From cd9e4458139658c4ce8f038e3a44bdecd17fa75d Mon Sep 17 00:00:00 2001 From: Nilay Vaish Date: Wed, 16 Sep 2015 11:59:56 -0500 Subject: [PATCH] ruby: message buffer, timer table: significant changes This patch changes MessageBuffer and TimerTable, two structures used for buffering messages by components in ruby. These structures would no longer maintain pointers to clock objects. Functions in these structures have been changed to take as input current time in Tick. Similarly, these structures will not operate on Cycle valued latencies for different operations. The corresponding functions would need to be provided with these latencies by components invoking the relevant functions. These latencies should also be in Ticks. I felt the need for these changes while trying to speed up ruby. The ultimate aim is to eliminate Consumer class and replace it with an EventManager object in the MessageBuffer and TimerTable classes. This object would be used for scheduling events. The event itself would contain information on the object and function to be invoked. In hindsight, it seems I should have done this while I was moving away from use of a single global clock in the memory system. That change led to introduction of clock objects that replaced the global clock object. It never crossed my mind that having clock object pointers is not a good design. And now I really don't like the fact that we have separate consumer, receiver and sender pointers in message buffers. --- src/mem/protocol/MESI_Three_Level-L0cache.sm | 14 ++- src/mem/protocol/MESI_Three_Level-L1cache.sm | 16 ++- src/mem/protocol/MESI_Two_Level-L1cache.sm | 20 ++-- src/mem/protocol/MESI_Two_Level-L2cache.sm | 21 ++-- src/mem/protocol/MESI_Two_Level-dir.sm | 18 +-- src/mem/protocol/MESI_Two_Level-dma.sm | 9 +- src/mem/protocol/MI_example-cache.sm | 16 ++- src/mem/protocol/MI_example-dir.sm | 19 ++-- src/mem/protocol/MI_example-dma.sm | 11 +- .../protocol/MOESI_CMP_directory-L1cache.sm | 33 +++--- .../protocol/MOESI_CMP_directory-L2cache.sm | 24 ++-- src/mem/protocol/MOESI_CMP_directory-dir.sm | 16 +-- src/mem/protocol/MOESI_CMP_directory-dma.sm | 13 ++- src/mem/protocol/MOESI_CMP_token-L1cache.sm | 77 +++++++------ src/mem/protocol/MOESI_CMP_token-L2cache.sm | 17 +-- src/mem/protocol/MOESI_CMP_token-dir.sm | 43 ++++---- src/mem/protocol/MOESI_CMP_token-dma.sm | 10 +- src/mem/protocol/MOESI_hammer-cache.sm | 17 +-- src/mem/protocol/MOESI_hammer-dir.sm | 25 +++-- src/mem/protocol/MOESI_hammer-dma.sm | 10 +- src/mem/protocol/Network_test-cache.sm | 5 +- src/mem/protocol/Network_test-dir.sm | 16 +-- src/mem/protocol/RubySlicc_Defines.sm | 2 +- src/mem/protocol/RubySlicc_Exports.sm | 1 + src/mem/protocol/RubySlicc_Types.sm | 12 +- src/mem/ruby/network/MessageBuffer.cc | 104 ++++++++---------- src/mem/ruby/network/MessageBuffer.hh | 57 +++------- src/mem/ruby/network/MessageBuffer.py | 1 - .../fixed-pipeline/NetworkInterface_d.cc | 16 +-- .../flexible-pipeline/NetworkInterface.cc | 15 +-- src/mem/ruby/network/simple/PerfectSwitch.cc | 12 +- src/mem/ruby/network/simple/SimpleNetwork.py | 6 - src/mem/ruby/network/simple/Switch.cc | 13 --- src/mem/ruby/network/simple/Throttle.cc | 20 ++-- .../slicc_interface/AbstractController.cc | 13 +-- src/mem/ruby/structures/TBETable.hh | 4 +- src/mem/ruby/structures/TimerTable.cc | 17 +-- src/mem/ruby/structures/TimerTable.hh | 21 +--- src/mem/ruby/system/DMASequencer.cc | 5 +- src/mem/ruby/system/RubyPort.cc | 1 - src/mem/ruby/system/Sequencer.cc | 2 +- src/mem/slicc/ast/EnqueueStatementAST.py | 5 +- src/mem/slicc/ast/ObjDeclAST.py | 2 + src/mem/slicc/ast/PeekStatementAST.py | 2 +- src/mem/slicc/ast/StallAndWaitStatementAST.py | 2 +- src/mem/slicc/symbols/StateMachine.py | 19 +--- 46 files changed, 383 insertions(+), 419 deletions(-) diff --git a/src/mem/protocol/MESI_Three_Level-L0cache.sm b/src/mem/protocol/MESI_Three_Level-L0cache.sm index 7e8626dc9..3f22a4906 100644 --- a/src/mem/protocol/MESI_Three_Level-L0cache.sm +++ b/src/mem/protocol/MESI_Three_Level-L0cache.sm @@ -135,6 +135,8 @@ machine(L0Cache, "MESI Directory L0 Cache") TBETable TBEs, template="", constructor="m_number_of_TBEs"; + Tick clockEdge(); + Cycles ticksToCycles(Tick t); void set_cache_entry(AbstractCacheEntry a); void unset_cache_entry(); void set_tbe(TBE a); @@ -255,7 +257,7 @@ machine(L0Cache, "MESI Directory L0 Cache") // Messages for this L0 cache from the L1 cache in_port(messgeBuffer_in, CoherenceMsg, bufferFromL1, rank = 1) { - if (messgeBuffer_in.isReady()) { + if (messgeBuffer_in.isReady(clockEdge())) { peek(messgeBuffer_in, CoherenceMsg, block_on="addr") { assert(in_msg.Dest == machineID); @@ -289,7 +291,7 @@ machine(L0Cache, "MESI Directory L0 Cache") // Mandatory Queue betweens Node's CPU and it's L0 caches in_port(mandatoryQueue_in, RubyRequest, mandatoryQueue, desc="...", rank = 0) { - if (mandatoryQueue_in.isReady()) { + if (mandatoryQueue_in.isReady(clockEdge())) { peek(mandatoryQueue_in, RubyRequest, block_on="LineAddress") { // Check for data access to blocks in I-cache and ifetchs to blocks in D-cache @@ -513,17 +515,19 @@ machine(L0Cache, "MESI Directory L0 Cache") } action(k_popMandatoryQueue, "k", desc="Pop mandatory queue.") { - mandatoryQueue_in.dequeue(); + mandatoryQueue_in.dequeue(clockEdge()); } action(l_popRequestQueue, "l", desc="Pop incoming request queue and profile the delay within this virtual network") { - profileMsgDelay(2, messgeBuffer_in.dequeue()); + Tick delay := messgeBuffer_in.dequeue(clockEdge()); + profileMsgDelay(2, ticksToCycles(delay)); } action(o_popIncomingResponseQueue, "o", desc="Pop Incoming Response queue and profile the delay within this virtual network") { - profileMsgDelay(1, messgeBuffer_in.dequeue()); + Tick delay := messgeBuffer_in.dequeue(clockEdge()); + profileMsgDelay(1, ticksToCycles(delay)); } action(s_deallocateTBE, "s", desc="Deallocate TBE") { diff --git a/src/mem/protocol/MESI_Three_Level-L1cache.sm b/src/mem/protocol/MESI_Three_Level-L1cache.sm index 6c8df8d75..0eb9a43b5 100644 --- a/src/mem/protocol/MESI_Three_Level-L1cache.sm +++ b/src/mem/protocol/MESI_Three_Level-L1cache.sm @@ -151,6 +151,8 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") int l2_select_low_bit, default="RubySystem::getBlockSizeBits()"; + Tick clockEdge(); + Cycles ticksToCycles(Tick t); void set_cache_entry(AbstractCacheEntry a); void unset_cache_entry(); void set_tbe(TBE a); @@ -266,7 +268,7 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") // Response From the L2 Cache to this L1 cache in_port(responseNetwork_in, ResponseMsg, responseFromL2, rank = 3) { - if (responseNetwork_in.isReady()) { + if (responseNetwork_in.isReady(clockEdge())) { peek(responseNetwork_in, ResponseMsg) { assert(in_msg.Destination.isElement(machineID)); @@ -303,7 +305,7 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") // Request to this L1 cache from the shared L2 in_port(requestNetwork_in, RequestMsg, requestFromL2, rank = 2) { - if(requestNetwork_in.isReady()) { + if(requestNetwork_in.isReady(clockEdge())) { peek(requestNetwork_in, RequestMsg) { assert(in_msg.Destination.isElement(machineID)); Entry cache_entry := getCacheEntry(in_msg.addr); @@ -340,7 +342,7 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") // Requests to this L1 cache from the L0 cache. in_port(messageBufferFromL0_in, CoherenceMsg, bufferFromL0, rank = 0) { - if (messageBufferFromL0_in.isReady()) { + if (messageBufferFromL0_in.isReady(clockEdge())) { peek(messageBufferFromL0_in, CoherenceMsg) { Entry cache_entry := getCacheEntry(in_msg.addr); TBE tbe := TBEs[in_msg.addr]; @@ -634,17 +636,19 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") } action(k_popL0RequestQueue, "k", desc="Pop mandatory queue.") { - messageBufferFromL0_in.dequeue(); + messageBufferFromL0_in.dequeue(clockEdge()); } action(l_popL2RequestQueue, "l", desc="Pop incoming request queue and profile the delay within this virtual network") { - profileMsgDelay(2, requestNetwork_in.dequeue()); + Tick delay := requestNetwork_in.dequeue(clockEdge()); + profileMsgDelay(2, ticksToCycles(delay)); } action(o_popL2ResponseQueue, "o", desc="Pop Incoming Response queue and profile the delay within this virtual network") { - profileMsgDelay(1, responseNetwork_in.dequeue()); + Tick delay := responseNetwork_in.dequeue(clockEdge()); + profileMsgDelay(1, ticksToCycles(delay)); } action(s_deallocateTBE, "s", desc="Deallocate TBE") { diff --git a/src/mem/protocol/MESI_Two_Level-L1cache.sm b/src/mem/protocol/MESI_Two_Level-L1cache.sm index b9be4663f..c40a47cae 100644 --- a/src/mem/protocol/MESI_Two_Level-L1cache.sm +++ b/src/mem/protocol/MESI_Two_Level-L1cache.sm @@ -156,6 +156,8 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") int l2_select_low_bit, default="RubySystem::getBlockSizeBits()"; + Tick clockEdge(); + Cycles ticksToCycles(Tick t); void set_cache_entry(AbstractCacheEntry a); void unset_cache_entry(); void set_tbe(TBE a); @@ -296,7 +298,7 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") // searches of all entries in the queue, not just the head msg. All // msgs in the structure can be invalidated if a demand miss matches. in_port(optionalQueue_in, RubyRequest, optionalQueue, desc="...", rank = 3) { - if (optionalQueue_in.isReady()) { + if (optionalQueue_in.isReady(clockEdge())) { peek(optionalQueue_in, RubyRequest) { // Instruction Prefetch if (in_msg.Type == RubyRequestType:IFETCH) { @@ -373,7 +375,7 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") // Response L1 Network - response msg to this L1 cache in_port(responseL1Network_in, ResponseMsg, responseToL1Cache, rank = 2) { - if (responseL1Network_in.isReady()) { + if (responseL1Network_in.isReady(clockEdge())) { peek(responseL1Network_in, ResponseMsg, block_on="addr") { assert(in_msg.Destination.isElement(machineID)); @@ -413,7 +415,7 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") // Request InterChip network - request from this L1 cache to the shared L2 in_port(requestL1Network_in, RequestMsg, requestToL1Cache, rank = 1) { - if(requestL1Network_in.isReady()) { + if(requestL1Network_in.isReady(clockEdge())) { peek(requestL1Network_in, RequestMsg, block_on="addr") { assert(in_msg.Destination.isElement(machineID)); @@ -439,7 +441,7 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") // Mandatory Queue betweens Node's CPU and it's L1 caches in_port(mandatoryQueue_in, RubyRequest, mandatoryQueue, desc="...", rank = 0) { - if (mandatoryQueue_in.isReady()) { + if (mandatoryQueue_in.isReady(clockEdge())) { peek(mandatoryQueue_in, RubyRequest, block_on="LineAddress") { // Check for data access to blocks in I-cache and ifetchs to blocks in D-cache @@ -866,17 +868,19 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") } action(k_popMandatoryQueue, "k", desc="Pop mandatory queue.") { - mandatoryQueue_in.dequeue(); + mandatoryQueue_in.dequeue(clockEdge()); } action(l_popRequestQueue, "l", desc="Pop incoming request queue and profile the delay within this virtual network") { - profileMsgDelay(2, requestL1Network_in.dequeue()); + Tick delay := requestL1Network_in.dequeue(clockEdge()); + profileMsgDelay(2, ticksToCycles(delay)); } action(o_popIncomingResponseQueue, "o", desc="Pop Incoming Response queue and profile the delay within this virtual network") { - profileMsgDelay(1, responseL1Network_in.dequeue()); + Tick delay := responseL1Network_in.dequeue(clockEdge()); + profileMsgDelay(1, ticksToCycles(delay)); } action(s_deallocateTBE, "s", desc="Deallocate TBE") { @@ -963,7 +967,7 @@ machine(L1Cache, "MESI Directory L1 Cache CMP") } action(pq_popPrefetchQueue, "\pq", desc="Pop the prefetch request queue") { - optionalQueue_in.dequeue(); + optionalQueue_in.dequeue(clockEdge()); } action(mp_markPrefetched, "mp", desc="Write data from response queue to cache") { diff --git a/src/mem/protocol/MESI_Two_Level-L2cache.sm b/src/mem/protocol/MESI_Two_Level-L2cache.sm index e4f719d9f..4134b7964 100644 --- a/src/mem/protocol/MESI_Two_Level-L2cache.sm +++ b/src/mem/protocol/MESI_Two_Level-L2cache.sm @@ -148,6 +148,10 @@ machine(L2Cache, "MESI Directory L2 Cache CMP") TBETable TBEs, template="", constructor="m_number_of_TBEs"; + Tick clockEdge(); + Tick cyclesToTicks(Cycles c); + Cycles ticksToCycles(Tick t); + void set_cache_entry(AbstractCacheEntry a); void unset_cache_entry(); void set_tbe(TBE a); @@ -285,7 +289,7 @@ machine(L2Cache, "MESI Directory L2 Cache CMP") in_port(L1unblockNetwork_in, ResponseMsg, unblockToL2Cache, rank = 2) { - if(L1unblockNetwork_in.isReady()) { + if(L1unblockNetwork_in.isReady(clockEdge())) { peek(L1unblockNetwork_in, ResponseMsg) { Entry cache_entry := getCacheEntry(in_msg.addr); TBE tbe := TBEs[in_msg.addr]; @@ -307,7 +311,7 @@ machine(L2Cache, "MESI Directory L2 Cache CMP") // Response L2 Network - response msg to this particular L2 bank in_port(responseL2Network_in, ResponseMsg, responseToL2Cache, rank = 1) { - if (responseL2Network_in.isReady()) { + if (responseL2Network_in.isReady(clockEdge())) { peek(responseL2Network_in, ResponseMsg) { // test wether it's from a local L1 or an off chip source assert(in_msg.Destination.isElement(machineID)); @@ -348,7 +352,7 @@ machine(L2Cache, "MESI Directory L2 Cache CMP") // L1 Request in_port(L1RequestL2Network_in, RequestMsg, L1RequestToL2Cache, rank = 0) { - if(L1RequestL2Network_in.isReady()) { + if(L1RequestL2Network_in.isReady(clockEdge())) { peek(L1RequestL2Network_in, RequestMsg) { Entry cache_entry := getCacheEntry(in_msg.addr); TBE tbe := TBEs[in_msg.addr]; @@ -604,15 +608,18 @@ machine(L2Cache, "MESI Directory L2 Cache CMP") } action(jj_popL1RequestQueue, "\j", desc="Pop incoming L1 request queue") { - profileMsgDelay(0, L1RequestL2Network_in.dequeue()); + Tick delay := L1RequestL2Network_in.dequeue(clockEdge()); + profileMsgDelay(0, ticksToCycles(delay)); } action(k_popUnblockQueue, "k", desc="Pop incoming unblock queue") { - profileMsgDelay(0, L1unblockNetwork_in.dequeue()); + Tick delay := L1unblockNetwork_in.dequeue(clockEdge()); + profileMsgDelay(0, ticksToCycles(delay)); } action(o_popIncomingResponseQueue, "o", desc="Pop Incoming Response queue") { - profileMsgDelay(1, responseL2Network_in.dequeue()); + Tick delay := responseL2Network_in.dequeue(clockEdge()); + profileMsgDelay(1, ticksToCycles(delay)); } action(m_writeDataToCache, "m", desc="Write data from response queue to cache") { @@ -769,7 +776,7 @@ machine(L2Cache, "MESI Directory L2 Cache CMP") } action(zn_recycleResponseNetwork, "zn", desc="recycle memory request") { - responseL2Network_in.recycle(); + responseL2Network_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(kd_wakeUpDependents, "kd", desc="wake-up dependents") { diff --git a/src/mem/protocol/MESI_Two_Level-dir.sm b/src/mem/protocol/MESI_Two_Level-dir.sm index 7484d001c..c9fbe3875 100644 --- a/src/mem/protocol/MESI_Two_Level-dir.sm +++ b/src/mem/protocol/MESI_Two_Level-dir.sm @@ -98,6 +98,8 @@ machine(Directory, "MESI Two Level directory protocol") // ** OBJECTS ** TBETable TBEs, template="", constructor="m_number_of_TBEs"; + Tick clockEdge(); + Tick cyclesToTicks(Cycles c); void set_tbe(TBE tbe); void unset_tbe(); void wakeUpBuffers(Addr a); @@ -190,7 +192,7 @@ machine(Directory, "MESI Two Level directory protocol") // ** IN_PORTS ** in_port(requestNetwork_in, RequestMsg, requestToDir, rank = 0) { - if (requestNetwork_in.isReady()) { + if (requestNetwork_in.isReady(clockEdge())) { peek(requestNetwork_in, RequestMsg) { assert(in_msg.Destination.isElement(machineID)); if (isGETRequest(in_msg.Type)) { @@ -210,7 +212,7 @@ machine(Directory, "MESI Two Level directory protocol") } in_port(responseNetwork_in, ResponseMsg, responseToDir, rank = 1) { - if (responseNetwork_in.isReady()) { + if (responseNetwork_in.isReady(clockEdge())) { peek(responseNetwork_in, ResponseMsg) { assert(in_msg.Destination.isElement(machineID)); if (in_msg.Type == CoherenceResponseType:MEMORY_DATA) { @@ -227,7 +229,7 @@ machine(Directory, "MESI Two Level directory protocol") // off-chip memory request/response is done in_port(memQueue_in, MemoryMsg, responseFromMemory, rank = 2) { - if (memQueue_in.isReady()) { + if (memQueue_in.isReady(clockEdge())) { peek(memQueue_in, MemoryMsg) { if (in_msg.Type == MemoryRequestType:MEMORY_READ) { trigger(Event:Memory_Data, in_msg.addr, TBEs[in_msg.addr]); @@ -286,15 +288,15 @@ machine(Directory, "MESI Two Level directory protocol") } action(j_popIncomingRequestQueue, "j", desc="Pop incoming request queue") { - requestNetwork_in.dequeue(); + requestNetwork_in.dequeue(clockEdge()); } action(k_popIncomingResponseQueue, "k", desc="Pop incoming request queue") { - responseNetwork_in.dequeue(); + responseNetwork_in.dequeue(clockEdge()); } action(l_popMemQueue, "q", desc="Pop off-chip request queue") { - memQueue_in.dequeue(); + memQueue_in.dequeue(clockEdge()); } action(kd_wakeUpDependents, "kd", desc="wake-up dependents") { @@ -322,7 +324,7 @@ machine(Directory, "MESI Two Level directory protocol") } action(p_popIncomingDMARequestQueue, "p", desc="Pop incoming DMA queue") { - requestNetwork_in.dequeue(); + requestNetwork_in.dequeue(clockEdge()); } action(dr_sendDMAData, "dr", desc="Send Data to DMA controller from directory") { @@ -359,7 +361,7 @@ machine(Directory, "MESI Two Level directory protocol") } action(zz_recycleDMAQueue, "zz", desc="recycle DMA queue") { - requestNetwork_in.recycle(); + requestNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(inv_sendCacheInvalidate, "inv", desc="Invalidate a cache block") { diff --git a/src/mem/protocol/MESI_Two_Level-dma.sm b/src/mem/protocol/MESI_Two_Level-dma.sm index cbd32cd44..84774ede8 100644 --- a/src/mem/protocol/MESI_Two_Level-dma.sm +++ b/src/mem/protocol/MESI_Two_Level-dma.sm @@ -51,6 +51,7 @@ machine(DMA, "DMA Controller") } State cur_state; + Tick clockEdge(); State getState(Addr addr) { return cur_state; @@ -78,7 +79,7 @@ machine(DMA, "DMA Controller") out_port(requestToDir_out, RequestMsg, requestToDir, desc="..."); in_port(dmaRequestQueue_in, SequencerMsg, mandatoryQueue, desc="...") { - if (dmaRequestQueue_in.isReady()) { + if (dmaRequestQueue_in.isReady(clockEdge())) { peek(dmaRequestQueue_in, SequencerMsg) { if (in_msg.Type == SequencerRequestType:LD ) { trigger(Event:ReadRequest, in_msg.LineAddress); @@ -92,7 +93,7 @@ machine(DMA, "DMA Controller") } in_port(dmaResponseQueue_in, ResponseMsg, responseFromDir, desc="...") { - if (dmaResponseQueue_in.isReady()) { + if (dmaResponseQueue_in.isReady(clockEdge())) { peek( dmaResponseQueue_in, ResponseMsg) { if (in_msg.Type == CoherenceResponseType:ACK) { trigger(Event:Ack, makeLineAddress(in_msg.addr)); @@ -142,11 +143,11 @@ machine(DMA, "DMA Controller") } action(p_popRequestQueue, "p", desc="Pop request queue") { - dmaRequestQueue_in.dequeue(); + dmaRequestQueue_in.dequeue(clockEdge()); } action(p_popResponseQueue, "\p", desc="Pop request queue") { - dmaResponseQueue_in.dequeue(); + dmaResponseQueue_in.dequeue(clockEdge()); } transition(READY, ReadRequest, BUSY_RD) { diff --git a/src/mem/protocol/MI_example-cache.sm b/src/mem/protocol/MI_example-cache.sm index 334106615..0a1570494 100644 --- a/src/mem/protocol/MI_example-cache.sm +++ b/src/mem/protocol/MI_example-cache.sm @@ -103,6 +103,8 @@ machine(L1Cache, "MI Example L1 Cache") TBETable TBEs, template="", constructor="m_number_of_TBEs"; // PROTOTYPES + Tick clockEdge(); + Cycles ticksToCycles(Tick t); void set_cache_entry(AbstractCacheEntry a); void unset_cache_entry(); void set_tbe(TBE b); @@ -200,7 +202,7 @@ machine(L1Cache, "MI Example L1 Cache") out_port(responseNetwork_out, ResponseMsg, responseFromCache); in_port(forwardRequestNetwork_in, RequestMsg, forwardToCache) { - if (forwardRequestNetwork_in.isReady()) { + if (forwardRequestNetwork_in.isReady(clockEdge())) { peek(forwardRequestNetwork_in, RequestMsg, block_on="addr") { Entry cache_entry := getCacheEntry(in_msg.addr); @@ -226,7 +228,7 @@ machine(L1Cache, "MI Example L1 Cache") } in_port(responseNetwork_in, ResponseMsg, responseToCache) { - if (responseNetwork_in.isReady()) { + if (responseNetwork_in.isReady(clockEdge())) { peek(responseNetwork_in, ResponseMsg, block_on="addr") { Entry cache_entry := getCacheEntry(in_msg.addr); @@ -244,7 +246,7 @@ machine(L1Cache, "MI Example L1 Cache") // Mandatory Queue in_port(mandatoryQueue_in, RubyRequest, mandatoryQueue, desc="...") { - if (mandatoryQueue_in.isReady()) { + if (mandatoryQueue_in.isReady(clockEdge())) { peek(mandatoryQueue_in, RubyRequest, block_on="LineAddress") { Entry cache_entry := getCacheEntry(in_msg.LineAddress); @@ -330,15 +332,17 @@ machine(L1Cache, "MI Example L1 Cache") } action(m_popMandatoryQueue, "m", desc="Pop the mandatory request queue") { - mandatoryQueue_in.dequeue(); + mandatoryQueue_in.dequeue(clockEdge()); } action(n_popResponseQueue, "n", desc="Pop the response queue") { - profileMsgDelay(1, responseNetwork_in.dequeue()); + Tick delay := responseNetwork_in.dequeue(clockEdge()); + profileMsgDelay(1, ticksToCycles(delay)); } action(o_popForwardedRequestQueue, "o", desc="Pop the forwarded request queue") { - profileMsgDelay(2, forwardRequestNetwork_in.dequeue()); + Tick delay := forwardRequestNetwork_in.dequeue(clockEdge()); + profileMsgDelay(2, ticksToCycles(delay)); } action(p_profileMiss, "pi", desc="Profile cache miss") { diff --git a/src/mem/protocol/MI_example-dir.sm b/src/mem/protocol/MI_example-dir.sm index bb4373901..f12e474b0 100644 --- a/src/mem/protocol/MI_example-dir.sm +++ b/src/mem/protocol/MI_example-dir.sm @@ -108,6 +108,9 @@ machine(Directory, "Directory protocol") // ** OBJECTS ** TBETable TBEs, template="", constructor="m_number_of_TBEs"; + Tick clockEdge(); + Cycles ticksToCycles(Tick t); + Tick cyclesToTicks(Cycles c); void set_tbe(TBE b); void unset_tbe(); @@ -204,7 +207,7 @@ machine(Directory, "Directory protocol") // ** IN_PORTS ** in_port(dmaRequestQueue_in, DMARequestMsg, dmaRequestToDir) { - if (dmaRequestQueue_in.isReady()) { + if (dmaRequestQueue_in.isReady(clockEdge())) { peek(dmaRequestQueue_in, DMARequestMsg) { TBE tbe := TBEs[in_msg.LineAddress]; if (in_msg.Type == DMARequestType:READ) { @@ -219,7 +222,7 @@ machine(Directory, "Directory protocol") } in_port(requestQueue_in, RequestMsg, requestToDir) { - if (requestQueue_in.isReady()) { + if (requestQueue_in.isReady(clockEdge())) { peek(requestQueue_in, RequestMsg) { TBE tbe := TBEs[in_msg.addr]; if (in_msg.Type == CoherenceRequestType:GETS) { @@ -242,7 +245,7 @@ machine(Directory, "Directory protocol") //added by SS // off-chip memory request/response is done in_port(memQueue_in, MemoryMsg, responseFromMemory) { - if (memQueue_in.isReady()) { + if (memQueue_in.isReady(clockEdge())) { peek(memQueue_in, MemoryMsg) { TBE tbe := TBEs[in_msg.addr]; if (in_msg.Type == MemoryRequestType:MEMORY_READ) { @@ -392,11 +395,11 @@ machine(Directory, "Directory protocol") } action(i_popIncomingRequestQueue, "i", desc="Pop incoming request queue") { - requestQueue_in.dequeue(); + requestQueue_in.dequeue(clockEdge()); } action(p_popIncomingDMARequestQueue, "p", desc="Pop incoming DMA queue") { - dmaRequestQueue_in.dequeue(); + dmaRequestQueue_in.dequeue(clockEdge()); } action(v_allocateTBE, "v", desc="Allocate TBE") { @@ -432,11 +435,11 @@ machine(Directory, "Directory protocol") } action(z_recycleRequestQueue, "z", desc="recycle request queue") { - requestQueue_in.recycle(); + requestQueue_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(y_recycleDMARequestQueue, "y", desc="recycle dma request queue") { - dmaRequestQueue_in.recycle(); + dmaRequestQueue_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } @@ -476,7 +479,7 @@ machine(Directory, "Directory protocol") } action(l_popMemQueue, "q", desc="Pop off-chip request queue") { - memQueue_in.dequeue(); + memQueue_in.dequeue(clockEdge()); } // TRANSITIONS diff --git a/src/mem/protocol/MI_example-dma.sm b/src/mem/protocol/MI_example-dma.sm index ce7b44630..76d87516a 100644 --- a/src/mem/protocol/MI_example-dma.sm +++ b/src/mem/protocol/MI_example-dma.sm @@ -52,6 +52,9 @@ machine(DMA, "DMA Controller") State cur_state; + Tick clockEdge(); + Cycles ticksToCycles(Tick t); + State getState(Addr addr) { return cur_state; } @@ -78,7 +81,7 @@ machine(DMA, "DMA Controller") out_port(requestToDir_out, DMARequestMsg, requestToDir, desc="..."); in_port(dmaRequestQueue_in, SequencerMsg, mandatoryQueue, desc="...") { - if (dmaRequestQueue_in.isReady()) { + if (dmaRequestQueue_in.isReady(clockEdge())) { peek(dmaRequestQueue_in, SequencerMsg) { if (in_msg.Type == SequencerRequestType:LD ) { trigger(Event:ReadRequest, in_msg.LineAddress); @@ -92,7 +95,7 @@ machine(DMA, "DMA Controller") } in_port(dmaResponseQueue_in, DMAResponseMsg, responseFromDir, desc="...") { - if (dmaResponseQueue_in.isReady()) { + if (dmaResponseQueue_in.isReady(clockEdge())) { peek( dmaResponseQueue_in, DMAResponseMsg) { if (in_msg.Type == DMAResponseType:ACK) { trigger(Event:Ack, in_msg.LineAddress); @@ -148,11 +151,11 @@ machine(DMA, "DMA Controller") } action(p_popRequestQueue, "p", desc="Pop request queue") { - dmaRequestQueue_in.dequeue(); + dmaRequestQueue_in.dequeue(clockEdge()); } action(p_popResponseQueue, "\p", desc="Pop request queue") { - dmaResponseQueue_in.dequeue(); + dmaResponseQueue_in.dequeue(clockEdge()); } transition(READY, ReadRequest, BUSY_RD) { diff --git a/src/mem/protocol/MOESI_CMP_directory-L1cache.sm b/src/mem/protocol/MOESI_CMP_directory-L1cache.sm index 2ef80efd2..1b1fd4ac7 100644 --- a/src/mem/protocol/MOESI_CMP_directory-L1cache.sm +++ b/src/mem/protocol/MOESI_CMP_directory-L1cache.sm @@ -133,6 +133,8 @@ machine(L1Cache, "Directory protocol") bool isPresent(Addr); } + Tick clockEdge(); + Tick cyclesToTicks(Cycles c); void set_cache_entry(AbstractCacheEntry b); void unset_cache_entry(); void set_tbe(TBE b); @@ -266,16 +268,16 @@ machine(L1Cache, "Directory protocol") // Use Timer in_port(useTimerTable_in, Addr, useTimerTable) { - if (useTimerTable_in.isReady()) { - trigger(Event:Use_Timeout, useTimerTable.readyAddress(), - getCacheEntry(useTimerTable.readyAddress()), - TBEs[useTimerTable.readyAddress()]); + if (useTimerTable_in.isReady(clockEdge())) { + Addr readyAddress := useTimerTable.nextAddress(); + trigger(Event:Use_Timeout, readyAddress, getCacheEntry(readyAddress), + TBEs.lookup(readyAddress)); } } // Trigger Queue in_port(triggerQueue_in, TriggerMsg, triggerQueue) { - if (triggerQueue_in.isReady()) { + if (triggerQueue_in.isReady(clockEdge())) { peek(triggerQueue_in, TriggerMsg) { if (in_msg.Type == TriggerType:ALL_ACKS) { trigger(Event:All_acks, in_msg.addr, @@ -291,7 +293,7 @@ machine(L1Cache, "Directory protocol") // Request Network in_port(requestNetwork_in, RequestMsg, requestToL1Cache) { - if (requestNetwork_in.isReady()) { + if (requestNetwork_in.isReady(clockEdge())) { peek(requestNetwork_in, RequestMsg, block_on="addr") { assert(in_msg.Destination.isElement(machineID)); DPRINTF(RubySlicc, "L1 received: %s\n", in_msg.Type); @@ -331,7 +333,7 @@ machine(L1Cache, "Directory protocol") // Response Network in_port(responseToL1Cache_in, ResponseMsg, responseToL1Cache) { - if (responseToL1Cache_in.isReady()) { + if (responseToL1Cache_in.isReady(clockEdge())) { peek(responseToL1Cache_in, ResponseMsg, block_on="addr") { if (in_msg.Type == CoherenceResponseType:ACK) { trigger(Event:Ack, in_msg.addr, @@ -352,7 +354,7 @@ machine(L1Cache, "Directory protocol") // Nothing from the unblock network // Mandatory Queue betweens Node's CPU and it's L1 caches in_port(mandatoryQueue_in, RubyRequest, mandatoryQueue, desc="...") { - if (mandatoryQueue_in.isReady()) { + if (mandatoryQueue_in.isReady(clockEdge())) { peek(mandatoryQueue_in, RubyRequest, block_on="LineAddress") { // Check for data access to blocks in I-cache and ifetchs to blocks in D-cache @@ -684,7 +686,7 @@ machine(L1Cache, "Directory protocol") } action(j_popTriggerQueue, "j", desc="Pop trigger queue.") { - triggerQueue_in.dequeue(); + triggerQueue_in.dequeue(clockEdge()); } action(jj_unsetUseTimer, "\jj", desc="Unset use timer.") { @@ -692,11 +694,11 @@ machine(L1Cache, "Directory protocol") } action(k_popMandatoryQueue, "k", desc="Pop mandatory queue.") { - mandatoryQueue_in.dequeue(); + mandatoryQueue_in.dequeue(clockEdge()); } action(l_popForwardQueue, "l", desc="Pop forwareded request queue.") { - requestNetwork_in.dequeue(); + requestNetwork_in.dequeue(clockEdge()); } action(m_decrementNumberOfMessages, "m", desc="Decrement the number of messages for which we're waiting") { @@ -715,7 +717,7 @@ machine(L1Cache, "Directory protocol") } action(n_popResponseQueue, "n", desc="Pop response queue") { - responseToL1Cache_in.dequeue(); + responseToL1Cache_in.dequeue(clockEdge()); } action(o_checkForCompletion, "o", desc="Check if we have received all the messages required for completion") { @@ -729,7 +731,8 @@ machine(L1Cache, "Directory protocol") } action(o_scheduleUseTimeout, "oo", desc="Schedule a use timeout.") { - useTimerTable.set(address, use_timeout_latency); + useTimerTable.set(address, + clockEdge() + cyclesToTicks(use_timeout_latency)); } action(ub_dmaUnblockL2Cache, "ub", desc="Send dma ack to l2 cache") { @@ -908,11 +911,11 @@ machine(L1Cache, "Directory protocol") } action(z_recycleRequestQueue, "z", desc="Send the head of the mandatory queue to the back of the queue.") { - requestNetwork_in.recycle(); + requestNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(zz_recycleMandatoryQueue, "\z", desc="Send the head of the mandatory queue to the back of the queue.") { - mandatoryQueue_in.recycle(); + mandatoryQueue_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } //***************************************************** diff --git a/src/mem/protocol/MOESI_CMP_directory-L2cache.sm b/src/mem/protocol/MOESI_CMP_directory-L2cache.sm index 0b288709e..84fb276e3 100644 --- a/src/mem/protocol/MOESI_CMP_directory-L2cache.sm +++ b/src/mem/protocol/MOESI_CMP_directory-L2cache.sm @@ -227,6 +227,8 @@ machine(L2Cache, "Token protocol") TBETable TBEs, template="", constructor="m_number_of_TBEs"; PerfectCacheMemory localDirectory, template=""; + Tick clockEdge(); + Tick cyclesToTicks(Cycles c); void set_cache_entry(AbstractCacheEntry b); void unset_cache_entry(); void set_tbe(TBE b); @@ -577,7 +579,7 @@ machine(L2Cache, "Token protocol") // Trigger Queue in_port(triggerQueue_in, TriggerMsg, triggerQueue) { - if (triggerQueue_in.isReady()) { + if (triggerQueue_in.isReady(clockEdge())) { peek(triggerQueue_in, TriggerMsg) { if (in_msg.Type == TriggerType:ALL_ACKS) { trigger(Event:All_Acks, in_msg.addr, @@ -592,7 +594,7 @@ machine(L2Cache, "Token protocol") // Request Network in_port(requestNetwork_in, RequestMsg, GlobalRequestToL2Cache) { - if (requestNetwork_in.isReady()) { + if (requestNetwork_in.isReady(clockEdge())) { peek(requestNetwork_in, RequestMsg) { if (in_msg.Type == CoherenceRequestType:GETX || in_msg.Type == CoherenceRequestType:DMA_WRITE) { if (in_msg.Requestor == machineID) { @@ -625,7 +627,7 @@ machine(L2Cache, "Token protocol") } in_port(L1requestNetwork_in, RequestMsg, L1RequestToL2Cache) { - if (L1requestNetwork_in.isReady()) { + if (L1requestNetwork_in.isReady(clockEdge())) { peek(L1requestNetwork_in, RequestMsg) { assert(in_msg.Destination.isElement(machineID)); if (in_msg.Type == CoherenceRequestType:GETX) { @@ -660,7 +662,7 @@ machine(L2Cache, "Token protocol") // Response Network in_port(responseNetwork_in, ResponseMsg, responseToL2Cache) { - if (responseNetwork_in.isReady()) { + if (responseNetwork_in.isReady(clockEdge())) { peek(responseNetwork_in, ResponseMsg) { assert(in_msg.Destination.isElement(machineID)); if (in_msg.Type == CoherenceResponseType:ACK) { @@ -1366,7 +1368,7 @@ machine(L2Cache, "Token protocol") } action(m_popRequestQueue, "m", desc="Pop request queue.") { - requestNetwork_in.dequeue(); + requestNetwork_in.dequeue(clockEdge()); } action(m_decrementNumberOfMessagesInt, "\m", desc="Decrement the number of messages for which we're waiting") { @@ -1391,15 +1393,15 @@ machine(L2Cache, "Token protocol") } action(n_popResponseQueue, "n", desc="Pop response queue") { - responseNetwork_in.dequeue(); + responseNetwork_in.dequeue(clockEdge()); } action(n_popTriggerQueue, "\n", desc="Pop trigger queue.") { - triggerQueue_in.dequeue(); + triggerQueue_in.dequeue(clockEdge()); } action(o_popL1RequestQueue, "o", desc="Pop L1 request queue.") { - L1requestNetwork_in.dequeue(); + L1requestNetwork_in.dequeue(clockEdge()); } @@ -1538,21 +1540,21 @@ machine(L2Cache, "Token protocol") peek(L1requestNetwork_in, RequestMsg) { APPEND_TRANSITION_COMMENT(in_msg.Requestor); } - L1requestNetwork_in.recycle(); + L1requestNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(zz_recycleRequestQueue, "\zz", desc="Send the head of the mandatory queue to the back of the queue.") { peek(requestNetwork_in, RequestMsg) { APPEND_TRANSITION_COMMENT(in_msg.Requestor); } - requestNetwork_in.recycle(); + requestNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(zz_recycleResponseQueue, "\z\z", desc="Send the head of the mandatory queue to the back of the queue.") { peek(responseNetwork_in, ResponseMsg) { APPEND_TRANSITION_COMMENT(in_msg.Sender); } - responseNetwork_in.recycle(); + responseNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(da_sendDmaAckUnblock, "da", desc="Send dma ack to global directory") { diff --git a/src/mem/protocol/MOESI_CMP_directory-dir.sm b/src/mem/protocol/MOESI_CMP_directory-dir.sm index 6ee7cd260..7175edc8d 100644 --- a/src/mem/protocol/MOESI_CMP_directory-dir.sm +++ b/src/mem/protocol/MOESI_CMP_directory-dir.sm @@ -119,6 +119,8 @@ machine(Directory, "Directory protocol") // ** OBJECTS ** TBETable TBEs, template="", constructor="m_number_of_TBEs"; + Tick clockEdge(); + Tick cyclesToTicks(Cycles c); void set_tbe(TBE b); void unset_tbe(); @@ -228,7 +230,7 @@ machine(Directory, "Directory protocol") // ** IN_PORTS ** in_port(unblockNetwork_in, ResponseMsg, responseToDir) { - if (unblockNetwork_in.isReady()) { + if (unblockNetwork_in.isReady(clockEdge())) { peek(unblockNetwork_in, ResponseMsg) { if (in_msg.Type == CoherenceResponseType:UNBLOCK) { if (getDirectoryEntry(in_msg.addr).WaitingUnblocks == 1) { @@ -261,7 +263,7 @@ machine(Directory, "Directory protocol") } in_port(requestQueue_in, RequestMsg, requestToDir) { - if (requestQueue_in.isReady()) { + if (requestQueue_in.isReady(clockEdge())) { peek(requestQueue_in, RequestMsg) { if (in_msg.Type == CoherenceRequestType:GETS) { trigger(Event:GETS, in_msg.addr, TBEs[in_msg.addr]); @@ -288,7 +290,7 @@ machine(Directory, "Directory protocol") // off-chip memory request/response is done in_port(memQueue_in, MemoryMsg, responseFromMemory) { - if (memQueue_in.isReady()) { + if (memQueue_in.isReady(clockEdge())) { peek(memQueue_in, MemoryMsg) { if (in_msg.Type == MemoryRequestType:MEMORY_READ) { trigger(Event:Memory_Data, in_msg.addr, TBEs[in_msg.addr]); @@ -438,11 +440,11 @@ machine(Directory, "Directory protocol") } action(i_popIncomingRequestQueue, "i", desc="Pop incoming request queue") { - requestQueue_in.dequeue(); + requestQueue_in.dequeue(clockEdge()); } action(j_popIncomingUnblockQueue, "j", desc="Pop incoming unblock queue") { - unblockNetwork_in.dequeue(); + unblockNetwork_in.dequeue(clockEdge()); } action(m_addUnlockerToSharers, "m", desc="Add the unlocker to the sharer list") { @@ -461,7 +463,7 @@ machine(Directory, "Directory protocol") } action(q_popMemQueue, "q", desc="Pop off-chip request queue") { - memQueue_in.dequeue(); + memQueue_in.dequeue(clockEdge()); } action(qf_queueMemoryFetchRequest, "qf", desc="Queue off-chip fetch request") { @@ -501,7 +503,7 @@ machine(Directory, "Directory protocol") } action(zz_recycleRequest, "\z", desc="Recycle the request queue") { - requestQueue_in.recycle(); + requestQueue_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(a_sendDMAAck, "\a", desc="Send DMA Ack that write completed, along with Inv Ack count") { diff --git a/src/mem/protocol/MOESI_CMP_directory-dma.sm b/src/mem/protocol/MOESI_CMP_directory-dma.sm index 10fc94abe..72dec6466 100644 --- a/src/mem/protocol/MOESI_CMP_directory-dma.sm +++ b/src/mem/protocol/MOESI_CMP_directory-dma.sm @@ -74,6 +74,7 @@ machine(DMA, "DMA Controller") TBETable TBEs, template="", constructor="m_number_of_TBEs"; State cur_state; + Tick clockEdge(); void set_tbe(TBE b); void unset_tbe(); @@ -104,7 +105,7 @@ machine(DMA, "DMA Controller") out_port(triggerQueue_out, TriggerMsg, triggerQueue, desc="..."); in_port(dmaRequestQueue_in, SequencerMsg, mandatoryQueue, desc="...") { - if (dmaRequestQueue_in.isReady()) { + if (dmaRequestQueue_in.isReady(clockEdge())) { peek(dmaRequestQueue_in, SequencerMsg) { if (in_msg.Type == SequencerRequestType:LD ) { trigger(Event:ReadRequest, in_msg.LineAddress, @@ -120,7 +121,7 @@ machine(DMA, "DMA Controller") } in_port(dmaResponseQueue_in, ResponseMsg, responseFromDir, desc="...") { - if (dmaResponseQueue_in.isReady()) { + if (dmaResponseQueue_in.isReady(clockEdge())) { peek( dmaResponseQueue_in, ResponseMsg) { if (in_msg.Type == CoherenceResponseType:DMA_ACK) { trigger(Event:DMA_Ack, makeLineAddress(in_msg.addr), @@ -141,7 +142,7 @@ machine(DMA, "DMA Controller") // Trigger Queue in_port(triggerQueue_in, TriggerMsg, triggerQueue) { - if (triggerQueue_in.isReady()) { + if (triggerQueue_in.isReady(clockEdge())) { peek(triggerQueue_in, TriggerMsg) { if (in_msg.Type == TriggerType:ALL_ACKS) { trigger(Event:All_Acks, in_msg.addr, TBEs[in_msg.addr]); @@ -215,15 +216,15 @@ machine(DMA, "DMA Controller") } action(p_popRequestQueue, "p", desc="Pop request queue") { - dmaRequestQueue_in.dequeue(); + dmaRequestQueue_in.dequeue(clockEdge()); } action(p_popResponseQueue, "\p", desc="Pop request queue") { - dmaResponseQueue_in.dequeue(); + dmaResponseQueue_in.dequeue(clockEdge()); } action(p_popTriggerQueue, "pp", desc="Pop trigger queue") { - triggerQueue_in.dequeue(); + triggerQueue_in.dequeue(clockEdge()); } action(t_updateTBEData, "t", desc="Update TBE Data") { diff --git a/src/mem/protocol/MOESI_CMP_token-L1cache.sm b/src/mem/protocol/MOESI_CMP_token-L1cache.sm index 230adfc4b..dac2027b9 100644 --- a/src/mem/protocol/MOESI_CMP_token-L1cache.sm +++ b/src/mem/protocol/MOESI_CMP_token-L1cache.sm @@ -184,6 +184,8 @@ machine(L1Cache, "Token protocol") int countReadStarvingForAddress(Addr); } + Tick clockEdge(); + Tick cyclesToTicks(Cycles c); void set_cache_entry(AbstractCacheEntry b); void unset_cache_entry(); void set_tbe(TBE b); @@ -456,25 +458,26 @@ machine(L1Cache, "Token protocol") // Use Timer in_port(useTimerTable_in, Addr, useTimerTable, rank=5) { - if (useTimerTable_in.isReady()) { - TBE tbe := L1_TBEs[useTimerTable.readyAddress()]; - - if (persistentTable.isLocked(useTimerTable.readyAddress()) && - (persistentTable.findSmallest(useTimerTable.readyAddress()) != machineID)) { - if (persistentTable.typeOfSmallest(useTimerTable.readyAddress()) == AccessType:Write) { - trigger(Event:Use_TimeoutStarverX, useTimerTable.readyAddress(), - getCacheEntry(useTimerTable.readyAddress()), tbe); + if (useTimerTable_in.isReady(clockEdge())) { + Addr readyAddress := useTimerTable.nextAddress(); + TBE tbe := L1_TBEs.lookup(readyAddress); + + if (persistentTable.isLocked(readyAddress) && + (persistentTable.findSmallest(readyAddress) != machineID)) { + if (persistentTable.typeOfSmallest(readyAddress) == AccessType:Write) { + trigger(Event:Use_TimeoutStarverX, readyAddress, + getCacheEntry(readyAddress), tbe); } else { - trigger(Event:Use_TimeoutStarverS, useTimerTable.readyAddress(), - getCacheEntry(useTimerTable.readyAddress()), tbe); + trigger(Event:Use_TimeoutStarverS, readyAddress, + getCacheEntry(readyAddress), tbe); } } else { if (no_mig_atomic && IsAtomic(tbe)) { - trigger(Event:Use_TimeoutNoStarvers_NoMig, useTimerTable.readyAddress(), - getCacheEntry(useTimerTable.readyAddress()), tbe); + trigger(Event:Use_TimeoutNoStarvers_NoMig, readyAddress, + getCacheEntry(readyAddress), tbe); } else { - trigger(Event:Use_TimeoutNoStarvers, useTimerTable.readyAddress(), - getCacheEntry(useTimerTable.readyAddress()), tbe); + trigger(Event:Use_TimeoutNoStarvers, readyAddress, + getCacheEntry(readyAddress), tbe); } } } @@ -482,16 +485,17 @@ machine(L1Cache, "Token protocol") // Reissue Timer in_port(reissueTimerTable_in, Addr, reissueTimerTable, rank=4) { - if (reissueTimerTable_in.isReady()) { - trigger(Event:Request_Timeout, reissueTimerTable.readyAddress(), - getCacheEntry(reissueTimerTable.readyAddress()), - L1_TBEs[reissueTimerTable.readyAddress()]); + Tick current_time := clockEdge(); + if (reissueTimerTable_in.isReady(current_time)) { + Addr addr := reissueTimerTable.nextAddress(); + trigger(Event:Request_Timeout, addr, getCacheEntry(addr), + L1_TBEs.lookup(addr)); } } // Persistent Network in_port(persistentNetwork_in, PersistentMsg, persistentToL1Cache, rank=3) { - if (persistentNetwork_in.isReady()) { + if (persistentNetwork_in.isReady(clockEdge())) { peek(persistentNetwork_in, PersistentMsg, block_on="addr") { assert(in_msg.Destination.isElement(machineID)); @@ -541,7 +545,7 @@ machine(L1Cache, "Token protocol") // Response Network in_port(responseNetwork_in, ResponseMsg, responseToL1Cache, rank=2) { - if (responseNetwork_in.isReady()) { + if (responseNetwork_in.isReady(clockEdge())) { peek(responseNetwork_in, ResponseMsg, block_on="addr") { assert(in_msg.Destination.isElement(machineID)); @@ -612,7 +616,7 @@ machine(L1Cache, "Token protocol") // Request Network in_port(requestNetwork_in, RequestMsg, requestToL1Cache) { - if (requestNetwork_in.isReady()) { + if (requestNetwork_in.isReady(clockEdge())) { peek(requestNetwork_in, RequestMsg, block_on="addr") { assert(in_msg.Destination.isElement(machineID)); @@ -659,7 +663,7 @@ machine(L1Cache, "Token protocol") // Mandatory Queue in_port(mandatoryQueue_in, RubyRequest, mandatoryQueue, desc="...", rank=0) { - if (mandatoryQueue_in.isReady()) { + if (mandatoryQueue_in.isReady(clockEdge())) { peek(mandatoryQueue_in, RubyRequest, block_on="LineAddress") { // Check for data access to blocks in I-cache and ifetchs to blocks in D-cache @@ -792,7 +796,8 @@ machine(L1Cache, "Token protocol") // IssueCount. // Set a wakeup timer - reissueTimerTable.set(address, reissue_wakeup_latency); + reissueTimerTable.set( + address, clockEdge() + cyclesToTicks(reissue_wakeup_latency)); } } else { @@ -844,9 +849,11 @@ machine(L1Cache, "Token protocol") // Set a wakeup timer if (dynamic_timeout_enabled) { - reissueTimerTable.set(address, (5 * averageLatencyEstimate()) / 4); + reissueTimerTable.set( + address, clockEdge() + cyclesToTicks(averageLatencyEstimate())); } else { - reissueTimerTable.set(address, fixed_timeout_latency); + reissueTimerTable.set( + address, clockEdge() + cyclesToTicks(fixed_timeout_latency)); } } @@ -911,7 +918,8 @@ machine(L1Cache, "Token protocol") // IssueCount. // Set a wakeup timer - reissueTimerTable.set(address, reissue_wakeup_latency); + reissueTimerTable.set( + address, clockEdge() + cyclesToTicks(reissue_wakeup_latency)); } } else { @@ -968,9 +976,11 @@ machine(L1Cache, "Token protocol") // Set a wakeup timer if (dynamic_timeout_enabled) { - reissueTimerTable.set(address, (5 * averageLatencyEstimate()) / 4); + reissueTimerTable.set( + address, clockEdge() + cyclesToTicks(averageLatencyEstimate())); } else { - reissueTimerTable.set(address, fixed_timeout_latency); + reissueTimerTable.set( + address, clockEdge() + cyclesToTicks(fixed_timeout_latency)); } } } @@ -1376,23 +1386,24 @@ machine(L1Cache, "Token protocol") } action(k_popMandatoryQueue, "k", desc="Pop mandatory queue.") { - mandatoryQueue_in.dequeue(); + mandatoryQueue_in.dequeue(clockEdge()); } action(l_popPersistentQueue, "l", desc="Pop persistent queue.") { - persistentNetwork_in.dequeue(); + persistentNetwork_in.dequeue(clockEdge()); } action(m_popRequestQueue, "m", desc="Pop request queue.") { - requestNetwork_in.dequeue(); + requestNetwork_in.dequeue(clockEdge()); } action(n_popResponseQueue, "n", desc="Pop response queue") { - responseNetwork_in.dequeue(); + responseNetwork_in.dequeue(clockEdge()); } action(o_scheduleUseTimeout, "o", desc="Schedule a use timeout.") { - useTimerTable.set(address, use_timeout_latency); + useTimerTable.set( + address, clockEdge() + cyclesToTicks(use_timeout_latency)); } action(p_informL2AboutTokenLoss, "p", desc="Inform L2 about loss of all tokens") { diff --git a/src/mem/protocol/MOESI_CMP_token-L2cache.sm b/src/mem/protocol/MOESI_CMP_token-L2cache.sm index 52bd19bcc..2ab593394 100644 --- a/src/mem/protocol/MOESI_CMP_token-L2cache.sm +++ b/src/mem/protocol/MOESI_CMP_token-L2cache.sm @@ -149,6 +149,7 @@ machine(L2Cache, "Token protocol") PersistentTable persistentTable; PerfectCacheMemory localDirectory, template=""; + Tick clockEdge(); void set_cache_entry(AbstractCacheEntry b); void unset_cache_entry(); @@ -326,7 +327,7 @@ machine(L2Cache, "Token protocol") // Persistent Network in_port(persistentNetwork_in, PersistentMsg, persistentToL2Cache) { - if (persistentNetwork_in.isReady()) { + if (persistentNetwork_in.isReady(clockEdge())) { peek(persistentNetwork_in, PersistentMsg) { assert(in_msg.Destination.isElement(machineID)); @@ -366,7 +367,7 @@ machine(L2Cache, "Token protocol") // Request Network in_port(requestNetwork_in, RequestMsg, GlobalRequestToL2Cache) { - if (requestNetwork_in.isReady()) { + if (requestNetwork_in.isReady(clockEdge())) { peek(requestNetwork_in, RequestMsg) { assert(in_msg.Destination.isElement(machineID)); @@ -389,7 +390,7 @@ machine(L2Cache, "Token protocol") } in_port(L1requestNetwork_in, RequestMsg, L1RequestToL2Cache) { - if (L1requestNetwork_in.isReady()) { + if (L1requestNetwork_in.isReady(clockEdge())) { peek(L1requestNetwork_in, RequestMsg) { assert(in_msg.Destination.isElement(machineID)); Entry cache_entry := getCacheEntry(in_msg.addr); @@ -413,7 +414,7 @@ machine(L2Cache, "Token protocol") // Response Network in_port(responseNetwork_in, ResponseMsg, responseToL2Cache) { - if (responseNetwork_in.isReady()) { + if (responseNetwork_in.isReady(clockEdge())) { peek(responseNetwork_in, ResponseMsg) { assert(in_msg.Destination.isElement(machineID)); Entry cache_entry := getCacheEntry(in_msg.addr); @@ -870,19 +871,19 @@ machine(L2Cache, "Token protocol") } action(l_popPersistentQueue, "l", desc="Pop persistent queue.") { - persistentNetwork_in.dequeue(); + persistentNetwork_in.dequeue(clockEdge()); } action(m_popRequestQueue, "m", desc="Pop request queue.") { - requestNetwork_in.dequeue(); + requestNetwork_in.dequeue(clockEdge()); } action(n_popResponseQueue, "n", desc="Pop response queue") { - responseNetwork_in.dequeue(); + responseNetwork_in.dequeue(clockEdge()); } action(o_popL1RequestQueue, "o", desc="Pop L1 request queue.") { - L1requestNetwork_in.dequeue(); + L1requestNetwork_in.dequeue(clockEdge()); } diff --git a/src/mem/protocol/MOESI_CMP_token-dir.sm b/src/mem/protocol/MOESI_CMP_token-dir.sm index ffef01eb0..63790531f 100644 --- a/src/mem/protocol/MOESI_CMP_token-dir.sm +++ b/src/mem/protocol/MOESI_CMP_token-dir.sm @@ -172,6 +172,8 @@ machine(Directory, "Token protocol") bool starving, default="false"; int l2_select_low_bit, default="RubySystem::getBlockSizeBits()"; + Tick clockEdge(); + Tick cyclesToTicks(Cycles c); void set_tbe(TBE b); void unset_tbe(); @@ -276,7 +278,7 @@ machine(Directory, "Token protocol") // ** IN_PORTS ** // off-chip memory request/response is done in_port(memQueue_in, MemoryMsg, responseFromMemory) { - if (memQueue_in.isReady()) { + if (memQueue_in.isReady(clockEdge())) { peek(memQueue_in, MemoryMsg) { if (in_msg.Type == MemoryRequestType:MEMORY_READ) { trigger(Event:Memory_Data, in_msg.addr, TBEs[in_msg.addr]); @@ -292,14 +294,15 @@ machine(Directory, "Token protocol") // Reissue Timer in_port(reissueTimerTable_in, Addr, reissueTimerTable) { - if (reissueTimerTable_in.isReady()) { - trigger(Event:Request_Timeout, reissueTimerTable.readyAddress(), - TBEs[reissueTimerTable.readyAddress()]); + Tick current_time := clockEdge(); + if (reissueTimerTable_in.isReady(current_time)) { + Addr addr := reissueTimerTable.nextAddress(); + trigger(Event:Request_Timeout, addr, TBEs.lookup(addr)); } } in_port(responseNetwork_in, ResponseMsg, responseToDir) { - if (responseNetwork_in.isReady()) { + if (responseNetwork_in.isReady(clockEdge())) { peek(responseNetwork_in, ResponseMsg) { assert(in_msg.Destination.isElement(machineID)); if (getDirectoryEntry(in_msg.addr).Tokens + in_msg.Tokens == max_tokens()) { @@ -338,7 +341,7 @@ machine(Directory, "Token protocol") } in_port(persistentNetwork_in, PersistentMsg, persistentToDir) { - if (persistentNetwork_in.isReady()) { + if (persistentNetwork_in.isReady(clockEdge())) { peek(persistentNetwork_in, PersistentMsg) { assert(in_msg.Destination.isElement(machineID)); @@ -400,7 +403,7 @@ machine(Directory, "Token protocol") } in_port(requestNetwork_in, RequestMsg, requestToDir) { - if (requestNetwork_in.isReady()) { + if (requestNetwork_in.isReady(clockEdge())) { peek(requestNetwork_in, RequestMsg) { assert(in_msg.Destination.isElement(machineID)); if (in_msg.Type == CoherenceRequestType:GETS) { @@ -415,7 +418,7 @@ machine(Directory, "Token protocol") } in_port(dmaRequestQueue_in, DMARequestMsg, dmaRequestToDir) { - if (dmaRequestQueue_in.isReady()) { + if (dmaRequestQueue_in.isReady(clockEdge())) { peek(dmaRequestQueue_in, DMARequestMsg) { if (in_msg.Type == DMARequestType:READ) { trigger(Event:DMA_READ, in_msg.LineAddress, TBEs[in_msg.LineAddress]); @@ -490,7 +493,7 @@ machine(Directory, "Token protocol") // IssueCount. // Set a wakeup timer - reissueTimerTable.set(address, reissue_wakeup_latency); + reissueTimerTable.set(address, cyclesToTicks(reissue_wakeup_latency)); } } @@ -558,7 +561,7 @@ machine(Directory, "Token protocol") // IssueCount. // Set a wakeup timer - reissueTimerTable.set(address, reissue_wakeup_latency); + reissueTimerTable.set(address, cyclesToTicks(reissue_wakeup_latency)); } } @@ -752,35 +755,35 @@ machine(Directory, "Token protocol") } action(j_popIncomingRequestQueue, "j", desc="Pop incoming request queue") { - requestNetwork_in.dequeue(); + requestNetwork_in.dequeue(clockEdge()); } action(z_recycleRequest, "z", desc="Recycle the request queue") { - requestNetwork_in.recycle(); + requestNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(k_popIncomingResponseQueue, "k", desc="Pop incoming response queue") { - responseNetwork_in.dequeue(); + responseNetwork_in.dequeue(clockEdge()); } action(kz_recycleResponse, "kz", desc="Recycle incoming response queue") { - responseNetwork_in.recycle(); + responseNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(l_popIncomingPersistentQueue, "l", desc="Pop incoming persistent queue") { - persistentNetwork_in.dequeue(); + persistentNetwork_in.dequeue(clockEdge()); } action(p_popDmaRequestQueue, "pd", desc="pop dma request queue") { - dmaRequestQueue_in.dequeue(); + dmaRequestQueue_in.dequeue(clockEdge()); } action(y_recycleDmaRequestQueue, "y", desc="recycle dma request queue") { - dmaRequestQueue_in.recycle(); + dmaRequestQueue_in.recycle(clockEdge(), cyclesToTicks(recycle_latency)); } action(l_popMemQueue, "q", desc="Pop off-chip request queue") { - memQueue_in.dequeue(); + memQueue_in.dequeue(clockEdge()); } action(r_bounceResponse, "r", desc="Bounce response to starving processor") { @@ -804,7 +807,7 @@ machine(Directory, "Token protocol") // if (reissueTimerTable.isSet(address)) { reissueTimerTable.unset(address); - reissueTimerTable.set(address, fixed_timeout_latency); + reissueTimerTable.set(address, cyclesToTicks(fixed_timeout_latency)); } } @@ -812,7 +815,7 @@ machine(Directory, "Token protocol") // // currently only support a fixed timeout latency // - reissueTimerTable.set(address, fixed_timeout_latency); + reissueTimerTable.set(address, cyclesToTicks(fixed_timeout_latency)); } action(ut_unsetReissueTimer, "ut", desc="Unset reissue timer.") { diff --git a/src/mem/protocol/MOESI_CMP_token-dma.sm b/src/mem/protocol/MOESI_CMP_token-dma.sm index 4bb80d4ba..efe3db3cd 100644 --- a/src/mem/protocol/MOESI_CMP_token-dma.sm +++ b/src/mem/protocol/MOESI_CMP_token-dma.sm @@ -54,6 +54,8 @@ machine(DMA, "DMA Controller") State cur_state; + Tick clockEdge(); + State getState(Addr addr) { return cur_state; } @@ -80,7 +82,7 @@ machine(DMA, "DMA Controller") out_port(reqToDirectory_out, DMARequestMsg, reqToDirectory, desc="..."); in_port(dmaRequestQueue_in, SequencerMsg, mandatoryQueue, desc="...") { - if (dmaRequestQueue_in.isReady()) { + if (dmaRequestQueue_in.isReady(clockEdge())) { peek(dmaRequestQueue_in, SequencerMsg) { if (in_msg.Type == SequencerRequestType:LD ) { trigger(Event:ReadRequest, in_msg.LineAddress); @@ -94,7 +96,7 @@ machine(DMA, "DMA Controller") } in_port(dmaResponseQueue_in, DMAResponseMsg, responseFromDir, desc="...") { - if (dmaResponseQueue_in.isReady()) { + if (dmaResponseQueue_in.isReady(clockEdge())) { peek( dmaResponseQueue_in, DMAResponseMsg) { if (in_msg.Type == DMAResponseType:ACK) { trigger(Event:Ack, in_msg.LineAddress); @@ -150,11 +152,11 @@ machine(DMA, "DMA Controller") } action(p_popRequestQueue, "p", desc="Pop request queue") { - dmaRequestQueue_in.dequeue(); + dmaRequestQueue_in.dequeue(clockEdge()); } action(p_popResponseQueue, "\p", desc="Pop request queue") { - dmaResponseQueue_in.dequeue(); + dmaResponseQueue_in.dequeue(clockEdge()); } transition(READY, ReadRequest, BUSY_RD) { diff --git a/src/mem/protocol/MOESI_hammer-cache.sm b/src/mem/protocol/MOESI_hammer-cache.sm index 88b7308ed..5d2383541 100644 --- a/src/mem/protocol/MOESI_hammer-cache.sm +++ b/src/mem/protocol/MOESI_hammer-cache.sm @@ -181,6 +181,7 @@ machine({L1Cache, L2Cache}, "AMD Hammer-like protocol") TBETable TBEs, template="", constructor="m_number_of_TBEs"; + Tick clockEdge(); void set_cache_entry(AbstractCacheEntry b); void unset_cache_entry(); void set_tbe(TBE b); @@ -329,7 +330,7 @@ machine({L1Cache, L2Cache}, "AMD Hammer-like protocol") // Trigger Queue in_port(triggerQueue_in, TriggerMsg, triggerQueue, rank=3) { - if (triggerQueue_in.isReady()) { + if (triggerQueue_in.isReady(clockEdge())) { peek(triggerQueue_in, TriggerMsg) { Entry cache_entry := getCacheEntry(in_msg.addr); @@ -352,7 +353,7 @@ machine({L1Cache, L2Cache}, "AMD Hammer-like protocol") // Response Network in_port(responseToCache_in, ResponseMsg, responseToCache, rank=2) { - if (responseToCache_in.isReady()) { + if (responseToCache_in.isReady(clockEdge())) { peek(responseToCache_in, ResponseMsg, block_on="addr") { Entry cache_entry := getCacheEntry(in_msg.addr); @@ -377,7 +378,7 @@ machine({L1Cache, L2Cache}, "AMD Hammer-like protocol") // Forward Network in_port(forwardToCache_in, RequestMsg, forwardToCache, rank=1) { - if (forwardToCache_in.isReady()) { + if (forwardToCache_in.isReady(clockEdge())) { peek(forwardToCache_in, RequestMsg, block_on="addr") { Entry cache_entry := getCacheEntry(in_msg.addr); @@ -421,7 +422,7 @@ machine({L1Cache, L2Cache}, "AMD Hammer-like protocol") // Mandatory Queue in_port(mandatoryQueue_in, RubyRequest, mandatoryQueue, desc="...", rank=0) { - if (mandatoryQueue_in.isReady()) { + if (mandatoryQueue_in.isReady(clockEdge())) { peek(mandatoryQueue_in, RubyRequest, block_on="LineAddress") { // Check for data access to blocks in I-cache and ifetchs to blocks in D-cache @@ -950,15 +951,15 @@ machine({L1Cache, L2Cache}, "AMD Hammer-like protocol") } action(j_popTriggerQueue, "j", desc="Pop trigger queue.") { - triggerQueue_in.dequeue(); + triggerQueue_in.dequeue(clockEdge()); } action(k_popMandatoryQueue, "k", desc="Pop mandatory queue.") { - mandatoryQueue_in.dequeue(); + mandatoryQueue_in.dequeue(clockEdge()); } action(l_popForwardQueue, "l", desc="Pop forwareded request queue.") { - forwardToCache_in.dequeue(); + forwardToCache_in.dequeue(clockEdge()); } action(hp_copyFromTBEToL2, "li", desc="Copy data from TBE to L2 cache entry.") { @@ -1017,7 +1018,7 @@ machine({L1Cache, L2Cache}, "AMD Hammer-like protocol") } action(n_popResponseQueue, "n", desc="Pop response queue") { - responseToCache_in.dequeue(); + responseToCache_in.dequeue(clockEdge()); } action(ll_L2toL1Transfer, "ll", desc="") { diff --git a/src/mem/protocol/MOESI_hammer-dir.sm b/src/mem/protocol/MOESI_hammer-dir.sm index 4948a8108..4f5b00658 100644 --- a/src/mem/protocol/MOESI_hammer-dir.sm +++ b/src/mem/protocol/MOESI_hammer-dir.sm @@ -184,6 +184,7 @@ machine(Directory, "AMD Hammer-like protocol") bool isPresent(Addr); } + Tick clockEdge(); void set_cache_entry(AbstractCacheEntry b); void unset_cache_entry(); void set_tbe(TBE a); @@ -314,7 +315,7 @@ machine(Directory, "AMD Hammer-like protocol") // Trigger Queue in_port(triggerQueue_in, TriggerMsg, triggerQueue, rank=5) { - if (triggerQueue_in.isReady()) { + if (triggerQueue_in.isReady(clockEdge())) { peek(triggerQueue_in, TriggerMsg) { PfEntry pf_entry := getProbeFilterEntry(in_msg.addr); TBE tbe := TBEs[in_msg.addr]; @@ -338,7 +339,7 @@ machine(Directory, "AMD Hammer-like protocol") } in_port(unblockNetwork_in, ResponseMsg, unblockToDir, rank=4) { - if (unblockNetwork_in.isReady()) { + if (unblockNetwork_in.isReady(clockEdge())) { peek(unblockNetwork_in, ResponseMsg) { PfEntry pf_entry := getProbeFilterEntry(in_msg.addr); TBE tbe := TBEs[in_msg.addr]; @@ -367,7 +368,7 @@ machine(Directory, "AMD Hammer-like protocol") // Response Network in_port(responseToDir_in, ResponseMsg, responseToDir, rank=3) { - if (responseToDir_in.isReady()) { + if (responseToDir_in.isReady(clockEdge())) { peek(responseToDir_in, ResponseMsg) { PfEntry pf_entry := getProbeFilterEntry(in_msg.addr); TBE tbe := TBEs[in_msg.addr]; @@ -390,7 +391,7 @@ machine(Directory, "AMD Hammer-like protocol") // off-chip memory request/response is done in_port(memQueue_in, MemoryMsg, responseFromMemory, rank=2) { - if (memQueue_in.isReady()) { + if (memQueue_in.isReady(clockEdge())) { peek(memQueue_in, MemoryMsg) { PfEntry pf_entry := getProbeFilterEntry(in_msg.addr); TBE tbe := TBEs[in_msg.addr]; @@ -407,7 +408,7 @@ machine(Directory, "AMD Hammer-like protocol") } in_port(requestQueue_in, RequestMsg, requestToDir, rank=1) { - if (requestQueue_in.isReady()) { + if (requestQueue_in.isReady(clockEdge())) { peek(requestQueue_in, RequestMsg) { PfEntry pf_entry := getProbeFilterEntry(in_msg.addr); TBE tbe := TBEs[in_msg.addr]; @@ -441,7 +442,7 @@ machine(Directory, "AMD Hammer-like protocol") } in_port(dmaRequestQueue_in, DMARequestMsg, dmaRequestToDir, rank=0) { - if (dmaRequestQueue_in.isReady()) { + if (dmaRequestQueue_in.isReady(clockEdge())) { peek(dmaRequestQueue_in, DMARequestMsg) { PfEntry pf_entry := getProbeFilterEntry(in_msg.LineAddress); TBE tbe := TBEs[in_msg.LineAddress]; @@ -682,7 +683,7 @@ machine(Directory, "AMD Hammer-like protocol") } action(n_popResponseQueue, "n", desc="Pop response queue") { - responseToDir_in.dequeue(); + responseToDir_in.dequeue(clockEdge()); } action(o_checkForCompletion, "o", desc="Check if we have received all the messages required for completion") { @@ -1115,14 +1116,14 @@ machine(Directory, "AMD Hammer-like protocol") } action(i_popIncomingRequestQueue, "i", desc="Pop incoming request queue") { - requestQueue_in.dequeue(); + requestQueue_in.dequeue(clockEdge()); } action(j_popIncomingUnblockQueue, "j", desc="Pop incoming unblock queue") { peek(unblockNetwork_in, ResponseMsg) { APPEND_TRANSITION_COMMENT(in_msg.Sender); } - unblockNetwork_in.dequeue(); + unblockNetwork_in.dequeue(clockEdge()); } action(k_wakeUpDependents, "k", desc="wake-up dependents") { @@ -1130,15 +1131,15 @@ machine(Directory, "AMD Hammer-like protocol") } action(l_popMemQueue, "q", desc="Pop off-chip request queue") { - memQueue_in.dequeue(); + memQueue_in.dequeue(clockEdge()); } action(g_popTriggerQueue, "g", desc="Pop trigger queue") { - triggerQueue_in.dequeue(); + triggerQueue_in.dequeue(clockEdge()); } action(p_popDmaRequestQueue, "pd", desc="pop dma request queue") { - dmaRequestQueue_in.dequeue(); + dmaRequestQueue_in.dequeue(clockEdge()); } action(zd_stallAndWaitDMARequest, "zd", desc="Stall and wait the dma request queue") { diff --git a/src/mem/protocol/MOESI_hammer-dma.sm b/src/mem/protocol/MOESI_hammer-dma.sm index 4691e2490..7157082c4 100644 --- a/src/mem/protocol/MOESI_hammer-dma.sm +++ b/src/mem/protocol/MOESI_hammer-dma.sm @@ -52,6 +52,8 @@ machine(DMA, "DMA Controller") State cur_state; + Tick clockEdge(); + State getState(Addr addr) { return cur_state; } @@ -77,7 +79,7 @@ machine(DMA, "DMA Controller") out_port(requestToDir_out, DMARequestMsg, requestToDir, desc="..."); in_port(dmaRequestQueue_in, SequencerMsg, mandatoryQueue, desc="...") { - if (dmaRequestQueue_in.isReady()) { + if (dmaRequestQueue_in.isReady(clockEdge())) { peek(dmaRequestQueue_in, SequencerMsg) { if (in_msg.Type == SequencerRequestType:LD ) { trigger(Event:ReadRequest, in_msg.LineAddress); @@ -91,7 +93,7 @@ machine(DMA, "DMA Controller") } in_port(dmaResponseQueue_in, DMAResponseMsg, responseFromDir, desc="...") { - if (dmaResponseQueue_in.isReady()) { + if (dmaResponseQueue_in.isReady(clockEdge())) { peek( dmaResponseQueue_in, DMAResponseMsg) { if (in_msg.Type == DMAResponseType:ACK) { trigger(Event:Ack, in_msg.LineAddress); @@ -147,11 +149,11 @@ machine(DMA, "DMA Controller") } action(p_popRequestQueue, "p", desc="Pop request queue") { - dmaRequestQueue_in.dequeue(); + dmaRequestQueue_in.dequeue(clockEdge()); } action(p_popResponseQueue, "\p", desc="Pop request queue") { - dmaResponseQueue_in.dequeue(); + dmaResponseQueue_in.dequeue(clockEdge()); } transition(READY, ReadRequest, BUSY_RD) { diff --git a/src/mem/protocol/Network_test-cache.sm b/src/mem/protocol/Network_test-cache.sm index 82829a6ea..dab8f1089 100644 --- a/src/mem/protocol/Network_test-cache.sm +++ b/src/mem/protocol/Network_test-cache.sm @@ -68,6 +68,7 @@ machine(L1Cache, "Network_test L1 Cache") } // FUNCTIONS + Tick clockEdge(); // cpu/testers/networktest/networktest.cc generates packets of the type // ReadReq, INST_FETCH, and WriteReq. @@ -129,7 +130,7 @@ machine(L1Cache, "Network_test L1 Cache") // Mandatory Queue in_port(mandatoryQueue_in, RubyRequest, mandatoryQueue, desc="...") { - if (mandatoryQueue_in.isReady()) { + if (mandatoryQueue_in.isReady(clockEdge())) { peek(mandatoryQueue_in, RubyRequest) { trigger(mandatory_request_type_to_event(in_msg.Type), in_msg.LineAddress, getCacheEntry(in_msg.LineAddress)); @@ -174,7 +175,7 @@ machine(L1Cache, "Network_test L1 Cache") } action(m_popMandatoryQueue, "m", desc="Pop the mandatory request queue") { - mandatoryQueue_in.dequeue(); + mandatoryQueue_in.dequeue(clockEdge()); } action(r_load_hit, "r", desc="Notify sequencer the load completed.") { diff --git a/src/mem/protocol/Network_test-dir.sm b/src/mem/protocol/Network_test-dir.sm index d618e98ff..6bd6920b3 100644 --- a/src/mem/protocol/Network_test-dir.sm +++ b/src/mem/protocol/Network_test-dir.sm @@ -60,7 +60,9 @@ machine(Directory, "Network_test Directory") DataBlock DataBlk, desc="data for the block"; } - // ** OBJECTS ** + // ** FUNCTIONS ** + Tick clockEdge(); + State getState(Addr addr) { return State:I; } @@ -87,7 +89,7 @@ machine(Directory, "Network_test Directory") // ** IN_PORTS ** in_port(requestQueue_in, RequestMsg, requestToDir) { - if (requestQueue_in.isReady()) { + if (requestQueue_in.isReady(clockEdge())) { peek(requestQueue_in, RequestMsg) { if (in_msg.Type == CoherenceRequestType:MSG) { trigger(Event:Receive_Request, in_msg.addr); @@ -98,7 +100,7 @@ machine(Directory, "Network_test Directory") } } in_port(forwardQueue_in, RequestMsg, forwardToDir) { - if (forwardQueue_in.isReady()) { + if (forwardQueue_in.isReady(clockEdge())) { peek(forwardQueue_in, RequestMsg) { if (in_msg.Type == CoherenceRequestType:MSG) { trigger(Event:Receive_Forward, in_msg.addr); @@ -109,7 +111,7 @@ machine(Directory, "Network_test Directory") } } in_port(responseQueue_in, RequestMsg, responseToDir) { - if (responseQueue_in.isReady()) { + if (responseQueue_in.isReady(clockEdge())) { peek(responseQueue_in, RequestMsg) { if (in_msg.Type == CoherenceRequestType:MSG) { trigger(Event:Receive_Response, in_msg.addr); @@ -123,15 +125,15 @@ machine(Directory, "Network_test Directory") // Actions action(i_popIncomingRequestQueue, "i", desc="Pop incoming request queue") { - requestQueue_in.dequeue(); + requestQueue_in.dequeue(clockEdge()); } action(f_popIncomingForwardQueue, "f", desc="Pop incoming forward queue") { - forwardQueue_in.dequeue(); + forwardQueue_in.dequeue(clockEdge()); } action(r_popIncomingResponseQueue, "r", desc="Pop incoming response queue") { - responseQueue_in.dequeue(); + responseQueue_in.dequeue(clockEdge()); } // TRANSITIONS diff --git a/src/mem/protocol/RubySlicc_Defines.sm b/src/mem/protocol/RubySlicc_Defines.sm index d4f7fa58f..eb235f8f3 100644 --- a/src/mem/protocol/RubySlicc_Defines.sm +++ b/src/mem/protocol/RubySlicc_Defines.sm @@ -1,4 +1,3 @@ - /* * Copyright (c) 1999-2005 Mark D. Hill and David A. Wood * All rights reserved. @@ -31,6 +30,7 @@ NodeID version; MachineID machineID; NodeID clusterID; +Cycles recycle_latency; // Functions implemented in the AbstractController class for // making timing access to the memory maintained by the diff --git a/src/mem/protocol/RubySlicc_Exports.sm b/src/mem/protocol/RubySlicc_Exports.sm index eeec185fc..7c2069203 100644 --- a/src/mem/protocol/RubySlicc_Exports.sm +++ b/src/mem/protocol/RubySlicc_Exports.sm @@ -37,6 +37,7 @@ external_type(PacketPtr, primitive="yes"); external_type(Packet, primitive="yes"); external_type(Addr, primitive="yes"); external_type(Cycles, primitive="yes", default="Cycles(0)"); +external_type(Tick, primitive="yes", default="0"); structure(DataBlock, external = "yes", desc="..."){ void clear(); diff --git a/src/mem/protocol/RubySlicc_Types.sm b/src/mem/protocol/RubySlicc_Types.sm index f464b3c7d..a8bf93bcc 100644 --- a/src/mem/protocol/RubySlicc_Types.sm +++ b/src/mem/protocol/RubySlicc_Types.sm @@ -41,9 +41,9 @@ external_type(OutPort, primitive="yes"); external_type(Scalar, primitive="yes"); structure(InPort, external = "yes", primitive="yes") { - bool isReady(); - Cycles dequeue(); - void recycle(); + bool isReady(Tick current_time); + Tick dequeue(Tick current_time); + void recycle(Tick current_time, Tick recycle_latency); bool isEmpty(); bool isStallMapEmpty(); int getStallMapSize(); @@ -179,9 +179,9 @@ structure (DMASequencer, external = "yes") { } structure (TimerTable, inport="yes", external = "yes") { - bool isReady(); - Addr readyAddress(); - void set(Addr, Cycles); + bool isReady(Tick); + Addr nextAddress(); + void set(Addr, Tick); void unset(Addr); bool isSet(Addr); } diff --git a/src/mem/ruby/network/MessageBuffer.cc b/src/mem/ruby/network/MessageBuffer.cc index b07bdbdca..35850f61e 100644 --- a/src/mem/ruby/network/MessageBuffer.cc +++ b/src/mem/ruby/network/MessageBuffer.cc @@ -40,7 +40,7 @@ using namespace std; using m5::stl_helpers::operator<<; MessageBuffer::MessageBuffer(const Params *p) - : SimObject(p), m_recycle_latency(p->recycle_latency), + : SimObject(p), m_max_size(p->buffer_size), m_time_last_time_size_checked(0), m_time_last_time_enqueue(0), m_time_last_time_pop(0), m_last_arrival_time(0), m_strict_fifo(p->ordered), @@ -48,9 +48,6 @@ MessageBuffer::MessageBuffer(const Params *p) { m_msg_counter = 0; m_consumer = NULL; - m_sender = NULL; - m_receiver = NULL; - m_size_last_time_size_checked = 0; m_size_at_cycle_start = 0; m_msgs_this_cycle = 0; @@ -63,10 +60,10 @@ MessageBuffer::MessageBuffer(const Params *p) } unsigned int -MessageBuffer::getSize() +MessageBuffer::getSize(Tick curTime) { - if (m_time_last_time_size_checked != m_receiver->curCycle()) { - m_time_last_time_size_checked = m_receiver->curCycle(); + if (m_time_last_time_size_checked != curTime) { + m_time_last_time_size_checked = curTime; m_size_last_time_size_checked = m_prio_heap.size(); } @@ -74,7 +71,7 @@ MessageBuffer::getSize() } bool -MessageBuffer::areNSlotsAvailable(unsigned int n) +MessageBuffer::areNSlotsAvailable(unsigned int n, Tick current_time) { // fast path when message buffers have infinite size @@ -88,11 +85,11 @@ MessageBuffer::areNSlotsAvailable(unsigned int n) // size immediately unsigned int current_size = 0; - if (m_time_last_time_pop < m_sender->clockEdge()) { + if (m_time_last_time_pop < current_time) { // no pops this cycle - heap size is correct current_size = m_prio_heap.size(); } else { - if (m_time_last_time_enqueue < m_sender->curCycle()) { + if (m_time_last_time_enqueue < current_time) { // no enqueues this cycle - m_size_at_cycle_start is correct current_size = m_size_at_cycle_start; } else { @@ -118,8 +115,6 @@ const Message* MessageBuffer::peek() const { DPRINTF(RubyQueue, "Peeking at head of queue.\n"); - assert(isReady()); - const Message* msg_ptr = m_prio_heap.front().get(); assert(msg_ptr); @@ -128,24 +123,24 @@ MessageBuffer::peek() const } // FIXME - move me somewhere else -Cycles +Tick random_time() { - Cycles time(1); - time += Cycles(random_mt.random(0, 3)); // [0...3] + Tick time = 1; + time += random_mt.random(0, 3); // [0...3] if (random_mt.random(0, 7) == 0) { // 1 in 8 chance - time += Cycles(100 + random_mt.random(1, 15)); // 100 + [1...15] + time += 100 + random_mt.random(1, 15); // 100 + [1...15] } return time; } void -MessageBuffer::enqueue(MsgPtr message, Cycles delta) +MessageBuffer::enqueue(MsgPtr message, Tick current_time, Tick delta) { // record current time incase we have a pop that also adjusts my size - if (m_time_last_time_enqueue < m_sender->curCycle()) { + if (m_time_last_time_enqueue < current_time) { m_msgs_this_cycle = 0; // first msg this cycle - m_time_last_time_enqueue = m_sender->curCycle(); + m_time_last_time_enqueue = current_time; } m_msg_counter++; @@ -154,23 +149,20 @@ MessageBuffer::enqueue(MsgPtr message, Cycles delta) // Calculate the arrival time of the message, that is, the first // cycle the message can be dequeued. assert(delta > 0); - Tick current_time = m_sender->clockEdge(); Tick arrival_time = 0; if (!RubySystem::getRandomization() || !m_randomization) { // No randomization - arrival_time = current_time + delta * m_sender->clockPeriod(); + arrival_time = current_time + delta; } else { // Randomization - ignore delta if (m_strict_fifo) { if (m_last_arrival_time < current_time) { m_last_arrival_time = current_time; } - arrival_time = m_last_arrival_time + - random_time() * m_sender->clockPeriod(); + arrival_time = m_last_arrival_time + random_time(); } else { - arrival_time = current_time + - random_time() * m_sender->clockPeriod(); + arrival_time = current_time + random_time(); } } @@ -180,9 +172,8 @@ MessageBuffer::enqueue(MsgPtr message, Cycles delta) if (arrival_time < m_last_arrival_time) { panic("FIFO ordering violated: %s name: %s current time: %d " "delta: %d arrival_time: %d last arrival_time: %d\n", - *this, name(), current_time, - delta * m_sender->clockPeriod(), - arrival_time, m_last_arrival_time); + *this, name(), current_time, delta, arrival_time, + m_last_arrival_time); } } @@ -195,10 +186,10 @@ MessageBuffer::enqueue(MsgPtr message, Cycles delta) Message* msg_ptr = message.get(); assert(msg_ptr != NULL); - assert(m_sender->clockEdge() >= msg_ptr->getLastEnqueueTime() && + assert(current_time >= msg_ptr->getLastEnqueueTime() && "ensure we aren't dequeued early"); - msg_ptr->updateDelayedTicks(m_sender->clockEdge()); + msg_ptr->updateDelayedTicks(current_time); msg_ptr->setLastEnqueueTime(arrival_time); msg_ptr->setMsgCounter(m_msg_counter); @@ -215,32 +206,30 @@ MessageBuffer::enqueue(MsgPtr message, Cycles delta) m_consumer->storeEventInfo(m_vnet_id); } -Cycles -MessageBuffer::dequeue() +Tick +MessageBuffer::dequeue(Tick current_time) { DPRINTF(RubyQueue, "Popping\n"); - assert(isReady()); + assert(isReady(current_time)); // get MsgPtr of the message about to be dequeued MsgPtr message = m_prio_heap.front(); // get the delay cycles - message->updateDelayedTicks(m_receiver->clockEdge()); - Cycles delayCycles = - m_receiver->ticksToCycles(message->getDelayedTicks()); + message->updateDelayedTicks(current_time); + Tick delay = message->getDelayedTicks(); // record previous size and time so the current buffer size isn't // adjusted until schd cycle - if (m_time_last_time_pop < m_receiver->clockEdge()) { + if (m_time_last_time_pop < current_time) { m_size_at_cycle_start = m_prio_heap.size(); - m_time_last_time_pop = m_receiver->clockEdge(); + m_time_last_time_pop = current_time; } - pop_heap(m_prio_heap.begin(), m_prio_heap.end(), - greater()); + pop_heap(m_prio_heap.begin(), m_prio_heap.end(), greater()); m_prio_heap.pop_back(); - return delayCycles; + return delay; } void @@ -249,25 +238,26 @@ MessageBuffer::clear() m_prio_heap.clear(); m_msg_counter = 0; - m_time_last_time_enqueue = Cycles(0); + m_time_last_time_enqueue = 0; m_time_last_time_pop = 0; m_size_at_cycle_start = 0; m_msgs_this_cycle = 0; } void -MessageBuffer::recycle() +MessageBuffer::recycle(Tick current_time, Tick recycle_latency) { DPRINTF(RubyQueue, "Recycling.\n"); - assert(isReady()); + assert(isReady(current_time)); MsgPtr node = m_prio_heap.front(); pop_heap(m_prio_heap.begin(), m_prio_heap.end(), greater()); - node->setLastEnqueueTime(m_receiver->clockEdge(m_recycle_latency)); + Tick future_time = current_time + recycle_latency; + node->setLastEnqueueTime(future_time); + m_prio_heap.back() = node; push_heap(m_prio_heap.begin(), m_prio_heap.end(), greater()); - m_consumer-> - scheduleEventAbsolute(m_receiver->clockEdge(m_recycle_latency)); + m_consumer->scheduleEventAbsolute(future_time); } void @@ -289,11 +279,10 @@ MessageBuffer::reanalyzeList(list <, Tick schdTick) } void -MessageBuffer::reanalyzeMessages(Addr addr) +MessageBuffer::reanalyzeMessages(Addr addr, Tick current_time) { DPRINTF(RubyQueue, "ReanalyzeMessages %s\n", addr); assert(m_stall_msg_map.count(addr) > 0); - Tick curTick = m_receiver->clockEdge(); // // Put all stalled messages associated with this address back on the @@ -301,15 +290,14 @@ MessageBuffer::reanalyzeMessages(Addr addr) // scheduled for the current cycle so that the previously stalled messages // will be observed before any younger messages that may arrive this cycle // - reanalyzeList(m_stall_msg_map[addr], curTick); + reanalyzeList(m_stall_msg_map[addr], current_time); m_stall_msg_map.erase(addr); } void -MessageBuffer::reanalyzeAllMessages() +MessageBuffer::reanalyzeAllMessages(Tick current_time) { DPRINTF(RubyQueue, "ReanalyzeAllMessages\n"); - Tick curTick = m_receiver->clockEdge(); // // Put all stalled messages associated with this address back on the @@ -319,20 +307,20 @@ MessageBuffer::reanalyzeAllMessages() // for (StallMsgMapType::iterator map_iter = m_stall_msg_map.begin(); map_iter != m_stall_msg_map.end(); ++map_iter) { - reanalyzeList(map_iter->second, curTick); + reanalyzeList(map_iter->second, current_time); } m_stall_msg_map.clear(); } void -MessageBuffer::stallMessage(Addr addr) +MessageBuffer::stallMessage(Addr addr, Tick current_time) { DPRINTF(RubyQueue, "Stalling due to %s\n", addr); - assert(isReady()); + assert(isReady(current_time)); assert(getOffset(addr) == 0); MsgPtr message = m_prio_heap.front(); - dequeue(); + dequeue(current_time); // // Note: no event is scheduled to analyze the map at a later time. @@ -356,10 +344,10 @@ MessageBuffer::print(ostream& out) const } bool -MessageBuffer::isReady() const +MessageBuffer::isReady(Tick current_time) const { return ((m_prio_heap.size() > 0) && - (m_prio_heap.front()->getLastEnqueueTime() <= m_receiver->clockEdge())); + (m_prio_heap.front()->getLastEnqueueTime() <= current_time)); } bool diff --git a/src/mem/ruby/network/MessageBuffer.hh b/src/mem/ruby/network/MessageBuffer.hh index 4209aea0f..4fdf4978d 100644 --- a/src/mem/ruby/network/MessageBuffer.hh +++ b/src/mem/ruby/network/MessageBuffer.hh @@ -55,24 +55,24 @@ class MessageBuffer : public SimObject typedef MessageBufferParams Params; MessageBuffer(const Params *p); - void reanalyzeMessages(Addr addr); - void reanalyzeAllMessages(); - void stallMessage(Addr addr); + void reanalyzeMessages(Addr addr, Tick current_time); + void reanalyzeAllMessages(Tick current_time); + void stallMessage(Addr addr, Tick current_time); // TRUE if head of queue timestamp <= SystemTime - bool isReady() const; + bool isReady(Tick current_time) const; void - delayHead() + delayHead(Tick current_time, Tick delta) { MsgPtr m = m_prio_heap.front(); std::pop_heap(m_prio_heap.begin(), m_prio_heap.end(), std::greater()); m_prio_heap.pop_back(); - enqueue(m, Cycles(1)); + enqueue(m, current_time, delta); } - bool areNSlotsAvailable(unsigned int n); + bool areNSlotsAvailable(unsigned int n, Tick curTime); int getPriority() { return m_priority_rank; } void setPriority(int rank) { m_priority_rank = rank; } void setConsumer(Consumer* consumer) @@ -86,20 +86,6 @@ class MessageBuffer : public SimObject m_consumer = consumer; } - void setSender(ClockedObject* obj) - { - DPRINTF(RubyQueue, "Setting sender: %s\n", obj->name()); - assert(m_sender == NULL || m_sender == obj); - m_sender = obj; - } - - void setReceiver(ClockedObject* obj) - { - DPRINTF(RubyQueue, "Setting receiver: %s\n", obj->name()); - assert(m_receiver == NULL || m_receiver == obj); - m_receiver = obj; - } - Consumer* getConsumer() { return m_consumer; } bool getOrdered() { return m_strict_fifo; } @@ -108,26 +94,20 @@ class MessageBuffer : public SimObject //! message queue. The function assumes that the queue is nonempty. const Message* peek() const; - const MsgPtr& - peekMsgPtr() const - { - assert(isReady()); - return m_prio_heap.front(); - } + const MsgPtr &peekMsgPtr() const { return m_prio_heap.front(); } - void enqueue(MsgPtr message) { enqueue(message, Cycles(1)); } - void enqueue(MsgPtr message, Cycles delta); + void enqueue(MsgPtr message, Tick curTime, Tick delta); //! Updates the delay cycles of the message at the head of the queue, //! removes it from the queue and returns its total delay. - Cycles dequeue(); + Tick dequeue(Tick current_time); - void recycle(); + void recycle(Tick current_time, Tick recycle_latency); bool isEmpty() const { return m_prio_heap.size() == 0; } bool isStallMapEmpty() { return m_stall_msg_map.size() == 0; } unsigned int getStallMapSize() { return m_stall_msg_map.size(); } - unsigned int getSize(); + unsigned int getSize(Tick curTime); void clear(); void print(std::ostream& out) const; @@ -148,17 +128,10 @@ class MessageBuffer : public SimObject uint32_t functionalWrite(Packet *pkt); private: - //added by SS - const Cycles m_recycle_latency; - void reanalyzeList(std::list &, Tick); private: // Data Members (m_ prefix) - //! The two ends of the buffer. - ClockedObject* m_sender; - ClockedObject* m_receiver; - //! Consumer to signal a wakeup(), can be NULL Consumer* m_consumer; std::vector m_prio_heap; @@ -170,12 +143,12 @@ class MessageBuffer : public SimObject StallMsgMapType m_stall_msg_map; const unsigned int m_max_size; - Cycles m_time_last_time_size_checked; + Tick m_time_last_time_size_checked; unsigned int m_size_last_time_size_checked; // variables used so enqueues appear to happen immediately, while // pop happen the next cycle - Cycles m_time_last_time_enqueue; + Tick m_time_last_time_enqueue; Tick m_time_last_time_pop; Tick m_last_arrival_time; @@ -193,7 +166,7 @@ class MessageBuffer : public SimObject int m_vnet_id; }; -Cycles random_time(); +Tick random_time(); inline std::ostream& operator<<(std::ostream& out, const MessageBuffer& obj) diff --git a/src/mem/ruby/network/MessageBuffer.py b/src/mem/ruby/network/MessageBuffer.py index 88c528e30..d8a028532 100644 --- a/src/mem/ruby/network/MessageBuffer.py +++ b/src/mem/ruby/network/MessageBuffer.py @@ -37,7 +37,6 @@ class MessageBuffer(SimObject): ordered = Param.Bool(False, "Whether the buffer is ordered") buffer_size = Param.Unsigned(0, "Maximum number of entries to buffer \ (0 allows infinite entries)") - recycle_latency = Param.Cycles(Parent.recycle_latency, "") randomization = Param.Bool(False, "") master = MasterPort("Master port to MessageBuffer receiver") diff --git a/src/mem/ruby/network/garnet/fixed-pipeline/NetworkInterface_d.cc b/src/mem/ruby/network/garnet/fixed-pipeline/NetworkInterface_d.cc index c7bd6178a..e350eba6b 100644 --- a/src/mem/ruby/network/garnet/fixed-pipeline/NetworkInterface_d.cc +++ b/src/mem/ruby/network/garnet/fixed-pipeline/NetworkInterface_d.cc @@ -115,13 +115,6 @@ NetworkInterface_d::addNode(vector& in, for (auto& it : in) { if (it != nullptr) { it->setConsumer(this); - it->setReceiver(this); - } - } - - for (auto& it : out) { - if (it != nullptr) { - it->setSender(this); } } } @@ -222,6 +215,7 @@ NetworkInterface_d::wakeup() DPRINTF(RubyNetwork, "m_id: %d woke up at time: %lld", m_id, curCycle()); MsgPtr msg_ptr; + Tick curTime = clockEdge(); // Checking for messages coming from the protocol // can pick up a message/cycle for each virtual net @@ -231,10 +225,10 @@ NetworkInterface_d::wakeup() continue; } - while (b->isReady()) { // Is there a message waiting + while (b->isReady(curTime)) { // Is there a message waiting msg_ptr = b->peekMsgPtr(); if (flitisizeMessage(msg_ptr, vnet)) { - b->dequeue(); + b->dequeue(curTime); } else { break; } @@ -253,7 +247,7 @@ NetworkInterface_d::wakeup() free_signal = true; outNode_ptr[t_flit->get_vnet()]->enqueue( - t_flit->get_msg_ptr(), Cycles(1)); + t_flit->get_msg_ptr(), curTime, cyclesToTicks(Cycles(1))); } // Simply send a credit back since we are not buffering // this flit in the NI @@ -363,7 +357,7 @@ NetworkInterface_d::checkReschedule() continue; } - while (it->isReady()) { // Is there a message waiting + while (it->isReady(clockEdge())) { // Is there a message waiting scheduleEvent(Cycles(1)); return; } diff --git a/src/mem/ruby/network/garnet/flexible-pipeline/NetworkInterface.cc b/src/mem/ruby/network/garnet/flexible-pipeline/NetworkInterface.cc index d834ea1a3..3d75ef8c2 100644 --- a/src/mem/ruby/network/garnet/flexible-pipeline/NetworkInterface.cc +++ b/src/mem/ruby/network/garnet/flexible-pipeline/NetworkInterface.cc @@ -99,13 +99,6 @@ NetworkInterface::addNode(vector& in, for (auto& it: in) { if (it != nullptr) { it->setConsumer(this); - it->setReceiver(this); - } - } - - for (auto& it : out) { - if (it != nullptr) { - it->setSender(this); } } } @@ -250,10 +243,10 @@ NetworkInterface::wakeup() continue; } - while (b->isReady()) { // Is there a message waiting + while (b->isReady(clockEdge())) { // Is there a message waiting msg_ptr = b->peekMsgPtr(); if (flitisizeMessage(msg_ptr, vnet)) { - b->dequeue(); + b->dequeue(clockEdge()); } else { break; } @@ -272,7 +265,7 @@ NetworkInterface::wakeup() m_id, curCycle()); outNode_ptr[t_flit->get_vnet()]->enqueue( - t_flit->get_msg_ptr(), Cycles(1)); + t_flit->get_msg_ptr(), clockEdge(), cyclesToTicks(Cycles(1))); // signal the upstream router that this vc can be freed now inNetLink->release_vc_link(t_flit->get_vc(), @@ -334,7 +327,7 @@ NetworkInterface::checkReschedule() continue; } - while (it->isReady()) { // Is there a message waiting + while (it->isReady(clockEdge())) { // Is there a message waiting scheduleEvent(Cycles(1)); return; } diff --git a/src/mem/ruby/network/simple/PerfectSwitch.cc b/src/mem/ruby/network/simple/PerfectSwitch.cc index 697357ccb..301d453c5 100644 --- a/src/mem/ruby/network/simple/PerfectSwitch.cc +++ b/src/mem/ruby/network/simple/PerfectSwitch.cc @@ -144,8 +144,9 @@ PerfectSwitch::operateMessageBuffer(MessageBuffer *buffer, int incoming, // temporary vectors to store the routing results vector output_links; vector output_link_destinations; + Tick current_time = m_switch->clockEdge(); - while (buffer->isReady()) { + while (buffer->isReady(current_time)) { DPRINTF(RubyNetwork, "incoming: %d\n", incoming); // Peek at message @@ -176,7 +177,7 @@ PerfectSwitch::operateMessageBuffer(MessageBuffer *buffer, int incoming, for (int out = 0; out < m_out.size(); out++) { int out_queue_length = 0; for (int v = 0; v < m_virtual_networks; v++) { - out_queue_length += m_out[out][v]->getSize(); + out_queue_length += m_out[out][v]->getSize(current_time); } int value = (out_queue_length << 8) | @@ -220,7 +221,7 @@ PerfectSwitch::operateMessageBuffer(MessageBuffer *buffer, int incoming, for (int i = 0; i < output_links.size(); i++) { int outgoing = output_links[i]; - if (!m_out[outgoing][vnet]->areNSlotsAvailable(1)) + if (!m_out[outgoing][vnet]->areNSlotsAvailable(1, current_time)) enough = false; DPRINTF(RubyNetwork, "Checking if node is blocked ..." @@ -251,7 +252,7 @@ PerfectSwitch::operateMessageBuffer(MessageBuffer *buffer, int incoming, } // Dequeue msg - buffer->dequeue(); + buffer->dequeue(current_time); m_pending_message_count[vnet]--; // Enqueue it - for all outgoing queues @@ -273,7 +274,8 @@ PerfectSwitch::operateMessageBuffer(MessageBuffer *buffer, int incoming, "inport[%d][%d] to outport [%d][%d].\n", incoming, vnet, outgoing, vnet); - m_out[outgoing][vnet]->enqueue(msg_ptr); + m_out[outgoing][vnet]->enqueue(msg_ptr, current_time, + m_switch->cyclesToTicks(Cycles(1))); } } } diff --git a/src/mem/ruby/network/simple/SimpleNetwork.py b/src/mem/ruby/network/simple/SimpleNetwork.py index f4ec440a3..87de0fb46 100644 --- a/src/mem/ruby/network/simple/SimpleNetwork.py +++ b/src/mem/ruby/network/simple/SimpleNetwork.py @@ -41,9 +41,6 @@ class SimpleNetwork(RubyNetwork): endpoint_bandwidth = Param.Int(1000, "bandwidth adjustment factor"); adaptive_routing = Param.Bool(False, "enable adaptive routing"); int_link_buffers = VectorParam.MessageBuffer("Buffers for int_links") - # int_links do not recycle buffers, so this parameter is not used. - # TODO: Move recycle_latency out of MessageBuffers and into controllers - recycle_latency = Param.Cycles(0, "") def setup_buffers(self): # Note that all SimpleNetwork MessageBuffers are currently ordered @@ -82,6 +79,3 @@ class Switch(BasicRouter): virt_nets = Param.Int(Parent.number_of_virtual_networks, "number of virtual networks") port_buffers = VectorParam.MessageBuffer("Port buffers") - # Ports do not recycle buffers, so this parameter is not used. - # TODO: Move recycle_latency out of MessageBuffers and into controllers - recycle_latency = Param.Cycles(0, "") diff --git a/src/mem/ruby/network/simple/Switch.cc b/src/mem/ruby/network/simple/Switch.cc index b9d0b8010..0951ef138 100644 --- a/src/mem/ruby/network/simple/Switch.cc +++ b/src/mem/ruby/network/simple/Switch.cc @@ -69,12 +69,6 @@ void Switch::addInPort(const vector& in) { m_perfect_switch->addInPort(in); - - for (auto& it : in) { - if (it != nullptr) { - it->setReceiver(this); - } - } } void @@ -95,17 +89,10 @@ Switch::addOutPort(const vector& out, vector intermediateBuffers; for (int i = 0; i < out.size(); ++i) { - if (out[i] != nullptr) { - out[i]->setSender(this); - } - assert(m_num_connected_buffers < m_port_buffers.size()); MessageBuffer* buffer_ptr = m_port_buffers[m_num_connected_buffers]; m_num_connected_buffers++; intermediateBuffers.push_back(buffer_ptr); - - buffer_ptr->setSender(this); - buffer_ptr->setReceiver(this); } // Hook the queues to the PerfectSwitch diff --git a/src/mem/ruby/network/simple/Throttle.cc b/src/mem/ruby/network/simple/Throttle.cc index 01d1f6fbe..3863ab944 100644 --- a/src/mem/ruby/network/simple/Throttle.cc +++ b/src/mem/ruby/network/simple/Throttle.cc @@ -94,14 +94,16 @@ Throttle::operateVnet(int vnet, int &bw_remaining, bool &schedule_wakeup, if (out == nullptr || in == nullptr) { return; } - assert(m_units_remaining[vnet] >= 0); - while (bw_remaining > 0 && (in->isReady() || m_units_remaining[vnet] > 0) && - out->areNSlotsAvailable(1)) { + assert(m_units_remaining[vnet] >= 0); + Tick current_time = m_switch->clockEdge(); + while (bw_remaining > 0 && (in->isReady(current_time) || + m_units_remaining[vnet] > 0) && + out->areNSlotsAvailable(1, current_time)) { // See if we are done transferring the previous message on // this virtual network - if (m_units_remaining[vnet] == 0 && in->isReady()) { + if (m_units_remaining[vnet] == 0 && in->isReady(current_time)) { // Find the size of the message we are moving MsgPtr msg_ptr = in->peekMsgPtr(); Message *net_msg_ptr = msg_ptr.get(); @@ -114,8 +116,9 @@ Throttle::operateVnet(int vnet, int &bw_remaining, bool &schedule_wakeup, m_ruby_system->curCycle()); // Move the message - in->dequeue(); - out->enqueue(msg_ptr, m_link_latency); + in->dequeue(current_time); + out->enqueue(msg_ptr, current_time, + m_switch->cyclesToTicks(m_link_latency)); // Count the message m_msg_counts[net_msg_ptr->getMessageSize()][vnet]++; @@ -128,8 +131,9 @@ Throttle::operateVnet(int vnet, int &bw_remaining, bool &schedule_wakeup, bw_remaining = max(0, -diff); } - if (bw_remaining > 0 && (in->isReady() || m_units_remaining[vnet] > 0) && - !out->areNSlotsAvailable(1)) { + if (bw_remaining > 0 && (in->isReady(current_time) || + m_units_remaining[vnet] > 0) && + !out->areNSlotsAvailable(1, current_time)) { DPRINTF(RubyNetwork, "vnet: %d", vnet); // schedule me to wakeup again because I'm waiting for my diff --git a/src/mem/ruby/slicc_interface/AbstractController.cc b/src/mem/ruby/slicc_interface/AbstractController.cc index 5bd38195a..be6438711 100644 --- a/src/mem/ruby/slicc_interface/AbstractController.cc +++ b/src/mem/ruby/slicc_interface/AbstractController.cc @@ -60,9 +60,6 @@ AbstractController::init() m_delayVCHistogram.push_back(new Stats::Histogram()); m_delayVCHistogram[i]->init(10); } - if (getMemoryQueue()) { - getMemoryQueue()->setSender(this); - } } void @@ -118,7 +115,8 @@ AbstractController::wakeUpBuffers(Addr addr) in_port_rank >= 0; in_port_rank--) { if ((*(m_waiting_buffers[addr]))[in_port_rank] != NULL) { - (*(m_waiting_buffers[addr]))[in_port_rank]->reanalyzeMessages(addr); + (*(m_waiting_buffers[addr]))[in_port_rank]-> + reanalyzeMessages(addr, clockEdge()); } } delete m_waiting_buffers[addr]; @@ -138,7 +136,8 @@ AbstractController::wakeUpAllBuffers(Addr addr) in_port_rank >= 0; in_port_rank--) { if ((*(m_waiting_buffers[addr]))[in_port_rank] != NULL) { - (*(m_waiting_buffers[addr]))[in_port_rank]->reanalyzeMessages(addr); + (*(m_waiting_buffers[addr]))[in_port_rank]-> + reanalyzeMessages(addr, clockEdge()); } } delete m_waiting_buffers[addr]; @@ -168,7 +167,7 @@ AbstractController::wakeUpAllBuffers() // if (*vec_iter != NULL && (wokeUpMsgBufs.count(*vec_iter) == 0)) { - (*vec_iter)->reanalyzeAllMessages(); + (*vec_iter)->reanalyzeAllMessages(clockEdge()); wokeUpMsgBufs.insert(*vec_iter); } } @@ -328,7 +327,7 @@ AbstractController::recvTimingResp(PacketPtr pkt) panic("Incorrect packet type received from memory controller!"); } - getMemoryQueue()->enqueue(msg); + getMemoryQueue()->enqueue(msg, clockEdge(), cyclesToTicks(Cycles(1))); delete pkt; } diff --git a/src/mem/ruby/structures/TBETable.hh b/src/mem/ruby/structures/TBETable.hh index cbc51dae5..4a24a5b13 100644 --- a/src/mem/ruby/structures/TBETable.hh +++ b/src/mem/ruby/structures/TBETable.hh @@ -47,12 +47,12 @@ class TBETable void allocate(Addr address); void deallocate(Addr address); bool - areNSlotsAvailable(int n) const + areNSlotsAvailable(int n, Tick current_time) const { return (m_number_of_TBEs - m_map.size()) >= n; } - ENTRY* lookup(Addr address); + ENTRY *lookup(Addr address); // Print cache contents void print(std::ostream& out) const; diff --git a/src/mem/ruby/structures/TimerTable.cc b/src/mem/ruby/structures/TimerTable.cc index 17dac6fc0..4809c8a47 100644 --- a/src/mem/ruby/structures/TimerTable.cc +++ b/src/mem/ruby/structures/TimerTable.cc @@ -34,14 +34,12 @@ TimerTable::TimerTable() : m_next_time(0) { m_consumer_ptr = NULL; - m_clockobj_ptr = NULL; - m_next_valid = false; m_next_address = 0; } bool -TimerTable::isReady() const +TimerTable::isReady(Tick curTime) const { if (m_map.empty()) return false; @@ -50,14 +48,12 @@ TimerTable::isReady() const updateNext(); } assert(m_next_valid); - return (m_clockobj_ptr->curCycle() >= m_next_time); + return (curTime >= m_next_time); } Addr -TimerTable::readyAddress() const +TimerTable::nextAddress() const { - assert(isReady()); - if (!m_next_valid) { updateNext(); } @@ -66,17 +62,14 @@ TimerTable::readyAddress() const } void -TimerTable::set(Addr address, Cycles relative_latency) +TimerTable::set(Addr address, Tick ready_time) { assert(address == makeLineAddress(address)); - assert(relative_latency > 0); assert(!m_map.count(address)); - Cycles ready_time = m_clockobj_ptr->curCycle() + relative_latency; m_map[address] = ready_time; assert(m_consumer_ptr != NULL); - m_consumer_ptr-> - scheduleEventAbsolute(m_clockobj_ptr->clockPeriod() * ready_time); + m_consumer_ptr->scheduleEventAbsolute(ready_time); m_next_valid = false; // Don't always recalculate the next ready address diff --git a/src/mem/ruby/structures/TimerTable.hh b/src/mem/ruby/structures/TimerTable.hh index 606201eb4..9efe7ca04 100644 --- a/src/mem/ruby/structures/TimerTable.hh +++ b/src/mem/ruby/structures/TimerTable.hh @@ -49,25 +49,16 @@ class TimerTable m_consumer_ptr = consumer_ptr; } - void setClockObj(ClockedObject* obj) - { - assert(m_clockobj_ptr == NULL); - m_clockobj_ptr = obj; - } - void setDescription(const std::string& name) { m_name = name; } - bool isReady() const; - Addr readyAddress() const; + bool isReady(Tick curTime) const; + Addr nextAddress() const; bool isSet(Addr address) const { return !!m_map.count(address); } - void set(Addr address, Cycles relative_latency); - void set(Addr address, uint64_t relative_latency) - { set(address, Cycles(relative_latency)); } - + void set(Addr address, Tick ready_time); void unset(Addr address); void print(std::ostream& out) const; @@ -82,14 +73,12 @@ class TimerTable // use a std::map for the address map as this container is sorted // and ensures a well-defined iteration order - typedef std::map AddressMap; + typedef std::map AddressMap; AddressMap m_map; mutable bool m_next_valid; - mutable Cycles m_next_time; // Only valid if m_next_valid is true + mutable Tick m_next_time; // Only valid if m_next_valid is true mutable Addr m_next_address; // Only valid if m_next_valid is true - //! Object used for querying time. - ClockedObject* m_clockobj_ptr; //! Consumer to signal a wakeup() Consumer* m_consumer_ptr; diff --git a/src/mem/ruby/system/DMASequencer.cc b/src/mem/ruby/system/DMASequencer.cc index 85b476cfd..3c895e627 100644 --- a/src/mem/ruby/system/DMASequencer.cc +++ b/src/mem/ruby/system/DMASequencer.cc @@ -54,7 +54,6 @@ DMASequencer::init() MemObject::init(); assert(m_controller != NULL); m_mandatory_q_ptr = m_controller->getMandatoryQueue(); - m_mandatory_q_ptr->setSender(this); m_is_busy = false; m_data_block_mask = ~ (~0 << RubySystem::getBlockSizeBits()); @@ -256,7 +255,7 @@ DMASequencer::makeRequest(PacketPtr pkt) } assert(m_mandatory_q_ptr != NULL); - m_mandatory_q_ptr->enqueue(msg); + m_mandatory_q_ptr->enqueue(msg, clockEdge(), cyclesToTicks(Cycles(1))); active_request.bytes_issued += msg->getLen(); return RequestStatus_Issued; @@ -302,7 +301,7 @@ DMASequencer::issueNext() } assert(m_mandatory_q_ptr != NULL); - m_mandatory_q_ptr->enqueue(msg); + m_mandatory_q_ptr->enqueue(msg, clockEdge(), cyclesToTicks(Cycles(1))); active_request.bytes_issued += msg->getLen(); DPRINTF(RubyDma, "DMA request bytes issued %d, bytes completed %d, total len %d\n", diff --git a/src/mem/ruby/system/RubyPort.cc b/src/mem/ruby/system/RubyPort.cc index c13aed97e..e03d23774 100644 --- a/src/mem/ruby/system/RubyPort.cc +++ b/src/mem/ruby/system/RubyPort.cc @@ -81,7 +81,6 @@ RubyPort::init() { assert(m_controller != NULL); m_mandatory_q_ptr = m_controller->getMandatoryQueue(); - m_mandatory_q_ptr->setSender(this); } BaseMasterPort & diff --git a/src/mem/ruby/system/Sequencer.cc b/src/mem/ruby/system/Sequencer.cc index 186e62d55..815e270b6 100644 --- a/src/mem/ruby/system/Sequencer.cc +++ b/src/mem/ruby/system/Sequencer.cc @@ -629,7 +629,7 @@ Sequencer::issueRequest(PacketPtr pkt, RubyRequestType secondary_type) assert(latency > 0); assert(m_mandatory_q_ptr != NULL); - m_mandatory_q_ptr->enqueue(msg, latency); + m_mandatory_q_ptr->enqueue(msg, clockEdge(), cyclesToTicks(latency)); } template diff --git a/src/mem/slicc/ast/EnqueueStatementAST.py b/src/mem/slicc/ast/EnqueueStatementAST.py index 930540494..556643e4e 100644 --- a/src/mem/slicc/ast/EnqueueStatementAST.py +++ b/src/mem/slicc/ast/EnqueueStatementAST.py @@ -65,9 +65,10 @@ class EnqueueStatementAST(StatementAST): if self.latexpr != None: ret_type, rcode = self.latexpr.inline(True) code("(${{self.queue_name.var.code}}).enqueue(" \ - "out_msg, Cycles($rcode));") + "out_msg, clockEdge(), cyclesToTicks(Cycles($rcode)));") else: - code("(${{self.queue_name.var.code}}).enqueue(out_msg);") + code("(${{self.queue_name.var.code}}).enqueue(out_msg, "\ + "clockEdge(), cyclesToTicks(Cycles(1)));") # End scope self.symtab.popFrame() diff --git a/src/mem/slicc/ast/ObjDeclAST.py b/src/mem/slicc/ast/ObjDeclAST.py index 7cea70b32..efc7ef928 100644 --- a/src/mem/slicc/ast/ObjDeclAST.py +++ b/src/mem/slicc/ast/ObjDeclAST.py @@ -55,6 +55,8 @@ class ObjDeclAST(DeclAST): c_code = "m_machineID" elif self.ident == "clusterID": c_code = "m_clusterID" + elif self.ident == "recycle_latency": + c_code = "m_recycle_latency" else: c_code = "(*m_%s_ptr)" % (self.ident) diff --git a/src/mem/slicc/ast/PeekStatementAST.py b/src/mem/slicc/ast/PeekStatementAST.py index f5ef91daf..00d26e908 100644 --- a/src/mem/slicc/ast/PeekStatementAST.py +++ b/src/mem/slicc/ast/PeekStatementAST.py @@ -77,7 +77,7 @@ class PeekStatementAST(StatementAST): if (m_is_blocking && (m_block_map.count(in_msg_ptr->m_$address_field) == 1) && (m_block_map[in_msg_ptr->m_$address_field] != &$qcode)) { - $qcode.delayHead(); + $qcode.delayHead(clockEdge(), cyclesToTicks(Cycles(1))); continue; } ''') diff --git a/src/mem/slicc/ast/StallAndWaitStatementAST.py b/src/mem/slicc/ast/StallAndWaitStatementAST.py index b2f622871..6ab2888b7 100644 --- a/src/mem/slicc/ast/StallAndWaitStatementAST.py +++ b/src/mem/slicc/ast/StallAndWaitStatementAST.py @@ -45,5 +45,5 @@ class StallAndWaitStatementAST(StatementAST): address_code = self.address.var.code code(''' stallBuffer(&($in_port_code), $address_code); - $in_port_code.stallMessage($address_code); + $in_port_code.stallMessage($address_code, clockEdge()); ''') diff --git a/src/mem/slicc/symbols/StateMachine.py b/src/mem/slicc/symbols/StateMachine.py index 42a81c096..015d902b4 100644 --- a/src/mem/slicc/symbols/StateMachine.py +++ b/src/mem/slicc/symbols/StateMachine.py @@ -580,24 +580,10 @@ $c_ident::initNetQueues() m_net_ptr->set${network}NetQueue(m_version + base, $vid->getOrdered(), $vnet, "$vnet_type", $vid); ''') - # Set the end - if network == "To": - code('$vid->setSender(this);') - else: - code('$vid->setReceiver(this);') - # Set Priority if "rank" in var: code('$vid->setPriority(${{var["rank"]}})') - else: - if var.type_ast.type.c_ident == "MessageBuffer": - code('$vid->setReceiver(this);') - if var.ident.find("triggerQueue") >= 0: - code('$vid->setSender(this);') - elif var.ident.find("optionalQueue") >= 0: - code('$vid->setSender(this);') - code.dedent() code(''' } @@ -637,9 +623,6 @@ $c_ident::init() comment = "Type %s default" % vtype.ident code('*$vid = ${{vtype["default"]}}; // $comment') - if vtype.c_ident == "TimerTable": - code('$vid->setClockObj(this);') - # Set the prefetchers code() for prefetcher in self.prefetchers: @@ -1293,7 +1276,7 @@ ${ident}_Controller::doTransitionWorker(${ident}_Event event, res = trans.resources for key,val in res.iteritems(): val = ''' -if (!%s.areNSlotsAvailable(%s)) +if (!%s.areNSlotsAvailable(%s, clockEdge())) return TransitionResult_ResourceStall; ''' % (key.code, val) case_sorter.append(val) -- 2.30.2