#include "op_plugin/AclOpsInterface.h"
#include "op_plugin/utils/OpAdapter.h"
namespace acl_op {
using npu_preparation = at_npu::native::OpPreparation;
using npu_utils = at_npu::native::NpuUtils;
namespace {
int64_t adaptive_avg_pool3d_backward_safe_size(const at::Tensor& self)
{
c10::SmallVector<int64_t, N> dims = {-3, -2, -1};
int64_t size = 1;
if (self.sizes().empty()) {
return size;
}
for (int64_t ndim : dims) {
ndim = op_plugin::utils::make_warp_dim(ndim, self.sizes().size());
size *= self.sizes()[ndim];
}
return size;
}
at::Tensor& adaptive_avg_pool3d_backward_out_nocheck(
at::Tensor& result,
const at::Tensor& grad_output,
const at::Tensor& self)
{
TORCH_CHECK(grad_output.size(grad_output.dim() - 3) == 1 && grad_output.size(grad_output.dim() - 2) == 1 &&
grad_output.size(grad_output.dim() - 1) == 1,
"adaptive_avg_pool3d_backward only support D=1 && H=1 && W=1 current!" + OPS_ERROR(ErrCode::PARAM));
acl_op::fill_(result, 1.0 / adaptive_avg_pool3d_backward_safe_size(self));
acl_op::mul_(result, grad_output);
return result;
}
}
at::Tensor& adaptive_avg_pool3d_backward_out(
const at::Tensor& grad_output,
const at::Tensor& self,
at::Tensor& grad_input)
{
npu_preparation::CheckOut(
{grad_output, self},
grad_input,
self);
if (!npu_utils::check_match(&grad_input)) {
at::Tensor contiguous_result = npu_utils::format_contiguous(grad_input);
adaptive_avg_pool3d_backward_out_nocheck(contiguous_result, grad_output, self);
npu_utils::format_fresh_view(grad_input, contiguous_result);
} else {
adaptive_avg_pool3d_backward_out_nocheck(grad_input, grad_output, self);
}
return grad_input;
}
at::Tensor _adaptive_avg_pool3d_backward(const at::Tensor& grad_output, const at::Tensor& self)
{
at::Tensor result = npu_preparation::apply_tensor(self);
adaptive_avg_pool3d_backward_out_nocheck(result, grad_output, self);
return result;
}
}