Optimize BSSN CUDA resident AMR prolong path

This commit is contained in:
2026-04-30 10:58:15 +08:00
parent 1ee229a91f
commit 18e9c9cc50
3 changed files with 778 additions and 78 deletions

View File

@@ -180,6 +180,38 @@ int bssn_cuda_unpack_state_segments_from_device_buffer_for_host_views(void *bloc
int segment_count,
const int *segment_meta);
int bssn_cuda_restrict_state_segments_to_device_buffer(void *block_tag,
int state_count,
double *device_buffer,
int *ex,
int segment_count,
const int *segment_meta);
int bssn_cuda_restrict_state_segments_to_device_buffer_for_host_views(void *block_tag,
double **state_host_key,
int state_count,
double *device_buffer,
int *ex,
int segment_count,
const int *segment_meta,
const double *state_soa);
int bssn_cuda_prolong_state_segments_to_device_buffer(void *block_tag,
int state_count,
double *device_buffer,
int *ex,
int segment_count,
const int *segment_meta);
int bssn_cuda_prolong_state_segments_to_device_buffer_for_host_views(void *block_tag,
double **state_host_key,
int state_count,
double *device_buffer,
int *ex,
int segment_count,
const int *segment_meta,
const double *state_soa);
int bssn_cuda_restrict_state_batch_to_device_buffer(void *block_tag,
int state_count,
double *device_buffer,
@@ -193,7 +225,8 @@ int bssn_cuda_restrict_state_batch_to_device_buffer_for_host_views(void *block_t
double *device_buffer,
int *ex,
int sx, int sy, int sz,
int fi0, int fj0, int fk0);
int fi0, int fj0, int fk0,
const double *state_soa);
int bssn_cuda_prolong_state_batch_to_device_buffer(void *block_tag,
int state_count,
@@ -210,7 +243,8 @@ int bssn_cuda_prolong_state_batch_to_device_buffer_for_host_views(void *block_ta
int *ex,
int sx, int sy, int sz,
int ii0, int jj0, int kk0,
int lbc_i, int lbc_j, int lbc_k);
int lbc_i, int lbc_j, int lbc_k,
const double *state_soa);
int bssn_cuda_download_state_subset(void *block_tag,
int *ex,