Skip to content
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

Jump to bottom

Fix replication_pad for cuda launch configuration #50565

Closed
wants to merge 11 commits into from
Closed

Fix replication_pad for cuda launch configuration #50565

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
5be5f8f
replication pad 1d forward
xwang233 Jan 15, 2021
02296ab
test_replication_pad_large
xwang233 Jan 15, 2021
f1ed851
pl, pr
xwang233 Jan 15, 2021
9e82437
backward
xwang233 Jan 15, 2021
bdad5d7
fix for replication pad 2d
xwang233 Jan 15, 2021
e4b7b1c
test replication pad 2d
xwang233 Jan 15, 2021
2a98f21
fix 3d cuda launch configuration
xwang233 Jan 15, 2021
d701c38
test 3d
xwang233 Jan 15, 2021
0475869
Merge remote-tracking branch 'upstream/master' into replication_pad_f…
xwang233 Jan 19, 2021
c185745
use for loop instead of while loop
xwang233 Jan 19, 2021
2033ad5
static_cast int64_t for windows build problem
xwang233 Jan 19, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
272 changes: 140 additions & 132 deletions aten/src/ATen/native/cuda/ReplicationPadding.cu
View file
Open in desktop
Expand Up @@ -30,11 +30,11 @@ template <typename scalar_t>
__global__ void replication_pad_forward_kernel1d(
PackedTensorAccessor64<scalar_t, 3> input,
PackedTensorAccessor64<scalar_t, 3> output,
int padL, int padR) {
int padL, int padR, int y_shift, int z_shift) {

int outputPointId = threadIdx.x + blockIdx.x * blockDim.x;
int plane = blockIdx.y;
int batch = blockIdx.z;
int plane = blockIdx.y + y_shift;
int batch = blockIdx.z + z_shift;
if (outputPointId >= output.size(2)) {
return;
}
Expand All @@ -53,11 +53,11 @@ template <typename scalar_t>
__global__ void replication_pad_backward_kernel(
PackedTensorAccessor64<scalar_t, 3> gradInput,
PackedTensorAccessor64<scalar_t, 3> gradOutput,
int padL, int padR) {
int padL, int padR, int y_shift, int z_shift) {

int outputPointId = threadIdx.x + blockIdx.x * blockDim.x;
int plane = blockIdx.y;
int batch = blockIdx.z;
int plane = blockIdx.y + y_shift;
int batch = blockIdx.z + z_shift;
if (outputPointId >= gradOutput.size(2)) {
return;
}
Expand All @@ -76,11 +76,11 @@ template <typename scalar_t>
__global__ void replication_pad_forward_kernel2d(
PackedTensorAccessor64<scalar_t, 4> input,
PackedTensorAccessor64<scalar_t, 4> output,
int padT, int padB, int padL, int padR) {
int padT, int padB, int padL, int padR, int y_shift, int z_shift) {

int outputPointId = threadIdx.x + blockIdx.x * blockDim.x;
int plane = blockIdx.y;
int batch = blockIdx.z;
int plane = blockIdx.y + y_shift;
int batch = blockIdx.z + z_shift;
if (outputPointId >= output.size(2) * output.size(3)) {
return;
}
Expand All @@ -103,11 +103,11 @@ template <typename scalar_t>
__global__ void replication_pad_backward_kernel(
PackedTensorAccessor64<scalar_t, 4> gradInput,
PackedTensorAccessor64<scalar_t, 4> gradOutput,
int padT, int padB, int padL, int padR) {
int padT, int padB, int padL, int padR, int y_shift, int z_shift) {

int outputPointId = threadIdx.x + blockIdx.x * blockDim.x;
int plane = blockIdx.y;
int batch = blockIdx.z;
int plane = blockIdx.y + y_shift;
int batch = blockIdx.z + z_shift;
if (outputPointId >= gradOutput.size(2) * gradOutput.size(3)) {
return;
}
Expand All @@ -130,11 +130,11 @@ template <typename scalar_t>
__global__ void replication_pad_forward_kernel3d(
PackedTensorAccessor64<scalar_t, 5> input,
PackedTensorAccessor64<scalar_t, 5> output,
int pfront, int pback, int ptop, int pbottom, int pleft, int pright) {
int pfront, int pback, int ptop, int pbottom, int pleft, int pright, int y_shift, int z_shift) {

int outputPointId = threadIdx.x + blockIdx.x * blockDim.x;
int plane = blockIdx.y;
int batch = blockIdx.z;
int plane = blockIdx.y + y_shift;
int batch = blockIdx.z + z_shift;
if (outputPointId >= (output.size(2) * output.size(3) *
output.size(4))) {
return;
Expand Down Expand Up @@ -166,10 +166,10 @@ template <typename scalar_t>
__global__ void replication_pad_backward_kernel(
PackedTensorAccessor64<scalar_t, 5> gradInput,
PackedTensorAccessor64<scalar_t, 5> gradOutput,
int pfront, int pback, int ptop, int pbottom, int pleft, int pright) {
int pfront, int pback, int ptop, int pbottom, int pleft, int pright, int y_shift, int z_shift) {
int outputPointId = threadIdx.x + blockIdx.x * blockDim.x;
int plane = blockIdx.y;
int batch = blockIdx.z;
int plane = blockIdx.y + y_shift;
int batch = blockIdx.z + z_shift;

if (outputPointId >= (gradOutput.size(2) * gradOutput.size(3) *
gradOutput.size(4))) {
Expand Down Expand Up @@ -249,34 +249,32 @@ void replication_pad1d_out_cuda_template(

AT_DISPATCH_FLOATING_TYPES_AND_HALF(
input.scalar_type(), "replication_pad1d_cuda", [&] {
at::Tensor input_ = input;
at::Tensor output_ = output;
if (numInputDims == 2) {
auto input_ = input.unsqueeze(0);
auto output_ = output.unsqueeze(0);
auto devInput = input_.packed_accessor64<scalar_t, 3>();
auto devOutput = output_.packed_accessor64<scalar_t, 3>();

int outputPlaneSize = devOutput.size(2);
dim3 gridSize(THCCeilDiv(outputPlaneSize, 256),
devOutput.size(1),
devOutput.size(0));
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_forward_kernel1d <<<gridSize, blockSize, 0,
at::cuda::getCurrentCUDAStream()>>>(devInput, devOutput, padL, padR);
C10_CUDA_KERNEL_LAUNCH_CHECK();
} else {
auto devInput = input.packed_accessor64<scalar_t, 3>();
auto devOutput = output.packed_accessor64<scalar_t, 3>();

int outputPlaneSize = devOutput.size(2);
dim3 gridSize(THCCeilDiv(outputPlaneSize, 256),
devOutput.size(1),
devOutput.size(0));
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_forward_kernel1d <<<gridSize, blockSize, 0,
at::cuda::getCurrentCUDAStream()>>>(devInput, devOutput, padL, padR);
C10_CUDA_KERNEL_LAUNCH_CHECK();
input_ = input.unsqueeze(0);
output_ = output.unsqueeze(0);
}

auto devInput = input_.packed_accessor64<scalar_t, 3>();
auto devOutput = output_.packed_accessor64<scalar_t, 3>();

int64_t outputPlaneSize = devOutput.size(2);
int64_t size1 = devOutput.size(1);
int64_t size0 = devOutput.size(0);

for (int64_t block_y = 0; block_y < size1; block_y += 65535) {
int64_t block_y_size = std::min(size1 - block_y, static_cast<int64_t>(65535));
for (int64_t block_z = 0; block_z < size0; block_z += 65535) {
int64_t block_z_size = std::min(size0 - block_z, static_cast<int64_t>(65535));

dim3 gridSize(THCCeilDiv(outputPlaneSize, static_cast<int64_t>(256)), block_y_size, block_z_size);
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_forward_kernel1d <<<gridSize, blockSize, 0,
at::cuda::getCurrentCUDAStream()>>>(devInput, devOutput, padL, padR, block_y, block_z);
C10_CUDA_KERNEL_LAUNCH_CHECK();
}
}
}
);
Expand Down Expand Up @@ -330,16 +328,23 @@ void replication_pad1d_backward_out_cuda_template(
auto devGradInput = gradInput_.packed_accessor64<scalar_t, 3>();
auto devGradOutput = gradOutput_.packed_accessor64<scalar_t, 3>();

int outputPlaneSize = devGradOutput.size(2);
dim3 gridSize(THCCeilDiv(outputPlaneSize, 256),
devGradOutput.size(1),
devGradOutput.size(0));
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);
int64_t outputPlaneSize = devGradOutput.size(2);
int64_t size1 = devGradOutput.size(1);
int64_t size0 = devGradOutput.size(0);

for (int64_t block_y = 0; block_y < size1; block_y += 65535) {
int64_t block_y_size = std::min(size1 - block_y, static_cast<int64_t>(65535));
for (int64_t block_z = 0; block_z < size0; block_z += 65535) {
int64_t block_z_size = std::min(size0 - block_z, static_cast<int64_t>(65535));

replication_pad_backward_kernel <<<gridSize, blockSize, 0,
at::cuda::getCurrentCUDAStream()>>>(devGradInput, devGradOutput,
padL, padR);
C10_CUDA_KERNEL_LAUNCH_CHECK();
dim3 gridSize(THCCeilDiv(outputPlaneSize, static_cast<int64_t>(256)), block_y_size, block_z_size);
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_backward_kernel <<<gridSize, blockSize, 0, at::cuda::getCurrentCUDAStream()>>>(
devGradInput, devGradOutput, padL, padR, block_y, block_z);
C10_CUDA_KERNEL_LAUNCH_CHECK();
}
}
});
}

Expand Down Expand Up @@ -398,36 +403,31 @@ void replication_pad2d_out_cuda_template(

AT_DISPATCH_FLOATING_TYPES_AND_HALF(
input.scalar_type(), "replication_pad2d_cuda", [&] {
at::Tensor input_ = input;
at::Tensor output_ = output;
if (numInputDims == 3) {
auto input_ = input.unsqueeze(0);
auto output_ = output.unsqueeze(0);
auto devInput = input_.packed_accessor64<scalar_t, 4>();
auto devOutput = output_.packed_accessor64<scalar_t, 4>();

int outputPlaneSize = devOutput.size(2) * devOutput.size(3);
dim3 gridSize(THCCeilDiv(outputPlaneSize, 256),
devOutput.size(1),
devOutput.size(0));
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_forward_kernel2d <<<gridSize, blockSize, 0,
at::cuda::getCurrentCUDAStream()>>>(
devInput, devOutput, padT, padB, padL, padR);
C10_CUDA_KERNEL_LAUNCH_CHECK();
} else {
auto devInput = input.packed_accessor64<scalar_t, 4>();
auto devOutput = output.packed_accessor64<scalar_t, 4>();

int outputPlaneSize = devOutput.size(2) * devOutput.size(3);
dim3 gridSize(THCCeilDiv(outputPlaneSize, 256),
devOutput.size(1),
devOutput.size(0));
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_forward_kernel2d <<<gridSize, blockSize, 0,
at::cuda::getCurrentCUDAStream()>>>(devInput, devOutput,
padT, padB, padL, padR);
C10_CUDA_KERNEL_LAUNCH_CHECK();
input_ = input.unsqueeze(0);
output_ = output.unsqueeze(0);
}
auto devInput = input_.packed_accessor64<scalar_t, 4>();
auto devOutput = output_.packed_accessor64<scalar_t, 4>();

int64_t outputPlaneSize = devOutput.size(2) * devOutput.size(3);
int64_t size1 = devOutput.size(1);
int64_t size0 = devOutput.size(0);

for (int64_t block_y = 0; block_y < size1; block_y += 65535) {
int64_t block_y_size = std::min(size1 - block_y, static_cast<int64_t>(65535));
for (int64_t block_z = 0; block_z < size0; block_z += 65535) {
int64_t block_z_size = std::min(size0 - block_z, static_cast<int64_t>(65535));

dim3 gridSize(THCCeilDiv(outputPlaneSize, static_cast<int64_t>(256)), block_y_size, block_z_size);
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_forward_kernel2d <<<gridSize, blockSize, 0, at::cuda::getCurrentCUDAStream()>>>(
devInput, devOutput, padT, padB, padL, padR, block_y, block_z);
C10_CUDA_KERNEL_LAUNCH_CHECK();
}
}
}
);
Expand Down Expand Up @@ -490,15 +490,23 @@ void replication_pad2d_backward_out_cuda_template(
auto devGradInput = gradInput_.packed_accessor64<scalar_t, 4>();
auto devGradOutput = gradOutput_.packed_accessor64<scalar_t, 4>();

int outputPlaneSize = devGradOutput.size(2) * devGradOutput.size(3);
dim3 gridSize(THCCeilDiv(outputPlaneSize, 256),
devGradOutput.size(1),
devGradOutput.size(0));
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);
replication_pad_backward_kernel <<<gridSize, blockSize, 0,
at::cuda::getCurrentCUDAStream()>>>(devGradInput, devGradOutput,
padT, padB, padL, padR);
C10_CUDA_KERNEL_LAUNCH_CHECK();
int64_t outputPlaneSize = devGradOutput.size(2) * devGradOutput.size(3);
int64_t size1 = devGradOutput.size(1);
int64_t size0 = devGradOutput.size(0);

for (int64_t block_y = 0; block_y < size1; block_y += 65535) {
int64_t block_y_size = std::min(size1 - block_y, static_cast<int64_t>(65535));
for (int64_t block_z = 0; block_z < size0; block_z += 65535) {
int64_t block_z_size = std::min(size0 - block_z, static_cast<int64_t>(65535));

dim3 gridSize(THCCeilDiv(outputPlaneSize, static_cast<int64_t>(256)), block_y_size, block_z_size);
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_backward_kernel <<<gridSize, blockSize, 0, at::cuda::getCurrentCUDAStream()>>>(
devGradInput, devGradOutput, padT, padB, padL, padR, block_y, block_z);
C10_CUDA_KERNEL_LAUNCH_CHECK();
}
}
}
);
}
Expand Down Expand Up @@ -652,38 +660,32 @@ void replication_pad3d_out_cuda_template(

AT_DISPATCH_FLOATING_TYPES_AND_HALF(
input.scalar_type(), "replication_pad3d_cuda", [&] {
at::Tensor input_ = input;
at::Tensor output_ = output;
if (numInputDims == 4) {
auto input_ = input.unsqueeze(0);
auto output_ = output.unsqueeze(0);
auto devInput = input_.packed_accessor64<scalar_t, 5>();
auto devOutput = output_.packed_accessor64<scalar_t, 5>();

int outputPlaneSize = devOutput.size(2) * devOutput.size(3) *
devOutput.size(4);
dim3 gridSize(THCCeilDiv(outputPlaneSize, 256),
devOutput.size(1),
devOutput.size(0));
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_forward_kernel3d <<<gridSize, blockSize, 0,
at::cuda::getCurrentCUDAStream()>>>(
devInput, devOutput, pfront, pback, ptop, pbottom, pleft, pright);
C10_CUDA_KERNEL_LAUNCH_CHECK();
} else {
auto devInput = input.packed_accessor64<scalar_t, 5>();
auto devOutput = output.packed_accessor64<scalar_t, 5>();

int outputPlaneSize = devOutput.size(2) * devOutput.size(3) *
devOutput.size(4);
dim3 gridSize(THCCeilDiv(outputPlaneSize, 256),
devOutput.size(1),
devOutput.size(0));
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_forward_kernel3d <<<gridSize, blockSize, 0,
at::cuda::getCurrentCUDAStream()>>>(
devInput, devOutput, pfront, pback, ptop, pbottom, pleft, pright);
C10_CUDA_KERNEL_LAUNCH_CHECK();
}

auto devInput = input_.packed_accessor64<scalar_t, 5>();
auto devOutput = output_.packed_accessor64<scalar_t, 5>();

int64_t outputPlaneSize = devOutput.size(2) * devOutput.size(3) * devOutput.size(4);
int64_t size1 = devOutput.size(1);
int64_t size0 = devOutput.size(0);

for (int64_t block_y = 0; block_y < size1; block_y += 65535) {
int64_t block_y_size = std::min(size1 - block_y, static_cast<int64_t>(65535));
for (int64_t block_z = 0; block_z < size0; block_z += 65535) {
int64_t block_z_size = std::min(size0 - block_z, static_cast<int64_t>(65535));

dim3 gridSize(THCCeilDiv(outputPlaneSize, static_cast<int64_t>(256)), block_y_size, block_z_size);
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_forward_kernel3d <<<gridSize, blockSize, 0, at::cuda::getCurrentCUDAStream()>>>(
devInput, devOutput, pfront, pback, ptop, pbottom, pleft, pright, block_y, block_z);
C10_CUDA_KERNEL_LAUNCH_CHECK();
}
}
}
);
Expand Down Expand Up @@ -735,17 +737,23 @@ void replication_pad3d_backward_out_cuda_template(
auto devGradInput = gradInput_.packed_accessor64<scalar_t, 5>();
auto devGradOutput = gradOutput_.packed_accessor64<scalar_t, 5>();

int outputPlaneSize = devGradOutput.size(2) * devGradOutput.size(3) *
devGradOutput.size(4);
dim3 gridSize(THCCeilDiv(outputPlaneSize, 256),
devGradOutput.size(1),
devGradOutput.size(0));
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_backward_kernel <<<gridSize, blockSize, 0,
at::cuda::getCurrentCUDAStream()>>>(
devGradInput, devGradOutput, pfront, pback, ptop, pbottom, pleft, pright);
C10_CUDA_KERNEL_LAUNCH_CHECK();
int64_t outputPlaneSize = devGradOutput.size(2) * devGradOutput.size(3) * devGradOutput.size(4);
int64_t size1 = devGradOutput.size(1);
int64_t size0 = devGradOutput.size(0);

for (int64_t block_y = 0; block_y < size1; block_y += 65535) {
int64_t block_y_size = std::min(size1 - block_y, static_cast<int64_t>(65535));
for (int64_t block_z = 0; block_z < size0; block_z += 65535) {
int64_t block_z_size = std::min(size0 - block_z, static_cast<int64_t>(65535));

dim3 gridSize(THCCeilDiv(outputPlaneSize, static_cast<int64_t>(256)), block_y_size, block_z_size);
dim3 blockSize(outputPlaneSize > 256 ? 256 : outputPlaneSize);

replication_pad_backward_kernel <<<gridSize, blockSize, 0, at::cuda::getCurrentCUDAStream()>>>(
devGradInput, devGradOutput, pfront, pback, ptop, pbottom, pleft, pright, block_y, block_z);
C10_CUDA_KERNEL_LAUNCH_CHECK();
}
}
}
);
}
Expand Down