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(gpu): fix full prop with 1 block #1444

Merged
merged 1 commit into from
Aug 2, 2024
Merged

fix(gpu): fix full prop with 1 block #1444

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
15 changes: 7 additions & 8 deletions backends/tfhe-cuda-backend/cuda/include/integer.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -84,9 +84,8 @@ void scratch_cuda_full_propagation_64(
void **streams, uint32_t *gpu_indexes, uint32_t gpu_count, int8_t **mem_ptr,
uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
uint32_t ks_level, uint32_t ks_base_log, uint32_t pbs_level,
uint32_t pbs_base_log, uint32_t grouping_factor, uint32_t num_radix_blocks,
uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
bool allocate_gpu_memory);
uint32_t pbs_base_log, uint32_t grouping_factor, uint32_t message_modulus,
uint32_t carry_modulus, PBS_TYPE pbs_type, bool allocate_gpu_memory);

void cuda_full_propagation_64_inplace(void **streams, uint32_t *gpu_indexes,
uint32_t gpu_count, void *input_blocks,
Expand Down Expand Up @@ -1035,10 +1034,10 @@ template <typename Torus> struct int_fullprop_buffer {

int_fullprop_buffer(cudaStream_t *streams, uint32_t *gpu_indexes,
uint32_t gpu_count, int_radix_params params,
uint32_t num_radix_blocks, bool allocate_gpu_memory) {
bool allocate_gpu_memory) {
this->params = params;
lut = new int_radix_lut<Torus>(streams, gpu_indexes, 1, params, 2,
num_radix_blocks, allocate_gpu_memory);
lut = new int_radix_lut<Torus>(streams, gpu_indexes, 1, params, 2, 2,
allocate_gpu_memory);

if (allocate_gpu_memory) {

Expand All @@ -1064,9 +1063,9 @@ template <typename Torus> struct int_fullprop_buffer {
params.polynomial_size, params.message_modulus, params.carry_modulus,
lut_f_carry);

Torus lwe_indexes_size = num_radix_blocks * sizeof(Torus);
Torus lwe_indexes_size = 2 * sizeof(Torus);
Torus *h_lwe_indexes = (Torus *)malloc(lwe_indexes_size);
for (int i = 0; i < num_radix_blocks; i++)
for (int i = 0; i < 2; i++)
h_lwe_indexes[i] = i;
Torus *lwe_indexes = lut->get_lut_indexes(gpu_indexes[0], 0);
cuda_memcpy_async_to_gpu(lwe_indexes, h_lwe_indexes, lwe_indexes_size,
Expand Down
8 changes: 3 additions & 5 deletions backends/tfhe-cuda-backend/cuda/src/integer/integer.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,18 +19,16 @@ void scratch_cuda_full_propagation_64(
void **streams, uint32_t *gpu_indexes, uint32_t gpu_count, int8_t **mem_ptr,
uint32_t lwe_dimension, uint32_t glwe_dimension, uint32_t polynomial_size,
uint32_t ks_level, uint32_t ks_base_log, uint32_t pbs_level,
uint32_t pbs_base_log, uint32_t grouping_factor, uint32_t num_radix_blocks,
uint32_t message_modulus, uint32_t carry_modulus, PBS_TYPE pbs_type,
bool allocate_gpu_memory) {
uint32_t pbs_base_log, uint32_t grouping_factor, uint32_t message_modulus,
uint32_t carry_modulus, PBS_TYPE pbs_type, bool allocate_gpu_memory) {
int_radix_params params(pbs_type, glwe_dimension, polynomial_size,
glwe_dimension * polynomial_size, lwe_dimension,
ks_level, ks_base_log, pbs_level, pbs_base_log,
grouping_factor, message_modulus, carry_modulus);

scratch_cuda_full_propagation<uint64_t>(
(cudaStream_t *)streams, gpu_indexes, gpu_count,
(int_fullprop_buffer<uint64_t> **)mem_ptr, params, num_radix_blocks,
allocate_gpu_memory);
(int_fullprop_buffer<uint64_t> **)mem_ptr, params, allocate_gpu_memory);
}

void cleanup_cuda_full_propagation(void **streams, uint32_t *gpu_indexes,
Expand Down
7 changes: 2 additions & 5 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 @@ -672,7 +672,6 @@ void host_full_propagate_inplace(cudaStream_t *streams, uint32_t *gpu_indexes,
for (int i = 0; i < num_blocks; i++) {
auto cur_input_block = &input_blocks[i * big_lwe_size];

cudaSetDevice(gpu_indexes[0]);
/// Since the keyswitch is done on one input only, use only 1 GPU
execute_keyswitch_async<Torus>(
streams, gpu_indexes, 1, mem_ptr->tmp_small_lwe_vector,
Expand Down Expand Up @@ -713,12 +712,10 @@ void scratch_cuda_full_propagation(cudaStream_t *streams, uint32_t *gpu_indexes,
uint32_t gpu_count,
int_fullprop_buffer<Torus> **mem_ptr,
int_radix_params params,
uint32_t num_radix_blocks,
bool allocate_gpu_memory) {

*mem_ptr =
new int_fullprop_buffer<Torus>(streams, gpu_indexes, gpu_count, params,
num_radix_blocks, allocate_gpu_memory);
*mem_ptr = new int_fullprop_buffer<Torus>(streams, gpu_indexes, gpu_count,
params, allocate_gpu_memory);
}

// (lwe_dimension+1) threads
Expand Down
1 change: 0 additions & 1 deletion backends/tfhe-cuda-backend/src/cuda_bind.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -669,7 +669,6 @@ extern "C" {
pbs_level: u32,
pbs_base_log: u32,
grouping_factor: u32,
num_blocks: u32,
message_modulus: u32,
carry_modulus: u32,
pbs_type: u32,
Expand Down
1 change: 0 additions & 1 deletion tfhe/src/integer/gpu/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -828,7 +828,6 @@ pub unsafe fn full_propagate_assign_async<T: UnsignedInteger, B: Numeric>(
pbs_level.0 as u32,
pbs_base_log.0 as u32,
grouping_factor.0 as u32,
num_blocks,
message_modulus.0 as u32,
carry_modulus.0 as u32,
pbs_type as u32,
Expand Down
Loading