#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 {
at::Tensor binary_cross_entropy_with_logits_nocheck(
const at::Tensor& self,
const at::Tensor& target,
const at::Tensor& weight,
const at::Tensor& pos_weight,
int64_t reduction) {
at::IntArrayRef output_size;
int64_t result_format = npu_preparation::get_tensor_npu_format(self);
if (reduction == at::Reduction::None) {
output_size = self.sizes();
} else {
output_size = at::ArrayRef<int64_t>();
result_format = ACL_FORMAT_ND;
}
at::Tensor result = npu_preparation::apply_tensor_with_format(output_size, self.options(), result_format);
at::Tensor weight_tensor;
if (weight.defined()) {
weight_tensor = npu_utils::format_contiguous(weight);
weight_tensor = (weight.scalar_type() != self.scalar_type()) ?
at_npu::native::custom_ops::_npu_dtype_cast(weight_tensor, self.scalar_type()) : weight_tensor;
} else {
weight_tensor = at::ones(self.sizes(), self.options());
}
at::Tensor pos_weight_tensor;
if (pos_weight.defined()) {
pos_weight_tensor = npu_utils::format_contiguous(pos_weight);
pos_weight_tensor = (pos_weight_tensor.scalar_type() != self.scalar_type()) ?
at_npu::native::custom_ops::_npu_dtype_cast(pos_weight_tensor, self.scalar_type()) : pos_weight_tensor;
} else {
pos_weight_tensor = at::ones(self.sizes(), self.options());
}
std::string reduction_str = op_plugin::utils::get_reduction_str(reduction);
at_npu::native::OpCommand cmd;
cmd.Name("SigmoidCrossEntropyWithLogitsV2")
.Input(self.to(target.dtype()))
.Input(target)
.Input(weight_tensor)
.Input(pos_weight_tensor)
.Output(result)
.Attr("reduction", reduction_str)
.Run();
return result;
}
}
at::Tensor npu_binary_cross_entropy_with_logits_backward(
const at::Tensor& grad_output,
const at::Tensor& self,
const at::Tensor& target,
const c10::optional<at::Tensor>& weight_opt,
const c10::optional<at::Tensor>& pos_weight_opt,
int64_t reduction) {
at::Tensor grad_input = npu_preparation::apply_tensor(self);
const at::Tensor& weight = c10::value_or_else(weight_opt, [] {return at::Tensor();});
const at::Tensor& pos_weight = c10::value_or_else(pos_weight_opt, [] {return at::Tensor();});
at::Tensor weight_tensor;
if (weight.defined()) {
weight_tensor = npu_utils::format_contiguous(weight);
weight_tensor = (weight_tensor.scalar_type() != self.scalar_type()) ?
at_npu::native::custom_ops::_npu_dtype_cast(weight_tensor, self.scalar_type()) : weight_tensor;
} else {
weight_tensor = at::ones(self.sizes(), self.options());
}
at::Tensor pos_weight_tensor;
if (pos_weight.defined()) {
pos_weight_tensor = npu_utils::format_contiguous(pos_weight);
pos_weight_tensor = (pos_weight_tensor.scalar_type() != self.scalar_type()) ?
at_npu::native::custom_ops::_npu_dtype_cast(pos_weight_tensor, self.scalar_type()) : pos_weight_tensor;
} else {
pos_weight_tensor = at::ones(self.sizes(), self.options());
}
at::Tensor dout_tensor = acl_op::npu_broadcast(grad_output, self.sizes());
std::string reduction_str = op_plugin::utils::get_reduction_str(reduction);
at_npu::native::OpCommand cmd;
cmd.Name("SigmoidCrossEntropyWithLogitsGradV2")
.Input(self)
.Input(target)
.Input(dout_tensor)
.Input(weight_tensor)
.Input(pos_weight_tensor)
.Output(grad_input)
.Attr("reduction", reduction_str)
.Run();
return grad_input;
}
at::Tensor binary_cross_entropy_with_logits(
const at::Tensor& self,
const at::Tensor& target,
const c10::optional<at::Tensor>& weight_opt,
const c10::optional<at::Tensor>& pos_weight_opt,
int64_t reduction) {
const at::Tensor& weight = c10::value_or_else(weight_opt, [] {return at::Tensor();});
const at::Tensor& pos_weight = c10::value_or_else(pos_weight_opt, [] {return at::Tensor();});
return binary_cross_entropy_with_logits_nocheck(self, target, weight, pos_weight, reduction);
}
}