ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Merge pull request #6857 from raafatfeki/pr/ompio_coll_write_clean

Просмотр исходного кода 
Pr/ompio_fcoll_write_clean
Этот коммит содержится в:
Raafat Feki 2019-09-04 11:06:56 -05:00 коммит произвёл GitHub
родитель 73c5fe8f05 2c6a5eed29
Коммит 7877743784
Не найден ключ, соответствующий данной подписи
Идентификатор ключа GPG: 4AEE18F83AFDEB23
2 изменённых файлов: 22 добавлений и 18 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

20
ompi/mca/fcoll/dynamic_gen2/fcoll_dynamic_gen2_file_write_all.c
Просмотреть файл

@ -801,7 +801,7 @@ static int shuffle_init ( int index, int cycles, int aggregator, int rank, mca_i
}
for(l=0;l<data->procs_per_group;l++){
data->disp_index[l] = 1;
data->disp_index[l] = 0;
if(data->max_disp_index[l] == 0) {
data->blocklen_per_process[l] = (int *) calloc (INIT_LEN, sizeof(int));
@ -880,8 +880,8 @@ static int shuffle_init ( int index, int cycles, int aggregator, int rank, mca_i
if (data->bytes_remaining <= data->bytes_to_write_in_cycle) {
/* The data fits completely into the block */
if (aggregator == rank) {
data->blocklen_per_process[data->n][data->disp_index[data->n] - 1] = data->bytes_remaining;
data->displs_per_process[data->n][data->disp_index[data->n] - 1] =
data->blocklen_per_process[data->n][data->disp_index[data->n]] = data->bytes_remaining;
data->displs_per_process[data->n][data->disp_index[data->n]] =
(ptrdiff_t)data->global_iov_array[data->sorted[data->current_index]].iov_base +
(data->global_iov_array[data->sorted[data->current_index]].iov_len
- data->bytes_remaining);
@ -914,11 +914,12 @@ static int shuffle_init ( int index, int cycles, int aggregator, int rank, mca_i
/* the remaining data from the previous cycle is larger than the
data->bytes_to_write_in_cycle, so we have to segment again */
if (aggregator == rank) {
data->blocklen_per_process[data->n][data->disp_index[data->n] - 1] = data->bytes_to_write_in_cycle;
data->displs_per_process[data->n][data->disp_index[data->n] - 1] =
data->blocklen_per_process[data->n][data->disp_index[data->n]] = data->bytes_to_write_in_cycle;
data->displs_per_process[data->n][data->disp_index[data->n]] =
(ptrdiff_t)data->global_iov_array[data->sorted[data->current_index]].iov_base +
(data->global_iov_array[data->sorted[data->current_index]].iov_len
- data->bytes_remaining);
data->disp_index[data->n] += 1;
}
if (data->procs_in_group[data->n] == rank) {
@ -935,9 +936,10 @@ static int shuffle_init ( int index, int cycles, int aggregator, int rank, mca_i
(MPI_Aint) data->global_iov_array[data->sorted[data->current_index]].iov_len) {
/* This entry has more data than we can sendin one cycle */
if (aggregator == rank) {
data->blocklen_per_process[data->n][data->disp_index[data->n] - 1] = data->bytes_to_write_in_cycle;
data->displs_per_process[data->n][data->disp_index[data->n] - 1] =
data->blocklen_per_process[data->n][data->disp_index[data->n]] = data->bytes_to_write_in_cycle;
data->displs_per_process[data->n][data->disp_index[data->n]] =
(ptrdiff_t)data->global_iov_array[data->sorted[data->current_index]].iov_base ;
data->disp_index[data->n] += 1;
}
if (data->procs_in_group[data->n] == rank) {
bytes_sent += data->bytes_to_write_in_cycle;
@ -951,9 +953,9 @@ static int shuffle_init ( int index, int cycles, int aggregator, int rank, mca_i
else {
/* Next data entry is less than data->bytes_to_write_in_cycle */
if (aggregator == rank) {
data->blocklen_per_process[data->n][data->disp_index[data->n] - 1] =
data->blocklen_per_process[data->n][data->disp_index[data->n]] =
data->global_iov_array[data->sorted[data->current_index]].iov_len;
data->displs_per_process[data->n][data->disp_index[data->n] - 1] = (ptrdiff_t)
data->displs_per_process[data->n][data->disp_index[data->n]] = (ptrdiff_t)
data->global_iov_array[data->sorted[data->current_index]].iov_base;
data->disp_index[data->n] += 1;

20
ompi/mca/fcoll/vulcan/fcoll_vulcan_file_write_all.c
Просмотреть файл

@ -836,7 +836,7 @@ static int shuffle_init ( int index, int cycles, int aggregator, int rank, mca_i
for(l=0;l<data->procs_per_group;l++){
data->disp_index[l] = 1;
data->disp_index[l] = 0;
if ( data->max_disp_index[l] == 0 ) {
data->blocklen_per_process[l] = (int *) calloc (INIT_LEN, sizeof(int));
@ -915,8 +915,8 @@ static int shuffle_init ( int index, int cycles, int aggregator, int rank, mca_i
if (data->bytes_remaining <= data->bytes_to_write_in_cycle) {
/* The data fits completely into the block */
if (aggregator == rank) {
data->blocklen_per_process[data->n][data->disp_index[data->n] - 1] = data->bytes_remaining;
data->displs_per_process[data->n][data->disp_index[data->n] - 1] =
data->blocklen_per_process[data->n][data->disp_index[data->n]] = data->bytes_remaining;
data->displs_per_process[data->n][data->disp_index[data->n]] =
(ptrdiff_t)data->global_iov_array[data->sorted[data->current_index]].iov_base +
(data->global_iov_array[data->sorted[data->current_index]].iov_len
- data->bytes_remaining);
@ -950,11 +950,12 @@ static int shuffle_init ( int index, int cycles, int aggregator, int rank, mca_i
/* the remaining data from the previous cycle is larger than the
data->bytes_to_write_in_cycle, so we have to segment again */
if (aggregator == rank) {
data->blocklen_per_process[data->n][data->disp_index[data->n] - 1] = data->bytes_to_write_in_cycle;
data->displs_per_process[data->n][data->disp_index[data->n] - 1] =
data->blocklen_per_process[data->n][data->disp_index[data->n]] = data->bytes_to_write_in_cycle;
data->displs_per_process[data->n][data->disp_index[data->n]] =
(ptrdiff_t)data->global_iov_array[data->sorted[data->current_index]].iov_base +
(data->global_iov_array[data->sorted[data->current_index]].iov_len
- data->bytes_remaining);
data->disp_index[data->n] += 1;
}
if (data->procs_in_group[data->n] == rank) {
@ -971,9 +972,10 @@ static int shuffle_init ( int index, int cycles, int aggregator, int rank, mca_i
(MPI_Aint) data->global_iov_array[data->sorted[data->current_index]].iov_len) {
/* This entry has more data than we can sendin one cycle */
if (aggregator == rank) {
data->blocklen_per_process[data->n][data->disp_index[data->n] - 1] = data->bytes_to_write_in_cycle;
data->displs_per_process[data->n][data->disp_index[data->n] - 1] =
data->blocklen_per_process[data->n][data->disp_index[data->n]] = data->bytes_to_write_in_cycle;
data->displs_per_process[data->n][data->disp_index[data->n]] =
(ptrdiff_t)data->global_iov_array[data->sorted[data->current_index]].iov_base ;
data->disp_index[data->n] += 1;
}
if (data->procs_in_group[data->n] == rank) {
bytes_sent += data->bytes_to_write_in_cycle;
@ -987,9 +989,9 @@ static int shuffle_init ( int index, int cycles, int aggregator, int rank, mca_i
else {
/* Next data entry is less than data->bytes_to_write_in_cycle */
if (aggregator == rank) {
data->blocklen_per_process[data->n][data->disp_index[data->n] - 1] =
data->blocklen_per_process[data->n][data->disp_index[data->n]] =
data->global_iov_array[data->sorted[data->current_index]].iov_len;
data->displs_per_process[data->n][data->disp_index[data->n] - 1] = (ptrdiff_t)
data->displs_per_process[data->n][data->disp_index[data->n]] = (ptrdiff_t)
data->global_iov_array[data->sorted[data->current_index]].iov_base;
data->disp_index[data->n] += 1;