Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix multi-GPU performance for division #1559

Merged
merged 2 commits into from
Sep 19, 2024
Merged

Fix multi-GPU performance for division #1559

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
804b526
chore(gpu): remove useless syncs
agnesLeroy Sep 17, 2024
8f98df9
chore(gpu): fix multi-gpu div performance
agnesLeroy Sep 19, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 0 additions & 2 deletions backends/tfhe-cuda-backend/cuda/src/integer/integer.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -443,7 +443,6 @@ void generate_device_accumulator_bivariate(
message_modulus, carry_modulus, f);

// copy host lut and lut_indexes_vec to device
cuda_synchronize_stream(stream, gpu_index);
cuda_memcpy_async_to_gpu(acc_bivariate, h_lut,
(glwe_dimension + 1) * polynomial_size *
sizeof(Torus),
Expand Down Expand Up @@ -509,7 +508,6 @@ void generate_device_accumulator(cudaStream_t stream, uint32_t gpu_index,
generate_lookup_table<Torus>(h_lut, glwe_dimension, polynomial_size,
message_modulus, carry_modulus, f);

cuda_synchronize_stream(stream, gpu_index);
// copy host lut and lut_indexes_vec to device
cuda_memcpy_async_to_gpu(
acc, h_lut, (glwe_dimension + 1) * polynomial_size * sizeof(Torus),
Expand Down
1 change: 1 addition & 0 deletions backends/tfhe-cuda-backend/cuda/src/pbs/programmable_bootstrap_multibit.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -440,6 +440,7 @@ uint32_t get_lwe_chunk_size(uint32_t gpu_index, uint32_t max_num_pbs,

int max_blocks_per_sm;
int max_shared_memory = cuda_get_max_shared_memory(0);
cudaSetDevice(gpu_index);
if (max_shared_memory < full_sm_keybundle)
cudaOccupancyMaxActiveBlocksPerMultiprocessor(
&max_blocks_per_sm,
Expand Down
Loading