diff options
author | Alexandru Dutu <alexandru.dutu@amd.com> | 2016-09-16 12:26:52 -0400 |
---|---|---|
committer | Alexandru Dutu <alexandru.dutu@amd.com> | 2016-09-16 12:26:52 -0400 |
commit | 589e13a23b3969c1137d2170a8638356d0c0fc65 (patch) | |
tree | 3ccfcde3173d9d3c234e0fa9ce130931786981eb /src/gpu-compute | |
parent | e9fe1b838b94eb46f165b9adb6281a862b172dfe (diff) | |
download | gem5-589e13a23b3969c1137d2170a8638356d0c0fc65.tar.xz |
gpu-compute: Wavefront refactoring
Renaming members of the Wavefront class in accordance with the style guide.
Diffstat (limited to 'src/gpu-compute')
-rw-r--r-- | src/gpu-compute/compute_unit.cc | 72 | ||||
-rw-r--r-- | src/gpu-compute/dispatcher.cc | 2 | ||||
-rw-r--r-- | src/gpu-compute/fetch_unit.cc | 2 | ||||
-rw-r--r-- | src/gpu-compute/global_memory_pipeline.cc | 6 | ||||
-rw-r--r-- | src/gpu-compute/local_memory_pipeline.cc | 6 | ||||
-rw-r--r-- | src/gpu-compute/wavefront.cc | 132 | ||||
-rw-r--r-- | src/gpu-compute/wavefront.hh | 80 |
7 files changed, 150 insertions, 150 deletions
diff --git a/src/gpu-compute/compute_unit.cc b/src/gpu-compute/compute_unit.cc index 83e2414db..32fa3bd6a 100644 --- a/src/gpu-compute/compute_unit.cc +++ b/src/gpu-compute/compute_unit.cc @@ -178,13 +178,13 @@ ComputeUnit::FillKernelState(Wavefront *w, NDRange *ndr) { w->resizeRegFiles(ndr->q.cRegCount, ndr->q.sRegCount, ndr->q.dRegCount); - w->workgroupsz[0] = ndr->q.wgSize[0]; - w->workgroupsz[1] = ndr->q.wgSize[1]; - w->workgroupsz[2] = ndr->q.wgSize[2]; - w->wg_sz = w->workgroupsz[0] * w->workgroupsz[1] * w->workgroupsz[2]; - w->gridsz[0] = ndr->q.gdSize[0]; - w->gridsz[1] = ndr->q.gdSize[1]; - w->gridsz[2] = ndr->q.gdSize[2]; + w->workGroupSz[0] = ndr->q.wgSize[0]; + w->workGroupSz[1] = ndr->q.wgSize[1]; + w->workGroupSz[2] = ndr->q.wgSize[2]; + w->wgSz = w->workGroupSz[0] * w->workGroupSz[1] * w->workGroupSz[2]; + w->gridSz[0] = ndr->q.gdSize[0]; + w->gridSz[1] = ndr->q.gdSize[1]; + w->gridSz[2] = ndr->q.gdSize[2]; w->kernelArgs = ndr->q.args; w->privSizePerItem = ndr->q.privMemPerItem; w->spillSizePerItem = ndr->q.spillMemPerItem; @@ -236,29 +236,29 @@ ComputeUnit::StartWF(Wavefront *w, int trueWgSize[], int trueWgSizeTotal, init_mask[k] = 1; } - w->kern_id = ndr->dispatchId; - w->dynwaveid = cnt; - w->init_mask = init_mask.to_ullong(); + w->kernId = ndr->dispatchId; + w->dynWaveId = cnt; + w->initMask = init_mask.to_ullong(); for (int k = 0; k < wfSize(); ++k) { - w->workitemid[0][k] = (k+cnt*wfSize()) % trueWgSize[0]; - w->workitemid[1][k] = + w->workItemId[0][k] = (k+cnt*wfSize()) % trueWgSize[0]; + w->workItemId[1][k] = ((k + cnt * wfSize()) / trueWgSize[0]) % trueWgSize[1]; - w->workitemid[2][k] = + w->workItemId[2][k] = (k + cnt * wfSize()) / (trueWgSize[0] * trueWgSize[1]); - w->workitemFlatId[k] = w->workitemid[2][k] * trueWgSize[0] * - trueWgSize[1] + w->workitemid[1][k] * trueWgSize[0] + - w->workitemid[0][k]; + w->workItemFlatId[k] = w->workItemId[2][k] * trueWgSize[0] * + trueWgSize[1] + w->workItemId[1][k] * trueWgSize[0] + + w->workItemId[0][k]; } - w->barrier_slots = divCeil(trueWgSizeTotal, wfSize()); + w->barrierSlots = divCeil(trueWgSizeTotal, wfSize()); - w->bar_cnt.resize(wfSize(), 0); + w->barCnt.resize(wfSize(), 0); - w->max_bar_cnt = 0; - w->old_barrier_cnt = 0; - w->barrier_cnt = 0; + w->maxBarCnt = 0; + w->oldBarrierCnt = 0; + w->barrierCnt = 0; w->privBase = ndr->q.privMemStart; ndr->q.privMemStart += ndr->q.privMemPerItem * wfSize(); @@ -269,22 +269,22 @@ ComputeUnit::StartWF(Wavefront *w, int trueWgSize[], int trueWgSizeTotal, w->pushToReconvergenceStack(0, UINT32_MAX, init_mask.to_ulong()); // WG state - w->wg_id = ndr->globalWgId; - w->dispatchid = ndr->dispatchId; - w->workgroupid[0] = w->wg_id % ndr->numWg[0]; - w->workgroupid[1] = (w->wg_id / ndr->numWg[0]) % ndr->numWg[1]; - w->workgroupid[2] = w->wg_id / (ndr->numWg[0] * ndr->numWg[1]); + w->wgId = ndr->globalWgId; + w->dispatchId = ndr->dispatchId; + w->workGroupId[0] = w->wgId % ndr->numWg[0]; + w->workGroupId[1] = (w->wgId / ndr->numWg[0]) % ndr->numWg[1]; + w->workGroupId[2] = w->wgId / (ndr->numWg[0] * ndr->numWg[1]); - w->barrier_id = barrier_id; + w->barrierId = barrier_id; w->stalledAtBarrier = false; // set the wavefront context to have a pointer to this section of the LDS w->ldsChunk = ldsChunk; int32_t refCount M5_VAR_USED = - lds.increaseRefCounter(w->dispatchid, w->wg_id); + lds.increaseRefCounter(w->dispatchId, w->wgId); DPRINTF(GPUDisp, "CU%d: increase ref ctr wg[%d] to [%d]\n", - cu_id, w->wg_id, refCount); + cu_id, w->wgId, refCount); w->instructionBuffer.clear(); @@ -468,15 +468,15 @@ ComputeUnit::AllAtBarrier(uint32_t _barrier_id, uint32_t bcnt, uint32_t bslots) DPRINTF(GPUSync, "Checking WF[%d][%d]\n", i_simd, i_wf); DPRINTF(GPUSync, "wf->barrier_id = %d, _barrier_id = %d\n", - w->barrier_id, _barrier_id); + w->barrierId, _barrier_id); DPRINTF(GPUSync, "wf->barrier_cnt %d, bcnt = %d\n", - w->barrier_cnt, bcnt); + w->barrierCnt, bcnt); } if (w->status == Wavefront::S_RUNNING && - w->barrier_id == _barrier_id && w->barrier_cnt == bcnt && - !w->outstanding_reqs) { + w->barrierId == _barrier_id && w->barrierCnt == bcnt && + !w->outstandingReqs) { ++ccnt; DPRINTF(GPUSync, "WF[%d][%d] at barrier, increment ccnt to " @@ -646,17 +646,17 @@ ComputeUnit::DataPort::recvTimingResp(PacketPtr pkt) if (w->status == Wavefront::S_RETURNING) { DPRINTF(GPUDisp, "CU%d: WF[%d][%d][wv=%d]: WG id completed %d\n", computeUnit->cu_id, w->simdId, w->wfSlotId, - w->wfDynId, w->kern_id); + w->wfDynId, w->kernId); computeUnit->shader->dispatcher->notifyWgCompl(w); w->status = Wavefront::S_STOPPED; } else { - w->outstanding_reqs--; + w->outstandingReqs--; } DPRINTF(GPUSync, "CU%d: WF[%d][%d]: barrier_cnt = %d\n", computeUnit->cu_id, gpuDynInst->simdId, - gpuDynInst->wfSlotId, w->barrier_cnt); + gpuDynInst->wfSlotId, w->barrierCnt); if (gpuDynInst->useContinuation) { assert(gpuDynInst->scope != Enums::MEMORY_SCOPE_NONE); diff --git a/src/gpu-compute/dispatcher.cc b/src/gpu-compute/dispatcher.cc index d1d011c0d..adc1f51bd 100644 --- a/src/gpu-compute/dispatcher.cc +++ b/src/gpu-compute/dispatcher.cc @@ -305,7 +305,7 @@ GpuDispatcher::exec() void GpuDispatcher::notifyWgCompl(Wavefront *w) { - int kern_id = w->kern_id; + int kern_id = w->kernId; DPRINTF(GPUDisp, "notify WgCompl %d\n",kern_id); assert(ndRangeMap[kern_id].dispatchId == kern_id); ndRangeMap[kern_id].numWgCompleted++; diff --git a/src/gpu-compute/fetch_unit.cc b/src/gpu-compute/fetch_unit.cc index 1f0a7d78e..9104c400e 100644 --- a/src/gpu-compute/fetch_unit.cc +++ b/src/gpu-compute/fetch_unit.cc @@ -115,7 +115,7 @@ FetchUnit::initiateFetch(Wavefront *wavefront) { // calculate the virtual address to fetch from the SQC Addr vaddr = wavefront->pc() + wavefront->instructionBuffer.size(); - vaddr = wavefront->base_ptr + vaddr * sizeof(GPUStaticInst*); + vaddr = wavefront->basePtr + vaddr * sizeof(GPUStaticInst*); DPRINTF(GPUTLB, "CU%d: WF[%d][%d]: Initiating fetch translation: %#x\n", computeUnit->cu_id, wavefront->simdId, wavefront->wfSlotId, vaddr); diff --git a/src/gpu-compute/global_memory_pipeline.cc b/src/gpu-compute/global_memory_pipeline.cc index a6a4d86db..102905ec8 100644 --- a/src/gpu-compute/global_memory_pipeline.cc +++ b/src/gpu-compute/global_memory_pipeline.cc @@ -212,16 +212,16 @@ GlobalMemPipeline::doGmReturn(GPUDynInstPtr m) } // Decrement outstanding register count - computeUnit->shader->ScheduleAdd(&w->outstanding_reqs, m->time, -1); + computeUnit->shader->ScheduleAdd(&w->outstandingReqs, m->time, -1); if (m->m_op == Enums::MO_ST || MO_A(m->m_op) || MO_ANR(m->m_op) || MO_H(m->m_op)) { - computeUnit->shader->ScheduleAdd(&w->outstanding_reqs_wr_gm, m->time, + computeUnit->shader->ScheduleAdd(&w->outstandingReqsWrGm, m->time, -1); } if (m->m_op == Enums::MO_LD || MO_A(m->m_op) || MO_ANR(m->m_op)) { - computeUnit->shader->ScheduleAdd(&w->outstanding_reqs_rd_gm, m->time, + computeUnit->shader->ScheduleAdd(&w->outstandingReqsRdGm, m->time, -1); } diff --git a/src/gpu-compute/local_memory_pipeline.cc b/src/gpu-compute/local_memory_pipeline.cc index a970d8f9b..e2238bf45 100644 --- a/src/gpu-compute/local_memory_pipeline.cc +++ b/src/gpu-compute/local_memory_pipeline.cc @@ -170,16 +170,16 @@ LocalMemPipeline::doSmReturn(GPUDynInstPtr m) } // Decrement outstanding request count - computeUnit->shader->ScheduleAdd(&w->outstanding_reqs, m->time, -1); + computeUnit->shader->ScheduleAdd(&w->outstandingReqs, m->time, -1); if (m->m_op == Enums::MO_ST || MO_A(m->m_op) || MO_ANR(m->m_op) || MO_H(m->m_op)) { - computeUnit->shader->ScheduleAdd(&w->outstanding_reqs_wr_lm, + computeUnit->shader->ScheduleAdd(&w->outstandingReqsWrLm, m->time, -1); } if (m->m_op == Enums::MO_LD || MO_A(m->m_op) || MO_ANR(m->m_op)) { - computeUnit->shader->ScheduleAdd(&w->outstanding_reqs_rd_lm, + computeUnit->shader->ScheduleAdd(&w->outstandingReqsRdLm, m->time, -1); } diff --git a/src/gpu-compute/wavefront.cc b/src/gpu-compute/wavefront.cc index a20330082..c73307ac4 100644 --- a/src/gpu-compute/wavefront.cc +++ b/src/gpu-compute/wavefront.cc @@ -52,43 +52,43 @@ WavefrontParams::create() Wavefront::Wavefront(const Params *p) : SimObject(p), callArgMem(nullptr) { - last_trace = 0; + lastTrace = 0; simdId = p->simdId; wfSlotId = p->wf_slot_id; status = S_STOPPED; reservedVectorRegs = 0; startVgprIndex = 0; - outstanding_reqs = 0; - mem_reqs_in_pipe = 0; - outstanding_reqs_wr_gm = 0; - outstanding_reqs_wr_lm = 0; - outstanding_reqs_rd_gm = 0; - outstanding_reqs_rd_lm = 0; - rd_lm_reqs_in_pipe = 0; - rd_gm_reqs_in_pipe = 0; - wr_lm_reqs_in_pipe = 0; - wr_gm_reqs_in_pipe = 0; - - barrier_cnt = 0; - old_barrier_cnt = 0; + outstandingReqs = 0; + memReqsInPipe = 0; + outstandingReqsWrGm = 0; + outstandingReqsWrLm = 0; + outstandingReqsRdGm = 0; + outstandingReqsRdLm = 0; + rdLmReqsInPipe = 0; + rdGmReqsInPipe = 0; + wrLmReqsInPipe = 0; + wrGmReqsInPipe = 0; + + barrierCnt = 0; + oldBarrierCnt = 0; stalledAtBarrier = false; - mem_trace_busy = 0; - old_vgpr_tcnt = 0xffffffffffffffffll; - old_dgpr_tcnt = 0xffffffffffffffffll; - old_vgpr.resize(p->wfSize); + memTraceBusy = 0; + oldVgprTcnt = 0xffffffffffffffffll; + oldDgprTcnt = 0xffffffffffffffffll; + oldVgpr.resize(p->wfSize); pendingFetch = false; dropFetch = false; condRegState = new ConditionRegisterState(); maxSpVgprs = 0; maxDpVgprs = 0; - last_addr.resize(p->wfSize); - workitemFlatId.resize(p->wfSize); - old_dgpr.resize(p->wfSize); - bar_cnt.resize(p->wfSize); + lastAddr.resize(p->wfSize); + workItemFlatId.resize(p->wfSize); + oldDgpr.resize(p->wfSize); + barCnt.resize(p->wfSize); for (int i = 0; i < 3; ++i) { - workitemid[i].resize(p->wfSize); + workItemId[i].resize(p->wfSize); } } @@ -158,7 +158,7 @@ void Wavefront::start(uint64_t _wfDynId,uint64_t _base_ptr) { wfDynId = _wfDynId; - base_ptr = _base_ptr; + basePtr = _base_ptr; status = S_RUNNING; } @@ -333,12 +333,12 @@ Wavefront::ready(itype_e type) // Is the wave waiting at a barrier if (stalledAtBarrier) { - if (!computeUnit->AllAtBarrier(barrier_id,barrier_cnt, - computeUnit->getRefCounter(dispatchid, wg_id))) { + if (!computeUnit->AllAtBarrier(barrierId,barrierCnt, + computeUnit->getRefCounter(dispatchId, wgId))) { // Are all threads at barrier? return 0; } - old_barrier_cnt = barrier_cnt; + oldBarrierCnt = barrierCnt; stalledAtBarrier = false; } @@ -395,7 +395,7 @@ Wavefront::ready(itype_e type) } // Are there in pipe or outstanding memory requests? - if ((outstanding_reqs + mem_reqs_in_pipe) > 0) { + if ((outstandingReqs + memReqsInPipe) > 0) { return 0; } @@ -416,7 +416,7 @@ Wavefront::ready(itype_e type) } // Are there in pipe or outstanding memory requests? - if ((outstanding_reqs + mem_reqs_in_pipe) > 0) { + if ((outstandingReqs + memReqsInPipe) > 0) { return 0; } @@ -444,7 +444,7 @@ Wavefront::ready(itype_e type) // Here Global memory instruction if (IS_OT_READ_GM(ii->opType()) || IS_OT_ATOMIC_GM(ii->opType())) { // Are there in pipe or outstanding global memory write requests? - if ((outstanding_reqs_wr_gm + wr_gm_reqs_in_pipe) > 0) { + if ((outstandingReqsWrGm + wrGmReqsInPipe) > 0) { return 0; } } @@ -452,7 +452,7 @@ Wavefront::ready(itype_e type) if (IS_OT_WRITE_GM(ii->opType()) || IS_OT_ATOMIC_GM(ii->opType()) || IS_OT_HIST_GM(ii->opType())) { // Are there in pipe or outstanding global memory read requests? - if ((outstanding_reqs_rd_gm + rd_gm_reqs_in_pipe) > 0) + if ((outstandingReqsRdGm + rdGmReqsInPipe) > 0) return 0; } @@ -467,7 +467,7 @@ Wavefront::ready(itype_e type) } if (!computeUnit->globalMemoryPipe. - isGMReqFIFOWrRdy(rd_gm_reqs_in_pipe + wr_gm_reqs_in_pipe)) { + isGMReqFIFOWrRdy(rdGmReqsInPipe + wrGmReqsInPipe)) { // Can we insert a new request to the Global Mem Request FIFO? return 0; } @@ -484,14 +484,14 @@ Wavefront::ready(itype_e type) IS_OT_WRITE_LM(ii->opType()) || IS_OT_ATOMIC_LM(ii->opType()))) { // Here for Shared memory instruction if (IS_OT_READ_LM(ii->opType()) || IS_OT_ATOMIC_LM(ii->opType())) { - if ((outstanding_reqs_wr_lm + wr_lm_reqs_in_pipe) > 0) { + if ((outstandingReqsWrLm + wrLmReqsInPipe) > 0) { return 0; } } if (IS_OT_WRITE_LM(ii->opType()) || IS_OT_ATOMIC_LM(ii->opType()) || IS_OT_HIST_LM(ii->opType())) { - if ((outstanding_reqs_rd_lm + rd_lm_reqs_in_pipe) > 0) { + if ((outstandingReqsRdLm + rdLmReqsInPipe) > 0) { return 0; } } @@ -506,7 +506,7 @@ Wavefront::ready(itype_e type) } if (!computeUnit->localMemoryPipe. - isLMReqFIFOWrRdy(rd_lm_reqs_in_pipe + wr_lm_reqs_in_pipe)) { + isLMReqFIFOWrRdy(rdLmReqsInPipe + wrLmReqsInPipe)) { // Can we insert a new request to the LDS Request FIFO? return 0; } @@ -523,14 +523,14 @@ Wavefront::ready(itype_e type) IS_OT_WRITE_PM(ii->opType()) || IS_OT_ATOMIC_PM(ii->opType()))) { // Here for Private memory instruction ------------------------ // if (IS_OT_READ_PM(ii->opType()) || IS_OT_ATOMIC_PM(ii->opType())) { - if ((outstanding_reqs_wr_gm + wr_gm_reqs_in_pipe) > 0) { + if ((outstandingReqsWrGm + wrGmReqsInPipe) > 0) { return 0; } } if (IS_OT_WRITE_PM(ii->opType()) || IS_OT_ATOMIC_PM(ii->opType()) || IS_OT_HIST_PM(ii->opType())) { - if ((outstanding_reqs_rd_gm + rd_gm_reqs_in_pipe) > 0) { + if ((outstandingReqsRdGm + rdGmReqsInPipe) > 0) { return 0; } } @@ -546,7 +546,7 @@ Wavefront::ready(itype_e type) } if (!computeUnit->globalMemoryPipe. - isGMReqFIFOWrRdy(rd_gm_reqs_in_pipe + wr_gm_reqs_in_pipe)) { + isGMReqFIFOWrRdy(rdGmReqsInPipe + wrGmReqsInPipe)) { // Can we insert a new request to the Global Mem Request FIFO? return 0; } @@ -579,13 +579,13 @@ Wavefront::ready(itype_e type) return 0; } if (!computeUnit->globalMemoryPipe. - isGMReqFIFOWrRdy(rd_gm_reqs_in_pipe + wr_gm_reqs_in_pipe)) { + isGMReqFIFOWrRdy(rdGmReqsInPipe + wrGmReqsInPipe)) { // Can we insert a new request to the Global Mem Request FIFO? return 0; } if (!computeUnit->localMemoryPipe. - isLMReqFIFOWrRdy(rd_lm_reqs_in_pipe + wr_lm_reqs_in_pipe)) { + isLMReqFIFOWrRdy(rdLmReqsInPipe + wrLmReqsInPipe)) { // Can we insert a new request to the LDS Request FIFO? return 0; } @@ -636,8 +636,8 @@ Wavefront::updateResources() ticks(computeUnit->issuePeriod)); } else if (ii->opType() == Enums::OT_FLAT_READ) { assert(Enums::SC_NONE != ii->executedAs()); - mem_reqs_in_pipe++; - rd_gm_reqs_in_pipe++; + memReqsInPipe++; + rdGmReqsInPipe++; if ( Enums::SC_SHARED == ii->executedAs() ) { computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()]. preset(computeUnit->shader->ticks(4)); @@ -651,8 +651,8 @@ Wavefront::updateResources() } } else if (ii->opType() == Enums::OT_FLAT_WRITE) { assert(Enums::SC_NONE != ii->executedAs()); - mem_reqs_in_pipe++; - wr_gm_reqs_in_pipe++; + memReqsInPipe++; + wrGmReqsInPipe++; if (Enums::SC_SHARED == ii->executedAs()) { computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()]. preset(computeUnit->shader->ticks(8)); @@ -665,67 +665,67 @@ Wavefront::updateResources() preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); } } else if (IS_OT_READ_GM(ii->opType())) { - mem_reqs_in_pipe++; - rd_gm_reqs_in_pipe++; + memReqsInPipe++; + rdGmReqsInPipe++; computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. preset(computeUnit->shader->ticks(4)); computeUnit->wfWait[computeUnit->GlbMemUnitId()]. preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); } else if (IS_OT_WRITE_GM(ii->opType())) { - mem_reqs_in_pipe++; - wr_gm_reqs_in_pipe++; + memReqsInPipe++; + wrGmReqsInPipe++; computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. preset(computeUnit->shader->ticks(8)); computeUnit->wfWait[computeUnit->GlbMemUnitId()]. preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); } else if (IS_OT_ATOMIC_GM(ii->opType())) { - mem_reqs_in_pipe++; - wr_gm_reqs_in_pipe++; - rd_gm_reqs_in_pipe++; + memReqsInPipe++; + wrGmReqsInPipe++; + rdGmReqsInPipe++; computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. preset(computeUnit->shader->ticks(8)); computeUnit->wfWait[computeUnit->GlbMemUnitId()]. preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); } else if (IS_OT_READ_LM(ii->opType())) { - mem_reqs_in_pipe++; - rd_lm_reqs_in_pipe++; + memReqsInPipe++; + rdLmReqsInPipe++; computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()]. preset(computeUnit->shader->ticks(4)); computeUnit->wfWait[computeUnit->ShrMemUnitId()]. preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); } else if (IS_OT_WRITE_LM(ii->opType())) { - mem_reqs_in_pipe++; - wr_lm_reqs_in_pipe++; + memReqsInPipe++; + wrLmReqsInPipe++; computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()]. preset(computeUnit->shader->ticks(8)); computeUnit->wfWait[computeUnit->ShrMemUnitId()]. preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); } else if (IS_OT_ATOMIC_LM(ii->opType())) { - mem_reqs_in_pipe++; - wr_lm_reqs_in_pipe++; - rd_lm_reqs_in_pipe++; + memReqsInPipe++; + wrLmReqsInPipe++; + rdLmReqsInPipe++; computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()]. preset(computeUnit->shader->ticks(8)); computeUnit->wfWait[computeUnit->ShrMemUnitId()]. preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); } else if (IS_OT_READ_PM(ii->opType())) { - mem_reqs_in_pipe++; - rd_gm_reqs_in_pipe++; + memReqsInPipe++; + rdGmReqsInPipe++; computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. preset(computeUnit->shader->ticks(4)); computeUnit->wfWait[computeUnit->GlbMemUnitId()]. preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); } else if (IS_OT_WRITE_PM(ii->opType())) { - mem_reqs_in_pipe++; - wr_gm_reqs_in_pipe++; + memReqsInPipe++; + wrGmReqsInPipe++; computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. preset(computeUnit->shader->ticks(8)); computeUnit->wfWait[computeUnit->GlbMemUnitId()]. preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); } else if (IS_OT_ATOMIC_PM(ii->opType())) { - mem_reqs_in_pipe++; - wr_gm_reqs_in_pipe++; - rd_gm_reqs_in_pipe++; + memReqsInPipe++; + wrGmReqsInPipe++; + rdGmReqsInPipe++; computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. preset(computeUnit->shader->ticks(8)); computeUnit->wfWait[computeUnit->GlbMemUnitId()]. @@ -865,7 +865,7 @@ Wavefront::exec() bool Wavefront::waitingAtBarrier(int lane) { - return bar_cnt[lane] < max_bar_cnt; + return barCnt[lane] < maxBarCnt; } void diff --git a/src/gpu-compute/wavefront.hh b/src/gpu-compute/wavefront.hh index 5a5386a3d..db2e434a9 100644 --- a/src/gpu-compute/wavefront.hh +++ b/src/gpu-compute/wavefront.hh @@ -155,16 +155,16 @@ class Wavefront : public SimObject enum status_e {S_STOPPED,S_RETURNING,S_RUNNING}; // Base pointer for array of instruction pointers - uint64_t base_ptr; + uint64_t basePtr; - uint32_t old_barrier_cnt; - uint32_t barrier_cnt; - uint32_t barrier_id; - uint32_t barrier_slots; + uint32_t oldBarrierCnt; + uint32_t barrierCnt; + uint32_t barrierId; + uint32_t barrierSlots; status_e status; // HW slot id where the WF is mapped to inside a SIMD unit int wfSlotId; - int kern_id; + int kernId; // SIMD unit where the WV has been scheduled int simdId; // pointer to parent CU @@ -193,37 +193,37 @@ class Wavefront : public SimObject bool isOldestInstALU(); bool isOldestInstBarrier(); // used for passing spill address to DDInstGPU - std::vector<Addr> last_addr; - std::vector<uint32_t> workitemid[3]; - std::vector<uint32_t> workitemFlatId; - uint32_t workgroupid[3]; - uint32_t workgroupsz[3]; - uint32_t gridsz[3]; - uint32_t wg_id; - uint32_t wg_sz; - uint32_t dynwaveid; - uint32_t maxdynwaveid; - uint32_t dispatchid; + std::vector<Addr> lastAddr; + std::vector<uint32_t> workItemId[3]; + std::vector<uint32_t> workItemFlatId; + uint32_t workGroupId[3]; + uint32_t workGroupSz[3]; + uint32_t gridSz[3]; + uint32_t wgId; + uint32_t wgSz; + uint32_t dynWaveId; + uint32_t maxDynWaveId; + uint32_t dispatchId; // outstanding global+local memory requests - uint32_t outstanding_reqs; + uint32_t outstandingReqs; // memory requests between scoreboard // and execute stage not yet executed - uint32_t mem_reqs_in_pipe; + uint32_t memReqsInPipe; // outstanding global memory write requests - uint32_t outstanding_reqs_wr_gm; + uint32_t outstandingReqsWrGm; // outstanding local memory write requests - uint32_t outstanding_reqs_wr_lm; + uint32_t outstandingReqsWrLm; // outstanding global memory read requests - uint32_t outstanding_reqs_rd_gm; + uint32_t outstandingReqsRdGm; // outstanding local memory read requests - uint32_t outstanding_reqs_rd_lm; - uint32_t rd_lm_reqs_in_pipe; - uint32_t rd_gm_reqs_in_pipe; - uint32_t wr_lm_reqs_in_pipe; - uint32_t wr_gm_reqs_in_pipe; - - int mem_trace_busy; - uint64_t last_trace; + uint32_t outstandingReqsRdLm; + uint32_t rdLmReqsInPipe; + uint32_t rdGmReqsInPipe; + uint32_t wrLmReqsInPipe; + uint32_t wrGmReqsInPipe; + + int memTraceBusy; + uint64_t lastTrace; // number of vector registers reserved by WF int reservedVectorRegs; // Index into the Vector Register File's namespace where the WF's registers @@ -231,25 +231,25 @@ class Wavefront : public SimObject uint32_t startVgprIndex; // Old value of destination gpr (for trace) - std::vector<uint32_t> old_vgpr; + std::vector<uint32_t> oldVgpr; // Id of destination gpr (for trace) - uint32_t old_vgpr_id; + uint32_t oldVgprId; // Tick count of last old_vgpr copy - uint64_t old_vgpr_tcnt; + uint64_t oldVgprTcnt; // Old value of destination gpr (for trace) - std::vector<uint64_t> old_dgpr; + std::vector<uint64_t> oldDgpr; // Id of destination gpr (for trace) - uint32_t old_dgpr_id; + uint32_t oldDgprId; // Tick count of last old_vgpr copy - uint64_t old_dgpr_tcnt; + uint64_t oldDgprTcnt; // Execution mask at wavefront start - VectorMask init_mask; + VectorMask initMask; // number of barriers this WF has joined - std::vector<int> bar_cnt; - int max_bar_cnt; + std::vector<int> barCnt; + int maxBarCnt; // Flag to stall a wave on barrier bool stalledAtBarrier; @@ -333,7 +333,7 @@ class Wavefront : public SimObject int ready(itype_e type); bool instructionBufferHasBranch(); void regStats(); - VectorMask get_pred() { return execMask() & init_mask; } + VectorMask getPred() { return execMask() & initMask; } bool waitingAtBarrier(int lane); |