* 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.
*/
* NOTE: Portions of this code were AI-generated and have been
* technically reviewed for functional accuracy and security
*/
* @file aclnn_acos_grad.cpp
* @brief ACLNN L2 API 实现 - AcosGrad 算子
*/
#include "aclnn_acos_grad.h"
#include "acos_grad.h"
#include "aclnn_kernels/contiguous.h"
#include "aclnn_kernels/common/op_error_check.h"
#include "opdev/op_log.h"
#include "opdev/op_dfx.h"
#include "opdev/common_types.h"
#include "opdev/data_type_utils.h"
#include "opdev/make_op_executor.h"
#include "opdev/platform.h"
#include "op_api/aclnn_check.h"
using namespace op;
#define ACLNN_MAX_SHAPE_RANK 8
static const std::initializer_list<op::DataType> ASCEND950_AICORE_DTYPE_SUPPORT_LIST = {
DataType::DT_FLOAT16,
DataType::DT_FLOAT,
DataType::DT_BF16
};
static bool IsDtypeSupported(DataType dtype)
{
auto npuArch = GetCurrentPlatformInfo().GetCurNpuArch();
if (npuArch == NpuArch::DAV_2201) {
return CheckType(dtype, ASCEND950_AICORE_DTYPE_SUPPORT_LIST);
}
if (IsRegBase()) {
return CheckType(dtype, ASCEND950_AICORE_DTYPE_SUPPORT_LIST);
}
return false;
}
static bool HasEmptyTensor(const aclTensor* y_grad, const aclTensor* x)
{
return y_grad->IsEmpty() || x->IsEmpty();
}
static bool CheckNotNull(const aclTensor* y_grad, const aclTensor* x, const aclTensor* x_grad)
{
OP_CHECK_NULL(y_grad, return false);
OP_CHECK_NULL(x, return false);
OP_CHECK_NULL(x_grad, return false);
return true;
}
static bool CheckDtypeValid(const aclTensor* y_grad, const aclTensor* x, const aclTensor* x_grad)
{
OP_CHECK_DTYPE_NOT_MATCH(y_grad, x->GetDataType(), return false);
OP_CHECK_DTYPE_NOT_MATCH(x_grad, y_grad->GetDataType(), return false);
if (!IsDtypeSupported(y_grad->GetDataType())) {
auto npuArch = GetCurrentPlatformInfo().GetCurNpuArch();
OP_LOGE(ACLNN_ERR_PARAM_INVALID,
"AcosGrad: Dtype not supported: dtype=%d, npuArch=%d. "
"Supported: FLOAT16, FLOAT32, BF16.",
static_cast<int>(y_grad->GetDataType()), static_cast<int>(npuArch));
return false;
}
return true;
}
static bool CheckFormat(const aclTensor* y_grad, const aclTensor* x, const aclTensor* x_grad)
{
auto fmtYGrad = y_grad->GetStorageFormat();
auto fmtX = x->GetStorageFormat();
auto fmtXGrad = x_grad->GetStorageFormat();
if (IsPrivateFormat(fmtYGrad) || IsPrivateFormat(fmtX) || IsPrivateFormat(fmtXGrad)) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID,
"AcosGrad: Private format not supported: y_grad=%d, x=%d, x_grad=%d",
static_cast<int>(fmtYGrad), static_cast<int>(fmtX), static_cast<int>(fmtXGrad));
return false;
}
return true;
}
static bool CheckShape(const aclTensor* y_grad, const aclTensor* x, const aclTensor* x_grad)
{
OP_CHECK_MAX_DIM(y_grad, ACLNN_MAX_SHAPE_RANK, return false);
OP_CHECK_MAX_DIM(x, ACLNN_MAX_SHAPE_RANK, return false);
OP_CHECK_MAX_DIM(x_grad, ACLNN_MAX_SHAPE_RANK, return false);
auto yGradShape = y_grad->GetViewShape();
auto xShape = x->GetViewShape();
auto xGradShape = x_grad->GetViewShape();
if (yGradShape != xShape || yGradShape != xGradShape) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID,
"AcosGrad: Shape mismatch: y_grad=%s, x=%s, x_grad=%s",
op::ToString(yGradShape).GetString(),
op::ToString(xShape).GetString(),
op::ToString(xGradShape).GetString());
return false;
}
return true;
}
static aclnnStatus CheckParams(const aclTensor* y_grad, const aclTensor* x, const aclTensor* x_grad)
{
if (!CheckNotNull(y_grad, x, x_grad)) {
OP_LOGE(ACLNN_ERR_PARAM_NULLPTR, "AcosGrad: CheckNotNull failed");
return ACLNN_ERR_PARAM_NULLPTR;
}
if (!CheckDtypeValid(y_grad, x, x_grad)) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID,
"AcosGrad: CheckDtypeValid failed: y_grad_dtype=%d, x_dtype=%d, x_grad_dtype=%d",
static_cast<int>(y_grad->GetDataType()),
static_cast<int>(x->GetDataType()),
static_cast<int>(x_grad->GetDataType()));
return ACLNN_ERR_PARAM_INVALID;
}
if (!CheckFormat(y_grad, x, x_grad)) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID, "AcosGrad: CheckFormat failed");
return ACLNN_ERR_PARAM_INVALID;
}
if (!CheckShape(y_grad, x, x_grad)) {
OP_LOGE(ACLNN_ERR_PARAM_INVALID, "AcosGrad: CheckShape failed");
return ACLNN_ERR_PARAM_INVALID;
}
return ACLNN_SUCCESS;
}
extern "C" aclnnStatus aclnnAcosGradGetWorkspaceSize(
const aclTensor* y_grad,
const aclTensor* x,
const aclTensor* x_grad,
uint64_t* workspaceSize,
aclOpExecutor** executor)
{
L2_DFX_PHASE_1(aclnnAcosGrad, DFX_IN(y_grad, x), DFX_OUT(x_grad));
OP_CHECK_NULL(workspaceSize, return ACLNN_ERR_PARAM_NULLPTR);
OP_CHECK_NULL(executor, return ACLNN_ERR_PARAM_NULLPTR);
auto uniqueExecutor = CREATE_EXECUTOR();
CHECK_RET(uniqueExecutor.get() != nullptr, ACLNN_ERR_INNER_CREATE_EXECUTOR);
auto ret = CheckParams(y_grad, x, x_grad);
CHECK_RET(ret == ACLNN_SUCCESS, ret);
if (HasEmptyTensor(y_grad, x)) {
*workspaceSize = 0;
uniqueExecutor.ReleaseTo(executor);
return ACLNN_SUCCESS;
}
auto yGradContiguous = l0op::Contiguous(y_grad, uniqueExecutor.get());
CHECK_RET(yGradContiguous != nullptr, ACLNN_ERR_INNER_NULLPTR);
auto xContiguous = l0op::Contiguous(x, uniqueExecutor.get());
CHECK_RET(xContiguous != nullptr, ACLNN_ERR_INNER_NULLPTR);
const aclTensor* opResult = l0op::AcosGrad(yGradContiguous, xContiguous, uniqueExecutor.get());
CHECK_RET(opResult != nullptr, ACLNN_ERR_INNER_NULLPTR);
auto viewCopyResult = l0op::ViewCopy(opResult, x_grad, uniqueExecutor.get());
CHECK_RET(viewCopyResult != nullptr, ACLNN_ERR_INNER_NULLPTR);
*workspaceSize = uniqueExecutor->GetWorkspaceSize();
uniqueExecutor.ReleaseTo(executor);
return ACLNN_SUCCESS;
}
extern "C" aclnnStatus aclnnAcosGrad(
void* workspace,
uint64_t workspaceSize,
aclOpExecutor* executor,
aclrtStream stream)
{
L2_DFX_PHASE_2(aclnnAcosGrad);
return CommonOpExecutorRun(workspace, workspaceSize, executor, stream);
}