github/tfhe-rs
1
1
Fork 0
mirror of https://github.com/zama-ai/tfhe-rs.git synced 2026-01-08 14:23:53 -05:00
Code Issues Packages Projects Releases Wiki Activity

fix(gpu): No broadcast is needed because full prop is done on 1 single GPU

Browse Source
This commit is contained in:
Enzo Di Maria
2025-12-03 16:48:12 +01:00
committed by Agnès Leroy
parent 0e1b064182
commit a96d68323b
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
backends/tfhe-cuda-backend/cuda/include/integer/integer_utilities.h
View File

@@ -1101,8 +1101,9 @@ template <typename Torus> struct int_fullprop_buffer {
cuda_memcpy_with_size_tracking_async_to_gpu(
lwe_indexes, h_lwe_indexes, lwe_indexes_size, streams.stream(0),
streams.gpu_index(0), allocate_gpu_memory);
auto active_streams = streams.active_gpu_subset(2);
lut->broadcast_lut(active_streams);
//
// No broadcast is needed because full prop is done on 1 single GPU.
//
tmp_small_lwe_vector = new CudaRadixCiphertextFFI;
create_zero_radix_ciphertext_async<Torus>(