Skip to content

Update to batch processing delta update for kvzch and fix table idx issue. #5148

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

Closed
wants to merge 1 commit into from
Closed

Update to batch processing delta update for kvzch and fix table idx issue. #5148

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
44 changes: 27 additions & 17 deletions fbgemm_gpu/src/embedding_inplace_ops/embedding_inplace_update_cpu.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -156,27 +156,37 @@ void dram_kv_embedding_inplace_update_cpu(
const int64_t* update_row_idx_ptr = update_row_idx.data_ptr<int64_t>();
const int64_t* update_offsets_ptr = update_offsets.data_ptr<int64_t>();

for (int64_t n = 0; n < N; ++n) {
int32_t t = update_table_idx_ptr[n];
int64_t row_idx = update_row_idx_ptr[n];
SparseType weight_ty = static_cast<SparseType>(weights_tys_ptr[t]);
int32_t D_start = D_offsets_ptr[t];
int32_t D_end = D_offsets_ptr[t + 1];
int64_t window_start = 0;
while (window_start < N) {
int32_t cur_table = update_table_idx_ptr[window_start];
int64_t window_end = window_start;
while (window_end < N && update_table_idx_ptr[window_end] == cur_table) {
++window_end;
}
int window_size = window_end - window_start;

SparseType weight_ty = static_cast<SparseType>(weights_tys_ptr[0]);
int32_t D_start = D_offsets_ptr[0];
int32_t D_end = D_offsets_ptr[1];
int32_t D = D_end - D_start;
int32_t D_bytes =
nbit::padded_row_size_in_bytes(D, weight_ty, row_alignment);

int64_t update_weight_offset = update_offsets_ptr[n];
const uint8_t* update_weight_row =
update_weights_ptr + update_weight_offset;
std::vector<uint8_t> tmp(update_weight_row, update_weight_row + D_bytes);
at::Tensor update_weight =
at::from_blob(
tmp.data(), {1, D_bytes}, at::TensorOptions().dtype(at::kByte))
.clone();
at::Tensor row_id =
at::full({1}, row_idx, at::TensorOptions().dtype(at::kLong));
(*embedding_inplace_update_method)({t, row_id, update_weight});
uint8_t* batched_weights_ptr = const_cast<uint8_t*>(
update_weights_ptr + update_offsets_ptr[window_start]);
auto weights_tensor = at::from_blob(
batched_weights_ptr,
{window_size, D_bytes},
at::TensorOptions().dtype(at::kByte));

int64_t* row_ids_ptr =
const_cast<int64_t*>(update_row_idx_ptr + window_start);
auto row_id_tensor = at::from_blob(
row_ids_ptr, {window_size}, at::TensorOptions().dtype(at::kLong));

(*embedding_inplace_update_method)(
{cur_table, row_id_tensor, weights_tensor});
window_start = window_end;
}
if (embedding_log_inplace_update_stats_method.has_value()) {
(*embedding_log_inplace_update_stats_method)({});
Expand Down
Loading