Skip to content

Avoid aten::select calls in fbgemm::permute_multi_embedding #5017

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

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Avoid aten::select calls in fbgemm::permute_multi_embedding #5017

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
24 changes: 15 additions & 9 deletions fbgemm_gpu/src/permute_multi_embedding_ops/permute_multi_embedding_ops_cpu.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,8 +41,10 @@ std::vector<Tensor> permute_multi_embedding_function_cpu(
pooled_embs[0].scalar_type(), "permute_multi_embs_cpu", [&] {
at::parallel_for(0, B, 0, [&](int32_t start, int32_t end) {
int32_t in_tensor, out_tensor, in_offset, out_offset, length, next;
auto stride = permutes.stride(0);
for (const auto i : c10::irange(permutes.size(0))) {
int32_t* __restrict__ pp = permutes[i].data_ptr<int32_t>();
int32_t* __restrict__ pp =
permutes.data_ptr<int32_t>() + i * stride;
if (reverse_permute) {
out_tensor = pp[PermuteParam::in_tensor];
in_tensor = pp[PermuteParam::out_tensor];
Expand All @@ -58,20 +60,24 @@ std::vector<Tensor> permute_multi_embedding_function_cpu(
length = pp[PermuteParam::length];
if (reverse_permute && next < 0) {
for (auto b : c10::irange(start, end)) {
auto outp =
outputs[out_tensor][b].data_ptr<scalar_t>() + out_offset;
auto inp =
inputs[in_tensor][b].data_ptr<scalar_t>() + in_offset;
auto out_stride = outputs[out_tensor].stride(0);
auto outp = outputs[out_tensor].data_ptr<scalar_t>() +
out_stride * b + out_offset;
auto inp_stride = inputs[in_tensor].stride(0);
auto inp = inputs[in_tensor].data_ptr<scalar_t>() +
inp_stride * b + in_offset;
for (const auto j : c10::irange(length)) {
outp[j] += inp[j];
}
}
} else {
for (auto b : c10::irange(start, end)) {
auto outp =
outputs[out_tensor][b].data_ptr<scalar_t>() + out_offset;
auto inp =
inputs[in_tensor][b].data_ptr<scalar_t>() + in_offset;
auto out_stride = outputs[out_tensor].stride(0);
auto outp = outputs[out_tensor].data_ptr<scalar_t>() +
out_stride * b + out_offset;
auto inp_stride = inputs[in_tensor].stride(0);
auto inp = inputs[in_tensor].data_ptr<scalar_t>() +
inp_stride * b + in_offset;
std::memcpy(outp, inp, length * pooled_embs[0].itemsize());
}
}
Expand Down
Loading