static void exec_gaussblur_app(DATA *image_out, DATA *image_in, ARGUMENTS *settings) {
size_t size_in_bytes = settings->size*settings->size*(PIXEL_CHANNELS*sizeof(DATA));
DATA *image_in_gpu = (DATA *)acc_malloc(size_in_bytes);
DATA *image_out_gpu = (DATA *)acc_malloc(size_in_bytes);
uint32_t e;
uint32_t i;
for (e=0; e<settings->energy_loops; ++e) {
acc_memcpy_to_device(image_in_gpu , image_in , size_in_bytes);
acc_memcpy_to_device(image_out_gpu, image_out, size_in_bytes);
for (i=0; i<settings->checkpoints; ++i) {
gaussblur_calc_gpu_compute((uint16_t *)image_in_gpu , (uint16_t *)image_out_gpu, settings->size);
gaussblur_calc_gpu_compute((uint16_t *)image_out_gpu, (uint16_t *)image_in_gpu , settings->size);
}
acc_memcpy_from_device(image_in , image_in_gpu , size_in_bytes);
acc_memcpy_from_device(image_out, image_out_gpu, size_in_bytes);
}
acc_free(image_in_gpu);
acc_free(image_out_gpu);
image_in_gpu = NULL;
image_out_gpu = NULL;
}
int
main (int argc, char **argv)
{
const int N = 256;
int i;
unsigned char *h;
void *d;
acc_init (acc_device_nvidia);
h = (unsigned char *) malloc (N);
for (i = 0; i < N; i++)
{
h[i] = i;
}
d = acc_malloc (N);
acc_memcpy_to_device (d, h, N);
memset (&h[0], 0, N);
acc_memcpy_to_device (d, h, N << 1);
acc_memcpy_from_device (h, d, N);
for (i = 0; i < N; i++)
{
if (h[i] != i)
abort ();
}
acc_free (d);
free (h);
acc_shutdown (acc_device_nvidia);
return 0;
}
int
main (int argc, char **argv)
{
const int N = 256;
int i;
unsigned char *h;
void *d;
h = (unsigned char *) malloc (N);
for (i = 0; i < N; i++)
{
h[i] = i;
}
d = acc_malloc (N);
fprintf (stderr, "CheCKpOInT\n");
acc_memcpy_to_device (0, h, N);
memset (&h[0], 0, N);
acc_memcpy_from_device (h, d, N);
for (i = 0; i < N; i++)
{
if (h[i] != i)
abort ();
}
acc_free (d);
free (h);
return 0;
}
void nrn_init(NrnThread* _nt, Memb_list* _ml, int _type){
double* _p; Datum* _ppvar; ThreadDatum* _thread;
double _v, v; int* _ni; int _iml, _cntml_padded, _cntml_actual;
_ni = _ml->_nodeindices;
_cntml_actual = _ml->_nodecount;
_cntml_padded = _ml->_nodecount_padded;
_thread = _ml->_thread;
#pragma acc update device (_mechtype) if(_nt->compute_gpu)
if (!_thread[_spth1]._pvoid) {
_thread[_spth1]._pvoid = nrn_cons_sparseobj(_kinetic_kstates_NMDA16_2, 16, _ml, _threadargs_);
#ifdef _OPENACC
if (_nt->compute_gpu) {
void* _d_so = (void*) acc_deviceptr(_thread[_spth1]._pvoid);
ThreadDatum* _d_td = (ThreadDatum*)acc_deviceptr(_thread);
acc_memcpy_to_device(&(_d_td[_spth1]._pvoid), &_d_so, sizeof(void*));
}
#endif
}
_acc_globals_update();
double * _nt_data = _nt->_data;
double * _vec_v = _nt->_actual_v;
int stream_id = _nt->stream_id;
if (_nrn_skip_initmodel == 0) {
#if LAYOUT == 1 /*AoS*/
for (_iml = 0; _iml < _cntml_actual; ++_iml) {
_p = _ml->_data + _iml*_psize; _ppvar = _ml->_pdata + _iml*_ppsize;
#elif LAYOUT == 0 /*SoA*/
_p = _ml->_data; _ppvar = _ml->_pdata;
/* insert compiler dependent ivdep like pragma */
_PRAGMA_FOR_VECTOR_LOOP_
_PRAGMA_FOR_INIT_ACC_LOOP_
for (_iml = 0; _iml < _cntml_actual; ++_iml) {
#else /* LAYOUT > 1 */ /*AoSoA*/
#error AoSoA not implemented.
for (;;) { /* help clang-format properly indent */
#endif
int _nd_idx = _ni[_iml];
_tsav = -1e20;
_v = _vec_v[_nd_idx];
_PRCELLSTATE_V
v = _v;
_PRCELLSTATE_V
nao = _ion_nao;
initmodel(_threadargs_);
}
}
#if NET_RECEIVE_BUFFERING
NetSendBuffer_t* _nsb = _ml->_net_send_buffer;
#if defined(_OPENACC) && !defined(DISABLE_OPENACC)
#pragma acc wait(stream_id)
#pragma acc update self(_nsb->_cnt) if(_nt->compute_gpu)
update_net_send_buffer_on_host(_nt, _nsb);
#endif
{int _i;
for (_i=0; _i < _nsb->_cnt; ++_i) {
net_sem_from_gpu(_nsb->_sendtype[_i], _nsb->_vdata_index[_i],
_nsb->_weight_index[_i], _nt->_id, _nsb->_pnt_index[_i],
_nsb->_nsb_t[_i], _nsb->_nsb_flag[_i]);
}}
_nsb->_cnt = 0;
#if defined(_OPENACC) && !defined(DISABLE_OPENACC)
#pragma acc update device(_nsb->_cnt) if(_nt->compute_gpu)
#endif
#endif
}
static double _nrn_current(_threadargsproto_, double _v){double _current=0.;v=_v;{ {
g = w * gmax * O ;
i = g * ( v - Erev ) ;
}
_current += i;
} return _current;
}
#if defined(ENABLE_CUDA_INTERFACE) && defined(_OPENACC)
void nrn_state_launcher(NrnThread*, Memb_list*, int, int);
void nrn_jacob_launcher(NrnThread*, Memb_list*, int, int);
void nrn_cur_launcher(NrnThread*, Memb_list*, int, int);
#endif
void nrn_cur(NrnThread* _nt, Memb_list* _ml, int _type) {
double* _p; Datum* _ppvar; ThreadDatum* _thread;
int* _ni; double _rhs, _g, _v, v; int _iml, _cntml_padded, _cntml_actual;
_ni = _ml->_nodeindices;
_cntml_actual = _ml->_nodecount;
_cntml_padded = _ml->_nodecount_padded;
_thread = _ml->_thread;
double * _vec_rhs = _nt->_actual_rhs;
double * _vec_d = _nt->_actual_d;
double * _vec_shadow_rhs = _nt->_shadow_rhs;
double * _vec_shadow_d = _nt->_shadow_d;
#if defined(ENABLE_CUDA_INTERFACE) && defined(_OPENACC) && !defined(DISABLE_OPENACC)
NrnThread* d_nt = acc_deviceptr(_nt);
Memb_list* d_ml = acc_deviceptr(_ml);
nrn_cur_launcher(d_nt, d_ml, _type, _cntml_actual);
return;
#endif
double * _nt_data = _nt->_data;
double * _vec_v = _nt->_actual_v;
int stream_id = _nt->stream_id;
#if LAYOUT == 1 /*AoS*/
for (_iml = 0; _iml < _cntml_actual; ++_iml) {
_p = _ml->_data + _iml*_psize; _ppvar = _ml->_pdata + _iml*_ppsize;
#elif LAYOUT == 0 /*SoA*/
_p = _ml->_data; _ppvar = _ml->_pdata;
/* insert compiler dependent ivdep like pragma */
_PRAGMA_FOR_VECTOR_LOOP_
_PRAGMA_FOR_CUR_SYN_ACC_LOOP_
for (_iml = 0; _iml < _cntml_actual; ++_iml) {
#else /* LAYOUT > 1 */ /*AoSoA*/
#error AoSoA not implemented.
for (;;) { /* help clang-format properly indent */
#endif
int _nd_idx = _ni[_iml];
_v = _vec_v[_nd_idx];
_PRCELLSTATE_V
nao = _ion_nao;
_g = _nrn_current(_threadargs_, _v + .001);
{ _rhs = _nrn_current(_threadargs_, _v);
}
_g = (_g - _rhs)/.001;
double _mfact = 1.e2/(_nd_area);
_g *= _mfact;
_rhs *= _mfact;
_PRCELLSTATE_G
#ifdef _OPENACC
if(_nt->compute_gpu) {
#pragma acc atomic update
_vec_rhs[_nd_idx] -= _rhs;
#pragma acc atomic update
_vec_d[_nd_idx] += _g;
} else {
_vec_shadow_rhs[_iml] = _rhs;
_vec_shadow_d[_iml] = _g;
}
#else
_vec_shadow_rhs[_iml] = _rhs;
_vec_shadow_d[_iml] = _g;
#endif
}
#ifdef _OPENACC
if(!(_nt->compute_gpu)) {
for (_iml = 0; _iml < _cntml_actual; ++_iml) {
int _nd_idx = _ni[_iml];
_vec_rhs[_nd_idx] -= _vec_shadow_rhs[_iml];
_vec_d[_nd_idx] += _vec_shadow_d[_iml];
}
#else
for (_iml = 0; _iml < _cntml_actual; ++_iml) {
int _nd_idx = _ni[_iml];
_vec_rhs[_nd_idx] -= _vec_shadow_rhs[_iml];
_vec_d[_nd_idx] += _vec_shadow_d[_iml];
#endif
}
}
void nrn_state(NrnThread* _nt, Memb_list* _ml, int _type) {
double* _p; Datum* _ppvar; ThreadDatum* _thread;
double v, _v = 0.0; int* _ni; int _iml, _cntml_padded, _cntml_actual;
_ni = _ml->_nodeindices;
_cntml_actual = _ml->_nodecount;
_cntml_padded = _ml->_nodecount_padded;
_thread = _ml->_thread;
#if defined(ENABLE_CUDA_INTERFACE) && defined(_OPENACC) && !defined(DISABLE_OPENACC)
NrnThread* d_nt = acc_deviceptr(_nt);
Memb_list* d_ml = acc_deviceptr(_ml);
nrn_state_launcher(d_nt, d_ml, _type, _cntml_actual);
return;
#endif
double * _nt_data = _nt->_data;
double * _vec_v = _nt->_actual_v;
int stream_id = _nt->stream_id;
#if LAYOUT == 1 /*AoS*/
for (_iml = 0; _iml < _cntml_actual; ++_iml) {
_p = _ml->_data + _iml*_psize; _ppvar = _ml->_pdata + _iml*_ppsize;
#elif LAYOUT == 0 /*SoA*/
_p = _ml->_data; _ppvar = _ml->_pdata;
/* insert compiler dependent ivdep like pragma */
_PRAGMA_FOR_VECTOR_LOOP_
_PRAGMA_FOR_STATE_ACC_LOOP_
for (_iml = 0; _iml < _cntml_actual; ++_iml) {
#else /* LAYOUT > 1 */ /*AoSoA*/
#error AoSoA not implemented.
for (;;) { /* help clang-format properly indent */
#endif
int _nd_idx = _ni[_iml];
_v = _vec_v[_nd_idx];
_PRCELLSTATE_V
v=_v;
{
nao = _ion_nao;
{
#if !defined(_kinetic_kstates_NMDA16_2)
#define _kinetic_kstates_NMDA16_2 0
#endif
sparse_thread((SparseObj*)_thread[_spth1]._pvoid, 16, _slist1, _dlist1, &t, dt, _kinetic_kstates_NMDA16_2, _linmat1, _threadargs_);
}}}
}
static void terminal(){}
static void _initlists(){
double _x; double* _p = &_x;
int _i; static int _first = 1;
int _cntml_actual=1;
int _cntml_padded=1;
int _iml=0;
if (!_first) return;
_slist1 = (int*)malloc(sizeof(int)*16);
_dlist1 = (int*)malloc(sizeof(int)*16);
_slist1[0] = &(RA2sMg) - _p; _dlist1[0] = &(DRA2sMg) - _p;
_slist1[1] = &(OMg) - _p; _dlist1[1] = &(DOMg) - _p;
_slist1[2] = &(O) - _p; _dlist1[2] = &(DO) - _p;
_slist1[3] = &(RA2fMg) - _p; _dlist1[3] = &(DRA2fMg) - _p;
_slist1[4] = &(RA2d2Mg) - _p; _dlist1[4] = &(DRA2d2Mg) - _p;
_slist1[5] = &(RA2d1Mg) - _p; _dlist1[5] = &(DRA2d1Mg) - _p;
_slist1[6] = &(RA2Mg) - _p; _dlist1[6] = &(DRA2Mg) - _p;
_slist1[7] = &(RAMg) - _p; _dlist1[7] = &(DRAMg) - _p;
_slist1[8] = &(RMg) - _p; _dlist1[8] = &(DRMg) - _p;
_slist1[9] = &(RA2s) - _p; _dlist1[9] = &(DRA2s) - _p;
_slist1[10] = &(RA2f) - _p; _dlist1[10] = &(DRA2f) - _p;
_slist1[11] = &(RA2d2) - _p; _dlist1[11] = &(DRA2d2) - _p;
_slist1[12] = &(RA2d1) - _p; _dlist1[12] = &(DRA2d1) - _p;
_slist1[13] = &(RA2) - _p; _dlist1[13] = &(DRA2) - _p;
_slist1[14] = &(RA) - _p; _dlist1[14] = &(DRA) - _p;
_slist1[15] = &(R) - _p; _dlist1[15] = &(DR) - _p;
#pragma acc enter data copyin(_slist1[0:16])
#pragma acc enter data copyin(_dlist1[0:16])
_first = 0;
}
} // namespace coreneuron_lib
