Skip to content

Add cudaStream_t to deformable convolution kernels #9522

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

Add cudaStream_t to deformable convolution kernels #9522

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 @@ -236,6 +236,7 @@ void deformable_im2col(

const unsigned int threads = GET_THREADS();
const unsigned int blocks = GET_BLOCKS(threads, num_kernels);
const cudaStream_t stream = at::cuda::getCurrentCUDAStream();

// Checks if we should use 64bits indexing
// https://github.com/pytorch/vision/issues/4269
Expand All @@ -250,7 +251,7 @@ void deformable_im2col(
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 @@ -427,6 +428,7 @@ void compute_grad_input(

const unsigned int threads = GET_THREADS();
const unsigned int blocks = GET_BLOCKS(threads, num_kernels);
const cudaStream_t stream = at::cuda::getCurrentCUDAStream();

// Checks if we should use 64bits indexing
// https://github.com/pytorch/vision/issues/4269
Expand All @@ -439,7 +441,7 @@ void compute_grad_input(
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 @@ -668,6 +670,7 @@ void compute_grad_offset_and_mask(

const unsigned int threads = GET_THREADS();
const unsigned int blocks = GET_BLOCKS(threads, num_kernels);
const cudaStream_t stream = at::cuda::getCurrentCUDAStream();

// Checks if we should use 64bits indexing
// https://github.com/pytorch/vision/issues/4269
Expand All @@ -682,7 +685,7 @@ void compute_grad_offset_and_mask(
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