* 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 <memory>
#include "opdev/framework_op.h"
#include "opdev/aicpu/aicpu_task.h"
#include "opdev/make_op_executor.h"
#include "opdev/op_dfx.h"
#include "opdev/platform.h"
#include "op_api/aclnn_check.h"
namespace l0op {
OP_TYPE_REGISTER(TensorMove);
static const int64_t DATA_LIMIT_910 = 200000 * 4;
static const int64_t DATA_LIMIT_910B = 100000 * 4;
static const int64_t DATA_LIMIT_310P = 32;
static const std::initializer_list<op::DataType> AICORE_DTYPE_SUPPORT_LIST_REGBASE = {
op::DataType::DT_FLOAT, op::DataType::DT_FLOAT16, op::DataType::DT_INT32, op::DataType::DT_UINT32,
op::DataType::DT_INT8, op::DataType::DT_INT16, op::DataType::DT_UINT16, op::DataType::DT_UINT8,
op::DataType::DT_BOOL, op::DataType::DT_INT64, op::DataType::DT_UINT64, op::DataType::DT_BF16,
op::DataType::DT_DOUBLE, op::DataType::DT_HIFLOAT8, op::DataType::DT_FLOAT8_E5M2, op::DataType::DT_FLOAT8_E4M3FN,
op::DataType::DT_COMPLEX32, op::DataType::DT_COMPLEX64};
static const std::initializer_list<op::DataType> AICORE_DTYPE_SUPPORT_LIST_910B = {
op::DataType::DT_FLOAT, op::DataType::DT_FLOAT16, op::DataType::DT_INT32, op::DataType::DT_UINT32,
op::DataType::DT_INT8, op::DataType::DT_INT16, op::DataType::DT_UINT16, op::DataType::DT_UINT8,
op::DataType::DT_BOOL, op::DataType::DT_INT64, op::DataType::DT_UINT64, op::DataType::DT_BF16,
op::DataType::DT_DOUBLE};
static const std::initializer_list<op::DataType> AICORE_DTYPE_SUPPORT_LIST = {
op::DataType::DT_FLOAT, op::DataType::DT_FLOAT16, op::DataType::DT_INT32, op::DataType::DT_UINT32,
op::DataType::DT_INT8, op::DataType::DT_INT16, op::DataType::DT_UINT16, op::DataType::DT_UINT8,
op::DataType::DT_BOOL, op::DataType::DT_INT64, op::DataType::DT_UINT64, op::DataType::DT_DOUBLE};
static bool IsAiCoreSupport(const aclTensor* self)
{
auto dataType = self->GetDataType();
auto curArch = op::GetCurrentPlatformInfo().GetCurNpuArch();
if (curArch == NpuArch::DAV_2201) {
return op::CheckType(dataType, AICORE_DTYPE_SUPPORT_LIST_910B);
}
if (op::IsRegBase(curArch)) {
return op::CheckType(dataType, AICORE_DTYPE_SUPPORT_LIST_REGBASE);
}
return op::CheckType(dataType, AICORE_DTYPE_SUPPORT_LIST);
}
const aclTensor* TensorMoveAiCore(const aclTensor* x, const aclTensor* y, aclOpExecutor* executor)
{
L0_DFX(TensorMoveAiCore, x, y);
ADD_TO_LAUNCHER_LIST_AICORE(TensorMove, op::AI_CORE, OP_INPUT(x), OP_OUTPUT(y));
return y;
}
const aclTensor* TensorMoveAiCpu(const aclTensor* x, const aclTensor* y, aclOpExecutor* executor)
{
L0_DFX(TensorMoveAiCpu, x, y);
static op::internal::AicpuTaskSpace space("Identity");
ADD_TO_LAUNCHER_LIST_AICPU(TensorMove, OP_ATTR_NAMES(), OP_INPUT(x), OP_OUTPUT(y));
return y;
}
bool IsCopyNpuToNpu(const aclTensor* x)
{
auto dataSize = static_cast<uint64_t>(x->GetViewShape().GetShapeSize()) * op::TypeSize(x->GetDataType());
auto socVersion = op::GetCurrentPlatformInfo().GetSocVersion();
if (socVersion == op::SocVersion::ASCEND910B || socVersion == op::SocVersion::ASCEND910_93
|| op::IsRegBase()) {
return static_cast<int64_t>(dataSize) <= DATA_LIMIT_910B;
}
if (socVersion == op::SocVersion::ASCEND910) {
return static_cast<int64_t>(dataSize) <= DATA_LIMIT_910;
}
if (socVersion == op::SocVersion::ASCEND310P) {
return static_cast<int64_t>(dataSize) < DATA_LIMIT_310P;
}
return false;
}
const aclTensor* TensorMove(const aclTensor* x, const aclTensor* y, aclOpExecutor* executor)
{
if (IsCopyNpuToNpu(x)) {
auto status = op::CopyNpuToNpu(x, y, executor);
if (status == ACLNN_SUCCESS) {
return y;
}
}
if (IsAiCoreSupport(x)) {
return TensorMoveAiCore(x, y, executor);
}
return TensorMoveAiCpu(x, y, executor);
}
}