#include "op_plugin/AclOpsInterface.h"
#include "op_plugin/utils/OpAdapter.h"
namespace acl_op {
using npu_preparation = at_npu::native::OpPreparation;
using npu_compile_type = at_npu::native::CompileType;
using npu_utils = at_npu::native::NpuUtils;
namespace {
at::Tensor& argmax_out_nocheck(at::Tensor& result, const at::Tensor& input, at::Scalar& dim_scalar, bool keepdim_value)
{
at_npu::native::OpCommand cmd;
cmd.Name("ArgMaxV2")
.Input(input)
.Input(dim_scalar, at::kInt, npu_compile_type::MEMORY_HOST_COMPILE_DEPENDENT)
.Output(result)
.Attr("keep_dims", keepdim_value)
.Run();
return result;
}
}
at::Tensor& argmax_out(const at::Tensor& self, at::optional<int64_t> dim, bool keepdim, at::Tensor& out)
{
at::Tensor input = dim.has_value() ? self : self.reshape({-1});
int64_t dim_value = dim.has_value() ? dim.value() : 0;
bool keepdim_value = dim.has_value() ? keepdim : false;
auto output_size = op_infer::reduce_ops_npu_output_size(input, dim_value, keepdim_value);
npu_preparation::CheckOut(
{self},
out,
npu_preparation::get_tensor_npu_format(out),
at::kLong,
output_size);
at::Scalar dim_scalar = dim_value;
at::Tensor result_cast = at_npu::native::custom_ops::_npu_dtype_cast(out, at::kInt);
argmax_out_nocheck(result_cast, input, dim_scalar, keepdim_value);
if (!npu_utils::check_match(&out)) {
at::Tensor contiguous_result = at_npu::native::custom_ops::_npu_dtype_cast(result_cast, at::kLong);
npu_utils::format_fresh_view(out, contiguous_result);
} else {
out = at_npu::native::custom_ops::_npu_dtype_cast(result_cast, at::kLong);
}
return out;
}
}