* Copyright (c) 2026 Huawei Technologies Co., Ltd. All Rights Reserved.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
#include "tensorflow/core/framework/bounds_check.h"
#include "tensorflow/core/framework/op_kernel.h"
#include "tensorflow/core/framework/register_types.h"
#include "tf_adapter/common/adapter_logger.h"
namespace tensorflow {
template <typename T>
class LarsOp : public OpKernel {
public:
explicit LarsOp(OpKernelConstruction *context) : OpKernel(context) {
ADP_LOG(INFO) << "new LarsOp";
}
~LarsOp() override = default;
void Compute(OpKernelContext *context) override {
int32_t input_num = num_inputs();
ADP_LOG(INFO) << "LarsOp: input num " << input_num;
input_num = ((input_num - 1) / 2);
for (int32_t j = 0; j < input_num; j++) {
const Tensor &w_tensor = context->input(j);
auto w_input = w_tensor.flat<T>();
const Tensor &g_tensor = context->input(j + input_num);
auto g_input = g_tensor.flat<T>();
Tensor *output_tensor = nullptr;
OP_REQUIRES_OK(context, context->allocate_output(j, w_tensor.shape(), &output_tensor));
auto output_flat = output_tensor->flat<T>();
const int32_t N = static_cast<int32_t>(w_input.size());
ADP_LOG(INFO) << "LarsOp idx " << j << ", data num " << N;
auto sum_w = w_input(0);
auto sum_g = g_input(0);
for (int32_t i = 1; i < N; i++) {
auto w = w_input(i);
sum_w += w;
ADP_LOG(INFO) << "LarsOp w " << w << ", sum_w " << sum_w;
auto g = g_input(i);
sum_g += g;
ADP_LOG(INFO) << "LarsOp g " << g << ", sum_g " << sum_g;
}
auto w_norm = sqrt(sum_w);
auto g_norm = sqrt(sum_g);
auto b = g_norm + w_norm + T(0.00001);
for (int32_t i = 1; i < N; i++) {
auto w = w_input(i);
auto g = g_input(i);
output_flat(i) = b * (g + w);
}
}
ADP_LOG(INFO) << "in LarsOp";
}
bool IsExpensive() override {
return false;
}
};
REGISTER_KERNEL_BUILDER(Name("LARS").Device(DEVICE_CPU).TypeConstraint<float>("T"), LarsOp<float>);
}