hsail/insts/mem.hh

150         getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
180         void execute(GPUDynInstPtr gpuDynInst);
382         getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
436         initiateAcc(GPUDynInstPtr gpuDynInst) override
440             gpuDynInst->statusBitVector = gpuDynInst->exec_mask;
443                 for (int i = 0; i < gpuDynInst->computeUnit()->wfSize(); ++i)
444                     if (gpuDynInst->exec_mask[i])
445                         gpuDynInst->statusVector.push_back(num_dest_operands);
447                         gpuDynInst->statusVector.push_back(0);
452                 c0 *d = &((c0*)gpuDynInst->d_data)
453                     [k * gpuDynInst->computeUnit()->wfSize()];
455                 for (int i = 0; i < gpuDynInst->computeUnit()->wfSize(); ++i) {
456                     if (gpuDynInst->exec_mask[i]) {
457                         Addr vaddr = gpuDynInst->addr[i] + k * sizeof(c0);
461                             *d = gpuDynInst->wavefront()->ldsChunk->
466                                 gpuDynInst->computeUnit()->masterId(),
467                                 0, gpuDynInst->wfDynId);
469                             gpuDynInst->setRequestFlags(req);
473                             if (gpuDynInst->computeUnit()->shader->
475                                 gpuDynInst->isAcquire()) {
479                                 gpuDynInst->execContinuation =
482                                 gpuDynInst->useContinuation = true;
486                                 gpuDynInst->useContinuation = false;
489                             gpuDynInst->computeUnit()->sendRequest(gpuDynInst,
497             gpuDynInst->updateStats();
501         completeAcc(GPUDynInstPtr gpuDynInst) override
527             Wavefront *w = gpuDynInst->wavefront();
545                     &((c1*)gpuDynInst->d_data)[k * w->computeUnit->wfSize()];
548                     if (gpuDynInst->exec_mask[i]) {
566             int loadVrfBankConflictCycles = gpuDynInst->computeUnit()->
567                 vrf[w->simdId]->exec(gpuDynInst->seqNum(), w, regVec,
568                                      sizeof(c0), gpuDynInst->time);
571                 gpuDynInst->computeUnit()->globalMemoryPipe
575                 gpuDynInst->computeUnit()->localMemoryPipe
582         execLdAcq(GPUDynInstPtr gpuDynInst) override
587                 if (gpuDynInst->computeUnit()->shader->separate_acquire_release
588                     && gpuDynInst->isAcquire()) {
589                     gpuDynInst->statusBitVector = VectorMask(1);
590                     gpuDynInst->useContinuation = false;
593                                   gpuDynInst->computeUnit()->masterId(),
594                                   0, gpuDynInst->wfDynId);
596                     gpuDynInst->computeUnit()->injectGlobalMemFence(gpuDynInst, false, req);
677         getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
698         void execute(GPUDynInstPtr gpuDynInst) override;
942         getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
1006         initiateAcc(GPUDynInstPtr gpuDynInst) override
1011                 if (gpuDynInst->computeUnit()->shader->separate_acquire_release
1012                     && gpuDynInst->isRelease()) {
1014                     gpuDynInst->statusBitVector = VectorMask(1);
1015                     gpuDynInst->execContinuation = &GPUStaticInst::execSt;
1016                     gpuDynInst->useContinuation = true;
1019                                   gpuDynInst->computeUnit()->masterId(),
1020                                   0, gpuDynInst->wfDynId);
1022                     gpuDynInst->computeUnit()->injectGlobalMemFence(gpuDynInst, false, req);
1029             execSt(gpuDynInst);
1035         void completeAcc(GPUDynInstPtr gpuDynInst) override { }
1042         execSt(GPUDynInstPtr gpuDynInst) override
1046             gpuDynInst->statusBitVector = gpuDynInst->exec_mask;
1049                 for (int i = 0; i < gpuDynInst->computeUnit()->wfSize(); ++i)
1050                     if (gpuDynInst->exec_mask[i])
1051                         gpuDynInst->statusVector.push_back(num_src_operands);
1053                         gpuDynInst->statusVector.push_back(0);
1057                 c0 *d = &((c0*)gpuDynInst->d_data)
1058                     [k * gpuDynInst->computeUnit()->wfSize()];
1060                 for (int i = 0; i < gpuDynInst->computeUnit()->wfSize(); ++i) {
1061                     if (gpuDynInst->exec_mask[i]) {
1062                         Addr vaddr = gpuDynInst->addr[i] + k * sizeof(c0);
1066                             gpuDynInst->wavefront()->ldsChunk->write<c0>(vaddr,
1071                                 gpuDynInst->computeUnit()->masterId(),
1072                                 0, gpuDynInst->wfDynId);
1074                             gpuDynInst->setRequestFlags(req);
1080                             gpuDynInst->useContinuation = false;
1081                             gpuDynInst->computeUnit()->sendRequest(gpuDynInst,
1090             gpuDynInst->updateStats();
1153         getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst) override
1173         void execute(GPUDynInstPtr gpuDynInst) override;
1443         getRegisterIndex(int operandIndex, GPUDynInstPtr gpuDynInst)
1477         initiateAcc(GPUDynInstPtr gpuDynInst) override
1482                 if (gpuDynInst->computeUnit()->shader->separate_acquire_release
1483                     && (gpuDynInst->isRelease()
1484                     || gpuDynInst->isAcquireRelease())) {
1486                     gpuDynInst->statusBitVector = VectorMask(1);
1488                     gpuDynInst->execContinuation = &GPUStaticInst::execAtomic;
1489                     gpuDynInst->useContinuation = true;
1493                                   gpuDynInst->computeUnit()->masterId(),
1494                                   0, gpuDynInst->wfDynId);
1496                     gpuDynInst->computeUnit()->injectGlobalMemFence(gpuDynInst, false, req);
1503             execAtomic(gpuDynInst);
1508         completeAcc(GPUDynInstPtr gpuDynInst) override
1519                 Wavefront *w = gpuDynInst->wavefront();
1525                 CType *p1 = &((CType*)gpuDynInst->d_data)[0];
1528                     if (gpuDynInst->exec_mask[i]) {
1544                 int loadVrfBankConflictCycles = gpuDynInst->computeUnit()->
1545                     vrf[w->simdId]->exec(gpuDynInst->seqNum(), w, regVec,
1546                                          sizeof(CType), gpuDynInst->time);
1549                     gpuDynInst->computeUnit()->globalMemoryPipe
1553                     gpuDynInst->computeUnit()->localMemoryPipe
1559         void execute(GPUDynInstPtr gpuDynInst) override;
1566         execAtomic(GPUDynInstPtr gpuDynInst) override
1568             gpuDynInst->statusBitVector = gpuDynInst->exec_mask;
1572             c0 *d = &((c0*) gpuDynInst->d_data)[0];
1573             c0 *e = &((c0*) gpuDynInst->a_data)[0];
1574             c0 *f = &((c0*) gpuDynInst->x_data)[0];
1576             for (int i = 0; i < gpuDynInst->computeUnit()->wfSize(); ++i) {
1577                 if (gpuDynInst->exec_mask[i]) {
1578                     Addr vaddr = gpuDynInst->addr[i];
1581                         Wavefront *wavefront = gpuDynInst->wavefront();
1626                                         gpuDynInst->computeUnit()->masterId(),
1627                                         0, gpuDynInst->wfDynId,
1628                                         gpuDynInst->makeAtomicOpFunctor<c0>(e,
1631                         gpuDynInst->setRequestFlags(req);
1635                         if (gpuDynInst->computeUnit()->shader->
1637                             (gpuDynInst->isAcquire())) {
1641                             gpuDynInst->execContinuation =
1644                             gpuDynInst->useContinuation = true;
1647                             gpuDynInst->useContinuation = false;
1650                         gpuDynInst->computeUnit()->sendRequest(gpuDynInst, i,
1660             gpuDynInst->updateStats();
1666         execAtomicAcq(GPUDynInstPtr gpuDynInst) override
1671                 if (gpuDynInst->computeUnit()->shader->separate_acquire_release
1672                      && gpuDynInst->isAcquire()) {
1673                     gpuDynInst->statusBitVector = VectorMask(1);
1677                     gpuDynInst->useContinuation = false;
1680                                   gpuDynInst->computeUnit()->masterId(),
1681                                   0, gpuDynInst->wfDynId);
1683                     gpuDynInst->computeUnit()->injectGlobalMemFence(gpuDynInst, false, req);