diff options
Diffstat (limited to 'src/mem/protocol/MOESI_AMD_Base-RegionDir.sm')
| -rw-r--r-- | src/mem/protocol/MOESI_AMD_Base-RegionDir.sm | 1187 |
1 files changed, 1187 insertions, 0 deletions
diff --git a/src/mem/protocol/MOESI_AMD_Base-RegionDir.sm b/src/mem/protocol/MOESI_AMD_Base-RegionDir.sm new file mode 100644 index 000000000..b392311c5 --- /dev/null +++ b/src/mem/protocol/MOESI_AMD_Base-RegionDir.sm @@ -0,0 +1,1187 @@ +/* + * Copyright (c) 2012-2015 Advanced Micro Devices, Inc. + * All rights reserved. + * + * For use for simulation and test purposes only + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * + * 1. Redistributions of source code must retain the above copyright notice, + * this list of conditions and the following disclaimer. + * + * 2. 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. + * + * 3. Neither the name of the copyright holder 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 HOLDER 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. + * + * Author: Jason Power + */ + +machine(MachineType:RegionDir, "Region Directory for AMD_Base-like protocol") +: CacheMemory *cacheMemory; // stores only region addresses. Must set block size same as below + NodeID cpuRegionBufferNum; + NodeID gpuRegionBufferNum; + int blocksPerRegion := 64; // 4k regions + Cycles toDirLatency := 10; // Latency to fwd requests and send invs to directory + bool always_migrate := "False"; + bool sym_migrate := "False"; + bool asym_migrate := "False"; + bool noTCCdir := "False"; + int TCC_select_num_bits := 1; + + // To the directory + MessageBuffer * requestToDir, network="To", virtual_network="5", vnet_type="request"; + + // To the region buffers + MessageBuffer * notifyToRBuffer, network="To", virtual_network="7", vnet_type="request"; + MessageBuffer * probeToRBuffer, network="To", virtual_network="8", vnet_type="request"; + + // From the region buffers + MessageBuffer * responseFromRBuffer, network="From", virtual_network="2", vnet_type="response"; + MessageBuffer * requestFromRegBuf, network="From", virtual_network="0", vnet_type="request"; + + MessageBuffer * triggerQueue; +{ + + // States + state_declaration(State, desc="Region states", default="RegionDir_State_NP") { + NP, AccessPermission:Invalid, desc="Not present in region directory"; + P, AccessPermission:Invalid, desc="Region is private to owner"; + S, AccessPermission:Invalid, desc="Region is shared between CPU and GPU"; + + P_NP, AccessPermission:Invalid, desc="Evicting the region"; + NP_P, AccessPermission:Invalid, desc="Must wait for ack from R-buf"; + NP_S, AccessPermission:Invalid, desc="Must wait for ack from R-buf"; + P_P, AccessPermission:Invalid, desc="Waiting for ack from R-buf"; + S_S, AccessPermission:Invalid, desc="Waiting for ack from R-buf"; + P_S, AccessPermission:Invalid, desc="Downgrading the region"; + S_P, AccessPermission:Invalid, desc="Upgrading the region"; + P_AS, AccessPermission:Invalid, desc="Sent invalidates, waiting for acks"; + S_AP, AccessPermission:Invalid, desc="Sent invalidates, waiting for acks"; + P_AP, AccessPermission:Invalid, desc="Sent invalidates, waiting for acks"; + + SP_NP_W, AccessPermission:Invalid, desc="Last sharer writing back, waiting for ack"; + S_W, AccessPermission:Invalid, desc="Sharer writing back, waiting for ack"; + + P_AP_W, AccessPermission:Invalid, desc="Fwded request to dir, waiting for ack"; + P_AS_W, AccessPermission:Invalid, desc="Fwded request to dir, waiting for ack"; + S_AP_W, AccessPermission:Invalid, desc="Fwded request to dir, waiting for ack"; + } + + enumeration(Event, desc="Region directory events") { + SendInv, desc="Send inv message to any machine that has a region buffer"; + SendUpgrade, desc="Send upgrade message to any machine that has a region buffer"; + SendDowngrade, desc="Send downgrade message to any machine that has a region buffer"; + + Evict, desc="Evict this region"; + + UpgradeRequest, desc="Request from r-buf for an upgrade"; + SharedRequest, desc="Request from r-buf for read"; + PrivateRequest, desc="Request from r-buf for write"; + + InvAckCore, desc="Ack from region buffer to order the invalidate"; + InvAckCoreNoShare, desc="Ack from region buffer to order the invalidate, and it does not have the region"; + CPUPrivateAck, desc="Ack from region buffer to order private notification"; + + LastAck, desc="Done eviciting all the blocks"; + + StaleCleanWbRequest, desc="stale clean writeback reqeust"; + StaleCleanWbRequestNoShare, desc="stale clean wb req from a cache which should be removed from sharers"; + CleanWbRequest, desc="clean writeback reqeust, multiple sharers"; + CleanWbRequest_LastSharer, desc="clean writeback reqeust, last sharer"; + WritebackAck, desc="Writeback Ack from region buffer"; + DirReadyAck, desc="Directory is ready, waiting Ack from region buffer"; + + TriggerInv, desc="trigger invalidate message"; + TriggerDowngrade, desc="trigger downgrade message"; + } + + enumeration(RequestType, desc="To communicate stats from transitions to recordStats") { + DataArrayRead, desc="Read the data array"; + DataArrayWrite, desc="Write the data array"; + TagArrayRead, desc="Read the data array"; + TagArrayWrite, desc="Write the data array"; + } + + structure(BoolVec, external="yes") { + bool at(int); + void resize(int); + void clear(); + } + + structure(Entry, desc="Region entry", interface="AbstractCacheEntry") { + Addr addr, desc="Base address of this region"; + NetDest Sharers, desc="Set of machines that are sharing, but not owners"; + State RegionState, desc="Region state"; + DataBlock DataBlk, desc="Data for the block (always empty in region dir)"; + MachineID Owner, desc="Machine which owns all blocks in this region"; + Cycles ProbeStart, desc="Time when the first probe request was issued"; + bool LastWriten, default="false", desc="The last time someone accessed this region, it wrote it"; + bool LastWritenByCpu, default="false", desc="The last time the CPU accessed this region, it wrote it"; + bool LastWritenByGpu, default="false", desc="The last time the GPU accessed this region, it wrote it"; + } + + structure(TBE, desc="...") { + State TBEState, desc="Transient state"; + MachineID Owner, desc="Machine which owns all blocks in this region"; + NetDest Sharers, desc="Set of machines to send evicts"; + int NumValidBlocks, desc="Number of blocks valid so we don't have to count a BoolVec"; + bool AllAcksReceived, desc="Got all necessary acks from dir"; + CoherenceRequestType MsgType, desc="Msg type for the evicts could be inv or dwngrd"; + Cycles ProbeRequestTime, default="Cycles(0)", desc="Start of probe request"; + Cycles InitialRequestTime, default="Cycles(0)", desc="To forward back on out msg"; + Addr DemandAddress, desc="Demand address from original request"; + uint64_t probe_id, desc="probe id for lifetime profiling"; + } + + structure(TBETable, external="yes") { + TBE lookup(Addr); + void allocate(Addr); + void deallocate(Addr); + bool isPresent(Addr); + } + + // Stores only region addresses + TBETable TBEs, template="<RegionDir_TBE>", constructor="m_number_of_TBEs"; + int TCC_select_low_bit, default="RubySystem::getBlockSizeBits()"; + + Tick clockEdge(); + Tick cyclesToTicks(Cycles c); + + void set_cache_entry(AbstractCacheEntry b); + void unset_cache_entry(); + void set_tbe(TBE b); + void unset_tbe(); + void wakeUpAllBuffers(); + void wakeUpBuffers(Addr a); + Cycles curCycle(); + + int blockBits, default="RubySystem::getBlockSizeBits()"; + int blockBytes, default="RubySystem::getBlockSizeBytes()"; + int regionBits, default="log2(m_blocksPerRegion)"; + + // Functions + + MachineID getCoreMachine(MachineID rBuf, Addr address) { + if (machineIDToNodeID(rBuf) == cpuRegionBufferNum) { + return createMachineID(MachineType:CorePair, intToID(0)); + } else if (machineIDToNodeID(rBuf) == gpuRegionBufferNum) { + if (noTCCdir) { + return mapAddressToRange(address,MachineType:TCC, + TCC_select_low_bit, TCC_select_num_bits); + } else { + return createMachineID(MachineType:TCCdir, intToID(0)); + } + } else { + error("Unexpected region buffer number"); + } + } + + bool isCpuMachine(MachineID rBuf) { + if (machineIDToNodeID(rBuf) == cpuRegionBufferNum) { + return true; + } else if (machineIDToNodeID(rBuf) == gpuRegionBufferNum) { + return false; + } else { + error("Unexpected region buffer number"); + } + } + + bool symMigrate(Entry cache_entry) { + return cache_entry.LastWriten; + } + + bool asymMigrate(Entry cache_entry, MachineID requestor) { + if (isCpuMachine(requestor)) { + return cache_entry.LastWritenByCpu; + } else { + return cache_entry.LastWritenByGpu; + } + } + + int getRegionOffset(Addr addr) { + if (blocksPerRegion > 1) { + Addr offset := bitSelect(addr, blockBits, regionBits+blockBits-1); + int ret := addressToInt(offset); + assert(ret < blocksPerRegion); + return ret; + } else { + return 0; + } + } + + Addr getRegionBase(Addr addr) { + return maskLowOrderBits(addr, blockBits+regionBits); + } + + Addr getNextBlock(Addr addr) { + Addr a := addr; + makeNextStrideAddress(a, 1); + return a; + } + + bool presentOrAvail(Addr addr) { + DPRINTF(RubySlicc, "Present? %s, avail? %s\n", cacheMemory.isTagPresent(getRegionBase(addr)), cacheMemory.cacheAvail(getRegionBase(addr))); + return cacheMemory.isTagPresent(getRegionBase(addr)) || cacheMemory.cacheAvail(getRegionBase(addr)); + } + + // Returns a region entry! + Entry getCacheEntry(Addr addr), return_by_pointer="yes" { + return static_cast(Entry, "pointer", cacheMemory.lookup(getRegionBase(addr))); + } + + TBE getTBE(Addr addr), return_by_pointer="yes" { + return TBEs.lookup(getRegionBase(addr)); + } + + DataBlock getDataBlock(Addr addr), return_by_ref="yes" { + return getCacheEntry(getRegionBase(addr)).DataBlk; + } + + State getState(TBE tbe, Entry cache_entry, Addr addr) { + if (is_valid(tbe)) { + return tbe.TBEState; + } else if (is_valid(cache_entry)) { + return cache_entry.RegionState; + } + return State:NP; + } + + void setState(TBE tbe, Entry cache_entry, Addr addr, State state) { + if (is_valid(tbe)) { + tbe.TBEState := state; + } + if (is_valid(cache_entry)) { + cache_entry.RegionState := state; + } + } + + AccessPermission getAccessPermission(Addr addr) { + TBE tbe := getTBE(addr); + if(is_valid(tbe)) { + return RegionDir_State_to_permission(tbe.TBEState); + } + Entry cache_entry := getCacheEntry(addr); + if(is_valid(cache_entry)) { + return RegionDir_State_to_permission(cache_entry.RegionState); + } + return AccessPermission:NotPresent; + } + + void setAccessPermission(Entry cache_entry, Addr addr, State state) { + if (is_valid(cache_entry)) { + cache_entry.changePermission(RegionDir_State_to_permission(state)); + } + } + + void functionalRead(Addr addr, Packet *pkt) { + functionalMemoryRead(pkt); + } + + int functionalWrite(Addr addr, Packet *pkt) { + if (functionalMemoryWrite(pkt)) { + return 1; + } else { + return 0; + } + } + + void recordRequestType(RequestType request_type, Addr addr) { + if (request_type == RequestType:DataArrayRead) { + cacheMemory.recordRequestType(CacheRequestType:DataArrayRead, addr); + } else if (request_type == RequestType:DataArrayWrite) { + cacheMemory.recordRequestType(CacheRequestType:DataArrayWrite, addr); + } else if (request_type == RequestType:TagArrayRead) { + cacheMemory.recordRequestType(CacheRequestType:TagArrayRead, addr); + } else if (request_type == RequestType:TagArrayWrite) { + cacheMemory.recordRequestType(CacheRequestType:TagArrayWrite, addr); + } + } + + bool checkResourceAvailable(RequestType request_type, Addr addr) { + if (request_type == RequestType:DataArrayRead) { + return cacheMemory.checkResourceAvailable(CacheResourceType:DataArray, addr); + } else if (request_type == RequestType:DataArrayWrite) { + return cacheMemory.checkResourceAvailable(CacheResourceType:DataArray, addr); + } else if (request_type == RequestType:TagArrayRead) { + return cacheMemory.checkResourceAvailable(CacheResourceType:TagArray, addr); + } else if (request_type == RequestType:TagArrayWrite) { + return cacheMemory.checkResourceAvailable(CacheResourceType:TagArray, addr); + } else { + error("Invalid RequestType type in checkResourceAvailable"); + return true; + } + } + + out_port(triggerQueue_out, TriggerMsg, triggerQueue); + + out_port(requestNetwork_out, CPURequestMsg, requestToDir); + out_port(notifyNetwork_out, CPURequestMsg, notifyToRBuffer); + out_port(probeNetwork_out, NBProbeRequestMsg, probeToRBuffer); + + in_port(triggerQueue_in, TriggerMsg, triggerQueue, rank=2) { + if (triggerQueue_in.isReady(clockEdge())) { + peek(triggerQueue_in, TriggerMsg) { + assert(in_msg.addr == getRegionBase(in_msg.addr)); + Entry cache_entry := getCacheEntry(in_msg.addr); + TBE tbe := getTBE(in_msg.addr); + DPRINTF(RubySlicc, "trigger msg: %s (%s)\n", in_msg, getRegionBase(in_msg.addr)); + if (in_msg.Type == TriggerType:AcksComplete) { + assert(is_valid(tbe)); + trigger(Event:LastAck, in_msg.addr, cache_entry, tbe); + } else if (in_msg.Type == TriggerType:InvRegion) { + assert(is_valid(tbe)); + trigger(Event:TriggerInv, in_msg.addr, cache_entry, tbe); + } else if (in_msg.Type == TriggerType:DowngradeRegion) { + assert(is_valid(tbe)); + trigger(Event:TriggerDowngrade, in_msg.addr, cache_entry, tbe); + } else { + error("Unknown trigger message"); + } + } + } + } + + in_port(responseNetwork_in, ResponseMsg, responseFromRBuffer, rank=1) { + if (responseNetwork_in.isReady(clockEdge())) { + peek(responseNetwork_in, ResponseMsg) { + TBE tbe := getTBE(in_msg.addr); + Entry cache_entry := getCacheEntry(in_msg.addr); + if (in_msg.Type == CoherenceResponseType:CPUPrbResp) { + assert(in_msg.addr == getRegionBase(in_msg.addr)); + assert(is_valid(tbe)); + if (in_msg.NotCached) { + trigger(Event:InvAckCoreNoShare, in_msg.addr, cache_entry, tbe); + } else { + trigger(Event:InvAckCore, in_msg.addr, cache_entry, tbe); + } + } else if (in_msg.Type == CoherenceResponseType:PrivateAck) { + assert(in_msg.addr == getRegionBase(in_msg.addr)); + assert(is_valid(cache_entry)); + //Fix Me...add back in: assert(cache_entry.Sharers.isElement(in_msg.Sender)); + trigger(Event:CPUPrivateAck, in_msg.addr, cache_entry, tbe); + } else if (in_msg.Type == CoherenceResponseType:RegionWbAck) { + //Fix Me...add back in: assert(cache_entry.Sharers.isElement(in_msg.Sender) == false); + assert(in_msg.addr == getRegionBase(in_msg.addr)); + trigger(Event:WritebackAck, in_msg.addr, cache_entry, tbe); + } else if (in_msg.Type == CoherenceResponseType:DirReadyAck) { + assert(is_valid(tbe)); + trigger(Event:DirReadyAck, getRegionBase(in_msg.addr), cache_entry, tbe); + } else { + error("Invalid response type"); + } + } + } + } + + // In from cores + // NOTE: We get the cache / TBE entry based on the region address, + // but pass the block address to the actions + in_port(requestNetwork_in, CPURequestMsg, requestFromRegBuf, rank=0) { + if (requestNetwork_in.isReady(clockEdge())) { + peek(requestNetwork_in, CPURequestMsg) { + //assert(in_msg.addr == getRegionBase(in_msg.addr)); + Addr address := getRegionBase(in_msg.addr); + DPRINTF(RubySlicc, "Got %s, base %s\n", in_msg.addr, address); + if (presentOrAvail(address)) { + TBE tbe := getTBE(address); + Entry cache_entry := getCacheEntry(address); + if (in_msg.Type == CoherenceRequestType:PrivateRequest) { + if (is_valid(cache_entry) && (cache_entry.Owner != in_msg.Requestor || + getState(tbe, cache_entry, address) == State:S)) { + trigger(Event:SendInv, address, cache_entry, tbe); + } else { + trigger(Event:PrivateRequest, address, cache_entry, tbe); + } + } else if (in_msg.Type == CoherenceRequestType:SharedRequest) { + if (is_invalid(cache_entry)) { + // If no one has ever requested this region give private permissions + trigger(Event:PrivateRequest, address, cache_entry, tbe); + } else { + if (always_migrate || + (sym_migrate && symMigrate(cache_entry)) || + (asym_migrate && asymMigrate(cache_entry, in_msg.Requestor))) { + if (cache_entry.Sharers.count() == 1 && + cache_entry.Sharers.isElement(in_msg.Requestor)) { + trigger(Event:UpgradeRequest, address, cache_entry, tbe); + } else { + trigger(Event:SendInv, address, cache_entry, tbe); + } + } else { // don't migrate + if(cache_entry.Sharers.isElement(in_msg.Requestor) || + getState(tbe, cache_entry, address) == State:S) { + trigger(Event:SharedRequest, address, cache_entry, tbe); + } else { + trigger(Event:SendDowngrade, address, cache_entry, tbe); + } + } + } + } else if (in_msg.Type == CoherenceRequestType:UpgradeRequest) { + if (is_invalid(cache_entry)) { + trigger(Event:PrivateRequest, address, cache_entry, tbe); + } else if (cache_entry.Sharers.count() == 1 && cache_entry.Sharers.isElement(in_msg.Requestor)) { + trigger(Event:UpgradeRequest, address, cache_entry, tbe); + } else { + trigger(Event:SendUpgrade, address, cache_entry, tbe); + } + } else if (in_msg.Type == CoherenceRequestType:CleanWbRequest) { + if (is_invalid(cache_entry) || cache_entry.Sharers.isElement(in_msg.Requestor) == false) { + trigger(Event:StaleCleanWbRequest, address, cache_entry, tbe); + } else { + DPRINTF(RubySlicc, "wb address %s(%s) owner %s sharers %s requestor %s %d %d\n", in_msg.addr, getRegionBase(in_msg.addr), cache_entry.Owner, cache_entry.Sharers, in_msg.Requestor, cache_entry.Sharers.isElement(in_msg.Requestor), cache_entry.Sharers.count()); + if (cache_entry.Sharers.isElement(in_msg.Requestor) && cache_entry.Sharers.count() == 1) { + DPRINTF(RubySlicc, "last wb\n"); + trigger(Event:CleanWbRequest_LastSharer, address, cache_entry, tbe); + } else { + DPRINTF(RubySlicc, "clean wb\n"); + trigger(Event:CleanWbRequest, address, cache_entry, tbe); + } + } + } else { + error("unknown region dir request type"); + } + } else { + Addr victim := cacheMemory.cacheProbe(getRegionBase(in_msg.addr)); + TBE victim_tbe := getTBE(victim); + Entry victim_entry := getCacheEntry(victim); + DPRINTF(RubySlicc, "Evicting address %s for new region at address %s(%s)\n", victim, in_msg.addr, getRegionBase(in_msg.addr)); + assert(is_valid(victim_entry)); + trigger(Event:Evict, victim, victim_entry, victim_tbe); + } + } + } + } + + // Actions + + action(f_fwdReqToDir, "f", desc="Forward CPU request to directory") { + peek(requestNetwork_in, CPURequestMsg) { + enqueue(requestNetwork_out, CPURequestMsg, toDirLatency) { + out_msg.addr := in_msg.addr; // This is the block address. "address" is the region address + out_msg.Type := in_msg.OriginalType; + out_msg.DataBlk := in_msg.DataBlk; + out_msg.Dirty := in_msg.Dirty; + out_msg.Requestor := getCoreMachine(in_msg.Requestor,address); + out_msg.WTRequestor := in_msg.WTRequestor; + out_msg.Destination.add(map_Address_to_Directory(in_msg.addr)); + out_msg.Shared := in_msg.Shared; + out_msg.MessageSize := in_msg.MessageSize; + out_msg.Private := in_msg.Private; + out_msg.NoAckNeeded := true; + out_msg.InitialRequestTime := in_msg.InitialRequestTime; + out_msg.ProbeRequestStartTime := curCycle(); + out_msg.DemandRequest := true; + if (is_valid(cache_entry) && getState(tbe, cache_entry, address) != State:S) { + out_msg.Acks := cache_entry.Sharers.count(); + } else { + out_msg.Acks := 0; + } + } + } + } + + action(f_fwdReqToDirShared, "fs", desc="Forward CPU request to directory (shared)") { + peek(requestNetwork_in, CPURequestMsg) { + enqueue(requestNetwork_out, CPURequestMsg, toDirLatency) { + out_msg.addr := in_msg.addr; // This is the block address. "address" is the region address + out_msg.Type := in_msg.OriginalType; + out_msg.DataBlk := in_msg.DataBlk; + out_msg.Dirty := in_msg.Dirty; + out_msg.Requestor := getCoreMachine(in_msg.Requestor,address); + out_msg.WTRequestor := in_msg.WTRequestor; + out_msg.Destination.add(map_Address_to_Directory(in_msg.addr)); + out_msg.Shared := in_msg.Shared; + out_msg.MessageSize := in_msg.MessageSize; + out_msg.Private := in_msg.Private; + out_msg.NoAckNeeded := true; + out_msg.InitialRequestTime := in_msg.InitialRequestTime; + out_msg.ProbeRequestStartTime := curCycle(); + out_msg.DemandRequest := true; + out_msg.ForceShared := true; + if (is_valid(cache_entry) && getState(tbe, cache_entry, address) != State:S) { + out_msg.Acks := cache_entry.Sharers.count(); + } else { + out_msg.Acks := 0; + } + } + } + } + + action(f_fwdReqToDirWithAck, "fa", desc="Forward CPU request to directory with ack request") { + peek(requestNetwork_in, CPURequestMsg) { + enqueue(requestNetwork_out, CPURequestMsg, toDirLatency) { + out_msg.addr := in_msg.addr; // This is the block address. "address" is the region address + out_msg.Type := in_msg.OriginalType; + out_msg.DataBlk := in_msg.DataBlk; + out_msg.Dirty := in_msg.Dirty; + out_msg.Requestor := getCoreMachine(in_msg.Requestor,address); + out_msg.WTRequestor := in_msg.WTRequestor; + out_msg.Destination.add(map_Address_to_Directory(in_msg.addr)); + out_msg.Shared := in_msg.Shared; + out_msg.MessageSize := in_msg.MessageSize; + out_msg.Private := in_msg.Private; + out_msg.NoAckNeeded := false; + out_msg.InitialRequestTime := in_msg.InitialRequestTime; + out_msg.ProbeRequestStartTime := curCycle(); + out_msg.DemandRequest := true; + if (is_valid(cache_entry)) { + out_msg.Acks := cache_entry.Sharers.count(); + // Don't need an ack from the requestor! + if (cache_entry.Sharers.isElement(in_msg.Requestor)) { + out_msg.Acks := out_msg.Acks - 1; + } + } else { + out_msg.Acks := 0; + } + } + } + } + + action(f_fwdReqToDirWithAckShared, "fas", desc="Forward CPU request to directory with ack request") { + peek(requestNetwork_in, CPURequestMsg) { + enqueue(requestNetwork_out, CPURequestMsg, toDirLatency) { + out_msg.addr := in_msg.addr; // This is the block address. "address" is the region address + out_msg.Type := in_msg.OriginalType; + out_msg.DataBlk := in_msg.DataBlk; + out_msg.Dirty := in_msg.Dirty; + out_msg.Requestor := getCoreMachine(in_msg.Requestor,address); + out_msg.WTRequestor := in_msg.WTRequestor; + out_msg.Destination.add(map_Address_to_Directory(in_msg.addr)); + out_msg.Shared := in_msg.Shared; + out_msg.MessageSize := in_msg.MessageSize; + out_msg.Private := in_msg.Private; + out_msg.NoAckNeeded := false; + out_msg.InitialRequestTime := in_msg.InitialRequestTime; + out_msg.ProbeRequestStartTime := curCycle(); + out_msg.DemandRequest := true; + out_msg.ForceShared := true; + if (is_valid(cache_entry)) { + out_msg.Acks := cache_entry.Sharers.count(); + // Don't need an ack from the requestor! + if (cache_entry.Sharers.isElement(in_msg.Requestor)) { + out_msg.Acks := out_msg.Acks - 1; + } + } else { + out_msg.Acks := 0; + } + } + } + } + + action(a_allocateRegionEntry, "a", desc="Allocate a new entry") { + set_cache_entry(cacheMemory.allocate(getRegionBase(address), new Entry)); + peek(requestNetwork_in, CPURequestMsg) { + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + } + } + + action(d_deallocateRegionEntry, "d", desc="Deallocate region entry") { + cacheMemory.deallocate(getRegionBase(address)); + unset_cache_entry(); + } + + action(ra_receiveAck, "ra", desc="Mark TBE entry as received this ack") { + //assert(tbe.ValidBlocks.at(getRegionOffset(address))); + DPRINTF(RubySlicc, "received ack for %s reg: %s\n", address, getRegionBase(address)); + tbe.NumValidBlocks := tbe.NumValidBlocks - 1; + assert(tbe.NumValidBlocks >= 0); + if (tbe.NumValidBlocks == 0) { + tbe.AllAcksReceived := true; + enqueue(triggerQueue_out, TriggerMsg, 1) { + out_msg.Type := TriggerType:AcksComplete; + out_msg.addr := address; + } + } + APPEND_TRANSITION_COMMENT(getRegionBase(address)); + APPEND_TRANSITION_COMMENT(" Acks left receive "); + APPEND_TRANSITION_COMMENT(tbe.NumValidBlocks); + } + + action(ca_checkAcks, "ca", desc="Check to see if we need more acks") { + if (tbe.NumValidBlocks == 0) { + tbe.AllAcksReceived := true; + enqueue(triggerQueue_out, TriggerMsg, 1) { + out_msg.Type := TriggerType:AcksComplete; + out_msg.addr := address; + } + } + } + + action(ti_triggerInv, "ti", desc="") { + enqueue(triggerQueue_out, TriggerMsg, 1) { + out_msg.Type := TriggerType:InvRegion; + out_msg.addr := address; + } + } + + action(td_triggerDowngrade, "td", desc="") { + enqueue(triggerQueue_out, TriggerMsg, 1) { + out_msg.Type := TriggerType:DowngradeRegion; + out_msg.addr := address; + } + } + + action(t_allocateTBE, "t", desc="allocate TBE Entry") { + check_allocate(TBEs); + TBEs.allocate(getRegionBase(address)); + set_tbe(getTBE(address)); + if (is_valid(cache_entry)) { + tbe.Owner := cache_entry.Owner; + tbe.Sharers := cache_entry.Sharers; + tbe.AllAcksReceived := true; // assume no acks are required + } + tbe.ProbeRequestTime := curCycle(); + peek(requestNetwork_in, CPURequestMsg) { + tbe.InitialRequestTime := in_msg.InitialRequestTime; + tbe.DemandAddress := in_msg.addr; + } + APPEND_TRANSITION_COMMENT(getRegionBase(address)); + APPEND_TRANSITION_COMMENT(" Acks left "); + APPEND_TRANSITION_COMMENT(tbe.NumValidBlocks); + APPEND_TRANSITION_COMMENT(" Owner, "); + APPEND_TRANSITION_COMMENT(tbe.Owner); + APPEND_TRANSITION_COMMENT(" sharers, "); + APPEND_TRANSITION_COMMENT(tbe.Sharers); + } + + action(ss_setSharers, "ss", desc="Add requestor to sharers") { + peek(requestNetwork_in, CPURequestMsg) { + cache_entry.Sharers.add(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(cache_entry.Sharers); + } + } + + action(rs_removeSharer, "rs", desc="Remove requestor to sharers") { + peek(requestNetwork_in, CPURequestMsg) { + cache_entry.Sharers.remove(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" removing "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" sharers "); + APPEND_TRANSITION_COMMENT(cache_entry.Sharers); + } + } + + action(rsr_removeSharerResponse, "rsr", desc="Remove requestor to sharers") { + peek(responseNetwork_in, ResponseMsg) { + cache_entry.Sharers.remove(in_msg.Sender); + APPEND_TRANSITION_COMMENT(cache_entry.Sharers); + } + } + + action(cs_clearSharers, "cs", desc="Add requestor to sharers") { + cache_entry.Sharers.clear(); + } + + action(so_setOwner, "so", desc="Set the owner to the requestor") { + peek(requestNetwork_in, CPURequestMsg) { + cache_entry.Owner := in_msg.Requestor; + APPEND_TRANSITION_COMMENT(" Owner now: "); + APPEND_TRANSITION_COMMENT(cache_entry.Owner); + } + } + + action(rr_removeRequestorFromTBE, "rr", desc="Remove requestor from TBE sharers") { + peek(requestNetwork_in, CPURequestMsg) { + tbe.Sharers.remove(in_msg.Requestor); + } + } + + action(ur_updateDirtyStatusOnRequest, "ur", desc="Update dirty status on demand request") { + peek(requestNetwork_in, CPURequestMsg) { + if (is_valid(cache_entry)) { + if ((in_msg.Type == CoherenceRequestType:SharedRequest) && + (cache_entry.Sharers.isElement(in_msg.Requestor) == false)) { + cache_entry.LastWriten := false; + if (isCpuMachine(in_msg.Requestor)) { + cache_entry.LastWritenByCpu := false; + } else { + cache_entry.LastWritenByGpu := false; + } + } else if ((in_msg.Type == CoherenceRequestType:PrivateRequest) || + (in_msg.Type == CoherenceRequestType:UpgradeRequest)) { + cache_entry.LastWriten := true; + if (isCpuMachine(in_msg.Requestor)) { + cache_entry.LastWritenByCpu := true; + } else { + cache_entry.LastWritenByGpu := true; + } + } + } + } + } + + action(ud_updateDirtyStatusWithWb, "ud", desc="Update dirty status on writeback") { + peek(requestNetwork_in, CPURequestMsg) { + if (is_valid(cache_entry) && in_msg.Dirty) { + cache_entry.LastWriten := true; + if (isCpuMachine(in_msg.Requestor)) { + cache_entry.LastWritenByCpu := true; + } else { + cache_entry.LastWritenByGpu := true; + } + } + } + } + + action(sns_setNumAcksSharers, "sns", desc="Set number of acks to one per shared region buffer") { + assert(is_valid(tbe)); + assert(is_valid(cache_entry)); + tbe.NumValidBlocks := tbe.Sharers.count(); + } + + action(sno_setNumAcksOne, "sno", desc="Set number of acks to one per shared region buffer") { + assert(is_valid(tbe)); + assert(is_valid(cache_entry)); + tbe.NumValidBlocks := 1; + } + + action(dt_deallocateTBE, "dt", desc="deallocate TBE Entry") { + TBEs.deallocate(getRegionBase(address)); + APPEND_TRANSITION_COMMENT(" reg: "); + APPEND_TRANSITION_COMMENT(getRegionBase(address)); + unset_tbe(); + } + + action(wb_sendWbNotice, "wb", desc="Send notice to cache that writeback is acknowledged") { + peek(requestNetwork_in, CPURequestMsg) { + enqueue(notifyNetwork_out, CPURequestMsg, 1) { + out_msg.addr := getRegionBase(address); + out_msg.Type := CoherenceRequestType:WbNotify; + out_msg.Destination.add(in_msg.Requestor); + out_msg.Requestor := machineID; + out_msg.MessageSize := MessageSizeType:Request_Control; + out_msg.InitialRequestTime := in_msg.InitialRequestTime; + } + } + } + + action(wbn_sendWbNoticeNoAck, "wbn", desc="Send notice to cache that writeback is acknowledged (no ack needed)") { + peek(requestNetwork_in, CPURequestMsg) { + enqueue(notifyNetwork_out, CPURequestMsg, 1) { + out_msg.addr := getRegionBase(address); + out_msg.Type := CoherenceRequestType:WbNotify; + out_msg.Destination.add(in_msg.Requestor); + out_msg.Requestor := machineID; + out_msg.MessageSize := MessageSizeType:Request_Control; + out_msg.InitialRequestTime := in_msg.InitialRequestTime; + out_msg.NoAckNeeded := true; + } + } + } + + action(b_sendPrivateNotice, "b", desc="Send notice to private cache that it has private access") { + peek(requestNetwork_in, CPURequestMsg) { + enqueue(notifyNetwork_out, CPURequestMsg, 1) { + out_msg.addr := getRegionBase(address); + out_msg.Type := CoherenceRequestType:PrivateNotify; + out_msg.Destination.add(in_msg.Requestor); + out_msg.Requestor := machineID; + out_msg.MessageSize := MessageSizeType:Request_Control; + out_msg.InitialRequestTime := in_msg.InitialRequestTime; + } + } + } + + action(bs_sendSharedNotice, "bs", desc="Send notice to private cache that it has private access") { + peek(requestNetwork_in, CPURequestMsg) { + enqueue(notifyNetwork_out, CPURequestMsg, 1) { + out_msg.addr := getRegionBase(address); + out_msg.Type := CoherenceRequestType:SharedNotify; + out_msg.Destination.add(in_msg.Requestor); + out_msg.Requestor := machineID; + out_msg.MessageSize := MessageSizeType:Request_Control; + out_msg.InitialRequestTime := in_msg.InitialRequestTime; + } + } + } + + action(c_sendSharedNoticeToOrigReq, "c", desc="Send notice to private cache that it has shared access") { + assert(is_valid(tbe)); + enqueue(notifyNetwork_out, CPURequestMsg, 1) { + out_msg.addr := getRegionBase(address); + out_msg.Type := CoherenceRequestType:SharedNotify; + out_msg.Destination.add(tbe.Owner); + out_msg.Requestor := machineID; + out_msg.MessageSize := MessageSizeType:Request_Control; + out_msg.ProbeRequestStartTime := tbe.ProbeRequestTime; + out_msg.InitialRequestTime := tbe.InitialRequestTime; + APPEND_TRANSITION_COMMENT("dest: "); + APPEND_TRANSITION_COMMENT(out_msg.Destination); + } + } + + action(sp_sendPrivateNoticeToOrigReq, "sp", desc="Send notice to private cache that it has private access") { + assert(is_valid(tbe)); + enqueue(notifyNetwork_out, CPURequestMsg, 1) { + out_msg.addr := getRegionBase(address); + out_msg.Type := CoherenceRequestType:PrivateNotify; + out_msg.Destination.add(tbe.Owner); + out_msg.Requestor := machineID; + out_msg.MessageSize := MessageSizeType:Request_Control; + out_msg.ProbeRequestStartTime := tbe.ProbeRequestTime; + out_msg.InitialRequestTime := tbe.InitialRequestTime; + APPEND_TRANSITION_COMMENT("dest: "); + APPEND_TRANSITION_COMMENT(out_msg.Destination); + } + } + + action(i_RegionInvNotify, "i", desc="Send notice to private cache that it no longer has private access") { + enqueue(probeNetwork_out, NBProbeRequestMsg, 1) { + out_msg.addr := address; + out_msg.DemandAddress := tbe.DemandAddress; + //out_msg.Requestor := tbe.Requestor; + out_msg.Requestor := machineID; + out_msg.Type := ProbeRequestType:PrbInv; + //Fix me: assert(tbe.Sharers.count() > 0); + out_msg.DemandRequest := true; + out_msg.Destination := tbe.Sharers; + out_msg.MessageSize := MessageSizeType:Request_Control; + APPEND_TRANSITION_COMMENT("dest: "); + APPEND_TRANSITION_COMMENT(out_msg.Destination); + } + } + + action(i0_RegionInvNotifyDemand0, "i0", desc="Send notice to private cache that it no longer has private access") { + enqueue(probeNetwork_out, NBProbeRequestMsg, 1) { + out_msg.addr := address; + // Demand address should default to 0 -> out_msg.DemandAddress := 0; + out_msg.Requestor := machineID; + out_msg.Type := ProbeRequestType:PrbInv; + out_msg.Destination := tbe.Sharers; + out_msg.MessageSize := MessageSizeType:Request_Control; + APPEND_TRANSITION_COMMENT("dest: "); + APPEND_TRANSITION_COMMENT(out_msg.Destination); + } + } + + action(rd_RegionDowngrade, "rd", desc="Send notice to private cache that it only has shared access") { + enqueue(probeNetwork_out, NBProbeRequestMsg, 1) { + out_msg.addr := address; + out_msg.DemandAddress := tbe.DemandAddress; + out_msg.Requestor := machineID; + out_msg.Type := ProbeRequestType:PrbDowngrade; + out_msg.DemandRequest := true; + out_msg.Destination := tbe.Sharers; + out_msg.MessageSize := MessageSizeType:Request_Control; + APPEND_TRANSITION_COMMENT("dest: "); + APPEND_TRANSITION_COMMENT(out_msg.Destination); + } + } + + action(p_popRequestQueue, "p", desc="Pop the request queue") { + requestNetwork_in.dequeue(clockEdge()); + } + + action(pt_popTriggerQueue, "pt", desc="Pop the trigger queue") { + triggerQueue_in.dequeue(clockEdge()); + } + + action(pr_popResponseQueue, "pr", desc="Pop the response queue") { + responseNetwork_in.dequeue(clockEdge()); + } + + action(s_stallAndWaitRequest, "s", desc="Stall and wait on the region address") { + Addr regAddr := getRegionBase(address); + stall_and_wait(requestNetwork_in, regAddr); + } + + action(w_wakeUpRegionDependents, "w", desc="Wake up any requests waiting for this region") { + wakeUpBuffers(getRegionBase(address)); + } + + action(wa_wakeUpAllDependents, "wa", desc="Wake up any requests waiting for this region") { + wakeUpAllBuffers(); + } + + action(zz_recycleRequestQueue, "\z", desc="...") { + requestNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); + } + + action(z_stall, "z", desc="stall request queue") { + // fake state + } + + action(mru_setMRU, "mru", desc="set MRU") { + cacheMemory.setMRU(address); + } + + // Transistions + + transition({NP_P, P_P, NP_S, S_S, S_P, P_S, P_NP, S_AP, P_AS, P_AP, SP_NP_W, S_W, P_AP_W, P_AS_W, S_AP_W}, {PrivateRequest, SharedRequest, UpgradeRequest, SendInv, SendUpgrade, SendDowngrade, CleanWbRequest, CleanWbRequest_LastSharer, StaleCleanWbRequest}) { + s_stallAndWaitRequest + } + + transition({NP_P, P_P, NP_S, S_S, S_P, S_W, P_S, P_NP, S_AP, P_AS, P_AP, P_AP_W, P_AS_W, S_AP_W}, Evict) { + zz_recycleRequestQueue; + } + + transition(NP, {PrivateRequest, SendUpgrade}, NP_P) {TagArrayRead, TagArrayWrite} { + a_allocateRegionEntry; + ur_updateDirtyStatusOnRequest; + f_fwdReqToDir; + b_sendPrivateNotice; + so_setOwner; + ss_setSharers; + t_allocateTBE; + p_popRequestQueue; + } + + transition(P, {PrivateRequest, UpgradeRequest}, P_P) {TagArrayRead} { + mru_setMRU; + ur_updateDirtyStatusOnRequest; + f_fwdReqToDir; + b_sendPrivateNotice; + t_allocateTBE; + p_popRequestQueue; + } + + transition({NP_P, P_P}, CPUPrivateAck, P) { + dt_deallocateTBE; + w_wakeUpRegionDependents; + pr_popResponseQueue; + } + + transition({NP, P, S}, StaleCleanWbRequest) {TagArrayRead, TagArrayWrite} { + wbn_sendWbNoticeNoAck; + ud_updateDirtyStatusWithWb; + p_popRequestQueue; + } + + transition(NP, SharedRequest, NP_S) {TagArrayRead, TagArrayWrite} { + a_allocateRegionEntry; + ur_updateDirtyStatusOnRequest; + f_fwdReqToDirShared; + bs_sendSharedNotice; + so_setOwner; + ss_setSharers; + t_allocateTBE; + p_popRequestQueue; + } + + // Could probably do this in parallel with other shared requests + transition(S, SharedRequest, S_S) {TagArrayRead, TagArrayWrite} { + mru_setMRU; + ur_updateDirtyStatusOnRequest; + f_fwdReqToDirShared; + bs_sendSharedNotice; + ss_setSharers; + t_allocateTBE; + p_popRequestQueue; + } + + transition({P, S}, CleanWbRequest_LastSharer, SP_NP_W) {TagArrayRead, TagArrayWrite} { + ud_updateDirtyStatusWithWb; + wb_sendWbNotice; + rs_removeSharer; + t_allocateTBE; + d_deallocateRegionEntry; + p_popRequestQueue; + } + + transition(S, CleanWbRequest, S_W) {TagArrayRead, TagArrayWrite} { + ud_updateDirtyStatusWithWb; + wb_sendWbNotice; + rs_removeSharer; + t_allocateTBE; + p_popRequestQueue; + } + + transition(SP_NP_W, WritebackAck, NP) { + dt_deallocateTBE; + w_wakeUpRegionDependents; + pr_popResponseQueue; + } + + transition(S_W, WritebackAck, S) { + dt_deallocateTBE; + w_wakeUpRegionDependents; + pr_popResponseQueue; + } + + transition({NP_S, S_S}, CPUPrivateAck, S) { + dt_deallocateTBE; + w_wakeUpRegionDependents; + pr_popResponseQueue; + } + + transition(S, UpgradeRequest, S_P) {TagArrayRead, TagArrayWrite} { + mru_setMRU; + ur_updateDirtyStatusOnRequest; + f_fwdReqToDir; + b_sendPrivateNotice; + so_setOwner; + t_allocateTBE; + p_popRequestQueue; + } + + transition(S_P, CPUPrivateAck, P) { + dt_deallocateTBE; + w_wakeUpRegionDependents; + pr_popResponseQueue; + } + + transition(P, SendInv, P_AP_W) {TagArrayRead, TagArrayWrite} { + mru_setMRU; + ur_updateDirtyStatusOnRequest; + f_fwdReqToDirWithAck; + so_setOwner; + t_allocateTBE; + rr_removeRequestorFromTBE; + sns_setNumAcksSharers; + cs_clearSharers; + ss_setSharers; + //i_RegionInvNotify; + p_popRequestQueue; + } + + transition({P_AP_W, S_AP_W}, DirReadyAck) { + ti_triggerInv; + pr_popResponseQueue; + } + + transition(P_AS_W, DirReadyAck) { + td_triggerDowngrade; + pr_popResponseQueue; + } + + transition(P_AS_W, TriggerDowngrade, P_AS) { + rd_RegionDowngrade; + pt_popTriggerQueue; + } + + transition(P_AP_W, TriggerInv, P_AP) { + i_RegionInvNotify; + pt_popTriggerQueue; + } + + transition(S_AP_W, TriggerInv, S_AP) { + i_RegionInvNotify; + pt_popTriggerQueue; + } + + transition(P, SendUpgrade, P_AP_W) {TagArrayRead, TagArrayWrite} { + mru_setMRU; + ur_updateDirtyStatusOnRequest; + f_fwdReqToDirWithAck; + so_setOwner; + t_allocateTBE; + rr_removeRequestorFromTBE; + sns_setNumAcksSharers; + cs_clearSharers; + ss_setSharers; + p_popRequestQueue; + } + + transition(P, Evict, P_NP) {TagArrayRead, TagArrayWrite} { + t_allocateTBE; + sns_setNumAcksSharers; + i0_RegionInvNotifyDemand0; + d_deallocateRegionEntry; + } + + transition(S, SendInv, P_AP_W) {TagArrayRead, TagArrayWrite} { + mru_setMRU; + ur_updateDirtyStatusOnRequest; + f_fwdReqToDirWithAck; + so_setOwner; + t_allocateTBE; + rr_removeRequestorFromTBE; + sns_setNumAcksSharers; + cs_clearSharers; + ss_setSharers; + p_popRequestQueue; + } + + transition(S, Evict, P_NP) {TagArrayRead, TagArrayWrite} { + t_allocateTBE; + sns_setNumAcksSharers; + i0_RegionInvNotifyDemand0; + d_deallocateRegionEntry; + } + + transition(P_NP, LastAck, NP) { + dt_deallocateTBE; + wa_wakeUpAllDependents; + pt_popTriggerQueue; + } + + transition(S, SendUpgrade, S_AP_W) {TagArrayRead, TagArrayWrite} { + mru_setMRU; + ur_updateDirtyStatusOnRequest; + f_fwdReqToDirWithAck; + so_setOwner; + t_allocateTBE; + rr_removeRequestorFromTBE; + sns_setNumAcksSharers; + cs_clearSharers; + ss_setSharers; + p_popRequestQueue; + } + + transition(S_AP, LastAck, S_P) { + sp_sendPrivateNoticeToOrigReq; + pt_popTriggerQueue; + } + + transition(P_AP, LastAck, P_P) { + sp_sendPrivateNoticeToOrigReq; + pt_popTriggerQueue; + } + + transition(P, SendDowngrade, P_AS_W) {TagArrayRead, TagArrayWrite} { + mru_setMRU; + ur_updateDirtyStatusOnRequest; + f_fwdReqToDirWithAckShared; + so_setOwner; + t_allocateTBE; + sns_setNumAcksSharers; + ss_setSharers; //why do we set the sharers before sending the downgrade? Are we sending a downgrade to the requestor? + p_popRequestQueue; + } + + transition(P_AS, LastAck, P_S) { + c_sendSharedNoticeToOrigReq; + pt_popTriggerQueue; + } + + transition(P_S, CPUPrivateAck, S) { + dt_deallocateTBE; + w_wakeUpRegionDependents; + pr_popResponseQueue; + } + + transition({P_NP, P_AS, S_AP, P_AP}, InvAckCore) {} { + ra_receiveAck; + pr_popResponseQueue; + } + + transition({P_NP, S_AP, P_AP}, InvAckCoreNoShare) {} { + ra_receiveAck; + pr_popResponseQueue; + } + + transition(P_AS, InvAckCoreNoShare) {} { + ra_receiveAck; + rsr_removeSharerResponse; + pr_popResponseQueue; + } + +} + + |