sanchayanmaity/gem5
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

gpu-compute: Wavefront refactoring

Browse source 
Renaming members of the Wavefront class in accordance with the style guide.
This commit is contained in:
Alexandru Dutu 2016-09-16 12:26:52 -04:00
parent e9fe1b838b
commit 589e13a23b
12 changed files with 259 additions and 259 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

30
src/arch/hsail/gen.py
View file

@ -233,7 +233,7 @@ $class_name::execute(GPUDynInstPtr gpuDynInst)
typedef Base::DestCType DestCType; typedef Base::DestCType DestCType;
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
@ -254,7 +254,7 @@ $class_name::execute(GPUDynInstPtr gpuDynInst)
typedef Base::DestCType DestCType; typedef Base::DestCType DestCType;
typedef Base::SrcCType SrcCType; typedef Base::SrcCType SrcCType;
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
@ -275,7 +275,7 @@ $class_name<DataType>::execute(GPUDynInstPtr gpuDynInst)
{ {
Wavefront *w = gpuDynInst->wavefront(); Wavefront *w = gpuDynInst->wavefront();
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
@ -310,7 +310,7 @@ $class_name<DataType>::execute(GPUDynInstPtr gpuDynInst)
typedef typename Base::Src1CType Src1T; typedef typename Base::Src1CType Src1T;
typedef typename Base::Src2CType Src2T; typedef typename Base::Src2CType Src2T;
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
@ -344,7 +344,7 @@ $class_name<DataType>::execute(GPUDynInstPtr gpuDynInst)
typedef CType Src0T; typedef CType Src0T;
typedef typename Base::Src1CType Src1T; typedef typename Base::Src1CType Src1T;
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
@ -371,7 +371,7 @@ $class_name<DataType>::execute(GPUDynInstPtr gpuDynInst)
{ {
Wavefront *w = gpuDynInst->wavefront(); Wavefront *w = gpuDynInst->wavefront();
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
CType dest_val; CType dest_val;
@ -399,7 +399,7 @@ $class_name<DestDataType, SrcDataType>::execute(GPUDynInstPtr gpuDynInst)
{ {
Wavefront *w = gpuDynInst->wavefront(); Wavefront *w = gpuDynInst->wavefront();
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
@ -745,17 +745,17 @@ def gen_special(brig_opcode, expr, dest_type='U32'):
gen(brig_opcode, None, expr, base_class) gen(brig_opcode, None, expr, base_class)
gen_special('WorkItemId', 'w->workitemid[src0][lane]') gen_special('WorkItemId', 'w->workItemId[src0][lane]')
gen_special('WorkItemAbsId', gen_special('WorkItemAbsId',
'w->workitemid[src0][lane] + (w->workgroupid[src0] * w->workgroupsz[src0])') 'w->workItemId[src0][lane] + (w->workGroupId[src0] * w->workGroupSz[src0])')
gen_special('WorkGroupId', 'w->workgroupid[src0]') gen_special('WorkGroupId', 'w->workGroupId[src0]')
gen_special('WorkGroupSize', 'w->workgroupsz[src0]') gen_special('WorkGroupSize', 'w->workGroupSz[src0]')
gen_special('CurrentWorkGroupSize', 'w->workgroupsz[src0]') gen_special('CurrentWorkGroupSize', 'w->workGroupSz[src0]')
gen_special('GridSize', 'w->gridsz[src0]') gen_special('GridSize', 'w->gridSz[src0]')
gen_special('GridGroups', gen_special('GridGroups',
'divCeil(w->gridsz[src0],w->workgroupsz[src0])') 'divCeil(w->gridSz[src0],w->workGroupSz[src0])')
gen_special('LaneId', 'lane') gen_special('LaneId', 'lane')
gen_special('WaveId', 'w->dynwaveid') gen_special('WaveId', 'w->dynWaveId')
gen_special('Clock', 'w->computeUnit->shader->tick_cnt', 'U64') gen_special('Clock', 'w->computeUnit->shader->tick_cnt', 'U64')
# gen_special('CU'', ') # gen_special('CU'', ')

10
src/arch/hsail/insts/decl.hh
View file

@ -960,7 +960,7 @@ namespace HsailISA
gpuDynInst->simdId = w->simdId; gpuDynInst->simdId = w->simdId;
gpuDynInst->wfSlotId = w->wfSlotId; gpuDynInst->wfSlotId = w->wfSlotId;
gpuDynInst->wfDynId = w->wfDynId; gpuDynInst->wfDynId = w->wfDynId;
gpuDynInst->kern_id = w->kern_id; gpuDynInst->kern_id = w->kernId;
gpuDynInst->cu_id = w->computeUnit->cu_id; gpuDynInst->cu_id = w->computeUnit->cu_id;
gpuDynInst->memoryOrder = gpuDynInst->memoryOrder =
@ -971,10 +971,10 @@ namespace HsailISA
GlobalMemPipeline* gmp = &(w->computeUnit->globalMemoryPipe); GlobalMemPipeline* gmp = &(w->computeUnit->globalMemoryPipe);
gmp->getGMReqFIFO().push(gpuDynInst); gmp->getGMReqFIFO().push(gpuDynInst);
w->wr_gm_reqs_in_pipe--; w->wrGmReqsInPipe--;
w->rd_gm_reqs_in_pipe--; w->rdGmReqsInPipe--;
w->mem_reqs_in_pipe--; w->memReqsInPipe--;
w->outstanding_reqs++; w->outstandingReqs++;
} else if (o_type == Enums::OT_SHARED_MEMFENCE) { } else if (o_type == Enums::OT_SHARED_MEMFENCE) {
// no-op // no-op
} else { } else {

14
src/arch/hsail/insts/main.cc
View file

@ -131,12 +131,12 @@ namespace HsailISA
{ {
Wavefront *w = gpuDynInst->wavefront(); Wavefront *w = gpuDynInst->wavefront();
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
// mask off completed work-items // mask off completed work-items
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
w->init_mask[lane] = 0; w->initMask[lane] = 0;
} }
} }
@ -149,14 +149,14 @@ namespace HsailISA
} }
// if all work-items have completed, then wave-front is done // if all work-items have completed, then wave-front is done
if (w->init_mask.none()) { if (w->initMask.none()) {
w->status = Wavefront::S_STOPPED; w->status = Wavefront::S_STOPPED;
int32_t refCount = w->computeUnit->getLds(). int32_t refCount = w->computeUnit->getLds().
decreaseRefCounter(w->dispatchid, w->wg_id); decreaseRefCounter(w->dispatchId, w->wgId);
DPRINTF(GPUExec, "CU%d: decrease ref ctr WG[%d] to [%d]\n", DPRINTF(GPUExec, "CU%d: decrease ref ctr WG[%d] to [%d]\n",
w->computeUnit->cu_id, w->wg_id, refCount); w->computeUnit->cu_id, w->wgId, refCount);
// free the vector registers of the completed wavefront // free the vector registers of the completed wavefront
w->computeUnit->vectorRegsReserved[w->simdId] -= w->computeUnit->vectorRegsReserved[w->simdId] -=
@ -201,8 +201,8 @@ namespace HsailISA
{ {
Wavefront *w = gpuDynInst->wavefront(); Wavefront *w = gpuDynInst->wavefront();
assert(w->barrier_cnt == w->old_barrier_cnt); assert(w->barrierCnt == w->oldBarrierCnt);
w->barrier_cnt = w->old_barrier_cnt + 1; w->barrierCnt = w->oldBarrierCnt + 1;
w->stalledAtBarrier = true; w->stalledAtBarrier = true;
} }
} // namespace HsailISA } // namespace HsailISA

78
src/arch/hsail/insts/mem_impl.hh
View file

@ -59,7 +59,7 @@ namespace HsailISA
Wavefront *w = gpuDynInst->wavefront(); Wavefront *w = gpuDynInst->wavefront();
typedef typename DestDataType::CType CType M5_VAR_USED; typedef typename DestDataType::CType CType M5_VAR_USED;
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
std::vector<Addr> addr_vec; std::vector<Addr> addr_vec;
addr_vec.resize(w->computeUnit->wfSize(), (Addr)0); addr_vec.resize(w->computeUnit->wfSize(), (Addr)0);
this->addr.calcVector(w, addr_vec); this->addr.calcVector(w, addr_vec);
@ -159,7 +159,7 @@ namespace HsailISA
Wavefront *w = gpuDynInst->wavefront(); Wavefront *w = gpuDynInst->wavefront();
typedef typename MemDataType::CType MemCType; typedef typename MemDataType::CType MemCType;
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
// Kernarg references are handled uniquely for now (no Memory Request // Kernarg references are handled uniquely for now (no Memory Request
// is used), so special-case them up front. Someday we should // is used), so special-case them up front. Someday we should
@ -230,7 +230,7 @@ namespace HsailISA
m->simdId = w->simdId; m->simdId = w->simdId;
m->wfSlotId = w->wfSlotId; m->wfSlotId = w->wfSlotId;
m->wfDynId = w->wfDynId; m->wfDynId = w->wfDynId;
m->kern_id = w->kern_id; m->kern_id = w->kernId;
m->cu_id = w->computeUnit->cu_id; m->cu_id = w->computeUnit->cu_id;
m->latency.init(&w->computeUnit->shader->tick_cnt); m->latency.init(&w->computeUnit->shader->tick_cnt);
@ -261,8 +261,8 @@ namespace HsailISA
} }
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_rd_gm++; w->outstandingReqsRdGm++;
w->rd_gm_reqs_in_pipe--; w->rdGmReqsInPipe--;
break; break;
case Brig::BRIG_SEGMENT_SPILL: case Brig::BRIG_SEGMENT_SPILL:
@ -281,14 +281,14 @@ namespace HsailISA
m->addr[lane] = m->addr[lane] * w->spillWidth + m->addr[lane] = m->addr[lane] * w->spillWidth +
lane * sizeof(MemCType) + w->spillBase; lane * sizeof(MemCType) + w->spillBase;
w->last_addr[lane] = m->addr[lane]; w->lastAddr[lane] = m->addr[lane];
} }
} }
} }
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_rd_gm++; w->outstandingReqsRdGm++;
w->rd_gm_reqs_in_pipe--; w->rdGmReqsInPipe--;
break; break;
case Brig::BRIG_SEGMENT_GROUP: case Brig::BRIG_SEGMENT_GROUP:
@ -296,8 +296,8 @@ namespace HsailISA
m->pipeId = LDSMEM_PIPE; m->pipeId = LDSMEM_PIPE;
m->latency.set(w->computeUnit->shader->ticks(24)); m->latency.set(w->computeUnit->shader->ticks(24));
w->computeUnit->localMemoryPipe.getLMReqFIFO().push(m); w->computeUnit->localMemoryPipe.getLMReqFIFO().push(m);
w->outstanding_reqs_rd_lm++; w->outstandingReqsRdLm++;
w->rd_lm_reqs_in_pipe--; w->rdLmReqsInPipe--;
break; break;
case Brig::BRIG_SEGMENT_READONLY: case Brig::BRIG_SEGMENT_READONLY:
@ -313,8 +313,8 @@ namespace HsailISA
} }
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_rd_gm++; w->outstandingReqsRdGm++;
w->rd_gm_reqs_in_pipe--; w->rdGmReqsInPipe--;
break; break;
case Brig::BRIG_SEGMENT_PRIVATE: case Brig::BRIG_SEGMENT_PRIVATE:
@ -332,8 +332,8 @@ namespace HsailISA
} }
} }
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_rd_gm++; w->outstandingReqsRdGm++;
w->rd_gm_reqs_in_pipe--; w->rdGmReqsInPipe--;
break; break;
default: default:
@ -341,8 +341,8 @@ namespace HsailISA
m->addr[0]); m->addr[0]);
} }
w->outstanding_reqs++; w->outstandingReqs++;
w->mem_reqs_in_pipe--; w->memReqsInPipe--;
} }
template<typename OperationType, typename SrcDataType, template<typename OperationType, typename SrcDataType,
@ -355,7 +355,7 @@ namespace HsailISA
typedef typename OperationType::CType CType; typedef typename OperationType::CType CType;
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
// arg references are handled uniquely for now (no Memory Request // arg references are handled uniquely for now (no Memory Request
// is used), so special-case them up front. Someday we should // is used), so special-case them up front. Someday we should
@ -419,7 +419,7 @@ namespace HsailISA
m->simdId = w->simdId; m->simdId = w->simdId;
m->wfSlotId = w->wfSlotId; m->wfSlotId = w->wfSlotId;
m->wfDynId = w->wfDynId; m->wfDynId = w->wfDynId;
m->kern_id = w->kern_id; m->kern_id = w->kernId;
m->cu_id = w->computeUnit->cu_id; m->cu_id = w->computeUnit->cu_id;
m->latency.init(&w->computeUnit->shader->tick_cnt); m->latency.init(&w->computeUnit->shader->tick_cnt);
@ -448,8 +448,8 @@ namespace HsailISA
} }
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_wr_gm++; w->outstandingReqsWrGm++;
w->wr_gm_reqs_in_pipe--; w->wrGmReqsInPipe--;
break; break;
case Brig::BRIG_SEGMENT_SPILL: case Brig::BRIG_SEGMENT_SPILL:
@ -469,8 +469,8 @@ namespace HsailISA
} }
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_wr_gm++; w->outstandingReqsWrGm++;
w->wr_gm_reqs_in_pipe--; w->wrGmReqsInPipe--;
break; break;
case Brig::BRIG_SEGMENT_GROUP: case Brig::BRIG_SEGMENT_GROUP:
@ -478,8 +478,8 @@ namespace HsailISA
m->pipeId = LDSMEM_PIPE; m->pipeId = LDSMEM_PIPE;
m->latency.set(w->computeUnit->shader->ticks(24)); m->latency.set(w->computeUnit->shader->ticks(24));
w->computeUnit->localMemoryPipe.getLMReqFIFO().push(m); w->computeUnit->localMemoryPipe.getLMReqFIFO().push(m);
w->outstanding_reqs_wr_lm++; w->outstandingReqsWrLm++;
w->wr_lm_reqs_in_pipe--; w->wrLmReqsInPipe--;
break; break;
case Brig::BRIG_SEGMENT_PRIVATE: case Brig::BRIG_SEGMENT_PRIVATE:
@ -497,16 +497,16 @@ namespace HsailISA
} }
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_wr_gm++; w->outstandingReqsWrGm++;
w->wr_gm_reqs_in_pipe--; w->wrGmReqsInPipe--;
break; break;
default: default:
fatal("Store to unsupported segment %d\n", this->segment); fatal("Store to unsupported segment %d\n", this->segment);
} }
w->outstanding_reqs++; w->outstandingReqs++;
w->mem_reqs_in_pipe--; w->memReqsInPipe--;
} }
template<typename OperationType, typename SrcDataType, template<typename OperationType, typename SrcDataType,
@ -596,7 +596,7 @@ namespace HsailISA
m->simdId = w->simdId; m->simdId = w->simdId;
m->wfSlotId = w->wfSlotId; m->wfSlotId = w->wfSlotId;
m->wfDynId = w->wfDynId; m->wfDynId = w->wfDynId;
m->kern_id = w->kern_id; m->kern_id = w->kernId;
m->cu_id = w->computeUnit->cu_id; m->cu_id = w->computeUnit->cu_id;
m->latency.init(&w->computeUnit->shader->tick_cnt); m->latency.init(&w->computeUnit->shader->tick_cnt);
@ -607,10 +607,10 @@ namespace HsailISA
m->pipeId = GLBMEM_PIPE; m->pipeId = GLBMEM_PIPE;
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_wr_gm++; w->outstandingReqsWrGm++;
w->wr_gm_reqs_in_pipe--; w->wrGmReqsInPipe--;
w->outstanding_reqs_rd_gm++; w->outstandingReqsRdGm++;
w->rd_gm_reqs_in_pipe--; w->rdGmReqsInPipe--;
break; break;
case Brig::BRIG_SEGMENT_GROUP: case Brig::BRIG_SEGMENT_GROUP:
@ -618,10 +618,10 @@ namespace HsailISA
m->pipeId = LDSMEM_PIPE; m->pipeId = LDSMEM_PIPE;
m->latency.set(w->computeUnit->shader->ticks(24)); m->latency.set(w->computeUnit->shader->ticks(24));
w->computeUnit->localMemoryPipe.getLMReqFIFO().push(m); w->computeUnit->localMemoryPipe.getLMReqFIFO().push(m);
w->outstanding_reqs_wr_lm++; w->outstandingReqsWrLm++;
w->wr_lm_reqs_in_pipe--; w->wrLmReqsInPipe--;
w->outstanding_reqs_rd_lm++; w->outstandingReqsRdLm++;
w->rd_lm_reqs_in_pipe--; w->rdLmReqsInPipe--;
break; break;
default: default:
@ -629,8 +629,8 @@ namespace HsailISA
this->segment); this->segment);
} }
w->outstanding_reqs++; w->outstandingReqs++;
w->mem_reqs_in_pipe--; w->memReqsInPipe--;
} }
const char* atomicOpToString(Brig::BrigAtomicOperation atomicOp); const char* atomicOpToString(Brig::BrigAtomicOperation atomicOp);

90
src/arch/hsail/insts/pseudo_inst.cc
View file

@ -79,7 +79,7 @@ namespace HsailISA
void void
Call::execPseudoInst(Wavefront *w, GPUDynInstPtr gpuDynInst) Call::execPseudoInst(Wavefront *w, GPUDynInstPtr gpuDynInst)
{ {
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
int op = 0; int op = 0;
bool got_op = false; bool got_op = false;
@ -181,7 +181,7 @@ namespace HsailISA
Call::MagicPrintLane(Wavefront *w) Call::MagicPrintLane(Wavefront *w)
{ {
#if TRACING_ON #if TRACING_ON
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
int src_val1 = src1.get<int>(w, lane, 1); int src_val1 = src1.get<int>(w, lane, 1);
@ -204,7 +204,7 @@ namespace HsailISA
Call::MagicPrintLane64(Wavefront *w) Call::MagicPrintLane64(Wavefront *w)
{ {
#if TRACING_ON #if TRACING_ON
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
int64_t src_val1 = src1.get<int64_t>(w, lane, 1); int64_t src_val1 = src1.get<int64_t>(w, lane, 1);
@ -227,7 +227,7 @@ namespace HsailISA
Call::MagicPrintWF32(Wavefront *w) Call::MagicPrintWF32(Wavefront *w)
{ {
#if TRACING_ON #if TRACING_ON
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
std::string res_str; std::string res_str;
res_str = csprintf("krl_prt (%s)\n", disassemble()); res_str = csprintf("krl_prt (%s)\n", disassemble());
@ -265,7 +265,7 @@ namespace HsailISA
Call::MagicPrintWF32ID(Wavefront *w) Call::MagicPrintWF32ID(Wavefront *w)
{ {
#if TRACING_ON #if TRACING_ON
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
std::string res_str; std::string res_str;
int src_val3 = -1; int src_val3 = -1;
res_str = csprintf("krl_prt (%s)\n", disassemble()); res_str = csprintf("krl_prt (%s)\n", disassemble());
@ -307,7 +307,7 @@ namespace HsailISA
Call::MagicPrintWF64(Wavefront *w) Call::MagicPrintWF64(Wavefront *w)
{ {
#if TRACING_ON #if TRACING_ON
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
std::string res_str; std::string res_str;
res_str = csprintf("krl_prt (%s)\n", disassemble()); res_str = csprintf("krl_prt (%s)\n", disassemble());
@ -345,7 +345,7 @@ namespace HsailISA
Call::MagicPrintWFID64(Wavefront *w) Call::MagicPrintWFID64(Wavefront *w)
{ {
#if TRACING_ON #if TRACING_ON
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
std::string res_str; std::string res_str;
int src_val3 = -1; int src_val3 = -1;
res_str = csprintf("krl_prt (%s)\n", disassemble()); res_str = csprintf("krl_prt (%s)\n", disassemble());
@ -387,7 +387,7 @@ namespace HsailISA
Call::MagicPrintWFFloat(Wavefront *w) Call::MagicPrintWFFloat(Wavefront *w)
{ {
#if TRACING_ON #if TRACING_ON
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
std::string res_str; std::string res_str;
res_str = csprintf("krl_prt (%s)\n", disassemble()); res_str = csprintf("krl_prt (%s)\n", disassemble());
@ -425,7 +425,7 @@ namespace HsailISA
res_str = csprintf("Breakpoint encountered for wavefront %i\n", res_str = csprintf("Breakpoint encountered for wavefront %i\n",
w->wfSlotId); w->wfSlotId);
res_str += csprintf(" Kern ID: %i\n", w->kern_id); res_str += csprintf(" Kern ID: %i\n", w->kernId);
res_str += csprintf(" Phase ID: %i\n", w->simdId); res_str += csprintf(" Phase ID: %i\n", w->simdId);
res_str += csprintf(" Executing on CU #%i\n", w->computeUnit->cu_id); res_str += csprintf(" Executing on CU #%i\n", w->computeUnit->cu_id);
res_str += csprintf(" Exec mask: "); res_str += csprintf(" Exec mask: ");
@ -455,7 +455,7 @@ namespace HsailISA
void void
Call::MagicPrefixSum(Wavefront *w) Call::MagicPrefixSum(Wavefront *w)
{ {
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
int res = 0; int res = 0;
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
@ -474,7 +474,7 @@ namespace HsailISA
// The reduction instruction takes up to 64 inputs (one from // The reduction instruction takes up to 64 inputs (one from
// each thread in a WF) and sums them. It returns the sum to // each thread in a WF) and sums them. It returns the sum to
// each thread in the WF. // each thread in the WF.
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
int res = 0; int res = 0;
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
@ -494,7 +494,7 @@ namespace HsailISA
void void
Call::MagicMaskLower(Wavefront *w) Call::MagicMaskLower(Wavefront *w)
{ {
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
int res = 0; int res = 0;
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
@ -519,7 +519,7 @@ namespace HsailISA
void void
Call::MagicMaskUpper(Wavefront *w) Call::MagicMaskUpper(Wavefront *w)
{ {
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
int res = 0; int res = 0;
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
@ -544,42 +544,42 @@ namespace HsailISA
void void
Call::MagicJoinWFBar(Wavefront *w) Call::MagicJoinWFBar(Wavefront *w)
{ {
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
int max_cnt = 0; int max_cnt = 0;
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
w->bar_cnt[lane]++; w->barCnt[lane]++;
if (w->bar_cnt[lane] > max_cnt) { if (w->barCnt[lane] > max_cnt) {
max_cnt = w->bar_cnt[lane]; max_cnt = w->barCnt[lane];
} }
} }
} }
if (max_cnt > w->max_bar_cnt) { if (max_cnt > w->maxBarCnt) {
w->max_bar_cnt = max_cnt; w->maxBarCnt = max_cnt;
} }
} }
void void
Call::MagicWaitWFBar(Wavefront *w) Call::MagicWaitWFBar(Wavefront *w)
{ {
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
int max_cnt = 0; int max_cnt = 0;
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
w->bar_cnt[lane]--; w->barCnt[lane]--;
} }
if (w->bar_cnt[lane] > max_cnt) { if (w->barCnt[lane] > max_cnt) {
max_cnt = w->bar_cnt[lane]; max_cnt = w->barCnt[lane];
} }
} }
if (max_cnt < w->max_bar_cnt) { if (max_cnt < w->maxBarCnt) {
w->max_bar_cnt = max_cnt; w->maxBarCnt = max_cnt;
} }
w->instructionBuffer.erase(w->instructionBuffer.begin() + 1, w->instructionBuffer.erase(w->instructionBuffer.begin() + 1,
@ -591,7 +591,7 @@ namespace HsailISA
void void
Call::MagicPanic(Wavefront *w) Call::MagicPanic(Wavefront *w)
{ {
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
if (mask[lane]) { if (mask[lane]) {
@ -648,12 +648,12 @@ namespace HsailISA
m->pipeId = GLBMEM_PIPE; m->pipeId = GLBMEM_PIPE;
m->latency.set(w->computeUnit->shader->ticks(64)); m->latency.set(w->computeUnit->shader->ticks(64));
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_wr_gm++; w->outstandingReqsWrGm++;
w->wr_gm_reqs_in_pipe--; w->wrGmReqsInPipe--;
w->outstanding_reqs_rd_gm++; w->outstandingReqsRdGm++;
w->rd_gm_reqs_in_pipe--; w->rdGmReqsInPipe--;
w->outstanding_reqs++; w->outstandingReqs++;
w->mem_reqs_in_pipe--; w->memReqsInPipe--;
} }
void void
@ -687,12 +687,12 @@ namespace HsailISA
m->pipeId = GLBMEM_PIPE; m->pipeId = GLBMEM_PIPE;
m->latency.set(w->computeUnit->shader->ticks(64)); m->latency.set(w->computeUnit->shader->ticks(64));
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_wr_gm++; w->outstandingReqsWrGm++;
w->wr_gm_reqs_in_pipe--; w->wrGmReqsInPipe--;
w->outstanding_reqs_rd_gm++; w->outstandingReqsRdGm++;
w->rd_gm_reqs_in_pipe--; w->rdGmReqsInPipe--;
w->outstanding_reqs++; w->outstandingReqs++;
w->mem_reqs_in_pipe--; w->memReqsInPipe--;
} }
void void
@ -725,16 +725,16 @@ namespace HsailISA
m->pipeId = GLBMEM_PIPE; m->pipeId = GLBMEM_PIPE;
m->latency.set(w->computeUnit->shader->ticks(1)); m->latency.set(w->computeUnit->shader->ticks(1));
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m); w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->outstanding_reqs_rd_gm++; w->outstandingReqsRdGm++;
w->rd_gm_reqs_in_pipe--; w->rdGmReqsInPipe--;
w->outstanding_reqs++; w->outstandingReqs++;
w->mem_reqs_in_pipe--; w->memReqsInPipe--;
} }
void void
Call::MagicXactCasLd(Wavefront *w) Call::MagicXactCasLd(Wavefront *w)
{ {
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
int src_val1 = 0; int src_val1 = 0;
for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) { for (int lane = 0; lane < w->computeUnit->wfSize(); ++lane) {
@ -756,7 +756,7 @@ namespace HsailISA
void void
Call::MagicMostSigThread(Wavefront *w) Call::MagicMostSigThread(Wavefront *w)
{ {
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
unsigned mst = true; unsigned mst = true;
for (int lane = w->computeUnit->wfSize() - 1; lane >= 0; --lane) { for (int lane = w->computeUnit->wfSize() - 1; lane >= 0; --lane) {
@ -770,7 +770,7 @@ namespace HsailISA
void void
Call::MagicMostSigBroadcast(Wavefront *w) Call::MagicMostSigBroadcast(Wavefront *w)
{ {
const VectorMask &mask = w->get_pred(); const VectorMask &mask = w->getPred();
int res = 0; int res = 0;
bool got_res = false; bool got_res = false;

72
src/gpu-compute/compute_unit.cc
View file

@ -178,13 +178,13 @@ ComputeUnit::FillKernelState(Wavefront *w, NDRange *ndr)
{ {
w->resizeRegFiles(ndr->q.cRegCount, ndr->q.sRegCount, ndr->q.dRegCount); w->resizeRegFiles(ndr->q.cRegCount, ndr->q.sRegCount, ndr->q.dRegCount);
w->workgroupsz[0] = ndr->q.wgSize[0]; w->workGroupSz[0] = ndr->q.wgSize[0];
w->workgroupsz[1] = ndr->q.wgSize[1]; w->workGroupSz[1] = ndr->q.wgSize[1];
w->workgroupsz[2] = ndr->q.wgSize[2]; w->workGroupSz[2] = ndr->q.wgSize[2];
w->wg_sz = w->workgroupsz[0] * w->workgroupsz[1] * w->workgroupsz[2]; w->wgSz = w->workGroupSz[0] * w->workGroupSz[1] * w->workGroupSz[2];
w->gridsz[0] = ndr->q.gdSize[0]; w->gridSz[0] = ndr->q.gdSize[0];
w->gridsz[1] = ndr->q.gdSize[1]; w->gridSz[1] = ndr->q.gdSize[1];
w->gridsz[2] = ndr->q.gdSize[2]; w->gridSz[2] = ndr->q.gdSize[2];
w->kernelArgs = ndr->q.args; w->kernelArgs = ndr->q.args;
w->privSizePerItem = ndr->q.privMemPerItem; w->privSizePerItem = ndr->q.privMemPerItem;
w->spillSizePerItem = ndr->q.spillMemPerItem; w->spillSizePerItem = ndr->q.spillMemPerItem;
@ -236,29 +236,29 @@ ComputeUnit::StartWF(Wavefront *w, int trueWgSize[], int trueWgSizeTotal,
init_mask[k] = 1; init_mask[k] = 1;
} }
w->kern_id = ndr->dispatchId; w->kernId = ndr->dispatchId;
w->dynwaveid = cnt; w->dynWaveId = cnt;
w->init_mask = init_mask.to_ullong(); w->initMask = init_mask.to_ullong();
for (int k = 0; k < wfSize(); ++k) { for (int k = 0; k < wfSize(); ++k) {
w->workitemid[0][k] = (k+cnt*wfSize()) % trueWgSize[0]; w->workItemId[0][k] = (k+cnt*wfSize()) % trueWgSize[0];
w->workitemid[1][k] = w->workItemId[1][k] =
((k + cnt * wfSize()) / trueWgSize[0]) % trueWgSize[1]; ((k + cnt * wfSize()) / trueWgSize[0]) % trueWgSize[1];
w->workitemid[2][k] = w->workItemId[2][k] =
(k + cnt * wfSize()) / (trueWgSize[0] * trueWgSize[1]); (k + cnt * wfSize()) / (trueWgSize[0] * trueWgSize[1]);
w->workitemFlatId[k] = w->workitemid[2][k] * trueWgSize[0] * w->workItemFlatId[k] = w->workItemId[2][k] * trueWgSize[0] *
trueWgSize[1] + w->workitemid[1][k] * trueWgSize[0] + trueWgSize[1] + w->workItemId[1][k] * trueWgSize[0] +
w->workitemid[0][k]; 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->maxBarCnt = 0;
w->old_barrier_cnt = 0; w->oldBarrierCnt = 0;
w->barrier_cnt = 0; w->barrierCnt = 0;
w->privBase = ndr->q.privMemStart; w->privBase = ndr->q.privMemStart;
ndr->q.privMemStart += ndr->q.privMemPerItem * wfSize(); 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()); w->pushToReconvergenceStack(0, UINT32_MAX, init_mask.to_ulong());
// WG state // WG state
w->wg_id = ndr->globalWgId; w->wgId = ndr->globalWgId;
w->dispatchid = ndr->dispatchId; w->dispatchId = ndr->dispatchId;
w->workgroupid[0] = w->wg_id % ndr->numWg[0]; w->workGroupId[0] = w->wgId % ndr->numWg[0];
w->workgroupid[1] = (w->wg_id / ndr->numWg[0]) % ndr->numWg[1]; w->workGroupId[1] = (w->wgId / ndr->numWg[0]) % ndr->numWg[1];
w->workgroupid[2] = w->wg_id / (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; w->stalledAtBarrier = false;
// set the wavefront context to have a pointer to this section of the LDS // set the wavefront context to have a pointer to this section of the LDS
w->ldsChunk = ldsChunk; w->ldsChunk = ldsChunk;
int32_t refCount M5_VAR_USED = 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", 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(); 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, "Checking WF[%d][%d]\n", i_simd, i_wf);
DPRINTF(GPUSync, "wf->barrier_id = %d, _barrier_id = %d\n", 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", DPRINTF(GPUSync, "wf->barrier_cnt %d, bcnt = %d\n",
w->barrier_cnt, bcnt); w->barrierCnt, bcnt);
} }
if (w->status == Wavefront::S_RUNNING && if (w->status == Wavefront::S_RUNNING &&
w->barrier_id == _barrier_id && w->barrier_cnt == bcnt && w->barrierId == _barrier_id && w->barrierCnt == bcnt &&
!w->outstanding_reqs) { !w->outstandingReqs) {
++ccnt; ++ccnt;
DPRINTF(GPUSync, "WF[%d][%d] at barrier, increment ccnt to " 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) { if (w->status == Wavefront::S_RETURNING) {
DPRINTF(GPUDisp, "CU%d: WF[%d][%d][wv=%d]: WG id completed %d\n", DPRINTF(GPUDisp, "CU%d: WF[%d][%d][wv=%d]: WG id completed %d\n",
computeUnit->cu_id, w->simdId, w->wfSlotId, computeUnit->cu_id, w->simdId, w->wfSlotId,
w->wfDynId, w->kern_id); w->wfDynId, w->kernId);
computeUnit->shader->dispatcher->notifyWgCompl(w); computeUnit->shader->dispatcher->notifyWgCompl(w);
w->status = Wavefront::S_STOPPED; w->status = Wavefront::S_STOPPED;
} else { } else {
w->outstanding_reqs--; w->outstandingReqs--;
} }
DPRINTF(GPUSync, "CU%d: WF[%d][%d]: barrier_cnt = %d\n", DPRINTF(GPUSync, "CU%d: WF[%d][%d]: barrier_cnt = %d\n",
computeUnit->cu_id, gpuDynInst->simdId, computeUnit->cu_id, gpuDynInst->simdId,
gpuDynInst->wfSlotId, w->barrier_cnt); gpuDynInst->wfSlotId, w->barrierCnt);
if (gpuDynInst->useContinuation) { if (gpuDynInst->useContinuation) {
assert(gpuDynInst->scope != Enums::MEMORY_SCOPE_NONE); assert(gpuDynInst->scope != Enums::MEMORY_SCOPE_NONE);

2
src/gpu-compute/dispatcher.cc
View file

@ -305,7 +305,7 @@ GpuDispatcher::exec()
void void
GpuDispatcher::notifyWgCompl(Wavefront *w) GpuDispatcher::notifyWgCompl(Wavefront *w)
{ {
int kern_id = w->kern_id; int kern_id = w->kernId;
DPRINTF(GPUDisp, "notify WgCompl %d\n",kern_id); DPRINTF(GPUDisp, "notify WgCompl %d\n",kern_id);
assert(ndRangeMap[kern_id].dispatchId == kern_id); assert(ndRangeMap[kern_id].dispatchId == kern_id);
ndRangeMap[kern_id].numWgCompleted++; ndRangeMap[kern_id].numWgCompleted++;

2
src/gpu-compute/fetch_unit.cc
View file

@ -115,7 +115,7 @@ FetchUnit::initiateFetch(Wavefront *wavefront)
{ {
// calculate the virtual address to fetch from the SQC // calculate the virtual address to fetch from the SQC
Addr vaddr = wavefront->pc() + wavefront->instructionBuffer.size(); 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", DPRINTF(GPUTLB, "CU%d: WF[%d][%d]: Initiating fetch translation: %#x\n",
computeUnit->cu_id, wavefront->simdId, wavefront->wfSlotId, vaddr); computeUnit->cu_id, wavefront->simdId, wavefront->wfSlotId, vaddr);

6
src/gpu-compute/global_memory_pipeline.cc
View file

@ -212,16 +212,16 @@ GlobalMemPipeline::doGmReturn(GPUDynInstPtr m)
} }
// Decrement outstanding register count // 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) || if (m->m_op == Enums::MO_ST || MO_A(m->m_op) || MO_ANR(m->m_op) ||
MO_H(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); -1);
} }
if (m->m_op == Enums::MO_LD || MO_A(m->m_op) || MO_ANR(m->m_op)) { 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); -1);
} }

6
src/gpu-compute/local_memory_pipeline.cc
View file

@ -170,16 +170,16 @@ LocalMemPipeline::doSmReturn(GPUDynInstPtr m)
} }
// Decrement outstanding request count // 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) if (m->m_op == Enums::MO_ST || MO_A(m->m_op) || MO_ANR(m->m_op)
|| MO_H(m->m_op)) { || MO_H(m->m_op)) {
computeUnit->shader->ScheduleAdd(&w->outstanding_reqs_wr_lm, computeUnit->shader->ScheduleAdd(&w->outstandingReqsWrLm,
m->time, -1); m->time, -1);
} }
if (m->m_op == Enums::MO_LD || MO_A(m->m_op) || MO_ANR(m->m_op)) { 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); m->time, -1);
} }

130
src/gpu-compute/wavefront.cc
View file

@ -52,43 +52,43 @@ WavefrontParams::create()
Wavefront::Wavefront(const Params *p) Wavefront::Wavefront(const Params *p)
: SimObject(p), callArgMem(nullptr) : SimObject(p), callArgMem(nullptr)
{ {
last_trace = 0; lastTrace = 0;
simdId = p->simdId; simdId = p->simdId;
wfSlotId = p->wf_slot_id; wfSlotId = p->wf_slot_id;
status = S_STOPPED; status = S_STOPPED;
reservedVectorRegs = 0; reservedVectorRegs = 0;
startVgprIndex = 0; startVgprIndex = 0;
outstanding_reqs = 0; outstandingReqs = 0;
mem_reqs_in_pipe = 0; memReqsInPipe = 0;
outstanding_reqs_wr_gm = 0; outstandingReqsWrGm = 0;
outstanding_reqs_wr_lm = 0; outstandingReqsWrLm = 0;
outstanding_reqs_rd_gm = 0; outstandingReqsRdGm = 0;
outstanding_reqs_rd_lm = 0; outstandingReqsRdLm = 0;
rd_lm_reqs_in_pipe = 0; rdLmReqsInPipe = 0;
rd_gm_reqs_in_pipe = 0; rdGmReqsInPipe = 0;
wr_lm_reqs_in_pipe = 0; wrLmReqsInPipe = 0;
wr_gm_reqs_in_pipe = 0; wrGmReqsInPipe = 0;
barrier_cnt = 0; barrierCnt = 0;
old_barrier_cnt = 0; oldBarrierCnt = 0;
stalledAtBarrier = false; stalledAtBarrier = false;
mem_trace_busy = 0; memTraceBusy = 0;
old_vgpr_tcnt = 0xffffffffffffffffll; oldVgprTcnt = 0xffffffffffffffffll;
old_dgpr_tcnt = 0xffffffffffffffffll; oldDgprTcnt = 0xffffffffffffffffll;
old_vgpr.resize(p->wfSize); oldVgpr.resize(p->wfSize);
pendingFetch = false; pendingFetch = false;
dropFetch = false; dropFetch = false;
condRegState = new ConditionRegisterState(); condRegState = new ConditionRegisterState();
maxSpVgprs = 0; maxSpVgprs = 0;
maxDpVgprs = 0; maxDpVgprs = 0;
last_addr.resize(p->wfSize); lastAddr.resize(p->wfSize);
workitemFlatId.resize(p->wfSize); workItemFlatId.resize(p->wfSize);
old_dgpr.resize(p->wfSize); oldDgpr.resize(p->wfSize);
bar_cnt.resize(p->wfSize); barCnt.resize(p->wfSize);
for (int i = 0; i < 3; ++i) { 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) Wavefront::start(uint64_t _wfDynId,uint64_t _base_ptr)
{ {
wfDynId = _wfDynId; wfDynId = _wfDynId;
base_ptr = _base_ptr; basePtr = _base_ptr;
status = S_RUNNING; status = S_RUNNING;
} }
@ -333,12 +333,12 @@ Wavefront::ready(itype_e type)
// Is the wave waiting at a barrier // Is the wave waiting at a barrier
if (stalledAtBarrier) { if (stalledAtBarrier) {
if (!computeUnit->AllAtBarrier(barrier_id,barrier_cnt, if (!computeUnit->AllAtBarrier(barrierId,barrierCnt,
computeUnit->getRefCounter(dispatchid, wg_id))) { computeUnit->getRefCounter(dispatchId, wgId))) {
// Are all threads at barrier? // Are all threads at barrier?
return 0; return 0;
} }
old_barrier_cnt = barrier_cnt; oldBarrierCnt = barrierCnt;
stalledAtBarrier = false; stalledAtBarrier = false;
} }
@ -395,7 +395,7 @@ Wavefront::ready(itype_e type)
} }
// Are there in pipe or outstanding memory requests? // Are there in pipe or outstanding memory requests?
if ((outstanding_reqs + mem_reqs_in_pipe) > 0) { if ((outstandingReqs + memReqsInPipe) > 0) {
return 0; return 0;
} }
@ -416,7 +416,7 @@ Wavefront::ready(itype_e type)
} }
// Are there in pipe or outstanding memory requests? // Are there in pipe or outstanding memory requests?
if ((outstanding_reqs + mem_reqs_in_pipe) > 0) { if ((outstandingReqs + memReqsInPipe) > 0) {
return 0; return 0;
} }
@ -444,7 +444,7 @@ Wavefront::ready(itype_e type)
// Here Global memory instruction // Here Global memory instruction
if (IS_OT_READ_GM(ii->opType()) || IS_OT_ATOMIC_GM(ii->opType())) { if (IS_OT_READ_GM(ii->opType()) || IS_OT_ATOMIC_GM(ii->opType())) {
// Are there in pipe or outstanding global memory write requests? // 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; return 0;
} }
} }
@ -452,7 +452,7 @@ Wavefront::ready(itype_e type)
if (IS_OT_WRITE_GM(ii->opType()) || IS_OT_ATOMIC_GM(ii->opType()) || if (IS_OT_WRITE_GM(ii->opType()) || IS_OT_ATOMIC_GM(ii->opType()) ||
IS_OT_HIST_GM(ii->opType())) { IS_OT_HIST_GM(ii->opType())) {
// Are there in pipe or outstanding global memory read requests? // 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; return 0;
} }
@ -467,7 +467,7 @@ Wavefront::ready(itype_e type)
} }
if (!computeUnit->globalMemoryPipe. 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? // Can we insert a new request to the Global Mem Request FIFO?
return 0; return 0;
} }
@ -484,14 +484,14 @@ Wavefront::ready(itype_e type)
IS_OT_WRITE_LM(ii->opType()) || IS_OT_ATOMIC_LM(ii->opType()))) { IS_OT_WRITE_LM(ii->opType()) || IS_OT_ATOMIC_LM(ii->opType()))) {
// Here for Shared memory instruction // Here for Shared memory instruction
if (IS_OT_READ_LM(ii->opType()) || IS_OT_ATOMIC_LM(ii->opType())) { 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; return 0;
} }
} }
if (IS_OT_WRITE_LM(ii->opType()) || IS_OT_ATOMIC_LM(ii->opType()) || if (IS_OT_WRITE_LM(ii->opType()) || IS_OT_ATOMIC_LM(ii->opType()) ||
IS_OT_HIST_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; return 0;
} }
} }
@ -506,7 +506,7 @@ Wavefront::ready(itype_e type)
} }
if (!computeUnit->localMemoryPipe. 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? // Can we insert a new request to the LDS Request FIFO?
return 0; return 0;
} }
@ -523,14 +523,14 @@ Wavefront::ready(itype_e type)
IS_OT_WRITE_PM(ii->opType()) || IS_OT_ATOMIC_PM(ii->opType()))) { IS_OT_WRITE_PM(ii->opType()) || IS_OT_ATOMIC_PM(ii->opType()))) {
// Here for Private memory instruction ------------------------ // // Here for Private memory instruction ------------------------ //
if (IS_OT_READ_PM(ii->opType()) || IS_OT_ATOMIC_PM(ii->opType())) { 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; return 0;
} }
} }
if (IS_OT_WRITE_PM(ii->opType()) || IS_OT_ATOMIC_PM(ii->opType()) || if (IS_OT_WRITE_PM(ii->opType()) || IS_OT_ATOMIC_PM(ii->opType()) ||
IS_OT_HIST_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; return 0;
} }
} }
@ -546,7 +546,7 @@ Wavefront::ready(itype_e type)
} }
if (!computeUnit->globalMemoryPipe. 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? // Can we insert a new request to the Global Mem Request FIFO?
return 0; return 0;
} }
@ -579,13 +579,13 @@ Wavefront::ready(itype_e type)
return 0; return 0;
} }
if (!computeUnit->globalMemoryPipe. 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? // Can we insert a new request to the Global Mem Request FIFO?
return 0; return 0;
} }
if (!computeUnit->localMemoryPipe. 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? // Can we insert a new request to the LDS Request FIFO?
return 0; return 0;
} }
@ -636,8 +636,8 @@ Wavefront::updateResources()
ticks(computeUnit->issuePeriod)); ticks(computeUnit->issuePeriod));
} else if (ii->opType() == Enums::OT_FLAT_READ) { } else if (ii->opType() == Enums::OT_FLAT_READ) {
assert(Enums::SC_NONE != ii->executedAs()); assert(Enums::SC_NONE != ii->executedAs());
mem_reqs_in_pipe++; memReqsInPipe++;
rd_gm_reqs_in_pipe++; rdGmReqsInPipe++;
if ( Enums::SC_SHARED == ii->executedAs() ) { if ( Enums::SC_SHARED == ii->executedAs() ) {
computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()]. computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()].
preset(computeUnit->shader->ticks(4)); preset(computeUnit->shader->ticks(4));
@ -651,8 +651,8 @@ Wavefront::updateResources()
} }
} else if (ii->opType() == Enums::OT_FLAT_WRITE) { } else if (ii->opType() == Enums::OT_FLAT_WRITE) {
assert(Enums::SC_NONE != ii->executedAs()); assert(Enums::SC_NONE != ii->executedAs());
mem_reqs_in_pipe++; memReqsInPipe++;
wr_gm_reqs_in_pipe++; wrGmReqsInPipe++;
if (Enums::SC_SHARED == ii->executedAs()) { if (Enums::SC_SHARED == ii->executedAs()) {
computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()]. computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()].
preset(computeUnit->shader->ticks(8)); preset(computeUnit->shader->ticks(8));
@ -665,67 +665,67 @@ Wavefront::updateResources()
preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); preset(computeUnit->shader->ticks(computeUnit->issuePeriod));
} }
} else if (IS_OT_READ_GM(ii->opType())) { } else if (IS_OT_READ_GM(ii->opType())) {
mem_reqs_in_pipe++; memReqsInPipe++;
rd_gm_reqs_in_pipe++; rdGmReqsInPipe++;
computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()].
preset(computeUnit->shader->ticks(4)); preset(computeUnit->shader->ticks(4));
computeUnit->wfWait[computeUnit->GlbMemUnitId()]. computeUnit->wfWait[computeUnit->GlbMemUnitId()].
preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); preset(computeUnit->shader->ticks(computeUnit->issuePeriod));
} else if (IS_OT_WRITE_GM(ii->opType())) { } else if (IS_OT_WRITE_GM(ii->opType())) {
mem_reqs_in_pipe++; memReqsInPipe++;
wr_gm_reqs_in_pipe++; wrGmReqsInPipe++;
computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()].
preset(computeUnit->shader->ticks(8)); preset(computeUnit->shader->ticks(8));
computeUnit->wfWait[computeUnit->GlbMemUnitId()]. computeUnit->wfWait[computeUnit->GlbMemUnitId()].
preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); preset(computeUnit->shader->ticks(computeUnit->issuePeriod));
} else if (IS_OT_ATOMIC_GM(ii->opType())) { } else if (IS_OT_ATOMIC_GM(ii->opType())) {
mem_reqs_in_pipe++; memReqsInPipe++;
wr_gm_reqs_in_pipe++; wrGmReqsInPipe++;
rd_gm_reqs_in_pipe++; rdGmReqsInPipe++;
computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()].
preset(computeUnit->shader->ticks(8)); preset(computeUnit->shader->ticks(8));
computeUnit->wfWait[computeUnit->GlbMemUnitId()]. computeUnit->wfWait[computeUnit->GlbMemUnitId()].
preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); preset(computeUnit->shader->ticks(computeUnit->issuePeriod));
} else if (IS_OT_READ_LM(ii->opType())) { } else if (IS_OT_READ_LM(ii->opType())) {
mem_reqs_in_pipe++; memReqsInPipe++;
rd_lm_reqs_in_pipe++; rdLmReqsInPipe++;
computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()]. computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()].
preset(computeUnit->shader->ticks(4)); preset(computeUnit->shader->ticks(4));
computeUnit->wfWait[computeUnit->ShrMemUnitId()]. computeUnit->wfWait[computeUnit->ShrMemUnitId()].
preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); preset(computeUnit->shader->ticks(computeUnit->issuePeriod));
} else if (IS_OT_WRITE_LM(ii->opType())) { } else if (IS_OT_WRITE_LM(ii->opType())) {
mem_reqs_in_pipe++; memReqsInPipe++;
wr_lm_reqs_in_pipe++; wrLmReqsInPipe++;
computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()]. computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()].
preset(computeUnit->shader->ticks(8)); preset(computeUnit->shader->ticks(8));
computeUnit->wfWait[computeUnit->ShrMemUnitId()]. computeUnit->wfWait[computeUnit->ShrMemUnitId()].
preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); preset(computeUnit->shader->ticks(computeUnit->issuePeriod));
} else if (IS_OT_ATOMIC_LM(ii->opType())) { } else if (IS_OT_ATOMIC_LM(ii->opType())) {
mem_reqs_in_pipe++; memReqsInPipe++;
wr_lm_reqs_in_pipe++; wrLmReqsInPipe++;
rd_lm_reqs_in_pipe++; rdLmReqsInPipe++;
computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()]. computeUnit->vrfToLocalMemPipeBus[computeUnit->nextLocRdBus()].
preset(computeUnit->shader->ticks(8)); preset(computeUnit->shader->ticks(8));
computeUnit->wfWait[computeUnit->ShrMemUnitId()]. computeUnit->wfWait[computeUnit->ShrMemUnitId()].
preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); preset(computeUnit->shader->ticks(computeUnit->issuePeriod));
} else if (IS_OT_READ_PM(ii->opType())) { } else if (IS_OT_READ_PM(ii->opType())) {
mem_reqs_in_pipe++; memReqsInPipe++;
rd_gm_reqs_in_pipe++; rdGmReqsInPipe++;
computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()].
preset(computeUnit->shader->ticks(4)); preset(computeUnit->shader->ticks(4));
computeUnit->wfWait[computeUnit->GlbMemUnitId()]. computeUnit->wfWait[computeUnit->GlbMemUnitId()].
preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); preset(computeUnit->shader->ticks(computeUnit->issuePeriod));
} else if (IS_OT_WRITE_PM(ii->opType())) { } else if (IS_OT_WRITE_PM(ii->opType())) {
mem_reqs_in_pipe++; memReqsInPipe++;
wr_gm_reqs_in_pipe++; wrGmReqsInPipe++;
computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()].
preset(computeUnit->shader->ticks(8)); preset(computeUnit->shader->ticks(8));
computeUnit->wfWait[computeUnit->GlbMemUnitId()]. computeUnit->wfWait[computeUnit->GlbMemUnitId()].
preset(computeUnit->shader->ticks(computeUnit->issuePeriod)); preset(computeUnit->shader->ticks(computeUnit->issuePeriod));
} else if (IS_OT_ATOMIC_PM(ii->opType())) { } else if (IS_OT_ATOMIC_PM(ii->opType())) {
mem_reqs_in_pipe++; memReqsInPipe++;
wr_gm_reqs_in_pipe++; wrGmReqsInPipe++;
rd_gm_reqs_in_pipe++; rdGmReqsInPipe++;
computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()]. computeUnit->vrfToGlobalMemPipeBus[computeUnit->nextGlbRdBus()].
preset(computeUnit->shader->ticks(8)); preset(computeUnit->shader->ticks(8));
computeUnit->wfWait[computeUnit->GlbMemUnitId()]. computeUnit->wfWait[computeUnit->GlbMemUnitId()].
@ -865,7 +865,7 @@ Wavefront::exec()
bool bool
Wavefront::waitingAtBarrier(int lane) Wavefront::waitingAtBarrier(int lane)
{ {
return bar_cnt[lane] < max_bar_cnt; return barCnt[lane] < maxBarCnt;
} }
void void

78
src/gpu-compute/wavefront.hh
View file

@ -155,16 +155,16 @@ class Wavefront : public SimObject
enum status_e {S_STOPPED,S_RETURNING,S_RUNNING}; enum status_e {S_STOPPED,S_RETURNING,S_RUNNING};
// Base pointer for array of instruction pointers // Base pointer for array of instruction pointers
uint64_t base_ptr; uint64_t basePtr;
uint32_t old_barrier_cnt; uint32_t oldBarrierCnt;
uint32_t barrier_cnt; uint32_t barrierCnt;
uint32_t barrier_id; uint32_t barrierId;
uint32_t barrier_slots; uint32_t barrierSlots;
status_e status; status_e status;
// HW slot id where the WF is mapped to inside a SIMD unit // HW slot id where the WF is mapped to inside a SIMD unit
int wfSlotId; int wfSlotId;
int kern_id; int kernId;
// SIMD unit where the WV has been scheduled // SIMD unit where the WV has been scheduled
int simdId; int simdId;
// pointer to parent CU // pointer to parent CU
@ -193,37 +193,37 @@ class Wavefront : public SimObject
bool isOldestInstALU(); bool isOldestInstALU();
bool isOldestInstBarrier(); bool isOldestInstBarrier();
// used for passing spill address to DDInstGPU // used for passing spill address to DDInstGPU
std::vector<Addr> last_addr; std::vector<Addr> lastAddr;
std::vector<uint32_t> workitemid[3]; std::vector<uint32_t> workItemId[3];
std::vector<uint32_t> workitemFlatId; std::vector<uint32_t> workItemFlatId;
uint32_t workgroupid[3]; uint32_t workGroupId[3];
uint32_t workgroupsz[3]; uint32_t workGroupSz[3];
uint32_t gridsz[3]; uint32_t gridSz[3];
uint32_t wg_id; uint32_t wgId;
uint32_t wg_sz; uint32_t wgSz;
uint32_t dynwaveid; uint32_t dynWaveId;
uint32_t maxdynwaveid; uint32_t maxDynWaveId;
uint32_t dispatchid; uint32_t dispatchId;
// outstanding global+local memory requests // outstanding global+local memory requests
uint32_t outstanding_reqs; uint32_t outstandingReqs;
// memory requests between scoreboard // memory requests between scoreboard
// and execute stage not yet executed // and execute stage not yet executed
uint32_t mem_reqs_in_pipe; uint32_t memReqsInPipe;
// outstanding global memory write requests // outstanding global memory write requests
uint32_t outstanding_reqs_wr_gm; uint32_t outstandingReqsWrGm;
// outstanding local memory write requests // outstanding local memory write requests
uint32_t outstanding_reqs_wr_lm; uint32_t outstandingReqsWrLm;
// outstanding global memory read requests // outstanding global memory read requests
uint32_t outstanding_reqs_rd_gm; uint32_t outstandingReqsRdGm;
// outstanding local memory read requests // outstanding local memory read requests
uint32_t outstanding_reqs_rd_lm; uint32_t outstandingReqsRdLm;
uint32_t rd_lm_reqs_in_pipe; uint32_t rdLmReqsInPipe;
uint32_t rd_gm_reqs_in_pipe; uint32_t rdGmReqsInPipe;
uint32_t wr_lm_reqs_in_pipe; uint32_t wrLmReqsInPipe;
uint32_t wr_gm_reqs_in_pipe; uint32_t wrGmReqsInPipe;
int mem_trace_busy; int memTraceBusy;
uint64_t last_trace; uint64_t lastTrace;
// number of vector registers reserved by WF // number of vector registers reserved by WF
int reservedVectorRegs; int reservedVectorRegs;
// Index into the Vector Register File's namespace where the WF's registers // Index into the Vector Register File's namespace where the WF's registers
@ -231,25 +231,25 @@ class Wavefront : public SimObject
uint32_t startVgprIndex; uint32_t startVgprIndex;
// Old value of destination gpr (for trace) // Old value of destination gpr (for trace)
std::vector<uint32_t> old_vgpr; std::vector<uint32_t> oldVgpr;
// Id of destination gpr (for trace) // Id of destination gpr (for trace)
uint32_t old_vgpr_id; uint32_t oldVgprId;
// Tick count of last old_vgpr copy // Tick count of last old_vgpr copy
uint64_t old_vgpr_tcnt; uint64_t oldVgprTcnt;
// Old value of destination gpr (for trace) // Old value of destination gpr (for trace)
std::vector<uint64_t> old_dgpr; std::vector<uint64_t> oldDgpr;
// Id of destination gpr (for trace) // Id of destination gpr (for trace)
uint32_t old_dgpr_id; uint32_t oldDgprId;
// Tick count of last old_vgpr copy // Tick count of last old_vgpr copy
uint64_t old_dgpr_tcnt; uint64_t oldDgprTcnt;
// Execution mask at wavefront start // Execution mask at wavefront start
VectorMask init_mask; VectorMask initMask;
// number of barriers this WF has joined // number of barriers this WF has joined
std::vector<int> bar_cnt; std::vector<int> barCnt;
int max_bar_cnt; int maxBarCnt;
// Flag to stall a wave on barrier // Flag to stall a wave on barrier
bool stalledAtBarrier; bool stalledAtBarrier;
@ -333,7 +333,7 @@ class Wavefront : public SimObject
int ready(itype_e type); int ready(itype_e type);
bool instructionBufferHasBranch(); bool instructionBufferHasBranch();
void regStats(); void regStats();
VectorMask get_pred() { return execMask() & init_mask; } VectorMask getPred() { return execMask() & initMask; }
bool waitingAtBarrier(int lane); bool waitingAtBarrier(int lane);