* Copyright (c) 2026 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.
*/
* \file aclnn_cat.cpp
* \brief
*/
#include "aclnn_chunk_cat.h"
#include "chunk_cat.h"
#include "aclnn_kernels/cast.h"
#include "aclnn_kernels/contiguous.h"
#include "aclnn_kernels/common/op_error_check.h"
#include "opdev/common_types.h"
#include "opdev/data_type_utils.h"
#include "opdev/format_utils.h"
#include "opdev/op_dfx.h"
#include "opdev/op_executor.h"
#include "opdev/op_log.h"
#include "opdev/shape_utils.h"
#include "opdev/tensor_view_utils.h"
#include "opdev/platform.h"
#include "op_api/op_api_def.h"
#include "op_api/aclnn_check.h"
#include <iostream>
using namespace op;
#ifdef __cplusplus
extern "C" {
#endif
constexpr size_t CAT_INPUT_NUM = 512;
static const std::initializer_list<op::DataType> ASCEND910B_DTYPE_SUPPORT_LIST = {
DataType::DT_FLOAT, DataType::DT_FLOAT16, DataType::DT_BF16};
static const inline std::initializer_list<DataType>& GetSupportDtypeList(NpuArch npuArch)
{
static const std::initializer_list<DataType> emptyDtypes = {};
if (
npuArch == NpuArch::DAV_2201) {
return ASCEND910B_DTYPE_SUPPORT_LIST;
} else {
return emptyDtypes;
}
}
static bool CheckDtypeValid(const aclTensorList* tensors, const aclTensor* out)
{
op::DataType inputType = (*tensors)[0]->GetDataType();
if (!CheckType(inputType, ASCEND910B_DTYPE_SUPPORT_LIST)) {
OP_LOGE(
ACLNN_ERR_PARAM_INVALID, "tensor %lu not implemented for %s, should be in dtype support list %s.", 0,
op::ToString(inputType).GetString(), op::ToString(ASCEND910B_DTYPE_SUPPORT_LIST).GetString());
return false;
}
for (uint64_t i = 1; i < tensors->Size(); i++) {
if ((*tensors)[i]->GetDataType() != inputType) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID, "expects all input tensors with the same dtype.");
return false;
}
}
OP_CHECK_DTYPE_NOT_SUPPORT(out, ASCEND910B_DTYPE_SUPPORT_LIST, return false);
if (inputType == DataType::DT_FLOAT && out->GetDataType() != DataType::DT_FLOAT) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID, "output dtype must be float when input dtype is float.");
return false;
}
return true;
}
static bool CheckNotNull(const aclTensorList* tensors, const aclTensor* out)
{
OP_CHECK_NULL(tensors, return false);
for (uint64_t i = 0; i < tensors->Size(); i++) {
OP_CHECK_NULL((*tensors)[i], return false);
}
OP_CHECK_NULL(out, return false);
return true;
}
static bool CheckFormat(const aclTensorList* tensors, const aclTensor* out)
{
for (uint64_t i = 0; i < tensors->Size(); i++) {
op::Format format = (*tensors)[i]->GetStorageFormat();
if (op::IsPrivateFormat(format)) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID, "Format only support ND、NCHW、NHWC、HWCN、NDHWC、NCDHW.");
return false;
}
}
if (op::IsPrivateFormat(out->GetStorageFormat())) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID, "Format only support ND、NCHW、NHWC、HWCN、NDHWC、NCDHW.");
return false;
}
return true;
}
static bool CheckShape(const aclTensorList* tensors)
{
for (uint64_t i = 0; i < tensors->Size(); i++) {
OP_CHECK_MAX_DIM((*tensors)[i], MAX_SUPPORT_DIMS_NUMS, return false);
op::Shape shape = (*tensors)[i]->GetViewShape();
if (shape.GetDimNum() == 0) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID, "input tensor %lu dimnum is 0.", i);
return false;
}
}
return true;
}
static aclnnStatus CheckParams(const aclTensorList* tensors, const aclTensor* out)
{
CHECK_RET(CheckNotNull(tensors, out), ACLNN_ERR_PARAM_NULLPTR);
CHECK_RET(CheckDtypeValid(tensors, out), ACLNN_ERR_PARAM_INVALID);
CHECK_RET(CheckFormat(tensors, out), ACLNN_ERR_PARAM_INVALID);
CHECK_RET(CheckShape(tensors), ACLNN_ERR_PARAM_INVALID);
return ACLNN_SUCCESS;
}
const aclTensor* MergeLastDims(const aclTensor* tensor, int64_t dim, aclOpExecutor* executor) {
op::Shape shapeTensor = tensor->GetViewShape();
int64_t dimNum = shapeTensor.GetDimNum();
op::Shape newShape;
for (int64_t i = 0; i <= dim; i++) {
newShape.AppendDim(static_cast<int64_t>(shapeTensor.GetDim(i)));
}
int64_t catdimSize = 1;
for (int64_t i = dim + 1; i < dimNum; i++) {
catdimSize *= shapeTensor.GetDim(i);
}
newShape.AppendDim(catdimSize);
auto reshapeTensor = executor->CreateView(tensor, tensor->GetViewShape(), tensor->GetViewOffset());
reshapeTensor->SetViewShape(newShape);
reshapeTensor->SetOriginalShape(newShape);
reshapeTensor->SetStorageShape(newShape);
return reshapeTensor;
}
const aclTensor* PostProcess(const aclTensor* tensor, const aclTensor* out, aclOpExecutor* executor) {
auto reFormatTensor = executor->CreateView(tensor, tensor->GetViewShape(), tensor->GetViewOffset());
reFormatTensor->SetViewFormat(out->GetViewFormat());
reFormatTensor->SetOriginalFormat(out->GetOriginalFormat());
reFormatTensor->SetStorageFormat(out->GetStorageFormat());
return reFormatTensor;
}
static aclnnStatus ProcessOneTensor(const aclTensorList* tensors, int64_t dim, int64_t numChunks,
aclTensor* out, aclOpExecutor* executor)
{
auto concatTensor = l0op::ChunkCat(tensors, dim, numChunks, out->GetDataType(), executor);
CHECK_RET(concatTensor != nullptr, ACLNN_ERR_INNER_NULLPTR);
CHECK_RET(CheckShapeAndScalarSame(concatTensor, out), ACLNN_ERR_PARAM_INVALID);
auto outTensor = PostProcess(concatTensor, out, executor);
auto viewCopyResult = l0op::ViewCopy(outTensor, out, executor);
CHECK_RET(viewCopyResult != nullptr, ACLNN_ERR_INNER_NULLPTR);
return ACLNN_SUCCESS;
}
static aclnnStatus SplitToChunkCat(const aclTensorList* tensors, int64_t dim, int64_t numChunks,
aclTensor* out, aclOpExecutor* executor)
{
op::FVector<const aclTensor*> tensorListA;
auto outType = out ->GetDataType();
for (uint64_t i = 0; i < tensors->Size(); i++) {
if (!(*tensors)[i]->IsEmpty()) {
auto contiguous = l0op::Contiguous((*tensors)[i], executor);
CHECK_RET(contiguous != nullptr, ACLNN_ERR_INNER_NULLPTR);
contiguous = MergeLastDims(contiguous, dim, executor);
tensorListA.emplace_back(contiguous);
}
}
if (tensorListA.size() == 1) {
auto tensorList = executor->AllocTensorList(tensorListA.data(), tensorListA.size());
return ProcessOneTensor(tensorList, dim, numChunks, out, executor);
}
while (tensorListA.size() > 1) {
op::FVector<const aclTensor*> tensorListOnce;
op::FVector<const aclTensor*> tensorListB;
for (auto tensor : tensorListA) {
tensorListOnce.emplace_back(tensor);
if (tensorListOnce.size() == CAT_INPUT_NUM) {
auto tensorList = executor->AllocTensorList(tensorListOnce.data(), tensorListOnce.size());
auto concatTensor = l0op::ChunkCat(tensorList, dim, numChunks, outType, executor);
CHECK_RET(concatTensor != nullptr, ACLNN_ERR_INNER_NULLPTR);
tensorListB.emplace_back(concatTensor);
tensorListOnce.clear();
}
}
if (!tensorListOnce.empty()) {
auto aclTensorListTail = executor->AllocTensorList(tensorListOnce.data(), tensorListOnce.size());
auto concatTensorTail = l0op::ChunkCat(aclTensorListTail, dim, numChunks, outType, executor);
CHECK_RET(concatTensorTail != nullptr, ACLNN_ERR_INNER_NULLPTR);
tensorListB.emplace_back(concatTensorTail);
tensorListOnce.clear();
}
tensorListA = tensorListB;
}
if (tensorListA.empty()) {
return ACLNN_SUCCESS;
}
CHECK_RET(CheckShapeAndScalarSame(tensorListA.front(), out), ACLNN_ERR_PARAM_INVALID);
auto outTensor = PostProcess(tensorListA.front(), out, executor);
auto viewCopyResult = l0op::ViewCopy(outTensor, out, executor);
CHECK_RET(viewCopyResult != nullptr, ACLNN_ERR_INNER_NULLPTR);
return ACLNN_SUCCESS;
}
aclnnStatus aclnnChunkCatGetWorkspaceSize(
const aclTensorList* tensors, int64_t dim, int64_t numChunks, aclTensor* out,
uint64_t* workspaceSize, aclOpExecutor** executor)
{
L2_DFX_PHASE_1(aclnnChunkCat, DFX_IN(tensors, dim, numChunks), DFX_OUT(out));
auto uniqueExecutor = CREATE_EXECUTOR();
CHECK_RET(uniqueExecutor.get() != nullptr, ACLNN_ERR_INNER_CREATE_EXECUTOR);
if (dim != 0) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID, "dim only support 0 now.");
}
if (tensors->Size() == 0) {
uniqueExecutor.ReleaseTo(executor);
return ACLNN_SUCCESS;
}
auto ret = CheckParams(tensors, out);
CHECK_RET(ret == ACLNN_SUCCESS, ret);
ret = SplitToChunkCat(tensors, dim, numChunks, out, uniqueExecutor.get());
CHECK_RET(ret == ACLNN_SUCCESS, ret);
*workspaceSize = uniqueExecutor->GetWorkspaceSize();
uniqueExecutor.ReleaseTo(executor);
return ACLNN_SUCCESS;
}
aclnnStatus aclnnChunkCat(void* workspace, uint64_t workspaceSize, aclOpExecutor* executor, aclrtStream stream)
{
L2_DFX_PHASE_2(aclnnChunkCat);
return CommonOpExecutorRun(workspace, workspaceSize, executor, stream);
}
#ifdef __cplusplus
}
#endif