Skip to content

Fix deform_conv2d kernels to use current CUDA stream #9515

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

Fix deform_conv2d kernels to use current CUDA stream #9515

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: 9 additions & 6 deletions torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -247,10 +247,11 @@ void deformable_im2col(
out_w >
std::numeric_limits<int32_t>::max());

const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
if (use_64bits_indexing) {
AT_DISPATCH_FLOATING_TYPES_AND_HALF(
input.scalar_type(), "deformable_im2col", ([&] {
deformable_im2col_kernel<scalar_t, int64_t><<<blocks, threads>>>(
deformable_im2col_kernel<scalar_t, int64_t><<<blocks, threads, 0, stream>>>(
num_kernels,
input.data_ptr<scalar_t>(),
data_offset.data_ptr<scalar_t>(),
Expand All @@ -277,7 +278,7 @@ void deformable_im2col(
} else {
AT_DISPATCH_FLOATING_TYPES_AND_HALF(
input.scalar_type(), "deformable_im2col", ([&] {
deformable_im2col_kernel<scalar_t, int><<<blocks, threads>>>(
deformable_im2col_kernel<scalar_t, int><<<blocks, threads, 0, stream>>>(
num_kernels,
input.data_ptr<scalar_t>(),
data_offset.data_ptr<scalar_t>(),
Expand Down Expand Up @@ -436,10 +437,11 @@ void compute_grad_input(

at::globalContext().alertNotDeterministic("compute_grad_input");

const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
if (use_64bits_indexing) {
AT_DISPATCH_FLOATING_TYPES_AND_HALF(
columns.scalar_type(), "compute_grad_input", ([&] {
deformable_col2im_kernel<scalar_t, int64_t><<<blocks, threads>>>(
deformable_col2im_kernel<scalar_t, int64_t><<<blocks, threads, 0, stream>>>(
num_kernels,
columns.data_ptr<scalar_t>(),
offset.data_ptr<scalar_t>(),
Expand All @@ -465,7 +467,7 @@ void compute_grad_input(
} else {
AT_DISPATCH_FLOATING_TYPES_AND_HALF(
columns.scalar_type(), "compute_grad_input", ([&] {
deformable_col2im_kernel<scalar_t, int><<<blocks, threads>>>(
deformable_col2im_kernel<scalar_t, int><<<blocks, threads, 0, stream>>>(
num_kernels,
columns.data_ptr<scalar_t>(),
offset.data_ptr<scalar_t>(),
Expand Down Expand Up @@ -678,11 +680,12 @@ void compute_grad_offset_and_mask(
((int64_t)channels * weight_h * weight_w * parallel_imgs * out_h * out_w >
std::numeric_limits<int32_t>::max());

const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
if (use_64bits_indexing) {
AT_DISPATCH_FLOATING_TYPES_AND_HALF(
columns.scalar_type(), "compute_grad_offset_and_mask", ([&] {
deformable_col2im_coord_kernel<scalar_t, int64_t>
<<<blocks, threads>>>(
<<<blocks, threads, 0, stream>>>(
num_kernels,
columns.data_ptr<scalar_t>(),
input.data_ptr<scalar_t>(),
Expand Down Expand Up @@ -711,7 +714,7 @@ void compute_grad_offset_and_mask(
} else {
AT_DISPATCH_FLOATING_TYPES_AND_HALF(
columns.scalar_type(), "compute_grad_offset_and_mask", ([&] {
deformable_col2im_coord_kernel<scalar_t, int><<<blocks, threads>>>(
deformable_col2im_coord_kernel<scalar_t, int><<<blocks, threads, 0, stream>>>(
num_kernels,
columns.data_ptr<scalar_t>(),
input.data_ptr<scalar_t>(),
Expand Down