Skip to content

Commit

Permalink
Fix memory leaks.
Browse files Browse the repository at this point in the history
  • Loading branch information
@ohearnk
ohearnk committed Aug 22, 2024
1 parent 55d9ccd commit e2b3f9b
Show file tree
Hide file tree
Showing 4 changed files with 290 additions and 274 deletions.
2 changes: 2 additions & 0 deletions src/gpu/cuda/gpu_get2e_grad_ffff.cu
View file
Original file line number Diff line number Diff line change
Expand Up @@ -251,6 +251,8 @@ void ResortERIs(_gpu_type gpu){
gpu -> gpu_sim.sorted_YCutoffIJ = gpu -> gpu_cutoff -> sorted_YCutoffIJ -> _devData;
gpu -> gpu_sim.ffStart = ffStart;

free(resorted_YCutoffIJ);
free(partial_eris);
}

void getGrad_ffff(_gpu_type gpu)
Expand Down
280 changes: 143 additions & 137 deletions src/gpu/cuda/gpu_libxc.h
View file
Original file line number Diff line number Diff line change
@@ -1,154 +1,160 @@
/*
!---------------------------------------------------------------------!
! Written by Madu Manathunga on 04/06/2021 !
! !
! Copyright (C) 2020-2021 Merz lab !
! Copyright (C) 2020-2021 Götz lab !
! !
! This Source Code Form is subject to the terms of the Mozilla Public !
! License, v. 2.0. If a copy of the MPL was not distributed with this !
! file, You can obtain one at http://mozilla.org/MPL/2.0/. !
!_____________________________________________________________________!
!---------------------------------------------------------------------!
! This source file contains functions that initialize and upload !
! libxc data and delete them. !
!---------------------------------------------------------------------!
*/
!---------------------------------------------------------------------!
! Written by Madu Manathunga on 04/06/2021 !
! !
! Copyright (C) 2020-2021 Merz lab !
! Copyright (C) 2020-2021 Götz lab !
! !
! This Source Code Form is subject to the terms of the Mozilla Public !
! License, v. 2.0. If a copy of the MPL was not distributed with this !
! file, You can obtain one at http://mozilla.org/MPL/2.0/. !
!_____________________________________________________________________!
!---------------------------------------------------------------------!
! This source file contains functions that initialize and upload !
! libxc data and delete them. !
!---------------------------------------------------------------------!
*/

#include "util.h"
#include "gpu_upload.cu"
#include "gpu_cleanup.cu"

gpu_libxc_info** init_gpu_libxc(int * num_of_funcs, int * arr_func_id, int* xc_polarization){

xc_func_type hyb_func;

int num_of_funcs_;
int * arr_func_id_;
double * arr_mix_coeffs_;

if(*num_of_funcs == 1){
if(*xc_polarization > 0){
xc_func_init(&hyb_func, arr_func_id[0], XC_POLARIZED);
}else{
xc_func_init(&hyb_func, arr_func_id[0], XC_UNPOLARIZED);
}

if( hyb_func.info->family == XC_FAMILY_HYB_GGA){
num_of_funcs_ = hyb_func.n_func_aux;
//Head-gordon's dispersion functionals has been categorized as HYB but they have no aux_funcs.
//This should be fixed with another conditional statement.
if(num_of_funcs_ > 0){

*num_of_funcs = num_of_funcs_;

arr_func_id_ = (int*) malloc(sizeof(int)*num_of_funcs_);
arr_mix_coeffs_ = (double*) malloc(sizeof(double)*num_of_funcs_);

for(int i=0;i<num_of_funcs_; i++){
arr_func_id_[i] = (hyb_func.func_aux[i])->info->number;
arr_mix_coeffs_[i] = hyb_func.mix_coef[i];

}
}else{
num_of_funcs_ = *num_of_funcs;
arr_func_id_ = arr_func_id;
arr_mix_coeffs_ = (double*) malloc(sizeof(double)*num_of_funcs_);
for(int i=0; i<num_of_funcs_; i++){
arr_mix_coeffs_[i]=1.0;
}

}
}else{
num_of_funcs_ = *num_of_funcs;
arr_func_id_ = arr_func_id;
arr_mix_coeffs_ = (double*) malloc(sizeof(double)*num_of_funcs_);

for(int i=0; i<num_of_funcs_; i++){
arr_mix_coeffs_[i]=1.0;
}

}

xc_func_end(&hyb_func);

}else{
num_of_funcs_ = *num_of_funcs;
arr_func_id_ = arr_func_id;
arr_mix_coeffs_ = (double*) malloc(sizeof(double)*num_of_funcs_);
for(int i=0; i<num_of_funcs_; i++){
arr_mix_coeffs_[i]=1.0;
}
}

int n_bytes = num_of_funcs_*sizeof(gpu_libxc_info*);

//A device array of pointers for gpu_libxc_info data
gpu_libxc_info** h_glinfo_array;
cudaHostAlloc((void**)&h_glinfo_array, n_bytes, cudaHostAllocMapped);


for(int i=0; i<num_of_funcs_; i++){
xc_func_type func;

if(*xc_polarization > 0){
xc_func_init(&func, arr_func_id_[i], XC_POLARIZED);
}else{
xc_func_init(&func, arr_func_id_[i], XC_UNPOLARIZED);

gpu_libxc_info** init_gpu_libxc(int * const num_of_funcs, int * arr_func_id,
int const * const xc_polarization) {
int num_of_funcs_;
int *arr_func_id_;
double *arr_mix_coeffs_;
xc_func_type hyb_func;
//A device array of pointers for gpu_libxc_info data
gpu_libxc_info** h_glinfo_array;

if (*num_of_funcs == 1) {
if (*xc_polarization > 0) {
xc_func_init(&hyb_func, arr_func_id[0], XC_POLARIZED);
} else {
xc_func_init(&hyb_func, arr_func_id[0], XC_UNPOLARIZED);
}

if (hyb_func.info->family == XC_FAMILY_HYB_GGA) {
num_of_funcs_ = hyb_func.n_func_aux;
//Head-gordon's dispersion functionals has been categorized as HYB but they have no aux_funcs.
//This should be fixed with another conditional statement.
if (num_of_funcs_ > 0) {
*num_of_funcs = num_of_funcs_;

arr_func_id_ = (int*) malloc(sizeof(int) * num_of_funcs_);
arr_mix_coeffs_ = (double*) malloc(sizeof(double) * num_of_funcs_);

for (int i = 0; i < num_of_funcs_; i++) {
arr_func_id_[i] = (hyb_func.func_aux[i])->info->number;
arr_mix_coeffs_[i] = hyb_func.mix_coef[i];
}
} else {
num_of_funcs_ = *num_of_funcs;
arr_func_id_ = arr_func_id;
arr_mix_coeffs_ = (double*) malloc(sizeof(double) * num_of_funcs_);
for (int i = 0; i < num_of_funcs_; i++) {
arr_mix_coeffs_[i] = 1.0;
}
//Madu: Put this inside a new libxc param init method
gpu_libxc_info* unkptr;

switch(func.info->family){
case(XC_FAMILY_LDA):
gpu_lda_work_params *ldawp;
ldawp = (gpu_lda_work_params*)malloc(sizeof(gpu_lda_work_params));
get_gpu_work_params(&func, (void*)ldawp);
unkptr = gpu_upload_libxc_info(&func, (void*)ldawp, arr_mix_coeffs_[i], 1);
break;

case(XC_FAMILY_GGA):
case(XC_FAMILY_HYB_GGA):
switch(func.info->kind){
case(XC_EXCHANGE):

gpu_ggax_work_params *ggaxwp;
ggaxwp = (gpu_ggax_work_params*) malloc(sizeof(gpu_ggax_work_params));
get_gpu_work_params(&func, (void*)ggaxwp);
unkptr = gpu_upload_libxc_info(&func, (void*)ggaxwp, arr_mix_coeffs_[i], 1);

break;

case(XC_CORRELATION):
case(XC_EXCHANGE_CORRELATION):
gpu_ggac_work_params *ggacwp;
ggacwp = (gpu_ggac_work_params*) malloc(sizeof(gpu_ggac_work_params));
get_gpu_work_params(&func, (void*)ggacwp);

unkptr = gpu_upload_libxc_info(&func, (void*)ggacwp, arr_mix_coeffs_[i], 1);

break;
}
break;
}

h_glinfo_array[i] = unkptr;
xc_func_end(&func);
}
} else {
num_of_funcs_ = *num_of_funcs;
arr_func_id_ = arr_func_id;
arr_mix_coeffs_ = (double*) malloc(sizeof(double) * num_of_funcs_);

for (int i = 0; i < num_of_funcs_; i++) {
arr_mix_coeffs_[i] = 1.0;
}
}

return h_glinfo_array;
xc_func_end(&hyb_func);
} else {
num_of_funcs_ = *num_of_funcs;
arr_func_id_ = arr_func_id;
arr_mix_coeffs_ = (double*) malloc(sizeof(double) * num_of_funcs_);
for (int i = 0; i < num_of_funcs_; i++) {
arr_mix_coeffs_[i] = 1.0;
}
}

}
cudaHostAlloc((void**) &h_glinfo_array, sizeof(gpu_libxc_info*) * num_of_funcs_, cudaHostAllocMapped);

void libxc_cleanup(gpu_libxc_info** d_glinfo, int n_func){
for (int i = 0; i < num_of_funcs_; i++) {
xc_func_type func;

for(int i=0; i< n_func; i++){
if (*xc_polarization > 0) {
xc_func_init(&func, arr_func_id_[i], XC_POLARIZED);
} else {
xc_func_init(&func, arr_func_id_[i], XC_UNPOLARIZED);
}
//Madu: Put this inside a new libxc param init method
gpu_libxc_info* unkptr;

switch(func.info->family) {
case(XC_FAMILY_LDA):
gpu_lda_work_params *ldawp;
ldawp = (gpu_lda_work_params*)malloc(sizeof(gpu_lda_work_params));
get_gpu_work_params(&func, (void*)ldawp);
unkptr = gpu_upload_libxc_info(&func, (void*)ldawp, arr_mix_coeffs_[i], 1);
break;

case(XC_FAMILY_GGA):
case(XC_FAMILY_HYB_GGA):
switch(func.info->kind) {
case(XC_EXCHANGE):
gpu_ggax_work_params *ggaxwp;
ggaxwp = (gpu_ggax_work_params*) malloc(sizeof(gpu_ggax_work_params));
get_gpu_work_params(&func, (void*)ggaxwp);
unkptr = gpu_upload_libxc_info(&func, (void*)ggaxwp, arr_mix_coeffs_[i], 1);
break;

case(XC_CORRELATION):
case(XC_EXCHANGE_CORRELATION):
gpu_ggac_work_params *ggacwp;
ggacwp = (gpu_ggac_work_params*) malloc(sizeof(gpu_ggac_work_params));
get_gpu_work_params(&func, (void*)ggacwp);

unkptr = gpu_upload_libxc_info(&func, (void*)ggacwp, arr_mix_coeffs_[i], 1);
break;

default:
break;
}
break;

default:
break;
}

gpu_libxc_cleanup(d_glinfo[i]);
h_glinfo_array[i] = unkptr;
xc_func_end(&func);
}

if (*num_of_funcs == 1) {
if (hyb_func.info->family == XC_FAMILY_HYB_GGA) {
if (hyb_func.n_func_aux > 0) {
free(arr_func_id_);
free(arr_mix_coeffs_);
} else {
free(arr_mix_coeffs_);
}
} else {
free(arr_mix_coeffs_);
}
} else {
free(arr_mix_coeffs_);
}

return h_glinfo_array;
}

}

cudaFreeHost(d_glinfo);
void libxc_cleanup(gpu_libxc_info** d_glinfo, int n_func) {
for (int i = 0; i < n_func; i++) {
gpu_libxc_cleanup(d_glinfo[i]);
}

cudaFreeHost(d_glinfo);
}
2 changes: 2 additions & 0 deletions src/gpu/hip/gpu_get2e_grad_ffff.cu
View file
Original file line number Diff line number Diff line change
Expand Up @@ -251,6 +251,8 @@ void ResortERIs(_gpu_type gpu){
gpu -> gpu_sim.sorted_YCutoffIJ = gpu -> gpu_cutoff -> sorted_YCutoffIJ -> _devData;
gpu -> gpu_sim.ffStart = ffStart;

free(resorted_YCutoffIJ);
free(partial_eris);
}

void getGrad_ffff(_gpu_type gpu)
Expand Down
Loading

0 comments on commit e2b3f9b

Please sign in to comment.