Skip to content

GitLab

Unverified Commit ae87c1e4 authored by Vasilis Vryniotis's avatar Vasilis Vryniotis Committed by GitHub
Browse files

Update to clang-format 11. (#3254)

parent 7bf6e7b1
Hide whitespace changes
Inline Side-by-side
Showing with 233 additions and 195 deletions
torchvision/csrc/ops/autograd/ps_roi_align_kernel.cpp
View file @ ae87c1e4
...@@ -62,12 +62,13 @@ class PSROIAlignFunction ...@@ -62,12 +62,13 @@ class PSROIAlignFunction
input_shape[2], input_shape[2],
input_shape[3]); input_shape[3]);
return {grad_in, return {
torch::autograd::Variable(), grad_in,
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable()}; torch::autograd::Variable(),
torch::autograd::Variable()};
} }
}; };
......
torchvision/csrc/ops/autograd/ps_roi_pool_kernel.cpp
View file @ ae87c1e4
...@@ -53,11 +53,12 @@ class PSROIPoolFunction : public torch::autograd::Function<PSROIPoolFunction> { ...@@ -53,11 +53,12 @@ class PSROIPoolFunction : public torch::autograd::Function<PSROIPoolFunction> {
input_shape[2], input_shape[2],
input_shape[3]); input_shape[3]);
return {grad_in, return {
torch::autograd::Variable(), grad_in,
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable()}; torch::autograd::Variable(),
torch::autograd::Variable()};
} }
}; };
......
torchvision/csrc/ops/autograd/roi_align_kernel.cpp
View file @ ae87c1e4
...@@ -57,13 +57,14 @@ class ROIAlignFunction : public torch::autograd::Function<ROIAlignFunction> { ...@@ -57,13 +57,14 @@ class ROIAlignFunction : public torch::autograd::Function<ROIAlignFunction> {
input_shape[3], input_shape[3],
ctx->saved_data["sampling_ratio"].toInt(), ctx->saved_data["sampling_ratio"].toInt(),
ctx->saved_data["aligned"].toBool()); ctx->saved_data["aligned"].toBool());
return {grad_in, return {
torch::autograd::Variable(), grad_in,
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable()}; torch::autograd::Variable(),
torch::autograd::Variable()};
} }
}; };
......
torchvision/csrc/ops/autograd/roi_pool_kernel.cpp
View file @ ae87c1e4
...@@ -53,11 +53,12 @@ class ROIPoolFunction : public torch::autograd::Function<ROIPoolFunction> { ...@@ -53,11 +53,12 @@ class ROIPoolFunction : public torch::autograd::Function<ROIPoolFunction> {
input_shape[2], input_shape[2],
input_shape[3]); input_shape[3]);
return {grad_in, return {
torch::autograd::Variable(), grad_in,
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable(), torch::autograd::Variable(),
torch::autograd::Variable()}; torch::autograd::Variable(),
torch::autograd::Variable()};
} }
}; };
......
torchvision/csrc/ops/cpu/deform_conv2d_kernel.cpp
View file @ ae87c1e4
...@@ -634,44 +634,50 @@ std::tuple<at::Tensor, at::Tensor, at::Tensor> backward_gradient_inputs( ...@@ -634,44 +634,50 @@ std::tuple<at::Tensor, at::Tensor, at::Tensor> backward_gradient_inputs(
input = input.reshape( input = input.reshape(
{batch_sz / n_parallel_imgs, n_parallel_imgs, n_in_channels, in_h, in_w}); {batch_sz / n_parallel_imgs, n_parallel_imgs, n_in_channels, in_h, in_w});
grad_offset = grad_offset.reshape({batch_sz / n_parallel_imgs, grad_offset = grad_offset.reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * 2 * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * 2 * weight_h * weight_w,
out_w}); out_h,
offset = offset.reshape({batch_sz / n_parallel_imgs, out_w});
n_parallel_imgs, offset = offset.reshape(
n_offset_grps * 2 * weight_h * weight_w, {batch_sz / n_parallel_imgs,
out_h, n_parallel_imgs,
out_w}); n_offset_grps * 2 * weight_h * weight_w,
out_h,
out_w});
if (use_mask) { if (use_mask) {
grad_mask = grad_mask.reshape({batch_sz / n_parallel_imgs, grad_mask = grad_mask.reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * weight_h * weight_w,
out_w}); out_h,
mask = mask.reshape({batch_sz / n_parallel_imgs, out_w});
n_parallel_imgs, mask = mask.reshape(
n_offset_grps * weight_h * weight_w, {batch_sz / n_parallel_imgs,
out_h, n_parallel_imgs,
out_w}); n_offset_grps * weight_h * weight_w,
out_h,
out_w});
} }
grad_out = grad_out grad_out = grad_out
.reshape({batch_sz / n_parallel_imgs, .reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_weight_grps, n_parallel_imgs,
n_out_channels / n_weight_grps, n_weight_grps,
out_h, n_out_channels / n_weight_grps,
out_w}) out_h,
out_w})
.permute({0, 2, 3, 1, 4, 5}); .permute({0, 2, 3, 1, 4, 5});
weight = weight.reshape({n_weight_grps, weight = weight.reshape(
weight.size(0) / n_weight_grps, {n_weight_grps,
weight.size(1), weight.size(0) / n_weight_grps,
weight.size(2), weight.size(1),
weight.size(3)}); weight.size(2),
weight.size(3)});
columns = columns.view( columns = columns.view(
{n_weight_grps, columns.size(0) / n_weight_grps, columns.size(1)}); {n_weight_grps, columns.size(0) / n_weight_grps, columns.size(1)});
...@@ -775,37 +781,41 @@ at::Tensor backward_gradient_parameters( ...@@ -775,37 +781,41 @@ at::Tensor backward_gradient_parameters(
} }
at::Tensor grad_out_buf = grad_out at::Tensor grad_out_buf = grad_out
.reshape({batch_sz / n_parallel_imgs, .reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_weight_grps, n_parallel_imgs,
n_out_channels / n_weight_grps, n_weight_grps,
out_h, n_out_channels / n_weight_grps,
out_w}) out_h,
out_w})
.permute({0, 2, 3, 1, 4, 5}) .permute({0, 2, 3, 1, 4, 5})
.contiguous(); .contiguous();
input = input.reshape( input = input.reshape(
{batch_sz / n_parallel_imgs, n_parallel_imgs, n_in_channels, in_h, in_w}); {batch_sz / n_parallel_imgs, n_parallel_imgs, n_in_channels, in_h, in_w});
offset = offset.reshape({batch_sz / n_parallel_imgs, offset = offset.reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * 2 * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * 2 * weight_h * weight_w,
out_w}); out_h,
out_w});
if (use_mask) { if (use_mask) {
mask = mask.reshape({batch_sz / n_parallel_imgs, mask = mask.reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * weight_h * weight_w,
out_w}); out_h,
out_w});
} }
grad_weight = grad_weight.view({n_weight_grps, grad_weight = grad_weight.view(
grad_weight.size(0) / n_weight_grps, {n_weight_grps,
grad_weight.size(1), grad_weight.size(0) / n_weight_grps,
grad_weight.size(2), grad_weight.size(1),
grad_weight.size(3)}); grad_weight.size(2),
grad_weight.size(3)});
auto columns = at::empty( auto columns = at::empty(
{n_weight_grps, {n_weight_grps,
...@@ -846,10 +856,11 @@ at::Tensor backward_gradient_parameters( ...@@ -846,10 +856,11 @@ at::Tensor backward_gradient_parameters(
} }
} }
grad_weight = grad_weight.view({grad_weight.size(0) * grad_weight.size(1), grad_weight = grad_weight.view(
grad_weight.size(2), {grad_weight.size(0) * grad_weight.size(1),
grad_weight.size(3), grad_weight.size(2),
grad_weight.size(4)}); grad_weight.size(3),
grad_weight.size(4)});
return grad_weight; return grad_weight;
} }
...@@ -976,26 +987,29 @@ at::Tensor deform_conv2d_forward_kernel( ...@@ -976,26 +987,29 @@ at::Tensor deform_conv2d_forward_kernel(
} }
// Separate batches into blocks // Separate batches into blocks
out = out.view({batch_sz / n_parallel_imgs, out = out.view(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
out_channels, n_parallel_imgs,
out_h, out_channels,
out_w}); out_h,
out_w});
input_c = input_c.view( input_c = input_c.view(
{batch_sz / n_parallel_imgs, n_parallel_imgs, n_in_channels, in_h, in_w}); {batch_sz / n_parallel_imgs, n_parallel_imgs, n_in_channels, in_h, in_w});
offset_c = offset_c.view({batch_sz / n_parallel_imgs, offset_c = offset_c.view(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * 2 * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * 2 * weight_h * weight_w,
out_w}); out_h,
out_w});
if (use_mask) { if (use_mask) {
mask_c = mask_c.view({batch_sz / n_parallel_imgs, mask_c = mask_c.view(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * weight_h * weight_w,
out_w}); out_h,
out_w});
} }
at::Tensor out_buf = at::zeros( at::Tensor out_buf = at::zeros(
...@@ -1006,16 +1020,18 @@ at::Tensor deform_conv2d_forward_kernel( ...@@ -1006,16 +1020,18 @@ at::Tensor deform_conv2d_forward_kernel(
out.options()); out.options());
// Separate channels into convolution groups // Separate channels into convolution groups
out_buf = out_buf.view({out_buf.size(0), out_buf = out_buf.view(
n_weight_grps, {out_buf.size(0),
out_buf.size(1) / n_weight_grps, n_weight_grps,
out_buf.size(2), out_buf.size(1) / n_weight_grps,
out_buf.size(3)}); out_buf.size(2),
weight_c = weight_c.view({n_weight_grps, out_buf.size(3)});
weight_c.size(0) / n_weight_grps, weight_c = weight_c.view(
weight_c.size(1), {n_weight_grps,
weight_c.size(2), weight_c.size(0) / n_weight_grps,
weight_c.size(3)}); weight_c.size(1),
weight_c.size(2),
weight_c.size(3)});
// Sample points and perform convolution // Sample points and perform convolution
auto columns = at::zeros( auto columns = at::zeros(
...@@ -1056,11 +1072,12 @@ at::Tensor deform_conv2d_forward_kernel( ...@@ -1056,11 +1072,12 @@ at::Tensor deform_conv2d_forward_kernel(
columns.view({columns.size(0) * columns.size(1), columns.size(2)}); columns.view({columns.size(0) * columns.size(1), columns.size(2)});
} }
out_buf = out_buf.view({batch_sz / n_parallel_imgs, out_buf = out_buf.view(
out_channels, {batch_sz / n_parallel_imgs,
n_parallel_imgs, out_channels,
out_h, n_parallel_imgs,
out_w}); out_h,
out_w});
out_buf.transpose_(1, 2); out_buf.transpose_(1, 2);
out.copy_(out_buf); out.copy_(out_buf);
out = out.view({batch_sz, out_channels, out_h, out_w}); out = out.view({batch_sz, out_channels, out_h, out_w});
......
torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu
View file @ ae87c1e4
...@@ -677,44 +677,50 @@ std::tuple<at::Tensor, at::Tensor, at::Tensor> backward_gradient_inputs( ...@@ -677,44 +677,50 @@ std::tuple<at::Tensor, at::Tensor, at::Tensor> backward_gradient_inputs(
input = input.reshape( input = input.reshape(
{batch_sz / n_parallel_imgs, n_parallel_imgs, n_in_channels, in_h, in_w}); {batch_sz / n_parallel_imgs, n_parallel_imgs, n_in_channels, in_h, in_w});
grad_offset = grad_offset.reshape({batch_sz / n_parallel_imgs, grad_offset = grad_offset.reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * 2 * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * 2 * weight_h * weight_w,
out_w}); out_h,
offset = offset.reshape({batch_sz / n_parallel_imgs, out_w});
n_parallel_imgs, offset = offset.reshape(
n_offset_grps * 2 * weight_h * weight_w, {batch_sz / n_parallel_imgs,
out_h, n_parallel_imgs,
out_w}); n_offset_grps * 2 * weight_h * weight_w,
out_h,
out_w});
if (use_mask) { if (use_mask) {
grad_mask = grad_mask.reshape({batch_sz / n_parallel_imgs, grad_mask = grad_mask.reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * weight_h * weight_w,
out_w}); out_h,
mask = mask.reshape({batch_sz / n_parallel_imgs, out_w});
n_parallel_imgs, mask = mask.reshape(
n_offset_grps * weight_h * weight_w, {batch_sz / n_parallel_imgs,
out_h, n_parallel_imgs,
out_w}); n_offset_grps * weight_h * weight_w,
out_h,
out_w});
} }
grad_out = grad_out grad_out = grad_out
.reshape({batch_sz / n_parallel_imgs, .reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_weight_grps, n_parallel_imgs,
n_out_channels / n_weight_grps, n_weight_grps,
out_h, n_out_channels / n_weight_grps,
out_w}) out_h,
out_w})
.permute({0, 2, 3, 1, 4, 5}); .permute({0, 2, 3, 1, 4, 5});
weight = weight.reshape({n_weight_grps, weight = weight.reshape(
weight.size(0) / n_weight_grps, {n_weight_grps,
weight.size(1), weight.size(0) / n_weight_grps,
weight.size(2), weight.size(1),
weight.size(3)}); weight.size(2),
weight.size(3)});
columns = columns.view( columns = columns.view(
{n_weight_grps, columns.size(0) / n_weight_grps, columns.size(1)}); {n_weight_grps, columns.size(0) / n_weight_grps, columns.size(1)});
...@@ -819,37 +825,41 @@ at::Tensor backward_gradient_parameters( ...@@ -819,37 +825,41 @@ at::Tensor backward_gradient_parameters(
} }
at::Tensor grad_out_buf = grad_out at::Tensor grad_out_buf = grad_out
.reshape({batch_sz / n_parallel_imgs, .reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_weight_grps, n_parallel_imgs,
n_out_channels / n_weight_grps, n_weight_grps,
out_h, n_out_channels / n_weight_grps,
out_w}) out_h,
out_w})
.permute({0, 2, 3, 1, 4, 5}) .permute({0, 2, 3, 1, 4, 5})
.contiguous(); .contiguous();
input = input.reshape( input = input.reshape(
{batch_sz / n_parallel_imgs, n_parallel_imgs, n_in_channels, in_h, in_w}); {batch_sz / n_parallel_imgs, n_parallel_imgs, n_in_channels, in_h, in_w});
offset = offset.reshape({batch_sz / n_parallel_imgs, offset = offset.reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * 2 * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * 2 * weight_h * weight_w,
out_w}); out_h,
out_w});
if (use_mask) { if (use_mask) {
mask = mask.reshape({batch_sz / n_parallel_imgs, mask = mask.reshape(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * weight_h * weight_w,
out_w}); out_h,
out_w});
} }
grad_weight = grad_weight.reshape({n_weight_grps, grad_weight = grad_weight.reshape(
grad_weight.size(0) / n_weight_grps, {n_weight_grps,
grad_weight.size(1), grad_weight.size(0) / n_weight_grps,
grad_weight.size(2), grad_weight.size(1),
grad_weight.size(3)}); grad_weight.size(2),
grad_weight.size(3)});
auto columns = at::empty( auto columns = at::empty(
{n_weight_grps, {n_weight_grps,
...@@ -890,10 +900,11 @@ at::Tensor backward_gradient_parameters( ...@@ -890,10 +900,11 @@ at::Tensor backward_gradient_parameters(
} }
} }
grad_weight = grad_weight.view({grad_weight.size(0) * grad_weight.size(1), grad_weight = grad_weight.view(
grad_weight.size(2), {grad_weight.size(0) * grad_weight.size(1),
grad_weight.size(3), grad_weight.size(2),
grad_weight.size(4)}); grad_weight.size(3),
grad_weight.size(4)});
return grad_weight; return grad_weight;
} }
...@@ -1021,26 +1032,29 @@ at::Tensor deform_conv2d_forward_kernel( ...@@ -1021,26 +1032,29 @@ at::Tensor deform_conv2d_forward_kernel(
} }
// Separate batches into blocks // Separate batches into blocks
out = out.view({batch_sz / n_parallel_imgs, out = out.view(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
out_channels, n_parallel_imgs,
out_h, out_channels,
out_w}); out_h,
out_w});
input_c = input_c.view( input_c = input_c.view(
{batch_sz / n_parallel_imgs, n_parallel_imgs, in_channels, in_h, in_w}); {batch_sz / n_parallel_imgs, n_parallel_imgs, in_channels, in_h, in_w});
offset_c = offset_c.view({batch_sz / n_parallel_imgs, offset_c = offset_c.view(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * 2 * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * 2 * weight_h * weight_w,
out_w}); out_h,
out_w});
if (use_mask) { if (use_mask) {
mask_c = mask_c.view({batch_sz / n_parallel_imgs, mask_c = mask_c.view(
n_parallel_imgs, {batch_sz / n_parallel_imgs,
n_offset_grps * weight_h * weight_w, n_parallel_imgs,
out_h, n_offset_grps * weight_h * weight_w,
out_w}); out_h,
out_w});
} }
at::Tensor out_buf = at::zeros( at::Tensor out_buf = at::zeros(
...@@ -1051,16 +1065,18 @@ at::Tensor deform_conv2d_forward_kernel( ...@@ -1051,16 +1065,18 @@ at::Tensor deform_conv2d_forward_kernel(
out.options()); out.options());
// Separate channels into convolution groups // Separate channels into convolution groups
out_buf = out_buf.view({out_buf.size(0), out_buf = out_buf.view(
n_weight_grps, {out_buf.size(0),
out_buf.size(1) / n_weight_grps, n_weight_grps,
out_buf.size(2), out_buf.size(1) / n_weight_grps,
out_buf.size(3)}); out_buf.size(2),
weight_c = weight_c.view({n_weight_grps, out_buf.size(3)});
weight_c.size(0) / n_weight_grps, weight_c = weight_c.view(
weight_c.size(1), {n_weight_grps,
weight_c.size(2), weight_c.size(0) / n_weight_grps,
weight_c.size(3)}); weight_c.size(1),
weight_c.size(2),
weight_c.size(3)});
// Sample points and perform convolution // Sample points and perform convolution
auto columns = at::zeros( auto columns = at::zeros(
...@@ -1101,11 +1117,12 @@ at::Tensor deform_conv2d_forward_kernel( ...@@ -1101,11 +1117,12 @@ at::Tensor deform_conv2d_forward_kernel(
columns.view({columns.size(0) * columns.size(1), columns.size(2)}); columns.view({columns.size(0) * columns.size(1), columns.size(2)});
} }
out_buf = out_buf.view({batch_sz / n_parallel_imgs, out_buf = out_buf.view(
out_channels, {batch_sz / n_parallel_imgs,
n_parallel_imgs, out_channels,
out_h, n_parallel_imgs,
out_w}); out_h,
out_w});
out_buf.transpose_(1, 2); out_buf.transpose_(1, 2);
out.copy_(out_buf); out.copy_(out_buf);
out = out.view({batch_sz, out_channels, out_h, out_w}); out = out.view({batch_sz, out_channels, out_h, out_w});
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment