64-BitBrainstorm_2026/AMSS-NCKU
8
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Broaden cached CUDA sync paths

Browse Source
This commit is contained in:
CGH0S7
2026-05-01 18:03:04 +08:00
parent 51f3819892
commit 35b6ceff02
6 changed files with 122 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
AMSS_NCKU_source/Parallel.C
View File

@@ -608,6 +608,24 @@ bool cuda_direct_pack_segment_to_device(double *buffer,
MyList<var> *VarLists,
int Symmetry)
{
#if USE_CUDA_Z4C && (ABEtype == 2)
if (state_count == Z4C_CUDA_STATE_COUNT)
{
if (type != 1)
return false;
const double t0 = sync_profile_enabled() ? MPI_Wtime() : 0.0;
const int i0 = cuda_seg_begin(dst, src->Bg, 0);
const int j0 = cuda_seg_begin(dst, src->Bg, 1);
const int k0 = cuda_seg_begin(dst, src->Bg, 2);
const bool ok = z4c_cuda_pack_state_batch_to_device_buffer(
src->Bg, state_count, buffer, src->Bg->shape,
i0, j0, k0,
dst->shape[0], dst->shape[1], dst->shape[2]) == 0;
if (sync_profile_enabled())
sync_profile_stats().direct_pack_sec += MPI_Wtime() - t0;
return ok;
}
#endif
#if USE_CUDA_BSSN
if (state_count <= 0 || state_count > BSSN_CUDA_STATE_COUNT)
return false;
@@ -777,6 +795,22 @@ bool cuda_direct_unpack_segment_from_device(double *buffer,
int state_count,
MyList<var> *VarListd)
{
#if USE_CUDA_Z4C && (ABEtype == 2)
if (state_count == Z4C_CUDA_STATE_COUNT)
{
const double t0 = sync_profile_enabled() ? MPI_Wtime() : 0.0;
const int i0 = cuda_seg_begin(dst, dst->Bg, 0);
const int j0 = cuda_seg_begin(dst, dst->Bg, 1);
const int k0 = cuda_seg_begin(dst, dst->Bg, 2);
const bool ok = z4c_cuda_unpack_state_batch_from_device_buffer(
dst->Bg, state_count, buffer, dst->Bg->shape,
i0, j0, k0,
dst->shape[0], dst->shape[1], dst->shape[2]) == 0;
if (sync_profile_enabled())
sync_profile_stats().direct_unpack_sec += MPI_Wtime() - t0;
return ok;
}
#endif
#if USE_CUDA_BSSN
if (state_count <= 0 || state_count > BSSN_CUDA_STATE_COUNT)
return false;
@@ -856,6 +890,10 @@ bool cuda_unpack_host_region_to_resident(Block *block,
bool cuda_device_state_count_supported(int state_count)
{
#if USE_CUDA_Z4C && (ABEtype == 2)
if (state_count == Z4C_CUDA_STATE_COUNT)
return true;
#endif
#if USE_CUDA_BSSN
return state_count > 0 && state_count <= BSSN_CUDA_STATE_COUNT;
#else
@@ -933,6 +971,10 @@ int cuda_data_packer_device_batched(double *data,
const int state_count = cuda_state_var_count(VarLists, VarListd);
if (!cuda_device_state_count_supported(state_count))
return -1;
#if USE_CUDA_Z4C && (ABEtype == 2)
if (state_count == Z4C_CUDA_STATE_COUNT)
return -1;
#endif
int size_out = 0;
Block *batch_block = 0;