|
#include <torch/extension.h>
|
|
|
|
|
|
torch::Tensor upfirdn2d_op(const torch::Tensor& input, const torch::Tensor& kernel,
|
|
int up_x, int up_y, int down_x, int down_y,
|
|
int pad_x0, int pad_x1, int pad_y0, int pad_y1);
|
|
|
|
#define CHECK_CUDA(x) TORCH_CHECK(x.type().is_cuda(), #x " must be a CUDA tensor")
|
|
#define CHECK_CONTIGUOUS(x) TORCH_CHECK(x.is_contiguous(), #x " must be contiguous")
|
|
#define CHECK_INPUT(x) CHECK_CUDA(x); CHECK_CONTIGUOUS(x)
|
|
|
|
torch::Tensor upfirdn2d(const torch::Tensor& input, const torch::Tensor& kernel,
|
|
int up_x, int up_y, int down_x, int down_y,
|
|
int pad_x0, int pad_x1, int pad_y0, int pad_y1) {
|
|
CHECK_CUDA(input);
|
|
CHECK_CUDA(kernel);
|
|
|
|
return upfirdn2d_op(input, kernel, up_x, up_y, down_x, down_y, pad_x0, pad_x1, pad_y0, pad_y1);
|
|
}
|
|
|
|
PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
|
|
m.def("upfirdn2d", &upfirdn2d, "upfirdn2d (CUDA)");
|
|
} |