]=blk; + blk->set = i; + } + } +} + +LRU::~LRU() +{ + delete [] dataBlks; + delete [] blks; + delete [] sets; +} + +// probe cache for presence of given block. +bool +LRU::probe(int asid, Addr addr) const +{ + // return(findBlock(Read, addr, asid) != 0); + Addr tag = extractTag(addr); + unsigned myset = extractSet(addr); + + LRUBlk *blk = sets[myset].findBlk(asid, tag); + + return (blk != NULL); // true if in cache +} + +LRUBlk* +LRU::findBlock(Addr addr, int asid, int &lat) +{ + Addr tag = extractTag(addr); + unsigned set = extractSet(addr); + LRUBlk *blk = sets[set].findBlk(asid, tag); + lat = hitLatency; + if (blk != NULL) { + // move this block to head of the MRU list + sets[set].moveToHead(blk); + if (blk->whenReady > curTick + && blk->whenReady - curTick > hitLatency) { + lat = blk->whenReady - curTick; + } + blk->refCount += 1; + } + + return blk; +} + +LRUBlk* +LRU::findBlock(Packet * &pkt, int &lat) +{ + Addr addr = pkt->getAddr(); + int asid = pkt->req->getAsid(); + + Addr tag = extractTag(addr); + unsigned set = extractSet(addr); + LRUBlk *blk = sets[set].findBlk(asid, tag); + lat = hitLatency; + if (blk != NULL) { + // move this block to head of the MRU list + sets[set].moveToHead(blk); + if (blk->whenReady > curTick + && blk->whenReady - curTick > hitLatency) { + lat = blk->whenReady - curTick; + } + blk->refCount += 1; + } + + return blk; +} + +LRUBlk* +LRU::findBlock(Addr addr, int asid) const +{ + Addr tag = extractTag(addr); + unsigned set = extractSet(addr); + LRUBlk *blk = sets[set].findBlk(asid, tag); + return blk; +} + +LRUBlk* +LRU::findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks) +{ + unsigned set = extractSet(pkt->getAddr()); + // grab a replacement candidate + LRUBlk *blk = sets[set].blks[assoc-1]; + sets[set].moveToHead(blk); + if (blk->isValid()) { + replacements[0]++; + totalRefs += blk->refCount; + ++sampledRefs; + blk->refCount = 0; + } else if (!blk->isTouched) { + tagsInUse++; + blk->isTouched = true; + if (!warmedUp && tagsInUse.value() >= warmupBound) { + warmedUp = true; + warmupCycle = curTick; + } + } + + return blk; +} + +void +LRU::invalidateBlk(int asid, Addr addr) +{ + LRUBlk *blk = findBlock(addr, asid); + if (blk) { + blk->status = 0; + blk->isTouched = false; + tagsInUse--; + } +} + +void +LRU::doCopy(Addr source, Addr dest, int asid, PacketList &writebacks) +{ + assert(source == blkAlign(source)); + assert(dest == blkAlign(dest)); + LRUBlk *source_blk = findBlock(source, asid); + assert(source_blk); + LRUBlk *dest_blk = findBlock(dest, asid); + if (dest_blk == NULL) { + // Need to do a replacement + Request *search = new Request(dest,1,0); + Packet * pkt = new Packet(search, Packet::ReadReq, -1); + BlkList dummy_list; + dest_blk = findReplacement(pkt, writebacks, dummy_list); + if (dest_blk->isValid() && dest_blk->isModified()) { + // Need to writeback data. +/* pkt = buildWritebackReq(regenerateBlkAddr(dest_blk->tag, + dest_blk->set), + dest_blk->req->asid, + dest_blk->xc, + blkSize, + dest_blk->data, + dest_blk->size); +*/ + Request *writebackReq = new Request(regenerateBlkAddr(dest_blk->tag, + dest_blk->set), + blkSize, 0); + Packet *writeback = new Packet(writebackReq, Packet::Writeback, -1); + writeback->dataDynamic<uint8_t>(dest_blk->data); + writebacks.push_back(writeback); + } + dest_blk->tag = extractTag(dest); + dest_blk->asid = asid; + delete search; + delete pkt; + } + /** + * @todo Can't assume the status once we have coherence on copies. + */ + + // Set this block as readable, writeable, and dirty. + dest_blk->status = 7; + memcpy(dest_blk->data, source_blk->data, blkSize); +} + +void +LRU::cleanupRefs() +{ + for (int i = 0; i < numSets*assoc; ++i) { + if (blks[i].isValid()) { + totalRefs += blks[i].refCount; + ++sampledRefs; + } + } +} diff --git a/src/mem/cache/tags/lru.hh b/src/mem/cache/tags/lru.hh new file mode 100644 index 000000000..437244660 --- /dev/null +++ b/src/mem/cache/tags/lru.hh @@ -0,0 +1,327 @@ +/* + * Copyright (c) 2003-2005 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: Erik Hallnor + */ + +/** + * @file + * Declaration of a LRU tag store. + */ + +#ifndef __LRU_HH__ +#define __LRU_HH__ + +#include <list> + +#include "mem/cache/cache_blk.hh" // base class +#include "mem/packet.hh" // for inlined functions +#include <assert.h> +#include "mem/cache/tags/base_tags.hh" + +class BaseCache; + +/** + * LRU cache block. + */ +class LRUBlk : public CacheBlk { + public: + /** Has this block been touched? Used to aid calculation of warmup time. */ + bool isTouched; +}; + +/** + * An associative set of cache blocks. + */ +class CacheSet +{ + public: + /** The associativity of this set. */ + int assoc; + + /** Cache blocks in this set, maintained in LRU order 0 = MRU. */ + LRUBlk **blks; + + /** + * Find a block matching the tag in this set. + * @param asid The address space ID. + * @param tag The Tag to find. + * @return Pointer to the block if found. + */ + LRUBlk* findBlk(int asid, Addr tag) const; + + /** + * Move the given block to the head of the list. + * @param blk The block to move. + */ + void moveToHead(LRUBlk *blk); +}; + +/** + * A LRU cache tag store. + */ +class LRU : public BaseTags +{ + public: + /** Typedef the block type used in this tag store. */ + typedef LRUBlk BlkType; + /** Typedef for a list of pointers to the local block class. */ + typedef std::list<LRUBlk*> BlkList; + protected: + /** The number of sets in the cache. */ + const int numSets; + /** The number of bytes in a block. */ + const int blkSize; + /** The associativity of the cache. */ + const int assoc; + /** The hit latency. */ + const int hitLatency; + + /** The cache sets. */ + CacheSet *sets; + + /** The cache blocks. */ + LRUBlk *blks; + /** The data blocks, 1 per cache block. */ + uint8_t *dataBlks; + + /** The amount to shift the address to get the set. */ + int setShift; + /** The amount to shift the address to get the tag. */ + int tagShift; + /** Mask out all bits that aren't part of the set index. */ + unsigned setMask; + /** Mask out all bits that aren't part of the block offset. */ + unsigned blkMask; + +public: + /** + * Construct and initialize this tag store. + * @param _numSets The number of sets in the cache. + * @param _blkSize The number of bytes in a block. + * @param _assoc The associativity of the cache. + * @param _hit_latency The latency in cycles for a hit. + */ + LRU(int _numSets, int _blkSize, int _assoc, int _hit_latency); + + /** + * Destructor + */ + virtual ~LRU(); + + /** + * Return the block size. + * @return the block size. + */ + int getBlockSize() + { + return blkSize; + } + + /** + * Return the subblock size. In the case of LRU it is always the block + * size. + * @return The block size. + */ + int getSubBlockSize() + { + return blkSize; + } + + /** + * Search for the address in the cache. + * @param asid The address space ID. + * @param addr The address to find. + * @return True if the address is in the cache. + */ + bool probe(int asid, Addr addr) const; + + /** + * Invalidate the block containing the given address. + * @param asid The address space ID. + * @param addr The address to invalidate. + */ + void invalidateBlk(int asid, Addr addr); + + /** + * Finds the given address in the cache and update replacement data. + * Returns the access latency as a side effect. + * @param req The request whose block to find. + * @param lat The access latency. + * @return Pointer to the cache block if found. + */ + LRUBlk* findBlock(Packet * &pkt, int &lat); + + /** + * Finds the given address in the cache and update replacement data. + * Returns the access latency as a side effect. + * @param addr The address to find. + * @param asid The address space ID. + * @param lat The access latency. + * @return Pointer to the cache block if found. + */ + LRUBlk* findBlock(Addr addr, int asid, int &lat); + + /** + * Finds the given address in the cache, do not update replacement data. + * @param addr The address to find. + * @param asid The address space ID. + * @return Pointer to the cache block if found. + */ + LRUBlk* findBlock(Addr addr, int asid) const; + + /** + * Find a replacement block for the address provided. + * @param req The request to a find a replacement candidate for. + * @param writebacks List for any writebacks to be performed. + * @param compress_blocks List of blocks to compress, for adaptive comp. + * @return The block to place the replacement in. + */ + LRUBlk* findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks); + + /** + * Generate the tag from the given address. + * @param addr The address to get the tag from. + * @return The tag of the address. + */ + Addr extractTag(Addr addr) const + { + return (addr >> tagShift); + } + + /** + * Generate the tag from the given address. + * @param addr The address to get the tag from. + * @param blk Ignored. + * @return The tag of the address. + */ + Addr extractTag(Addr addr, LRUBlk *blk) const + { + return (addr >> tagShift); + } + + /** + * Calculate the set index from the address. + * @param addr The address to get the set from. + * @return The set index of the address. + */ + int extractSet(Addr addr) const + { + return ((addr >> setShift) & setMask); + } + + /** + * Get the block offset from an address. + * @param addr The address to get the offset of. + * @return The block offset. + */ + int extractBlkOffset(Addr addr) const + { + return (addr & blkMask); + } + + /** + * Align an address to the block size. + * @param addr the address to align. + * @return The block address. + */ + Addr blkAlign(Addr addr) const + { + return (addr & ~(Addr)blkMask); + } + + /** + * Regenerate the block address from the tag. + * @param tag The tag of the block. + * @param set The set of the block. + * @return The block address. + */ + Addr regenerateBlkAddr(Addr tag, unsigned set) const + { + return ((tag << tagShift) | ((Addr)set << setShift)); + } + + /** + * Return the hit latency. + * @return the hit latency. + */ + int getHitLatency() const + { + return hitLatency; + } + + /** + * Read the data out of the internal storage of the given cache block. + * @param blk The cache block to read. + * @param data The buffer to read the data into. + * @return The cache block's data. + */ + void readData(LRUBlk *blk, uint8_t *data) + { + memcpy(data, blk->data, blk->size); + } + + /** + * Write data into the internal storage of the given cache block. Since in + * LRU does not store data differently this just needs to update the size. + * @param blk The cache block to write. + * @param data The data to write. + * @param size The number of bytes to write. + * @param writebacks A list for any writebacks to be performed. May be + * needed when writing to a compressed block. + */ + void writeData(LRUBlk *blk, uint8_t *data, int size, + PacketList & writebacks) + { + assert(size <= blkSize); + blk->size = size; + } + + /** + * Perform a block aligned copy from the source address to the destination. + * @param source The block-aligned source address. + * @param dest The block-aligned destination address. + * @param asid The address space DI. + * @param writebacks List for any generated writeback requests. + */ + void doCopy(Addr source, Addr dest, int asid, PacketList &writebacks); + + /** + * No impl. + */ + void fixCopy(Packet * &pkt, PacketList &writebacks) + { + } + + /** + * Called at end of simulation to complete average block reference stats. + */ + virtual void cleanupRefs(); +}; + +#endif diff --git a/src/mem/cache/tags/repl/gen.cc b/src/mem/cache/tags/repl/gen.cc new file mode 100644 index 000000000..ec1c2aaf3 --- /dev/null +++ b/src/mem/cache/tags/repl/gen.cc @@ -0,0 +1,277 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + * Steve Reinhardt + */ + +/** + * @file + * Definitions of the Generational replacement policy. + */ + +#include <string> + +#include "base/misc.hh" +#include "mem/cache/tags/iic.hh" +#include "mem/cache/tags/repl/gen.hh" +#include "sim/builder.hh" +#include "sim/host.hh" + +using namespace std; + +GenRepl::GenRepl(const string &_name, + int _num_pools, + int _fresh_res, + int _pool_res) // fix this, should be set by cache + : Repl(_name) +{ + num_pools = _num_pools; + fresh_res = _fresh_res; + pool_res = _pool_res; + num_entries = 0; + num_pool_entries = 0; + misses = 0; + pools = new GenPool[num_pools+1]; +} + +GenRepl::~GenRepl() +{ + delete [] pools; +} + +unsigned long +GenRepl::getRepl() +{ + unsigned long tmp; + GenReplEntry *re; + int i; + int num_seen = 0; + if (!(num_pool_entries>0)) { + fatal("No blks available to replace"); + } + num_entries--; + num_pool_entries--; + for (i = 0; i < num_pools; i++) { + while ((re = pools[i].pop())) { + num_seen++; + // Remove invalidated entries + if (!re->valid) { + delete re; + continue; + } + if (iic->clearRef(re->tag_ptr)) { + pools[(((i+1)== num_pools)? i :i+1)].push(re, misses); + } + else { + tmp = re->tag_ptr; + delete re; + + repl_pool.sample(i); + + return tmp; + } + } + } + fatal("No replacement found"); + return 0xffffffff; +} + +unsigned long * +GenRepl::getNRepl(int n) +{ + unsigned long *tmp; + GenReplEntry *re; + int i; + if (!(num_pool_entries>(n-1))) { + fatal("Not enough blks available to replace"); + } + num_entries -= n; + num_pool_entries -= n; + tmp = new unsigned long[n]; /* array of cache_blk pointers */ + int blk_index = 0; + for (i = 0; i < num_pools && blk_index < n; i++) { + while (blk_index < n && (re = pools[i].pop())) { + // Remove invalidated entries + if (!re->valid) { + delete re; + continue; + } + if (iic->clearRef(re->tag_ptr)) { + pools[(((i+1)== num_pools)? i :i+1)].push(re, misses); + } + else { + tmp[blk_index] = re->tag_ptr; + blk_index++; + delete re; + repl_pool.sample(i); + } + } + } + if (blk_index >= n) + return tmp; + /* search the fresh pool */ + + fatal("No N replacements found"); + return NULL; +} + +void +GenRepl::doAdvance(std::list<unsigned long> &demoted) +{ + int i; + int num_seen = 0; + GenReplEntry *re; + misses++; + for (i=0; i<num_pools; i++) { + while (misses-pools[i].oldest > pool_res && (re = pools[i].pop())!=NULL) { + if (iic->clearRef(re->tag_ptr)) { + pools[(((i+1)== num_pools)? i :i+1)].push(re, misses); + /** @todo Not really demoted, but use it for now. */ + demoted.push_back(re->tag_ptr); + advance_pool.sample(i); + } + else { + pools[(((i-1)<0)?i:i-1)].push(re, misses); + demoted.push_back(re->tag_ptr); + demote_pool.sample(i); + } + } + num_seen += pools[i].size; + } + while (misses-pools[num_pools].oldest > fresh_res + && (re = pools[num_pools].pop())!=NULL) { + num_pool_entries++; + if (iic->clearRef(re->tag_ptr)) { + pools[num_pools/2].push(re, misses); + /** @todo Not really demoted, but use it for now. */ + demoted.push_back(re->tag_ptr); + advance_pool.sample(num_pools); + } + else { + pools[num_pools/2-1].push(re, misses); + demoted.push_back(re->tag_ptr); + demote_pool.sample(num_pools); + } + } +} + +void* +GenRepl::add(unsigned long tag_index) +{ + GenReplEntry *re = new GenReplEntry; + re->tag_ptr = tag_index; + re->valid = true; + pools[num_pools].push(re, misses); + num_entries++; + return (void*)re; +} + +void +GenRepl::regStats(const string name) +{ + using namespace Stats; + + /** GEN statistics */ + repl_pool + .init(0, 16, 1) + .name(name + ".repl_pool_dist") + .desc("Dist. of Repl. across pools") + .flags(pdf) + ; + + advance_pool + .init(0, 16, 1) + .name(name + ".advance_pool_dist") + .desc("Dist. of Repl. across pools") + .flags(pdf) + ; + + demote_pool + .init(0, 16, 1) + .name(name + ".demote_pool_dist") + .desc("Dist. of Repl. across pools") + .flags(pdf) + ; +} + +int +GenRepl::fixTag(void* _re, unsigned long old_index, unsigned long new_index) +{ + GenReplEntry *re = (GenReplEntry*)_re; + assert(re->valid); + if (re->tag_ptr == old_index) { + re->tag_ptr = new_index; + return 1; + } + fatal("Repl entry: tag ptrs do not match"); + return 0; +} + +bool +GenRepl::findTagPtr(unsigned long index) +{ + for (int i = 0; i < num_pools + 1; ++i) { + list<GenReplEntry*>::const_iterator iter = pools[i].entries.begin(); + list<GenReplEntry*>::const_iterator end = pools[i].entries.end(); + for (; iter != end; ++iter) { + if ((*iter)->valid && (*iter)->tag_ptr == index) { + return true; + } + } + } + return false; +} + +#ifndef DOXYGEN_SHOULD_SKIP_THIS + +BEGIN_DECLARE_SIM_OBJECT_PARAMS(GenRepl) + + Param<int> num_pools; + Param<int> fresh_res; + Param<int> pool_res; + +END_DECLARE_SIM_OBJECT_PARAMS(GenRepl) + + +BEGIN_INIT_SIM_OBJECT_PARAMS(GenRepl) + + INIT_PARAM(num_pools, "capacity in bytes"), + INIT_PARAM(fresh_res, "associativity"), + INIT_PARAM(pool_res, "block size in bytes") + +END_INIT_SIM_OBJECT_PARAMS(GenRepl) + + +CREATE_SIM_OBJECT(GenRepl) +{ + return new GenRepl(getInstanceName(), num_pools, fresh_res, pool_res); +} + +REGISTER_SIM_OBJECT("GenRepl", GenRepl) + +#endif // DOXYGEN_SHOULD_SKIP_THIS diff --git a/src/mem/cache/tags/repl/gen.hh b/src/mem/cache/tags/repl/gen.hh new file mode 100644 index 000000000..c1ceb3f4e --- /dev/null +++ b/src/mem/cache/tags/repl/gen.hh @@ -0,0 +1,247 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + */ + +/** + * @file + * Declarations of generational replacement policy + */ + +#ifndef ___GEN_HH__ +#define __GEN_HH__ + +#include <list> + +#include "base/statistics.hh" +#include "mem/cache/tags/repl/repl.hh" + +/** + * Generational Replacement entry. + */ +class GenReplEntry +{ + public: + /** Valid flag, used to quickly invalidate bogus entries. */ + bool valid; + /** The difference between this entry and the previous in the pool. */ + int delta; + /** Pointer to the corresponding tag in the IIC. */ + unsigned long tag_ptr; +}; + +/** + * Generational replacement pool + */ +class GenPool +{ + public: + /** The time the last entry was added. */ + Tick newest; + /** The time the oldest entry was added. */ + Tick oldest; + /** List of the replacement entries in this pool. */ + std::list<GenReplEntry*> entries; + + /** The number of entries in this pool. */ + int size; + + /** + * Simple constructor. + */ + GenPool() { + newest = 0; + oldest = 0; + size = 0; + } + + /** + * Add an entry to this pool. + * @param re The entry to add. + * @param now The current time. + */ + void push(GenReplEntry *re, Tick now) { + ++size; + if (!entries.empty()) { + re->delta = now - newest; + newest = now; + } else { + re->delta = 0; + newest = oldest = now; + } + entries.push_back(re); + } + + /** + * Remove an entry from the pool. + * @return The entry at the front of the list. + */ + GenReplEntry* pop() { + GenReplEntry *tmp = NULL; + if (!entries.empty()) { + --size; + tmp = entries.front(); + entries.pop_front(); + oldest += tmp->delta; + } + return tmp; + } + + /** + * Return the entry at the front of the list. + * @return the entry at the front of the list. + */ + GenReplEntry* top() { + return entries.front(); + } + + /** + * Destructor. + */ + ~GenPool() { + while (!entries.empty()) { + GenReplEntry *tmp = entries.front(); + entries.pop_front(); + delete tmp; + } + } +}; + +/** + * Generational replacement policy for use with the IIC. + * @todo update to use STL and for efficiency + */ +class GenRepl : public Repl +{ + public: + /** The array of pools. */ + GenPool *pools; + /** The number of pools. */ + int num_pools; + /** The amount of time to stay in the fresh pool. */ + int fresh_res; + /** The amount of time to stay in the normal pools. */ + int pool_res; + /** The maximum number of entries */ + int num_entries; + /** The number of entries currently in the pools. */ + int num_pool_entries; + /** The number of misses. Used as the internal time. */ + Tick misses; + + // Statistics + + /** + * @addtogroup CacheStatistics + * @{ + */ + /** The number of replacements from each pool. */ + Stats::Distribution<> repl_pool; + /** The number of advances out of each pool. */ + Stats::Distribution<> advance_pool; + /** The number of demotions from each pool. */ + Stats::Distribution<> demote_pool; + /** + * @} + */ + + /** + * Constructs and initializes this replacement policy. + * @param name The name of the policy. + * @param num_pools The number of pools to use. + * @param fresh_res The amount of time to wait in the fresh pool. + * @param pool_res The amount of time to wait in the normal pools. + */ + GenRepl(const std::string &name, int num_pools, + int fresh_res, int pool_res); + + /** + * Destructor. + */ + ~GenRepl(); + + /** + * Returns the tag pointer of the cache block to replace. + * @return The tag to replace. + */ + virtual unsigned long getRepl(); + + /** + * Return an array of N tag pointers to replace. + * @param n The number of tag pointer to return. + * @return An array of tag pointers to replace. + */ + virtual unsigned long *getNRepl(int n); + + /** + * Update replacement data + */ + virtual void doAdvance(std::list<unsigned long> &demoted); + + /** + * Add a tag to the replacement policy and return a pointer to the + * replacement entry. + * @param tag_index The tag to add. + * @return The replacement entry. + */ + virtual void* add(unsigned long tag_index); + + /** + * Register statistics. + * @param name The name to prepend to statistic descriptions. + */ + virtual void regStats(const std::string name); + + /** + * Update the tag pointer to when the tag moves. + * @param re The replacement entry of the tag. + * @param old_index The old tag pointer. + * @param new_index The new tag pointer. + * @return 1 if successful, 0 otherwise. + */ + virtual int fixTag(void *re, unsigned long old_index, + unsigned long new_index); + + /** + * Remove this entry from the replacement policy. + * @param re The replacement entry to remove + */ + virtual void removeEntry(void *re) + { + ((GenReplEntry*)re)->valid = false; + } + + protected: + /** + * Debug function to verify that there is only one repl entry per tag. + * @param index The tag index to check. + */ + bool findTagPtr(unsigned long index); +}; + +#endif /* __GEN_HH__ */ diff --git a/src/mem/cache/tags/repl/repl.cc b/src/mem/cache/tags/repl/repl.cc new file mode 100644 index 000000000..ce781eb9f --- /dev/null +++ b/src/mem/cache/tags/repl/repl.cc @@ -0,0 +1,43 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + * Nathan Binkert + */ + +/** + * Definitions of the base replacement class. + */ + +#include "sim/param.hh" +#include "mem/cache/tags/repl/repl.hh" + +#ifndef DOXYGEN_SHOULD_SKIP_THIS + +DEFINE_SIM_OBJECT_CLASS_NAME("Repl", Repl) + +#endif //DOXYGEN_SHOULD_SKIP_THIS diff --git a/src/mem/cache/tags/repl/repl.hh b/src/mem/cache/tags/repl/repl.hh new file mode 100644 index 000000000..7c289a5c1 --- /dev/null +++ b/src/mem/cache/tags/repl/repl.hh @@ -0,0 +1,129 @@ +/* + * Copyright (c) 2002-2005 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: Erik Hallnor + * Steve Reinhardt + * Nathan Binkert + */ + +/** + * @file + * Declaration of a base replacement policy class. + */ + +#ifndef __REPL_HH__ +#define __REPL_HH__ + +#include <string> +#include <list> + +#include "cpu/smt.hh" +#include "sim/host.hh" +#include "sim/sim_object.hh" + + +class IIC; + +/** + * A pure virtual base class that defines the interface of a replacement + * policy. + */ +class Repl : public SimObject +{ + public: + /** Pointer to the IIC using this policy. */ + IIC *iic; + + /** + * Construct and initialize this polixy. + * @param name The instance name of this policy. + */ + Repl (const std::string &name) + : SimObject(name) + { + iic = NULL; + } + + /** + * Set the back pointer to the IIC. + * @param iic_ptr Pointer to the IIC. + */ + void setIIC(IIC *iic_ptr) + { + iic = iic_ptr; + } + + /** + * Returns the tag pointer of the cache block to replace. + * @return The tag to replace. + */ + virtual unsigned long getRepl() = 0; + + /** + * Return an array of N tag pointers to replace. + * @param n The number of tag pointer to return. + * @return An array of tag pointers to replace. + */ + virtual unsigned long *getNRepl(int n) = 0; + + /** + * Update replacement data + */ + virtual void doAdvance(std::list<unsigned long> &demoted) = 0; + + /** + * Add a tag to the replacement policy and return a pointer to the + * replacement entry. + * @param tag_index The tag to add. + * @return The replacement entry. + */ + virtual void* add(unsigned long tag_index) = 0; + + /** + * Register statistics. + * @param name The name to prepend to statistic descriptions. + */ + virtual void regStats(const std::string name) = 0; + + /** + * Update the tag pointer to when the tag moves. + * @param re The replacement entry of the tag. + * @param old_index The old tag pointer. + * @param new_index The new tag pointer. + * @return 1 if successful, 0 otherwise. + */ + virtual int fixTag(void *re, unsigned long old_index, + unsigned long new_index) = 0; + + /** + * Remove this entry from the replacement policy. + * @param re The replacement entry to remove + */ + virtual void removeEntry(void *re) = 0; +}; + +#endif /* SMT_REPL_HH */ diff --git a/src/mem/cache/tags/split.cc b/src/mem/cache/tags/split.cc new file mode 100644 index 000000000..bf23fb8cb --- /dev/null +++ b/src/mem/cache/tags/split.cc @@ -0,0 +1,478 @@ +/* + * Copyright (c) 2004-2005 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: Lisa Hsu + */ + +/** + * @file + * Definitions of split cache tag store. + */ + +#include <string> +#include <iostream> +#include <fstream> + +#include "base/cprintf.hh" +#include "base/intmath.hh" +#include "base/output.hh" +#include "base/trace.hh" +#include "mem/cache/base_cache.hh" +#include "mem/cache/tags/split.hh" +#include "mem/cache/tags/split_lifo.hh" +#include "mem/cache/tags/split_lru.hh" + + +using namespace std; +using namespace TheISA; + +// create and initialize a partitioned cache structure +Split::Split(int _numSets, int _blkSize, int total_ways, int LRU1_assoc, + bool _lifo, bool _two_queue, int _hit_latency) : + numSets(_numSets), blkSize(_blkSize), lifo(_lifo), hitLatency(_hit_latency) +{ + DPRINTF(Split, "new split cache!!\n"); + + DPRINTF(Split, "lru has %d numSets, %d blkSize, %d assoc, and %d hit_latency\n", + numSets, blkSize, LRU1_assoc, hitLatency); + + lru = new SplitLRU(_numSets, _blkSize, LRU1_assoc, _hit_latency, 1); + + if (total_ways - LRU1_assoc == 0) { + lifo_net = NULL; + lru_net = NULL; + } else { + if (lifo) { + DPRINTF(Split, "Other partition is a LIFO with size %d in bytes. it gets %d ways\n", + (total_ways - LRU1_assoc)*_numSets*_blkSize, (total_ways - LRU1_assoc)); + lifo_net = new SplitLIFO(_blkSize, (total_ways - LRU1_assoc)*_numSets*_blkSize, + (total_ways - LRU1_assoc), _hit_latency, _two_queue, 2); + lru_net = NULL; + } + else { + DPRINTF(Split, "other LRU gets %d ways\n", total_ways - LRU1_assoc); + lru_net = new SplitLRU(_numSets, _blkSize, total_ways - LRU1_assoc, _hit_latency, 2); + lifo_net = NULL; + } + } + + blkMask = blkSize - 1; + + if (!isPowerOf2(total_ways)) + warn("total cache ways/columns %d should be power of 2", + total_ways); + + warmedUp = false; + /** @todo Make warmup percentage a parameter. */ + warmupBound = numSets * total_ways; + +} + +Split::~Split() +{ + delete lru; + if (lifo) + delete lifo_net; + else + delete lru_net; +} + +void +Split::regStats(const string &name) +{ + using namespace Stats; + + BaseTags::regStats(name); + + usedEvictDist.init(0,3000,40); + unusedEvictDist.init(0,3000,40); + useByCPUCycleDist.init(0,35,1); + + nic_repl + .name(name + ".nic_repl") + .desc("number of replacements in the nic partition") + .precision(0) + ; + + cpu_repl + .name(name + ".cpu_repl") + .desc("number of replacements in the cpu partition") + .precision(0) + ; + + lru->regStats(name + ".lru"); + + if (lifo && lifo_net) { + lifo_net->regStats(name + ".lifo_net"); + } else if (lru_net) { + lru_net->regStats(name + ".lru_net"); + } + + nicUsedWhenEvicted + .name(name + ".nicUsedWhenEvicted") + .desc("number of NIC blks that were used before evicted") + ; + + nicUsedTotLatency + .name(name + ".nicUsedTotLatency") + .desc("total cycles before eviction of used NIC blks") + ; + + nicUsedTotEvicted + .name(name + ".nicUsedTotEvicted") + .desc("total number of used NIC blks evicted") + ; + + nicUsedAvgLatency + .name(name + ".nicUsedAvgLatency") + .desc("avg number of cycles a used NIC blk is in cache") + .precision(0) + ; + nicUsedAvgLatency = nicUsedTotLatency / nicUsedTotEvicted; + + usedEvictDist + .name(name + ".usedEvictDist") + .desc("distribution of used NIC blk eviction times") + .flags(pdf | cdf) + ; + + nicUnusedWhenEvicted + .name(name + ".nicUnusedWhenEvicted") + .desc("number of NIC blks that were unused when evicted") + ; + + nicUnusedTotLatency + .name(name + ".nicUnusedTotLatency") + .desc("total cycles before eviction of unused NIC blks") + ; + + nicUnusedTotEvicted + .name(name + ".nicUnusedTotEvicted") + .desc("total number of unused NIC blks evicted") + ; + + nicUnusedAvgLatency + .name(name + ".nicUnusedAvgLatency") + .desc("avg number of cycles an unused NIC blk is in cache") + .precision(0) + ; + nicUnusedAvgLatency = nicUnusedTotLatency / nicUnusedTotEvicted; + + unusedEvictDist + .name(name + ".unusedEvictDist") + .desc("distribution of unused NIC blk eviction times") + .flags(pdf | cdf) + ; + + nicUseByCPUCycleTotal + .name(name + ".nicUseByCPUCycleTotal") + .desc("total latency of NIC blks til usage time") + ; + + nicBlksUsedByCPU + .name(name + ".nicBlksUsedByCPU") + .desc("total number of NIC blks used") + ; + + nicAvgUsageByCPULatency + .name(name + ".nicAvgUsageByCPULatency") + .desc("average number of cycles before a NIC blk that is used gets used") + .precision(0) + ; + nicAvgUsageByCPULatency = nicUseByCPUCycleTotal / nicBlksUsedByCPU; + + useByCPUCycleDist + .name(name + ".useByCPUCycleDist") + .desc("the distribution of cycle time in cache before NIC blk is used") + .flags(pdf | cdf) + ; + + cpuUsedBlks + .name(name + ".cpuUsedBlks") + .desc("number of cpu blks that were used before evicted") + ; + + cpuUnusedBlks + .name(name + ".cpuUnusedBlks") + .desc("number of cpu blks that were unused before evicted") + ; + + nicAvgLatency + .name(name + ".nicAvgLatency") + .desc("avg number of cycles a NIC blk is in cache before evicted") + .precision(0) + ; + nicAvgLatency = (nicUnusedTotLatency + nicUsedTotLatency) / + (nicUnusedTotEvicted + nicUsedTotEvicted); + + NR_CP_hits + .name(name + ".NR_CP_hits") + .desc("NIC requests hitting in CPU Partition") + ; + + NR_NP_hits + .name(name + ".NR_NP_hits") + .desc("NIC requests hitting in NIC Partition") + ; + + CR_CP_hits + .name(name + ".CR_CP_hits") + .desc("CPU requests hitting in CPU partition") + ; + + CR_NP_hits + .name(name + ".CR_NP_hits") + .desc("CPU requests hitting in NIC partition") + ; + +} + +// probe cache for presence of given block. +bool +Split::probe(int asid, Addr addr) const +{ + bool success = lru->probe(asid, addr); + if (!success) { + if (lifo && lifo_net) + success = lifo_net->probe(asid, addr); + else if (lru_net) + success = lru_net->probe(asid, addr); + } + + return success; +} + +SplitBlk* +Split::findBlock(Packet * &pkt, int &lat) +{ + + Addr aligned = blkAlign(pkt->getAddr()); + + if (memHash.count(aligned)) { + memHash[aligned]++; + } else if (pkt->nic_pkt()) { + memHash[aligned] = 1; + } + + SplitBlk *blk = lru->findBlock(pkt->getAddr(), pkt->req->getAsid(), lat); + if (blk) { + if (pkt->nic_pkt()) { + NR_CP_hits++; + } else { + CR_CP_hits++; + } + } else { + if (lifo && lifo_net) { + blk = lifo_net->findBlock(pkt->getAddr(), pkt->req->getAsid(), lat); + + } else if (lru_net) { + blk = lru_net->findBlock(pkt->getAddr(), pkt->req->getAsid(), lat); + } + if (blk) { + if (pkt->nic_pkt()) { + NR_NP_hits++; + } else { + CR_NP_hits++; + } + } + } + + if (blk) { + Tick latency = curTick - blk->ts; + if (blk->isNIC) { + if (!blk->isUsed && !pkt->nic_pkt()) { + useByCPUCycleDist.sample(latency); + nicUseByCPUCycleTotal += latency; + nicBlksUsedByCPU++; + } + } + blk->isUsed = true; + + if (pkt->nic_pkt()) { + DPRINTF(Split, "found block in partition %d\n", blk->part); + } + } + return blk; +} + +SplitBlk* +Split::findBlock(Addr addr, int asid, int &lat) +{ + SplitBlk *blk = lru->findBlock(addr, asid, lat); + if (!blk) { + if (lifo && lifo_net) { + blk = lifo_net->findBlock(addr, asid, lat); + } else if (lru_net) { + blk = lru_net->findBlock(addr, asid, lat); + } + } + + return blk; +} + +SplitBlk* +Split::findBlock(Addr addr, int asid) const +{ + SplitBlk *blk = lru->findBlock(addr, asid); + if (!blk) { + if (lifo && lifo_net) { + blk = lifo_net->findBlock(addr, asid); + } else if (lru_net) { + blk = lru_net->findBlock(addr, asid); + } + } + + return blk; +} + +SplitBlk* +Split::findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks) +{ + SplitBlk *blk; + + if (pkt->nic_pkt()) { + DPRINTF(Split, "finding a replacement for nic_req\n"); + nic_repl++; + if (lifo && lifo_net) + blk = lifo_net->findReplacement(pkt, writebacks, + compress_blocks); + else if (lru_net) + blk = lru_net->findReplacement(pkt, writebacks, + compress_blocks); + // in this case, this is an LRU only cache, it's non partitioned + else + blk = lru->findReplacement(pkt, writebacks, compress_blocks); + } else { + DPRINTF(Split, "finding replacement for cpu_req\n"); + blk = lru->findReplacement(pkt, writebacks, + compress_blocks); + cpu_repl++; + } + + Tick latency = curTick - blk->ts; + if (blk->isNIC) { + if (blk->isUsed) { + nicUsedWhenEvicted++; + usedEvictDist.sample(latency); + nicUsedTotLatency += latency; + nicUsedTotEvicted++; + } else { + nicUnusedWhenEvicted++; + unusedEvictDist.sample(latency); + nicUnusedTotLatency += latency; + nicUnusedTotEvicted++; + } + } else { + if (blk->isUsed) { + cpuUsedBlks++; + } else { + cpuUnusedBlks++; + } + } + + // blk attributes for the new blk coming IN + blk->ts = curTick; + blk->isNIC = (pkt->nic_pkt()) ? true : false; + + return blk; +} + +void +Split::invalidateBlk(int asid, Addr addr) +{ + SplitBlk *blk = lru->findBlock(addr, asid); + if (!blk) { + if (lifo && lifo_net) + blk = lifo_net->findBlock(addr, asid); + else if (lru_net) + blk = lru_net->findBlock(addr, asid); + + if (!blk) + return; + } + + blk->status = 0; + blk->isTouched = false; + tagsInUse--; +} + +void +Split::doCopy(Addr source, Addr dest, int asid, PacketList &writebacks) +{ + if (lru->probe(asid, source)) + lru->doCopy(source, dest, asid, writebacks); + else { + if (lifo && lifo_net) + lifo_net->doCopy(source, dest, asid, writebacks); + else if (lru_net) + lru_net->doCopy(source, dest, asid, writebacks); + } +} + +void +Split::cleanupRefs() +{ + lru->cleanupRefs(); + if (lifo && lifo_net) + lifo_net->cleanupRefs(); + else if (lru_net) + lru_net->cleanupRefs(); + + ofstream memPrint(simout.resolve("memory_footprint.txt").c_str(), + ios::trunc); + + // this shouldn't be here but it happens at the end, which is what i want + memIter end = memHash.end(); + for (memIter iter = memHash.begin(); iter != end; ++iter) { + ccprintf(memPrint, "%8x\t%d\n", (*iter).first, (*iter).second); + } +} + +Addr +Split::regenerateBlkAddr(Addr tag, int set) const +{ + if (lifo_net) + return lifo_net->regenerateBlkAddr(tag, set); + else + return lru->regenerateBlkAddr(tag, set); +} + +Addr +Split::extractTag(Addr addr, SplitBlk *blk) const +{ + if (blk->part == 2) { + if (lifo_net) + return lifo_net->extractTag(addr); + else if (lru_net) + return lru_net->extractTag(addr); + else + panic("this shouldn't happen"); + } else + return lru->extractTag(addr); +} + diff --git a/src/mem/cache/tags/split.hh b/src/mem/cache/tags/split.hh new file mode 100644 index 000000000..5e0340269 --- /dev/null +++ b/src/mem/cache/tags/split.hh @@ -0,0 +1,335 @@ +/* + * Copyright (c) 2004-2005 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: Lisa Hsu + */ + +/** + * @file + * Declaration of a split/partitioned tag store. + */ + +#ifndef __SPLIT_HH__ +#define __SPLIT_HH__ + +#include <list> + +#include "mem/cache/cache_blk.hh" // base class +#include "mem/cache/tags/split_blk.hh" +#include "mem/packet.hh" // for inlined functions +#include <assert.h> +#include "mem/cache/tags/base_tags.hh" +#include "base/hashmap.hh" + +class BaseCache; +class SplitLRU; +class SplitLIFO; + +/** + * A cache tag store. + */ +class Split : public BaseTags +{ + public: + /** Typedef the block type used in this tag store. */ + typedef SplitBlk BlkType; + /** Typedef for a list of pointers to the local block class. */ + typedef std::list<SplitBlk*> BlkList; + protected: + /** The number of sets in the cache. */ + const int numSets; + /** The number of bytes in a block. */ + const int blkSize; + /** Whether the 2nd partition (for the nic) is LIFO or not */ + const bool lifo; + /** The hit latency. */ + const int hitLatency; + + Addr blkMask; + + /** Number of NIC requests that hit in the NIC partition */ + Stats::Scalar<> NR_NP_hits; + /** Number of NIC requests that hit in the CPU partition */ + Stats::Scalar<> NR_CP_hits; + /** Number of CPU requests that hit in the NIC partition */ + Stats::Scalar<> CR_NP_hits; + /** Number of CPU requests that hit in the CPU partition */ + Stats::Scalar<> CR_CP_hits; + /** The number of nic replacements (i.e. misses) */ + Stats::Scalar<> nic_repl; + /** The number of cpu replacements (i.e. misses) */ + Stats::Scalar<> cpu_repl; + + //For latency studies + /** the number of NIC blks that were used before evicted */ + Stats::Scalar<> nicUsedWhenEvicted; + /** the total latency of used NIC blocks in the cache */ + Stats::Scalar<> nicUsedTotLatency; + /** the total number of used NIC blocks evicted */ + Stats::Scalar<> nicUsedTotEvicted; + /** the average number of cycles a used NIC blk is in the cache */ + Stats::Formula nicUsedAvgLatency; + /** the Distribution of used NIC blk eviction times */ + Stats::Distribution<> usedEvictDist; + + /** the number of NIC blks that were unused before evicted */ + Stats::Scalar<> nicUnusedWhenEvicted; + /** the total latency of unused NIC blks in the cache */ + Stats::Scalar<> nicUnusedTotLatency; + /** the total number of unused NIC blocks evicted */ + Stats::Scalar<> nicUnusedTotEvicted; + /** the average number of cycles an unused NIC blk is in the cache */ + Stats::Formula nicUnusedAvgLatency; + /** the Distribution of unused NIC blk eviction times */ + Stats::Distribution<> unusedEvictDist; + + /** The total latency of NIC blocks to 1st usage time by CPU */ + Stats::Scalar<> nicUseByCPUCycleTotal; + /** The total number of NIC blocks used */ + Stats::Scalar<> nicBlksUsedByCPU; + /** the average number of cycles before a NIC blk that is used gets used by CPU */ + Stats::Formula nicAvgUsageByCPULatency; + /** the Distribution of cycles time before a NIC blk is used by CPU*/ + Stats::Distribution<> useByCPUCycleDist; + + /** the number of CPU blks that were used before evicted */ + Stats::Scalar<> cpuUsedBlks; + /** the number of CPU blks that were unused before evicted */ + Stats::Scalar<> cpuUnusedBlks; + + /** the avg number of cycles before a NIC blk is evicted */ + Stats::Formula nicAvgLatency; + + typedef m5::hash_map<Addr, int, m5::hash<Addr> > hash_t; + typedef hash_t::const_iterator memIter; + hash_t memHash; + + + private: + SplitLRU *lru; + SplitLRU *lru_net; + SplitLIFO *lifo_net; + + public: + /** + * Construct and initialize this tag store. + * @param _numSets The number of sets in the cache. + * @param _blkSize The number of bytes in a block. + * @param _assoc The associativity of the cache. + * @param _hit_latency The latency in cycles for a hit. + */ + Split(int _numSets, int _blkSize, int total_ways, int LRU1_assoc, + bool _lifo, bool _two_queue, int _hit_latency); + + /** + * Destructor + */ + virtual ~Split(); + + /** + * Register the stats for this object + * @param name The name to prepend to the stats name. + */ + void regStats(const std::string &name); + + /** + * Return the block size. + * @return the block size. + */ + int getBlockSize() + { + return blkSize; + } + + /** + * Return the subblock size. In the case of Split it is always the block + * size. + * @return The block size. + */ + int getSubBlockSize() + { + return blkSize; + } + + /** + * Search for the address in the cache. + * @param asid The address space ID. + * @param addr The address to find. + * @return True if the address is in the cache. + */ + bool probe(int asid, Addr addr) const; + + /** + * Invalidate the block containing the given address. + * @param asid The address space ID. + * @param addr The address to invalidate. + */ + void invalidateBlk(int asid, Addr addr); + + /** + * Finds the given address in the cache and update replacement data. + * Returns the access latency as a side effect. + * @param addr The address to find. + * @param asid The address space ID. + * @param lat The access latency. + * @return Pointer to the cache block if found. + */ + SplitBlk* findBlock(Addr addr, int asid, int &lat); + + /** + * Finds the given address in the cache and update replacement data. + * Returns the access latency as a side effect. + * @param req The memory request whose block to find + * @param lat The access latency. + * @return Pointer to the cache block if found. + */ + SplitBlk* findBlock(Packet * &pkt, int &lat); + + /** + * Finds the given address in the cache, do not update replacement data. + * @param addr The address to find. + * @param asid The address space ID. + * @return Pointer to the cache block if found. + */ + SplitBlk* findBlock(Addr addr, int asid) const; + + /** + * Find a replacement block for the address provided. + * @param req The request to a find a replacement candidate for. + * @param writebacks List for any writebacks to be performed. + * @param compress_blocks List of blocks to compress, for adaptive comp. + * @return The block to place the replacement in. + */ + SplitBlk* findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks); + + + /** + * Generate the tag from the given address. + * @param addr The address to get the tag from. + * @param blk The block to find the partition it's in + * @return The tag of the address. + */ + Addr extractTag(Addr addr, SplitBlk *blk) const; + + /** + * Calculate the set index from the address. + * @param addr The address to get the set from. + * @return The set index of the address. + */ + int extractSet(Addr addr) const + { + panic("should never call this!\n"); + } + + /** + * Get the block offset from an address. + * @param addr The address to get the offset of. + * @return The block offset. + */ + int extractBlkOffset(Addr addr) const + { + return (addr & blkMask); + } + + /** + * Align an address to the block size. + * @param addr the address to align. + * @return The block address. + */ + Addr blkAlign(Addr addr) const + { + return (addr & ~(Addr) (blkMask)); + } + + /** + * Regenerate the block address from the tag. + * @param tag The tag of the block. + * @param set The set of the block. + * @return The block address. + */ + Addr regenerateBlkAddr(Addr tag, int set) const; + + /** + * Return the hit latency. + * @return the hit latency. + */ + int getHitLatency() const + { + return hitLatency; + } + + /** + * Read the data out of the internal storage of the given cache block. + * @param blk The cache block to read. + * @param data The buffer to read the data into. + * @return The cache block's data. + */ + void readData(SplitBlk *blk, uint8_t *data) + { + memcpy(data, blk->data, blk->size); + } + + /** + * Write data into the internal storage of the given cache block. Since in + * Split does not store data differently this just needs to update the size. + * @param blk The cache block to write. + * @param data The data to write. + * @param size The number of bytes to write. + * @param writebacks A list for any writebacks to be performed. May be + * needed when writing to a compressed block. + */ + void writeData(SplitBlk *blk, uint8_t *data, int size, + PacketList & writebacks) + { + assert(size <= blkSize); + blk->size = size; + } + + /** + * Perform a block aligned copy from the source address to the destination. + * @param source The block-aligned source address. + * @param dest The block-aligned destination address. + * @param asid The address space DI. + * @param writebacks List for any generated writeback requests. + */ + void doCopy(Addr source, Addr dest, int asid, PacketList &writebacks); + + /** + * No impl. + */ + void fixCopy(Packet * &pkt, PacketList &writebacks) + { + } + + /** + * Called at end of simulation to complete average block reference stats. + */ + virtual void cleanupRefs(); +}; + +#endif diff --git a/src/mem/cache/tags/split_blk.hh b/src/mem/cache/tags/split_blk.hh new file mode 100644 index 000000000..f38516180 --- /dev/null +++ b/src/mem/cache/tags/split_blk.hh @@ -0,0 +1,68 @@ +/* + * Copyright (c) 2004-2005 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: Lisa Hsu + */ + +/** + * @file + * Declaration of partitioned tag store cache block class. + */ + +#ifndef __SPLIT_BLK_HH__ +#define __SPLIT_BLK_HH__ + +#include "mem/cache/cache_blk.hh" // base class + +/** + * Split cache block. + */ +class SplitBlk : public CacheBlk { + public: + /** Has this block been touched? Used to aid calculation of warmup time. */ + bool isTouched; + /** Has this block been used after being brought in? (for LIFO partition) */ + bool isUsed; + /** is this blk a NIC block? (i.e. requested by the NIC) */ + bool isNIC; + /** timestamp of the arrival of this block into the cache */ + Tick ts; + /** the previous block in the LIFO partition (brought in before than me) */ + SplitBlk *prev; + /** the next block in the LIFO partition (brought in later than me) */ + SplitBlk *next; + /** which partition this block is in */ + int part; + + SplitBlk() + : isTouched(false), isUsed(false), isNIC(false), ts(0), prev(NULL), next(NULL), + part(0) + {} +}; + +#endif + diff --git a/src/mem/cache/tags/split_lifo.cc b/src/mem/cache/tags/split_lifo.cc new file mode 100644 index 000000000..f6493fdd2 --- /dev/null +++ b/src/mem/cache/tags/split_lifo.cc @@ -0,0 +1,407 @@ +/* + * Copyright (c) 2004-2005 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: Lisa Hsu + */ + +/** + * @file + * Definitions of LIFO tag store usable in a partitioned cache. + */ + +#include <string> + +#include "mem/cache/base_cache.hh" +#include "base/intmath.hh" +#include "mem/cache/tags/split_lifo.hh" +#include "sim/root.hh" +#include "base/trace.hh" + +using namespace std; + +SplitBlk* +LIFOSet::findBlk(int asid, Addr tag) const +{ + for (SplitBlk *blk = firstIn; blk != NULL; blk = blk->next) { + if (blk->tag == tag && blk->isValid()) { + return blk; + } + } + return NULL; +} + +void +LIFOSet::moveToLastIn(SplitBlk *blk) +{ + if (blk == lastIn) + return; + + if (blk == firstIn) { + blk->next->prev = NULL; + } else { + blk->prev->next = blk->next; + blk->next->prev = blk->prev; + } + blk->next = NULL; + blk->prev = lastIn; + lastIn->next = blk; + + lastIn = blk; +} + +void +LIFOSet::moveToFirstIn(SplitBlk *blk) +{ + if (blk == firstIn) + return; + + if (blk == lastIn) { + blk->prev->next = NULL; + } else { + blk->next->prev = blk->prev; + blk->prev->next = blk->next; + } + + blk->prev = NULL; + blk->next = firstIn; + firstIn->prev = blk; + + firstIn = blk; +} + +// create and initialize a LIFO cache structure +SplitLIFO::SplitLIFO(int _blkSize, int _size, int _ways, int _hit_latency, bool two_Queue, int _part) : + blkSize(_blkSize), size(_size), numBlks(_size/_blkSize), numSets((_size/_ways)/_blkSize), ways(_ways), + hitLatency(_hit_latency), twoQueue(two_Queue), part(_part) +{ + if (!isPowerOf2(blkSize)) + fatal("cache block size (in bytes) must be a power of 2"); + if (!(hitLatency > 0)) + fatal("access latency in cycles must be at least on cycle"); + if (_ways == 0) + fatal("if instantiating a splitLIFO, needs non-zero size!"); + + + SplitBlk *blk; + int i, j, blkIndex; + + setShift = floorLog2(blkSize); + blkMask = blkSize - 1; + setMask = numSets - 1; + tagShift = setShift + floorLog2(numSets); + + warmedUp = false; + /** @todo Make warmup percentage a parameter. */ + warmupBound = size/blkSize; + + // allocate data blocks + blks = new SplitBlk[numBlks]; + sets = new LIFOSet[numSets]; + dataBlks = new uint8_t[size]; + +/* + // these start off point to same blk + top = &(blks[0]); + head = top; +*/ + + blkIndex = 0; + for (i=0; i < numSets; ++i) { + sets[i].ways = ways; + sets[i].lastIn = &blks[blkIndex]; + sets[i].firstIn = &blks[blkIndex + ways - 1]; + + /* 3 cases: if there is 1 way, if there are 2 ways, or if there are 3+. + in the case of 1 way, last in and first out point to the same blocks, + and the next and prev pointers need to be assigned specially. and so on + */ + /* deal with the first way */ + blk = &blks[blkIndex]; + blk->prev = &blks[blkIndex + 1]; + blk->next = NULL; + blk->data = &dataBlks[blkSize*blkIndex]; + blk->size = blkSize; + blk->part = part; + blk->set = i; + ++blkIndex; + + /* if there are "middle" ways, do them here */ + if (ways > 2) { + for (j=1; j < ways-1; ++j) { + blk = &blks[blkIndex]; + blk->data = &dataBlks[blkSize*blkIndex]; + blk->prev = &blks[blkIndex+1]; + blk->next = &blks[blkIndex-1]; + blk->data = &(dataBlks[blkSize*blkIndex]); + blk->size = blkSize; + blk->part = part; + blk->set = i; + ++blkIndex; + } + } + + /* do the final way here, depending on whether the final way is the only + way or not + */ + if (ways > 1) { + blk = &blks[blkIndex]; + blk->prev = NULL; + blk->next = &blks[blkIndex - 1]; + blk->data = &dataBlks[blkSize*blkIndex]; + blk->size = blkSize; + blk->part = part; + blk->set = i; + ++blkIndex; + } else { + blk->prev = NULL; + } + } + assert(blkIndex == numBlks); +} + +SplitLIFO::~SplitLIFO() +{ + delete [] blks; + delete [] sets; + delete [] dataBlks; +} + +void +SplitLIFO::regStats(const std::string &name) +{ + BaseTags::regStats(name); + + hits + .name(name + ".hits") + .desc("number of hits on this partition") + .precision(0) + ; + + misses + .name(name + ".misses") + .desc("number of misses in this partition") + .precision(0) + ; + + invalidations + .name(name + ".invalidations") + .desc("number of invalidations in this partition") + .precision(0) + ; +} + +// probe cache for presence of given block. +bool +SplitLIFO::probe(int asid, Addr addr) const +{ + Addr tag = extractTag(addr); + unsigned myset = extractSet(addr); + + SplitBlk* blk = sets[myset].findBlk(asid, tag); + return (blk != NULL); +} + +SplitBlk* +SplitLIFO::findBlock(Addr addr, int asid, int &lat) +{ + Addr tag = extractTag(addr); + unsigned set = extractSet(addr); + SplitBlk *blk = sets[set].findBlk(asid, tag); + + lat = hitLatency; + + if (blk) { + DPRINTF(Split, "Found LIFO blk %#x in set %d, with tag %#x\n", + addr, set, tag); + hits++; + + if (blk->whenReady > curTick && blk->whenReady - curTick > hitLatency) + lat = blk->whenReady - curTick; + blk->refCount +=1; + + if (twoQueue) { + blk->isUsed = true; + sets[set].moveToFirstIn(blk); + } else { + sets[set].moveToLastIn(blk); + } + } + + return blk; +} + +SplitBlk* +SplitLIFO::findBlock(Packet * &pkt, int &lat) +{ + Addr addr = pkt->getAddr(); + int asid = pkt->req->getAsid(); + + Addr tag = extractTag(addr); + unsigned set = extractSet(addr); + SplitBlk *blk = sets[set].findBlk(asid, tag); + + if (blk) { + DPRINTF(Split, "Found LIFO blk %#x in set %d, with tag %#x\n", + addr, set, tag); + hits++; + + if (twoQueue) { + blk->isUsed = true; + sets[set].moveToFirstIn(blk); + } else { + sets[set].moveToLastIn(blk); + } + } + lat = hitLatency; + + return blk; +} + +SplitBlk* +SplitLIFO::findBlock(Addr addr, int asid) const +{ + Addr tag = extractTag(addr); + unsigned set = extractSet(addr); + SplitBlk *blk = sets[set].findBlk(asid, tag); + + return blk; +} + +SplitBlk* +SplitLIFO::findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks) +{ + unsigned set = extractSet(pkt->getAddr()); + + SplitBlk *firstIn = sets[set].firstIn; + SplitBlk *lastIn = sets[set].lastIn; + + SplitBlk *blk; + if (twoQueue && firstIn->isUsed) { + blk = firstIn; + blk->isUsed = false; + sets[set].moveToLastIn(blk); + } else { + int withValue = sets[set].withValue; + if (withValue == ways) { + blk = lastIn; + } else { + blk = &(sets[set].firstIn[ways - ++withValue]); + } + } + + DPRINTF(Split, "just assigned %#x addr into LIFO, replacing %#x status %#x\n", + pkt->getAddr(), regenerateBlkAddr(blk->tag, set), blk->status); + if (blk->isValid()) { + replacements[0]++; + totalRefs += blk->refCount; + ++sampledRefs; + blk->refCount = 0; + } else { + tagsInUse++; + blk->isTouched = true; + if (!warmedUp && tagsInUse.value() >= warmupBound) { + warmedUp = true; + warmupCycle = curTick; + } + } + + misses++; + + return blk; +} + +void +SplitLIFO::invalidateBlk(int asid, Addr addr) +{ + SplitBlk *blk = findBlock(addr, asid); + if (blk) { + blk->status = 0; + blk->isTouched = false; + tagsInUse--; + invalidations++; + } +} + +void +SplitLIFO::doCopy(Addr source, Addr dest, int asid, PacketList &writebacks) +{ +//Copy Unsuported for now +#if 0 + assert(source == blkAlign(source)); + assert(dest == blkAlign(dest)); + SplitBlk *source_blk = findBlock(source, asid); + assert(source_blk); + SplitBlk *dest_blk = findBlock(dest, asid); + if (dest_blk == NULL) { + // Need to do a replacement + Packet * pkt = new Packet(); + pkt->paddr = dest; + BlkList dummy_list; + dest_blk = findReplacement(pkt, writebacks, dummy_list); + if (dest_blk->isValid() && dest_blk->isModified()) { + // Need to writeback data. + pkt = buildWritebackReq(regenerateBlkAddr(dest_blk->tag, + dest_blk->set), + dest_blk->req->asid, + dest_blk->xc, + blkSize, + (cache->doData())?dest_blk->data:0, + dest_blk->size); + writebacks.push_back(pkt); + } + dest_blk->tag = extractTag(dest); + dest_blk->req->asid = asid; + /** + * @todo Do we need to pass in the execution context, or can we + * assume its the same? + */ + assert(source_blk->xc); + dest_blk->xc = source_blk->xc; + } + /** + * @todo Can't assume the status once we have coherence on copies. + */ + + // Set this block as readable, writeable, and dirty. + dest_blk->status = 7; + if (cache->doData()) { + memcpy(dest_blk->data, source_blk->data, blkSize); + } +#endif +} + +void +SplitLIFO::cleanupRefs() +{ + for (int i = 0; i < numBlks; ++i) { + if (blks[i].isValid()) { + totalRefs += blks[i].refCount; + ++sampledRefs; + } + } +} diff --git a/src/mem/cache/tags/split_lifo.hh b/src/mem/cache/tags/split_lifo.hh new file mode 100644 index 000000000..dfcaa0b67 --- /dev/null +++ b/src/mem/cache/tags/split_lifo.hh @@ -0,0 +1,350 @@ +/* + * Copyright (c) 2004-2005 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: Lisa Hsu + */ + +/** + * @file + * Declaration of a LIFO tag store usable in a partitioned cache. + */ + +#ifndef __SPLIT_LIFO_HH__ +#define __SPLIT_LIFO_HH__ + +#include <list> + +#include "mem/cache/cache_blk.hh" // base class +#include "mem/cache/tags/split_blk.hh" +#include "mem/packet.hh" // for inlined functions +#include "base/hashmap.hh" +#include <assert.h> +#include "mem/cache/tags/base_tags.hh" + +class BaseCache; + +/** + * A LIFO set of cache blks + */ +class LIFOSet { + public: + /** the number of blocks in this set */ + int ways; + + /** Cache blocks in this set, maintained in LIFO order where + 0 = Last in (head) */ + SplitBlk *lastIn; + SplitBlk *firstIn; + + /** has the initial "filling" of this set finished? i.e., have you had + * 'ways' number of compulsory misses in this set yet? if withValue == ways, + * then yes. withValue is meant to be the number of blocks in the set that have + * gone through their first compulsory miss. + */ + int withValue; + + /** + * Find a block matching the tag in this set. + * @param asid The address space ID. + * @param tag the Tag you are looking for + * @return Pointer to the block, if found, NULL otherwise + */ + SplitBlk* findBlk(int asid, Addr tag) const; + + void moveToLastIn(SplitBlk *blk); + void moveToFirstIn(SplitBlk *blk); + + LIFOSet() + : ways(-1), lastIn(NULL), firstIn(NULL), withValue(0) + {} +}; + +/** + * A LIFO cache tag store. + */ +class SplitLIFO : public BaseTags +{ + public: + /** Typedef the block type used in this tag store. */ + typedef SplitBlk BlkType; + /** Typedef for a list of pointers to the local block class. */ + typedef std::list<SplitBlk*> BlkList; + protected: + /** The number of bytes in a block. */ + const int blkSize; + /** the size of the cache in bytes */ + const int size; + /** the number of blocks in the cache */ + const int numBlks; + /** the number of sets in the cache */ + const int numSets; + /** the number of ways in the cache */ + const int ways; + /** The hit latency. */ + const int hitLatency; + /** whether this is a "2 queue" replacement @sa moveToLastIn @sa moveToFirstIn */ + const bool twoQueue; + /** indicator for which partition this is */ + const int part; + + /** The cache blocks. */ + SplitBlk *blks; + /** The Cache sets */ + LIFOSet *sets; + /** The data blocks, 1 per cache block. */ + uint8_t *dataBlks; + + /** The amount to shift the address to get the set. */ + int setShift; + /** The amount to shift the address to get the tag. */ + int tagShift; + /** Mask out all bits that aren't part of the set index. */ + unsigned setMask; + /** Mask out all bits that aren't part of the block offset. */ + unsigned blkMask; + + + /** the number of hit in this partition */ + Stats::Scalar<> hits; + /** the number of blocks brought into this partition (i.e. misses) */ + Stats::Scalar<> misses; + /** the number of invalidations in this partition */ + Stats::Scalar<> invalidations; + +public: + /** + * Construct and initialize this tag store. + * @param _numSets The number of sets in the cache. + * @param _blkSize The number of bytes in a block. + * @param _assoc The associativity of the cache. + * @param _hit_latency The latency in cycles for a hit. + */ + SplitLIFO(int _blkSize, int _size, int _ways, int _hit_latency, bool twoQueue, int _part); + + /** + * Destructor + */ + virtual ~SplitLIFO(); + + /** + * Register the statistics for this object + * @param name The name to precede the stat + */ + void regStats(const std::string &name); + + /** + * Return the block size. + * @return the block size. + */ + int getBlockSize() + { + return blkSize; + } + + /** + * Return the subblock size. In the case of LIFO it is always the block + * size. + * @return The block size. + */ + int getSubBlockSize() + { + return blkSize; + } + + /** + * Search for the address in the cache. + * @param asid The address space ID. + * @param addr The address to find. + * @return True if the address is in the cache. + */ + bool probe(int asid, Addr addr) const; + + /** + * Invalidate the block containing the given address. + * @param asid The address space ID. + * @param addr The address to invalidate. + */ + void invalidateBlk(int asid, Addr addr); + + /** + * Finds the given address in the cache and update replacement data. + * Returns the access latency as a side effect. + * @param addr The address to find. + * @param asid The address space ID. + * @param lat The access latency. + * @return Pointer to the cache block if found. + */ + SplitBlk* findBlock(Addr addr, int asid, int &lat); + + /** + * Finds the given address in the cache and update replacement data. + * Returns the access latency as a side effect. + * @param req The req whose block to find + * @param lat The access latency. + * @return Pointer to the cache block if found. + */ + SplitBlk* findBlock(Packet * &pkt, int &lat); + + /** + * Finds the given address in the cache, do not update replacement data. + * @param addr The address to find. + * @param asid The address space ID. + * @return Pointer to the cache block if found. + */ + SplitBlk* findBlock(Addr addr, int asid) const; + + /** + * Find a replacement block for the address provided. + * @param req The request to a find a replacement candidate for. + * @param writebacks List for any writebacks to be performed. + * @param compress_blocks List of blocks to compress, for adaptive comp. + * @return The block to place the replacement in. + */ + SplitBlk* findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks); + + /** + * Generate the tag from the given address. + * @param addr The address to get the tag from. + * @return The tag of the address. + */ + Addr extractTag(Addr addr) const + { + return (addr >> tagShift); + } + + /** + * Generate the tag from the given address. + * @param addr The address to get the tag from. + * @param blk Ignored + * @return The tag of the address. + */ + Addr extractTag(Addr addr, SplitBlk *blk) const + { + return (addr >> tagShift); + } + + /** + * Calculate the set index from the address. + * @param addr The address to get the set from. + * @return The set index of the address. + */ + int extractSet(Addr addr) const + { + return ((addr >> setShift) & setMask); + } + + /** + * Get the block offset from an address. + * @param addr The address to get the offset of. + * @return The block offset. + */ + int extractBlkOffset(Addr addr) const + { + return (addr & blkMask); + } + + /** + * Align an address to the block size. + * @param addr the address to align. + * @return The block address. + */ + Addr blkAlign(Addr addr) const + { + return (addr & ~(Addr)blkMask); + } + + /** + * Regenerate the block address from the tag. + * @param tag The tag of the block. + * @param set The set of the block. + * @return The block address. + */ + Addr regenerateBlkAddr(Addr tag, unsigned set) const + { + return ((tag << tagShift) | ((Addr)set << setShift)); + } + + /** + * Return the hit latency. + * @return the hit latency. + */ + int getHitLatency() const + { + return hitLatency; + } + + /** + * Read the data out of the internal storage of the given cache block. + * @param blk The cache block to read. + * @param data The buffer to read the data into. + * @return The cache block's data. + */ + void readData(SplitBlk *blk, uint8_t *data) + { + memcpy(data, blk->data, blk->size); + } + + /** + * Write data into the internal storage of the given cache block. Since in + * LIFO does not store data differently this just needs to update the size. + * @param blk The cache block to write. + * @param data The data to write. + * @param size The number of bytes to write. + * @param writebacks A list for any writebacks to be performed. May be + * needed when writing to a compressed block. + */ + void writeData(SplitBlk *blk, uint8_t *data, int size, + PacketList & writebacks) + { + assert(size <= blkSize); + blk->size = size; + } + + /** + * Perform a block aligned copy from the source address to the destination. + * @param source The block-aligned source address. + * @param dest The block-aligned destination address. + * @param asid The address space DI. + * @param writebacks List for any generated writeback requests. + */ + void doCopy(Addr source, Addr dest, int asid, PacketList &writebacks); + + /** + * No impl. + */ + void fixCopy(Packet * &pkt, PacketList &writebacks) + { + } + + /** + * Called at end of simulation to complete average block reference stats. + */ + virtual void cleanupRefs(); +}; + +#endif diff --git a/src/mem/cache/tags/split_lru.cc b/src/mem/cache/tags/split_lru.cc new file mode 100644 index 000000000..7fc742001 --- /dev/null +++ b/src/mem/cache/tags/split_lru.cc @@ -0,0 +1,333 @@ +/* + * Copyright (c) 2004-2005 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: Lisa Hsu + */ + +/** + * @file + * Definitions of LRU tag store for a partitioned cache. + */ + +#include <string> + +#include "mem/cache/base_cache.hh" +#include "base/intmath.hh" +#include "mem/cache/tags/split_lru.hh" +#include "sim/root.hh" + +using namespace std; + +SplitBlk* +SplitCacheSet::findBlk(int asid, Addr tag) const +{ + for (int i = 0; i < assoc; ++i) { + if (blks[i]->tag == tag && blks[i]->isValid()) { + return blks[i]; + } + } + return 0; +} + + +void +SplitCacheSet::moveToHead(SplitBlk *blk) +{ + // nothing to do if blk is already head + if (blks[0] == blk) + return; + + // write 'next' block into blks[i], moving up from MRU toward LRU + // until we overwrite the block we moved to head. + + // start by setting up to write 'blk' into blks[0] + int i = 0; + SplitBlk *next = blk; + + do { + assert(i < assoc); + // swap blks[i] and next + SplitBlk *tmp = blks[i]; + blks[i] = next; + next = tmp; + ++i; + } while (next != blk); +} + + +// create and initialize a LRU/MRU cache structure +SplitLRU::SplitLRU(int _numSets, int _blkSize, int _assoc, int _hit_latency, int _part) : + numSets(_numSets), blkSize(_blkSize), assoc(_assoc), hitLatency(_hit_latency), part(_part) +{ + // Check parameters + if (blkSize < 4 || !isPowerOf2(blkSize)) { + fatal("Block size must be at least 4 and a power of 2"); + } + if (numSets <= 0 || !isPowerOf2(numSets)) { + fatal("# of sets must be non-zero and a power of 2"); + } + if (assoc <= 0) { + fatal("associativity must be greater than zero"); + } + if (hitLatency <= 0) { + fatal("access latency must be greater than zero"); + } + + SplitBlk *blk; + int i, j, blkIndex; + + blkMask = blkSize - 1; + setShift = floorLog2(blkSize); + setMask = numSets - 1; + tagShift = setShift + floorLog2(numSets); + warmedUp = false; + /** @todo Make warmup percentage a parameter. */ + warmupBound = numSets * assoc; + + sets = new SplitCacheSet[numSets]; + blks = new SplitBlk[numSets * assoc]; + // allocate data storage in one big chunk + dataBlks = new uint8_t[numSets*assoc*blkSize]; + + blkIndex = 0; // index into blks array + for (i = 0; i < numSets; ++i) { + sets[i].assoc = assoc; + + sets[i].blks = new SplitBlk*[assoc]; + + // link in the data blocks + for (j = 0; j < assoc; ++j) { + // locate next cache block + blk = &blks[blkIndex]; + blk->data = &dataBlks[blkSize*blkIndex]; + ++blkIndex; + + // invalidate new cache block + blk->status = 0; + + //EGH Fix Me : do we need to initialize blk? + + // Setting the tag to j is just to prevent long chains in the hash + // table; won't matter because the block is invalid + blk->tag = j; + blk->whenReady = 0; + blk->asid = -1; + blk->isTouched = false; + blk->size = blkSize; + sets[i].blks[j]=blk; + blk->set = i; + blk->part = part; + } + } +} + +SplitLRU::~SplitLRU() +{ + delete [] dataBlks; + delete [] blks; + delete [] sets; +} + +void +SplitLRU::regStats(const std::string &name) +{ + BaseTags::regStats(name); + + hits + .name(name + ".hits") + .desc("number of hits on this partition") + .precision(0) + ; + + misses + .name(name + ".misses") + .desc("number of misses in this partition") + .precision(0) + ; +} + +// probe cache for presence of given block. +bool +SplitLRU::probe(int asid, Addr addr) const +{ + // return(findBlock(Read, addr, asid) != 0); + Addr tag = extractTag(addr); + unsigned myset = extractSet(addr); + + SplitBlk *blk = sets[myset].findBlk(asid, tag); + + return (blk != NULL); // true if in cache +} + +SplitBlk* +SplitLRU::findBlock(Addr addr, int asid, int &lat) +{ + Addr tag = extractTag(addr); + unsigned set = extractSet(addr); + SplitBlk *blk = sets[set].findBlk(asid, tag); + lat = hitLatency; + if (blk != NULL) { + // move this block to head of the MRU list + sets[set].moveToHead(blk); + if (blk->whenReady > curTick && blk->whenReady - curTick > hitLatency){ + lat = blk->whenReady - curTick; + } + blk->refCount += 1; + hits++; + } + + return blk; +} + +SplitBlk* +SplitLRU::findBlock(Packet * &pkt, int &lat) +{ + Addr addr = pkt->getAddr(); + int asid = pkt->req->getAsid(); + + Addr tag = extractTag(addr); + unsigned set = extractSet(addr); + SplitBlk *blk = sets[set].findBlk(asid, tag); + lat = hitLatency; + if (blk != NULL) { + // move this block to head of the MRU list + sets[set].moveToHead(blk); + if (blk->whenReady > curTick && blk->whenReady - curTick > hitLatency){ + lat = blk->whenReady - curTick; + } + blk->refCount += 1; + hits++; + } + + return blk; +} + +SplitBlk* +SplitLRU::findBlock(Addr addr, int asid) const +{ + Addr tag = extractTag(addr); + unsigned set = extractSet(addr); + SplitBlk *blk = sets[set].findBlk(asid, tag); + return blk; +} + +SplitBlk* +SplitLRU::findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks) +{ + unsigned set = extractSet(pkt->getAddr()); + // grab a replacement candidate + SplitBlk *blk = sets[set].blks[assoc-1]; + sets[set].moveToHead(blk); + if (blk->isValid()) { + replacements[0]++; + totalRefs += blk->refCount; + ++sampledRefs; + blk->refCount = 0; + } else if (!blk->isTouched) { + tagsInUse++; + blk->isTouched = true; + if (!warmedUp && tagsInUse.value() >= warmupBound) { + warmedUp = true; + warmupCycle = curTick; + } + } + + misses++; + + return blk; +} + +void +SplitLRU::invalidateBlk(int asid, Addr addr) +{ + SplitBlk *blk = findBlock(addr, asid); + if (blk) { + blk->status = 0; + blk->isTouched = false; + tagsInUse--; + } +} + +void +SplitLRU::doCopy(Addr source, Addr dest, int asid, PacketList &writebacks) +{ +//Copy not supported for now +#if 0 + assert(source == blkAlign(source)); + assert(dest == blkAlign(dest)); + SplitBlk *source_blk = findBlock(source, asid); + assert(source_blk); + SplitBlk *dest_blk = findBlock(dest, asid); + if (dest_blk == NULL) { + // Need to do a replacement + Packet * pkt = new Packet(); + pkt->paddr = dest; + BlkList dummy_list; + dest_blk = findReplacement(pkt, writebacks, dummy_list); + if (dest_blk->isValid() && dest_blk->isModified()) { + // Need to writeback data. + pkt = buildWritebackReq(regenerateBlkAddr(dest_blk->tag, + dest_blk->set), + dest_blk->req->asid, + dest_blk->xc, + blkSize, + (cache->doData())?dest_blk->data:0, + dest_blk->size); + writebacks.push_back(pkt); + } + dest_blk->tag = extractTag(dest); + dest_blk->req->asid = asid; + /** + * @todo Do we need to pass in the execution context, or can we + * assume its the same? + */ + assert(source_blk->xc); + dest_blk->xc = source_blk->xc; + } + /** + * @todo Can't assume the status once we have coherence on copies. + */ + + // Set this block as readable, writeable, and dirty. + dest_blk->status = 7; + if (cache->doData()) { + memcpy(dest_blk->data, source_blk->data, blkSize); + } +#endif +} + +void +SplitLRU::cleanupRefs() +{ + for (int i = 0; i < numSets*assoc; ++i) { + if (blks[i].isValid()) { + totalRefs += blks[i].refCount; + ++sampledRefs; + } + } +} diff --git a/src/mem/cache/tags/split_lru.hh b/src/mem/cache/tags/split_lru.hh new file mode 100644 index 000000000..03886b1d8 --- /dev/null +++ b/src/mem/cache/tags/split_lru.hh @@ -0,0 +1,333 @@ +/* + * Copyright (c) 2004-2005 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: Lisa Hsu + */ + +/** + * @file + * Declaration of a LRU tag store for a partitioned cache. + */ + +#ifndef __SPLIT_LRU_HH__ +#define __SPLIT_LRU_HH__ + +#include <list> + +#include "mem/cache/cache_blk.hh" // base class +#include "mem/cache/tags/split_blk.hh" +#include "mem/packet.hh" // for inlined functions +#include <assert.h> +#include "mem/cache/tags/base_tags.hh" + +class BaseCache; + +/** + * An associative set of cache blocks. + */ + +class SplitCacheSet +{ + public: + /** The associativity of this set. */ + int assoc; + + /** Cache blocks in this set, maintained in LRU order 0 = MRU. */ + SplitBlk **blks; + + /** + * Find a block matching the tag in this set. + * @param asid The address space ID. + * @param tag The Tag to find. + * @return Pointer to the block if found. + */ + SplitBlk* findBlk(int asid, Addr tag) const; + + /** + * Move the given block to the head of the list. + * @param blk The block to move. + */ + void moveToHead(SplitBlk *blk); +}; + +/** + * A LRU cache tag store. + */ +class SplitLRU : public BaseTags +{ + public: + /** Typedef the block type used in this tag store. */ + typedef SplitBlk BlkType; + /** Typedef for a list of pointers to the local block class. */ + typedef std::list<SplitBlk*> BlkList; + protected: + /** The number of sets in the cache. */ + const int numSets; + /** The number of bytes in a block. */ + const int blkSize; + /** The associativity of the cache. */ + const int assoc; + /** The hit latency. */ + const int hitLatency; + /** indicator for which partition this is */ + const int part; + + /** The cache sets. */ + SplitCacheSet *sets; + + /** The cache blocks. */ + SplitBlk *blks; + /** The data blocks, 1 per cache block. */ + uint8_t *dataBlks; + + /** The amount to shift the address to get the set. */ + int setShift; + /** The amount to shift the address to get the tag. */ + int tagShift; + /** Mask out all bits that aren't part of the set index. */ + unsigned setMask; + /** Mask out all bits that aren't part of the block offset. */ + unsigned blkMask; + + /** number of hits in this partition */ + Stats::Scalar<> hits; + /** number of blocks brought into this partition (i.e. misses) */ + Stats::Scalar<> misses; + +public: + /** + * Construct and initialize this tag store. + * @param _numSets The number of sets in the cache. + * @param _blkSize The number of bytes in a block. + * @param _assoc The associativity of the cache. + * @param _hit_latency The latency in cycles for a hit. + */ + SplitLRU(int _numSets, int _blkSize, int _assoc, int _hit_latency, int _part); + + /** + * Destructor + */ + virtual ~SplitLRU(); + + /** + * Register the statistics for this object + * @param name The name to precede the stat + */ + void regStats(const std::string &name); + + /** + * Return the block size. + * @return the block size. + */ + int getBlockSize() + { + return blkSize; + } + + /** + * Return the subblock size. In the case of LRU it is always the block + * size. + * @return The block size. + */ + int getSubBlockSize() + { + return blkSize; + } + + /** + * Search for the address in the cache. + * @param asid The address space ID. + * @param addr The address to find. + * @return True if the address is in the cache. + */ + bool probe(int asid, Addr addr) const; + + /** + * Invalidate the block containing the given address. + * @param asid The address space ID. + * @param addr The address to invalidate. + */ + void invalidateBlk(int asid, Addr addr); + + /** + * Finds the given address in the cache and update replacement data. + * Returns the access latency as a side effect. + * @param addr The address to find. + * @param asid The address space ID. + * @param lat The access latency. + * @return Pointer to the cache block if found. + */ + SplitBlk* findBlock(Addr addr, int asid, int &lat); + + /** + * Finds the given address in the cache and update replacement data. + * Returns the access latency as a side effect. + * @param req The req whose block to find. + * @param lat The access latency. + * @return Pointer to the cache block if found. + */ + SplitBlk* findBlock(Packet * &pkt, int &lat); + + /** + * Finds the given address in the cache, do not update replacement data. + * @param addr The address to find. + * @param asid The address space ID. + * @return Pointer to the cache block if found. + */ + SplitBlk* findBlock(Addr addr, int asid) const; + + /** + * Find a replacement block for the address provided. + * @param req The request to a find a replacement candidate for. + * @param writebacks List for any writebacks to be performed. + * @param compress_blocks List of blocks to compress, for adaptive comp. + * @return The block to place the replacement in. + */ + SplitBlk* findReplacement(Packet * &pkt, PacketList &writebacks, + BlkList &compress_blocks); + + /** + * Generate the tag from the given address. + * @param addr The address to get the tag from. + * @return The tag of the address. + */ + Addr extractTag(Addr addr) const + { + return (addr >> tagShift); + } + + /** + * Generate the tag from the given address. + * @param addr The address to get the tag from. + * @param blk Ignored. + * @return The tag of the address. + */ + Addr extractTag(Addr addr, SplitBlk *blk) const + { + return (addr >> tagShift); + } + + /** + * Calculate the set index from the address. + * @param addr The address to get the set from. + * @return The set index of the address. + */ + int extractSet(Addr addr) const + { + return ((addr >> setShift) & setMask); + } + + /** + * Get the block offset from an address. + * @param addr The address to get the offset of. + * @return The block offset. + */ + int extractBlkOffset(Addr addr) const + { + return (addr & blkMask); + } + + /** + * Align an address to the block size. + * @param addr the address to align. + * @return The block address. + */ + Addr blkAlign(Addr addr) const + { + return (addr & ~(Addr)blkMask); + } + + /** + * Regenerate the block address from the tag. + * @param tag The tag of the block. + * @param set The set of the block. + * @return The block address. + */ + Addr regenerateBlkAddr(Addr tag, unsigned set) const + { + return ((tag << tagShift) | ((Addr)set << setShift)); + } + + /** + * Return the hit latency. + * @return the hit latency. + */ + int getHitLatency() const + { + return hitLatency; + } + + /** + * Read the data out of the internal storage of the given cache block. + * @param blk The cache block to read. + * @param data The buffer to read the data into. + * @return The cache block's data. + */ + void readData(SplitBlk *blk, uint8_t *data) + { + memcpy(data, blk->data, blk->size); + } + + /** + * Write data into the internal storage of the given cache block. Since in + * LRU does not store data differently this just needs to update the size. + * @param blk The cache block to write. + * @param data The data to write. + * @param size The number of bytes to write. + * @param writebacks A list for any writebacks to be performed. May be + * needed when writing to a compressed block. + */ + void writeData(SplitBlk *blk, uint8_t *data, int size, + PacketList & writebacks) + { + assert(size <= blkSize); + blk->size = size; + } + + /** + * Perform a block aligned copy from the source address to the destination. + * @param source The block-aligned source address. + * @param dest The block-aligned destination address. + * @param asid The address space DI. + * @param writebacks List for any generated writeback requests. + */ + void doCopy(Addr source, Addr dest, int asid, PacketList &writebacks); + + /** + * No impl. + */ + void fixCopy(Packet * &pkt, PacketList &writebacks) + { + } + + /** + * Called at end of simulation to complete average block reference stats. + */ + virtual void cleanupRefs(); +}; + +#endif diff --git a/src/mem/config/cache.hh b/src/mem/config/cache.hh new file mode 100644 index 000000000..24da04021 --- /dev/null +++ b/src/mem/config/cache.hh @@ -0,0 +1,42 @@ +/* + * Copyright (c) 2004-2005 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: Nathan Binkert + */ + +/** + * @file + * Central location to configure which cache types we want to build + * into the simulator. In the future, this should probably be + * autogenerated by some sort of configuration script. + */ +#define USE_CACHE_LRU 1 +#define USE_CACHE_FALRU 1 +// #define USE_CACHE_SPLIT 1 +// #define USE_CACHE_SPLIT_LIFO 1 +#define USE_CACHE_IIC 1 + diff --git a/src/mem/packet.cc b/src/mem/packet.cc index 56dd2bdfa..91298df8c 100644 --- a/src/mem/packet.cc +++ b/src/mem/packet.cc @@ -57,6 +57,19 @@ Packet::cmdString() const } } +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; + default: return OtherCmdString; + } +} + /** delete the data pointed to in the data pointer. Ok to call to matter how * data was allocted. */ void diff --git a/src/mem/packet.hh b/src/mem/packet.hh index 403039d96..2b97ab0c1 100644 --- a/src/mem/packet.hh +++ b/src/mem/packet.hh @@ -41,10 +41,24 @@ #include "mem/request.hh" #include "arch/isa_traits.hh" #include "sim/root.hh" +#include <list> struct Packet; typedef Packet* PacketPtr; typedef uint8_t* PacketDataPtr; +typedef std::list<PacketPtr> PacketList; + +//Coherence Flags +#define NACKED_LINE 1 << 0 +#define SATISFIED 1 << 1 +#define SHARED_LINE 1 << 2 +#define CACHE_LINE_FILL 1 << 3 +#define COMPRESSED 1 << 4 +#define NO_ALLOCATE 1 << 5 + +//For statistics we need max number of commands, hard code it at +//20 for now. @todo fix later +#define NUM_MEM_CMDS 1 << 9 /** * A Packet is used to encapsulate a transfer between two objects in @@ -55,6 +69,10 @@ typedef uint8_t* PacketDataPtr; */ class Packet { + public: + /** Temporary FLAGS field until cache gets working, this should be in coherence/sender state. */ + uint64_t flags; + private: /** A pointer to the data being transfered. It can be differnt * sizes at each level of the heirarchy so it belongs in the @@ -100,8 +118,12 @@ class Packet /** Is the 'src' field valid? */ bool srcValid; + public: + /** Used to calculate latencies for each packet.*/ + Tick time; + /** The special destination address indicating that the packet * should be routed based on its address. */ static const short Broadcast = -1; @@ -149,30 +171,58 @@ class Packet IsRequest = 1 << 4, IsResponse = 1 << 5, NeedsResponse = 1 << 6, + IsSWPrefetch = 1 << 7, + IsHWPrefetch = 1 << 8 }; public: /** List of all commands associated with a packet. */ enum Command { + InvalidCmd = 0, ReadReq = IsRead | IsRequest | NeedsResponse, WriteReq = IsWrite | IsRequest | NeedsResponse, WriteReqNoAck = IsWrite | IsRequest, ReadResp = IsRead | IsResponse, - WriteResp = IsWrite | IsResponse + WriteResp = IsWrite | IsResponse, + Writeback = IsWrite | IsRequest, + SoftPFReq = IsRead | IsRequest | IsSWPrefetch | NeedsResponse, + HardPFReq = IsRead | IsRequest | IsHWPrefetch | NeedsResponse, + SoftPFResp = IsRead | IsRequest | IsSWPrefetch | IsResponse, + HardPFResp = IsRead | IsRequest | IsHWPrefetch | IsResponse, + InvalidateReq = IsInvalidate | IsRequest, + WriteInvalidateReq = IsWrite | IsInvalidate | IsRequest, + UpgradeReq = IsInvalidate | NeedsResponse, + UpgradeResp = IsInvalidate | IsResponse, + ReadExReq = IsRead | IsInvalidate | NeedsResponse, + ReadExResp = IsRead | IsInvalidate | IsResponse }; /** 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); + + /** Return the index of this command. */ + inline int cmdToIndex() const { return (int) cmd; } + /** The command field of the packet. */ Command cmd; bool isRead() { return (cmd & IsRead) != 0; } + bool isWrite() { return (cmd & IsWrite) != 0; } bool isRequest() { return (cmd & IsRequest) != 0; } bool isResponse() { return (cmd & IsResponse) != 0; } bool needsResponse() { return (cmd & NeedsResponse) != 0; } + bool isInvalidate() { return (cmd * IsInvalidate) != 0; } + + bool isCacheFill() { return (flags & CACHE_LINE_FILL) != 0; } + bool isNoAllocate() { return (flags & NO_ALLOCATE) != 0; } + bool isCompressed() { return (flags & COMPRESSED) != 0; } + + bool nic_pkt() { assert("Unimplemented\n" && 0); } /** Possible results of a packet's request. */ enum Result @@ -197,6 +247,10 @@ class Packet Addr getAddr() const { assert(addrSizeValid); return addr; } int getSize() const { assert(addrSizeValid); return size; } + Addr getOffset(int blkSize) const { return req->getPaddr() & (Addr)(blkSize - 1); } + + void addrOverride(Addr newAddr) { assert(addrSizeValid); addr = newAddr; } + void cmdOverride(Command newCmd) { cmd = newCmd; } /** Constructor. Note that a Request object must be constructed * first, but the Requests's physical address and size fields @@ -210,6 +264,21 @@ class Packet req(_req), coherence(NULL), senderState(NULL), cmd(_cmd), result(Unknown) { + flags = 0; + } + + /** 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) + : data(NULL), staticData(false), dynamicData(false), arrayData(false), + addr(_req->paddr & ~(_blkSize - 1)), size(_blkSize), + dest(_dest), + addrSizeValid(_req->validPaddr), srcValid(false), + req(_req), coherence(NULL), senderState(NULL), cmd(_cmd), + result(Unknown) + { + flags = 0; } /** Destructor. */ diff --git a/src/mem/request.hh b/src/mem/request.hh index af1d6d8a8..a1524f807 100644 --- a/src/mem/request.hh +++ b/src/mem/request.hh @@ -44,6 +44,7 @@ class Request; typedef Request* RequestPtr; + /** The request is a Load locked/store conditional. */ const unsigned LOCKED = 0x001; /** The virtual address is also the physical address. */ @@ -62,6 +63,8 @@ const unsigned PF_EXCLUSIVE = 0x100; const unsigned EVICT_NEXT = 0x200; /** The request should ignore unaligned access faults */ const unsigned NO_ALIGN_FAULT = 0x400; +/** The request was an instruction read. */ +const unsigned INST_READ = 0x800; class Request { @@ -224,6 +227,11 @@ class Request /** Accessor function for pc.*/ Addr getPC() { assert(validPC); return pc; } + /** Accessor Function to Check Cacheability. */ + bool isUncacheable() { return getFlags() & UNCACHEABLE; } + + bool isInstRead() { return getFlags() & INST_READ; } + friend class Packet; }; |