* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* 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 "transdata_kernel.h"
#include <memory>
#include <vector>
#include "formats/formats.h"
#include "formats/utils/formats_trans_utils.h"
#include "framework/common/framework_types_internal.h"
#include "host_kernels/kernel_utils.h"
#include "graph/utils/type_utils.h"
#include "host_kernels/kernel_factory.h"
namespace ge {
namespace {
const size_t kTransdataInputSize = 1;
}
Status TransdataKernel::ValidateInput(const OpDescPtr &op_desc_ptr, const std::vector<ConstGeTensorPtr> &input) const {
if (input.empty()) {
GELOGE(PARAM_INVALID, "Input tensor vector is empty");
return PARAM_INVALID;
}
ConstGeTensorPtr const_weight_ptr = input[0];
if (const_weight_ptr == nullptr) {
GELOGE(PARAM_INVALID, "Input const_weight_ptr is nullptr.");
return PARAM_INVALID;
}
if (op_desc_ptr == nullptr) {
GELOGE(PARAM_INVALID, "Input opDescPtr is nullptr.");
return PARAM_INVALID;
}
if (op_desc_ptr->GetInputsSize() != kTransdataInputSize) {
GELOGW("trans_op has more than 1 input_size.");
return NOT_CHANGED;
}
return SUCCESS;
}
Status TransdataKernel::Compute(const OpDescPtr op_desc_ptr, const std::vector<ConstGeTensorPtr> &input,
std::vector<GeTensorPtr> &v_output) {
GE_CHECK_NOTNULL(op_desc_ptr);
GELOGD("TransdataKernel begin");
Status status = ValidateInput(op_desc_ptr, input);
if (status != SUCCESS) {
return status;
}
ConstGeTensorPtr const_weight_ptr = input[0];
const auto &op_desc = op_desc_ptr->MutableOutputDesc(0);
const auto &op_desc_in = op_desc_ptr->MutableInputDesc(0);
GE_CHECK_NOTNULL(op_desc);
GE_CHECK_NOTNULL(op_desc_in);
const auto &src_format = op_desc_in->GetFormat();
const auto &src_shape = op_desc_in->GetShape().GetDims();
const auto &src_data_type = op_desc_in->GetDataType();
const auto &data_shape = op_desc->GetShape().GetDims();
const auto &data_format = op_desc->GetFormat();
const auto &data_type = op_desc->GetDataType();
const Format src_primary_format = static_cast<Format>(GetPrimaryFormat(src_format));
const Format dst_primary_format = static_cast<Format>(GetPrimaryFormat(data_format));
const Format src_sub_format = static_cast<Format>(GetSubFormat(src_format));
const Format dst_sub_format = static_cast<Format>(GetSubFormat(data_format));
const int64_t src_c0_format = GetC0Value(static_cast<int32_t>(src_format));
const int64_t dts_c0_format = GetC0Value(static_cast<int32_t>(data_format));
GELOGD(
"current node %s, format %s, primary format %s, input shape %s, data type %s, weight format %s, shape %s, "
"data type %s. output format %s, output primary format %s, shape %s, data type %s.",
op_desc_ptr->GetName().c_str(), TypeUtils::FormatToSerialString(src_format).c_str(),
TypeUtils::FormatToSerialString(src_primary_format).c_str(),
formats::ShapeToString(src_shape).c_str(), TypeUtils::DataTypeToSerialString(src_data_type).c_str(),
TypeUtils::FormatToSerialString(const_weight_ptr->GetTensorDesc().GetFormat()).c_str(),
formats::ShapeToString(const_weight_ptr->GetTensorDesc().GetShape()).c_str(),
TypeUtils::DataTypeToSerialString(const_weight_ptr->GetTensorDesc().GetDataType()).c_str(),
TypeUtils::FormatToSerialString(data_format).c_str(),
TypeUtils::FormatToSerialString(dst_primary_format).c_str(), formats::ShapeToString(data_shape).c_str(),
TypeUtils::DataTypeToSerialString(data_type).c_str());
const uint8_t *src_data = const_weight_ptr->GetData().data();
const formats::TransArgs trans_args{src_data, src_format, data_format, src_primary_format, dst_primary_format,
src_sub_format, dst_sub_format, src_c0_format, dts_c0_format, src_shape,
data_shape, src_data_type};
formats::TransResult trans_result;
GELOGD("Trans formats from %s to %s, primary formats from %s to %s, shape %s to %s, data type %s.",
TypeUtils::FormatToSerialString(src_format).c_str(), TypeUtils::FormatToSerialString(data_format).c_str(),
TypeUtils::FormatToSerialString(src_primary_format).c_str(),
TypeUtils::FormatToSerialString(dst_primary_format).c_str(),
formats::ShapeToString(src_shape).c_str(), formats::ShapeToString(data_shape).c_str(),
TypeUtils::DataTypeToSerialString(src_data_type).c_str());
if (src_data_type != data_type || data_shape.empty() || !formats::IsTransFormatSupport(trans_args)) {
GELOGW("Transfer from format %s to %s, primary formats from %s to %s, shape %s to %s, data type %s to %s "
"is not supported", TypeUtils::FormatToSerialString(src_format).c_str(),
TypeUtils::FormatToSerialString(data_format).c_str(),
TypeUtils::FormatToSerialString(src_primary_format).c_str(),
TypeUtils::FormatToSerialString(dst_primary_format).c_str(),
formats::ShapeToString(src_shape).c_str(), formats::ShapeToString(data_shape).c_str(),
TypeUtils::DataTypeToSerialString(src_data_type).c_str(),
TypeUtils::DataTypeToSerialString(data_type).c_str());
return NOT_CHANGED;
}
if (!KernelUtils::CheckSizeForTransOp(const_weight_ptr, op_desc_ptr)) {
GELOGI("CheckSize result is: input size is not equal to weight size");
return NOT_CHANGED;
}
if (formats::TransDataFormat(trans_args, trans_result) != SUCCESS) {
GELOGW("Failed to trans formats from %s to %s, primary formats from %s to %s, shape %s to %s, data type %s",
TypeUtils::FormatToSerialString(src_format).c_str(), TypeUtils::FormatToSerialString(data_format).c_str(),
TypeUtils::FormatToSerialString(src_primary_format).c_str(),
TypeUtils::FormatToSerialString(dst_primary_format).c_str(),
formats::ShapeToString(src_shape).c_str(), formats::ShapeToString(data_shape).c_str(),
TypeUtils::DataTypeToSerialString(src_data_type).c_str());
return NOT_CHANGED;
}
GeTensorPtr output_ptr = MakeShared<GeTensor>(op_desc_ptr->GetOutputDesc(0));
if (output_ptr == nullptr) {
GELOGE(ge::PARAM_INVALID, "Make shared failed");
return ge::PARAM_INVALID;
}
if (output_ptr->SetData(trans_result.data.get(), trans_result.length) != GRAPH_SUCCESS) {
GELOGW("Compute: SetData failed");
}
v_output.push_back(output_ptr);
return SUCCESS;
}
REGISTER_COMPUTE_NODE_KERNEL(TRANSDATA, TransdataKernel);
}
