diff options
Diffstat (limited to 'src/mem/protocol/GPU_RfO-TCP.sm')
| -rw-r--r-- | src/mem/protocol/GPU_RfO-TCP.sm | 1009 |
1 files changed, 1009 insertions, 0 deletions
diff --git a/src/mem/protocol/GPU_RfO-TCP.sm b/src/mem/protocol/GPU_RfO-TCP.sm new file mode 100644 index 000000000..6cf9224a6 --- /dev/null +++ b/src/mem/protocol/GPU_RfO-TCP.sm @@ -0,0 +1,1009 @@ +/* + * Copyright (c) 2011-2015 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: Lisa Hsu + */ + +machine(MachineType:TCP, "GPU TCP (L1 Data Cache)") + : GPUCoalescer* coalescer; + Sequencer* sequencer; + bool use_seq_not_coal; + CacheMemory * L1cache; + int TCC_select_num_bits; + Cycles issue_latency := 40; // time to send data down to TCC + Cycles l2_hit_latency := 18; + + MessageBuffer * requestFromTCP, network="To", virtual_network="1", vnet_type="request"; + MessageBuffer * responseFromTCP, network="To", virtual_network="3", vnet_type="response"; + MessageBuffer * unblockFromCore, network="To", virtual_network="5", vnet_type="unblock"; + + MessageBuffer * probeToTCP, network="From", virtual_network="1", vnet_type="request"; + MessageBuffer * responseToTCP, network="From", virtual_network="3", vnet_type="response"; + + MessageBuffer * mandatoryQueue; +{ + state_declaration(State, desc="TCP Cache States", default="TCP_State_I") { + I, AccessPermission:Invalid, desc="Invalid"; + S, AccessPermission:Read_Only, desc="Shared"; + E, AccessPermission:Read_Write, desc="Exclusive"; + O, AccessPermission:Read_Only, desc="Owner state in core, both clusters and other cores may be sharing line"; + M, AccessPermission:Read_Write, desc="Modified"; + + I_M, AccessPermission:Busy, desc="Invalid, issued RdBlkM, have not seen response yet"; + I_ES, AccessPermission:Busy, desc="Invalid, issued RdBlk, have not seen response yet"; + S_M, AccessPermission:Read_Only, desc="Shared, issued CtoD, have not seen response yet"; + O_M, AccessPermission:Read_Only, desc="Shared, issued CtoD, have not seen response yet"; + + ES_I, AccessPermission:Read_Only, desc="L1 replacement, waiting for clean WB ack"; + MO_I, AccessPermission:Read_Only, desc="L1 replacement, waiting for dirty WB ack"; + + MO_PI, AccessPermission:Read_Only, desc="L1 downgrade, waiting for CtoD ack (or ProbeInvalidateData)"; + + I_C, AccessPermission:Invalid, desc="Invalid, waiting for WBAck from TCC for canceled WB"; + } + + enumeration(Event, desc="TCP Events") { + // Core initiated + Load, desc="Load"; + Store, desc="Store"; + + // TCC initiated + TCC_AckS, desc="TCC Ack to Core Request"; + TCC_AckE, desc="TCC Ack to Core Request"; + TCC_AckM, desc="TCC Ack to Core Request"; + TCC_AckCtoD, desc="TCC Ack to Core Request"; + TCC_AckWB, desc="TCC Ack for clean WB"; + TCC_NackWB, desc="TCC Nack for clean WB"; + + // Mem sys initiated + Repl, desc="Replacing block from cache"; + + // Probe Events + PrbInvData, desc="probe, return O or M data"; + PrbInv, desc="probe, no need for data"; + LocalPrbInv, desc="local probe, no need for data"; + PrbShrData, desc="probe downgrade, return O or M data"; + } + + 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(Entry, desc="...", interface="AbstractCacheEntry") { + State CacheState, desc="cache state"; + bool Dirty, desc="Is the data dirty (diff than memory)?"; + DataBlock DataBlk, desc="data for the block"; + bool FromL2, default="false", desc="block just moved from L2"; + } + + structure(TBE, desc="...") { + State TBEState, desc="Transient state"; + DataBlock DataBlk, desc="data for the block, required for concurrent writebacks"; + bool Dirty, desc="Is the data dirty (different than memory)?"; + int NumPendingMsgs, desc="Number of acks/data messages that this processor is waiting for"; + bool Shared, desc="Victim hit by shared probe"; + } + + structure(TBETable, external="yes") { + TBE lookup(Addr); + void allocate(Addr); + void deallocate(Addr); + bool isPresent(Addr); + } + + TBETable TBEs, template="<TCP_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(); + + // Internal functions + Entry getCacheEntry(Addr address), return_by_pointer="yes" { + Entry cache_entry := static_cast(Entry, "pointer", L1cache.lookup(address)); + return cache_entry; + } + + DataBlock getDataBlock(Addr addr), return_by_ref="yes" { + TBE tbe := TBEs.lookup(addr); + if(is_valid(tbe)) { + return tbe.DataBlk; + } else { + return getCacheEntry(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.CacheState; + } + return State:I; + } + + 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.CacheState := state; + } + } + + AccessPermission getAccessPermission(Addr addr) { + TBE tbe := TBEs.lookup(addr); + if(is_valid(tbe)) { + return TCP_State_to_permission(tbe.TBEState); + } + + Entry cache_entry := getCacheEntry(addr); + if(is_valid(cache_entry)) { + return TCP_State_to_permission(cache_entry.CacheState); + } + + return AccessPermission:NotPresent; + } + + bool isValid(Addr addr) { + AccessPermission perm := getAccessPermission(addr); + if (perm == AccessPermission:NotPresent || + perm == AccessPermission:Invalid || + perm == AccessPermission:Busy) { + return false; + } else { + return true; + } + } + + void setAccessPermission(Entry cache_entry, Addr addr, State state) { + if (is_valid(cache_entry)) { + cache_entry.changePermission(TCP_State_to_permission(state)); + } + } + + void functionalRead(Addr addr, Packet *pkt) { + TBE tbe := TBEs.lookup(addr); + if(is_valid(tbe)) { + testAndRead(addr, tbe.DataBlk, pkt); + } else { + functionalMemoryRead(pkt); + } + } + + int functionalWrite(Addr addr, Packet *pkt) { + int num_functional_writes := 0; + + TBE tbe := TBEs.lookup(addr); + if(is_valid(tbe)) { + num_functional_writes := num_functional_writes + + testAndWrite(addr, tbe.DataBlk, pkt); + } + + num_functional_writes := num_functional_writes + functionalMemoryWrite(pkt); + return num_functional_writes; + } + + void recordRequestType(RequestType request_type, Addr addr) { + if (request_type == RequestType:DataArrayRead) { + L1cache.recordRequestType(CacheRequestType:DataArrayRead, addr); + } else if (request_type == RequestType:DataArrayWrite) { + L1cache.recordRequestType(CacheRequestType:DataArrayWrite, addr); + } else if (request_type == RequestType:TagArrayRead) { + L1cache.recordRequestType(CacheRequestType:TagArrayRead, addr); + } else if (request_type == RequestType:TagArrayWrite) { + L1cache.recordRequestType(CacheRequestType:TagArrayWrite, addr); + } + } + + bool checkResourceAvailable(RequestType request_type, Addr addr) { + if (request_type == RequestType:DataArrayRead) { + return L1cache.checkResourceAvailable(CacheResourceType:DataArray, addr); + } else if (request_type == RequestType:DataArrayWrite) { + return L1cache.checkResourceAvailable(CacheResourceType:DataArray, addr); + } else if (request_type == RequestType:TagArrayRead) { + return L1cache.checkResourceAvailable(CacheResourceType:TagArray, addr); + } else if (request_type == RequestType:TagArrayWrite) { + return L1cache.checkResourceAvailable(CacheResourceType:TagArray, addr); + } else { + error("Invalid RequestType type in checkResourceAvailable"); + return true; + } + } + + MachineType getCoherenceType(MachineID myMachID, + MachineID senderMachID) { + if(myMachID == senderMachID) { + return MachineType:TCP; + } else if(machineIDToMachineType(senderMachID) == MachineType:TCP) { + return MachineType:L1Cache_wCC; + } else if(machineIDToMachineType(senderMachID) == MachineType:TCC) { + return MachineType:TCC; + } else { + return MachineType:TCCdir; + } + } + + // Out Ports + + out_port(requestNetwork_out, CPURequestMsg, requestFromTCP); + out_port(responseNetwork_out, ResponseMsg, responseFromTCP); + out_port(unblockNetwork_out, UnblockMsg, unblockFromCore); + + // In Ports + + in_port(probeNetwork_in, TDProbeRequestMsg, probeToTCP) { + if (probeNetwork_in.isReady(clockEdge())) { + peek(probeNetwork_in, TDProbeRequestMsg, block_on="addr") { + DPRINTF(RubySlicc, "%s\n", in_msg); + DPRINTF(RubySlicc, "machineID: %s\n", machineID); + Entry cache_entry := getCacheEntry(in_msg.addr); + TBE tbe := TBEs.lookup(in_msg.addr); + + if (in_msg.Type == ProbeRequestType:PrbInv) { + if (in_msg.ReturnData) { + trigger(Event:PrbInvData, in_msg.addr, cache_entry, tbe); + } else { + if(in_msg.localCtoD) { + trigger(Event:LocalPrbInv, in_msg.addr, cache_entry, tbe); + } else { + trigger(Event:PrbInv, in_msg.addr, cache_entry, tbe); + } + } + } else if (in_msg.Type == ProbeRequestType:PrbDowngrade) { + assert(in_msg.ReturnData); + trigger(Event:PrbShrData, in_msg.addr, cache_entry, tbe); + } + } + } + } + + in_port(responseToTCP_in, ResponseMsg, responseToTCP) { + if (responseToTCP_in.isReady(clockEdge())) { + peek(responseToTCP_in, ResponseMsg, block_on="addr") { + + Entry cache_entry := getCacheEntry(in_msg.addr); + TBE tbe := TBEs.lookup(in_msg.addr); + + if (in_msg.Type == CoherenceResponseType:TDSysResp) { + if (in_msg.State == CoherenceState:Modified) { + if (in_msg.CtoD) { + trigger(Event:TCC_AckCtoD, in_msg.addr, cache_entry, tbe); + } else { + trigger(Event:TCC_AckM, in_msg.addr, cache_entry, tbe); + } + } else if (in_msg.State == CoherenceState:Shared) { + trigger(Event:TCC_AckS, in_msg.addr, cache_entry, tbe); + } else if (in_msg.State == CoherenceState:Exclusive) { + trigger(Event:TCC_AckE, in_msg.addr, cache_entry, tbe); + } + } else if (in_msg.Type == CoherenceResponseType:TDSysWBAck) { + trigger(Event:TCC_AckWB, in_msg.addr, cache_entry, tbe); + } else if (in_msg.Type == CoherenceResponseType:TDSysWBNack) { + trigger(Event:TCC_NackWB, in_msg.addr, cache_entry, tbe); + } else { + error("Unexpected Response Message to Core"); + } + } + } + } + + in_port(mandatoryQueue_in, RubyRequest, mandatoryQueue, desc="...") { + if (mandatoryQueue_in.isReady(clockEdge())) { + peek(mandatoryQueue_in, RubyRequest, block_on="LineAddress") { + Entry cache_entry := getCacheEntry(in_msg.LineAddress); + TBE tbe := TBEs.lookup(in_msg.LineAddress); + DPRINTF(RubySlicc, "%s\n", in_msg); + if (in_msg.Type == RubyRequestType:LD) { + if (is_valid(cache_entry) || L1cache.cacheAvail(in_msg.LineAddress)) { + trigger(Event:Load, in_msg.LineAddress, cache_entry, tbe); + } else { + Addr victim := L1cache.cacheProbe(in_msg.LineAddress); + trigger(Event:Repl, victim, getCacheEntry(victim), TBEs.lookup(victim)); + } + } else { + if (is_valid(cache_entry) || L1cache.cacheAvail(in_msg.LineAddress)) { + trigger(Event:Store, in_msg.LineAddress, cache_entry, tbe); + } else { + Addr victim := L1cache.cacheProbe(in_msg.LineAddress); + trigger(Event:Repl, victim, getCacheEntry(victim), TBEs.lookup(victim)); + } + } + } + } + } + + // Actions + + action(ic_invCache, "ic", desc="invalidate cache") { + if(is_valid(cache_entry)) { + L1cache.deallocate(address); + } + unset_cache_entry(); + } + + action(n_issueRdBlk, "n", desc="Issue RdBlk") { + enqueue(requestNetwork_out, CPURequestMsg, issue_latency) { + out_msg.addr := address; + out_msg.Type := CoherenceRequestType:RdBlk; + out_msg.Requestor := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.MessageSize := MessageSizeType:Request_Control; + out_msg.InitialRequestTime := curCycle(); + } + } + + action(nM_issueRdBlkM, "nM", desc="Issue RdBlkM") { + enqueue(requestNetwork_out, CPURequestMsg, issue_latency) { + out_msg.addr := address; + out_msg.Type := CoherenceRequestType:RdBlkM; + out_msg.Requestor := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.MessageSize := MessageSizeType:Request_Control; + out_msg.InitialRequestTime := curCycle(); + } + } + + action(vd_victim, "vd", desc="Victimize M/O Data") { + enqueue(requestNetwork_out, CPURequestMsg, issue_latency) { + out_msg.addr := address; + out_msg.Requestor := machineID; + assert(is_valid(cache_entry)); + out_msg.DataBlk := cache_entry.DataBlk; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.MessageSize := MessageSizeType:Request_Control; + out_msg.Type := CoherenceRequestType:VicDirty; + out_msg.InitialRequestTime := curCycle(); + if (cache_entry.CacheState == State:O) { + out_msg.Shared := true; + } else { + out_msg.Shared := false; + } + out_msg.Dirty := cache_entry.Dirty; + } + } + + action(vc_victim, "vc", desc="Victimize E/S Data") { + enqueue(requestNetwork_out, CPURequestMsg, issue_latency) { + out_msg.addr := address; + out_msg.Requestor := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.MessageSize := MessageSizeType:Request_Control; + out_msg.Type := CoherenceRequestType:VicClean; + out_msg.InitialRequestTime := curCycle(); + if (cache_entry.CacheState == State:S) { + out_msg.Shared := true; + } else { + out_msg.Shared := false; + } + } + } + + action(a_allocate, "a", desc="allocate block") { + if (is_invalid(cache_entry)) { + set_cache_entry(L1cache.allocate(address, new Entry)); + } + } + + action(t_allocateTBE, "t", desc="allocate TBE Entry") { + check_allocate(TBEs); + assert(is_valid(cache_entry)); + TBEs.allocate(address); + set_tbe(TBEs.lookup(address)); + tbe.DataBlk := cache_entry.DataBlk; // Data only used for WBs + tbe.Dirty := cache_entry.Dirty; + tbe.Shared := false; + } + + action(d_deallocateTBE, "d", desc="Deallocate TBE") { + TBEs.deallocate(address); + unset_tbe(); + } + + action(p_popMandatoryQueue, "pm", desc="Pop Mandatory Queue") { + mandatoryQueue_in.dequeue(clockEdge()); + } + + action(pr_popResponseQueue, "pr", desc="Pop Response Queue") { + responseToTCP_in.dequeue(clockEdge()); + } + + action(pp_popProbeQueue, "pp", desc="pop probe queue") { + probeNetwork_in.dequeue(clockEdge()); + } + + action(l_loadDone, "l", desc="local load done") { + assert(is_valid(cache_entry)); + if (use_seq_not_coal) { + sequencer.readCallback(address, cache_entry.DataBlk, + false, MachineType:TCP); + } else { + coalescer.readCallback(address, MachineType:TCP, cache_entry.DataBlk); + } + } + + action(xl_loadDone, "xl", desc="remote load done") { + peek(responseToTCP_in, ResponseMsg) { + assert(is_valid(cache_entry)); + if (use_seq_not_coal) { + coalescer.recordCPReadCallBack(machineID, in_msg.Sender); + sequencer.readCallback(address, + cache_entry.DataBlk, + false, + machineIDToMachineType(in_msg.Sender), + in_msg.InitialRequestTime, + in_msg.ForwardRequestTime, + in_msg.ProbeRequestStartTime); + } else { + MachineType cc_mach_type := getCoherenceType(machineID, + in_msg.Sender); + coalescer.readCallback(address, + cc_mach_type, + cache_entry.DataBlk, + in_msg.InitialRequestTime, + in_msg.ForwardRequestTime, + in_msg.ProbeRequestStartTime); + } + } + } + + action(s_storeDone, "s", desc="local store done") { + assert(is_valid(cache_entry)); + if (use_seq_not_coal) { + coalescer.recordCPWriteCallBack(machineID, machineID); + sequencer.writeCallback(address, cache_entry.DataBlk, + false, MachineType:TCP); + } else { + coalescer.writeCallback(address, MachineType:TCP, cache_entry.DataBlk); + } + cache_entry.Dirty := true; + } + + action(xs_storeDone, "xs", desc="remote store done") { + peek(responseToTCP_in, ResponseMsg) { + assert(is_valid(cache_entry)); + if (use_seq_not_coal) { + coalescer.recordCPWriteCallBack(machineID, in_msg.Sender); + sequencer.writeCallback(address, + cache_entry.DataBlk, + false, + machineIDToMachineType(in_msg.Sender), + in_msg.InitialRequestTime, + in_msg.ForwardRequestTime, + in_msg.ProbeRequestStartTime); + } else { + MachineType cc_mach_type := getCoherenceType(machineID, + in_msg.Sender); + coalescer.writeCallback(address, + cc_mach_type, + cache_entry.DataBlk, + in_msg.InitialRequestTime, + in_msg.ForwardRequestTime, + in_msg.ProbeRequestStartTime); + } + cache_entry.Dirty := true; + } + } + + action(w_writeCache, "w", desc="write data to cache") { + peek(responseToTCP_in, ResponseMsg) { + assert(is_valid(cache_entry)); + cache_entry.DataBlk := in_msg.DataBlk; + cache_entry.Dirty := in_msg.Dirty; + } + } + + action(ss_sendStaleNotification, "ss", desc="stale data; nothing to writeback") { + peek(responseToTCP_in, ResponseMsg) { + enqueue(responseNetwork_out, ResponseMsg, issue_latency) { + out_msg.addr := address; + out_msg.Type := CoherenceResponseType:StaleNotif; + out_msg.Sender := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCC, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.MessageSize := MessageSizeType:Response_Control; + DPRINTF(RubySlicc, "%s\n", out_msg); + } + } + } + + action(wb_data, "wb", desc="write back data") { + peek(responseToTCP_in, ResponseMsg) { + enqueue(responseNetwork_out, ResponseMsg, issue_latency) { + out_msg.addr := address; + out_msg.Type := CoherenceResponseType:CPUData; + out_msg.Sender := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCC, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.DataBlk := tbe.DataBlk; + out_msg.Dirty := tbe.Dirty; + if (tbe.Shared) { + out_msg.NbReqShared := true; + } else { + out_msg.NbReqShared := false; + } + out_msg.State := CoherenceState:Shared; // faux info + out_msg.MessageSize := MessageSizeType:Writeback_Data; + DPRINTF(RubySlicc, "%s\n", out_msg); + } + } + } + + action(piu_sendProbeResponseInvUntransferredOwnership, "piu", desc="send probe ack inv, no data, retain ownership") { + enqueue(responseNetwork_out, ResponseMsg, issue_latency) { + out_msg.addr := address; + out_msg.Type := CoherenceResponseType:CPUPrbResp; // TCC, L3 respond in same way to probes + out_msg.Sender := machineID; + // will this always be ok? probably not for multisocket + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.Dirty := false; + out_msg.Hit := false; + out_msg.Ntsl := true; + out_msg.State := CoherenceState:NA; + out_msg.UntransferredOwner :=true; + out_msg.MessageSize := MessageSizeType:Response_Control; + } + } + + action(pi_sendProbeResponseInv, "pi", desc="send probe ack inv, no data") { + enqueue(responseNetwork_out, ResponseMsg, issue_latency) { + out_msg.addr := address; + out_msg.Type := CoherenceResponseType:CPUPrbResp; // TCC, L3 respond in same way to probes + out_msg.Sender := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.Dirty := false; + out_msg.Hit := false; + out_msg.Ntsl := true; + out_msg.State := CoherenceState:NA; + out_msg.MessageSize := MessageSizeType:Response_Control; + out_msg.isValid := isValid(address); + } + } + + action(pim_sendProbeResponseInvMs, "pim", desc="send probe ack inv, no data") { + enqueue(responseNetwork_out, ResponseMsg, issue_latency) { + out_msg.addr := address; + out_msg.Type := CoherenceResponseType:CPUPrbResp; // L3 and TCC respond in same way to probes + out_msg.Sender := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.Dirty := false; + out_msg.Ntsl := true; + out_msg.Hit := false; + out_msg.State := CoherenceState:NA; + out_msg.MessageSize := MessageSizeType:Response_Control; + out_msg.isValid := isValid(address); + } + } + + action(prm_sendProbeResponseMiss, "prm", desc="send probe ack PrbShrData, no data") { + enqueue(responseNetwork_out, ResponseMsg, issue_latency) { + out_msg.addr := address; + out_msg.Type := CoherenceResponseType:CPUPrbResp; // L3 and TCC respond in same way to probes + out_msg.Sender := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.Dirty := false; // only true if sending back data i think + out_msg.Hit := false; + out_msg.Ntsl := false; + out_msg.State := CoherenceState:NA; + out_msg.MessageSize := MessageSizeType:Response_Control; + out_msg.isValid := isValid(address); + } + } + + action(pd_sendProbeResponseData, "pd", desc="send probe ack, with data") { + enqueue(responseNetwork_out, ResponseMsg, issue_latency) { + assert(is_valid(cache_entry) || is_valid(tbe)); + out_msg.addr := address; + out_msg.Type := CoherenceResponseType:CPUPrbResp; + out_msg.Sender := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.DataBlk := getDataBlock(address); + if (is_valid(tbe)) { + out_msg.Dirty := tbe.Dirty; + } else { + out_msg.Dirty := cache_entry.Dirty; + } + out_msg.Hit := true; + out_msg.State := CoherenceState:NA; + out_msg.MessageSize := MessageSizeType:Response_Data; + out_msg.isValid := isValid(address); + APPEND_TRANSITION_COMMENT("Sending ack with dirty "); + APPEND_TRANSITION_COMMENT(out_msg.Dirty); + } + } + + action(pdm_sendProbeResponseDataMs, "pdm", desc="send probe ack, with data") { + enqueue(responseNetwork_out, ResponseMsg, issue_latency) { + assert(is_valid(cache_entry) || is_valid(tbe)); + assert(is_valid(cache_entry)); + out_msg.addr := address; + out_msg.Type := CoherenceResponseType:CPUPrbResp; + out_msg.Sender := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.DataBlk := getDataBlock(address); + if (is_valid(tbe)) { + out_msg.Dirty := tbe.Dirty; + } else { + out_msg.Dirty := cache_entry.Dirty; + } + out_msg.Hit := true; + out_msg.State := CoherenceState:NA; + out_msg.MessageSize := MessageSizeType:Response_Data; + out_msg.isValid := isValid(address); + APPEND_TRANSITION_COMMENT("Sending ack with dirty "); + APPEND_TRANSITION_COMMENT(out_msg.Dirty); + DPRINTF(RubySlicc, "Data is %s\n", out_msg.DataBlk); + } + } + + action(sf_setSharedFlip, "sf", desc="hit by shared probe, status may be different") { + assert(is_valid(tbe)); + tbe.Shared := true; + } + + action(mru_updateMRU, "mru", desc="Touch block for replacement policy") { + L1cache.setMRU(address); + } + + action(uu_sendUnblock, "uu", desc="state changed, unblock") { + enqueue(unblockNetwork_out, UnblockMsg, issue_latency) { + out_msg.addr := address; + out_msg.Sender := machineID; + out_msg.Destination.add(mapAddressToRange(address,MachineType:TCCdir, + TCC_select_low_bit, TCC_select_num_bits)); + out_msg.MessageSize := MessageSizeType:Unblock_Control; + out_msg.wasValid := isValid(address); + DPRINTF(RubySlicc, "%s\n", out_msg); + } + } + + action(yy_recycleProbeQueue, "yy", desc="recycle probe queue") { + probeNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); + } + + action(zz_recycleMandatoryQueue, "\z", desc="recycle mandatory queue") { + mandatoryQueue_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); + } + + // Transitions + + // transitions from base + transition(I, Load, I_ES) {TagArrayRead} { + a_allocate; + n_issueRdBlk; + p_popMandatoryQueue; + } + + transition(I, Store, I_M) {TagArrayRead, TagArrayWrite} { + a_allocate; + nM_issueRdBlkM; + p_popMandatoryQueue; + } + + transition(S, Store, S_M) {TagArrayRead} { + mru_updateMRU; + nM_issueRdBlkM; + p_popMandatoryQueue; + } + + transition(E, Store, M) {TagArrayRead, TagArrayWrite, DataArrayWrite} { + mru_updateMRU; + s_storeDone; + p_popMandatoryQueue; + } + + transition(O, Store, O_M) {TagArrayRead, DataArrayWrite} { + mru_updateMRU; + nM_issueRdBlkM; + p_popMandatoryQueue; + } + + transition(M, Store) {TagArrayRead, DataArrayWrite} { + mru_updateMRU; + s_storeDone; + p_popMandatoryQueue; + } + + // simple hit transitions + transition({S, E, O, M}, Load) {TagArrayRead, DataArrayRead} { + l_loadDone; + mru_updateMRU; + p_popMandatoryQueue; + } + + // recycles from transients + transition({I_M, I_ES, ES_I, MO_I, S_M, O_M, MO_PI, I_C}, {Load, Store, Repl}) {} { + zz_recycleMandatoryQueue; + } + + transition({S, E}, Repl, ES_I) {TagArrayRead} { + t_allocateTBE; + vc_victim; + ic_invCache; + } + + transition({O, M}, Repl, MO_I) {TagArrayRead, DataArrayRead} { + t_allocateTBE; + vd_victim; + ic_invCache; + } + + // TD event transitions + transition(I_M, {TCC_AckM, TCC_AckCtoD}, M) {TagArrayRead, TagArrayWrite, DataArrayWrite} { + w_writeCache; + xs_storeDone; + uu_sendUnblock; + pr_popResponseQueue; + } + + transition(I_ES, TCC_AckS, S) {TagArrayWrite, DataArrayWrite} { + w_writeCache; + xl_loadDone; + uu_sendUnblock; + pr_popResponseQueue; + } + + transition(I_ES, TCC_AckE, E) {TagArrayWrite, DataArrayWrite} { + w_writeCache; + xl_loadDone; + uu_sendUnblock; + pr_popResponseQueue; + } + + transition({S_M, O_M}, TCC_AckM, M) {TagArrayWrite, DataArrayWrite} { + xs_storeDone; + uu_sendUnblock; + pr_popResponseQueue; + } + + transition({MO_I, ES_I}, TCC_NackWB, I){TagArrayWrite} { + d_deallocateTBE; + pr_popResponseQueue; + } + + transition({MO_I, ES_I}, TCC_AckWB, I) {TagArrayWrite, DataArrayRead} { + wb_data; + d_deallocateTBE; + pr_popResponseQueue; + } + + transition(I_C, TCC_AckWB, I) {TagArrayWrite} { + ss_sendStaleNotification; + d_deallocateTBE; + pr_popResponseQueue; + } + + transition(I_C, TCC_NackWB, I) {TagArrayWrite} { + d_deallocateTBE; + pr_popResponseQueue; + } + + // Probe transitions + transition({M, O}, PrbInvData, I) {TagArrayRead, TagArrayWrite} { + pd_sendProbeResponseData; + ic_invCache; + pp_popProbeQueue; + } + + transition(I, PrbInvData) {TagArrayRead, TagArrayWrite} { + prm_sendProbeResponseMiss; + pp_popProbeQueue; + } + + transition({E, S}, PrbInvData, I) {TagArrayRead, TagArrayWrite} { + pd_sendProbeResponseData; + ic_invCache; + pp_popProbeQueue; + } + + transition(I_C, PrbInvData, I_C) {} { + pi_sendProbeResponseInv; + ic_invCache; + pp_popProbeQueue; + } + + // Needed for TCC-based protocols. Must hold on to ownership till transfer complete + transition({M, O}, LocalPrbInv, MO_PI){TagArrayRead, TagArrayWrite} { + piu_sendProbeResponseInvUntransferredOwnership; + pp_popProbeQueue; + } + + // If there is a race and we see a probe invalidate, handle normally. + transition(MO_PI, PrbInvData, I){TagArrayWrite} { + pd_sendProbeResponseData; + ic_invCache; + pp_popProbeQueue; + } + + transition(MO_PI, PrbInv, I){TagArrayWrite} { + pi_sendProbeResponseInv; + ic_invCache; + pp_popProbeQueue; + } + + // normal exit when ownership is successfully transferred + transition(MO_PI, TCC_AckCtoD, I) {TagArrayWrite} { + ic_invCache; + pr_popResponseQueue; + } + + transition({M, O, E, S, I}, PrbInv, I) {TagArrayRead, TagArrayWrite} { + pi_sendProbeResponseInv; + ic_invCache; + pp_popProbeQueue; + } + + transition({E, S, I}, LocalPrbInv, I){TagArrayRead, TagArrayWrite} { + pi_sendProbeResponseInv; + ic_invCache; + pp_popProbeQueue; + } + + + transition({M, E, O}, PrbShrData, O) {TagArrayRead, TagArrayWrite, DataArrayRead} { + pd_sendProbeResponseData; + pp_popProbeQueue; + } + + transition(MO_PI, PrbShrData) {DataArrayRead} { + pd_sendProbeResponseData; + pp_popProbeQueue; + } + + + transition(S, PrbShrData, S) {TagArrayRead, DataArrayRead} { + pd_sendProbeResponseData; + pp_popProbeQueue; + } + + transition({I, I_C}, PrbShrData) {TagArrayRead} { + prm_sendProbeResponseMiss; + pp_popProbeQueue; + } + + transition(I_C, PrbInv, I_C) {} { + pi_sendProbeResponseInv; + ic_invCache; + pp_popProbeQueue; + } + + transition({I_M, I_ES}, {PrbInv, PrbInvData}){TagArrayRead} { + pi_sendProbeResponseInv; + ic_invCache; + a_allocate; // but make sure there is room for incoming data when it arrives + pp_popProbeQueue; + } + + transition({I_M, I_ES}, PrbShrData) {} { + prm_sendProbeResponseMiss; + pp_popProbeQueue; + } + + transition(S_M, PrbInvData, I_M) {TagArrayRead} { + pim_sendProbeResponseInvMs; + ic_invCache; + a_allocate; + pp_popProbeQueue; + } + + transition(O_M, PrbInvData, I_M) {TagArrayRead,DataArrayRead} { + pdm_sendProbeResponseDataMs; + ic_invCache; + a_allocate; + pp_popProbeQueue; + } + + transition({S_M, O_M}, {PrbInv}, I_M) {TagArrayRead} { + pim_sendProbeResponseInvMs; + ic_invCache; + a_allocate; + pp_popProbeQueue; + } + + transition(S_M, {LocalPrbInv}, I_M) {TagArrayRead} { + pim_sendProbeResponseInvMs; + ic_invCache; + a_allocate; + pp_popProbeQueue; + } + + transition(O_M, LocalPrbInv, I_M) {TagArrayRead} { + piu_sendProbeResponseInvUntransferredOwnership; + ic_invCache; + a_allocate; + pp_popProbeQueue; + } + + transition({S_M, O_M}, PrbShrData) {DataArrayRead} { + pd_sendProbeResponseData; + pp_popProbeQueue; + } + + transition(ES_I, PrbInvData, I_C){ + pd_sendProbeResponseData; + ic_invCache; + pp_popProbeQueue; + } + + transition(MO_I, PrbInvData, I_C) {DataArrayRead} { + pd_sendProbeResponseData; + ic_invCache; + pp_popProbeQueue; + } + + transition(MO_I, PrbInv, I_C) { + pi_sendProbeResponseInv; + ic_invCache; + pp_popProbeQueue; + } + + transition(ES_I, PrbInv, I_C) { + pi_sendProbeResponseInv; + ic_invCache; + pp_popProbeQueue; + } + + transition(ES_I, PrbShrData, ES_I) {DataArrayRead} { + pd_sendProbeResponseData; + sf_setSharedFlip; + pp_popProbeQueue; + } + + transition(MO_I, PrbShrData, MO_I) {DataArrayRead} { + pd_sendProbeResponseData; + sf_setSharedFlip; + pp_popProbeQueue; + } + +} |