Skip to content

Commit

Permalink
Fix non-legacy atomics in gradient calculations.
Browse files Browse the repository at this point in the history
  • Loading branch information
@ohearnk
ohearnk committed Oct 26, 2024
1 parent 337a584 commit 6affe35
Show file tree
Hide file tree
Showing 10 changed files with 1,514 additions and 1,185 deletions.
235 changes: 202 additions & 33 deletions src/gpu/cuda/gpu.cu
View file
Original file line number Diff line number Diff line change
Expand Up @@ -51,12 +51,55 @@ extern "C" void gpu_startup_(int* ierr)

gpu = new gpu_type;

#if defined(DEBUG) || defined(DEBUGTIME)
gpu->debugFile = debugFile;
#endif
gpu->totalCPUMemory = 0;
gpu->totalGPUMemory = 0;
gpu->gpu_dev_id = -1;
gpu->blocks = 0;
gpu->threadsPerBlock = 0;
gpu->twoEThreadsPerBlock = 0;
gpu->XCThreadsPerBlock = 0;
gpu->gradThreadsPerBlock = 0;
gpu->xc_blocks = 0;
gpu->xc_threadsPerBlock = 0;
gpu->sswGradThreadsPerBlock = 0;
gpu->mpirank = -1;
gpu->mpisize = 0;
gpu->timer = NULL;
gpu->natom = 0;
gpu->nextatom = 0;
gpu->nbasis = 0;
gpu->nElec = 0;
gpu->imult = 0;
gpu->molchg = 0;
gpu->iAtomType = 0;
gpu->nshell = 0;
gpu->nprim = 0;
gpu->jshell = 0;
gpu->jbasis = 0;
gpu->maxL = 0;
gpu->iattype = NULL;
gpu->xyz = NULL;
gpu->allxyz = NULL;
gpu->chg = NULL;
gpu->allchg = NULL;
gpu->DFT_calculated = NULL;
gpu->grad = NULL;
gpu->ptchg_grad = NULL;
gpu->gradULL = NULL;
gpu->ptchg_gradULL = NULL;
gpu->cew_grad = NULL;
gpu->gpu_calculated = NULL;
gpu->gpu_basis = NULL;
gpu->gpu_cutoff = NULL;
gpu->gpu_xcq = NULL;
gpu->aoint_buffer = NULL;
gpu->intCount = NULL;
gpu->scratch = NULL;
gpu->lri_data = NULL;

#if defined DEBUG || defined DEBUGTIME
gpu->debugFile = debugFile;
#endif
PRINTDEBUG("CREATE NEW GPU");
}

Expand Down Expand Up @@ -234,17 +277,16 @@ extern "C" void gpu_allocate_scratch_(bool* allocate_gradient_scratch)
gpu->scratch->store = new gpu_buffer_type<QUICKDouble>(store_size);
gpu->scratch->store->DeleteCPU( );

gpu->scratch->store2 = new gpu_buffer_type<QUICKDouble>(store_size);
gpu->scratch->store2->DeleteCPU( );

gpu->scratch->YVerticalTemp = new gpu_buffer_type<QUICKDouble>(
gpu->blocks * gpu->twoEThreadsPerBlock * VDIM1 * VDIM2 * VDIM3);
gpu->scratch->YVerticalTemp->DeleteCPU( );

gpu->gpu_sim.store = gpu->scratch->store->_devData;
gpu->gpu_sim.YVerticalTemp = gpu->scratch->YVerticalTemp->_devData;

gpu->scratch->store2 = new gpu_buffer_type<QUICKDouble>(store_size);
gpu->scratch->store2->DeleteCPU( );

gpu->gpu_sim.store2 = gpu->scratch->store2->_devData;
gpu->gpu_sim.YVerticalTemp = gpu->scratch->YVerticalTemp->_devData;

if (*allocate_gradient_scratch) {
gpu->scratch->storeAA = new gpu_buffer_type<QUICKDouble>(store_size);
Expand All @@ -259,10 +301,22 @@ extern "C" void gpu_allocate_scratch_(bool* allocate_gradient_scratch)
gpu->gpu_sim.storeAA = gpu->scratch->storeAA->_devData;
gpu->gpu_sim.storeBB = gpu->scratch->storeBB->_devData;
gpu->gpu_sim.storeCC = gpu->scratch->storeCC->_devData;
} else {
gpu->scratch->storeAA = NULL;
gpu->scratch->storeBB = NULL;
gpu->scratch->storeCC = NULL;
gpu->gpu_sim.storeAA = NULL;
gpu->gpu_sim.storeBB = NULL;
gpu->gpu_sim.storeCC = NULL;
}

cudaMemsetAsync(gpu->gpu_sim.store, 0, sizeof(QUICKDouble) * gpu->scratch->store->_length);
cudaMemsetAsync(gpu->gpu_sim.store2, 0, sizeof(QUICKDouble) * gpu->scratch->store2->_length);
if (*allocate_gradient_scratch) {
cudaMemsetAsync(gpu->gpu_sim.storeAA, 0, sizeof(QUICKDouble) * gpu->scratch->storeAA->_length);
cudaMemsetAsync(gpu->gpu_sim.storeBB, 0, sizeof(QUICKDouble) * gpu->scratch->storeBB->_length);
cudaMemsetAsync(gpu->gpu_sim.storeCC, 0, sizeof(QUICKDouble) * gpu->scratch->storeCC->_length);
}
}


Expand Down Expand Up @@ -307,7 +361,6 @@ extern "C" void gpu_shutdown_(int* ierr)
extern "C" void gpu_setup_(int* natom, int* nbasis, int* nElec, int* imult, int* molchg,
int* iAtomType)
{

#if defined(DEBUG)
cudaEvent_t start,end;
cudaEventCreate(&start);
Expand All @@ -326,14 +379,84 @@ extern "C" void gpu_setup_(int* natom, int* nbasis, int* nElec, int* imult, int*
gpu->imult = *imult;
gpu->molchg = *molchg;
gpu->iAtomType = *iAtomType;

gpu->gpu_calculated = new gpu_calculated_type;
gpu->gpu_basis = new gpu_basis_type;
gpu->gpu_cutoff = new gpu_cutoff_type;
gpu->gpu_xcq = new XC_quadrature_type;
gpu->gpu_calculated->natom = *natom;
gpu->gpu_basis->natom = *natom;
gpu->gpu_calculated->nbasis = *nbasis;
gpu->gpu_calculated->o = NULL;
gpu->gpu_calculated->ob = NULL;
gpu->gpu_calculated->dense = NULL;
gpu->gpu_calculated->denseb = NULL;
#if defined(USE_LEGACY_ATOMICS)
gpu->gpu_calculated->oULL = NULL;
gpu->gpu_calculated->obULL = NULL;
#endif
gpu->gpu_calculated->distance = NULL;

gpu->gpu_basis = new gpu_basis_type;
gpu->gpu_basis->natom = *natom;
gpu->gpu_basis->nbasis = *nbasis;
gpu->gpu_basis->nshell = 0;
gpu->gpu_basis->nprim = 0;
gpu->gpu_basis->jshell = 0;
gpu->gpu_basis->jbasis = 0;
gpu->gpu_basis->Qshell = 0;
gpu->gpu_basis->maxcontract = 0;
gpu->gpu_basis->prim_total = 0;
gpu->gpu_basis->fStart = 0;
gpu->gpu_basis->ffStart = 0;
gpu->gpu_basis->ncontract = NULL;
gpu->gpu_basis->itype = NULL;
gpu->gpu_basis->aexp = NULL;
gpu->gpu_basis->dcoeff = NULL;
gpu->gpu_basis->ncenter = NULL;
gpu->gpu_basis->sigrad2 = NULL;
gpu->gpu_basis->kstart = NULL;
gpu->gpu_basis->katom = NULL;
gpu->gpu_basis->kprim = NULL;
gpu->gpu_basis->Ksumtype = NULL;
gpu->gpu_basis->Qnumber = NULL;
gpu->gpu_basis->Qstart = NULL;
gpu->gpu_basis->Qfinal = NULL;
gpu->gpu_basis->Qsbasis = NULL;
gpu->gpu_basis->Qfbasis = NULL;
gpu->gpu_basis->sorted_Qnumber = NULL;
gpu->gpu_basis->sorted_Q = NULL;
gpu->gpu_basis->gccoeff = NULL;
gpu->gpu_basis->Xcoeff = NULL;
gpu->gpu_basis->Xcoeff_oei = NULL;
gpu->gpu_basis->expoSum = NULL;
gpu->gpu_basis->weightedCenterX = NULL;
gpu->gpu_basis->weightedCenterY = NULL;
gpu->gpu_basis->weightedCenterZ = NULL;
gpu->gpu_basis->cons = NULL;
gpu->gpu_basis->gcexpo = NULL;
gpu->gpu_basis->KLMN = NULL;
gpu->gpu_basis->Apri = NULL;
gpu->gpu_basis->Kpri = NULL;
gpu->gpu_basis->PpriX = NULL;
gpu->gpu_basis->PpriY = NULL;
gpu->gpu_basis->PpriZ = NULL;
gpu->gpu_basis->prim_start = NULL;
gpu->gpu_basis->mpi_bcompute = NULL;
gpu->gpu_basis->mpi_boeicompute = NULL;

gpu->gpu_cutoff = new gpu_cutoff_type;
gpu->gpu_cutoff->natom = 0;
gpu->gpu_cutoff->nbasis = 0;
gpu->gpu_cutoff->nshell = 0;
gpu->gpu_cutoff->sqrQshell = 0;
gpu->gpu_cutoff->sorted_YCutoffIJ = NULL;
gpu->gpu_cutoff->cutMatrix = NULL;
gpu->gpu_cutoff->YCutoff = NULL;
gpu->gpu_cutoff->cutPrim = NULL;
gpu->gpu_cutoff->integralCutoff = 0.0;
gpu->gpu_cutoff->coreIntegralCutoff = 0.0;
gpu->gpu_cutoff->primLimit = 0.0;
gpu->gpu_cutoff->DMCutoff = 0.0;
gpu->gpu_cutoff->XCCutoff = 0.0;
gpu->gpu_cutoff->gradCutoff = 0.0;
gpu->gpu_cutoff->sorted_OEICutoffIJ = NULL;

gpu->gpu_sim.natom = *natom;
gpu->gpu_sim.nbasis = *nbasis;
Expand All @@ -343,6 +466,57 @@ extern "C" void gpu_setup_(int* natom, int* nbasis, int* nElec, int* imult, int*
gpu->gpu_sim.iAtomType = *iAtomType;
gpu->gpu_sim.use_cew = false;

gpu->gpu_xcq = new XC_quadrature_type;
gpu->gpu_xcq->npoints = 0;
gpu->gpu_xcq->nbins = 0;
gpu->gpu_xcq->ntotbf = 0;
gpu->gpu_xcq->ntotpf = 0;
gpu->gpu_xcq->bin_size = 0;
gpu->gpu_xcq->gridy = NULL;
gpu->gpu_xcq->gridz = NULL;
gpu->gpu_xcq->sswt = NULL;
gpu->gpu_xcq->weight = NULL;
gpu->gpu_xcq->gatm = NULL;
gpu->gpu_xcq->bin_counter = NULL;
gpu->gpu_xcq->dweight_ssd = NULL;
gpu->gpu_xcq->basf = NULL;
gpu->gpu_xcq->primf = NULL;
gpu->gpu_xcq->primfpbin = NULL;
gpu->gpu_xcq->basf_locator = NULL;
gpu->gpu_xcq->primf_locator = NULL;
gpu->gpu_xcq->bin_locator = NULL;
gpu->gpu_xcq->densa = NULL;
gpu->gpu_xcq->densb = NULL;
gpu->gpu_xcq->gax = NULL;
gpu->gpu_xcq->gbx = NULL;
gpu->gpu_xcq->gay = NULL;
gpu->gpu_xcq->gby = NULL;
gpu->gpu_xcq->gaz = NULL;
gpu->gpu_xcq->gbz = NULL;
gpu->gpu_xcq->exc = NULL;
gpu->gpu_xcq->xc_grad = NULL;
gpu->gpu_xcq->gxc_grad = NULL;
gpu->gpu_xcq->phi = NULL;
gpu->gpu_xcq->dphidx = NULL;
gpu->gpu_xcq->dphidy = NULL;
gpu->gpu_xcq->dphidz = NULL;
gpu->gpu_xcq->phi_loc = NULL;
gpu->gpu_xcq->npoints_ssd = 0;
gpu->gpu_xcq->gridy_ssd = NULL;
gpu->gpu_xcq->gridz_ssd = NULL;
gpu->gpu_xcq->exc_ssd = NULL;
gpu->gpu_xcq->quadwt = NULL;
gpu->gpu_xcq->gatm_ssd = NULL;
gpu->gpu_xcq->uw_ssd = NULL;
gpu->gpu_xcq->wtang = NULL;
gpu->gpu_xcq->rwt = NULL;
gpu->gpu_xcq->rad3 = NULL;
gpu->gpu_xcq->gpweight = NULL;
gpu->gpu_xcq->cfweight = NULL;
gpu->gpu_xcq->pfweight = NULL;
gpu->gpu_xcq->mpi_bxccompute = NULL;
gpu->gpu_xcq->smem_size = 0;

upload_para_to_const( );

#if defined(DEBUG)
Expand Down Expand Up @@ -1874,8 +2048,10 @@ extern "C" void gpu_upload_lri_(QUICKDouble* zeta, QUICKDouble* cc, int *ierr)
{
gpu->lri_data = new lri_data_type;

gpu->lri_data->cc = new gpu_buffer_type<QUICKDouble>(cc, gpu->natom+gpu->nextatom);
gpu->lri_data->zeta = 0;
gpu->lri_data->cc = new gpu_buffer_type<QUICKDouble>(cc, gpu->natom+gpu->nextatom);
gpu->lri_data->cc->Upload();
gpu->lri_data->vrecip = NULL;

gpu->gpu_sim.lri_zeta = *zeta;

Expand Down Expand Up @@ -2266,41 +2442,34 @@ void print_uploaded_dft_info()

PRINTDEBUG("GRID POINTS & WEIGHTS");

for(int i=0; i<gpu->gpu_xcq->npoints; i++){
fprintf(gpu->debugFile,"Grid: %i x=%f y=%f z=%f sswt=%f weight=%f gatm=%i dweight_ssd=%i \n",i,
for (int i = 0; i < gpu->gpu_xcq->npoints; i++) {
fprintf(gpu->debugFile, "Grid: %i x=%f y=%f z=%f sswt=%f weight=%f gatm=%i dweight_ssd=%i \n", i,
gpu->gpu_xcq->gridx->_hostData[i], gpu->gpu_xcq->gridy->_hostData[i], gpu->gpu_xcq->gridz->_hostData[i],
gpu->gpu_xcq->sswt->_hostData[i], gpu->gpu_xcq->weight->_hostData[i], gpu->gpu_xcq->gatm->_hostData[i],
gpu->gpu_xcq->dweight_ssd->_hostData[i]);
}

PRINTDEBUG("BASIS & PRIMITIVE FUNCTION LISTS");

for(int bin_id=0; bin_id<gpu->gpu_xcq->nbins; bin_id++){

for(int i=gpu->gpu_xcq->basf_locator->_hostData[bin_id]; i<gpu->gpu_xcq->basf_locator->_hostData[bin_id+1] ; i++){

for(int j=gpu->gpu_xcq->primf_locator->_hostData[i]; j< gpu->gpu_xcq->primf_locator->_hostData[i+1]; j++){

fprintf(gpu->debugFile,"Bin ID= %i basf location= %i ibas= %i primf location= %i jprim= %i \n", bin_id, i,
for (int bin_id = 0; bin_id < gpu->gpu_xcq->nbins; bin_id++) {
for (int i = gpu->gpu_xcq->basf_locator->_hostData[bin_id]; i < gpu->gpu_xcq->basf_locator->_hostData[bin_id + 1]; i++) {
for (int j = gpu->gpu_xcq->primf_locator->_hostData[i]; j < gpu->gpu_xcq->primf_locator->_hostData[i + 1]; j++) {
fprintf(gpu->debugFile, "Bin ID= %i basf location= %i ibas= %i primf location= %i jprim= %i \n", bin_id, i,
gpu->gpu_xcq->basf->_hostData[i], j, gpu->gpu_xcq->primf->_hostData[j]);

}

}

}

PRINTDEBUG("RADIUS OF SIGNIFICANCE");

for(int i=0; i<gpu->nbasis; i++){
fprintf(gpu->debugFile,"ibas=%i sigrad2=%f \n", i, gpu->gpu_basis->sigrad2->_hostData[i]);
for (int i = 0; i < gpu->nbasis; i++) {
fprintf(gpu->debugFile, "ibas=%i sigrad2=%f \n", i, gpu->gpu_basis->sigrad2->_hostData[i]);
}

for(int i=0; i<gpu->nbasis; i++){
for(int j=0; j<gpu->gpu_basis->maxcontract; j++){

fprintf(gpu->debugFile,"ibas=%i jprim=%i dcoeff=%f \n",i,j, gpu->gpu_basis->dcoeff->_hostData[ j + i * gpu->gpu_basis->maxcontract]);

for (int i = 0; i < gpu->nbasis; i++) {
for (int j = 0; j < gpu->gpu_basis->maxcontract; j++) {
fprintf(gpu->debugFile, "ibas=%i jprim=%i dcoeff=%f \n", i, j,
gpu->gpu_basis->dcoeff->_hostData[j + i * gpu->gpu_basis->maxcontract]);
}
}

Expand Down
2 changes: 1 addition & 1 deletion src/gpu/cuda/gpu_get2e_getxc_drivers.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -394,7 +394,7 @@ extern "C" void gpu_get_cshell_xcgrad_(QUICKDouble *grad)
#endif

delete gpu->grad;
#ifdef USE_LEGACY_ATOMICS
#if defined(USE_LEGACY_ATOMICS)
delete gpu->gradULL;
#endif
delete gpu->gpu_calculated->dense;
Expand Down
Loading

0 comments on commit 6affe35

Please sign in to comment.