From 9357e353fc976a409fb0cb3a875b402f452577f7 Mon Sep 17 00:00:00 2001 From: Korey Sewell Date: Sun, 31 Jan 2010 18:30:48 -0500 Subject: [PATCH] inorder: inst count mgmt --- src/cpu/inorder/SConscript | 2 + src/cpu/inorder/cpu.cc | 38 ++++--- src/cpu/inorder/cpu.hh | 2 + src/cpu/inorder/inorder_dyn_inst.cc | 8 +- src/cpu/inorder/inorder_dyn_inst.hh | 5 +- src/cpu/inorder/pipeline_stage.cc | 26 ++++- src/cpu/inorder/reg_dep_map.cc | 24 +++++ src/cpu/inorder/reg_dep_map.hh | 2 + src/cpu/inorder/resource.cc | 34 +++--- src/cpu/inorder/resource.hh | 11 +- src/cpu/inorder/resources/cache_unit.cc | 105 ++++++++++++++----- src/cpu/inorder/resources/cache_unit.hh | 5 +- src/cpu/inorder/resources/graduation_unit.cc | 2 - src/cpu/inorder/resources/use_def.cc | 15 ++- 14 files changed, 211 insertions(+), 68 deletions(-) diff --git a/src/cpu/inorder/SConscript b/src/cpu/inorder/SConscript index afc6a29e4..f222350af 100644 --- a/src/cpu/inorder/SConscript +++ b/src/cpu/inorder/SConscript @@ -54,6 +54,8 @@ if 'InOrderCPU' in env['CPU_MODELS']: TraceFlag('InOrderGraduation') TraceFlag('ThreadModel') TraceFlag('RefCount') + TraceFlag('AddrDep') + CompoundFlag('InOrderCPUAll', [ 'InOrderStage', 'InOrderStall', 'InOrderCPU', 'InOrderMDU', 'InOrderAGEN', 'InOrderFetchSeq', 'InOrderTLB', 'InOrderBPred', diff --git a/src/cpu/inorder/cpu.cc b/src/cpu/inorder/cpu.cc index e864c8c86..e28af9e7a 100644 --- a/src/cpu/inorder/cpu.cc +++ b/src/cpu/inorder/cpu.cc @@ -333,6 +333,12 @@ InOrderCPU::InOrderCPU(Params *params) 0); } + dummyReqInst = new InOrderDynInst(this, NULL, 0, 0, 0); + dummyReqInst->setSquashed(); + + dummyBufferInst = new InOrderDynInst(this, NULL, 0, 0, 0); + dummyBufferInst->setSquashed(); + lastRunningCycle = curTick; // Reset CPU to reset state. @@ -343,6 +349,8 @@ InOrderCPU::InOrderCPU(Params *params) reset(); #endif + dummyBufferInst->resetInstCount(); + // Schedule First Tick Event, CPU will reschedule itself from here on out. scheduleTickEvent(0); } @@ -1176,6 +1184,8 @@ InOrderCPU::instDone(DynInstPtr inst, ThreadID tid) removeInst(inst); } +// currently unused function, but substitute repetitive code w/this function +// call void InOrderCPU::addToRemoveList(DynInstPtr &inst) { @@ -1194,6 +1204,10 @@ InOrderCPU::removeInst(DynInstPtr &inst) removeInstsThisCycle = true; // Remove the instruction. + + DPRINTF(RefCount, "Pushing instruction [tid:%i] PC %#x " + "[sn:%lli] to remove list\n", + inst->threadNumber, inst->readPC(), inst->seqNum); removeList.push(inst->getInstListIt()); } @@ -1208,7 +1222,7 @@ InOrderCPU::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) inst_iter--; - DPRINTF(InOrderCPU, "Deleting instructions from CPU instruction " + DPRINTF(InOrderCPU, "Squashing instructions from CPU instruction " "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", tid, seq_num, (*inst_iter)->seqNum); @@ -1238,6 +1252,9 @@ InOrderCPU::squashInstIt(const ListIt &instIt, ThreadID tid) (*instIt)->setSquashed(); + DPRINTF(RefCount, "Pushing instruction [tid:%i] PC %#x " + "[sn:%lli] to remove list\n", + (*instIt)->threadNumber, (*instIt)->readPC(), (*instIt)->seqNum); removeList.push(instIt); } } @@ -1251,7 +1268,7 @@ InOrderCPU::cleanUpRemovedInsts() "[tid:%i] [sn:%lli] PC %#x\n", (*removeList.front())->threadNumber, (*removeList.front())->seqNum, - (*removeList.front())->readPC()); + (*removeList.front())->readPC()); DynInstPtr inst = *removeList.front(); ThreadID tid = inst->threadNumber; @@ -1279,11 +1296,6 @@ InOrderCPU::cleanUpRemovedInsts() instList[tid].erase(removeList.front()); removeList.pop(); - - DPRINTF(RefCount, "pop from remove list: [sn:%i]: Refcount = %i.\n", - inst->seqNum, - 0/*inst->curCount()*/); - } removeInstsThisCycle = false; @@ -1295,22 +1307,18 @@ InOrderCPU::cleanUpRemovedReqs() while (!reqRemoveList.empty()) { ResourceRequest *res_req = reqRemoveList.front(); - DPRINTF(RefCount, "[tid:%i]: Removing Request, " - "[sn:%lli] [slot:%i] [stage_num:%i] [res:%s] [refcount:%i].\n", + DPRINTF(InOrderCPU, "[tid:%i] [sn:%lli]: Removing Request " + "[stage_num:%i] [res:%s] [slot:%i] [completed:%i].\n", res_req->inst->threadNumber, res_req->inst->seqNum, - res_req->getSlot(), res_req->getStageNum(), res_req->res->name(), - 0/*res_req->inst->curCount()*/); + (res_req->isCompleted()) ? res_req->getComplSlot() : res_req->getSlot(), + res_req->isCompleted()); reqRemoveList.pop(); delete res_req; - - DPRINTF(RefCount, "after remove request: [sn:%i]: Refcount = %i.\n", - res_req->inst->seqNum, - 0/*res_req->inst->curCount()*/); } } diff --git a/src/cpu/inorder/cpu.hh b/src/cpu/inorder/cpu.hh index 253b5b18f..0c42f349e 100644 --- a/src/cpu/inorder/cpu.hh +++ b/src/cpu/inorder/cpu.hh @@ -247,6 +247,8 @@ class InOrderCPU : public BaseCPU /** Instruction used to signify that there is no *real* instruction in buffer slot */ DynInstPtr dummyInst[ThePipeline::MaxThreads]; + DynInstPtr dummyBufferInst; + DynInstPtr dummyReqInst; /** Used by resources to signify a denied access to a resource. */ ResourceRequest *dummyReq[ThePipeline::MaxThreads]; diff --git a/src/cpu/inorder/inorder_dyn_inst.cc b/src/cpu/inorder/inorder_dyn_inst.cc index c0e5aa69b..75e1c570f 100644 --- a/src/cpu/inorder/inorder_dyn_inst.cc +++ b/src/cpu/inorder/inorder_dyn_inst.cc @@ -164,7 +164,7 @@ InOrderDynInst::initVars() // Update Instruction Count for this instruction ++instcount; - if (instcount > 500) { + if (instcount > 100) { fatal("Number of Active Instructions in CPU is too high. " "(Not Dereferencing Ptrs. Correctly?)\n"); } @@ -175,6 +175,12 @@ InOrderDynInst::initVars() threadNumber, seqNum, instcount); } +void +InOrderDynInst::resetInstCount() +{ + instcount = 0; +} + InOrderDynInst::~InOrderDynInst() { diff --git a/src/cpu/inorder/inorder_dyn_inst.hh b/src/cpu/inorder/inorder_dyn_inst.hh index ad4da9aab..8a5f9cf25 100644 --- a/src/cpu/inorder/inorder_dyn_inst.hh +++ b/src/cpu/inorder/inorder_dyn_inst.hh @@ -1032,14 +1032,15 @@ class InOrderDynInst : public FastAlloc, public RefCounted /** Count of total number of dynamic instructions. */ static int instcount; + void resetInstCount(); + /** Dumps out contents of this BaseDynInst. */ void dump(); /** Dumps out contents of this BaseDynInst into given string. */ void dump(std::string &outstring); - - //inline int curCount() { return curCount(); } + //inline int curCount() { return curCount(); } }; diff --git a/src/cpu/inorder/pipeline_stage.cc b/src/cpu/inorder/pipeline_stage.cc index c991fe1bd..571cf10bb 100644 --- a/src/cpu/inorder/pipeline_stage.cc +++ b/src/cpu/inorder/pipeline_stage.cc @@ -101,8 +101,6 @@ PipelineStage::setCPU(InOrderCPU *cpu_ptr) { cpu = cpu_ptr; - dummyBufferInst = new InOrderDynInst(cpu_ptr, NULL, 0, 0, 0); - DPRINTF(InOrderStage, "Set CPU pointer.\n"); tracer = dynamic_cast(cpu->getTracer()); @@ -388,6 +386,8 @@ PipelineStage::squashPrevStageInsts(InstSeqNum squash_seq_num, ThreadID tid) prevStage->insts[i]->seqNum, prevStage->insts[i]->readPC()); prevStage->insts[i]->setSquashed(); + + prevStage->insts[i] = cpu->dummyBufferInst; } } } @@ -609,7 +609,7 @@ PipelineStage::sortInsts() skidBuffer[tid].push(prevStage->insts[i]); - prevStage->insts[i] = dummyBufferInst; + prevStage->insts[i] = cpu->dummyBufferInst; } } @@ -816,7 +816,7 @@ PipelineStage::processThread(bool &status_change, ThreadID tid) // call processInsts() // If status is Unblocking, // buffer any instructions coming from fetch - // continue trying to empty skid buffer + // continue trying to empty skid buffer // check if stall conditions have passed // Stage should try to process as many instructions as its bandwidth @@ -960,6 +960,8 @@ PipelineStage::processInstSchedule(DynInstPtr inst,int &reqs_processed) } reqs_processed++; + + req->stagePasses++; } else { DPRINTF(InOrderStage, "[tid:%i]: [sn:%i] request to %s failed." "\n", tid, inst->seqNum, cpu->resPool->name(res_num)); @@ -969,7 +971,7 @@ PipelineStage::processInstSchedule(DynInstPtr inst,int &reqs_processed) if (req->isMemStall() && cpu->threadModel == InOrderCPU::SwitchOnCacheMiss) { // Save Stalling Instruction - DPRINTF(ThreadModel, "[tid:%i] Detected cache miss.\n", tid); + DPRINTF(ThreadModel, "[tid:%i] [sn:%i] Detected cache miss.\n", tid, inst->seqNum); DPRINTF(InOrderStage, "Inserting [tid:%i][sn:%i] into switch out buffer.\n", tid, inst->seqNum); @@ -994,6 +996,20 @@ PipelineStage::processInstSchedule(DynInstPtr inst,int &reqs_processed) cpu->activateNextReadyContext(); } + // Mark request for deletion + // if it isnt currently being used by a resource + if (!req->hasSlot()) { + DPRINTF(InOrderStage, "[sn:%i] Deleting Request, has no slot in resource.\n", + inst->seqNum); + + cpu->reqRemoveList.push(req); + } else { + DPRINTF(InOrderStage, "[sn:%i] Ignoring Request Deletion, in resource [slot:%i].\n", + inst->seqNum, req->getSlot()); + //req = cpu->dummyReq[tid]; + } + + break; } diff --git a/src/cpu/inorder/reg_dep_map.cc b/src/cpu/inorder/reg_dep_map.cc index 51782a588..7fac0a905 100644 --- a/src/cpu/inorder/reg_dep_map.cc +++ b/src/cpu/inorder/reg_dep_map.cc @@ -235,3 +235,27 @@ RegDepMap::findBypassInst(unsigned idx) return NULL; } + +void +RegDepMap::dump() +{ + + for (int idx=0; idx < regMap.size(); idx++) { + + if (regMap[idx].size() > 0) { + cprintf("Reg #%i (size:%i): ", idx, regMap[idx].size()); + + std::list::iterator list_it = regMap[idx].begin(); + std::list::iterator list_end = regMap[idx].end(); + + while (list_it != list_end) { + cprintf("[sn:%i] ", (*list_it)->seqNum); + + list_it++; + } + + cprintf("\n"); + } + + } +} diff --git a/src/cpu/inorder/reg_dep_map.hh b/src/cpu/inorder/reg_dep_map.hh index b78e211bb..cb9d35bf4 100644 --- a/src/cpu/inorder/reg_dep_map.hh +++ b/src/cpu/inorder/reg_dep_map.hh @@ -88,6 +88,8 @@ class RegDepMap /** Size of Dependency of Map */ int depSize(unsigned idx); + void dump(); + protected: // Eventually make this a map of lists for // efficiency sake! diff --git a/src/cpu/inorder/resource.cc b/src/cpu/inorder/resource.cc index dcf5f3117..e63925fe8 100644 --- a/src/cpu/inorder/resource.cc +++ b/src/cpu/inorder/resource.cc @@ -101,12 +101,6 @@ Resource::slotsInUse() void Resource::freeSlot(int slot_idx) { - DPRINTF(RefCount, "Removing [tid:%i] [sn:%i]'s request from resource " - "[slot:%i].\n", - reqMap[slot_idx]->inst->readTid(), - reqMap[slot_idx]->inst->seqNum, - slot_idx); - // Put slot number on this resource's free list availSlots.push_back(slot_idx); @@ -181,7 +175,7 @@ Resource::request(DynInstPtr inst) // See if the resource is already serving this instruction. // If so, use that request; bool try_request = false; - int slot_num; + int slot_num = -1; int stage_num; ResReqPtr inst_req = findRequest(inst); @@ -440,6 +434,10 @@ ResourceRequest::ResourceRequest(Resource *_res, DynInstPtr _inst, } #endif + + stagePasses = 0; + complSlotNum = -1; + } ResourceRequest::~ResourceRequest() @@ -454,17 +452,29 @@ ResourceRequest::~ResourceRequest() void ResourceRequest::done(bool completed) { - DPRINTF(Resource, "%s done with request from [sn:%i] [tid:%i].\n", - res->name(), inst->seqNum, inst->readTid()); + DPRINTF(Resource, "%s [slot:%i] done with request from [sn:%i] [tid:%i].\n", + res->name(), slotNum, inst->seqNum, inst->readTid()); setCompleted(completed); - // Add to remove list - res->cpu->reqRemoveList.push(res->reqMap[slotNum]); - + // Used for debugging purposes + if (completed) { + complSlotNum = slotNum; + + // Would like to start a convention such as all requests deleted in resources/pipeline + // but a little more complex then it seems... + // For now, all COMPLETED requests deleted in resource.. + // all FAILED requests deleted in pipeline stage + // *all SQUASHED requests deleted in resource + res->cpu->reqRemoveList.push(res->reqMap[slotNum]); + } + // Free Slot So Another Instruction Can Use This Resource res->freeSlot(slotNum); + // change slot # to -1, since we check slotNum to see if request is still valid + slotNum = -1; + res->instReqsProcessed++; } diff --git a/src/cpu/inorder/resource.hh b/src/cpu/inorder/resource.hh index 383340df2..b9650df18 100644 --- a/src/cpu/inorder/resource.hh +++ b/src/cpu/inorder/resource.hh @@ -331,6 +331,8 @@ class ResourceRequest */ void done(bool completed = true); + short stagePasses; + ///////////////////////////////////////////// // // GET RESOURCE REQUEST IDENTIFICATION / INFO @@ -339,8 +341,11 @@ class ResourceRequest /** Get Resource Index */ int getResIdx() { return resIdx; } + /** Get Slot Number */ int getSlot() { return slotNum; } + int getComplSlot() { return complSlotNum; } + bool hasSlot() { return slotNum >= 0; } /** Get Stage Number */ int getStageNum() { return stageNum; } @@ -363,6 +368,9 @@ class ResourceRequest /** Instruction being used */ DynInstPtr inst; + /** Not guaranteed to be set, used for debugging */ + InstSeqNum seqNum; + /** Fault Associated With This Resource Request */ Fault fault; @@ -396,7 +404,8 @@ class ResourceRequest int stageNum; int resIdx; int slotNum; - + int complSlotNum; + /** Resource Request Status */ bool completed; bool squashed; diff --git a/src/cpu/inorder/resources/cache_unit.cc b/src/cpu/inorder/resources/cache_unit.cc index 3fa1ed180..00058163f 100644 --- a/src/cpu/inorder/resources/cache_unit.cc +++ b/src/cpu/inorder/resources/cache_unit.cc @@ -155,14 +155,11 @@ CacheUnit::getSlot(DynInstPtr inst) return -1; inst->memTime = curTick; - addrList[tid].push_back(req_addr); - addrMap[tid][req_addr] = inst->seqNum; - DPRINTF(InOrderCachePort, - "[tid:%i]: [sn:%i]: Address %08p added to dependency list\n", - inst->readTid(), inst->seqNum, req_addr); + setAddrDependency(inst); return new_slot; } else { // Allow same instruction multiple accesses to same address + // should only happen maybe after a squashed inst. needs to replay if (addrMap[tid][req_addr] == inst->seqNum) { int new_slot = Resource::getSlot(inst); @@ -183,31 +180,45 @@ CacheUnit::getSlot(DynInstPtr inst) } void -CacheUnit::freeSlot(int slot_num) +CacheUnit::setAddrDependency(DynInstPtr inst) { - ThreadID tid = reqMap[slot_num]->inst->readTid(); - - vector::iterator vect_it = - find(addrList[tid].begin(), addrList[tid].end(), - reqMap[slot_num]->inst->getMemAddr()); - - assert(vect_it != addrList[tid].end() || - reqMap[slot_num]->inst->splitInst); + Addr req_addr = inst->getMemAddr(); + ThreadID tid = inst->readTid(); + addrList[tid].push_back(req_addr); + addrMap[tid][req_addr] = inst->seqNum; DPRINTF(InOrderCachePort, - "[tid:%i]: Address %08p removed from dependency list\n", - reqMap[slot_num]->inst->readTid(), (*vect_it)); + "[tid:%i]: [sn:%i]: Address %08p added to dependency list\n", + inst->readTid(), inst->seqNum, req_addr); + DPRINTF(AddrDep, + "[tid:%i]: [sn:%i]: Address %08p added to dependency list\n", + inst->readTid(), inst->seqNum, req_addr); +} + +void +CacheUnit::removeAddrDependency(DynInstPtr inst) +{ + ThreadID tid = inst->readTid(); + + Addr mem_addr = inst->getMemAddr(); + + // Erase from Address List + vector::iterator vect_it = find(addrList[tid].begin(), addrList[tid].end(), + mem_addr); + assert(vect_it != addrList[tid].end() || inst->splitInst); if (vect_it != addrList[tid].end()) { - - DPRINTF(InOrderCachePort, - "[tid:%i]: Address %08p removed from dependency list\n", - reqMap[slot_num]->inst->readTid(), (*vect_it)); - - addrList[tid].erase(vect_it); - } + DPRINTF(AddrDep, + "[tid:%i]: [sn:%i] Address %08p removed from dependency list\n", + inst->readTid(), inst->seqNum, (*vect_it)); + + addrList[tid].erase(vect_it); + + // Erase From Address Map (Used for Debugging) + addrMap[tid].erase(addrMap[tid].find(mem_addr)); + } + - Resource::freeSlot(slot_num); } ResReqPtr @@ -687,8 +698,14 @@ CacheUnit::execute(int slot_num) DPRINTF(InOrderCachePort, "[tid:%i]: Instruction [sn:%i] is: %s\n", tid, seq_num, inst->staticInst->disassemble(inst->PC)); + removeAddrDependency(inst); + delete cache_req->dataPkt; - //cache_req->setMemStall(false); + + // Do not stall and switch threads for fetch... for now.. + // TODO: We need to detect cache misses for latencies > 1 + // cache_req->setMemStall(false); + cache_req->done(); } else { DPRINTF(InOrderCachePort, @@ -711,6 +728,7 @@ CacheUnit::execute(int slot_num) if (cache_req->isMemAccComplete() || inst->isDataPrefetch() || inst->isInstPrefetch()) { + removeAddrDependency(inst); cache_req->setMemStall(false); cache_req->done(); } else { @@ -729,6 +747,7 @@ CacheUnit::execute(int slot_num) if (cache_req->isMemAccComplete() || inst->isDataPrefetch() || inst->isInstPrefetch()) { + removeAddrDependency(inst); cache_req->setMemStall(false); cache_req->done(); } else { @@ -747,6 +766,7 @@ CacheUnit::execute(int slot_num) if (cache_req->isMemAccComplete() || inst->isDataPrefetch() || inst->isInstPrefetch()) { + removeAddrDependency(inst); cache_req->setMemStall(false); cache_req->done(); } else { @@ -911,6 +931,10 @@ CacheUnit::processCacheCompletion(PacketPtr pkt) "Ignoring completion of squashed access, [tid:%i] [sn:%i]\n", cache_pkt->cacheReq->getInst()->readTid(), cache_pkt->cacheReq->getInst()->seqNum); + DPRINTF(RefCount, + "Ignoring completion of squashed access, [tid:%i] [sn:%i]\n", + cache_pkt->cacheReq->getTid(), + cache_pkt->cacheReq->seqNum); cache_pkt->cacheReq->done(); delete cache_pkt; @@ -1154,6 +1178,14 @@ CacheUnit::squash(DynInstPtr inst, int stage_num, "[tid:%i] Squashing request from [sn:%i]\n", req_ptr->getInst()->readTid(), req_ptr->getInst()->seqNum); + if (req_ptr->isSquashed()) { + DPRINTF(AddrDep, "Request for [tid:%i] [sn:%i] already squashed, ignoring squash process.\n", + req_ptr->getInst()->readTid(), + req_ptr->getInst()->seqNum); + map_it++; + continue; + } + req_ptr->setSquashed(); req_ptr->getInst()->setSquashed(); @@ -1178,7 +1210,29 @@ CacheUnit::squash(DynInstPtr inst, int stage_num, // Mark slot for removal from resource slot_remove_list.push_back(req_ptr->getSlot()); + + DPRINTF(InOrderCachePort, + "[tid:%i] Squashing request from [sn:%i]\n", + req_ptr->getInst()->readTid(), req_ptr->getInst()->seqNum); + } else { + DPRINTF(InOrderCachePort, + "[tid:%i] Request from [sn:%i] squashed, but still pending completion.\n", + req_ptr->getInst()->readTid(), req_ptr->getInst()->seqNum); + DPRINTF(RefCount, + "[tid:%i] Request from [sn:%i] squashed (split:%i), but still pending completion.\n", + req_ptr->getInst()->readTid(), req_ptr->getInst()->seqNum, + req_ptr->getInst()->splitInst); } + + if (req_ptr->getInst()->validMemAddr()) { + DPRINTF(AddrDep, "Squash of [tid:%i] [sn:%i], attempting to remove addr. %08p dependencies.\n", + req_ptr->getInst()->readTid(), + req_ptr->getInst()->seqNum, + req_ptr->getInst()->getMemAddr()); + + removeAddrDependency(req_ptr->getInst()); + } + } map_it++; @@ -1320,3 +1374,4 @@ CacheUnit::write(DynInstPtr inst, int32_t data, Addr addr, unsigned flags, { return write(inst, (uint32_t)data, addr, flags, res); } + diff --git a/src/cpu/inorder/resources/cache_unit.hh b/src/cpu/inorder/resources/cache_unit.hh index 8200ace87..9004f3b93 100644 --- a/src/cpu/inorder/resources/cache_unit.hh +++ b/src/cpu/inorder/resources/cache_unit.hh @@ -135,8 +135,6 @@ class CacheUnit : public Resource int getSlot(DynInstPtr inst); - void freeSlot(int slot_num); - /** Execute the function of this resource. The Default is action * is to do nothing. More specific models will derive from this * class and define their own execute function. @@ -184,6 +182,9 @@ class CacheUnit : public Resource uint64_t getMemData(Packet *packet); + void setAddrDependency(DynInstPtr inst); + void removeAddrDependency(DynInstPtr inst); + protected: /** Cache interface. */ CachePort *cachePort; diff --git a/src/cpu/inorder/resources/graduation_unit.cc b/src/cpu/inorder/resources/graduation_unit.cc index 2d7cd5c8c..2dad9889a 100644 --- a/src/cpu/inorder/resources/graduation_unit.cc +++ b/src/cpu/inorder/resources/graduation_unit.cc @@ -79,8 +79,6 @@ GraduationUnit::execute(int slot_num) "[tid:%i] Graduating instruction [sn:%i].\n", tid, inst->seqNum); - DPRINTF(RefCount, "Refcount = %i.\n", 0/*inst->curCount()*/); - // Release Non-Speculative "Block" on instructions that could not execute // because there was a non-speculative inst. active. // @TODO: Fix this functionality. Probably too conservative. diff --git a/src/cpu/inorder/resources/use_def.cc b/src/cpu/inorder/resources/use_def.cc index a4f3a0d21..5fd6a4724 100644 --- a/src/cpu/inorder/resources/use_def.cc +++ b/src/cpu/inorder/resources/use_def.cc @@ -191,6 +191,7 @@ UseDefUnit::execute(int slot_idx) DPRINTF(InOrderStall, "STALL: [tid:%i]: waiting for " "[sn:%i] to write\n", tid, outReadSeqNum[tid]); + ud_req->done(false); } } else { @@ -249,6 +250,7 @@ UseDefUnit::execute(int slot_idx) DPRINTF(InOrderStall, "STALL: [tid:%i]: waiting for " "[sn:%i] to forward\n", tid, outReadSeqNum[tid]); + ud_req->done(false); } } else { DPRINTF(InOrderUseDef, "[tid:%i]: Source register idx: %i" @@ -258,6 +260,7 @@ UseDefUnit::execute(int slot_idx) "register (idx=%i)\n", tid, reg_idx); outReadSeqNum[tid] = inst->seqNum; + ud_req->done(false); } } } @@ -360,6 +363,7 @@ UseDefUnit::execute(int slot_idx) DPRINTF(InOrderStall, "STALL: [tid:%i]: waiting for " "[sn:%i] to read\n", tid, outReadSeqNum); + ud_req->done(false); } } else { DPRINTF(InOrderUseDef, "[tid:%i]: Dest. register idx: %i is " @@ -369,6 +373,7 @@ UseDefUnit::execute(int slot_idx) "register (idx=%i)\n", tid, reg_idx); outWriteSeqNum[tid] = inst->seqNum; + ud_req->done(false); } } break; @@ -402,12 +407,16 @@ UseDefUnit::squash(DynInstPtr inst, int stage_num, InstSeqNum squash_seq_num, req_ptr->getInst()->readTid(), req_ptr->getInst()->seqNum); - regDepMap[tid]->remove(req_ptr->getInst()); - int req_slot_num = req_ptr->getSlot(); - if (latency > 0) + if (latency > 0) { + assert(0); + unscheduleEvent(req_slot_num); + } + + // Mark request for later removal + cpu->reqRemoveList.push(req_ptr); // Mark slot for removal from resource slot_remove_list.push_back(req_ptr->getSlot());