Skip to content

fix test_lookup_table_op unittests #21

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
windstamp wants to merge 1 commit into from
Closed

fix test_lookup_table_op unittests #21

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
29 changes: 29 additions & 0 deletions paddle/fluid/operators/lookup_table_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,9 +105,26 @@ class LookupTableCUDAKernel : public framework::OpKernel<T> {
auto *table = table_t->data<T>();
auto *output = output_t->mutable_data<T>(context.GetPlace());

#ifdef PADDLE_WITH_HIP
dim3 threads(64, 4);
#else
dim3 threads(128, 8);
#endif // PADDLE_WITH_HIP

dim3 grids(8, 1);

#ifdef PADDLE_WITH_HIP
if (padding_idx == -1)
LookupTable<
T, 64, 4, 8,
false><<<grids, threads, 0, context.cuda_device_context().stream()>>>(
output, table, ids, N, K, D, padding_idx);
else
LookupTable<
T, 64, 4, 8,
true><<<grids, threads, 0, context.cuda_device_context().stream()>>>(
output, table, ids, N, K, D, padding_idx);
#else
if (padding_idx == -1)
LookupTable<
T, 128, 8, 8,
Expand All @@ -118,6 +135,7 @@ class LookupTableCUDAKernel : public framework::OpKernel<T> {
T, 128, 8, 8,
true><<<grids, threads, 0, context.cuda_device_context().stream()>>>(
output, table, ids, N, K, D, padding_idx);
#endif // PADDLE_WITH_HIP
}
};

Expand Down Expand Up @@ -185,10 +203,21 @@ class LookupTableGradCUDAKernel : public framework::OpKernel<T> {
auto t = framework::EigenVector<T>::Flatten(*d_table_t);
t.device(*dev_ctx.eigen_device()) = t.constant(static_cast<T>(0));

#ifdef PADDLE_WITH_HIP
dim3 threads(64, 4);
#else
dim3 threads(128, 8);
#endif // PADDLE_WITH_HIP

dim3 grids(8, 1);

#ifdef PADDLE_WITH_HIP
LookupTableGrad<T, 64, 4, 8><<<grids, threads, 0, dev_ctx.stream()>>>(
d_table, d_output, ids, N, K, D);
#else
LookupTableGrad<T, 128, 8, 8><<<grids, threads, 0, dev_ctx.stream()>>>(
d_table, d_output, ids, N, K, D);
#endif // PADDLE_WITH_HIP
}
}
};
Expand Down