* 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 "pack.h"
#include "opdev/aicpu/aicpu_task.h"
#include "opdev/data_type_utils.h"
#include "opdev/format_utils.h"
#include "opdev/make_op_executor.h"
#include "opdev/op_def.h"
#include "opdev/op_dfx.h"
#include "opdev/op_executor.h"
#include "opdev/op_log.h"
#include "opdev/shape_utils.h"
using namespace op;
namespace l0op {
OP_TYPE_REGISTER(Pack);
static const std::initializer_list<op::DataType> AICORE_DTYPE_SUPPORT_LIST = {
op::DataType::DT_INT8, op::DataType::DT_INT16, op::DataType::DT_INT32, op::DataType::DT_INT64,
op::DataType::DT_UINT8, op::DataType::DT_UINT16, op::DataType::DT_UINT32, op::DataType::DT_UINT64,
op::DataType::DT_FLOAT16, op::DataType::DT_FLOAT, op::DataType::DT_DOUBLE, op::DataType::DT_BOOL,
op::DataType::DT_COMPLEX64, op::DataType::DT_BF16};
static bool IsAiCoreSupport(const aclTensorList* inputs)
{
for (uint64_t i = 0; i < inputs->Size(); i++) {
if (!CheckType((*inputs)[i]->GetDataType(), AICORE_DTYPE_SUPPORT_LIST)) {
return false;
}
}
return true;
}
static const aclTensor* PackAiCore(
const aclTensorList* inputs, int64_t dim, aclTensor* out, op::DataType out_dtype, aclOpExecutor* executor)
{
L0_DFX(PackAiCore, inputs, dim, out, out_dtype);
ADD_TO_LAUNCHER_LIST_AICORE(Pack, OP_INPUT(inputs), OP_OUTPUT(out), OP_ATTR(dim));
return out;
}
static const aclTensor* PackAiCpu(
const aclTensorList* inputs, int64_t dim, aclTensor* out, op::DataType out_dtype, aclOpExecutor* executor)
{
L0_DFX(PackAiCpu, inputs, dim, out, out_dtype);
static internal::AicpuTaskSpace space("Pack", ge::DEPEND_IN_SHAPE, true);
auto ret = ADD_TO_LAUNCHER_LIST_AICPU(
Pack, OP_ATTR_NAMES({"N", "axis"}), OP_INPUT(inputs), OP_OUTPUT(out), OP_ATTR(static_cast<int64_t>(inputs->Size()), dim));
CHECK_RET(ret == ACLNN_SUCCESS, nullptr);
return out;
}
static bool PackInferShape(const aclTensorList* inputs, int64_t dim, op::Shape& pack_shape)
{
auto shape = (*inputs)[0]->GetViewShape();
size_t dim_num = shape.GetDimNum();
pack_shape.SetDimNum(dim_num + 1);
size_t inputs_size = inputs->Size();
size_t dim_new = dim >= 0 ? dim : dim + dim_num + 1;
for (size_t i = 0, j = 0; i <= dim_num; i++, j++) {
if (i == dim_new) {
pack_shape.SetDim(i, inputs_size);
j--;
} else {
pack_shape.SetDim(i, shape.GetDim(j));
}
}
return true;
}
const aclTensor* Pack(const aclTensorList* inputs, int64_t dim, op::DataType out_dtype, aclOpExecutor* executor)
{
op::Shape pack_shape;
if (!PackInferShape(inputs, dim, pack_shape)) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID, "inference shape for pack failed.");
return nullptr;
}
auto out = executor->AllocTensor(pack_shape, out_dtype);
if (IsAiCoreSupport(inputs)) {
return PackAiCore(inputs, dim, out, out_dtype, executor);
} else {
return PackAiCpu(inputs, dim, out, out_dtype, executor);
}
}
}