#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& ge_out_nocheck(at::Tensor& result, const at::Tensor& self, const at::Tensor& other) {
auto unified_result = npu_preparation::comparison_op_check(result, self, other, true);
at_npu::native::OpCommand cmd;
cmd.Name("GreaterEqual")
.Expect(unified_result)
.Input(self)
.Input(other)
.Output(result)
.Run();
return result;
}
at::Tensor& ge_out_nocheck(at::Tensor& result, const at::Tensor& self, at::Scalar other) {
at_npu::native::OpCommand cmd;
cmd.Name("GreaterEqual")
.Input(self)
.Input(other, self.scalar_type())
.Output(result)
.Run();
return result;
}
at::ScalarType get_ge_calculate_type(const at::Tensor& self, const at::Tensor& other) {
at::ScalarType calculate_type = at::native::result_type(self, other);
if (calculate_type == at::kInt || calculate_type == at::kBool) {
calculate_type = at::kFloat;
}
return calculate_type;
}
at::ScalarType get_ge_calculate_type(const at::Tensor& self, const at::Scalar& other) {
at::ScalarType calculate_type = at::native::result_type(self, other);
if (calculate_type == at::kInt || calculate_type == at::kBool) {
calculate_type = at::kFloat;
}
return calculate_type;
}
}
at::Tensor& ge_out(const at::Tensor& self, const at::Tensor& other, at::Tensor& result) {
if (npu_preparation::IsCPUScalar(other)) {
return acl_op::ge_out(self, other.item(), result);
} else if (npu_preparation::IsCPUScalar(self)) {
return acl_op::le_out(other, self.item(), result);
} else {
auto output_size = op_infer::broadcast_ops_npu_output_size(self, other);
npu_preparation::CheckOut(
{self, other},
result,
result,
output_size);
TORCH_CHECK(self.device() == other.device(),
"Expected all tensors to be on the same device, but found at least two devices, ",
self.device(), " and ", other.device(),
OPS_ERROR(ErrCode::PARAM));
auto calculate_type = get_ge_calculate_type(self, other);
auto self_cast = op_plugin::utils::get_cast_input(self, calculate_type);
auto other_cast = op_plugin::utils::get_cast_input(other, calculate_type);
auto result_type = result.scalar_type();
at::Tensor result_cast = (result_type != at::kBool) ?
at_npu::native::custom_ops::_npu_dtype_cast(result, at::kBool) : result;
if (!npu_utils::check_match(&result_cast)) {
at::Tensor contiguous_result = npu_utils::format_contiguous(result_cast);
ge_out_nocheck(contiguous_result, self_cast, other_cast);
npu_utils::format_fresh_view(result_cast, contiguous_result);
} else {
ge_out_nocheck(result_cast, self_cast, other_cast);
}
if (result_type != at::kBool) {
result_cast = at_npu::native::custom_ops::_npu_dtype_cast(result_cast, result_type);
result.copy_(result_cast);
}
return result;
}
}
at::Tensor& ge_out(const at::Tensor& self, const at::Scalar& other, at::Tensor& result) {
auto calculate_type = get_ge_calculate_type(self, other);
auto self_cast = op_plugin::utils::get_cast_input(self, calculate_type);
npu_preparation::CheckOut(
{self_cast},
result,
result,
self.sizes());
auto result_type = result.scalar_type();
at::Tensor result_cast = (result_type != at::kBool) ?
at_npu::native::custom_ops::_npu_dtype_cast(result, at::kBool) : result;
if (!npu_utils::check_match(&result_cast)) {
at::Tensor contiguous_result = npu_utils::format_contiguous(result_cast);
ge_out_nocheck(contiguous_result, self_cast, other);
npu_utils::format_fresh_view(result_cast, contiguous_result);
} else {
ge_out_nocheck(result_cast, self_cast, other);
}
if (result_type != at::kBool) {
result_cast = at_npu::native::custom_ops::_npu_dtype_cast(result_cast, result_type);
result.copy_(result_cast);
}
return result;
}
at::Tensor ge(const at::Tensor& self, const at::Tensor& other) {
if (npu_preparation::IsCPUScalar(other)) {
return acl_op::ge(self, other.item());
} else if (npu_preparation::IsCPUScalar(self)) {
return acl_op::le(other, self.item());
} else {
TORCH_CHECK(self.device() == other.device(),
"Expected all tensors to be on the same device, but found at least two devices, ",
self.device(), " and ", other.device(),
OPS_ERROR(ErrCode::PARAM));
auto calculate_type = get_ge_calculate_type(self, other);
auto self_cast = op_plugin::utils::get_cast_input(self, calculate_type);
auto other_cast = op_plugin::utils::get_cast_input(other, calculate_type);
auto output_size = op_infer::broadcast_ops_npu_output_size(self_cast, other_cast);
at::Tensor result = npu_preparation::apply_tensor(output_size, self_cast.options().dtype(at::kBool), self_cast);
ge_out_nocheck(result, self_cast, other_cast);
return result;
}
}
at::Tensor ge(const at::Tensor& self, const at::Scalar& other) {
auto calculate_type = get_ge_calculate_type(self, other);
auto self_cast = op_plugin::utils::get_cast_input(self, calculate_type);
at::Tensor result = npu_preparation::apply_tensor(self, self.options().dtype(at::kBool));
ge_out_nocheck(result, self_cast, other);
return result;
}
at::Tensor& ge_(at::Tensor& self, const at::Tensor& other) {
return acl_op::ge_out(self, other, self);
}
at::Tensor& ge_(at::Tensor& self, const at::Scalar& other) {
return acl_op::ge_out(self, other, self);
}
}