Use header files:

- Create header files for kernel implementation and remove definitions from vision_*.h files.
- Eliminate unnecessary headers and ensure all cpp include their headers.
- Move internal implementations in detail namespaces.

torchvision/csrc/cpu/deform_conv2d_cpu.cpp

@@ -66,15 +66,9 @@
 // modified from
 // https://github.com/open-mmlab/mmdetection/blob/master/mmdet/ops/dcn/src/deform_conv_cuda.cpp
 
-#include <ATen/ATen.h>
-#include <ATen/TensorUtils.h>
-#include <TH/TH.h>
+#include "deform_conv2d_cpu.h"
 
-#include <cmath>
-#include <iostream>
-#include <tuple>
-
-namespace {
+namespace detail {
 
 const int kMaxParallelImgs = 32;
 
@@ -851,7 +845,7 @@ at::Tensor backward_gradient_parameters(
   return grad_weight;
 }
 
-} // namespace
+} // namespace detail
 
 at::Tensor deform_conv2d_forward_cpu(
     const at::Tensor& input_param,
@@ -885,8 +879,8 @@ at::Tensor deform_conv2d_forward_cpu(
   int in_h = input.size(2);
   int in_w = input.size(3);
 
-  int n_parallel_imgs =
-      get_greatest_divisor_below_bound(batch_sz, kMaxParallelImgs);
+  int n_parallel_imgs = detail::get_greatest_divisor_below_bound(
+      batch_sz, detail::kMaxParallelImgs);
 
   // Unpack shapes and args
   int out_channels = weight.size(0);
@@ -1015,7 +1009,7 @@ at::Tensor deform_conv2d_forward_cpu(
       {n_in_channels * weight_h * weight_w, n_parallel_imgs * out_h * out_w},
       input.options());
   for (int b = 0; b < batch_sz / n_parallel_imgs; b++) {
-    deformable_im2col(
+    detail::deformable_im2col(
         input[b],
         offset[b],
         mask[b],
@@ -1086,10 +1080,10 @@ deform_conv2d_backward_cpu(
   at::Tensor bias = bias_param.contiguous();
 
   const int batch_sz = input.size(0);
-  const int n_parallel_imgs =
-      get_greatest_divisor_below_bound(batch_sz, kMaxParallelImgs);
+  const int n_parallel_imgs = detail::get_greatest_divisor_below_bound(
+      batch_sz, detail::kMaxParallelImgs);
 
-  auto grad_input_and_offset_and_mask = backward_gradient_inputs(
+  auto grad_input_and_offset_and_mask = detail::backward_gradient_inputs(
       input,
       weight,
       offset,
@@ -1110,7 +1104,7 @@ deform_conv2d_backward_cpu(
   auto grad_offset = std::get<1>(grad_input_and_offset_and_mask);
   auto grad_mask = std::get<2>(grad_input_and_offset_and_mask);
 
-  auto grad_weight = backward_gradient_parameters(
+  auto grad_weight = detail::backward_gradient_parameters(
       input,
       weight,
       offset,

torchvision/csrc/cpu/deform_conv2d_cpu.h

@@ -0,0 +1,39 @@
+#pragma once
+
+#include <ATen/ATen.h>
+#include "../macros.h"
+
+VISION_API at::Tensor deform_conv2d_forward_cpu(
+    const at::Tensor& input_param,
+    const at::Tensor& weight_param,
+    const at::Tensor& offset_param,
+    const at::Tensor& mask_param,
+    const at::Tensor& bias_param,
+    int64_t stride_h,
+    int64_t stride_w,
+    int64_t pad_h,
+    int64_t pad_w,
+    int64_t dil_h,
+    int64_t dil_w,
+    int64_t n_weight_grps,
+    int64_t n_offset_grps,
+    bool use_mask);
+
+VISION_API std::
+    tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor, at::Tensor>
+    deform_conv2d_backward_cpu(
+        const at::Tensor& grad_out_param,
+        const at::Tensor& input_param,
+        const at::Tensor& weight_param,
+        const at::Tensor& offset_param,
+        const at::Tensor& mask_param,
+        const at::Tensor& bias_param,
+        int64_t stride_h,
+        int64_t stride_w,
+        int64_t pad_h,
+        int64_t pad_w,
+        int64_t dil_h,
+        int64_t dil_w,
+        int64_t n_weight_grps,
+        int64_t n_offset_grps,
+        bool use_mask);

torchvision/csrc/cpu/vision_cpu.h

@@ -2,40 +2,7 @@
 #include <torch/extension.h>
 #include "../macros.h"
 
-VISION_API at::Tensor deform_conv2d_forward_cpu(
-    const at::Tensor& input_param,
-    const at::Tensor& weight_param,
-    const at::Tensor& offset_param,
-    const at::Tensor& mask_param,
-    const at::Tensor& bias_param,
-    int64_t stride_h,
-    int64_t stride_w,
-    int64_t pad_h,
-    int64_t pad_w,
-    int64_t dil_h,
-    int64_t dil_w,
-    int64_t n_weight_grps,
-    int64_t n_offset_grps,
-    bool use_mask);
-
-VISION_API std::
-    tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor, at::Tensor>
-    deform_conv2d_backward_cpu(
-        const at::Tensor& grad_out_param,
-        const at::Tensor& input_param,
-        const at::Tensor& weight_param,
-        const at::Tensor& offset_param,
-        const at::Tensor& mask_param,
-        const at::Tensor& bias_param,
-        int64_t stride_h,
-        int64_t stride_w,
-        int64_t pad_h,
-        int64_t pad_w,
-        int64_t dil_h,
-        int64_t dil_w,
-        int64_t n_weight_grps,
-        int64_t n_offset_grps,
-        bool use_mask);
+// TODO: Delete this file once all the methods are gone
 
 VISION_API at::Tensor nms_cpu(
     const at::Tensor& dets,

torchvision/csrc/cuda/deform_conv2d_cuda.cu

@@ -67,18 +67,14 @@
 // https://github.com/open-mmlab/mmdetection/blob/master/mmdet/ops/dcn/src/deform_conv_cuda.cpp
 
 #include <ATen/ATen.h>
-#include <ATen/TensorUtils.h>
 #include <ATen/cuda/CUDAContext.h>
 #include <c10/cuda/CUDAGuard.h>
 #include <THC/THCAtomics.cuh>
 
+#include "deform_conv2d_cuda.h"
 #include "cuda_helpers.h"
 
-#include <cmath>
-#include <iostream>
-#include <tuple>
-
-namespace {
+namespace detail {
 
 const int kMaxParallelImgs = 32;
 
@@ -898,7 +894,7 @@ at::Tensor backward_gradient_parameters(
   return grad_weight;
 }
 
-} // namespace
+} // namespace detail
 
 at::Tensor deform_conv2d_forward_cuda(
     const at::Tensor& input_param,
@@ -935,7 +931,7 @@ at::Tensor deform_conv2d_forward_cuda(
   int in_w = input.size(3);
 
   int n_parallel_imgs =
-      get_greatest_divisor_below_bound(batch_sz, kMaxParallelImgs);
+      detail::get_greatest_divisor_below_bound(batch_sz, detail::kMaxParallelImgs);
 
   int out_channels = weight.size(0);
   int weight_h = weight.size(2);
@@ -1063,7 +1059,7 @@ at::Tensor deform_conv2d_forward_cuda(
       {in_channels * weight_h * weight_w, n_parallel_imgs * out_h * out_w},
       input.options());
   for (int b = 0; b < batch_sz / n_parallel_imgs; b++) {
-    deformable_im2col(
+    detail::deformable_im2col(
         input[b],
         offset[b],
         mask[b],
@@ -1134,10 +1130,10 @@ deform_conv2d_backward_cuda(
   at::Tensor bias = bias_param.contiguous();
 
   const int batch_sz = input.size(0);
-  const int n_parallel_imgs =
-      get_greatest_divisor_below_bound(batch_sz, kMaxParallelImgs);
+  const int n_parallel_imgs = detail::get_greatest_divisor_below_bound(
+      batch_sz, detail::kMaxParallelImgs);
 
-  auto grad_input_and_offset_and_mask = backward_gradient_inputs(
+  auto grad_input_and_offset_and_mask = detail::backward_gradient_inputs(
       input,
       weight,
       offset,
@@ -1158,7 +1154,7 @@ deform_conv2d_backward_cuda(
   auto grad_offset = std::get<1>(grad_input_and_offset_and_mask);
   auto grad_mask = std::get<2>(grad_input_and_offset_and_mask);
 
-  auto grad_weight = backward_gradient_parameters(
+  auto grad_weight = detail::backward_gradient_parameters(
       input,
       weight,
       offset,

torchvision/csrc/cuda/deform_conv2d_cuda.h

@@ -0,0 +1,39 @@
+#pragma once
+
+#include <ATen/ATen.h>
+#include "../macros.h"
+
+VISION_API at::Tensor deform_conv2d_forward_cuda(
+    const at::Tensor& input_param,
+    const at::Tensor& weight_param,
+    const at::Tensor& offset_param,
+    const at::Tensor& mask_param,
+    const at::Tensor& bias_param,
+    int64_t stride_h,
+    int64_t stride_w,
+    int64_t pad_h,
+    int64_t pad_w,
+    int64_t dil_h,
+    int64_t dil_w,
+    int64_t n_weight_grps,
+    int64_t n_offset_grps,
+    bool use_mask);
+
+VISION_API std::
+    tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor, at::Tensor>
+    deform_conv2d_backward_cuda(
+        const at::Tensor& grad_out_param,
+        const at::Tensor& input_param,
+        const at::Tensor& weight_param,
+        const at::Tensor& offset_param,
+        const at::Tensor& mask_param,
+        const at::Tensor& bias_param,
+        int64_t stride_h,
+        int64_t stride_w,
+        int64_t pad_h,
+        int64_t pad_w,
+        int64_t dil_h,
+        int64_t dil_w,
+        int64_t n_weight_grps,
+        int64_t n_offset_grps,
+        bool use_mask);

torchvision/csrc/cuda/vision_cuda.h

@@ -2,40 +2,7 @@
 #include <torch/extension.h>
 #include "../macros.h"
 
-VISION_API at::Tensor deform_conv2d_forward_cuda(
-    const at::Tensor& input_param,
-    const at::Tensor& weight_param,
-    const at::Tensor& offset_param,
-    const at::Tensor& mask_param,
-    const at::Tensor& bias_param,
-    int64_t stride_h,
-    int64_t stride_w,
-    int64_t pad_h,
-    int64_t pad_w,
-    int64_t dil_h,
-    int64_t dil_w,
-    int64_t n_weight_grps,
-    int64_t n_offset_grps,
-    bool use_mask);
-
-VISION_API std::
-    tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor, at::Tensor>
-    deform_conv2d_backward_cuda(
-        const at::Tensor& grad_out_param,
-        const at::Tensor& input_param,
-        const at::Tensor& weight_param,
-        const at::Tensor& offset_param,
-        const at::Tensor& mask_param,
-        const at::Tensor& bias_param,
-        int64_t stride_h,
-        int64_t stride_w,
-        int64_t pad_h,
-        int64_t pad_w,
-        int64_t dil_h,
-        int64_t dil_w,
-        int64_t n_weight_grps,
-        int64_t n_offset_grps,
-        bool use_mask);
+// TODO: Delete this file once all the methods are gone
 
 VISION_API at::Tensor nms_cuda(
     const at::Tensor& dets,

torchvision/csrc/deform_conv2d.cpp

@@ -1,17 +1,10 @@
 #pragma once
 
-#include "cpu/vision_cpu.h"
-
-#ifdef WITH_CUDA
-#include "autocast.h"
-#include "cuda/vision_cuda.h"
-#endif
-#ifdef WITH_HIP
+#include "deform_conv2d.h"
+#include <torch/extension.h>
 #include "autocast.h"
-#include "hip/vision_cuda.h"
-#endif
 
-namespace {
+namespace detail {
 
 at::Tensor deform_conv2d(
     const at::Tensor& input,
@@ -261,7 +254,7 @@ class DeformConv2dBackwardFunction
   }
 };
 
-} // namespace
+} // namespace detail
 
 #if defined(WITH_CUDA) || defined(WITH_HIP)
 at::Tensor deform_conv2d_autocast(
@@ -280,7 +273,7 @@ at::Tensor deform_conv2d_autocast(
     int64_t offset_groups,
     bool use_mask) {
   c10::impl::ExcludeDispatchKeyGuard no_autocast(c10::DispatchKey::Autocast);
-  return deform_conv2d(
+  return detail::deform_conv2d(
              at::autocast::cached_cast(at::kFloat, input),
              at::autocast::cached_cast(at::kFloat, weight),
              at::autocast::cached_cast(at::kFloat, offset),
@@ -314,7 +307,7 @@ at::Tensor deform_conv2d_autograd(
     int64_t groups,
     int64_t offset_groups,
     bool use_mask) {
-  return DeformConv2dFunction::apply(
+  return detail::DeformConv2dFunction::apply(
       input,
       weight,
       offset,
@@ -348,7 +341,7 @@ deform_conv2d_backward_autograd(
     int64_t groups,
     int64_t offset_groups,
     bool use_mask) {
-  auto result = DeformConv2dBackwardFunction::apply(
+  auto result = detail::DeformConv2dBackwardFunction::apply(
       grad,
       input,
       weight,