Star1309
1430
代码
代码
Issues118
Pull Requests248
Actions
项目讨论
Wiki
分析
Star1309
1430
已合并
新增950算子 #1641
huangyuxiaaaaa创建于 3月12日
新增950算子 #1641
已合并
huangyuxiaaaaa创建于 3月12日
从huangyuxiaaaaa/ops-matherf合入到cann/ops-mathmaster
共 114 个文件变更+6917-2448
Mmath/atan/CMakeLists.txt+11-7
@@ -1,11 +1,15 @@
1-# ---------------------------------------------------------------------------------------------------------1+# ----------------------------------------------------------------------------
2-# Copyright (c) 2025 Huawei Technologies Co., Ltd.2+# Copyright (c) 2026 Huawei Technologies Co., Ltd.
3-# This program is free software, you can redistribute it and/or modify it under the terms and conditions of 3+# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4# CANN Open Software License Agreement Version 2.0 (the "License").4# CANN Open Software License Agreement Version 2.0 (the "License").
5# Please refer to the License for details. You may not use this file except in compliance with the License.5# Please refer to the License for details. You may not use this file except in compliance with the License.
6-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, 6+# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7-# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. 7+# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9-# ---------------------------------------------------------------------------------------------------------9+# ----------------------------------------------------------------------------
10 10 
11-add_all_modules_sources(OPTYPE atan ACLNNTYPE aclnn_exclude)11+# 设置算子定义时支持的芯片类型
12+set(SUPPORT_COMPUTE_UNIT "ascend950" "mc62cm12a")
13+# 设置每种芯片类型对应的tiling文件目录，即采用op_host目录下哪个文件夹下的tiling文件编译
14+set(SUPPORT_TILING_DIR "arch35" "arch35")
15+add_all_modules_sources(OPTYPE atan ACLNNTYPE aclnn_exclude COMPUTE_UNIT ${SUPPORT_COMPUTE_UNIT} TILING_DIR ${SUPPORT_TILING_DIR} DISABLE_IN_OPP TRUE)
Mmath/atan/op_graph/atan_graph_infer.cpp+16-16
@@ -1,25 +1,25 @@
1-/* Copyright (c) Huawei Technologies Co., Ltd. 2025. All rights reserved.1+/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * See LICENSE in the root of the software repository for the full text of the License.
9- */9+ */
10 10 
11#include "register/op_impl_registry.h"11#include "register/op_impl_registry.h"
12#include "log/log.h"12#include "log/log.h"
13 13 
14using namespace ge;14using namespace ge;
15namespace ops {15namespace ops {
16-static ge::graphStatus InferDataTypeForAtan(gert::InferDataTypeContext* context) {16+static ge::graphStatus InferDataTypeForAtan(gert::InferDataTypeContext* context)
17- OP_LOGI("Begin InferDataTypeForAtan");17+{
18- const ge::DataType xDataType = context->GetInputDataType(0);18+ OP_LOGI("Begin InferDataTypeForAtan");
19- context->SetOutputDataType(0, xDataType);19+ const ge::DataType xDataType = context->GetInputDataType(0);
20- return ge::GRAPH_SUCCESS;20+ context->SetOutputDataType(0, xDataType);
21+ return ge::GRAPH_SUCCESS;
21}22}
22 23 
23IMPL_OP(Atan).InferDataType(InferDataTypeForAtan);24IMPL_OP(Atan).InferDataType(InferDataTypeForAtan);
24-}25+} // namespace ops
25-// namespace ops
Mmath/atan/op_graph/atan_proto.h+2-3
@@ -1,5 +1,5 @@
1/**1/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3 * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3 * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4 * CANN Open Software License Agreement Version 2.0 (the "License").4 * CANN Open Software License Agreement Version 2.0 (the "License").
5 * Please refer to the License for details. You may not use this file except in compliance with the License.5 * Please refer to the License for details. You may not use this file except in compliance with the License.
@@ -17,8 +17,7 @@
17 17 
18#include "graph/operator_reg.h"18#include "graph/operator_reg.h"
19 19 
20-namespace ge20+namespace ge {
21-{
22/**21/**
23*@brief Computes the trignometric inverse tangent of x element-wise.22*@brief Computes the trignometric inverse tangent of x element-wise.
24* The atan operation returns the inverse of tan, such that if y = tan(x) then, x = atan(y).23* The atan operation returns the inverse of tan, such that if y = tan(x) then, x = atan(y).
Amath/atan/op_host/arch35/atan_tiling_arch35.cpp+159-0
@@ -0,0 +1,159 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file atan_tiling_arch35.cpp
13+ * \brief atan_tiling source file
14+ */
15+ 
16+#include <iostream>
17+#include <graph/utils/type_utils.h>
18+ 
19+#include "tiling/platform/platform_ascendc.h"
20+#include "register/op_impl_registry.h"
21+#include "register/tilingdata_base.h"
22+#include "op_host/tiling_util.h"
23+#include "atvoss/elewise/elewise_tiling.h"
24+#include "log/log.h"
25+#include "util/math_util.h"
26+#include "platform/platform_info.h"
27+#include "math/atan/op_kernel/arch35/atan_dag.h"
28+#include "math/atan/op_kernel/arch35/atan_struct.h"
29+#include "atan_tiling_arch35.h"
30+ 
31+using namespace ge;
32+using namespace AtanOp;
33+using namespace Ops::Math::OpTiling;
34+ 
35+namespace optiling {
36+constexpr int64_t ASCEND_API_BUFFER = 122880;
37+const int64_t ASCEND_WORKSPACE = 16 * 1024 * 1024;
38+ 
39+ge::graphStatus AtanTiling::CalcInputDtype()
40+{
41+ auto inputDesc = tilingContext->GetInputDesc(0);
42+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, inputDesc);
43+ this->inputDtype = inputDesc->GetDataType();
44+ OP_CHECK_IF(
45+ this->inputDtype != ge::DT_FLOAT16 && this->inputDtype != ge::DT_BF16 && this->inputDtype != ge::DT_FLOAT,
46+ OP_LOGE(
47+ tilingContext->GetNodeName(),
48+ "input x dtype [%s] not supported, only support [DT_FLOAT16, DT_BF16, DT_FLOAT]",
49+ ge::TypeUtils::DataTypeToSerialString(this->inputDtype).c_str()),
50+ return ge::GRAPH_FAILED);
51+ return ge::GRAPH_SUCCESS;
52+}
53+ 
54+ge::graphStatus AtanTiling::CalcOutputDtype()
55+{
56+ auto outputDesc = tilingContext->GetOutputDesc(0);
57+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, outputDesc);
58+ this->outputDtype = outputDesc->GetDataType();
59+ OP_CHECK_IF(
60+ this->outputDtype != ge::DT_FLOAT16 && this->outputDtype != ge::DT_BF16 && this->outputDtype != ge::DT_FLOAT,
61+ OP_LOGE(
62+ tilingContext->GetNodeName(),
63+ "output y dtype [%s] not supported, only support [DT_FLOAT16, DT_BF16, DT_FLOAT]",
64+ ge::TypeUtils::DataTypeToSerialString(this->outputDtype).c_str()),
65+ return ge::GRAPH_FAILED);
66+ OP_CHECK_IF(
67+ this->outputDtype != this->inputDtype,
68+ OP_LOGE(
69+ tilingContext->GetNodeName(), "input x dtype [%s] should be the same as output y dtype [%s]",
70+ ge::TypeUtils::DataTypeToSerialString(this->inputDtype).c_str(),
71+ ge::TypeUtils::DataTypeToSerialString(this->outputDtype).c_str()),
72+ return ge::GRAPH_FAILED);
73+ return ge::GRAPH_SUCCESS;
74+}
75+ 
76+ge::graphStatus AtanTiling::CheckShape()
77+{
78+ auto inputStorageShape = tilingContext->GetInputShape(0);
79+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, inputStorageShape);
80+ const gert::Shape& inputXShape = EnsureNotScalar(inputStorageShape->GetStorageShape());
81+ 
82+ auto outputStorageShape = tilingContext->GetOutputShape(0);
83+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, outputStorageShape);
84+ const gert::Shape& outputYShape = EnsureNotScalar(outputStorageShape->GetStorageShape());
85+ 
86+ OP_CHECK_IF(
87+ inputXShape != outputYShape, OP_LOGE(tilingContext->GetNodeName(), "input x and output y shape not same"),
88+ return ge::GRAPH_FAILED);
89+ return ge::GRAPH_SUCCESS;
90+}
91+ 
92+ge::graphStatus AtanTiling::RunTiling()
93+{
94+ OP_LOGD(tilingContext->GetNodeName(), "Enter TilingForAtan");
95+ ElewiseBaseTiling elewiseBaseTiling(tilingContext);
96+ OP_CHECK_IF(
97+ CalcInputDtype() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "get input dtype failed"),
98+ return ge::GRAPH_FAILED);
99+ OP_CHECK_IF(
100+ CalcOutputDtype() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "get output dtype failed"),
101+ return ge::GRAPH_FAILED);
102+ OP_CHECK_IF(
103+ CheckShape() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "check shape failed"), return ge::GRAPH_FAILED);
104+ 
105+ auto tiling = tilingContext->GetTilingData<EleBaseTilingDataV2>();
106+ OP_CHECK_IF(
107+ (tiling == nullptr), OP_LOGE(tilingContext->GetNodeName(), "Get AtanTiling from GE context failed"),
108+ return ge::GRAPH_FAILED);
109+ 
110+ ge::graphStatus ret = ge::GRAPH_SUCCESS;
111+ 
112+ if (inputDtype == ge::DT_FLOAT) {
113+ dType = TPL_FP32;
114+ ret = elewiseBaseTiling.DoTiling<AtanOp::AtanDag<float>::OpDag>(*tiling, ASCEND_API_BUFFER);
115+ } else if (inputDtype == ge::DT_FLOAT16) {
116+ dType = TPL_FP16;
117+ ret = elewiseBaseTiling.DoTiling<AtanOp::AtanDag<half>::OpDag>(*tiling, ASCEND_API_BUFFER);
118+ } else if (inputDtype == ge::DT_BF16) {
119+ dType = TPL_BF16;
120+ ret = elewiseBaseTiling.DoTiling<AtanOp::AtanDag<bfloat16_t>::OpDag>(*tiling, ASCEND_API_BUFFER);
121+ } else {
122+ OP_LOGE(
123+ tilingContext->GetNodeName(),
124+ "Input dtype is only support fp16, bf16, fp32, "
125+ "while got %s!",
126+ ge::TypeUtils::DataTypeToSerialString(this->inputDtype).c_str());
127+ return ge::GRAPH_FAILED;
128+ }
129+ 
130+ OP_CHECK_IF(ret == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "elewiseBaseTiling failed"), return ge::GRAPH_FAILED);
131+ 
132+ // set workspace/tilingkey/blockdim
133+ size_t* currentWorkspace = tilingContext->GetWorkspaceSizes(1);
134+ currentWorkspace[0] = ASCEND_WORKSPACE;
135+ const uint64_t tilingKey = GET_TPL_TILING_KEY(static_cast<uint64_t>(tiling->scheMode), dType);
136+ OP_LOGD(tilingContext->GetNodeName(), "[TilingData] : tilingKey=%lu.", tilingKey);
137+ tilingContext->SetTilingKey(tilingKey);
138+ tilingContext->SetBlockDim(tiling->blockNum);
139+ 
140+ return ret;
141+}
142+ 
143+ge::graphStatus TilingForAtan(gert::TilingContext* context)
144+{
145+ OP_LOGD(context->GetNodeName(), "Tiling4Atan rt2.0 is running.");
146+ auto compileInfo = reinterpret_cast<const ElewiseCompileInfo*>(context->GetCompileInfo());
147+ OP_CHECK_NULL_WITH_CONTEXT(context, compileInfo);
148+ AtanTiling baseOpTiling(context);
149+ return baseOpTiling.RunTiling();
150+}
151+ 
152+static ge::graphStatus TilingPrepareForAtan([[maybe_unused]] gert::TilingParseContext* context)
153+{
154+ return ge::GRAPH_SUCCESS;
155+}
156+ 
157+IMPL_OP_OPTILING(Atan).Tiling(TilingForAtan).TilingParse<ElewiseCompileInfo>(TilingPrepareForAtan);
158+ 
159+} // namespace optiling
Amath/atan/op_host/arch35/atan_tiling_arch35.h+41-0
@@ -0,0 +1,41 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file atan_tiling_arch35.h
13+ * \brief atan_tiling head file
14+ */
15+#ifndef OPS_MATH_ATAN_TILING_ARCH35_H
16+#define OPS_MATH_ATAN_TILING_ARCH35_H
17+ 
18+#include "register/tilingdata_base.h"
19+#include "atvoss/elewise/elewise_tiling.h"
20+ 
21+namespace optiling {
22+ 
23+class AtanTiling {
24+public:
25+ explicit AtanTiling(gert::TilingContext* context) : tilingContext(context)
26+ {}
27+ ge::graphStatus RunTiling();
28+ 
29+protected:
30+ ge::graphStatus CalcOutputDtype();
31+ ge::graphStatus CalcInputDtype();
32+ ge::graphStatus CheckShape();
33+ 
34+private:
35+ uint64_t dType = 0;
36+ gert::TilingContext* tilingContext;
37+ ge::DataType inputDtype;
38+ ge::DataType outputDtype;
39+};
40+} // namespace optiling
41+#endif // OPS_MATH_ATAN_TILING_ARCH35_H
Amath/atan/op_host/atan_def.cpp+49-0
@@ -0,0 +1,49 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file atan_def.cpp
13+ * \brief atan def
14+ */
15+ 
16+#include <cstdint>
17+#include "register/op_def_registry.h"
18+ 
19+namespace ops {
20+class Atan : public OpDef {
21+public:
22+ explicit Atan(const char* name) : OpDef(name)
23+ {
24+ this->Input("x")
25+ .ParamType(REQUIRED)
26+ .DataType({ge::DT_BF16, ge::DT_FLOAT16, ge::DT_FLOAT})
27+ .Format({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND})
28+ .UnknownShapeFormat({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND});
29+ this->Output("y")
30+ .ParamType(REQUIRED)
31+ .DataType({ge::DT_BF16, ge::DT_FLOAT16, ge::DT_FLOAT})
32+ .Format({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND})
33+ .UnknownShapeFormat({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND});
34+ 
35+ OpAICoreConfig aicoreConfig;
36+ aicoreConfig.DynamicCompileStaticFlag(true)
37+ .DynamicFormatFlag(false)
38+ .DynamicRankSupportFlag(true)
39+ .DynamicShapeSupportFlag(true)
40+ .NeedCheckSupportFlag(false)
41+ .PrecisionReduceFlag(true)
42+ .ExtendCfgInfo("opFile.value", "atan_apt");
43+ this->AICore().AddConfig("ascend950", aicoreConfig);
44+ this->AICore().AddConfig("mc62cm12a", aicoreConfig);
45+ }
46+};
47+ 
48+OP_ADD(Atan);
49+} // namespace ops
Mmath/atan/op_host/atan_infershape.cpp+7-7
@@ -1,5 +1,5 @@
1-/* Copyright (c) Huawei Technologies Co., Ltd. 2025. All rights reserved.1+/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3 * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3 * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4 * CANN Open Software License Agreement Version 2.0 (the "License").4 * CANN Open Software License Agreement Version 2.0 (the "License").
5 * Please refer to the License for details. You may not use this file except in compliance with the License.5 * Please refer to the License for details. You may not use this file except in compliance with the License.
@@ -14,11 +14,11 @@
14 14 
15using namespace ge;15using namespace ge;
16namespace ops {16namespace ops {
17-static ge::graphStatus InferShapeForAtan(gert::InferShapeContext *context) {17+static ge::graphStatus InferShapeForAtan(gert::InferShapeContext* context)
18- OP_LOGI("Begin InferShapeForAtan");18+{
19- return Ops::Base::InferShape4Elewise(context);19+ OP_LOGI("Begin InferShapeForAtan");
20+ return Ops::Base::InferShape4Elewise(context);
20}21}
21 22 
22IMPL_OP_INFERSHAPE(Atan).InferShape(InferShapeForAtan);23IMPL_OP_INFERSHAPE(Atan).InferShape(InferShapeForAtan);
23-}24+} // namespace ops
24-// namespace ops
Amath/atan/op_host/config/ascend950/atan_binary.json+92-0
@@ -0,0 +1,92 @@
1+{
2+ "op_type": "Atan",
3+ "op_list": [
4+ {
5+ "bin_filename": "Atan_bfloat16",
6+ "inputs": [
7+ {
8+ "name": "x",
9+ "index": 0,
10+ "dtype": "bfloat16",
11+ "format": "ND",
12+ "paramType": "required",
13+ "shape": [
14+ -2
15+ ],
16+ "format_match_mode": "FormatAgnostic"
17+ }
18+ ],
19+ "outputs": [
20+ {
21+ "name": "y",
22+ "index": 0,
23+ "dtype": "bfloat16",
24+ "format": "ND",
25+ "paramType": "required",
26+ "shape": [
27+ -2
28+ ],
29+ "format_match_mode": "FormatAgnostic"
30+ }
31+ ]
32+ },
33+ {
34+ "bin_filename": "Atan_float16",
35+ "inputs": [
36+ {
37+ "name": "x",
38+ "index": 0,
39+ "dtype": "float16",
40+ "format": "ND",
41+ "paramType": "required",
42+ "shape": [
43+ -2
44+ ],
45+ "format_match_mode": "FormatAgnostic"
46+ }
47+ ],
48+ "outputs": [
49+ {
50+ "name": "y",
51+ "index": 0,
52+ "dtype": "float16",
53+ "format": "ND",
54+ "paramType": "required",
55+ "shape": [
56+ -2
57+ ],
58+ "format_match_mode": "FormatAgnostic"
59+ }
60+ ]
61+ },
62+ {
63+ "bin_filename": "Atan_float32",
64+ "inputs": [
65+ {
66+ "name": "x",
67+ "index": 0,
68+ "dtype": "float32",
69+ "format": "ND",
70+ "paramType": "required",
71+ "shape": [
72+ -2
73+ ],
74+ "format_match_mode": "FormatAgnostic"
75+ }
76+ ],
77+ "outputs": [
78+ {
79+ "name": "y",
80+ "index": 0,
81+ "dtype": "float32",
82+ "format": "ND",
83+ "paramType": "required",
84+ "shape": [
85+ -2
86+ ],
87+ "format_match_mode": "FormatAgnostic"
88+ }
89+ ]
90+ }
91+ ]
92+}
Amath/atan/op_host/config/ascend950/atan_simplified_key.ini+2-0
@@ -0,0 +1,2 @@
1+[Atan]
2+default=0
Dmath/atan/op_host/op_api/aclnn_atan.h+0-95
@@ -1,95 +0,0 @@
1-/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4- * CANN Open Software License Agreement Version 2.0 (the "License").
5- * Please refer to the License for details. You may not use this file except in compliance with the License.
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8- * See LICENSE in the root of the software repository for the full text of the License.
9- */
10-#ifndef OP_API_INC_ATAN_H_
11-#define OP_API_INC_ATAN_H_
12- 
13-#include "aclnn/aclnn_base.h"
14-#include "aclnn_util.h"
15- 
16-#ifdef __cplusplus
17-extern "C" {
18-#endif
19- 
20-/**
21- * @brief aclnnAtan的第一段接口，根据具体的计算流程，计算workspace大小。
22- * @domain aclnn_math
23- * 功能描述：从输入矩阵的每一个元素进行反正切操作后输出。
24- * 计算公式：
25- * out_{i}=tan^{-1}(input_{i})
26- * 参数描述：
27- * @param [in] input
28- * 输入Tensor，数据类型支持INT8，INT16，INT32，INT64，UINT8，BOOL，FLOAT，FLOAT16，DOUBLE，BFLOAT16。支持非连续Tensor，数据格式支持ND。
29- * @param [in] out
30- * 输出Tensor，数据类型支持FLOAT，BFLOAT16, FLOAT16，DOUBLE。支持非连续Tensor，数据格式支持ND。
31- * @param [out] workspaceSize 返回用户需要在npu device侧申请的workspace大小。
32- * @param [out] executor 返回op执行器，包含了算子计算流程。
33- * @return aclnnStatus 返回状态码
34- */
35-ACLNN_API aclnnStatus aclnnAtanGetWorkspaceSize(const aclTensor* input, aclTensor* out, uint64_t* workspaceSize,
36- aclOpExecutor** executor);
37-/**
38- * @brief aclnnAtan的第二段接口，用于执行计算。
39- * 功能描述：从输入矩阵的每一个元素进行反正切操作后输出。。
40- * 计算公式：
41- * out_{i}=tan^{-1}(input_{i})
42- * 实现说明：
43- * api计算的基本路径：
44-```mermaid
45-graph LR
46- A[(Self)] -->B([l0op::Contiguous])
47- B --> C([l0op::Atan])
48- C --> G([l0op::Cast])
49- G --> E([l0op::ViewCopy])
50- E --> S[(Out)]
51-```
52- * @param [in] workspace: 在npu device侧申请的workspace内存起址。
53- * @param [in] workspaceSize: 在npu device侧申请的workspace大小，由第一段接口aclnnAddGetWorkspaceSize获取。
54- * @param [in] stream: acl stream流。
55- * @param [in] executor: op执行器，包含了算子计算流程。
56- * @return aclnnStatus: 返回状态码。
57- */
58-ACLNN_API aclnnStatus aclnnAtan(void* workspace, uint64_t workspaceSize, aclOpExecutor* executor,
59- const aclrtStream stream);
60- 
61-/**
62- * @brief aclnnInplaceAtan的第一段接口，根据具体的计算流程，计算workspace大小。
63- * @domain aclnn_math
64- * 功能描述：从输入矩阵的每一个元素进行反余弦操作后输出。
65- * 计算公式：
66- * out_{i}=tan^{-1}(input_{i})
67- * 参数描述：
68- * @param [in] input
69- * 输入Tensor，数据类型支持INT8，INT16，INT32，INT64，UINT8，BOOL，FLOAT，FLOAT16，DOUBLE。支持非连续Tensor，数据格式支持ND。
70- * @param [in] out
71- * 输出Tensor，数据类型支持FLOAT，FLOAT16，DOUBLE。支持非连续Tensor，数据格式支持ND。
72- * @param [out] workspaceSize 返回用户需要在npu device侧申请的workspace大小。
73- * @param [out] executor 返回op执行器，包含了算子计算流程。
74- * @return aclnnStatus 返回状态码
75- */
76-ACLNN_API aclnnStatus aclnnInplaceAtanGetWorkspaceSize(aclTensor* inputRef, uint64_t* workspaceSize,
77- aclOpExecutor** executor);
78- 
79-/**
80- * @brief: aclnnInplaceAtan的第二段接口，用于执行计算
81- *
82- * 算子功能： 对输入Tensor原地完成atan操作
83- * @param [in] workspace: 在npu device侧申请的workspace内存起址。
84- * @param [in] workspaceSize: 在npu device侧申请的workspace大小，由第一段接口aclnnAtanGetWorkspaceSize获取。
85- * @param [in] stream: acl stream流。
86- * @param [in] executor: op执行器，包含了算子计算流程。
87- * @return aclnnStatus: 返回状态码。
88- */
89-ACLNN_API aclnnStatus aclnnInplaceAtan(void* workspace, uint64_t workspaceSize, aclOpExecutor* executor,
90- const aclrtStream stream);
91-#ifdef __cplusplus
92-}
93-#endif
94- 
95-#endif
Dmath/atan/op_host/op_api/atan.h+0-19
@@ -1,19 +0,0 @@
1-/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4- * CANN Open Software License Agreement Version 2.0 (the "License").
5- * Please refer to the License for details. You may not use this file except in compliance with the License.
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8- * See LICENSE in the root of the software repository for the full text of the License.
9- */
10-#ifndef PTA_NPU_OP_API_INC_LEVEL0_OP_ATAN_OP_H_
11-#define PTA_NPU_OP_API_INC_LEVEL0_OP_ATAN_OP_H_
12- 
13-#include "opdev/op_executor.h"
14- 
15-namespace l0op {
16-const aclTensor* Atan(const aclTensor* input, aclOpExecutor* executor);
17-}
18- 
19-#endif
Amath/atan/op_kernel/arch35/atan_dag.h+47-0
@@ -0,0 +1,47 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file atan_dag.h
13+ * atan:
14+ * atan(x) = x - x^3/3 + x^5/5 - x^7/7 + x^9/9 - x^11/11 + x^13/13... (|x|<=1)
15+ * = pi/4 + atan((x-1)/(x+1)) (x>1)
16+ * = pi/8 + atan((x-tan(pi/8))/(1+tan(pi/8)*x)) (tan(pi/8)=0.4142135623730950) (x > tan(pi/8) and x <
17+ * tan(pi/4)))
18+ */
19+ 
20+#ifndef ATAN_DAG_H
21+#define ATAN_DAG_H
22+#include "atvoss/util/dag.h"
23+#include "atvoss/util/vec.h"
24+#include "atvoss/util/placeholder.h"
25+ 
26+namespace AtanOp {
27+using namespace Ops::Base;
28+ 
29+template <typename T>
30+struct AtanDag {
31+ using InputX1T = Bind<Vec::CopyIn<T>, Placeholder::In0<T>>;
32+ 
33+ // cast
34+ using InputX1 = Bind<Vec::Cast<float, T, 0>, InputX1T>;
35+ 
36+ using AtanRes = Bind<Vec::AtanPolyApprox<float>, InputX1>;
37+ using OpCastRes = Bind<Vec::Cast<T, float, 1>, AtanRes>;
38+ using OpCopyOut = Bind<Vec::CopyOut<T>, Placeholder::Out0<T>, OpCastRes>;
39+ 
40+ // 指定输出节点
41+ using Outputs = Elems<OpCopyOut>;
42+ using MemCfg = MemOptCfg<MemLevel::LEVEL_2>;
43+ using OpDag = DAGSch<Outputs, void, MemCfg>;
44+};
45+} // namespace AtanOp
46+ 
47+#endif // ATAN_DAG_H
Amath/atan/op_kernel/arch35/atan_struct.h+39-0
@@ -0,0 +1,39 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file atan_struct.h
13+ * \brief atan_struct
14+ */
15+ 
16+#include "ascendc/host_api/tiling/template_argument.h"
17+ 
18+#ifndef OPS_MATH_ATAN_STRUCT_H_
19+#define OPS_MATH_ATAN_STRUCT_H_
20+ 
21+namespace AtanOp {
22+#define TPL_FP16 101
23+#define TPL_BF16 102
24+#define TPL_FP32 103
25+ 
26+#define TPL_SCH_MODE_0 0
27+#define TPL_SCH_MODE_1 1
28+ 
29+// placeholder to avoid tilingkey overlap
30+ 
31+ASCENDC_TPL_ARGS_DECL(
32+ Atan, ASCENDC_TPL_UINT_DECL(schMode, 1, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
33+ ASCENDC_TPL_DTYPE_DECL(dType, TPL_FP16, TPL_BF16, TPL_FP32));
34+ 
35+ASCENDC_TPL_SEL(ASCENDC_TPL_ARGS_SEL(
36+ ASCENDC_TPL_UINT_SEL(schMode, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
37+ ASCENDC_TPL_DTYPE_SEL(dType, TPL_FP16, TPL_BF16, TPL_FP32)));
38+} // namespace AtanOp
39+#endif // OPS_MATH_ATAN_STRUCT_H_
Amath/atan/op_kernel/atan_apt.cpp+46-0
@@ -0,0 +1,46 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file atan_apt.cpp
13+ * \brief atan kernel
14+ */
15+ 
16+#include "kernel_operator.h"
17+#include "arch35/atan_dag.h"
18+#include "arch35/atan_struct.h"
19+#include "atvoss/elewise/elewise_sch.h"
20+ 
21+using namespace Ops::Base;
22+using namespace AscendC;
23+ 
24+template <uint64_t schMode, uint64_t dType>
25+__global__ __aicore__ void atan(GM_ADDR x, GM_ADDR y, GM_ADDR workspace, GM_ADDR tiling)
26+{
27+ REGISTER_TILING_DEFAULT(EleBaseTilingDataV2);
28+ GET_TILING_DATA_WITH_STRUCT(EleBaseTilingDataV2, tilingData, tiling);
29+ KERNEL_TASK_TYPE_DEFAULT(KERNEL_TYPE_AIV_ONLY);
30+ TPipe pipe;
31+ 
32+ if constexpr (dType == TPL_FP16) {
33+ ElementwiseSch<schMode, AtanOp::AtanDag<half>::OpDag> sch(&tilingData, &pipe);
34+ sch.Init(x, y);
35+ sch.Process();
36+ } else if constexpr (dType == TPL_BF16) {
37+ ElementwiseSch<schMode, AtanOp::AtanDag<bfloat16_t>::OpDag> sch(&tilingData, &pipe);
38+ sch.Init(x, y);
39+ sch.Process();
40+ } else if constexpr (dType == TPL_FP32) {
41+ ElementwiseSch<schMode, AtanOp::AtanDag<float>::OpDag> sch(&tilingData, &pipe);
42+ sch.Init(x, y);
43+ sch.Process();
44+ }
45+ return;
46+}
Amath/atan/tests/ut/op_api/test_aclnn_atan.cpp+153-0
@@ -0,0 +1,153 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <array>
12+#include <vector>
13+#include "gtest/gtest.h"
14+ 
15+#include "math/atan/op_api/aclnn_atan.h"
16+ 
17+#include "op_api_ut_common/inner/types.h"
18+#include "op_api_ut_common/op_api_ut.h"
19+#include "op_api_ut_common/scalar_desc.h"
20+#include "op_api_ut_common/tensor_desc.h"
21+ 
22+using namespace std;
23+ 
24+class l2_atan_test : public testing::Test {
25+protected:
26+ static void SetUpTestCase()
27+ {
28+ cout << "l2_atan_test SetUp" << endl;
29+ }
30+ 
31+ static void TearDownTestCase()
32+ {
33+ cout << "l2_atan_test TearDown" << endl;
34+ }
35+};
36+ 
37+TEST_F(l2_atan_test, ascend910B2_atan_dtype_all)
38+{
39+ vector<aclDataType> input_vaild_dtype_list{ACL_INT8, ACL_INT32, ACL_UINT8, ACL_INT16, ACL_INT64,
40+ ACL_BOOL, ACL_FLOAT, ACL_FLOAT16, ACL_DOUBLE, ACL_BF16};
41+ vector<aclDataType> output_vaild_dtype_list{ACL_FLOAT, ACL_FLOAT16, ACL_DOUBLE};
42+ vector<aclDataType> invaild_dtype_list{ACL_COMPLEX64, ACL_COMPLEX128};
43+ for (auto dtype1 : input_vaild_dtype_list) {
44+ auto self_tensor_desc = TensorDesc({3, 5}, dtype1, ACL_FORMAT_ND).ValueRange(-20, 20);
45+ for (auto dtype2 : output_vaild_dtype_list) {
46+ auto out_tensor_desc = TensorDesc({3, 5}, dtype2, ACL_FORMAT_ND).Precision(0.001, 0.001);
47+ 
48+ auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
49+ uint64_t workspace_size = 0;
50+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
51+ EXPECT_EQ(aclRet, ACL_SUCCESS);
52+ }
53+ }
54+ for (auto dtype : invaild_dtype_list) {
55+ auto self_tensor_desc = TensorDesc({3, 5}, dtype, ACL_FORMAT_ND).ValueRange(-20, 20);
56+ auto out_tensor_desc = TensorDesc({3, 5}, dtype, ACL_FORMAT_ND);
57+ 
58+ auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
59+ uint64_t workspace_size = 0;
60+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
61+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
62+ }
63+}
64+ 
65+TEST_F(l2_atan_test, atan_nullptr)
66+{
67+ auto ut = OP_API_UT(aclnnAtan, INPUT((aclTensor*)nullptr), OUTPUT((aclTensor*)nullptr));
68+ uint64_t workspace_size = 0;
69+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
70+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_NULLPTR);
71+}
72+ 
73+TEST_F(l2_atan_test, atan_precision)
74+{
75+ auto self_tensor_desc = TensorDesc({13, 16, 9}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-20, 20);
76+ auto out_tensor_desc = TensorDesc({13, 16, 9}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.001, 0.001);
77+ 
78+ auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
79+ uint64_t workspace_size = 0;
80+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
81+ EXPECT_EQ(aclRet, ACL_SUCCESS);
82+}
83+ 
84+TEST_F(l2_atan_test, ascend910B2_atan_bf16_precision)
85+{
86+ auto self_tensor_desc = TensorDesc({13, 16, 9}, ACL_BF16, ACL_FORMAT_ND).ValueRange(-20, 20);
87+ auto out_tensor_desc = TensorDesc({13, 16, 9}, ACL_BF16, ACL_FORMAT_ND).Precision(0.001, 0.001);
88+ 
89+ auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
90+ uint64_t workspace_size = 0;
91+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
92+ EXPECT_EQ(aclRet, ACL_SUCCESS);
93+}
94+ 
95+TEST_F(l2_atan_test, atan_empty_tensor)
96+{
97+ auto self_tensor_desc = TensorDesc({13, 0, 9}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-20, 20);
98+ auto out_tensor_desc = TensorDesc({13, 0, 9}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.001, 0.001);
99+ 
100+ auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
101+ uint64_t workspace_size = 0;
102+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
103+ EXPECT_EQ(aclRet, ACL_SUCCESS);
104+}
105+ 
106+TEST_F(l2_atan_test, atan_non_contiguous)
107+{
108+ auto self_tensor_desc = TensorDesc({5, 3}, ACL_FLOAT, ACL_FORMAT_ND, {1, 5}, 0, {3, 5}).ValueRange(-20, 20);
109+ auto out_tensor_desc = TensorDesc({5, 3}, ACL_FLOAT, ACL_FORMAT_ND, {1, 5}, 0, {3, 5}).Precision(0.001, 0.001);
110+ 
111+ auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
112+ uint64_t workspace_size = 0;
113+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
114+ EXPECT_EQ(aclRet, ACL_SUCCESS);
115+}
116+ 
117+TEST_F(l2_atan_test, atan_bigDim)
118+{
119+ auto self_tensor_desc = TensorDesc({2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-20, 20);
120+ auto out_tensor_desc = TensorDesc({2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.001, 0.001);
121+ 
122+ auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
123+ uint64_t workspace_size = 0;
124+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
125+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
126+}
127+ 
128+TEST_F(l2_atan_test, atan_bigDim_non_contiguous)
129+{
130+ auto self_tensor_desc = TensorDesc(
131+ {2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND, {128, 64, 32, 16, 8, 4, 2, 1, 0},
132+ 0, {2, 2, 2, 2, 2, 2, 2, 2, 1})
133+ .ValueRange(-20, 20);
134+ auto out_tensor_desc = TensorDesc(
135+ {2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND, {128, 64, 32, 16, 8, 4, 2, 1, 0},
136+ 0, {2, 2, 2, 2, 2, 2, 2, 2, 1})
137+ .Precision(0.001, 0.001);
138+ 
139+ auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
140+ uint64_t workspace_size = 0;
141+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
142+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
143+}
144+ 
145+TEST_F(l2_atan_test, atan_inplace)
146+{
147+ auto self_tensor_desc = TensorDesc({3, 5}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-20, 20);
148+ auto ut = OP_API_UT(aclnnInplaceAtan, INPUT(self_tensor_desc), OUTPUT());
149+ 
150+ uint64_t workspace_size = 0;
151+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
152+ EXPECT_EQ(aclRet, ACL_SUCCESS);
153+}
Dmath/atan/tests/ut/op_api/test_atan.cpp+0-139
@@ -1,139 +0,0 @@
1-/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4- * CANN Open Software License Agreement Version 2.0 (the "License").
5- * Please refer to the License for details. You may not use this file except in compliance with the License.
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8- * See LICENSE in the root of the software repository for the full text of the License.
9- */
10-#include <array>
11-#include <vector>
12-#include "gtest/gtest.h"
13- 
14-#include "../../../../op_host/op_api/aclnn_atan.h"
15- 
16-#include "op_api_ut_common/inner/types.h"
17-#include "op_api_ut_common/op_api_ut.h"
18-#include "op_api_ut_common/scalar_desc.h"
19-#include "op_api_ut_common/tensor_desc.h"
20- 
21-using namespace std;
22- 
23-class l2_atan_test : public testing::Test {
24- protected:
25- static void SetUpTestCase() {
26- cout << "l2_atan_test SetUp" << endl;
27- }
28- 
29- static void TearDownTestCase() {
30- cout << "l2_atan_test TearDown" << endl;
31- }
32-};
33- 
34-TEST_F(l2_atan_test, ascend910B2_atan_dtype_all) {
35- vector<aclDataType> input_vaild_dtype_list{ACL_INT8, ACL_INT32, ACL_UINT8, ACL_INT16, ACL_INT64,
36- ACL_BOOL, ACL_FLOAT, ACL_FLOAT16, ACL_DOUBLE, ACL_BF16};
37- vector<aclDataType> output_vaild_dtype_list{ACL_FLOAT, ACL_FLOAT16, ACL_DOUBLE};
38- vector<aclDataType> invaild_dtype_list{ACL_COMPLEX64, ACL_COMPLEX128};
39- for (auto dtype1 : input_vaild_dtype_list) {
40- auto self_tensor_desc = TensorDesc({3, 5}, dtype1, ACL_FORMAT_ND).ValueRange(-20, 20);
41- for (auto dtype2 : output_vaild_dtype_list) {
42- auto out_tensor_desc = TensorDesc({3, 5}, dtype2, ACL_FORMAT_ND).Precision(0.001, 0.001);
43- 
44- auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
45- uint64_t workspace_size = 0;
46- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
47- EXPECT_EQ(aclRet, ACL_SUCCESS);
48- }
49- }
50- for (auto dtype : invaild_dtype_list) {
51- auto self_tensor_desc = TensorDesc({3, 5}, dtype, ACL_FORMAT_ND).ValueRange(-20, 20);
52- auto out_tensor_desc = TensorDesc({3, 5}, dtype, ACL_FORMAT_ND);
53- 
54- auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
55- uint64_t workspace_size = 0;
56- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
57- EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
58- }
59-}
60- 
61-TEST_F(l2_atan_test, atan_nullptr) {
62- auto ut = OP_API_UT(aclnnAtan, INPUT((aclTensor*)nullptr), OUTPUT((aclTensor*)nullptr));
63- uint64_t workspace_size = 0;
64- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
65- EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_NULLPTR);
66-}
67- 
68-TEST_F(l2_atan_test, atan_precision) {
69- auto self_tensor_desc = TensorDesc({13, 16, 9}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-20, 20);
70- auto out_tensor_desc = TensorDesc({13, 16, 9}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.001, 0.001);
71- 
72- auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
73- uint64_t workspace_size = 0;
74- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
75- EXPECT_EQ(aclRet, ACL_SUCCESS);
76-}
77- 
78-TEST_F(l2_atan_test, ascend910B2_atan_bf16_precision) {
79- auto self_tensor_desc = TensorDesc({13, 16, 9}, ACL_BF16, ACL_FORMAT_ND).ValueRange(-20, 20);
80- auto out_tensor_desc = TensorDesc({13, 16, 9}, ACL_BF16, ACL_FORMAT_ND).Precision(0.001, 0.001);
81- 
82- auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
83- uint64_t workspace_size = 0;
84- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
85- EXPECT_EQ(aclRet, ACL_SUCCESS);
86-}
87- 
88-TEST_F(l2_atan_test, atan_empty_tensor) {
89- auto self_tensor_desc = TensorDesc({13, 0, 9}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-20, 20);
90- auto out_tensor_desc = TensorDesc({13, 0, 9}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.001, 0.001);
91- 
92- auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
93- uint64_t workspace_size = 0;
94- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
95- EXPECT_EQ(aclRet, ACL_SUCCESS);
96-}
97- 
98-TEST_F(l2_atan_test, atan_non_contiguous) {
99- auto self_tensor_desc = TensorDesc({5, 3}, ACL_FLOAT, ACL_FORMAT_ND, {1, 5}, 0, {3, 5}).ValueRange(-20, 20);
100- auto out_tensor_desc = TensorDesc({5, 3}, ACL_FLOAT, ACL_FORMAT_ND, {1, 5}, 0, {3, 5}).Precision(0.001, 0.001);
101- 
102- auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
103- uint64_t workspace_size = 0;
104- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
105- EXPECT_EQ(aclRet, ACL_SUCCESS);
106-}
107- 
108-TEST_F(l2_atan_test, atan_bigDim) {
109- auto self_tensor_desc = TensorDesc({2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-20, 20);
110- auto out_tensor_desc = TensorDesc({2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.001, 0.001);
111- 
112- auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
113- uint64_t workspace_size = 0;
114- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
115- EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
116-}
117- 
118-TEST_F(l2_atan_test, atan_bigDim_non_contiguous) {
119- auto self_tensor_desc = TensorDesc({2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND,
120- {128, 64, 32, 16, 8, 4, 2, 1, 0}, 0, {2, 2, 2, 2, 2, 2, 2, 2, 1})
121- .ValueRange(-20, 20);
122- auto out_tensor_desc = TensorDesc({2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND,
123- {128, 64, 32, 16, 8, 4, 2, 1, 0}, 0, {2, 2, 2, 2, 2, 2, 2, 2, 1})
124- .Precision(0.001, 0.001);
125- 
126- auto ut = OP_API_UT(aclnnAtan, INPUT(self_tensor_desc), OUTPUT(out_tensor_desc));
127- uint64_t workspace_size = 0;
128- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
129- EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
130-}
131- 
132-TEST_F(l2_atan_test, atan_inplace) {
133- auto self_tensor_desc = TensorDesc({3, 5}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-20, 20);
134- auto ut = OP_API_UT(aclnnInplaceAtan, INPUT(self_tensor_desc), OUTPUT());
135-
136- uint64_t workspace_size = 0;
137- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
138- EXPECT_EQ(aclRet, ACL_SUCCESS);
139-}
Amath/atan/tests/ut/op_host/arch35/test_atan_tiling.cpp+151-0
@@ -0,0 +1,151 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <gtest/gtest.h>
12+#include <iostream>
13+#include "math/atan/op_host/arch35/atan_tiling_arch35.h"
14+#include "tiling_context_faker.h"
15+#include "tiling_case_executor.h"
16+#include "atvoss/elewise/elewise_tiling.h"
17+ 
18+using namespace std;
19+ 
20+class AtanTilingTest : public testing::Test {
21+protected:
22+ static void SetUpTestCase()
23+ {
24+ std::cout << "AtanTilingTest SetUp" << std::endl;
25+ }
26+ 
27+ static void TearDownTestCase()
28+ {
29+ std::cout << "AtanTilingTest TearDown" << std::endl;
30+ }
31+};
32+ 
33+TEST_F(AtanTilingTest, atan_test_tiling_fp16_input)
34+{
35+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
36+ gert::TilingContextPara tilingContextPara(
37+ "Atan",
38+ {
39+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT16, ge::FORMAT_ND},
40+ },
41+ {
42+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT16, ge::FORMAT_ND},
43+ },
44+ {}, &compileInfo);
45+ 
46+ uint64_t expectTilingKey = 203;
47+ string expectTilingData = "8192 4 5760 2048 4 1 1 2048 2048 5760 1 ";
48+ std::vector<size_t> expectWorkspaces = {16777216};
49+ ExecuteTestCaseForEle(
50+ tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
51+}
52+ 
53+TEST_F(AtanTilingTest, atan_test_tiling_bf16_input)
54+{
55+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
56+ gert::TilingContextPara tilingContextPara(
57+ "Atan",
58+ {
59+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
60+ },
61+ {
62+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
63+ },
64+ {}, &compileInfo);
65+ 
66+ uint64_t expectTilingKey = 205;
67+ string expectTilingData = "8192 4 5760 2048 4 1 1 2048 2048 5760 1 ";
68+ std::vector<size_t> expectWorkspaces = {16777216};
69+ ExecuteTestCaseForEle(
70+ tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
71+}
72+ 
73+TEST_F(AtanTilingTest, atan_test_tiling_fp32_input)
74+{
75+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
76+ gert::TilingContextPara tilingContextPara(
77+ "Atan",
78+ {
79+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
80+ },
81+ {
82+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
83+ },
84+ {}, &compileInfo);
85+ 
86+ uint64_t expectTilingKey = 207;
87+ string expectTilingData = "8192 8 8704 1024 8 1 1 1024 1024 8704 1 ";
88+ std::vector<size_t> expectWorkspaces = {16777216};
89+ ExecuteTestCaseForEle(
90+ tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
91+}
92+ 
93+TEST_F(AtanTilingTest, atan_test_tiling_invalid_input_dtype)
94+{
95+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
96+ gert::TilingContextPara tilingContextPara(
97+ "Atan",
98+ {
99+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT32, ge::FORMAT_ND},
100+ },
101+ {
102+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
103+ },
104+ {}, &compileInfo);
105+ 
106+ uint64_t expectTilingKey = 7;
107+ string expectTilingData = "";
108+ std::vector<size_t> expectWorkspaces = {16777216};
109+ ExecuteTestCaseForEle(
110+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
111+}
112+ 
113+TEST_F(AtanTilingTest, atan_test_tiling_invalid_output_dtype)
114+{
115+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
116+ gert::TilingContextPara tilingContextPara(
117+ "Atan",
118+ {
119+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
120+ },
121+ {
122+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
123+ },
124+ {}, &compileInfo);
125+ 
126+ uint64_t expectTilingKey = 7;
127+ string expectTilingData = "";
128+ std::vector<size_t> expectWorkspaces = {16777216};
129+ ExecuteTestCaseForEle(
130+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
131+}
132+ 
133+TEST_F(AtanTilingTest, atan_test_tiling_invalid_shape)
134+{
135+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
136+ gert::TilingContextPara tilingContextPara(
137+ "Atan",
138+ {
139+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
140+ },
141+ {
142+ {{{1, 64, 2, 6}, {1, 64, 2, 6}}, ge::DT_FLOAT, ge::FORMAT_ND},
143+ },
144+ {}, &compileInfo);
145+ 
146+ uint64_t expectTilingKey = 7;
147+ string expectTilingData = "";
148+ std::vector<size_t> expectWorkspaces = {16777216};
149+ ExecuteTestCaseForEle(
150+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
151+}
Amath/atan/tests/ut/op_host/test_atan_infershape.cpp+69-0
@@ -0,0 +1,69 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <gtest/gtest.h>
12+#include <iostream>
13+#include "infershape_context_faker.h"
14+#include "infershape_case_executor.h"
15+ 
16+class AtanInferShape : public testing::Test {
17+protected:
18+ static void SetUpTestCase()
19+ {
20+ std::cout << "Atan InferShape SetUp" << std::endl;
21+ }
22+ 
23+ static void TearDownTestCase()
24+ {
25+ std::cout << "Atan InferShape TearDown" << std::endl;
26+ }
27+};
28+ 
29+TEST_F(AtanInferShape, atan_infershape_diff_test)
30+{
31+ gert::InfershapeContextPara infershapeContextPara(
32+ "Atan",
33+ {
34+ {{{4, 3, 4}, {4, 3, 4}}, ge::DT_FLOAT16, ge::FORMAT_ND},
35+ },
36+ {
37+ {{{}, {}}, ge::DT_FLOAT16, ge::FORMAT_ND},
38+ });
39+ std::vector<std::vector<int64_t>> expectOutputShape = {{4, 3, 4}};
40+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
41+}
42+ 
43+TEST_F(AtanInferShape, atan_infershape_same_test)
44+{
45+ gert::InfershapeContextPara infershapeContextPara(
46+ "Atan",
47+ {
48+ {{{1, 3, 4}, {1, 3, 4}}, ge::DT_FLOAT16, ge::FORMAT_ND},
49+ },
50+ {
51+ {{{}, {}}, ge::DT_FLOAT16, ge::FORMAT_ND},
52+ });
53+ std::vector<std::vector<int64_t>> expectOutputShape = {{1, 3, 4}};
54+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
55+}
56+ 
57+TEST_F(AtanInferShape, atan_infershape_dynamic_test)
58+{
59+ gert::InfershapeContextPara infershapeContextPara(
60+ "Atan",
61+ {
62+ {{{-1}, {-1}}, ge::DT_FLOAT16, ge::FORMAT_ND},
63+ },
64+ {
65+ {{{}, {}}, ge::DT_FLOAT16, ge::FORMAT_ND},
66+ });
67+ std::vector<std::vector<int64_t>> expectOutputShape = {{-1}};
68+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
69+}
Mmath/erf/CMakeLists.txt+10-15
@@ -1,20 +1,15 @@
1# ----------------------------------------------------------------------------1# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.2+# Copyright (c) 2026 Huawei Technologies Co., Ltd.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.3+# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4-# This file is a part of the CANN Open Software.4+# CANN Open Software License Agreement Version 2.0 (the "License").
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6# Please refer to the License for details. You may not use this file except in compliance with the License.5# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING6+# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
10# ----------------------------------------------------------------------------9# ----------------------------------------------------------------------------
11 10 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)11+# 设置算子定义时支持的芯片类型
13-if(NOT ENABLE_TEST AND NOT BENCHMARK)12+set(SUPPORT_COMPUTE_UNIT "ascend950" "mc62cm12a")
14- list(REMOVE_ITEM CURRENT_DIRS tests)13+# 设置每种芯片类型对应的tiling文件目录，即采用op_host目录下哪个文件夹下的tiling文件编译
15-endif()14+set(SUPPORT_TILING_DIR "arch35" "arch35")
16-foreach(SUB_DIR ${CURRENT_DIRS})15+add_all_modules_sources(OPTYPE erf ACLNNTYPE aclnn_exclude COMPUTE_UNIT ${SUPPORT_COMPUTE_UNIT} TILING_DIR ${SUPPORT_TILING_DIR} DISABLE_IN_OPP TRUE)
17- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
18- add_subdirectory(${SUB_DIR})
19- endif()
20-endforeach()
Rmath/erf/op_host/op_api/aclnn_erf.cpp→math/erf/op_api/aclnn_erf.cpp+0-0
文件重命名但无更改。
Rmath/erf/op_host/op_api/aclnn_erf.h→math/erf/op_api/aclnn_erf.h+0-0
文件重命名但无更改。
Rmath/erf/op_host/op_api/erf.cpp→math/erf/op_api/erf.cpp+0-0
文件重命名但无更改。
Rmath/erf/op_host/op_api/erf.h→math/erf/op_api/erf.h+0-0
文件重命名但无更改。
Amath/erf/op_graph/erf_proto.h+43-0
@@ -0,0 +1,43 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file erf_proto.h
13+ * \brief
14+ */
15+#ifndef OPS_OP_PROTO_INC_ERF_H_
16+#define OPS_OP_PROTO_INC_ERF_H_
17+ 
18+#include "graph/operator_reg.h"
19+#include "graph/types.h"
20+ 
21+namespace ge {
22+ 
23+/**
24+* @brief Computes the Gauss error function of 'x' element-wise. \n
25+ 
26+* @par Inputs:
27+* x: A Tensor of type bfloat16, float16, float32 or double. the format can be
28+* [NCHW,NHWC,ND]
29+ 
30+* @par Outputs:
31+* y: A Tensor. Has the same type, format and shape as 'x'. \n
32+ 
33+* @par Third-party framework compatibility
34+* Compatible with the TensorFlow operator Erf.
35+*/
36+REG_OP(Erf)
37+ .INPUT(x, TensorType({FloatingDataType, DT_BF16}))
38+ .OUTPUT(y, TensorType({FloatingDataType, DT_BF16}))
39+ .OP_END_FACTORY_REG(Erf)
40+ 
41+} // namespace ge
42+ 
43+#endif // OPS_OP_PROTO_INC_ERF_H_
Dmath/erf/op_host/CMakeLists.txt+0-12
@@ -1,12 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-add_modules_sources()
Amath/erf/op_host/arch35/erf_tiling_arch35.cpp+145-0
@@ -0,0 +1,145 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file erf_tiling_arch35.cpp
13+ * \brief
14+ */
15+#include "erf_tiling_arch35.h"
16+#include <graph/utils/type_utils.h>
17+#include "register/op_impl_registry.h"
18+#include "tiling/platform/platform_ascendc.h"
19+#include "log/log.h"
20+#include "util/math_util.h"
21+#include "platform/platform_info.h"
22+#include "op_host/tiling_util.h"
23+#include "register/tilingdata_base.h"
24+#include "math/erf/op_kernel/arch35/erf_dag.h"
25+#include "math/erf/op_kernel/arch35/erf_struct.h"
26+ 
27+#include <iostream>
28+ 
29+using namespace ErfOp;
30+using namespace Ops::Math::OpTiling;
31+ 
32+namespace optiling {
33+const size_t ASCEND_WORKSPACE = 16777216; // 16M
34+const int64_t ASCEND_API_BUFFER = 122880; // 120K
35+ 
36+ge::graphStatus ErfTiling::CalcInputDtype()
37+{
38+ OP_LOGD(tilingContext->GetNodeName(), "ErfTiling CalcInputDtype enter.");
39+ auto inputDesc = tilingContext->GetInputDesc(0);
40+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, inputDesc);
41+ this->inputDtype = inputDesc->GetDataType();
42+ OP_CHECK_IF(
43+ this->inputDtype != ge::DT_FLOAT16 && this->inputDtype != ge::DT_BF16 && this->inputDtype != ge::DT_FLOAT,
44+ OP_LOGE(
45+ tilingContext->GetNodeName(),
46+ "input x dtype [%s] not supported, only support [DT_FLOAT16, DT_BF16, DT_FLOAT]",
47+ ge::TypeUtils::DataTypeToSerialString(this->inputDtype).c_str()),
48+ return ge::GRAPH_FAILED);
49+ return ge::GRAPH_SUCCESS;
50+}
51+ 
52+ge::graphStatus ErfTiling::CalcOutputDtype()
53+{
54+ OP_LOGD(tilingContext->GetNodeName(), "ErfTiling CalcOutputDtype enter.");
55+ auto outputDesc = tilingContext->GetOutputDesc(0);
56+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, outputDesc);
57+ this->outputDtype = outputDesc->GetDataType();
58+ OP_CHECK_IF(
59+ this->outputDtype != this->inputDtype,
60+ OP_LOGE(tilingContext->GetNodeName(), "output y dtype not same as input x"), return ge::GRAPH_FAILED);
61+ return ge::GRAPH_SUCCESS;
62+}
63+ 
64+ge::graphStatus ErfTiling::CheckShape()
65+{
66+ OP_LOGD(tilingContext->GetNodeName(), "ErfTiling CheckShape enter.");
67+ auto inputStorageShape = tilingContext->GetInputShape(0);
68+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, inputStorageShape);
69+ const gert::Shape& inputXShape = EnsureNotScalar(inputStorageShape->GetStorageShape());
70+ 
71+ auto outputStorageShape = tilingContext->GetOutputShape(0);
72+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, outputStorageShape);
73+ const gert::Shape& outputYShape = EnsureNotScalar(outputStorageShape->GetStorageShape());
74+ 
75+ OP_CHECK_IF(
76+ inputXShape != outputYShape, OP_LOGE(tilingContext->GetNodeName(), "input x and output y shape not same"),
77+ return ge::GRAPH_FAILED);
78+ return ge::GRAPH_SUCCESS;
79+}
80+ 
81+ge::graphStatus ErfTiling::RunTiling()
82+{
83+ OP_LOGD(tilingContext->GetNodeName(), "ErfTiling RunTiling enter.");
84+ ElewiseBaseTiling elewiseBaseTiling(tilingContext);
85+ OP_CHECK_IF(
86+ CalcInputDtype() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "get input dtype failed"),
87+ return ge::GRAPH_FAILED);
88+ OP_CHECK_IF(
89+ CalcOutputDtype() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "get output dtype failed"),
90+ return ge::GRAPH_FAILED);
91+ OP_CHECK_IF(
92+ CheckShape() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "check shape failed"), return ge::GRAPH_FAILED);
93+ 
94+ auto tiling = tilingContext->GetTilingData<EleBaseTilingDataV2>();
95+ OP_CHECK_IF(
96+ (tiling == nullptr), OP_LOGE(tilingContext->GetNodeName(), "Get AssignAddTiling from GE context failed"),
97+ return ge::GRAPH_FAILED);
98+ 
99+ ge::graphStatus baseTilingResult = ge::GRAPH_FAILED;
100+ if (this->outputDtype == ge::DT_FLOAT16) {
101+ dType = TPL_FP16;
102+ baseTilingResult = elewiseBaseTiling.DoTiling<ErfOp::ErfDAG<half, half>::OpDag>(*tiling, ASCEND_API_BUFFER);
103+ } else if (this->outputDtype == ge::DT_BF16) {
104+ dType = TPL_BF16;
105+ baseTilingResult = elewiseBaseTiling.DoTiling<ErfOp::ErfDAG<bfloat16_t>::OpDag>(*tiling, ASCEND_API_BUFFER);
106+ } else if (this->outputDtype == ge::DT_FLOAT) {
107+ dType = TPL_FP32;
108+ baseTilingResult = elewiseBaseTiling.DoTiling<ErfOp::ErfDAG<float>::OpDag>(*tiling, ASCEND_API_BUFFER);
109+ } else {
110+ OP_LOGE(
111+ tilingContext->GetNodeName(),
112+ "output y dtype [%s] not supported, only support [DT_FLOAT16, DT_BF16, DT_FLOAT]",
113+ ge::TypeUtils::DataTypeToSerialString(this->outputDtype).c_str());
114+ return ge::GRAPH_FAILED;
115+ }
116+ OP_CHECK_IF(
117+ baseTilingResult == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "elewiseBaseTiling failed"),
118+ return ge::GRAPH_FAILED);
119+ 
120+ size_t* currentWorkspace = tilingContext->GetWorkspaceSizes(1);
121+ currentWorkspace[0] = ASCEND_WORKSPACE;
122+ 
123+ const uint64_t tilingKey = GET_TPL_TILING_KEY(static_cast<uint64_t>(tiling->scheMode), dType);
124+ OP_LOGD(tilingContext->GetNodeName(), "[TilingData] : tilingKey=%lu", tilingKey);
125+ tilingContext->SetTilingKey(tilingKey);
126+ tilingContext->SetBlockDim(tiling->blockNum);
127+ return ge::GRAPH_SUCCESS;
128+}
129+ 
130+static ge::graphStatus Tiling4Erf(gert::TilingContext* tilingContextGen)
131+{
132+ OP_LOGD(tilingContextGen->GetNodeName(), "Tiling4Erf rt2.0 is running.");
133+ auto compileInfo = reinterpret_cast<const ElewiseCompileInfo*>(tilingContextGen->GetCompileInfo());
134+ OP_CHECK_NULL_WITH_CONTEXT(tilingContextGen, compileInfo);
135+ ErfTiling baseOpTiling(tilingContextGen);
136+ return baseOpTiling.RunTiling();
137+}
138+ 
139+static ge::graphStatus TilingPrepareForErf([[maybe_unused]] gert::TilingParseContext* context)
140+{
141+ return ge::GRAPH_SUCCESS;
142+}
143+ 
144+IMPL_OP_OPTILING(Erf).Tiling(Tiling4Erf).TilingParse<ElewiseCompileInfo>(TilingPrepareForErf);
145+} // namespace optiling
Amath/erf/op_host/arch35/erf_tiling_arch35.h+42-0
@@ -0,0 +1,42 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file erf_tiling_arch35.h
13+ * \brief
14+ */
15+#ifndef OPS_MATH_ERF_TILING_ARCH35_H
16+#define OPS_MATH_ERF_TILING_ARCH35_H
17+ 
18+#include "register/tilingdata_base.h"
19+#include "atvoss/elewise/elewise_tiling.h"
20+#include "math/erf/op_kernel/arch35/erf_struct.h"
21+ 
22+namespace optiling {
23+using namespace Ops::Base;
24+ 
25+class ErfTiling {
26+public:
27+ explicit ErfTiling(gert::TilingContext* context) : tilingContext(context) {};
28+ ge::graphStatus RunTiling();
29+ 
30+protected:
31+ ge::graphStatus CalcOutputDtype();
32+ ge::graphStatus CalcInputDtype();
33+ ge::graphStatus CheckShape();
34+ 
35+private:
36+ gert::TilingContext* tilingContext;
37+ ge::DataType outputDtype;
38+ ge::DataType inputDtype;
39+ uint64_t dType = 0;
40+};
41+} // namespace optiling
42+#endif // OPS_MATH_ERF_TILING_ARCH35_H
Amath/erf/op_host/config/ascend950/erf_binary.json+92-0
@@ -0,0 +1,92 @@
1+{
2+ "op_type": "Erf",
3+ "op_list": [
4+ {
5+ "bin_filename": "Erf_edc28b7a93481966e91becc3c30d25f0",
6+ "inputs": [
7+ {
8+ "name": "x",
9+ "index": 0,
10+ "dtype": "bfloat16",
11+ "format": "ND",
12+ "paramType": "required",
13+ "shape": [
14+ -2
15+ ],
16+ "format_match_mode": "FormatAgnostic"
17+ }
18+ ],
19+ "outputs": [
20+ {
21+ "name": "y",
22+ "index": 0,
23+ "dtype": "bfloat16",
24+ "format": "ND",
25+ "paramType": "required",
26+ "shape": [
27+ -2
28+ ],
29+ "format_match_mode": "FormatAgnostic"
30+ }
31+ ]
32+ },
33+ {
34+ "bin_filename": "Erf_e73d8e0483dd2b837ef71ecbee13cf01",
35+ "inputs": [
36+ {
37+ "name": "x",
38+ "index": 0,
39+ "dtype": "float16",
40+ "format": "ND",
41+ "paramType": "required",
42+ "shape": [
43+ -2
44+ ],
45+ "format_match_mode": "FormatAgnostic"
46+ }
47+ ],
48+ "outputs": [
49+ {
50+ "name": "y",
51+ "index": 0,
52+ "dtype": "float16",
53+ "format": "ND",
54+ "paramType": "required",
55+ "shape": [
56+ -2
57+ ],
58+ "format_match_mode": "FormatAgnostic"
59+ }
60+ ]
61+ },
62+ {
63+ "bin_filename": "Erf_61f744b9d0e233e3b5200517ad55da6f",
64+ "inputs": [
65+ {
66+ "name": "x",
67+ "index": 0,
68+ "dtype": "float32",
69+ "format": "ND",
70+ "paramType": "required",
71+ "shape": [
72+ -2
73+ ],
74+ "format_match_mode": "FormatAgnostic"
75+ }
76+ ],
77+ "outputs": [
78+ {
79+ "name": "y",
80+ "index": 0,
81+ "dtype": "float32",
82+ "format": "ND",
83+ "paramType": "required",
84+ "shape": [
85+ -2
86+ ],
87+ "format_match_mode": "FormatAgnostic"
88+ }
89+ ]
90+ }
91+ ]
92+}
Amath/erf/op_host/config/ascend950/erf_simplified_key.ini+2-0
@@ -0,0 +1,2 @@
1+[Erf]
2+default=0
Amath/erf/op_host/erf_def.cpp+49-0
@@ -0,0 +1,49 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file erf_def.cpp
13+ * \brief erf def
14+ */
15+ 
16+#include <cstdint>
17+#include "register/op_def_registry.h"
18+ 
19+namespace ops {
20+class Erf : public OpDef {
21+public:
22+ explicit Erf(const char* name) : OpDef(name)
23+ {
24+ this->Input("x")
25+ .ParamType(REQUIRED)
26+ .DataType({ge::DT_BF16, ge::DT_FLOAT16, ge::DT_FLOAT})
27+ .Format({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND})
28+ .UnknownShapeFormat({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND});
29+ this->Output("y")
30+ .ParamType(REQUIRED)
31+ .DataType({ge::DT_BF16, ge::DT_FLOAT16, ge::DT_FLOAT})
32+ .Format({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND})
33+ .UnknownShapeFormat({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND});
34+ 
35+ OpAICoreConfig aicoreConfig;
36+ aicoreConfig.DynamicCompileStaticFlag(true)
37+ .DynamicFormatFlag(false)
38+ .DynamicRankSupportFlag(true)
39+ .DynamicShapeSupportFlag(true)
40+ .NeedCheckSupportFlag(false)
41+ .PrecisionReduceFlag(true)
42+ .ExtendCfgInfo("opFile.value", "erf_apt");
43+ this->AICore().AddConfig("ascend950", aicoreConfig);
44+ this->AICore().AddConfig("mc62cm12a", aicoreConfig);
45+ }
46+};
47+ 
48+OP_ADD(Erf);
49+} // namespace ops
Rmath/right_shift/op_host/op_api/right_shift.h→math/erf/op_host/erf_infershape.cpp+12-13
@@ -1,5 +1,5 @@
1/**1/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3 * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3 * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4 * CANN Open Software License Agreement Version 2.0 (the "License").4 * CANN Open Software License Agreement Version 2.0 (the "License").
5 * Please refer to the License for details. You may not use this file except in compliance with the License.5 * Please refer to the License for details. You may not use this file except in compliance with the License.
@@ -8,15 +8,14 @@
8 * See LICENSE in the root of the software repository for the full text of the License.8 * See LICENSE in the root of the software repository for the full text of the License.
9 */9 */
10 10 
11- #ifndef OP_API_INC_LEVEL0_RIGHT_SHIFT_H_11+/*!
12- #define OP_API_INC_LEVEL0_RIGHT_SHIFT_H_12+ * \file erf_infershape.cpp
13- 13+ * \brief
14- #include "opdev/op_executor.h"14+ */
15- 15+#include "infershape_elewise_util.h"
16- namespace l0op {16+#include "register/op_impl_registry.h"
17- bool IsSupported(const aclTensor* x, const aclTensor* y);17+ 
18- const aclTensor* RightShift(const aclTensor* x, const aclTensor* y, aclOpExecutor* executor);18+using namespace ge;
19- } // namespace l0op19+namespace ops {
20- 20+IMPL_OP_INFERSHAPE(Erf).InferShape(Ops::Base::InferShape4Elewise);
21- #endif // OP_API_INC_LEVEL0_RIGHT_SHIFT_H_21+} // namespace ops
22-
Amath/erf/op_kernel/arch35/erf_dag.h+42-0
@@ -0,0 +1,42 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/* !
12+ * \file erf_dag.h
13+ * \brief
14+ */
15+ 
16+#ifndef OPS_MATH_ERF_DAG_H
17+#define OPS_MATH_ERF_DAG_H
18+ 
19+#include "atvoss/util/dag.h"
20+#include "atvoss/util/vec.h"
21+#include "atvoss/util/placeholder.h"
22+ 
23+namespace ErfOp {
24+using namespace Ops::Base;
25+const int CAST_MODE_NONE = 0;
26+const int CAST_MODE_RINT = 1;
27+ 
28+template <typename U, typename T = float>
29+struct ErfDAG {
30+ using OpCopyIn0 = Bind<Vec::CopyIn<U>, Placeholder::In0<U>>;
31+ using OpCopyIn0Cast = Bind<Vec::Cast<T, U, CAST_MODE_NONE>, OpCopyIn0>;
32+ using OpResult = Bind<Vec::Erf<T>, OpCopyIn0Cast>;
33+ using OpResultCast = Bind<Vec::Cast<U, T, CAST_MODE_RINT>, OpResult>;
34+ using OpCopyOut = Bind<Vec::CopyOut<U>, Placeholder::Out0<U>, OpResultCast>;
35+ 
36+ using Outputs = Elems<OpCopyOut>;
37+ using MemCfg = MemOptCfg<MemLevel::LEVEL_2>;
38+ using OpDag = DAGSch<Outputs, void, MemCfg>;
39+};
40+} // namespace ErfOp
41+ 
42+#endif // OPS_MATH_ERF_DAG_H
Amath/erf/op_kernel/arch35/erf_struct.h+45-0
@@ -0,0 +1,45 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/* !
12+ * \file erf_struct.h
13+ * \brief
14+ */
15+ 
16+#include "ascendc/host_api/tiling/template_argument.h"
17+ 
18+#ifndef OPS_MATH_ERF_STRUCT_H_
19+#define OPS_MATH_ERF_STRUCT_H_
20+ 
21+namespace ErfOp {
22+#define TPL_FP16 1
23+#define TPL_BF16 2
24+#define TPL_FP32 3
25+ 
26+#define TPL_SCH_MODE_0 0
27+#define TPL_SCH_MODE_1 1
28+ 
29+ASCENDC_TPL_ARGS_DECL(
30+ Erf, ASCENDC_TPL_UINT_DECL(schMode, 1, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
31+ ASCENDC_TPL_DTYPE_DECL(dType, TPL_FP16, TPL_BF16, TPL_FP32));
32+ 
33+ASCENDC_TPL_SEL(
34+ ASCENDC_TPL_ARGS_SEL(
35+ ASCENDC_TPL_UINT_SEL(schMode, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
36+ ASCENDC_TPL_DTYPE_SEL(dType, TPL_FP16)),
37+ ASCENDC_TPL_ARGS_SEL(
38+ ASCENDC_TPL_UINT_SEL(schMode, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
39+ ASCENDC_TPL_DTYPE_SEL(dType, TPL_BF16)),
40+ ASCENDC_TPL_ARGS_SEL(
41+ ASCENDC_TPL_UINT_SEL(schMode, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
42+ ASCENDC_TPL_DTYPE_SEL(dType, TPL_FP32)));
43+} // namespace ErfOp
44+ 
45+#endif // OPS_MATH_ERF_STRUCT_H_
Amath/erf/op_kernel/erf_apt.cpp+48-0
@@ -0,0 +1,48 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file erf_apt.cpp
13+ * \brief z = erf(x)
14+ */
15+#include "kernel_operator.h"
16+#include "kernel_tiling/kernel_tiling.h"
17+#include "arch35/erf_dag.h"
18+#include "arch35/erf_struct.h"
19+#include "atvoss/elewise/elewise_sch.h"
20+#include "atvoss/elewise/elewise_base_struct.h"
21+ 
22+using namespace Ops::Base;
23+namespace AscendC {
24+ 
25+template <uint64_t schMode, uint64_t dType>
26+__global__ __aicore__ void erf(GM_ADDR x, GM_ADDR y, GM_ADDR workspace, GM_ADDR tiling)
27+{
28+ REGISTER_TILING_DEFAULT(EleBaseTilingDataV2);
29+ GET_TILING_DATA_WITH_STRUCT(EleBaseTilingDataV2, tilingData, tiling);
30+ KERNEL_TASK_TYPE_DEFAULT(KERNEL_TYPE_AIV_ONLY);
31+ TPipe pipe;
32+ if constexpr (dType == TPL_FP16) {
33+ ElementwiseSch<schMode, ErfOp::ErfDAG<half, half>::OpDag> sch(&tilingData, &pipe);
34+ sch.Init(x, y);
35+ sch.Process();
36+ } else if constexpr (dType == TPL_BF16) {
37+ ElementwiseSch<schMode, ErfOp::ErfDAG<bfloat16_t>::OpDag> sch(&tilingData, &pipe);
38+ sch.Init(x, y);
39+ sch.Process();
40+ } else if constexpr (dType == TPL_FP32) {
41+ ElementwiseSch<schMode, ErfOp::ErfDAG<float>::OpDag> sch(&tilingData, &pipe);
42+ sch.Init(x, y);
43+ sch.Process();
44+ }
45+ return;
46+}
47+ 
48+} // namespace AscendC
Dmath/erf/tests/CMakeLists.txt+0-17
@@ -1,17 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
13-foreach(SUB_DIR ${CURRENT_DIRS})
14- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
15- add_subdirectory(${SUB_DIR})
16- endif()
17-endforeach()
Dmath/erf/tests/ut/CMakeLists.txt+0-17
@@ -1,17 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
13-foreach(SUB_DIR ${CURRENT_DIRS})
14- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
15- add_subdirectory(${SUB_DIR})
16- endif()
17-endforeach()
Dmath/erf/tests/ut/graph_plugin/CMakeLists.txt+0-18
@@ -1,18 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
13-foreach(SUB_DIR ${CURRENT_DIRS})
14- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
15- add_subdirectory(${SUB_DIR})
16- endif()
17-endforeach()
18- 
Dmath/erf/tests/ut/graph_plugin/fusion_pass/CMakeLists.txt+0-13
@@ -1,13 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-if(PASS_UT OR (UT_TEST_ALL AND NOT AICPU_ONLY))
13-endif()
Amath/erf/tests/ut/op_api/test_aclnn_erf.cpp+352-0
@@ -0,0 +1,352 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include "gtest/gtest.h"
12+#include "math/erf/op_api/aclnn_erf.h"
13+#include "op_api_ut_common/tensor_desc.h"
14+#include "op_api_ut_common/op_api_ut.h"
15+ 
16+class l2ErfTest : public testing::Test {
17+protected:
18+ static void SetUpTestCase()
19+ {
20+ std::cout << "l2ErfTest SetUp" << std::endl;
21+ }
22+ 
23+ static void TearDownTestCase()
24+ {
25+ std::cout << "l2ErfTest TearDown" << std::endl;
26+ }
27+};
28+ 
29+// self的数据类型不在支持范围内
30+TEST_F(l2ErfTest, l2_erf_test_001)
31+{
32+ auto selfDesc = TensorDesc({2, 3}, ACL_INT8, ACL_FORMAT_ND);
33+ auto outDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
34+ 
35+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
36+ 
37+ uint64_t workspaceSize = 0;
38+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
39+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
40+}
41+ 
42+// out的数据类型不在支持范围内
43+TEST_F(l2ErfTest, l2_erf_test_002)
44+{
45+ auto selfDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
46+ auto outDesc = TensorDesc({2, 3}, ACL_INT8, ACL_FORMAT_ND);
47+ 
48+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
49+ 
50+ uint64_t workspaceSize = 0;
51+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
52+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
53+}
54+ 
55+// self和out的shape不一致
56+TEST_F(l2ErfTest, l2_erf_test_004)
57+{
58+ auto selfDesc = TensorDesc({2, 3, 5}, ACL_FLOAT, ACL_FORMAT_ND);
59+ auto outDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
60+ 
61+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
62+ 
63+ uint64_t workspaceSize = 0;
64+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
65+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
66+}
67+ 
68+// self和out的shape不一致
69+TEST_F(l2ErfTest, l2_erf_test_005)
70+{
71+ auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND);
72+ auto outDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
73+ 
74+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
75+ 
76+ uint64_t workspaceSize = 0;
77+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
78+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
79+}
80+ 
81+// 空tensor
82+TEST_F(l2ErfTest, l2_erf_test_006)
83+{
84+ auto selfDesc = TensorDesc({2, 0}, ACL_FLOAT, ACL_FORMAT_ND);
85+ auto outDesc = TensorDesc({2, 0}, ACL_FLOAT, ACL_FORMAT_ND);
86+ 
87+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
88+ 
89+ uint64_t workspaceSize = 0;
90+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
91+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
92+}
93+ 
94+// 正常路径，float32,万分之一精度不合格，用千分之一精度。
95+TEST_F(l2ErfTest, l2_erf_test_007)
96+{
97+ auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND);
98+ auto outDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.01, 0.01);
99+ 
100+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
101+ 
102+ uint64_t workspaceSize = 0;
103+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
104+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
105+}
106+ 
107+// 正常路径，float16 由于NUMPY目前不支持ERF函数，无法进行精度测试
108+TEST_F(l2ErfTest, l2_erf_test_008)
109+{
110+ auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT16, ACL_FORMAT_ND);
111+ auto outDesc = TensorDesc({2, 4}, ACL_FLOAT16, ACL_FORMAT_ND);
112+ 
113+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
114+ 
115+ uint64_t workspaceSize = 0;
116+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
117+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
118+}
119+ 
120+// self为空
121+TEST_F(l2ErfTest, l2_erf_test_009)
122+{
123+ auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_FLOAT16, ACL_FORMAT_ND);
124+ 
125+ auto ut = OP_API_UT(aclnnErf, INPUT((aclTensor*)nullptr), OUTPUT(outDesc));
126+ 
127+ uint64_t workspaceSize = 0;
128+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
129+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_NULLPTR);
130+}
131+ 
132+// out为空
133+TEST_F(l2ErfTest, l2_erf_test_010)
134+{
135+ auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_FLOAT16, ACL_FORMAT_ND);
136+ 
137+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT((aclTensor*)nullptr));
138+ 
139+ uint64_t workspaceSize = 0;
140+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
141+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_NULLPTR);
142+}
143+ 
144+// dtype int32
145+TEST_F(l2ErfTest, l2_erf_test_013)
146+{
147+ auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_INT32, ACL_FORMAT_NDHWC);
148+ auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_INT32, ACL_FORMAT_NDHWC);
149+ 
150+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
151+ 
152+ uint64_t workspaceSize = 0;
153+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
154+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
155+}
156+ 
157+// dtype int64
158+TEST_F(l2ErfTest, l2_erf_test_014)
159+{
160+ auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_INT64, ACL_FORMAT_NDHWC);
161+ auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_INT64, ACL_FORMAT_NDHWC);
162+ 
163+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
164+ 
165+ uint64_t workspaceSize = 0;
166+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
167+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
168+}
169+ 
170+// dtype int16
171+TEST_F(l2ErfTest, l2_erf_test_015)
172+{
173+ auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_INT16, ACL_FORMAT_NDHWC);
174+ auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_INT16, ACL_FORMAT_NDHWC);
175+ 
176+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
177+ 
178+ uint64_t workspaceSize = 0;
179+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
180+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
181+}
182+ 
183+// dtype int8
184+TEST_F(l2ErfTest, l2_erf_test_016)
185+{
186+ auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_INT8, ACL_FORMAT_NDHWC);
187+ auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_INT8, ACL_FORMAT_NDHWC);
188+ 
189+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
190+ 
191+ uint64_t workspaceSize = 0;
192+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
193+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
194+}
195+ 
196+// dtype uint8
197+TEST_F(l2ErfTest, l2_erf_test_017)
198+{
199+ auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_UINT8, ACL_FORMAT_NDHWC);
200+ auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_UINT8, ACL_FORMAT_NDHWC);
201+ 
202+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
203+ 
204+ uint64_t workspaceSize = 0;
205+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
206+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
207+}
208+ 
209+// dtype bool
210+TEST_F(l2ErfTest, l2_erf_test_018)
211+{
212+ auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_BOOL, ACL_FORMAT_NDHWC);
213+ auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_FLOAT, ACL_FORMAT_NDHWC);
214+ 
215+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
216+ 
217+ uint64_t workspaceSize = 0;
218+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
219+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
220+}
221+ 
222+// dtype complex64
223+TEST_F(l2ErfTest, l2_erf_test_019)
224+{
225+ auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_COMPLEX64, ACL_FORMAT_NDHWC);
226+ auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_COMPLEX64, ACL_FORMAT_NDHWC);
227+ 
228+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
229+ 
230+ uint64_t workspaceSize = 0;
231+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
232+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
233+}
234+ 
235+// dtype complex128
236+TEST_F(l2ErfTest, l2_erf_test_020)
237+{
238+ auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_COMPLEX128, ACL_FORMAT_NDHWC);
239+ auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_COMPLEX128, ACL_FORMAT_NDHWC);
240+ 
241+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
242+ 
243+ uint64_t workspaceSize = 0;
244+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
245+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
246+}
247+ 
248+// dtype undefine
249+TEST_F(l2ErfTest, l2_erf_test_021)
250+{
251+ auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_DT_UNDEFINED, ACL_FORMAT_NDHWC);
252+ auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_DT_UNDEFINED, ACL_FORMAT_NDHWC);
253+ 
254+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
255+ 
256+ uint64_t workspaceSize = 0;
257+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
258+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
259+}
260+ 
261+// format self是私有格式
262+TEST_F(l2ErfTest, l2_erf_test_025)
263+{
264+ auto selfDesc = TensorDesc({2, 4, 6, 8, 5}, ACL_FLOAT, ACL_FORMAT_NC1HWC0);
265+ auto outDesc = TensorDesc({2, 4, 6, 8}, ACL_FLOAT, ACL_FORMAT_HWCN);
266+ 
267+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
268+ 
269+ uint64_t workspaceSize = 0;
270+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
271+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
272+}
273+ 
274+// format out是私有格式
275+TEST_F(l2ErfTest, l2_erf_test_026)
276+{
277+ auto selfDesc = TensorDesc({2, 4, 6, 8}, ACL_FLOAT, ACL_FORMAT_HWCN);
278+ auto outDesc = TensorDesc({2, 4, 6, 8, 8}, ACL_FLOAT, ACL_FORMAT_NC1HWC0);
279+ 
280+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
281+ 
282+ uint64_t workspaceSize = 0;
283+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
284+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
285+}
286+ 
287+// 正常路径，float32，inplace
288+TEST_F(l2ErfTest, l2_erf_test_027)
289+{
290+ auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.01, 0.01);
291+ 
292+ auto ut = OP_API_UT(aclnnInplaceErf, INPUT(selfDesc), OUTPUT());
293+ 
294+ uint64_t workspaceSize = 0;
295+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
296+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
297+}
298+ 
299+// 非连续
300+TEST_F(l2ErfTest, l2_erf_test_029)
301+{
302+ auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND, {1, 2}, 0, {4, 2});
303+ auto outDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND, {1, 2}, 0, {4, 2}).Precision(0.01, 0.01);
304+ 
305+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
306+ 
307+ uint64_t workspaceSize = 0;
308+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
309+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
310+}
311+ 
312+// CheckShape_10D
313+TEST_F(l2ErfTest, l2_erf_test_030)
314+{
315+ auto selfDesc = TensorDesc({2, 2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND);
316+ auto outDesc = TensorDesc({2, 2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND);
317+ 
318+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
319+ 
320+ // only test GetWorkspaceSize
321+ uint64_t workspaceSize = 0;
322+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
323+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
324+}
325+ 
326+// Check dtype
327+TEST_F(l2ErfTest, l2_erf_test_031)
328+{
329+ auto selfDesc = TensorDesc({2, 4}, ACL_BOOL, ACL_FORMAT_ND);
330+ auto outDesc = TensorDesc({2, 4}, ACL_INT32, ACL_FORMAT_ND);
331+ 
332+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
333+ 
334+ // only test GetWorkspaceSize
335+ uint64_t workspaceSize = 0;
336+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
337+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
338+}
339+ 
340+// Check dtype
341+TEST_F(l2ErfTest, l2_erf_test_032)
342+{
343+ auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT16, ACL_FORMAT_ND);
344+ auto outDesc = TensorDesc({2, 4}, ACL_INT32, ACL_FORMAT_ND);
345+ 
346+ auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
347+ 
348+ // only test GetWorkspaceSize
349+ uint64_t workspaceSize = 0;
350+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
351+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
352+}
Dmath/erf/tests/ut/op_host/CMakeLists.txt+0-20
@@ -1,20 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
13-foreach(SUB_DIR ${CURRENT_DIRS})
14- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
15- add_subdirectory(${SUB_DIR})
16- endif()
17-endforeach()
18- 
19-if(OP_API_UT OR (UT_TEST_ALL AND NOT AICPU_ONLY))
20-endif()
Amath/erf/tests/ut/op_host/arch35/test_erf_tiling.cpp+151-0
@@ -0,0 +1,151 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <gtest/gtest.h>
12+#include <iostream>
13+#include "math/erf/op_host/arch35/erf_tiling_arch35.h"
14+#include "tiling_context_faker.h"
15+#include "tiling_case_executor.h"
16+#include "atvoss/elewise/elewise_tiling.h"
17+ 
18+using namespace std;
19+ 
20+class ErfTilingTest : public testing::Test {
21+protected:
22+ static void SetUpTestCase()
23+ {
24+ std::cout << "ErfTilingTest SetUp" << std::endl;
25+ }
26+ 
27+ static void TearDownTestCase()
28+ {
29+ std::cout << "ErfTilingTest TearDown" << std::endl;
30+ }
31+};
32+ 
33+TEST_F(ErfTilingTest, erf_test_tiling_fp16_input)
34+{
35+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
36+ gert::TilingContextPara tilingContextPara(
37+ "Erf",
38+ {
39+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT16, ge::FORMAT_ND},
40+ },
41+ {
42+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT16, ge::FORMAT_ND},
43+ },
44+ {}, &compileInfo);
45+ 
46+ uint64_t expectTilingKey = 3;
47+ string expectTilingData = "8192 4 17408 2048 4 1 1 2048 2048 17408 1 ";
48+ std::vector<size_t> expectWorkspaces = {16777216};
49+ ExecuteTestCaseForEle(
50+ tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
51+}
52+ 
53+TEST_F(ErfTilingTest, erf_test_tiling_bf16_input)
54+{
55+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
56+ gert::TilingContextPara tilingContextPara(
57+ "Erf",
58+ {
59+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
60+ },
61+ {
62+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
63+ },
64+ {}, &compileInfo);
65+ 
66+ uint64_t expectTilingKey = 5;
67+ string expectTilingData = "8192 4 5760 2048 4 1 1 2048 2048 5760 1 ";
68+ std::vector<size_t> expectWorkspaces = {16777216};
69+ ExecuteTestCaseForEle(
70+ tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
71+}
72+ 
73+TEST_F(ErfTilingTest, erf_test_tiling_fp32_input)
74+{
75+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
76+ gert::TilingContextPara tilingContextPara(
77+ "Erf",
78+ {
79+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
80+ },
81+ {
82+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
83+ },
84+ {}, &compileInfo);
85+ 
86+ uint64_t expectTilingKey = 7;
87+ string expectTilingData = "8192 8 8704 1024 8 1 1 1024 1024 8704 1 ";
88+ std::vector<size_t> expectWorkspaces = {16777216};
89+ ExecuteTestCaseForEle(
90+ tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
91+}
92+ 
93+TEST_F(ErfTilingTest, erf_test_tiling_invalid_input_dtype)
94+{
95+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
96+ gert::TilingContextPara tilingContextPara(
97+ "Erf",
98+ {
99+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT32, ge::FORMAT_ND},
100+ },
101+ {
102+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
103+ },
104+ {}, &compileInfo);
105+ 
106+ uint64_t expectTilingKey = 7;
107+ string expectTilingData = "";
108+ std::vector<size_t> expectWorkspaces = {16777216};
109+ ExecuteTestCaseForEle(
110+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
111+}
112+ 
113+TEST_F(ErfTilingTest, erf_test_tiling_invalid_output_dtype)
114+{
115+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
116+ gert::TilingContextPara tilingContextPara(
117+ "Erf",
118+ {
119+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
120+ },
121+ {
122+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
123+ },
124+ {}, &compileInfo);
125+ 
126+ uint64_t expectTilingKey = 7;
127+ string expectTilingData = "";
128+ std::vector<size_t> expectWorkspaces = {16777216};
129+ ExecuteTestCaseForEle(
130+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
131+}
132+ 
133+TEST_F(ErfTilingTest, erf_test_tiling_invalid_shape)
134+{
135+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
136+ gert::TilingContextPara tilingContextPara(
137+ "Erf",
138+ {
139+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
140+ },
141+ {
142+ {{{1, 64, 2, 6}, {1, 64, 2, 6}}, ge::DT_FLOAT, ge::FORMAT_ND},
143+ },
144+ {}, &compileInfo);
145+ 
146+ uint64_t expectTilingKey = 7;
147+ string expectTilingData = "";
148+ std::vector<size_t> expectWorkspaces = {16777216};
149+ ExecuteTestCaseForEle(
150+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
151+}
Dmath/erf/tests/ut/op_host/op_api/CMakeLists.txt+0-10
@@ -1,10 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
Dmath/erf/tests/ut/op_host/op_api/test_aclnn_erf.cpp+0-332
@@ -1,332 +0,0 @@
1-/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4- * CANN Open Software License Agreement Version 2.0 (the "License").
5- * Please refer to the License for details. You may not use this file except in compliance with the License.
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8- * See LICENSE in the root of the software repository for the full text of the License.
9- */
10- 
11-#include "gtest/gtest.h"
12-#include "aclnn_erf.h"
13-#include "op_api_ut_common/tensor_desc.h"
14-#include "op_api_ut_common/op_api_ut.h"
15- 
16-class l2ErfTest : public testing::Test {
17-protected:
18- static void SetUpTestCase() {
19- std::cout << "l2ErfTest SetUp" << std::endl;
20- }
21- 
22- static void TearDownTestCase() { std::cout << "l2ErfTest TearDown" << std::endl; }
23-};
24- 
25-// self的数据类型不在支持范围内
26-TEST_F(l2ErfTest, l2_erf_test_001) {
27- auto selfDesc = TensorDesc({2, 3}, ACL_INT8, ACL_FORMAT_ND);
28- auto outDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
29- 
30- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
31- 
32- uint64_t workspaceSize = 0;
33- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
34- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
35-}
36- 
37-// out的数据类型不在支持范围内
38-TEST_F(l2ErfTest, l2_erf_test_002) {
39- auto selfDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
40- auto outDesc = TensorDesc({2, 3}, ACL_INT8, ACL_FORMAT_ND);
41- 
42- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
43- 
44- uint64_t workspaceSize = 0;
45- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
46- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
47-}
48- 
49-// self和out的shape不一致
50-TEST_F(l2ErfTest, l2_erf_test_004) {
51- auto selfDesc = TensorDesc({2, 3, 5}, ACL_FLOAT, ACL_FORMAT_ND);
52- auto outDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
53- 
54- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
55- 
56- uint64_t workspaceSize = 0;
57- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
58- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
59-}
60- 
61-// self和out的shape不一致
62-TEST_F(l2ErfTest, l2_erf_test_005) {
63- auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND);
64- auto outDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
65- 
66- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
67- 
68- uint64_t workspaceSize = 0;
69- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
70- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
71-}
72- 
73-// 空tensor
74-TEST_F(l2ErfTest, l2_erf_test_006) {
75- auto selfDesc = TensorDesc({2, 0}, ACL_FLOAT, ACL_FORMAT_ND);
76- auto outDesc = TensorDesc({2, 0}, ACL_FLOAT, ACL_FORMAT_ND);
77- 
78- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
79- 
80- uint64_t workspaceSize = 0;
81- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
82- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
83- 
84-}
85- 
86-// 正常路径，float32,万分之一精度不合格，用千分之一精度。
87-TEST_F(l2ErfTest, l2_erf_test_007) {
88- auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND);
89- auto outDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.01, 0.01);
90- 
91- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
92- 
93- uint64_t workspaceSize = 0;
94- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
95- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
96- 
97-}
98- 
99-// 正常路径，float16 由于NUMPY目前不支持ERF函数，无法进行精度测试
100-TEST_F(l2ErfTest, l2_erf_test_008) {
101- auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT16, ACL_FORMAT_ND);
102- auto outDesc = TensorDesc({2, 4}, ACL_FLOAT16, ACL_FORMAT_ND);
103- 
104- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
105- 
106- uint64_t workspaceSize = 0;
107- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
108- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
109- 
110-}
111- 
112- 
113-// self为空
114-TEST_F(l2ErfTest, l2_erf_test_009) {
115- auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_FLOAT16, ACL_FORMAT_ND);
116- 
117- auto ut = OP_API_UT(aclnnErf, INPUT((aclTensor*)nullptr), OUTPUT(outDesc));
118- 
119- uint64_t workspaceSize = 0;
120- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
121- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_NULLPTR);
122-}
123- 
124-// out为空
125-TEST_F(l2ErfTest, l2_erf_test_010) {
126- auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_FLOAT16, ACL_FORMAT_ND);
127- 
128- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT((aclTensor*)nullptr));
129- 
130- uint64_t workspaceSize = 0;
131- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
132- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_NULLPTR);
133-}
134- 
135-// dtype int32
136-TEST_F(l2ErfTest, l2_erf_test_013) {
137- auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_INT32, ACL_FORMAT_NDHWC);
138- auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_INT32, ACL_FORMAT_NDHWC);
139- 
140- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
141- 
142- uint64_t workspaceSize = 0;
143- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
144- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
145-}
146- 
147-// dtype int64
148-TEST_F(l2ErfTest, l2_erf_test_014) {
149- auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_INT64, ACL_FORMAT_NDHWC);
150- auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_INT64, ACL_FORMAT_NDHWC);
151- 
152- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
153- 
154- uint64_t workspaceSize = 0;
155- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
156- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
157-}
158- 
159-// dtype int16
160-TEST_F(l2ErfTest, l2_erf_test_015) {
161- auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_INT16, ACL_FORMAT_NDHWC);
162- auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_INT16, ACL_FORMAT_NDHWC);
163- 
164- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
165- 
166- uint64_t workspaceSize = 0;
167- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
168- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
169-}
170- 
171-// dtype int8
172-TEST_F(l2ErfTest, l2_erf_test_016) {
173- auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_INT8, ACL_FORMAT_NDHWC);
174- auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_INT8, ACL_FORMAT_NDHWC);
175- 
176- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
177- 
178- uint64_t workspaceSize = 0;
179- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
180- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
181-}
182- 
183-// dtype uint8
184-TEST_F(l2ErfTest, l2_erf_test_017) {
185- auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_UINT8, ACL_FORMAT_NDHWC);
186- auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_UINT8, ACL_FORMAT_NDHWC);
187- 
188- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
189- 
190- uint64_t workspaceSize = 0;
191- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
192- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
193-}
194- 
195-// dtype bool
196-TEST_F(l2ErfTest, l2_erf_test_018) {
197- auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_BOOL, ACL_FORMAT_NDHWC);
198- auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_FLOAT, ACL_FORMAT_NDHWC);
199- 
200- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
201- 
202- uint64_t workspaceSize = 0;
203- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
204- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
205- 
206-}
207- 
208-// dtype complex64
209-TEST_F(l2ErfTest, l2_erf_test_019) {
210- auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_COMPLEX64, ACL_FORMAT_NDHWC);
211- auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_COMPLEX64, ACL_FORMAT_NDHWC);
212- 
213- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
214- 
215- uint64_t workspaceSize = 0;
216- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
217- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
218-}
219- 
220-// dtype complex128
221-TEST_F(l2ErfTest, l2_erf_test_020) {
222- auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_COMPLEX128, ACL_FORMAT_NDHWC);
223- auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_COMPLEX128, ACL_FORMAT_NDHWC);
224- 
225- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
226- 
227- uint64_t workspaceSize = 0;
228- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
229- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
230-}
231- 
232-// dtype undefine
233-TEST_F(l2ErfTest, l2_erf_test_021) {
234- auto selfDesc = TensorDesc({2, 4, 6, 7}, ACL_DT_UNDEFINED, ACL_FORMAT_NDHWC);
235- auto outDesc = TensorDesc({2, 4, 6, 7}, ACL_DT_UNDEFINED, ACL_FORMAT_NDHWC);
236- 
237- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
238- 
239- uint64_t workspaceSize = 0;
240- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
241- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
242-}
243- 
244-// format self是私有格式
245-TEST_F(l2ErfTest, l2_erf_test_025) {
246- auto selfDesc = TensorDesc({2, 4, 6, 8, 5}, ACL_FLOAT, ACL_FORMAT_NC1HWC0);
247- auto outDesc = TensorDesc({2, 4, 6, 8}, ACL_FLOAT, ACL_FORMAT_HWCN);
248- 
249- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
250- 
251- uint64_t workspaceSize = 0;
252- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
253- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
254-}
255- 
256-// format out是私有格式
257-TEST_F(l2ErfTest, l2_erf_test_026) {
258- auto selfDesc = TensorDesc({2, 4, 6, 8}, ACL_FLOAT, ACL_FORMAT_HWCN);
259- auto outDesc = TensorDesc({2, 4, 6, 8, 8}, ACL_FLOAT, ACL_FORMAT_NC1HWC0);
260- 
261- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
262- 
263- uint64_t workspaceSize = 0;
264- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
265- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
266-}
267- 
268-// 正常路径，float32，inplace
269-TEST_F(l2ErfTest, l2_erf_test_027) {
270- auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.01, 0.01);
271- 
272- auto ut = OP_API_UT(aclnnInplaceErf, INPUT(selfDesc), OUTPUT());
273- 
274- uint64_t workspaceSize = 0;
275- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
276- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
277- 
278-}
279- 
280-// 非连续
281-TEST_F(l2ErfTest, l2_erf_test_029) {
282- auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND, {1, 2}, 0, {4, 2});
283- auto outDesc = TensorDesc({2, 4}, ACL_FLOAT, ACL_FORMAT_ND, {1, 2}, 0, {4, 2}).Precision(0.01, 0.01);
284- 
285- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
286- 
287- uint64_t workspaceSize = 0;
288- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
289- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
290- 
291-}
292- 
293-// CheckShape_10D
294-TEST_F(l2ErfTest, l2_erf_test_030) {
295- auto selfDesc = TensorDesc({2, 2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND);
296- auto outDesc = TensorDesc({2, 2, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND);
297- 
298- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
299- 
300- // only test GetWorkspaceSize
301- uint64_t workspaceSize = 0;
302- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
303- EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
304-}
305- 
306-// Check dtype
307-TEST_F(l2ErfTest, l2_erf_test_031) {
308- auto selfDesc = TensorDesc({2, 4}, ACL_BOOL, ACL_FORMAT_ND);
309- auto outDesc = TensorDesc({2, 4}, ACL_INT32, ACL_FORMAT_ND);
310- 
311- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
312- 
313- // only test GetWorkspaceSize
314- uint64_t workspaceSize = 0;
315- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
316- EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
317-}
318- 
319-// Check dtype
320-TEST_F(l2ErfTest, l2_erf_test_032) {
321- auto selfDesc = TensorDesc({2, 4}, ACL_FLOAT16, ACL_FORMAT_ND);
322- auto outDesc = TensorDesc({2, 4}, ACL_INT32, ACL_FORMAT_ND);
323- 
324- auto ut = OP_API_UT(aclnnErf, INPUT(selfDesc), OUTPUT(outDesc));
325- 
326- // only test GetWorkspaceSize
327- uint64_t workspaceSize = 0;
328- aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
329- EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
330-}
331- 
332- 
Amath/erf/tests/ut/op_host/test_erf_infershape.cpp+69-0
@@ -0,0 +1,69 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <gtest/gtest.h>
12+#include <iostream>
13+#include "infershape_context_faker.h"
14+#include "infershape_case_executor.h"
15+ 
16+class ErfInferShape : public testing::Test {
17+protected:
18+ static void SetUpTestCase()
19+ {
20+ std::cout << "Erf InferShape SetUp" << std::endl;
21+ }
22+ 
23+ static void TearDownTestCase()
24+ {
25+ std::cout << "Erf InferShape TearDown" << std::endl;
26+ }
27+};
28+ 
29+TEST_F(ErfInferShape, erf_infershape_diff_test)
30+{
31+ gert::InfershapeContextPara infershapeContextPara(
32+ "Erf",
33+ {
34+ {{{4, 3, 4}, {4, 3, 4}}, ge::DT_FLOAT16, ge::FORMAT_ND},
35+ },
36+ {
37+ {{{}, {}}, ge::DT_FLOAT16, ge::FORMAT_ND},
38+ });
39+ std::vector<std::vector<int64_t>> expectOutputShape = {{4, 3, 4}};
40+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
41+}
42+ 
43+TEST_F(ErfInferShape, erf_infershape_same_test)
44+{
45+ gert::InfershapeContextPara infershapeContextPara(
46+ "Erf",
47+ {
48+ {{{1, 3, 4}, {1, 3, 4}}, ge::DT_FLOAT16, ge::FORMAT_ND},
49+ },
50+ {
51+ {{{}, {}}, ge::DT_FLOAT16, ge::FORMAT_ND},
52+ });
53+ std::vector<std::vector<int64_t>> expectOutputShape = {{1, 3, 4}};
54+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
55+}
56+ 
57+TEST_F(ErfInferShape, erf_infershape_dynamic_test)
58+{
59+ gert::InfershapeContextPara infershapeContextPara(
60+ "Erf",
61+ {
62+ {{{-1}, {-1}}, ge::DT_FLOAT16, ge::FORMAT_ND},
63+ },
64+ {
65+ {{{}, {}}, ge::DT_FLOAT16, ge::FORMAT_ND},
66+ });
67+ std::vector<std::vector<int64_t>> expectOutputShape = {{-1}};
68+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
69+}
Mmath/expm1/CMakeLists.txt+10-15
@@ -1,20 +1,15 @@
1# ----------------------------------------------------------------------------1# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.2+# Copyright (c) 2026 Huawei Technologies Co., Ltd.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.3+# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4-# This file is a part of the CANN Open Software.4+# CANN Open Software License Agreement Version 2.0 (the "License").
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6# Please refer to the License for details. You may not use this file except in compliance with the License.5# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING6+# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
10# ----------------------------------------------------------------------------9# ----------------------------------------------------------------------------
11 10 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)11+# 设置算子定义时支持的芯片类型
13-if(NOT ENABLE_TEST AND NOT BENCHMARK)12+set(SUPPORT_COMPUTE_UNIT "ascend950")
14- list(REMOVE_ITEM CURRENT_DIRS tests)13+# 设置每种芯片类型对应的tiling文件目录，即采用op_host目录下哪个文件夹下的tiling文件编译
15-endif()14+set(SUPPORT_TILING_DIR "arch35")
16-foreach(SUB_DIR ${CURRENT_DIRS})15+add_all_modules_sources(OPTYPE expm1 ACLNNTYPE aclnn_exclude COMPUTE_UNIT ${SUPPORT_COMPUTE_UNIT} TILING_DIR ${SUPPORT_TILING_DIR} DISABLE_IN_OPP TRUE)
17- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
18- add_subdirectory(${SUB_DIR})
19- endif()
20-endforeach()
Rmath/expm1/op_host/op_api/aclnn_expm1.cpp→math/expm1/op_api/aclnn_expm1.cpp+0-0
文件重命名但无更改。
Rmath/expm1/op_host/op_api/aclnn_expm1.h→math/expm1/op_api/aclnn_expm1.h+0-0
文件重命名但无更改。
Rmath/expm1/op_host/op_api/expm1.cpp→math/expm1/op_api/expm1.cpp+0-0
文件重命名但无更改。
Rmath/expm1/op_host/op_api/expm1.h→math/expm1/op_api/expm1.h+0-0
文件重命名但无更改。
Amath/expm1/op_graph/expm1_proto.h+43-0
@@ -0,0 +1,43 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file expm1_proto.h
13+ * \brief
14+ */
15+#ifndef OPS_OP_PROTO_INC_EXPM1_H_
16+#define OPS_OP_PROTO_INC_EXPM1_H_
17+ 
18+#include "graph/operator_reg.h"
19+#include "graph/types.h"
20+ 
21+namespace ge {
22+ 
23+/**
24+* @brief Computes the exp(x) - 1 element-wise, y = e^x - 1.
25+ 
26+* @par Inputs:
27+* One input:
28+* x: An ND or 5HD tensor. Support 1D~8D. Must be one of the following types:
29+* bfloat16, float16, float32, double, complex64, complex128.
30+ 
31+* @par Outputs:
32+* y: A ND Tensor of the same dtype as "x".
33+ 
34+* @par Third-party framework compatibility
35+* Compatible with TensorFlow operator Expm1.
36+*/
37+REG_OP(Expm1)
38+ .INPUT(x, TensorType::UnaryDataType())
39+ .OUTPUT(y, TensorType::UnaryDataType())
40+ .OP_END_FACTORY_REG(Expm1)
41+} // namespace ge
42+ 
43+#endif // OPS_OP_PROTO_INC_EXPM1_H_
Dmath/expm1/op_host/CMakeLists.txt+0-12
@@ -1,12 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-add_modules_sources(OPTYPE expm1 ACLNNTYPE aclnn_exclude)
Amath/expm1/op_host/arch35/expm1_tiling_arch35.cpp+162-0
@@ -0,0 +1,162 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file expm1_tiling_arch35.cpp
13+ * \brief
14+ */
15+#include "expm1_tiling_arch35.h"
16+#include <graph/utils/type_utils.h>
17+#include "register/op_impl_registry.h"
18+#include "tiling/platform/platform_ascendc.h"
19+#include "log/log.h"
20+#include "util/math_util.h"
21+#include "platform/platform_info.h"
22+#include "op_host/tiling_util.h"
23+#include "register/tilingdata_base.h"
24+#include "math/expm1/op_kernel/arch35/expm1_dag.h"
25+#include "math/expm1/op_kernel/arch35/expm1_struct.h"
26+ 
27+#include <iostream>
28+ 
29+using namespace Ops::Math::OpTiling;
30+ 
31+namespace optiling {
32+const int64_t ASCEND_WORKSPACE = 16777216; // 16M
33+const int64_t ASCEND_API_BUFFER = 122880; // 120K
34+const int64_t DCACHE_SIZE = 32768;
35+ 
36+ge::graphStatus Expm1Tiling::SetTilingData()
37+{
38+ OP_LOGD(tilingContext->GetNodeName(), "Expm1Tiling SetTilingData enter.");
39+ 
40+ size_t* currentWorkspace = tilingContext->GetWorkspaceSizes(1);
41+ currentWorkspace[0] = ASCEND_WORKSPACE;
42+ 
43+ const uint64_t tilingKey = GET_TPL_TILING_KEY(static_cast<uint64_t>(tiling_->scheMode), dType);
44+ OP_LOGD(tilingContext->GetNodeName(), "[TilingData] : tilingKey=%lu", tilingKey);
45+ tilingContext->SetTilingKey(tilingKey);
46+ tilingContext->SetBlockDim(tiling_->blockNum);
47+ 
48+ uint64_t ubSize = 0;
49+ auto platformInfo = tilingContext->GetPlatformInfo();
50+ if (platformInfo == nullptr) {
51+ auto compileInfoPtr = reinterpret_cast<const ElewiseCompileInfo*>(tilingContext->GetCompileInfo());
52+ OP_CHECK_IF(compileInfoPtr == nullptr, OP_LOGE(tilingContext, "compile info is null"), return ge::GRAPH_FAILED);
53+ ubSize = compileInfoPtr->ubSize;
54+ } else {
55+ auto ascendcPlatform = platform_ascendc::PlatformAscendC(platformInfo);
56+ uint64_t ubSizePlatForm = 0;
57+ ascendcPlatform.GetCoreMemSize(platform_ascendc::CoreMemType::UB, ubSizePlatForm);
58+ ubSize = ubSizePlatForm;
59+ }
60+ tilingContext->SetLocalMemorySize(static_cast<uint32_t>(ubSize - DCACHE_SIZE));
61+ 
62+ return ge::GRAPH_SUCCESS;
63+}
64+ 
65+ge::graphStatus Expm1Tiling::CalcInputDtype()
66+{
67+ OP_LOGD(tilingContext->GetNodeName(), "Expm1Tiling CalcInputDtype enter.");
68+ auto inputDesc = tilingContext->GetInputDesc(0);
69+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, inputDesc);
70+ this->inputDtype = inputDesc->GetDataType();
71+ OP_CHECK_IF(
72+ this->inputDtype != ge::DT_FLOAT16 && this->inputDtype != ge::DT_BF16 && this->inputDtype != ge::DT_FLOAT,
73+ OP_LOGE(
74+ tilingContext->GetNodeName(), "input x dtype[%s] dtype not support",
75+ ge::TypeUtils::DataTypeToSerialString(this->inputDtype).c_str()),
76+ return ge::GRAPH_FAILED);
77+ return ge::GRAPH_SUCCESS;
78+}
79+ 
80+ge::graphStatus Expm1Tiling::CheckShape()
81+{
82+ OP_LOGD(tilingContext->GetNodeName(), "Expm1Tiling CheckShape enter.");
83+ auto inputStorageShape = tilingContext->GetInputShape(0);
84+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, inputStorageShape);
85+ const gert::Shape& inputXShape = EnsureNotScalar(inputStorageShape->GetStorageShape());
86+ 
87+ auto outputStorageShape = tilingContext->GetOutputShape(0);
88+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, outputStorageShape);
89+ const gert::Shape& outputYShape = EnsureNotScalar(outputStorageShape->GetStorageShape());
90+ 
91+ OP_CHECK_IF(
92+ inputXShape != outputYShape, OP_LOGE(tilingContext->GetNodeName(), "input x and output y shape not same"),
93+ return ge::GRAPH_FAILED);
94+ return ge::GRAPH_SUCCESS;
95+}
96+ 
97+ge::graphStatus Expm1Tiling::CalcOutputDtype()
98+{
99+ OP_LOGD(tilingContext->GetNodeName(), "Expm1Tiling CalcOutputDtype enter.");
100+ auto outputDesc = tilingContext->GetOutputDesc(0);
101+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, outputDesc);
102+ this->outputDtype = outputDesc->GetDataType();
103+ OP_CHECK_IF(
104+ this->outputDtype != this->inputDtype,
105+ OP_LOGE(tilingContext->GetNodeName(), "output y dtype not same as input x"), return ge::GRAPH_FAILED);
106+ return ge::GRAPH_SUCCESS;
107+}
108+ 
109+ge::graphStatus Expm1Tiling::RunTiling()
110+{
111+ OP_LOGD(tilingContext->GetNodeName(), "Expm1Tiling RunTiling enter.");
112+ ElewiseBaseTiling elewiseBaseTiling(tilingContext);
113+ OP_CHECK_IF(
114+ CalcInputDtype() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "get input dtype failed"),
115+ return ge::GRAPH_FAILED);
116+ OP_CHECK_IF(
117+ CalcOutputDtype() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "get output dtype failed"),
118+ return ge::GRAPH_FAILED);
119+ OP_CHECK_IF(
120+ CheckShape() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "check shape failed"), return ge::GRAPH_FAILED);
121+ 
122+ tiling_ = tilingContext->GetTilingData<EleBaseTilingDataV2>();
123+ ge::graphStatus baseTilingResult = ge::GRAPH_FAILED;
124+ if (this->outputDtype == ge::DT_FLOAT16) {
125+ dType = TPL_FP16;
126+ baseTilingResult =
127+ elewiseBaseTiling.DoTiling<Expm1Op::Expm1DAG<half>::OpDag>(*tiling_, ASCEND_API_BUFFER + DCACHE_SIZE);
128+ } else if (this->outputDtype == ge::DT_BF16) {
129+ dType = TPL_BF16;
130+ baseTilingResult =
131+ elewiseBaseTiling.DoTiling<Expm1Op::Expm1DAG<bfloat16_t>::OpDag>(*tiling_, ASCEND_API_BUFFER + DCACHE_SIZE);
132+ } else if (this->outputDtype == ge::DT_FLOAT) {
133+ dType = TPL_FP32;
134+ baseTilingResult =
135+ elewiseBaseTiling.DoTiling<Expm1Op::Expm1DAG<float>::OpDag>(*tiling_, ASCEND_API_BUFFER + DCACHE_SIZE);
136+ } else {
137+ OP_LOGE(tilingContext->GetNodeName(), "output dtype not support");
138+ return ge::GRAPH_FAILED;
139+ }
140+ OP_CHECK_IF(
141+ baseTilingResult != ge::GRAPH_SUCCESS, OP_LOGE(tilingContext, "elewiseBaseTiling failed"),
142+ return ge::GRAPH_FAILED);
143+ 
144+ return SetTilingData();
145+}
146+ 
147+static ge::graphStatus Tiling4Expm1(gert::TilingContext* tilingContextGen)
148+{
149+ OP_LOGD(tilingContextGen->GetNodeName(), "Tiling4Expm1 rt2.0 is running.");
150+ auto compileInfo = reinterpret_cast<const ElewiseCompileInfo*>(tilingContextGen->GetCompileInfo());
151+ OP_CHECK_NULL_WITH_CONTEXT(tilingContextGen, compileInfo);
152+ Expm1Tiling baseOpTiling(tilingContextGen);
153+ return baseOpTiling.RunTiling();
154+}
155+ 
156+static ge::graphStatus TilingPrepareForExpm1([[maybe_unused]] gert::TilingParseContext* context)
157+{
158+ return ge::GRAPH_SUCCESS;
159+}
160+ 
161+IMPL_OP_OPTILING(Expm1).Tiling(Tiling4Expm1).TilingParse<ElewiseCompileInfo>(TilingPrepareForExpm1);
162+} // namespace optiling
Amath/expm1/op_host/arch35/expm1_tiling_arch35.h+44-0
@@ -0,0 +1,44 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file expm1_tiling_arch35.h
13+ * \brief
14+ */
15+#ifndef OPS_MATH_EXPM1_TILING_ARCH35_H
16+#define OPS_MATH_EXPM1_TILING_ARCH35_H
17+ 
18+#include "register/tilingdata_base.h"
19+#include "atvoss/elewise/elewise_tiling.h"
20+#include "math/expm1/op_kernel/arch35/expm1_struct.h"
21+ 
22+namespace optiling {
23+using namespace Ops::Base;
24+ 
25+class Expm1Tiling {
26+public:
27+ explicit Expm1Tiling(gert::TilingContext* context) : tilingContext(context) {};
28+ ge::graphStatus RunTiling();
29+ 
30+protected:
31+ ge::graphStatus CalcOutputDtype();
32+ ge::graphStatus CalcInputDtype();
33+ ge::graphStatus CheckShape();
34+ ge::graphStatus SetTilingData();
35+ 
36+private:
37+ EleBaseTilingDataV2* tiling_ = nullptr;
38+ gert::TilingContext* tilingContext;
39+ ge::DataType outputDtype;
40+ ge::DataType inputDtype;
41+ uint64_t dType = 0;
42+};
43+} // namespace optiling
44+#endif // OPS_MATH_EXPM1_TILING_ARCH35_H
Amath/expm1/op_host/config/ascend950/expm1_binary.json+92-0
@@ -0,0 +1,92 @@
1+{
2+ "op_type": "Expm1",
3+ "op_list": [
4+ {
5+ "bin_filename": "Expm1_d2db1a80c523e7e59a032c95969880af",
6+ "inputs": [
7+ {
8+ "name": "x",
9+ "index": 0,
10+ "dtype": "float16",
11+ "format": "ND",
12+ "paramType": "required",
13+ "shape": [
14+ -2
15+ ],
16+ "format_match_mode": "FormatAgnostic"
17+ }
18+ ],
19+ "outputs": [
20+ {
21+ "name": "y",
22+ "index": 0,
23+ "dtype": "float16",
24+ "format": "ND",
25+ "paramType": "required",
26+ "shape": [
27+ -2
28+ ],
29+ "format_match_mode": "FormatAgnostic"
30+ }
31+ ]
32+ },
33+ {
34+ "bin_filename": "Expm1_5c5e58e043224cccf51a06108010ae13",
35+ "inputs": [
36+ {
37+ "name": "x",
38+ "index": 0,
39+ "dtype": "float32",
40+ "format": "ND",
41+ "paramType": "required",
42+ "shape": [
43+ -2
44+ ],
45+ "format_match_mode": "FormatAgnostic"
46+ }
47+ ],
48+ "outputs": [
49+ {
50+ "name": "y",
51+ "index": 0,
52+ "dtype": "float32",
53+ "format": "ND",
54+ "paramType": "required",
55+ "shape": [
56+ -2
57+ ],
58+ "format_match_mode": "FormatAgnostic"
59+ }
60+ ]
61+ },
62+ {
63+ "bin_filename": "Expm1_5c5e58e043224cccf51a06108010ae17",
64+ "inputs": [
65+ {
66+ "name": "x",
67+ "index": 0,
68+ "dtype": "bfloat16",
69+ "format": "ND",
70+ "paramType": "required",
71+ "shape": [
72+ -2
73+ ],
74+ "format_match_mode": "FormatAgnostic"
75+ }
76+ ],
77+ "outputs": [
78+ {
79+ "name": "y",
80+ "index": 0,
81+ "dtype": "bfloat16",
82+ "format": "ND",
83+ "paramType": "required",
84+ "shape": [
85+ -2
86+ ],
87+ "format_match_mode": "FormatAgnostic"
88+ }
89+ ]
90+ }
91+ ]
92+}
Amath/expm1/op_host/config/ascend950/expm1_simplified_key.ini+2-0
@@ -0,0 +1,2 @@
1+[Expm1]
2+default=0
Amath/expm1/op_host/expm1_def.cpp+48-0
@@ -0,0 +1,48 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file expm1_def.cpp
13+ * \brief expm1 def
14+ */
15+ 
16+#include <cstdint>
17+#include "register/op_def_registry.h"
18+ 
19+namespace ops {
20+class Expm1 : public OpDef {
21+public:
22+ explicit Expm1(const char* name) : OpDef(name)
23+ {
24+ this->Input("x")
25+ .ParamType(REQUIRED)
26+ .DataType({ge::DT_BF16, ge::DT_FLOAT16, ge::DT_FLOAT})
27+ .Format({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND})
28+ .UnknownShapeFormat({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND});
29+ this->Output("y")
30+ .ParamType(REQUIRED)
31+ .DataType({ge::DT_BF16, ge::DT_FLOAT16, ge::DT_FLOAT})
32+ .Format({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND})
33+ .UnknownShapeFormat({ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND});
34+ 
35+ OpAICoreConfig aicoreConfig;
36+ aicoreConfig.DynamicCompileStaticFlag(true)
37+ .DynamicFormatFlag(false)
38+ .DynamicRankSupportFlag(true)
39+ .DynamicShapeSupportFlag(true)
40+ .NeedCheckSupportFlag(false)
41+ .PrecisionReduceFlag(true)
42+ .ExtendCfgInfo("opFile.value", "expm1_apt");
43+ this->AICore().AddConfig("ascend950", aicoreConfig);
44+ }
45+};
46+ 
47+OP_ADD(Expm1);
48+} // namespace ops
Amath/expm1/op_host/expm1_infershape.cpp+21-0
@@ -0,0 +1,21 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file expm1_infershape.cpp
13+ * \brief
14+ */
15+#include "infershape_elewise_util.h"
16+#include "register/op_impl_registry.h"
17+ 
18+using namespace ge;
19+namespace ops {
20+IMPL_OP_INFERSHAPE(Expm1).InferShape(Ops::Base::InferShape4Elewise);
21+} // namespace ops
Amath/expm1/op_kernel/arch35/expm1_dag.h+118-0
@@ -0,0 +1,118 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/* !
12+ * \file expm1_dag.h
13+ * \brief
14+ */
15+ 
16+#ifndef OPS_MATH_EXPM1_DAG_H
17+#define OPS_MATH_EXPM1_DAG_H
18+ 
19+#include "atvoss/util/dag.h"
20+#include "atvoss/util/vec.h"
21+#include "atvoss/util/placeholder.h"
22+#include <limits>
23+ 
24+namespace Expm1Op {
25+using namespace Ops::Base;
26+const int CAST_MODE_NONE = 0;
27+const int CAST_MODE_RINT = 1;
28+constexpr uint32_t THREAD_NUM = 1024;
29+constexpr float INV_LN2_APPROX = 1.4426950216293335f;
30+constexpr float LN2_HALF_APPROX = 0.4099999964237213f;
31+constexpr float LN2_APPROX = 0.693145751953125f;
32+constexpr float ONE_MINUS_LN2_APPROX = 0.000001428606765330187f;
33+constexpr float FLOAT_128 = 128.000000f;
34+constexpr float FLOAT_NEG_ONE = -1.00000000f;
35+constexpr float C5 = 0.008382412604987621f;
36+constexpr float C4 = 0.0013879507314413786f;
37+constexpr float C3 = 0.04166783019900322f;
38+constexpr float C2 = 0.1666639745235443f;
39+constexpr float C1 = 0.4999999403953552f;
40+constexpr float FLOAT_INF = std::numeric_limits<float>::infinity();
41+constexpr float FLOAT_NEG_25 = -25.0000000f;
42+constexpr float FLOAT_2 = 2.0000000f;
43+ 
44+#ifdef __CCE_AICORE__
45+template <typename T>
46+__simt_vf__ __aicore__
47+LAUNCH_BOUND(THREAD_NUM) inline void Expm1SimtCompute(__ubuf__ T* x, __ubuf__ T* y, const int64_t totalNum)
48+{
49+ for (int64_t i = Simt::GetThreadIdx(); i < totalNum; i += Simt::GetThreadNum()) {
50+ float f1 = x[i];
51+ float f0 = Simt::Expm1(f1);
52+ float f2 = f1 * INV_LN2_APPROX;
53+ float f3 = Simt::Round(f2);
54+ float f4 = Simt::Abs(f1);
55+ bool p1 = f4 < LN2_HALF_APPROX;
56+ float f5 = p1 ? 0.0f : f3;
57+ float f6 = -f5;
58+ float f7 = LN2_APPROX;
59+ float f8 = Simt::Fma(f6, f7, f1);
60+ float f9 = ONE_MINUS_LN2_APPROX;
61+ float f10 = Simt::Fma(f6, f9, f8);
62+ bool p2 = f5 == FLOAT_128;
63+ float f11 = f5 + FLOAT_NEG_ONE;
64+ float f12 = p2 ? f11 : f5;
65+ float f13 = C5;
66+ float f14 = C4;
67+ float f15 = Simt::Fma(f14, f10, f13);
68+ float f16 = C3;
69+ float f17 = Simt::Fma(f15, f10, f16);
70+ float f18 = C2;
71+ float f19 = Simt::Fma(f17, f10, f18);
72+ float f20 = C1;
73+ float f21 = Simt::Fma(f19, f10, f20);
74+ float f22 = f10 * f21;
75+ float f23 = Simt::Fma(f22, f10, f10);
76+ float f24 = Simt::Exp2(f12);
77+ float f25 = f24 + FLOAT_NEG_ONE;
78+ float f26 = Simt::Fma(f23, f24, f25);
79+ float f27 = f26 + f26;
80+ float f28 = p2 ? f27 : f26;
81+ bool p3 = f12 > FLOAT_128;
82+ float f29 = p3 ? FLOAT_INF : f28;
83+ bool p4 = f12 < FLOAT_NEG_25;
84+ float f30 = p4 ? FLOAT_NEG_ONE : f29;
85+ bool p5 = f1 == 0.0f;
86+ float f31 = f1 + f1;
87+ float f32 = p5 ? f31 : f30;
88+ y[i] = Simt::Abs(x[i]) > FLOAT_2 ? f0 : f32;
89+ }
90+}
91+#endif
92+ 
93+template <class T>
94+struct Expm1Custom : public Vec::ElemwiseUnaryOP<T, T> {
95+ __aicore__ inline Expm1Custom(LocalTensor<T>& dst, LocalTensor<T>& src, uint32_t count)
96+ {
97+#ifdef __CCE_AICORE__
98+ __ubuf__ T* srcAddr = (__ubuf__ T*)src.GetPhyAddr();
99+ __ubuf__ T* dstAddr = (__ubuf__ T*)dst.GetPhyAddr();
100+ Simt::VF_CALL<Expm1SimtCompute<T>>(Simt::Dim3(THREAD_NUM), srcAddr, dstAddr, count);
101+#endif
102+ }
103+};
104+ 
105+template <typename U, typename T = float>
106+struct Expm1DAG {
107+ using OpCopyIn0 = Bind<Vec::CopyIn<U>, Placeholder::In0<U>>;
108+ using OpCopyIn0Cast = Bind<Vec::Cast<T, U, CAST_MODE_NONE>, OpCopyIn0>;
109+ using OpResult1 = Bind<Expm1Custom<T>, OpCopyIn0Cast>;
110+ using OpResultCast = Bind<Vec::Cast<U, T, CAST_MODE_RINT>, OpResult1>;
111+ using OpCopyOut = Bind<Vec::CopyOut<U>, Placeholder::Out0<U>, OpResultCast>;
112+ 
113+ using Outputs = Elems<OpCopyOut>;
114+ using MemCfg = MemOptCfg<MemLevel::LEVEL_2>;
115+ using OpDag = DAGSch<Outputs, void, MemCfg>;
116+};
117+} // namespace Expm1Op
118+#endif // OPS_MATH_EXPM1_DAG_H
Amath/expm1/op_kernel/arch35/expm1_struct.h+35-0
@@ -0,0 +1,35 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/* !
12+ * \file expm1_struct.h
13+ * \brief
14+ */
15+ 
16+#include "ascendc/host_api/tiling/template_argument.h"
17+ 
18+#ifndef OPS_MATH_EXPM1_STRUCT_H_
19+#define OPS_MATH_EXPM1_STRUCT_H_
20+ 
21+#define TPL_FP16 1
22+#define TPL_BF16 2
23+#define TPL_FP32 3
24+ 
25+#define TPL_SCH_MODE_0 0
26+#define TPL_SCH_MODE_1 1
27+ 
28+ASCENDC_TPL_ARGS_DECL(
29+ Expm1, ASCENDC_TPL_UINT_DECL(schMode, 1, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
30+ ASCENDC_TPL_DTYPE_DECL(dType, TPL_FP16, TPL_BF16, TPL_FP32));
31+ 
32+ASCENDC_TPL_SEL(ASCENDC_TPL_ARGS_SEL(
33+ ASCENDC_TPL_UINT_SEL(schMode, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
34+ ASCENDC_TPL_DTYPE_SEL(dType, TPL_FP16, TPL_BF16, TPL_FP32)));
35+#endif // OPS_MATH_EXPM1_STRUCT_H_
Amath/expm1/op_kernel/expm1_apt.cpp+48-0
@@ -0,0 +1,48 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file expm1_apt.cpp
13+ * \brief z = expm1(x)
14+ */
15+#include "kernel_operator.h"
16+#include "kernel_tiling/kernel_tiling.h"
17+#include "arch35/expm1_struct.h"
18+#include "arch35/expm1_dag.h"
19+#include "atvoss/elewise/elewise_sch.h"
20+#include "atvoss/elewise/elewise_base_struct.h"
21+ 
22+using namespace Ops::Base;
23+namespace AscendC {
24+ 
25+template <uint64_t schMode, uint64_t dType>
26+__global__ __aicore__ void expm1(GM_ADDR x, GM_ADDR y, GM_ADDR workspace, GM_ADDR tiling)
27+{
28+ REGISTER_TILING_DEFAULT(EleBaseTilingDataV2);
29+ GET_TILING_DATA_WITH_STRUCT(EleBaseTilingDataV2, tilingData, tiling);
30+ KERNEL_TASK_TYPE_DEFAULT(KERNEL_TYPE_AIV_ONLY);
31+ 
32+ TPipe pipe;
33+ if constexpr (dType == TPL_FP32) {
34+ ElementwiseSch<schMode, Expm1Op::Expm1DAG<float>::OpDag> sch(&tilingData, &pipe);
35+ sch.Init(x, y);
36+ sch.Process();
37+ } else if constexpr (dType == TPL_BF16) {
38+ ElementwiseSch<schMode, Expm1Op::Expm1DAG<bfloat16_t>::OpDag> sch(&tilingData, &pipe);
39+ sch.Init(x, y);
40+ sch.Process();
41+ } else if constexpr (dType == TPL_FP16) {
42+ ElementwiseSch<schMode, Expm1Op::Expm1DAG<half>::OpDag> sch(&tilingData, &pipe);
43+ sch.Init(x, y);
44+ sch.Process();
45+ }
46+}
47+ 
48+} // namespace AscendC
Dmath/expm1/tests/CMakeLists.txt+0-17
@@ -1,17 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
13-foreach(SUB_DIR ${CURRENT_DIRS})
14- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
15- add_subdirectory(${SUB_DIR})
16- endif()
17-endforeach()
Dmath/expm1/tests/ut/CMakeLists.txt+0-17
@@ -1,17 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
13-foreach(SUB_DIR ${CURRENT_DIRS})
14- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
15- add_subdirectory(${SUB_DIR})
16- endif()
17-endforeach()
Rmath/expm1/tests/ut/op_host/op_api/test_aclnn_expm1.cpp→math/expm1/tests/ut/op_api/test_aclnn_expm1.cpp+42-41
@@ -1,18 +1,18 @@
1/**1/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * See LICENSE in the root of the software repository for the full text of the License.
9- */9+ */
10 10 
11#include <array>11#include <array>
12#include <vector>12#include <vector>
13#include "gtest/gtest.h"13#include "gtest/gtest.h"
14#include "opdev/op_log.h"14#include "opdev/op_log.h"
15-#include "aclnn_expm1.h"15+#include "math/expm1/op_api/aclnn_expm1.h"
16#include "op_api_ut_common/op_api_ut.h"16#include "op_api_ut_common/op_api_ut.h"
17#include "op_api_ut_common/scalar_desc.h"17#include "op_api_ut_common/scalar_desc.h"
18#include "op_api_ut_common/tensor_desc.h"18#include "op_api_ut_common/tensor_desc.h"
@@ -20,17 +20,22 @@
20using namespace op;20using namespace op;
21using namespace std;21using namespace std;
22 22 
23- 
24class l2_expm1_test : public testing::Test {23class l2_expm1_test : public testing::Test {
25protected:24protected:
26- static void SetUpTestCase() { std::cout << "Expm1 Test Setup" << std::endl; }25+ static void SetUpTestCase()
27- static void TearDownTestCase() { std::cout << "Expm1 Test TearDown" << std::endl; }26+ {
27+ std::cout << "Expm1 Test Setup" << std::endl;
28+ }
29+ static void TearDownTestCase()
30+ {
31+ std::cout << "Expm1 Test TearDown" << std::endl;
32+ }
28};33};
29 34 
30TEST_F(l2_expm1_test, expm1_testcase_001_normal_float32)35TEST_F(l2_expm1_test, expm1_testcase_001_normal_float32)
31{36{
32 auto outDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.0001, 0.0001);37 auto outDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.0001, 0.0001);
33- auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-1,1);38+ auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-1, 1);
34 39 
35 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));40 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
36 41 
@@ -38,14 +43,13 @@ TEST_F(l2_expm1_test, expm1_testcase_001_normal_float32)
38 uint64_t workspaceSize = 0;43 uint64_t workspaceSize = 0;
39 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);44 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
40 EXPECT_EQ(aclRet, ACL_SUCCESS);45 EXPECT_EQ(aclRet, ACL_SUCCESS);
41- 
42}46}
43 47 
44// float1648// float16
45TEST_F(l2_expm1_test, expm1_testcase_002_normal_float16)49TEST_F(l2_expm1_test, expm1_testcase_002_normal_float16)
46{50{
47 auto outDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND).Precision(0.001, 0.001);51 auto outDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND).Precision(0.001, 0.001);
48- auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND).ValueRange(-1,1);52+ auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND).ValueRange(-1, 1);
49 53 
50 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));54 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
51 55 
@@ -53,14 +57,13 @@ TEST_F(l2_expm1_test, expm1_testcase_002_normal_float16)
53 uint64_t workspaceSize = 0;57 uint64_t workspaceSize = 0;
54 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);58 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
55 EXPECT_EQ(aclRet, ACL_SUCCESS);59 EXPECT_EQ(aclRet, ACL_SUCCESS);
56- 
57}60}
58 61 
59// bfloat1662// bfloat16
60TEST_F(l2_expm1_test, ascend910B2_expm1_testcase_002_normal_float16)63TEST_F(l2_expm1_test, ascend910B2_expm1_testcase_002_normal_float16)
61{64{
62 auto outDesc = TensorDesc({2, 5}, ACL_BF16, ACL_FORMAT_ND).Precision(0.001, 0.001);65 auto outDesc = TensorDesc({2, 5}, ACL_BF16, ACL_FORMAT_ND).Precision(0.001, 0.001);
63- auto selfDesc = TensorDesc({2, 5}, ACL_BF16, ACL_FORMAT_ND).ValueRange(-1,1);66+ auto selfDesc = TensorDesc({2, 5}, ACL_BF16, ACL_FORMAT_ND).ValueRange(-1, 1);
64 67 
65 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));68 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
66 69 
@@ -68,7 +71,6 @@ TEST_F(l2_expm1_test, ascend910B2_expm1_testcase_002_normal_float16)
68 uint64_t workspaceSize = 0;71 uint64_t workspaceSize = 0;
69 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);72 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
70 EXPECT_EQ(aclRet, ACL_SUCCESS);73 EXPECT_EQ(aclRet, ACL_SUCCESS);
71- 
72}74}
73 75 
74// empty76// empty
@@ -88,7 +90,7 @@ TEST_F(l2_expm1_test, expm1_testcase_003_normal_empty_tensor)
88// CheckNotNull90// CheckNotNull
89TEST_F(l2_expm1_test, expm1_testcase_004_exception_null_out)91TEST_F(l2_expm1_test, expm1_testcase_004_exception_null_out)
90{92{
91- auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND).ValueRange(-1,1);93+ auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND).ValueRange(-1, 1);
92 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT((aclTensor*)nullptr));94 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT((aclTensor*)nullptr));
93 95 
94 // SAMPLE: only test GetWorkspaceSize96 // SAMPLE: only test GetWorkspaceSize
@@ -114,7 +116,7 @@ TEST_F(l2_expm1_test, expm1_testcase_005_exception_null_self)
114TEST_F(l2_expm1_test, expm1_testcase_007_normal_dtype_not_the_same)116TEST_F(l2_expm1_test, expm1_testcase_007_normal_dtype_not_the_same)
115{117{
116 auto outDesc = TensorDesc({5, 5}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.001, 0.001);118 auto outDesc = TensorDesc({5, 5}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.001, 0.001);
117- auto selfDesc = TensorDesc({5, 5}, ACL_FLOAT16, ACL_FORMAT_ND).ValueRange(-1,1);119+ auto selfDesc = TensorDesc({5, 5}, ACL_FLOAT16, ACL_FORMAT_ND).ValueRange(-1, 1);
118 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));120 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
119 121 
120 // SAMPLE: only test GetWorkspaceSize122 // SAMPLE: only test GetWorkspaceSize
@@ -122,14 +124,13 @@ TEST_F(l2_expm1_test, expm1_testcase_007_normal_dtype_not_the_same)
122 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);124 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
123 125 
124 EXPECT_EQ(aclRet, ACLNN_SUCCESS);126 EXPECT_EQ(aclRet, ACLNN_SUCCESS);
125- 
126}127}
127 128 
128// CheckDtype different dtype of input129// CheckDtype different dtype of input
129TEST_F(l2_expm1_test, expm1_testcase_008_normal_dtype_not_the_same)130TEST_F(l2_expm1_test, expm1_testcase_008_normal_dtype_not_the_same)
130{131{
131 auto outDesc = TensorDesc({1, 16, 1, 1}, ACL_FLOAT16, ACL_FORMAT_ND).Precision(0.001, 0.001);132 auto outDesc = TensorDesc({1, 16, 1, 1}, ACL_FLOAT16, ACL_FORMAT_ND).Precision(0.001, 0.001);
132- auto selfDesc = TensorDesc({1, 16, 1, 1}, ACL_BOOL, ACL_FORMAT_ND).ValueRange(-1,1);133+ auto selfDesc = TensorDesc({1, 16, 1, 1}, ACL_BOOL, ACL_FORMAT_ND).ValueRange(-1, 1);
133 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));134 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
134 135 
135 // SAMPLE: only test GetWorkspaceSize136 // SAMPLE: only test GetWorkspaceSize
@@ -156,7 +157,8 @@ TEST_F(l2_expm1_test, expm1_testcase_009_exception_different_shape)
156// not contiguous157// not contiguous
157TEST_F(l2_expm1_test, expm1_testcase_010_normal_not_contiguous_float)158TEST_F(l2_expm1_test, expm1_testcase_010_normal_not_contiguous_float)
158{159{
159- auto outDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND, {1, 2}, 0, {5, 2}).Value(vector<float>{1,1,1,1,1,1,1,1,1,1});160+ auto outDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND, {1, 2}, 0, {5, 2})
161+ .Value(vector<float>{1, 1, 1, 1, 1, 1, 1, 1, 1, 1});
160 auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND, {1, 2}, 0, {5, 2}).ValueRange(-1, 1);162 auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND, {1, 2}, 0, {5, 2}).ValueRange(-1, 1);
161 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));163 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
162 164 
@@ -168,7 +170,7 @@ TEST_F(l2_expm1_test, expm1_testcase_010_normal_not_contiguous_float)
168// not contiguous170// not contiguous
169TEST_F(l2_expm1_test, expm1_testcase_011_normal_not_contiguous_float16)171TEST_F(l2_expm1_test, expm1_testcase_011_normal_not_contiguous_float16)
170{172{
171- auto outDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND).Value(vector<float>{1,1,1,1,1,1,1,1,1,1});173+ auto outDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND).Value(vector<float>{1, 1, 1, 1, 1, 1, 1, 1, 1, 1});
172 auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND, {1, 2}, 0, {5, 2}).ValueRange(-1, 1);174 auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND, {1, 2}, 0, {5, 2}).ValueRange(-1, 1);
173 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));175 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
174 176 
@@ -180,7 +182,8 @@ TEST_F(l2_expm1_test, expm1_testcase_011_normal_not_contiguous_float16)
180// not contiguous182// not contiguous
181TEST_F(l2_expm1_test, expm1_testcase_012_normal_not_contiguous_float16)183TEST_F(l2_expm1_test, expm1_testcase_012_normal_not_contiguous_float16)
182{184{
183- auto outDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND, {1, 2}, 0, {5, 2}).Value(vector<float>{1,1,1,1,1,1,1,1,1,1});185+ auto outDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND, {1, 2}, 0, {5, 2})
186+ .Value(vector<float>{1, 1, 1, 1, 1, 1, 1, 1, 1, 1});
184 auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND).ValueRange(-1, 1);187 auto selfDesc = TensorDesc({2, 5}, ACL_FLOAT16, ACL_FORMAT_ND).ValueRange(-1, 1);
185 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));188 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
186 189 
@@ -218,8 +221,8 @@ TEST_F(l2_expm1_test, expm1_testcase_014_exception_complex128_dtype_not_supporte
218// largeDim221// largeDim
219TEST_F(l2_expm1_test, expm1_testcase_015_normal_large_dims)222TEST_F(l2_expm1_test, expm1_testcase_015_normal_large_dims)
220{223{
221- auto outDesc = TensorDesc({1,2,2,2,2,2,2,2,2}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.0001, 0.0001);224+ auto outDesc = TensorDesc({1, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.0001, 0.0001);
222- auto selfDesc = TensorDesc({1,2,2,2,2,2,2,2,2}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-1,1);225+ auto selfDesc = TensorDesc({1, 2, 2, 2, 2, 2, 2, 2, 2}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-1, 1);
223 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));226 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
224 227 
225 // SAMPLE: only test GetWorkspaceSize228 // SAMPLE: only test GetWorkspaceSize
@@ -231,7 +234,7 @@ TEST_F(l2_expm1_test, expm1_testcase_015_normal_large_dims)
231// empty234// empty
232TEST_F(l2_expm1_test, expm1_testcase_016_normal_self_empty_tensor)235TEST_F(l2_expm1_test, expm1_testcase_016_normal_self_empty_tensor)
233{236{
234- auto outDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-1,1);237+ auto outDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-1, 1);
235 auto selfDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND);238 auto selfDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND);
236 239 
237 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));240 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
@@ -246,7 +249,7 @@ TEST_F(l2_expm1_test, expm1_testcase_016_normal_self_empty_tensor)
246TEST_F(l2_expm1_test, expm1_testcase_017_exception_out_empty_tensor)249TEST_F(l2_expm1_test, expm1_testcase_017_exception_out_empty_tensor)
247{250{
248 auto outDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND);251 auto outDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND);
249- auto selfDesc = TensorDesc({16, 16}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-1,1);252+ auto selfDesc = TensorDesc({16, 16}, ACL_FLOAT, ACL_FORMAT_ND).ValueRange(-1, 1);
250 253 
251 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));254 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
252 255 
@@ -259,8 +262,8 @@ TEST_F(l2_expm1_test, expm1_testcase_017_exception_out_empty_tensor)
259// dtype can cast to out262// dtype can cast to out
260TEST_F(l2_expm1_test, expm1_testcase_018_can_cast_out)263TEST_F(l2_expm1_test, expm1_testcase_018_can_cast_out)
261{264{
262- auto selfDesc = TensorDesc({5,5}, ACL_INT64, ACL_FORMAT_ND).ValueRange(-1,1);265+ auto selfDesc = TensorDesc({5, 5}, ACL_INT64, ACL_FORMAT_ND).ValueRange(-1, 1);
263- auto outDesc = TensorDesc({5,5}, ACL_BOOL, ACL_FORMAT_ND);266+ auto outDesc = TensorDesc({5, 5}, ACL_BOOL, ACL_FORMAT_ND);
264 267 
265 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));268 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
266 269 
@@ -273,8 +276,8 @@ TEST_F(l2_expm1_test, expm1_testcase_018_can_cast_out)
273// dtype can cast to out276// dtype can cast to out
274TEST_F(l2_expm1_test, expm1_testcase_019_can_cast_out)277TEST_F(l2_expm1_test, expm1_testcase_019_can_cast_out)
275{278{
276- auto selfDesc = TensorDesc({5,5}, ACL_INT64, ACL_FORMAT_ND).ValueRange(-1,1);279+ auto selfDesc = TensorDesc({5, 5}, ACL_INT64, ACL_FORMAT_ND).ValueRange(-1, 1);
277- auto outDesc = TensorDesc({5,5}, ACL_FLOAT, ACL_FORMAT_ND);280+ auto outDesc = TensorDesc({5, 5}, ACL_FLOAT, ACL_FORMAT_ND);
278 281 
279 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));282 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
280 283 
@@ -287,8 +290,8 @@ TEST_F(l2_expm1_test, expm1_testcase_019_can_cast_out)
287// dtype can cast to out290// dtype can cast to out
288TEST_F(l2_expm1_test, expm1_testcase_020_can_cast_out)291TEST_F(l2_expm1_test, expm1_testcase_020_can_cast_out)
289{292{
290- auto selfDesc = TensorDesc({5,5}, ACL_BOOL, ACL_FORMAT_ND).ValueRange(-1,1);293+ auto selfDesc = TensorDesc({5, 5}, ACL_BOOL, ACL_FORMAT_ND).ValueRange(-1, 1);
291- auto outDesc = TensorDesc({5,5}, ACL_FLOAT, ACL_FORMAT_ND);294+ auto outDesc = TensorDesc({5, 5}, ACL_FLOAT, ACL_FORMAT_ND);
292 295 
293 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));296 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
294 297 
@@ -301,7 +304,7 @@ TEST_F(l2_expm1_test, expm1_testcase_020_can_cast_out)
301TEST_F(l2_expm1_test, expm1_testcase_021_normal_int64_float32)304TEST_F(l2_expm1_test, expm1_testcase_021_normal_int64_float32)
302{305{
303 auto outDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.0001, 0.0001);306 auto outDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.0001, 0.0001);
304- auto selfDesc = TensorDesc({2, 5}, ACL_INT64, ACL_FORMAT_ND).ValueRange(-1,1);307+ auto selfDesc = TensorDesc({2, 5}, ACL_INT64, ACL_FORMAT_ND).ValueRange(-1, 1);
305 308 
306 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));309 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
307 310 
@@ -309,13 +312,12 @@ TEST_F(l2_expm1_test, expm1_testcase_021_normal_int64_float32)
309 uint64_t workspaceSize = 0;312 uint64_t workspaceSize = 0;
310 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);313 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
311 EXPECT_EQ(aclRet, ACL_SUCCESS);314 EXPECT_EQ(aclRet, ACL_SUCCESS);
312- 
313}315}
314 316 
315TEST_F(l2_expm1_test, expm1_testcase_022_normal_bool_float32)317TEST_F(l2_expm1_test, expm1_testcase_022_normal_bool_float32)
316{318{
317 auto outDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.0001, 0.0001);319 auto outDesc = TensorDesc({2, 5}, ACL_FLOAT, ACL_FORMAT_ND).Precision(0.0001, 0.0001);
318- auto selfDesc = TensorDesc({2, 5}, ACL_BOOL, ACL_FORMAT_ND).ValueRange(-1,1);320+ auto selfDesc = TensorDesc({2, 5}, ACL_BOOL, ACL_FORMAT_ND).ValueRange(-1, 1);
319 321 
320 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));322 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
321 323 
@@ -323,14 +325,13 @@ TEST_F(l2_expm1_test, expm1_testcase_022_normal_bool_float32)
323 uint64_t workspaceSize = 0;325 uint64_t workspaceSize = 0;
324 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);326 aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspaceSize);
325 EXPECT_EQ(aclRet, ACL_SUCCESS);327 EXPECT_EQ(aclRet, ACL_SUCCESS);
326- 
327}328}
328 329 
329// dtype cannot cast to out330// dtype cannot cast to out
330TEST_F(l2_expm1_test, expm1_testcase_023_cannot_cast_out)331TEST_F(l2_expm1_test, expm1_testcase_023_cannot_cast_out)
331{332{
332- auto selfDesc = TensorDesc({5,5}, ACL_BOOL, ACL_FORMAT_ND).ValueRange(-1,1);333+ auto selfDesc = TensorDesc({5, 5}, ACL_BOOL, ACL_FORMAT_ND).ValueRange(-1, 1);
333- auto outDesc = TensorDesc({5,5}, ACL_INT64, ACL_FORMAT_ND);334+ auto outDesc = TensorDesc({5, 5}, ACL_INT64, ACL_FORMAT_ND);
334 335 
335 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));336 auto ut = OP_API_UT(aclnnExpm1, INPUT(selfDesc), OUTPUT(outDesc));
336 337 
Dmath/expm1/tests/ut/op_host/CMakeLists.txt+0-20
@@ -1,20 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
13-foreach(SUB_DIR ${CURRENT_DIRS})
14- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
15- add_subdirectory(${SUB_DIR})
16- endif()
17-endforeach()
18- 
19-if(OP_API_UT OR (UT_TEST_ALL AND NOT AICPU_ONLY))
20-endif()
Amath/expm1/tests/ut/op_host/arch35/test_expm1_tiling.cpp+191-0
@@ -0,0 +1,191 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <gtest/gtest.h>
12+#include <iostream>
13+#include "math/expm1/op_host/arch35/expm1_tiling_arch35.h"
14+#include "tiling_context_faker.h"
15+#include "tiling_case_executor.h"
16+#include "atvoss/elewise/elewise_tiling.h"
17+ 
18+using namespace std;
19+ 
20+class Expm1TilingTest : public testing::Test {
21+protected:
22+ static void SetUpTestCase()
23+ {
24+ std::cout << "Expm1TilingTest SetUp" << std::endl;
25+ }
26+ 
27+ static void TearDownTestCase()
28+ {
29+ std::cout << "Expm1TilingTest TearDown" << std::endl;
30+ }
31+};
32+ 
33+TEST_F(Expm1TilingTest, expm1_test_tiling_fp16_input)
34+{
35+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
36+ gert::TilingContextPara tilingContextPara(
37+ "Expm1",
38+ {
39+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT16, ge::FORMAT_ND},
40+ },
41+ {
42+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT16, ge::FORMAT_ND},
43+ },
44+ {}, &compileInfo);
45+ 
46+ uint64_t expectTilingKey = 3;
47+ string expectTilingData = "8192 4 4352 2048 4 1 1 2048 2048 4352 1 ";
48+ std::vector<size_t> expectWorkspaces = {16777216};
49+ ExecuteTestCaseForEle(
50+ tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
51+}
52+ 
53+TEST_F(Expm1TilingTest, expm1_test_tiling_bf16_input)
54+{
55+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
56+ gert::TilingContextPara tilingContextPara(
57+ "Expm1",
58+ {
59+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
60+ },
61+ {
62+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
63+ },
64+ {}, &compileInfo);
65+ 
66+ uint64_t expectTilingKey = 5;
67+ string expectTilingData = "8192 4 4352 2048 4 1 1 2048 2048 4352 1 ";
68+ std::vector<size_t> expectWorkspaces = {16777216};
69+ ExecuteTestCaseForEle(
70+ tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
71+}
72+ 
73+TEST_F(Expm1TilingTest, expm1_test_tiling_fp32_input)
74+{
75+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
76+ gert::TilingContextPara tilingContextPara(
77+ "Expm1",
78+ {
79+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
80+ },
81+ {
82+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
83+ },
84+ {}, &compileInfo);
85+ 
86+ uint64_t expectTilingKey = 7;
87+ string expectTilingData = "8192 8 6656 1024 8 1 1 1024 1024 6656 1 ";
88+ std::vector<size_t> expectWorkspaces = {16777216};
89+ ExecuteTestCaseForEle(
90+ tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
91+}
92+ 
93+TEST_F(Expm1TilingTest, expm1_test_tiling_invalid_input_dtype)
94+{
95+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
96+ gert::TilingContextPara tilingContextPara(
97+ "Expm1",
98+ {
99+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_DOUBLE, ge::FORMAT_ND},
100+ },
101+ {
102+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
103+ },
104+ {}, &compileInfo);
105+ 
106+ uint64_t expectTilingKey = 7;
107+ string expectTilingData = "";
108+ std::vector<size_t> expectWorkspaces = {16777216};
109+ ExecuteTestCaseForEle(
110+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
111+}
112+ 
113+TEST_F(Expm1TilingTest, expm1_test_tiling_invalid_output_dtype)
114+{
115+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
116+ gert::TilingContextPara tilingContextPara(
117+ "Expm1",
118+ {
119+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
120+ },
121+ {
122+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_DOUBLE, ge::FORMAT_ND},
123+ },
124+ {}, &compileInfo);
125+ 
126+ uint64_t expectTilingKey = 7;
127+ string expectTilingData = "";
128+ std::vector<size_t> expectWorkspaces = {16777216};
129+ ExecuteTestCaseForEle(
130+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
131+}
132+ 
133+TEST_F(Expm1TilingTest, expm1_test_tiling_invalid_shape)
134+{
135+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
136+ gert::TilingContextPara tilingContextPara(
137+ "Expm1",
138+ {
139+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
140+ },
141+ {
142+ {{{1, 64, 2, 32}, {1, 64, 2, 32}}, ge::DT_FLOAT, ge::FORMAT_ND},
143+ },
144+ {}, &compileInfo);
145+ 
146+ uint64_t expectTilingKey = 7;
147+ string expectTilingData = "";
148+ std::vector<size_t> expectWorkspaces = {16777216};
149+ ExecuteTestCaseForEle(
150+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
151+}
152+ 
153+TEST_F(Expm1TilingTest, expm1_test_tiling_empty_tensor)
154+{
155+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
156+ gert::TilingContextPara tilingContextPara(
157+ "Expm1",
158+ {
159+ {{{1, 0, 2, 64}, {1, 0, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
160+ },
161+ {
162+ {{{1, 0, 2, 64}, {1, 0, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
163+ },
164+ {}, &compileInfo);
165+ 
166+ uint64_t expectTilingKey = 7;
167+ string expectTilingData = "";
168+ std::vector<size_t> expectWorkspaces = {16777216};
169+ ExecuteTestCaseForEle(
170+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
171+}
172+ 
173+TEST_F(Expm1TilingTest, expm1_test_tiling_invalid_input_output_dtype)
174+{
175+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
176+ gert::TilingContextPara tilingContextPara(
177+ "Expm1",
178+ {
179+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
180+ },
181+ {
182+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
183+ },
184+ {}, &compileInfo);
185+ 
186+ uint64_t expectTilingKey = 7;
187+ string expectTilingData = "";
188+ std::vector<size_t> expectWorkspaces = {16777216};
189+ ExecuteTestCaseForEle(
190+ tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
191+}
Dmath/expm1/tests/ut/op_host/op_api/CMakeLists.txt+0-10
@@ -1,10 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
Amath/expm1/tests/ut/op_host/test_expm1_infershape.cpp+55-0
@@ -0,0 +1,55 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <gtest/gtest.h>
12+#include <iostream>
13+#include "infershape_context_faker.h"
14+#include "infershape_case_executor.h"
15+ 
16+class Expm1InferShape : public testing::Test {
17+protected:
18+ static void SetUpTestCase()
19+ {
20+ std::cout << "Expm1 InferShape SetUp" << std::endl;
21+ }
22+ 
23+ static void TearDownTestCase()
24+ {
25+ std::cout << "Expm1 InferShape TearDown" << std::endl;
26+ }
27+};
28+ 
29+TEST_F(Expm1InferShape, expm1_infershape_diff_test)
30+{
31+ gert::InfershapeContextPara infershapeContextPara(
32+ "Expm1",
33+ {
34+ {{{-1}, {-1}}, ge::DT_FLOAT16, ge::FORMAT_ND},
35+ },
36+ {
37+ {{{}, {}}, ge::DT_FLOAT16, ge::FORMAT_ND},
38+ });
39+ std::vector<std::vector<int64_t>> expectOutputShape = {{-1}};
40+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
41+}
42+ 
43+TEST_F(Expm1InferShape, expm1_infershape_static_test)
44+{
45+ gert::InfershapeContextPara infershapeContextPara(
46+ "Expm1",
47+ {
48+ {{{4, 3, 4}, {4, 3, 4}}, ge::DT_FLOAT16, ge::FORMAT_ND},
49+ },
50+ {
51+ {{{}, {}}, ge::DT_FLOAT16, ge::FORMAT_ND},
52+ });
53+ std::vector<std::vector<int64_t>> expectOutputShape = {{4, 3, 4}};
54+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
55+}
Mmath/nan_to_num/CMakeLists.txt+10-15
@@ -1,20 +1,15 @@
1# ----------------------------------------------------------------------------1# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.2+# Copyright (c) 2026 Huawei Technologies Co., Ltd.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.3+# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4-# This file is a part of the CANN Open Software.4+# CANN Open Software License Agreement Version 2.0 (the "License").
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6# Please refer to the License for details. You may not use this file except in compliance with the License.5# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING6+# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
10# ----------------------------------------------------------------------------9# ----------------------------------------------------------------------------
11 10 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)11+# 设置算子定义时支持的芯片类型
13-if(NOT ENABLE_TEST AND NOT BENCHMARK)12+set(SUPPORT_COMPUTE_UNIT "ascend950")
14- list(REMOVE_ITEM CURRENT_DIRS tests)13+# 设置每种芯片类型对应的tiling文件目录，即采用op_host目录下哪个文件夹下的tiling文件编译
15-endif()14+set(SUPPORT_TILING_DIR "arch35")
16-foreach(SUB_DIR ${CURRENT_DIRS})15+add_all_modules_sources(OPTYPE nan_to_num ACLNNTYPE aclnn_exclude COMPUTE_UNIT ${SUPPORT_COMPUTE_UNIT} TILING_DIR ${SUPPORT_TILING_DIR} DISABLE_IN_OPP TRUE)
17- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
18- add_subdirectory(${SUB_DIR})
19- endif()
20-endforeach()
Rmath/nan_to_num/op_host/op_api/aclnn_nan_to_num.cpp→math/nan_to_num/op_api/aclnn_nan_to_num.cpp+0-0
文件重命名但无更改。
Rmath/nan_to_num/op_host/op_api/aclnn_nan_to_num.h→math/nan_to_num/op_api/aclnn_nan_to_num.h+0-0
文件重命名但无更改。
Rmath/nan_to_num/op_host/op_api/nan_to_num.cpp→math/nan_to_num/op_api/nan_to_num.cpp+0-0
文件重命名但无更改。
Rmath/nan_to_num/op_host/op_api/nan_to_num.h→math/nan_to_num/op_api/nan_to_num.h+0-0
文件重命名但无更改。
Amath/nan_to_num/op_graph/nan_to_num_proto.h+42-0
@@ -0,0 +1,42 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file nan_to_num_proto.h
13+ * \brief nan_to_num proto
14+ */
15+#ifndef OPS_MATH_NAN_TO_NUM_PROTO_H_
16+#define OPS_MATH_NAN_TO_NUM_PROTO_H_
17+ 
18+#include "graph/operator_reg.h"
19+ 
20+namespace ge {
21+/**
22+ * @brief Replaces NaN, positive infinity, and negative infinity values in x with specified values.
23+ * @par Inputs
24+ * x: An ND tensor of type float16, bfloat16, float32.
25+ * @par Outputs
26+ * y: An ND tensor of the same type as x.
27+ * @par Attributes
28+ * nan: An attr of type float. The value to replace NaN values.
29+ * posinf: An attr of type float. The value to replace positive infinity values.
30+ * neginf: An attr of type float. The value to replace negative infinity values.
31+ * @par Third-party framework compatibility
32+ * Compatible with PyTorch operator nan_to_num.
33+ */
34+REG_OP(NanToNum)
35+ .INPUT(x, TensorType({DT_BF16, DT_FLOAT16, DT_FLOAT}))
36+ .OUTPUT(y, TensorType({DT_BF16, DT_FLOAT16, DT_FLOAT}))
37+ .REQUIRED_ATTR(nan, Float)
38+ .REQUIRED_ATTR(posinf, Float)
39+ .REQUIRED_ATTR(neginf, Float)
40+ .OP_END_FACTORY_REG(NanToNum)
41+} // namespace ge
42+#endif // OPS_MATH_NAN_TO_NUM_PROTO_H_
Dmath/nan_to_num/op_host/CMakeLists.txt+0-12
@@ -1,12 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-add_modules_sources()
Amath/nan_to_num/op_host/arch35/nan_to_num_tiling_arch35.cpp+178-0
@@ -0,0 +1,178 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file nan_to_num_tiling_arch35.cpp
13+ * \brief nan_to_num_tiling source file
14+ */
15+ 
16+#include "nan_to_num_tiling_arch35.h"
17+#include <graph/utils/type_utils.h>
18+#include <iostream>
19+#include "tiling/platform/platform_ascendc.h"
20+#include "register/op_impl_registry.h"
21+#include "register/tilingdata_base.h"
22+#include "op_host/tiling_util.h"
23+#include "atvoss/elewise/elewise_tiling.h"
24+#include "log/log.h"
25+#include "util/math_util.h"
26+#include "platform/platform_info.h"
27+#include "math/nan_to_num/op_kernel/arch35/nan_to_num_dag.h"
28+#include "math/nan_to_num/op_kernel/arch35/nan_to_num_struct.h"
29+ 
30+using namespace ge;
31+using namespace NanToNumOp;
32+using namespace Ops::Math::OpTiling;
33+ 
34+namespace optiling {
35+constexpr int64_t ASCEND_API_BUFFER = 122880;
36+const int64_t ASCEND_WORKSPACE = 16 * 1024 * 1024;
37+ 
38+ge::graphStatus NanToNumTiling::CalcInputDtype()
39+{
40+ OP_LOGD(tilingContext->GetNodeName(), "NanToNumTiling CalcInputDtype enter.");
41+ auto inputDesc = tilingContext->GetInputDesc(0);
42+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, inputDesc);
43+ this->inputDtype = inputDesc->GetDataType();
44+ OP_CHECK_IF(
45+ this->inputDtype != ge::DT_FLOAT16 && this->inputDtype != ge::DT_BF16 && this->inputDtype != ge::DT_FLOAT,
46+ OP_LOGE(
47+ tilingContext->GetNodeName(),
48+ "input x dtype [%s] not supported, only support [DT_FLOAT16, DT_BF16, DT_FLOAT]",
49+ ge::TypeUtils::DataTypeToSerialString(this->inputDtype).c_str()),
50+ return ge::GRAPH_FAILED);
51+ return ge::GRAPH_SUCCESS;
52+}
53+ 
54+ge::graphStatus NanToNumTiling::CalcOutputDtype()
55+{
56+ OP_LOGD(tilingContext->GetNodeName(), "NanToNumTiling CalcOutputDtype enter.");
57+ auto outputDesc = tilingContext->GetOutputDesc(0);
58+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, outputDesc);
59+ this->outputDtype = outputDesc->GetDataType();
60+ OP_CHECK_IF(
61+ this->outputDtype != this->inputDtype,
62+ OP_LOGE(tilingContext->GetNodeName(), "output y dtype not same as input x"), return ge::GRAPH_FAILED);
63+ return ge::GRAPH_SUCCESS;
64+}
65+ 
66+ge::graphStatus NanToNumTiling::CheckShape()
67+{
68+ OP_LOGD(tilingContext->GetNodeName(), "NanToNumTiling CheckShape enter.");
69+ auto inputStorageShape = tilingContext->GetInputShape(0);
70+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, inputStorageShape);
71+ const gert::Shape& inputXShape = EnsureNotScalar(inputStorageShape->GetStorageShape());
72+ 
73+ auto outputStorageShape = tilingContext->GetOutputShape(0);
74+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, outputStorageShape);
75+ const gert::Shape& outputYShape = EnsureNotScalar(outputStorageShape->GetStorageShape());
76+ 
77+ OP_CHECK_IF(
78+ inputXShape != outputYShape, OP_LOGE(tilingContext->GetNodeName(), "input x and output y shape not same"),
79+ return ge::GRAPH_FAILED);
80+ return ge::GRAPH_SUCCESS;
81+}
82+ 
83+ge::graphStatus NanToNumTiling::SetAttr()
84+{
85+ OP_LOGD(tilingContext->GetNodeName(), "NanToNumTiling SetAttr enter.");
86+ auto attrs = tilingContext->GetAttrs();
87+ OP_CHECK_NULL_WITH_CONTEXT(tilingContext, attrs);
88+ const float* nanValueAttr = attrs->GetAttrPointer<float>(NanToNumOp::PLACEHOLDER_INDEX_0);
89+ const float* posinfValueAttr = attrs->GetAttrPointer<float>(NanToNumOp::PLACEHOLDER_INDEX_1);
90+ const float* neginfValueAttr = attrs->GetAttrPointer<float>(NanToNumOp::PLACEHOLDER_INDEX_2);
91+ 
92+ OP_CHECK_IF(
93+ nanValueAttr == nullptr, OP_LOGE(tilingContext->GetNodeName(), "nan value must exist"),
94+ return ge::GRAPH_FAILED);
95+ OP_CHECK_IF(
96+ posinfValueAttr == nullptr, OP_LOGE(tilingContext->GetNodeName(), "posinf value must exist"),
97+ return ge::GRAPH_FAILED);
98+ OP_CHECK_IF(
99+ neginfValueAttr == nullptr, OP_LOGE(tilingContext->GetNodeName(), "neginf value must exist"),
100+ return ge::GRAPH_FAILED);
101+ 
102+ tiling->nan = *nanValueAttr;
103+ tiling->posinf = *posinfValueAttr;
104+ tiling->neginf = *neginfValueAttr;
105+ return ge::GRAPH_SUCCESS;
106+}
107+ 
108+ge::graphStatus NanToNumTiling::SetTilingData()
109+{
110+ OP_LOGD(tilingContext->GetNodeName(), "NanToNumTiling SetTilingData enter.");
111+ 
112+ size_t* currentWorkspace = tilingContext->GetWorkspaceSizes(1);
113+ currentWorkspace[0] = ASCEND_WORKSPACE;
114+ 
115+ const uint64_t tilingKey = GET_TPL_TILING_KEY(static_cast<uint64_t>(tiling->baseTiling.scheMode), dType);
116+ OP_LOGD(tilingContext->GetNodeName(), "[TilingData] : tilingKey=%lu", tilingKey);
117+ tilingContext->SetTilingKey(tilingKey);
118+ tilingContext->SetBlockDim(tiling->baseTiling.blockNum);
119+ return ge::GRAPH_SUCCESS;
120+}
121+ 
122+ge::graphStatus NanToNumTiling::RunTiling()
123+{
124+ OP_LOGD(tilingContext->GetNodeName(), "NanToNumTiling RunTiling enter.");
125+ ElewiseBaseTiling elewiseBaseTiling(tilingContext);
126+ tiling = tilingContext->GetTilingData<NanToNumTilingData>();
127+ 
128+ OP_CHECK_IF(
129+ CalcInputDtype() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "get input dtype failed"),
130+ return ge::GRAPH_FAILED);
131+ OP_CHECK_IF(
132+ CalcOutputDtype() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "get output dtype failed"),
133+ return ge::GRAPH_FAILED);
134+ OP_CHECK_IF(
135+ CheckShape() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "check shape failed"), return ge::GRAPH_FAILED);
136+ OP_CHECK_IF(SetAttr() == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "set Attr failed"), return ge::GRAPH_FAILED);
137+ 
138+ ge::graphStatus baseTilingResult = ge::GRAPH_FAILED;
139+ if (this->outputDtype == ge::DT_FLOAT16) {
140+ dType = TPL_FP16;
141+ baseTilingResult = elewiseBaseTiling.DoTiling<NanToNumOp::NanToNumDAG<half>::OpDag>(tiling->baseTiling);
142+ } else if (this->outputDtype == ge::DT_BF16) {
143+ dType = TPL_BF16;
144+ baseTilingResult = elewiseBaseTiling.DoTiling<NanToNumOp::NanToNumDAG<bfloat16_t>::OpDag>(tiling->baseTiling);
145+ } else if (this->outputDtype == ge::DT_FLOAT) {
146+ dType = TPL_FP32;
147+ baseTilingResult = elewiseBaseTiling.DoTiling<NanToNumOp::NanToNumDAG<float>::OpDag>(tiling->baseTiling);
148+ } else {
149+ OP_LOGE(
150+ tilingContext->GetNodeName(),
151+ "output y dtype [%s] not supported, only support [DT_FLOAT16, DT_BF16, DT_FLOAT]",
152+ ge::TypeUtils::DataTypeToSerialString(this->outputDtype).c_str());
153+ return ge::GRAPH_FAILED;
154+ }
155+ OP_CHECK_IF(
156+ baseTilingResult == ge::GRAPH_FAILED, OP_LOGE(tilingContext, "elewiseBaseTiling failed"),
157+ return ge::GRAPH_FAILED);
158+ 
159+ return SetTilingData();
160+}
161+ 
162+static ge::graphStatus Tiling4NanToNum(gert::TilingContext* context)
163+{
164+ OP_LOGD(context->GetNodeName(), "Tiling4NanToNum rt2.0 is running.");
165+ auto compileInfo = reinterpret_cast<const ElewiseCompileInfo*>(context->GetCompileInfo());
166+ OP_CHECK_NULL_WITH_CONTEXT(context, compileInfo);
167+ NanToNumTiling baseOpTiling(context);
168+ return baseOpTiling.RunTiling();
169+}
170+ 
171+static ge::graphStatus TilingPrepareForNanToNum([[maybe_unused]] gert::TilingParseContext* context)
172+{
173+ return ge::GRAPH_SUCCESS;
174+}
175+ 
176+IMPL_OP_OPTILING(NanToNum).Tiling(Tiling4NanToNum).TilingParse<ElewiseCompileInfo>(TilingPrepareForNanToNum);
177+ 
178+} // namespace optiling
Amath/nan_to_num/op_host/arch35/nan_to_num_tiling_arch35.h+44-0
@@ -0,0 +1,44 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file nan_to_num_tiling_arch35.h
13+ * \brief nan_to_num_tiling head file
14+ */
15+#ifndef OPS_MATH_NAN_TO_NUM_TILING_ARCH35_H
16+#define OPS_MATH_NAN_TO_NUM_TILING_ARCH35_H
17+ 
18+#include "register/tilingdata_base.h"
19+#include "atvoss/elewise/elewise_tiling.h"
20+#include "math/nan_to_num/op_kernel/arch35/nan_to_num_tiling_struct.h"
21+ 
22+namespace optiling {
23+using namespace NanToNum;
24+class NanToNumTiling {
25+public:
26+ explicit NanToNumTiling(gert::TilingContext* context) : tilingContext(context) {};
27+ ge::graphStatus RunTiling();
28+ NanToNumTilingData* tiling = nullptr;
29+ 
30+protected:
31+ ge::graphStatus CalcOutputDtype();
32+ ge::graphStatus CalcInputDtype();
33+ ge::graphStatus CheckShape();
34+ ge::graphStatus SetAttr();
35+ ge::graphStatus SetTilingData();
36+ 
37+private:
38+ uint64_t dType = 0;
39+ gert::TilingContext* tilingContext;
40+ ge::DataType outputDtype;
41+ ge::DataType inputDtype;
42+};
43+} // namespace optiling
44+#endif // OPS_MATH_NAN_TO_NUM_TILING_ARCH35_H
Amath/nan_to_num/op_host/config/ascend950/nan_to_num_binary.json+143-0
@@ -0,0 +1,143 @@
1+{
2+ "op_type": "NanToNum",
3+ "op_list": [
4+ {
5+ "bin_filename": "NanToNum_1380099fab80e737f7e8ae05be4056ee",
6+ "inputs": [
7+ {
8+ "name": "x",
9+ "index": 0,
10+ "dtype": "bfloat16",
11+ "format": "ND",
12+ "paramType": "required",
13+ "shape": [
14+ -2
15+ ],
16+ "format_match_mode": "FormatAgnostic"
17+ }
18+ ],
19+ "outputs": [
20+ {
21+ "name": "y",
22+ "index": 0,
23+ "dtype": "bfloat16",
24+ "format": "ND",
25+ "paramType": "required",
26+ "shape": [
27+ -2
28+ ],
29+ "format_match_mode": "FormatAgnostic"
30+ }
31+ ],
32+ "attrs": [
33+ {
34+ "name": "nan",
35+ "dtype": "float",
36+ "value": null
37+ },
38+ {
39+ "name": "posinf",
40+ "dtype": "float",
41+ "value": null
42+ },
43+ {
44+ "name": "neginf",
45+ "dtype": "float",
46+ "value": null
47+ }
48+ ]
49+ },
50+ {
51+ "bin_filename": "NanToNum_a75a498177a4fc56db6ce7bd9fe08211",
52+ "inputs": [
53+ {
54+ "name": "x",
55+ "index": 0,
56+ "dtype": "float16",
57+ "format": "ND",
58+ "paramType": "required",
59+ "shape": [
60+ -2
61+ ],
62+ "format_match_mode": "FormatAgnostic"
63+ }
64+ ],
65+ "outputs": [
66+ {
67+ "name": "y",
68+ "index": 0,
69+ "dtype": "float16",
70+ "format": "ND",
71+ "paramType": "required",
72+ "shape": [
73+ -2
74+ ],
75+ "format_match_mode": "FormatAgnostic"
76+ }
77+ ],
78+ "attrs": [
79+ {
80+ "name": "nan",
81+ "dtype": "float",
82+ "value": null
83+ },
84+ {
85+ "name": "posinf",
86+ "dtype": "float",
87+ "value": null
88+ },
89+ {
90+ "name": "neginf",
91+ "dtype": "float",
92+ "value": null
93+ }
94+ ]
95+ },
96+ {
97+ "bin_filename": "NanToNum_1266cdadbc347490c57865a77cde9902",
98+ "inputs": [
99+ {
100+ "name": "x",
101+ "index": 0,
102+ "dtype": "float32",
103+ "format": "ND",
104+ "paramType": "required",
105+ "shape": [
106+ -2
107+ ],
108+ "format_match_mode": "FormatAgnostic"
109+ }
110+ ],
111+ "outputs": [
112+ {
113+ "name": "y",
114+ "index": 0,
115+ "dtype": "float32",
116+ "format": "ND",
117+ "paramType": "required",
118+ "shape": [
119+ -2
120+ ],
121+ "format_match_mode": "FormatAgnostic"
122+ }
123+ ],
124+ "attrs": [
125+ {
126+ "name": "nan",
127+ "dtype": "float",
128+ "value": null
129+ },
130+ {
131+ "name": "posinf",
132+ "dtype": "float",
133+ "value": null
134+ },
135+ {
136+ "name": "neginf",
137+ "dtype": "float",
138+ "value": null
139+ }
140+ ]
141+ }
142+ ]
143+ }
Amath/nan_to_num/op_host/config/ascend950/nan_to_num_simplified_key.ini+2-0
@@ -0,0 +1,2 @@
1+[NanToNum]
2+default=0
Amath/nan_to_num/op_host/nan_to_num_def.cpp+53-0
@@ -0,0 +1,53 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file nan_to_num_def.cpp
13+ * \brief NanToNum def
14+ */
15+ 
16+#include <cstdint>
17+#include "register/op_def_registry.h"
18+ 
19+namespace ops {
20+static const std::vector<ge::DataType> nanToNumDataType = {ge::DT_BF16, ge::DT_FLOAT16, ge::DT_FLOAT};
21+static const std::vector<ge::Format> nanToNumFormat = {ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND};
22+class NanToNum : public OpDef {
23+public:
24+ explicit NanToNum(const char* name) : OpDef(name)
25+ {
26+ this->Input("x")
27+ .ParamType(REQUIRED)
28+ .DataType(nanToNumDataType)
29+ .Format(nanToNumFormat)
30+ .UnknownShapeFormat(nanToNumFormat);
31+ this->Output("y")
32+ .ParamType(REQUIRED)
33+ .DataType(nanToNumDataType)
34+ .Format(nanToNumFormat)
35+ .UnknownShapeFormat(nanToNumFormat);
36+ this->Attr("nan").AttrType(REQUIRED).Float();
37+ this->Attr("posinf").AttrType(REQUIRED).Float();
38+ this->Attr("neginf").AttrType(REQUIRED).Float();
39+ 
40+ OpAICoreConfig aicoreConfig;
41+ aicoreConfig.DynamicCompileStaticFlag(true)
42+ .DynamicFormatFlag(false)
43+ .DynamicRankSupportFlag(true)
44+ .DynamicShapeSupportFlag(true)
45+ .NeedCheckSupportFlag(false)
46+ .PrecisionReduceFlag(true)
47+ .ExtendCfgInfo("opFile.value", "nan_to_num_apt");
48+ this->AICore().AddConfig("ascend950", aicoreConfig);
49+ }
50+};
51+ 
52+OP_ADD(NanToNum);
53+} // namespace ops
Amath/nan_to_num/op_host/nan_to_num_infershape.cpp+21-0
@@ -0,0 +1,21 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file nan_to_num_infershape.cpp
13+ * \brief NanToNum infershape
14+ */
15+ 
16+#include "infershape_elewise_util.h"
17+#include "register/op_impl_registry.h"
18+using namespace ge;
19+namespace ops {
20+IMPL_OP_INFERSHAPE(NanToNum).InferShape(Ops::Base::InferShape4Elewise);
21+} // namespace ops
Amath/nan_to_num/op_kernel/arch35/nan_to_num_dag.h+133-0
@@ -0,0 +1,133 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file nan_to_num_dag.h
13+ * \brief nan_to_num dag
14+ */
15+ 
16+#ifndef OPS_MATH_NAN_TO_NUM_DAG_H
17+#define OPS_MATH_NAN_TO_NUM_DAG_H
18+ 
19+#include "atvoss/util/dag.h"
20+#include "atvoss/util/vec.h"
21+#include "atvoss/util/placeholder.h"
22+ 
23+#ifdef __CCE_AICORE__
24+constexpr static AscendC::MicroAPI::CastTrait castTrait0 = {
25+ AscendC::MicroAPI::RegLayout::ZERO, AscendC::MicroAPI::SatMode::UNKNOWN, AscendC::MicroAPI::MaskMergeMode::ZEROING,
26+ AscendC::RoundMode::UNKNOWN};
27+constexpr static AscendC::MicroAPI::CastTrait castTrait1 = {
28+ AscendC::MicroAPI::RegLayout::ZERO, AscendC::MicroAPI::SatMode::NO_SAT, AscendC::MicroAPI::MaskMergeMode::ZEROING,
29+ AscendC::RoundMode::CAST_RINT};
30+#endif
31+ 
32+namespace NanToNumOp {
33+using namespace Ops::Base;
34+const int PLACEHOLDER_INDEX_0 = 0;
35+const int PLACEHOLDER_INDEX_1 = 1;
36+const int PLACEHOLDER_INDEX_2 = 2;
37+ 
38+const uint32_t MAX_VALUE_FP32 = 0x7f800000;
39+const uint32_t MIN_VALUE_FP32 = 0xff800000;
40+ 
41+template <class T>
42+struct NanToNumCustom : public Vec::ElemwiseQuaternaryOP<T, T, float, float, float> {
43+ __aicore__ inline NanToNumCustom(
44+ LocalTensor<T>& dst, LocalTensor<T>& src, float nan, float posinf, float neginf, uint32_t count)
45+ {
46+#ifdef __CCE_AICORE__
47+ uint32_t dtypeSize = sizeof(float);
48+ uint32_t vl = VECTOR_REG_WIDTH / dtypeSize;
49+ uint16_t loopNum = (count + vl - 1) / vl;
50+ uint32_t vlSize = vl;
51+ 
52+ T maxValue = 0;
53+ T minValue = 0;
54+ maxValue = *reinterpret_cast<const float*>(&MAX_VALUE_FP32);
55+ minValue = *reinterpret_cast<const float*>(&MIN_VALUE_FP32);
56+ __ubuf__ T* srcAddr = (__ubuf__ T*)src.GetPhyAddr();
57+ __ubuf__ T* dstAddr = (__ubuf__ T*)dst.GetPhyAddr();
58+ MicroAPI::RegTensor<float, MicroAPI::RegTraitNumOne> vregInput;
59+ MicroAPI::RegTensor<float, MicroAPI::RegTraitNumOne> vregOutput;
60+ MicroAPI::RegTensor<T, MicroAPI::RegTraitNumOne> vregInput16;
61+ MicroAPI::RegTensor<T, MicroAPI::RegTraitNumOne> vregOutput16;
62+ MicroAPI::RegTensor<float> nanTensor;
63+ MicroAPI::RegTensor<float> posinfTensor;
64+ MicroAPI::RegTensor<float> neginfTensor;
65+ MicroAPI::MaskReg mask, cmpMaskNan, cmpMaskPosinf, cmpMaskNeginf;
66+ if constexpr (std::is_same_v<T, float>) {
67+ __VEC_SCOPE__
68+ {
69+ MicroAPI::Duplicate(nanTensor, nan);
70+ MicroAPI::Duplicate(posinfTensor, posinf);
71+ MicroAPI::Duplicate(neginfTensor, neginf);
72+ 
73+ for (uint16_t loopIdx = 0; loopIdx < loopNum; loopIdx++) {
74+ mask = MicroAPI::UpdateMask<float, MicroAPI::RegTraitNumOne>(count);
75+ // OpCopyIn
76+ MicroAPI::DataCopy<T, MicroAPI::LoadDist::DIST_NORM>(
77+ vregInput, (__ubuf__ T*)(srcAddr + loopIdx * vlSize));
78+ MicroAPI::Compare<T, CMPMODE::NE>(cmpMaskNan, vregInput, vregInput, mask);
79+ MicroAPI::CompareScalar<T, CMPMODE::EQ>(cmpMaskPosinf, vregInput, maxValue, mask);
80+ MicroAPI::CompareScalar<T, CMPMODE::EQ>(cmpMaskNeginf, vregInput, minValue, mask);
81+ MicroAPI::Select<T>(vregOutput, nanTensor, vregInput, cmpMaskNan);
82+ MicroAPI::Select<T>(vregOutput, posinfTensor, vregOutput, cmpMaskPosinf);
83+ MicroAPI::Select<T>(vregOutput, neginfTensor, vregOutput, cmpMaskNeginf);
84+ // OpCopyOut
85+ MicroAPI::DataCopy<T, MicroAPI::StoreDist::DIST_NORM_B32>(
86+ (__ubuf__ T*)(dstAddr + loopIdx * vlSize), vregOutput, mask);
87+ }
88+ }
89+ } else {
90+ __VEC_SCOPE__
91+ {
92+ MicroAPI::Duplicate(nanTensor, nan);
93+ MicroAPI::Duplicate(posinfTensor, posinf);
94+ MicroAPI::Duplicate(neginfTensor, neginf);
95+ 
96+ for (uint16_t loopIdx = 0; loopIdx < loopNum; loopIdx++) {
97+ mask = MicroAPI::UpdateMask<float, MicroAPI::RegTraitNumOne>(count);
98+ // OpCopyIn
99+ MicroAPI::DataCopy<T, MicroAPI::LoadDist::DIST_UNPACK_B16>(
100+ vregInput16, (__ubuf__ T*)(srcAddr + loopIdx * vlSize));
101+ MicroAPI::Cast<float, T, castTrait0>(vregInput, vregInput16, mask);
102+ MicroAPI::Compare<float, CMPMODE::NE>(cmpMaskNan, vregInput, vregInput, mask);
103+ MicroAPI::CompareScalar<float, CMPMODE::EQ>(cmpMaskPosinf, vregInput, maxValue, mask);
104+ MicroAPI::CompareScalar<float, CMPMODE::EQ>(cmpMaskNeginf, vregInput, minValue, mask);
105+ MicroAPI::Select<float>(vregOutput, nanTensor, vregInput, cmpMaskNan);
106+ MicroAPI::Select<float>(vregOutput, posinfTensor, vregOutput, cmpMaskPosinf);
107+ MicroAPI::Select<float>(vregOutput, neginfTensor, vregOutput, cmpMaskNeginf);
108+ MicroAPI::Cast<T, float, castTrait1>(vregOutput16, vregOutput, mask);
109+ // OpCopyOut
110+ MicroAPI::DataCopy<T, MicroAPI::StoreDist::DIST_PACK_B32>(
111+ (__ubuf__ T*)(dstAddr + loopIdx * vlSize), vregOutput16, mask);
112+ }
113+ }
114+ }
115+ 
116+#endif
117+ }
118+};
119+ 
120+template <typename U>
121+struct NanToNumDAG {
122+ using OpCopyIn0 = Bind<Vec::CopyIn<U>, Placeholder::In0<U>>;
123+ using OpResult = Bind<
124+ NanToNumCustom<U>, OpCopyIn0, Placeholder::Var<float, PLACEHOLDER_INDEX_0>,
125+ Placeholder::Var<float, PLACEHOLDER_INDEX_1>, Placeholder::Var<float, PLACEHOLDER_INDEX_2>>;
126+ using OpCopyOut = Bind<Vec::CopyOut<U>, Placeholder::Out0<U>, OpResult>;
127+ using Outputs = Elems<OpCopyOut>;
128+ using MemCfg = MemOptCfg<MemLevel::LEVEL_2>;
129+ using OpDag = DAGSch<Outputs, void, MemCfg>;
130+};
131+} // namespace NanToNumOp
132+ 
133+#endif // OPS_MATH_NAN_TO_NUM_DAG_H
Amath/nan_to_num/op_kernel/arch35/nan_to_num_struct.h+44-0
@@ -0,0 +1,44 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file nan_to_num_struct.h
13+ * \brief nan_to_num_struct
14+ */
15+ 
16+#include "ascendc/host_api/tiling/template_argument.h"
17+ 
18+#ifndef OPS_MATH_NAN_TO_NUM_STRUCT_H_
19+#define OPS_MATH_NAN_TO_NUM_STRUCT_H_
20+ 
21+namespace NanToNumOp {
22+#define TPL_FP16 1
23+#define TPL_BF16 2
24+#define TPL_FP32 3
25+ 
26+#define TPL_SCH_MODE_0 0
27+#define TPL_SCH_MODE_1 1
28+ 
29+ASCENDC_TPL_ARGS_DECL(
30+ NanToNum, ASCENDC_TPL_UINT_DECL(schMode, 1, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
31+ ASCENDC_TPL_DTYPE_DECL(dType, TPL_FP16, TPL_BF16, TPL_FP32));
32+ 
33+ASCENDC_TPL_SEL(
34+ ASCENDC_TPL_ARGS_SEL(
35+ ASCENDC_TPL_UINT_SEL(schMode, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
36+ ASCENDC_TPL_DTYPE_SEL(dType, TPL_FP16)),
37+ ASCENDC_TPL_ARGS_SEL(
38+ ASCENDC_TPL_UINT_SEL(schMode, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
39+ ASCENDC_TPL_DTYPE_SEL(dType, TPL_BF16)),
40+ ASCENDC_TPL_ARGS_SEL(
41+ ASCENDC_TPL_UINT_SEL(schMode, ASCENDC_TPL_UI_LIST, TPL_SCH_MODE_0, TPL_SCH_MODE_1),
42+ ASCENDC_TPL_DTYPE_SEL(dType, TPL_FP32)));
43+} // namespace NanToNumOp
44+#endif // OPS_MATH_NAN_TO_NUM_STRUCT_H_
Amath/nan_to_num/op_kernel/arch35/nan_to_num_tiling_struct.h+24-0
@@ -0,0 +1,24 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#ifndef OPS_MATH_NAN_TO_NUM_TILING_STRUCT_H
12+#define OPS_MATH_NAN_TO_NUM_TILING_STRUCT_H
13+#include "atvoss/elewise/elewise_base_struct.h"
14+ 
15+namespace NanToNum {
16+using namespace Ops::Base;
17+struct NanToNumTilingData {
18+ EleBaseTilingData baseTiling;
19+ float nan;
20+ float posinf;
21+ float neginf;
22+};
23+} // namespace NanToNum
24+#endif // OPS_MATH_NAN_TO_NUM_TILING_STRUCT_H
Amath/nan_to_num/op_kernel/nan_to_num_apt.cpp+57-0
@@ -0,0 +1,57 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file nan_to_num_apt.cpp
13+ * \brief z = nan_to_num(x)
14+ */
15+#include "kernel_operator.h"
16+#include "kernel_tiling/kernel_tiling.h"
17+#include "arch35/nan_to_num_dag.h"
18+#include "arch35/nan_to_num_struct.h"
19+#include "arch35/nan_to_num_tiling_struct.h"
20+#include "atvoss/elewise/elewise_sch.h"
21+ 
22+using namespace AscendC;
23+using namespace NanToNumOp;
24+using namespace NanToNum;
25+ 
26+template <uint64_t schMode, uint64_t dType>
27+__global__ __aicore__ void nan_to_num(GM_ADDR x, GM_ADDR y, GM_ADDR workspace, GM_ADDR tiling)
28+{
29+ REGISTER_TILING_DEFAULT(NanToNumTilingData);
30+ GET_TILING_DATA_WITH_STRUCT(NanToNumTilingData, tilingData, tiling);
31+ KERNEL_TASK_TYPE_DEFAULT(KERNEL_TYPE_AIV_ONLY);
32+ 
33+ TPipe pipe;
34+ if constexpr (dType == TPL_FP16) {
35+ ElementwiseSch<schMode, NanToNumOp::NanToNumDAG<half>::OpDag> sch(&(tilingData.baseTiling), &pipe);
36+ sch.template SetVar<float, NanToNumOp::PLACEHOLDER_INDEX_0>(tilingData.nan);
37+ sch.template SetVar<float, NanToNumOp::PLACEHOLDER_INDEX_1>(tilingData.posinf);
38+ sch.template SetVar<float, NanToNumOp::PLACEHOLDER_INDEX_2>(tilingData.neginf);
39+ sch.Init(x, y);
40+ sch.Process();
41+ } else if constexpr (dType == TPL_BF16) {
42+ ElementwiseSch<schMode, NanToNumOp::NanToNumDAG<bfloat16_t>::OpDag> sch(&(tilingData.baseTiling), &pipe);
43+ sch.template SetVar<float, NanToNumOp::PLACEHOLDER_INDEX_0>(tilingData.nan);
44+ sch.template SetVar<float, NanToNumOp::PLACEHOLDER_INDEX_1>(tilingData.posinf);
45+ sch.template SetVar<float, NanToNumOp::PLACEHOLDER_INDEX_2>(tilingData.neginf);
46+ sch.Init(x, y);
47+ sch.Process();
48+ } else if constexpr (dType == TPL_FP32) {
49+ ElementwiseSch<schMode, NanToNumOp::NanToNumDAG<float>::OpDag> sch(&(tilingData.baseTiling), &pipe);
50+ sch.template SetVar<float, NanToNumOp::PLACEHOLDER_INDEX_0>(tilingData.nan);
51+ sch.template SetVar<float, NanToNumOp::PLACEHOLDER_INDEX_1>(tilingData.posinf);
52+ sch.template SetVar<float, NanToNumOp::PLACEHOLDER_INDEX_2>(tilingData.neginf);
53+ sch.Init(x, y);
54+ sch.Process();
55+ }
56+ return;
57+}
Dmath/nan_to_num/tests/CMakeLists.txt+0-17
@@ -1,17 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
13-foreach(SUB_DIR ${CURRENT_DIRS})
14- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
15- add_subdirectory(${SUB_DIR})
16- endif()
17-endforeach()
Dmath/nan_to_num/tests/ut/CMakeLists.txt+0-17
@@ -1,17 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
13-foreach(SUB_DIR ${CURRENT_DIRS})
14- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
15- add_subdirectory(${SUB_DIR})
16- endif()
17-endforeach()
Amath/nan_to_num/tests/ut/op_api/test_aclnn_inplace_nan_to_num.cpp+341-0
@@ -0,0 +1,341 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include "gtest/gtest.h"
12+#include "../../../op_api/aclnn_nan_to_num.h"
13+#include "op_api_ut_common/tensor_desc.h"
14+#include "op_api_ut_common/op_api_ut.h"
15+ 
16+class l2InplaceNanToNumTest : public testing::Test {
17+protected:
18+ static void SetUpTestCase()
19+ {
20+ std::cout << "l2InplaceNanToNumTest SetUp" << std::endl;
21+ }
22+ 
23+ static void TearDownTestCase()
24+ {
25+ std::cout << "l2InplaceNanToNumTest TearDown" << std::endl;
26+ }
27+};
28+ 
29+// 异常场景：self为空指针，返回ACLNN_ERR_PARAM_NULLPTR
30+TEST_F(l2InplaceNanToNumTest, l2_nan_to_num_test_err_self_null)
31+{
32+ float nan = 0;
33+ float posinf = 1.0;
34+ float neginf = -1.0;
35+ 
36+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT((aclTensor*)nullptr, nan, posinf, neginf), OUTPUT());
37+ 
38+ uint64_t workspaceSize = 0;
39+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
40+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_NULLPTR);
41+}
42+ 
43+// 异常场景：self的数据类型不在支持范围内，返回ACLNN_ERR_PARAM_INVALID
44+TEST_F(l2InplaceNanToNumTest, l2_nan_to_num_test_err_self_double)
45+{
46+ auto selfDesc = TensorDesc({2, 3}, ACL_DOUBLE, ACL_FORMAT_ND);
47+ float nan = 0;
48+ float posinf = 1.0;
49+ float neginf = -1.0;
50+ 
51+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
52+ 
53+ uint64_t workspaceSize = 0;
54+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
55+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
56+}
57+ 
58+// 异常场景：self的数据类型不在支持范围内，返回ACLNN_ERR_PARAM_INVALID
59+TEST_F(l2InplaceNanToNumTest, l2_nan_to_num_test_err_self_complex)
60+{
61+ auto selfDesc = TensorDesc({2, 3}, ACL_COMPLEX64, ACL_FORMAT_ND);
62+ float nan = 0;
63+ float posinf = 1.0;
64+ float neginf = -1.0;
65+ 
66+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
67+ 
68+ uint64_t workspaceSize = 0;
69+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
70+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
71+}
72+ 
73+// 异常场景：Tensor为9维、返回ACLNN_ERR_PARAM_INVALID
74+TEST_F(l2InplaceNanToNumTest, l2_nan_to_num_test_err_dim_over)
75+{
76+ auto selfDesc = TensorDesc({1, 2, 3, 4, 5, 6, 7, 8, 9}, ACL_FLOAT, ACL_FORMAT_ND);
77+ float nan = 0;
78+ float posinf = 1.0;
79+ float neginf = -1.0;
80+ 
81+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
82+ 
83+ uint64_t workspaceSize = 0;
84+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
85+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
86+}
87+ 
88+// 正常场景：空tensor，返回ACLNN_SUCCESS
89+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_err_empty_tensor)
90+{
91+ auto selfDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND);
92+ float nan = 0;
93+ float posinf = 1.0;
94+ float neginf = -1.0;
95+ 
96+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
97+ 
98+ uint64_t workspaceSize = 0;
99+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
100+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
101+}
102+ 
103+// 正常场景：float16、6维、nan/posinf/neginf为int，返回ACLNN_SUCCESS，精度校验通过
104+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_fp16_dim6_para_int)
105+{
106+ auto selfDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
107+ int nan = 0;
108+ int posinf = 1;
109+ int neginf = -1;
110+ 
111+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
112+ 
113+ uint64_t workspaceSize = 0;
114+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
115+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
116+}
117+ 
118+// 正常场景：float16、6维，返回ACLNN_SUCCESS，精度校验通过
119+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_fp16_dim6)
120+{
121+ auto selfDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
122+ float nan = 0;
123+ float posinf = 1.0;
124+ float neginf = -1.0;
125+ 
126+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
127+ 
128+ uint64_t workspaceSize = 0;
129+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
130+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
131+}
132+ 
133+// 正常场景：float32、7维，返回ACLNN_SUCCESS，精度校验通过
134+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_fp32_dim7)
135+{
136+ auto selfDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12}, ACL_FLOAT, ACL_FORMAT_ND);
137+ float nan = 0;
138+ float posinf = 1.0;
139+ float neginf = -1.0;
140+ 
141+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
142+ 
143+ uint64_t workspaceSize = 0;
144+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
145+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
146+}
147+ 
148+// 正常场景：bfloat16、8维，返回ACLNN_SUCCESS，精度校验通过
149+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_bf16_dim8)
150+{
151+ auto selfDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12, 23}, ACL_BF16, ACL_FORMAT_ND);
152+ float nan = 0;
153+ float posinf = 1.0;
154+ float neginf = -1.0;
155+ 
156+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
157+ 
158+ uint64_t workspaceSize = 0;
159+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
160+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
161+}
162+ 
163+// 正常场景：ACL_INT8、1维，ACLNN_SUCCESS
164+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_int8_dim1)
165+{
166+ auto selfDesc = TensorDesc({2}, ACL_INT8, ACL_FORMAT_ND);
167+ float nan = 0;
168+ float posinf = 1.0;
169+ float neginf = -1.0;
170+ 
171+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
172+ 
173+ uint64_t workspaceSize = 0;
174+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
175+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
176+}
177+ 
178+// 正常场景：ACL_INT16、2维，ACLNN_SUCCESS
179+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_int16_dim2)
180+{
181+ auto selfDesc = TensorDesc({2, 4}, ACL_INT16, ACL_FORMAT_ND);
182+ float nan = 0;
183+ float posinf = 1.0;
184+ float neginf = -1.0;
185+ 
186+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
187+ 
188+ uint64_t workspaceSize = 0;
189+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
190+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
191+}
192+ 
193+// 正常场景：ACL_INT32、3维，ACLNN_SUCCESS
194+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_int32_dim3)
195+{
196+ auto selfDesc = TensorDesc({2, 4, 12}, ACL_INT32, ACL_FORMAT_ND);
197+ float nan = 0;
198+ float posinf = 1.0;
199+ float neginf = -1.0;
200+ 
201+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
202+ 
203+ uint64_t workspaceSize = 0;
204+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
205+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
206+}
207+ 
208+// 正常场景：ACL_INT64、4维，ACLNN_SUCCESS
209+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_int64_dim4)
210+{
211+ auto selfDesc = TensorDesc({2, 4, 12, 9}, ACL_INT64, ACL_FORMAT_ND);
212+ float nan = 0;
213+ float posinf = 1.0;
214+ float neginf = -1.0;
215+ 
216+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
217+ 
218+ uint64_t workspaceSize = 0;
219+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
220+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
221+}
222+ 
223+// 正常场景：ACL_UINT8、5维，ACLNN_SUCCESS
224+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_uint8_dim5)
225+{
226+ auto selfDesc = TensorDesc({2, 4, 12, 9, 21}, ACL_UINT8, ACL_FORMAT_ND);
227+ float nan = 0;
228+ float posinf = 1.0;
229+ float neginf = -1.0;
230+ 
231+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
232+ 
233+ uint64_t workspaceSize = 0;
234+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
235+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
236+}
237+ 
238+// 正常场景：ACL_BOOL、6维，ACLNN_SUCCESS
239+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_bool_dim6)
240+{
241+ auto selfDesc = TensorDesc({2, 4, 12, 9, 21, 32}, ACL_BOOL, ACL_FORMAT_ND);
242+ float nan = 0;
243+ float posinf = 1.0;
244+ float neginf = -1.0;
245+ 
246+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
247+ 
248+ uint64_t workspaceSize = 0;
249+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
250+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
251+}
252+ 
253+// 正常场景：float32、4维、ACL_FORMAT_NCHW，返回ACLNN_SUCCESS，精度校验通过
254+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nchw)
255+{
256+ auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NCHW);
257+ float nan = 0;
258+ float posinf = 1.0;
259+ float neginf = -1.0;
260+ 
261+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
262+ 
263+ uint64_t workspaceSize = 0;
264+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
265+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
266+}
267+ 
268+// 正常场景：float32、4维、ACL_FORMAT_NHWC，返回ACLNN_SUCCESS，精度校验通过
269+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nhwc)
270+{
271+ auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NHWC);
272+ float nan = 0;
273+ float posinf = 1.0;
274+ float neginf = -1.0;
275+ 
276+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
277+ 
278+ uint64_t workspaceSize = 0;
279+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
280+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
281+}
282+ 
283+// 正常场景：float32、4维、ACL_FORMAT_NC1HWC0，返回ACLNN_SUCCESS，精度校验通过
284+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nc1hwc0)
285+{
286+ auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NC1HWC0);
287+ float nan = 0;
288+ float posinf = 1.0;
289+ float neginf = -1.0;
290+ 
291+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
292+ 
293+ uint64_t workspaceSize = 0;
294+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
295+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
296+}
297+ 
298+// 正常场景：float32、4维、ACL_FORMAT_HWCN，返回ACLNN_SUCCESS，精度校验通过
299+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_float32_hwcn)
300+{
301+ auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_HWCN);
302+ float nan = 0;
303+ float posinf = 1.0;
304+ float neginf = -1.0;
305+ 
306+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
307+ 
308+ uint64_t workspaceSize = 0;
309+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
310+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
311+}
312+ 
313+// 成功场景：浮点型、non-continuous，返回ACLNN_SUCCESS，精度校验通过
314+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_float32_non_continuous)
315+{
316+ auto selfDesc = TensorDesc({5, 4}, ACL_FLOAT, ACL_FORMAT_ND, {1, 5}, 0, {4, 5}).ValueRange(-2, 2);
317+ float nan = 0;
318+ float posinf = 1.0;
319+ float neginf = -1.0;
320+ 
321+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
322+ 
323+ uint64_t workspaceSize = 0;
324+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
325+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
326+}
327+ 
328+// 成功场景：整型、non-continuous，返回ACLNN_SUCCESS，精度校验通过
329+TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_int_non_continuous)
330+{
331+ auto selfDesc = TensorDesc({5, 4}, ACL_INT32, ACL_FORMAT_ND, {1, 5}, 0, {4, 5}).ValueRange(-2, 2);
332+ float nan = 0;
333+ float posinf = 1.0;
334+ float neginf = -1.0;
335+ 
336+ auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
337+ 
338+ uint64_t workspaceSize = 0;
339+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
340+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
341+}
Amath/nan_to_num/tests/ut/op_api/test_aclnn_nan_to_num.cpp+329-0
@@ -0,0 +1,329 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include "gtest/gtest.h"
12+#include "../../../op_api/aclnn_nan_to_num.h"
13+#include "op_api_ut_common/tensor_desc.h"
14+#include "op_api_ut_common/op_api_ut.h"
15+ 
16+class l2NanToNumTest : public testing::Test {
17+protected:
18+ static void SetUpTestCase()
19+ {
20+ std::cout << "l2NanToNumTest SetUp" << std::endl;
21+ }
22+ 
23+ static void TearDownTestCase()
24+ {
25+ std::cout << "l2NanToNumTest TearDown" << std::endl;
26+ }
27+};
28+ 
29+// 异常场景：self为空指针，返回ACLNN_ERR_PARAM_NULLPTR
30+TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_self_null)
31+{
32+ auto outDesc = TensorDesc({2, 3}, ACL_INT32, ACL_FORMAT_ND);
33+ float nan = 0;
34+ float posinf = 1.0;
35+ float neginf = -1.0;
36+ 
37+ auto ut = OP_API_UT(aclnnNanToNum, INPUT((aclTensor*)nullptr, nan, posinf, neginf), OUTPUT(outDesc));
38+ 
39+ uint64_t workspaceSize = 0;
40+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
41+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_NULLPTR);
42+}
43+ 
44+// 异常场景：out为空指针，返回ACLNN_ERR_PARAM_NULLPTR
45+TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_out_null)
46+{
47+ auto selfDesc = TensorDesc({2, 3}, ACL_INT32, ACL_FORMAT_ND);
48+ float nan = 0;
49+ float posinf = 1.0;
50+ float neginf = -1.0;
51+ 
52+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT((aclTensor*)nullptr));
53+ 
54+ uint64_t workspaceSize = 0;
55+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
56+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_NULLPTR);
57+}
58+ 
59+// 异常场景：self和out的数据类型不一致，返回ACLNN_ERR_PARAM_INVALID
60+TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_self_out_datatype_inconsistent)
61+{
62+ auto selfDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
63+ auto outDesc = TensorDesc({2, 3}, ACL_INT8, ACL_FORMAT_ND);
64+ float nan = 0;
65+ float posinf = 1.0;
66+ float neginf = -1.0;
67+ 
68+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
69+ 
70+ uint64_t workspaceSize = 0;
71+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
72+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
73+}
74+ 
75+// 异常场景：self的数据类型不在支持范围内，返回ACLNN_ERR_PARAM_INVALID
76+TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_self_double)
77+{
78+ auto selfDesc = TensorDesc({2, 3}, ACL_DOUBLE, ACL_FORMAT_ND);
79+ auto outDesc = TensorDesc({2, 3}, ACL_DOUBLE, ACL_FORMAT_ND);
80+ float nan = 0;
81+ float posinf = 1.0;
82+ float neginf = -1.0;
83+ 
84+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
85+ 
86+ uint64_t workspaceSize = 0;
87+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
88+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
89+}
90+ 
91+// 异常场景：self的数据类型不在支持范围内，返回ACLNN_ERR_PARAM_INVALID
92+TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_self_complex)
93+{
94+ auto selfDesc = TensorDesc({2, 3}, ACL_COMPLEX64, ACL_FORMAT_ND);
95+ auto outDesc = TensorDesc({2, 3}, ACL_COMPLEX64, ACL_FORMAT_ND);
96+ float nan = 0;
97+ float posinf = 1.0;
98+ float neginf = -1.0;
99+ 
100+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
101+ 
102+ uint64_t workspaceSize = 0;
103+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
104+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
105+}
106+ 
107+// 异常场景：self和out的shape不一致，返回ACLNN_ERR_PARAM_INVALID
108+TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_self_fp32_out_fp16)
109+{
110+ auto selfDesc = TensorDesc({2, 3, 9}, ACL_FLOAT, ACL_FORMAT_ND);
111+ auto outDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
112+ float nan = 0;
113+ float posinf = 1.0;
114+ float neginf = -1.0;
115+ 
116+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
117+ 
118+ uint64_t workspaceSize = 0;
119+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
120+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
121+}
122+ 
123+// 异常场景：Tensor为9维、返回ACLNN_ERR_PARAM_INVALID
124+TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_dim_over)
125+{
126+ auto selfDesc = TensorDesc({1, 2, 3, 4, 5, 6, 7, 8, 9}, ACL_FLOAT, ACL_FORMAT_ND);
127+ auto outDesc = TensorDesc({1, 2, 3, 4, 5, 6, 7, 8, 9}, ACL_FLOAT, ACL_FORMAT_ND);
128+ float nan = 0;
129+ float posinf = 1.0;
130+ float neginf = -1.0;
131+ 
132+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
133+ 
134+ uint64_t workspaceSize = 0;
135+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
136+ EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
137+}
138+ 
139+// 正常场景：空tensor，返回ACLNN_SUCCESS
140+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_err_empty_tensor)
141+{
142+ auto selfDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND);
143+ auto outDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND);
144+ float nan = 0;
145+ float posinf = 1.0;
146+ float neginf = -1.0;
147+ 
148+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
149+ 
150+ uint64_t workspaceSize = 0;
151+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
152+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
153+}
154+ 
155+// 正常场景：float16、6维、nan/posinf/neginf为int，返回ACLNN_SUCCESS，精度校验通过
156+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_fp16_dim6_para_int)
157+{
158+ auto selfDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
159+ auto outDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
160+ int nan = 0;
161+ int posinf = 1;
162+ int neginf = -1;
163+ 
164+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
165+ 
166+ uint64_t workspaceSize = 0;
167+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
168+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
169+}
170+ 
171+// 正常场景：float16、6维，返回ACLNN_SUCCESS，精度校验通过
172+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_fp16_dim6)
173+{
174+ auto selfDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
175+ auto outDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
176+ float nan = 0;
177+ float posinf = 1.0;
178+ float neginf = -1.0;
179+ 
180+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
181+ 
182+ uint64_t workspaceSize = 0;
183+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
184+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
185+}
186+ 
187+// 正常场景：float32、7维，返回ACLNN_SUCCESS，精度校验通过
188+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_fp32_dim7)
189+{
190+ auto selfDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12}, ACL_FLOAT, ACL_FORMAT_ND);
191+ auto outDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12}, ACL_FLOAT, ACL_FORMAT_ND);
192+ float nan = 0;
193+ float posinf = 1.0;
194+ float neginf = -1.0;
195+ 
196+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
197+ 
198+ uint64_t workspaceSize = 0;
199+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
200+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
201+}
202+ 
203+// 正常场景：bfloat16、8维，返回ACLNN_SUCCESS，精度校验通过
204+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_bf16_dim8)
205+{
206+ auto selfDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12, 23}, ACL_BF16, ACL_FORMAT_ND);
207+ auto outDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12, 23}, ACL_BF16, ACL_FORMAT_ND);
208+ float nan = 0;
209+ float posinf = 1.0;
210+ float neginf = -1.0;
211+ 
212+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
213+ 
214+ uint64_t workspaceSize = 0;
215+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
216+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
217+}
218+ 
219+// 正常场景：ACL_BOOL、6维，ACLNN_SUCCESS
220+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_bool_dim6)
221+{
222+ auto selfDesc = TensorDesc({2, 4, 12, 9, 21, 32}, ACL_BOOL, ACL_FORMAT_ND);
223+ auto outDesc = TensorDesc({2, 4, 12, 9, 21, 32}, ACL_BOOL, ACL_FORMAT_ND);
224+ float nan = 0;
225+ float posinf = 1.0;
226+ float neginf = -1.0;
227+ 
228+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
229+ 
230+ uint64_t workspaceSize = 0;
231+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
232+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
233+}
234+ 
235+// 正常场景：float32、4维、ACL_FORMAT_NCHW，返回ACLNN_SUCCESS，精度校验通过
236+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nchw)
237+{
238+ auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NCHW);
239+ auto outDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NCHW);
240+ float nan = 0;
241+ float posinf = 1.0;
242+ float neginf = -1.0;
243+ 
244+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
245+ 
246+ uint64_t workspaceSize = 0;
247+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
248+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
249+}
250+ 
251+// 正常场景：float32、4维、ACL_FORMAT_NHWC，返回ACLNN_SUCCESS，精度校验通过
252+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nhwc)
253+{
254+ auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NHWC);
255+ auto outDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NHWC);
256+ float nan = 0;
257+ float posinf = 1.0;
258+ float neginf = -1.0;
259+ 
260+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
261+ 
262+ uint64_t workspaceSize = 0;
263+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
264+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
265+}
266+ 
267+// 正常场景：float32、4维、ACL_FORMAT_NC1HWC0，返回ACLNN_SUCCESS，精度校验通过
268+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nc1hwc0)
269+{
270+ auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NC1HWC0);
271+ auto outDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NC1HWC0);
272+ float nan = 0;
273+ float posinf = 1.0;
274+ float neginf = -1.0;
275+ 
276+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
277+ 
278+ uint64_t workspaceSize = 0;
279+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
280+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
281+}
282+ 
283+// 正常场景：float32、4维、ACL_FORMAT_HWCN，返回ACLNN_SUCCESS，精度校验通过
284+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_float32_hwcn)
285+{
286+ auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_HWCN);
287+ auto outDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_HWCN);
288+ float nan = 0;
289+ float posinf = 1.0;
290+ float neginf = -1.0;
291+ 
292+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
293+ 
294+ uint64_t workspaceSize = 0;
295+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
296+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
297+}
298+ 
299+// 成功场景：浮点型、non-continuous，返回ACLNN_SUCCESS，精度校验通过
300+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_float32_non_continuous)
301+{
302+ auto selfDesc = TensorDesc({5, 4}, ACL_FLOAT, ACL_FORMAT_ND, {1, 5}, 0, {4, 5}).ValueRange(-2, 2);
303+ auto outDesc = TensorDesc({5, 4}, ACL_FLOAT, ACL_FORMAT_ND, {1, 5}, 0, {4, 5}).ValueRange(-2, 2);
304+ float nan = 0;
305+ float posinf = 1.0;
306+ float neginf = -1.0;
307+ 
308+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
309+ 
310+ uint64_t workspaceSize = 0;
311+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
312+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
313+}
314+ 
315+// 成功场景：整型、non-continuous，返回ACLNN_SUCCESS，精度校验通过
316+TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_int_non_continuous)
317+{
318+ auto selfDesc = TensorDesc({5, 4}, ACL_INT32, ACL_FORMAT_ND, {1, 5}, 0, {4, 5}).ValueRange(-2, 2);
319+ auto outDesc = TensorDesc({5, 4}, ACL_INT32, ACL_FORMAT_ND, {1, 5}, 0, {4, 5}).ValueRange(-2, 2);
320+ float nan = 0;
321+ float posinf = 1.0;
322+ float neginf = -1.0;
323+ 
324+ auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
325+ 
326+ uint64_t workspaceSize = 0;
327+ aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
328+ EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
329+}
Dmath/nan_to_num/tests/ut/op_host/CMakeLists.txt+0-22
@@ -1,22 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
12-if(UT_TEST_ALL OR OP_HOST_UT)
13- add_modules_ut_sources(UT_NAME ${OP_TILING_MODULE_NAME} MODE PRIVATE DIR ${CMAKE_CURRENT_SOURCE_DIR})
14- add_modules_ut_sources(UT_NAME ${OP_INFERSHAPE_MODULE_NAME} MODE PRIVATE DIR ${CMAKE_CURRENT_SOURCE_DIR})
15-endif()
16- 
17-file(GLOB CURRENT_DIRS RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
18-foreach(SUB_DIR ${CURRENT_DIRS})
19- if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/${SUB_DIR}/CMakeLists.txt")
20- add_subdirectory(${SUB_DIR})
21- endif()
22-endforeach()
Amath/nan_to_num/tests/ut/op_host/arch35/test_nan_to_num_tiling.cpp+173-0
@@ -0,0 +1,173 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <gtest/gtest.h>
12+#include <iostream>
13+#include "../../../../op_host/arch35/nan_to_num_tiling_arch35.h"
14+#include "tiling_context_faker.h"
15+#include "tiling_case_executor.h"
16+#include "atvoss/elewise/elewise_tiling.h"
17+ 
18+using namespace std;
19+ 
20+class NanToNumTilingTest : public testing::Test {
21+ protected:
22+ static void SetUpTestCase() {
23+ std::cout << "NanToNumTilingTest SetUp" << std::endl;
24+ }
25+ 
26+ static void TearDownTestCase() {
27+ std::cout << "NanToNumTilingTest TearDown" << std::endl;
28+ }
29+};
30+ 
31+TEST_F(NanToNumTilingTest, nan_to_num_test_tiling_fp16_input)
32+{
33+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
34+ gert::TilingContextPara tilingContextPara(
35+ "NanToNum",
36+ {
37+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT16, ge::FORMAT_ND},
38+ },
39+ {
40+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT16, ge::FORMAT_ND},
41+ },
42+ {
43+ gert::TilingContextPara::OpAttr("nan", Ops::Math::AnyValue::CreateFrom<float>(-1.0)),
44+ gert::TilingContextPara::OpAttr("posinf", Ops::Math::AnyValue::CreateFrom<float>(1.0)),
45+ gert::TilingContextPara::OpAttr("neginf", Ops::Math::AnyValue::CreateFrom<float>(0.0)),
46+ },
47+ &compileInfo);
48+ 
49+ uint64_t expectTilingKey = 3;
50+ string expectTilingData = "8192 4 32768 2048 4 1 1 2048 2048 32768 1 ";
51+ std::vector<size_t> expectWorkspaces = {16777216};
52+ ExecuteTestCaseForEle(tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
53+}
54+ 
55+TEST_F(NanToNumTilingTest, nan_to_num_test_tiling_bf16_input)
56+{
57+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
58+ gert::TilingContextPara tilingContextPara(
59+ "NanToNum",
60+ {
61+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
62+ },
63+ {
64+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
65+ },
66+ {
67+ gert::TilingContextPara::OpAttr("nan", Ops::Math::AnyValue::CreateFrom<float>(-1.0)),
68+ gert::TilingContextPara::OpAttr("posinf", Ops::Math::AnyValue::CreateFrom<float>(1.0)),
69+ gert::TilingContextPara::OpAttr("neginf", Ops::Math::AnyValue::CreateFrom<float>(0.0)),
70+ },
71+ &compileInfo);
72+ 
73+ uint64_t expectTilingKey = 5;
74+ string expectTilingData = "8192 4 32768 2048 4 1 1 2048 2048 32768 1 ";
75+ std::vector<size_t> expectWorkspaces = {16777216};
76+ ExecuteTestCaseForEle(tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
77+}
78+ 
79+TEST_F(NanToNumTilingTest, nan_to_num_test_tiling_fp32_input)
80+{
81+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
82+ gert::TilingContextPara tilingContextPara(
83+ "NanToNum",
84+ {
85+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
86+ },
87+ {
88+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
89+ },
90+ {
91+ gert::TilingContextPara::OpAttr("nan", Ops::Math::AnyValue::CreateFrom<float>(-1.0)),
92+ gert::TilingContextPara::OpAttr("posinf", Ops::Math::AnyValue::CreateFrom<float>(1.0)),
93+ gert::TilingContextPara::OpAttr("neginf", Ops::Math::AnyValue::CreateFrom<float>(0.0)),
94+ },
95+ &compileInfo);
96+ 
97+ uint64_t expectTilingKey = 7;
98+ string expectTilingData = "8192 8 16384 1024 8 1 1 1024 1024 16384 1 ";
99+ std::vector<size_t> expectWorkspaces = {16777216};
100+ ExecuteTestCaseForEle(tilingContextPara, ge::GRAPH_SUCCESS, true, expectTilingKey, true, expectTilingData, expectWorkspaces);
101+}
102+ 
103+TEST_F(NanToNumTilingTest, nan_to_num_test_tiling_invalid_input_dtype)
104+{
105+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
106+ gert::TilingContextPara tilingContextPara(
107+ "NanToNum",
108+ {
109+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT32, ge::FORMAT_ND},
110+ },
111+ {
112+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
113+ },
114+ {
115+ gert::TilingContextPara::OpAttr("nan", Ops::Math::AnyValue::CreateFrom<float>(-1.0)),
116+ gert::TilingContextPara::OpAttr("posinf", Ops::Math::AnyValue::CreateFrom<float>(1.0)),
117+ gert::TilingContextPara::OpAttr("neginf", Ops::Math::AnyValue::CreateFrom<float>(0.0)),
118+ },
119+ &compileInfo);
120+ 
121+ uint64_t expectTilingKey = 7;
122+ string expectTilingData = "";
123+ std::vector<size_t> expectWorkspaces = {16777216};
124+ ExecuteTestCaseForEle(tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
125+}
126+ 
127+TEST_F(NanToNumTilingTest, nan_to_num_test_tiling_invalid_output_dtype)
128+{
129+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
130+ gert::TilingContextPara tilingContextPara(
131+ "NanToNum",
132+ {
133+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
134+ },
135+ {
136+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_BF16, ge::FORMAT_ND},
137+ },
138+ {
139+ gert::TilingContextPara::OpAttr("nan", Ops::Math::AnyValue::CreateFrom<float>(-1.0)),
140+ gert::TilingContextPara::OpAttr("posinf", Ops::Math::AnyValue::CreateFrom<float>(1.0)),
141+ gert::TilingContextPara::OpAttr("neginf", Ops::Math::AnyValue::CreateFrom<float>(0.0)),
142+ },
143+ &compileInfo);
144+ 
145+ uint64_t expectTilingKey = 7;
146+ string expectTilingData = "";
147+ std::vector<size_t> expectWorkspaces = {16777216};
148+ ExecuteTestCaseForEle(tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
149+}
150+ 
151+TEST_F(NanToNumTilingTest, nan_to_num_test_tiling_invalid_shape)
152+{
153+ Ops::Base::ElewiseCompileInfo compileInfo = {64, 253952};
154+ gert::TilingContextPara tilingContextPara(
155+ "NanToNum",
156+ {
157+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT, ge::FORMAT_ND},
158+ },
159+ {
160+ {{{1, 64, 2, 6}, {1, 64, 2, 6}}, ge::DT_FLOAT, ge::FORMAT_ND},
161+ },
162+ {
163+ gert::TilingContextPara::OpAttr("nan", Ops::Math::AnyValue::CreateFrom<float>(-1.0)),
164+ gert::TilingContextPara::OpAttr("posinf", Ops::Math::AnyValue::CreateFrom<float>(1.0)),
165+ gert::TilingContextPara::OpAttr("neginf", Ops::Math::AnyValue::CreateFrom<float>(0.0)),
166+ },
167+ &compileInfo);
168+ 
169+ uint64_t expectTilingKey = 7;
170+ string expectTilingData = "";
171+ std::vector<size_t> expectWorkspaces = {16777216};
172+ ExecuteTestCaseForEle(tilingContextPara, ge::GRAPH_FAILED, true, expectTilingKey, false, expectTilingData, expectWorkspaces);
173+}
Dmath/nan_to_num/tests/ut/op_host/op_api/CMakeLists.txt+0-11
@@ -1,11 +0,0 @@
1-# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify it.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
9-# See LICENSE in the root of the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------
11- 
Dmath/nan_to_num/tests/ut/op_host/op_api/test_aclnn_inplace_nan_to_num.cpp+0-347
@@ -1,347 +0,0 @@
1-/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4- * CANN Open Software License Agreement Version 2.0 (the "License").
5- * Please refer to the License for details. You may not use this file except in compliance with the License.
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8- * See LICENSE in the root of the software repository for the full text of the License.
9- */
10- 
11-#include "gtest/gtest.h"
12-#include "aclnn_nan_to_num.h"
13-#include "op_api_ut_common/tensor_desc.h"
14-#include "op_api_ut_common/op_api_ut.h"
15- 
16-class l2InplaceNanToNumTest : public testing::Test {
17- protected:
18- static void SetUpTestCase() {
19- std::cout << "l2InplaceNanToNumTest SetUp" << std::endl;
20- }
21- 
22- static void TearDownTestCase() {
23- std::cout << "l2InplaceNanToNumTest TearDown" << std::endl;
24- }
25-};
26- 
27-// 异常场景：self为空指针，返回ACLNN_ERR_PARAM_NULLPTR
28-TEST_F(l2InplaceNanToNumTest, l2_nan_to_num_test_err_self_null) {
29- float nan = 0;
30- float posinf = 1.0;
31- float neginf = -1.0;
32- 
33- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT((aclTensor *)nullptr, nan, posinf, neginf), OUTPUT());
34- 
35- uint64_t workspaceSize = 0;
36- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
37- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_NULLPTR);
38-}
39- 
40-// 异常场景：self的数据类型不在支持范围内，返回ACLNN_ERR_PARAM_INVALID
41-TEST_F(l2InplaceNanToNumTest, l2_nan_to_num_test_err_self_double) {
42- auto selfDesc = TensorDesc({2, 3}, ACL_DOUBLE, ACL_FORMAT_ND);
43- float nan = 0;
44- float posinf = 1.0;
45- float neginf = -1.0;
46- 
47- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
48- 
49- uint64_t workspaceSize = 0;
50- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
51- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
52-}
53- 
54-// 异常场景：self的数据类型不在支持范围内，返回ACLNN_ERR_PARAM_INVALID
55-TEST_F(l2InplaceNanToNumTest, l2_nan_to_num_test_err_self_complex) {
56- auto selfDesc = TensorDesc({2, 3}, ACL_COMPLEX64, ACL_FORMAT_ND);
57- float nan = 0;
58- float posinf = 1.0;
59- float neginf = -1.0;
60- 
61- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
62- 
63- uint64_t workspaceSize = 0;
64- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
65- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
66-}
67- 
68-// 异常场景：Tensor为9维、返回ACLNN_ERR_PARAM_INVALID
69-TEST_F(l2InplaceNanToNumTest, l2_nan_to_num_test_err_dim_over) {
70- auto selfDesc = TensorDesc({1, 2, 3, 4, 5, 6, 7, 8, 9}, ACL_FLOAT, ACL_FORMAT_ND);
71- float nan = 0;
72- float posinf = 1.0;
73- float neginf = -1.0;
74- 
75- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
76- 
77- uint64_t workspaceSize = 0;
78- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
79- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
80-}
81- 
82-// 正常场景：空tensor，返回ACLNN_SUCCESS
83-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_err_empty_tensor) {
84- auto selfDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND);
85- float nan = 0;
86- float posinf = 1.0;
87- float neginf = -1.0;
88- 
89- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
90- 
91- uint64_t workspaceSize = 0;
92- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
93- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
94- 
95- ut.TestPrecision();
96-}
97- 
98-// 正常场景：float16、6维、nan/posinf/neginf为int，返回ACLNN_SUCCESS，精度校验通过
99-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_fp16_dim6_para_int) {
100- auto selfDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
101- int nan = 0;
102- int posinf = 1;
103- int neginf = -1;
104- 
105- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
106- 
107- uint64_t workspaceSize = 0;
108- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
109- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
110- 
111- ut.TestPrecision();
112-}
113- 
114-// 正常场景：float16、6维，返回ACLNN_SUCCESS，精度校验通过
115-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_fp16_dim6) {
116- auto selfDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
117- float nan = 0;
118- float posinf = 1.0;
119- float neginf = -1.0;
120- 
121- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
122- 
123- uint64_t workspaceSize = 0;
124- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
125- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
126- 
127- ut.TestPrecision();
128-}
129- 
130-// 正常场景：float32、7维，返回ACLNN_SUCCESS，精度校验通过
131-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_fp32_dim7) {
132- auto selfDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12}, ACL_FLOAT, ACL_FORMAT_ND);
133- float nan = 0;
134- float posinf = 1.0;
135- float neginf = -1.0;
136- 
137- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
138- 
139- uint64_t workspaceSize = 0;
140- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
141- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
142- 
143- ut.TestPrecision();
144-}
145- 
146-// 正常场景：bfloat16、8维，返回ACLNN_SUCCESS，精度校验通过
147-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_bf16_dim8) {
148- auto selfDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12, 23}, ACL_BF16, ACL_FORMAT_ND);
149- float nan = 0;
150- float posinf = 1.0;
151- float neginf = -1.0;
152- 
153- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
154- 
155- uint64_t workspaceSize = 0;
156- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
157- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
158- 
159- // ut.TestPrecision();
160-}
161- 
162-// 正常场景：ACL_INT8、1维，ACLNN_SUCCESS
163-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_int8_dim1) {
164- auto selfDesc = TensorDesc({2}, ACL_INT8, ACL_FORMAT_ND);
165- float nan = 0;
166- float posinf = 1.0;
167- float neginf = -1.0;
168- 
169- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
170- 
171- uint64_t workspaceSize = 0;
172- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
173- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
174- 
175- ut.TestPrecision();
176-}
177- 
178-// 正常场景：ACL_INT16、2维，ACLNN_SUCCESS
179-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_int16_dim2) {
180- auto selfDesc = TensorDesc({2, 4}, ACL_INT16, ACL_FORMAT_ND);
181- float nan = 0;
182- float posinf = 1.0;
183- float neginf = -1.0;
184- 
185- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
186- 
187- uint64_t workspaceSize = 0;
188- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
189- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
190- 
191- ut.TestPrecision();
192-}
193- 
194-// 正常场景：ACL_INT32、3维，ACLNN_SUCCESS
195-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_int32_dim3) {
196- auto selfDesc = TensorDesc({2, 4, 12}, ACL_INT32, ACL_FORMAT_ND);
197- float nan = 0;
198- float posinf = 1.0;
199- float neginf = -1.0;
200- 
201- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
202- 
203- uint64_t workspaceSize = 0;
204- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
205- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
206- 
207- ut.TestPrecision();
208-}
209- 
210-// 正常场景：ACL_INT64、4维，ACLNN_SUCCESS
211-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_int64_dim4) {
212- auto selfDesc = TensorDesc({2, 4, 12, 9}, ACL_INT64, ACL_FORMAT_ND);
213- float nan = 0;
214- float posinf = 1.0;
215- float neginf = -1.0;
216- 
217- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
218- 
219- uint64_t workspaceSize = 0;
220- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
221- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
222- 
223- ut.TestPrecision();
224-}
225- 
226-// 正常场景：ACL_UINT8、5维，ACLNN_SUCCESS
227-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_uint8_dim5) {
228- auto selfDesc = TensorDesc({2, 4, 12, 9, 21}, ACL_UINT8, ACL_FORMAT_ND);
229- float nan = 0;
230- float posinf = 1.0;
231- float neginf = -1.0;
232- 
233- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
234- 
235- uint64_t workspaceSize = 0;
236- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
237- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
238- 
239- ut.TestPrecision();
240-}
241- 
242-// 正常场景：ACL_BOOL、6维，ACLNN_SUCCESS
243-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_bool_dim6) {
244- auto selfDesc = TensorDesc({2, 4, 12, 9, 21, 32}, ACL_BOOL, ACL_FORMAT_ND);
245- float nan = 0;
246- float posinf = 1.0;
247- float neginf = -1.0;
248- 
249- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
250- 
251- uint64_t workspaceSize = 0;
252- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
253- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
254- 
255- ut.TestPrecision();
256-}
257- 
258-// 正常场景：float32、4维、ACL_FORMAT_NCHW，返回ACLNN_SUCCESS，精度校验通过
259-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nchw) {
260- auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NCHW);
261- float nan = 0;
262- float posinf = 1.0;
263- float neginf = -1.0;
264- 
265- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
266- 
267- uint64_t workspaceSize = 0;
268- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
269- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
270- ut.TestPrecision();
271-}
272- 
273-// 正常场景：float32、4维、ACL_FORMAT_NHWC，返回ACLNN_SUCCESS，精度校验通过
274-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nhwc) {
275- auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NHWC);
276- float nan = 0;
277- float posinf = 1.0;
278- float neginf = -1.0;
279- 
280- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
281- 
282- uint64_t workspaceSize = 0;
283- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
284- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
285- ut.TestPrecision();
286-}
287- 
288-// 正常场景：float32、4维、ACL_FORMAT_NC1HWC0，返回ACLNN_SUCCESS，精度校验通过
289-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nc1hwc0) {
290- auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NC1HWC0);
291- float nan = 0;
292- float posinf = 1.0;
293- float neginf = -1.0;
294- 
295- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
296- 
297- uint64_t workspaceSize = 0;
298- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
299- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
300- ut.TestPrecision();
301-}
302- 
303-// 正常场景：float32、4维、ACL_FORMAT_HWCN，返回ACLNN_SUCCESS，精度校验通过
304-TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_float32_hwcn) {
305- auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_HWCN);
306- float nan = 0;
307- float posinf = 1.0;
308- float neginf = -1.0;
309- 
310- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
311- 
312- uint64_t workspaceSize = 0;
313- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
314- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
315- ut.TestPrecision();
316-}
317- 
318-// 成功场景：浮点型、non-continuous，返回ACLNN_SUCCESS，精度校验通过
319- TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_float32_non_continuous) {
320- auto selfDesc = TensorDesc({5, 4}, ACL_FLOAT, ACL_FORMAT_ND, {1, 5}, 0, {4,5}).ValueRange(-2, 2);
321- float nan = 0;
322- float posinf = 1.0;
323- float neginf = -1.0;
324- 
325- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
326- 
327- uint64_t workspaceSize = 0;
328- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
329- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
330- ut.TestPrecision();
331-}
332- 
333-// 成功场景：整型、non-continuous，返回ACLNN_SUCCESS，精度校验通过
334- TEST_F(l2InplaceNanToNumTest, ascend910B2_l2_nan_to_num_test_int_non_continuous) {
335- auto selfDesc = TensorDesc({5, 4}, ACL_INT32, ACL_FORMAT_ND, {1, 5}, 0, {4,5}).ValueRange(-2, 2);
336- float nan = 0;
337- float posinf = 1.0;
338- float neginf = -1.0;
339- 
340- auto ut = OP_API_UT(aclnnInplaceNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT());
341- 
342- uint64_t workspaceSize = 0;
343- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
344- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
345- ut.TestPrecision();
346-}
347- 
Dmath/nan_to_num/tests/ut/op_host/op_api/test_aclnn_nan_to_num.cpp+0-327
@@ -1,327 +0,0 @@
1-/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4- * CANN Open Software License Agreement Version 2.0 (the "License").
5- * Please refer to the License for details. You may not use this file except in compliance with the License.
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8- * See LICENSE in the root of the software repository for the full text of the License.
9- */
10- 
11-#include "gtest/gtest.h"
12-#include "aclnn_nan_to_num.h"
13-#include "op_api_ut_common/tensor_desc.h"
14-#include "op_api_ut_common/op_api_ut.h"
15- 
16-class l2NanToNumTest : public testing::Test {
17- protected:
18- static void SetUpTestCase() {
19- std::cout << "l2NanToNumTest SetUp" << std::endl;
20- }
21- 
22- static void TearDownTestCase() {
23- std::cout << "l2NanToNumTest TearDown" << std::endl;
24- }
25-};
26- 
27-// 异常场景：self为空指针，返回ACLNN_ERR_PARAM_NULLPTR
28-TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_self_null) {
29- auto outDesc = TensorDesc({2, 3}, ACL_INT32, ACL_FORMAT_ND);
30- float nan = 0;
31- float posinf = 1.0;
32- float neginf = -1.0;
33- 
34- auto ut = OP_API_UT(aclnnNanToNum, INPUT((aclTensor *)nullptr, nan, posinf, neginf), OUTPUT(outDesc));
35- 
36- uint64_t workspaceSize = 0;
37- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
38- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_NULLPTR);
39-}
40- 
41-// 异常场景：out为空指针，返回ACLNN_ERR_PARAM_NULLPTR
42-TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_out_null) {
43- auto selfDesc = TensorDesc({2, 3}, ACL_INT32, ACL_FORMAT_ND);
44- float nan = 0;
45- float posinf = 1.0;
46- float neginf = -1.0;
47- 
48- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT((aclTensor *)nullptr));
49- 
50- uint64_t workspaceSize = 0;
51- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
52- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_NULLPTR);
53-}
54- 
55-// 异常场景：self和out的数据类型不一致，返回ACLNN_ERR_PARAM_INVALID
56-TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_self_out_datatype_inconsistent) {
57- auto selfDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
58- auto outDesc = TensorDesc({2, 3}, ACL_INT8, ACL_FORMAT_ND);
59- float nan = 0;
60- float posinf = 1.0;
61- float neginf = -1.0;
62- 
63- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
64- 
65- uint64_t workspaceSize = 0;
66- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
67- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
68-}
69- 
70-// 异常场景：self的数据类型不在支持范围内，返回ACLNN_ERR_PARAM_INVALID
71-TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_self_double) {
72- auto selfDesc = TensorDesc({2, 3}, ACL_DOUBLE, ACL_FORMAT_ND);
73- auto outDesc = TensorDesc({2, 3}, ACL_DOUBLE, ACL_FORMAT_ND);
74- float nan = 0;
75- float posinf = 1.0;
76- float neginf = -1.0;
77- 
78- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
79- 
80- uint64_t workspaceSize = 0;
81- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
82- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
83-}
84- 
85-// 异常场景：self的数据类型不在支持范围内，返回ACLNN_ERR_PARAM_INVALID
86-TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_self_complex) {
87- auto selfDesc = TensorDesc({2, 3}, ACL_COMPLEX64, ACL_FORMAT_ND);
88- auto outDesc = TensorDesc({2, 3}, ACL_COMPLEX64, ACL_FORMAT_ND);
89- float nan = 0;
90- float posinf = 1.0;
91- float neginf = -1.0;
92- 
93- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
94- 
95- uint64_t workspaceSize = 0;
96- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
97- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
98-}
99- 
100-// 异常场景：self和out的shape不一致，返回ACLNN_ERR_PARAM_INVALID
101-TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_self_fp32_out_fp16) {
102- auto selfDesc = TensorDesc({2, 3, 9}, ACL_FLOAT, ACL_FORMAT_ND);
103- auto outDesc = TensorDesc({2, 3}, ACL_FLOAT, ACL_FORMAT_ND);
104- float nan = 0;
105- float posinf = 1.0;
106- float neginf = -1.0;
107- 
108- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
109- 
110- uint64_t workspaceSize = 0;
111- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
112- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
113-}
114- 
115-// 异常场景：Tensor为9维、返回ACLNN_ERR_PARAM_INVALID
116-TEST_F(l2NanToNumTest, l2_nan_to_num_test_err_dim_over) {
117- auto selfDesc = TensorDesc({1, 2, 3, 4, 5, 6, 7, 8, 9}, ACL_FLOAT, ACL_FORMAT_ND);
118- auto outDesc = TensorDesc({1, 2, 3, 4, 5, 6, 7, 8, 9}, ACL_FLOAT, ACL_FORMAT_ND);
119- float nan = 0;
120- float posinf = 1.0;
121- float neginf = -1.0;
122- 
123- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
124- 
125- uint64_t workspaceSize = 0;
126- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
127- EXPECT_EQ(getWorkspaceResult, ACLNN_ERR_PARAM_INVALID);
128-}
129- 
130-// 正常场景：空tensor，返回ACLNN_SUCCESS
131-TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_err_empty_tensor) {
132- auto selfDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND);
133- auto outDesc = TensorDesc({0}, ACL_FLOAT, ACL_FORMAT_ND);
134- float nan = 0;
135- float posinf = 1.0;
136- float neginf = -1.0;
137- 
138- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
139- 
140- uint64_t workspaceSize = 0;
141- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
142- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
143- 
144- ut.TestPrecision();
145-}
146- 
147-// 正常场景：float16、6维、nan/posinf/neginf为int，返回ACLNN_SUCCESS，精度校验通过
148-TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_fp16_dim6_para_int) {
149- auto selfDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
150- auto outDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
151- int nan = 0;
152- int posinf = 1;
153- int neginf = -1;
154- 
155- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
156- 
157- uint64_t workspaceSize = 0;
158- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
159- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
160- 
161- ut.TestPrecision();
162-}
163- 
164-// 正常场景：float16、6维，返回ACLNN_SUCCESS，精度校验通过
165-TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_fp16_dim6) {
166- auto selfDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
167- auto outDesc = TensorDesc({2, 4, 6, 7, 9, 1}, ACL_FLOAT16, ACL_FORMAT_ND);
168- float nan = 0;
169- float posinf = 1.0;
170- float neginf = -1.0;
171- 
172- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
173- 
174- uint64_t workspaceSize = 0;
175- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
176- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
177- 
178- ut.TestPrecision();
179-}
180- 
181-// 正常场景：float32、7维，返回ACLNN_SUCCESS，精度校验通过
182-TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_fp32_dim7) {
183- auto selfDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12}, ACL_FLOAT, ACL_FORMAT_ND);
184- auto outDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12}, ACL_FLOAT, ACL_FORMAT_ND);
185- float nan = 0;
186- float posinf = 1.0;
187- float neginf = -1.0;
188- 
189- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
190- 
191- uint64_t workspaceSize = 0;
192- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
193- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
194- 
195- ut.TestPrecision();
196-}
197- 
198-// 正常场景：bfloat16、8维，返回ACLNN_SUCCESS，精度校验通过
199-TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_bf16_dim8) {
200- auto selfDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12, 23}, ACL_BF16, ACL_FORMAT_ND);
201- auto outDesc = TensorDesc({2, 4, 7, 9, 1, 2, 12, 23}, ACL_BF16, ACL_FORMAT_ND);
202- float nan = 0;
203- float posinf = 1.0;
204- float neginf = -1.0;
205- 
206- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
207- 
208- uint64_t workspaceSize = 0;
209- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
210- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
211- 
212- // ut.TestPrecision();
213-}
214- 
215-// 正常场景：ACL_BOOL、6维，ACLNN_SUCCESS
216-TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_bool_dim6) {
217- auto selfDesc = TensorDesc({2, 4, 12, 9, 21, 32}, ACL_BOOL, ACL_FORMAT_ND);
218- auto outDesc = TensorDesc({2, 4, 12, 9, 21, 32}, ACL_BOOL, ACL_FORMAT_ND);
219- float nan = 0;
220- float posinf = 1.0;
221- float neginf = -1.0;
222- 
223- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
224- 
225- uint64_t workspaceSize = 0;
226- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
227- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
228- 
229- ut.TestPrecision();
230-}
231- 
232-// 正常场景：float32、4维、ACL_FORMAT_NCHW，返回ACLNN_SUCCESS，精度校验通过
233-TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nchw) {
234- auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NCHW);
235- auto outDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NCHW);
236- float nan = 0;
237- float posinf = 1.0;
238- float neginf = -1.0;
239- 
240- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
241- 
242- uint64_t workspaceSize = 0;
243- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
244- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
245- ut.TestPrecision();
246-}
247- 
248-// 正常场景：float32、4维、ACL_FORMAT_NHWC，返回ACLNN_SUCCESS，精度校验通过
249-TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nhwc) {
250- auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NHWC);
251- auto outDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NHWC);
252- float nan = 0;
253- float posinf = 1.0;
254- float neginf = -1.0;
255- 
256- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
257- 
258- uint64_t workspaceSize = 0;
259- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
260- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
261- ut.TestPrecision();
262-}
263- 
264-// 正常场景：float32、4维、ACL_FORMAT_NC1HWC0，返回ACLNN_SUCCESS，精度校验通过
265-TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_float32_nc1hwc0) {
266- auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NC1HWC0);
267- auto outDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_NC1HWC0);
268- float nan = 0;
269- float posinf = 1.0;
270- float neginf = -1.0;
271- 
272- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
273- 
274- uint64_t workspaceSize = 0;
275- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
276- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
277- ut.TestPrecision();
278-}
279- 
280-// 正常场景：float32、4维、ACL_FORMAT_HWCN，返回ACLNN_SUCCESS，精度校验通过
281-TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_float32_hwcn) {
282- auto selfDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_HWCN);
283- auto outDesc = TensorDesc({2, 3, 4, 5}, ACL_FLOAT, ACL_FORMAT_HWCN);
284- float nan = 0;
285- float posinf = 1.0;
286- float neginf = -1.0;
287- 
288- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
289- 
290- uint64_t workspaceSize = 0;
291- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
292- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
293- ut.TestPrecision();
294-}
295- 
296-// 成功场景：浮点型、non-continuous，返回ACLNN_SUCCESS，精度校验通过
297- TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_float32_non_continuous) {
298- auto selfDesc = TensorDesc({5, 4}, ACL_FLOAT, ACL_FORMAT_ND, {1, 5}, 0, {4,5}).ValueRange(-2, 2);
299- auto outDesc = TensorDesc({5, 4}, ACL_FLOAT, ACL_FORMAT_ND, {1, 5}, 0, {4,5}).ValueRange(-2, 2);
300- float nan = 0;
301- float posinf = 1.0;
302- float neginf = -1.0;
303- 
304- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
305- 
306- uint64_t workspaceSize = 0;
307- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
308- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
309- ut.TestPrecision();
310-}
311- 
312-// 成功场景：整型、non-continuous，返回ACLNN_SUCCESS，精度校验通过
313- TEST_F(l2NanToNumTest, ascend910B2_l2_nan_to_num_test_int_non_continuous) {
314- auto selfDesc = TensorDesc({5, 4}, ACL_INT32, ACL_FORMAT_ND, {1, 5}, 0, {4,5}).ValueRange(-2, 2);
315- auto outDesc = TensorDesc({5, 4}, ACL_INT32, ACL_FORMAT_ND, {1, 5}, 0, {4,5}).ValueRange(-2, 2);
316- float nan = 0;
317- float posinf = 1.0;
318- float neginf = -1.0;
319- 
320- auto ut = OP_API_UT(aclnnNanToNum, INPUT(selfDesc, nan, posinf, neginf), OUTPUT(outDesc));
321- 
322- uint64_t workspaceSize = 0;
323- aclnnStatus getWorkspaceResult = ut.TestGetWorkspaceSize(&workspaceSize);
324- EXPECT_EQ(getWorkspaceResult, ACLNN_SUCCESS);
325- ut.TestPrecision();
326-}
327- 
Amath/nan_to_num/tests/ut/op_host/test_nan_to_num_infershape.cpp+81-0
@@ -0,0 +1,81 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <gtest/gtest.h>
12+#include <iostream>
13+#include "infershape_context_faker.h"
14+#include "infershape_case_executor.h"
15+ 
16+class NanToNumInferShape : public testing::Test {
17+ protected:
18+ static void SetUpTestCase() {
19+ std::cout << "NanToNum InferShape SetUp" << std::endl;
20+ }
21+ 
22+ static void TearDownTestCase() {
23+ std::cout << "NanToNum InferShape TearDown" << std::endl;
24+ }
25+};
26+ 
27+TEST_F(NanToNumInferShape, nan_to_num_infershape_diff_test)
28+{
29+ gert::InfershapeContextPara infershapeContextPara(
30+ "NanToNum",
31+ {
32+ {{{4, 3, 4}, {4, 3, 4}}, ge::DT_FLOAT16, ge::FORMAT_ND},
33+ },
34+ {
35+ {{{}, {}}, ge::DT_FLOAT16, ge::FORMAT_ND},
36+ });
37+ std::vector<std::vector<int64_t>> expectOutputShape = {{4, 3, 4}};
38+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
39+}
40+ 
41+TEST_F(NanToNumInferShape, nan_to_num_infershape_same_test)
42+{
43+ gert::InfershapeContextPara infershapeContextPara(
44+ "NanToNum",
45+ {
46+ {{{1, 3, 4}, {1, 3, 4}}, ge::DT_FLOAT16, ge::FORMAT_ND},
47+ },
48+ {
49+ {{{}, {}}, ge::DT_FLOAT16, ge::FORMAT_ND},
50+ });
51+ std::vector<std::vector<int64_t>> expectOutputShape = {{1, 3, 4}};
52+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
53+}
54+ 
55+TEST_F(NanToNumInferShape, nan_to_num_infershape_test2)
56+{
57+ gert::InfershapeContextPara infershapeContextPara(
58+ "NanToNum",
59+ {
60+ {{{5, -1}, {5, -1}}, ge::DT_FLOAT, ge::FORMAT_ND},
61+ },
62+ {
63+ {{{}, {}}, ge::DT_FLOAT, ge::FORMAT_ND},
64+ });
65+ std::vector<std::vector<int64_t>> expectOutputShape = {{5, -1}};
66+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
67+}
68+ 
69+TEST_F(NanToNumInferShape, nan_to_num_infershape_test3)
70+{
71+ gert::InfershapeContextPara infershapeContextPara(
72+ "NanToNum",
73+ {
74+ {{{-2}, {-2}}, ge::DT_FLOAT, ge::FORMAT_ND},
75+ },
76+ {
77+ {{{}, {}}, ge::DT_FLOAT, ge::FORMAT_ND},
78+ });
79+ std::vector<std::vector<int64_t>> expectOutputShape = {{-2}};
80+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
81+}
Mmath/right_shift/CMakeLists.txt+12-8
@@ -1,12 +1,16 @@
1-# ---------------------------------------------------------------------------------------------------------1+# ----------------------------------------------------------------------------
2-# Copyright (c) 2025 Huawei Technologies Co., Ltd.2+# Copyright (c) 2026 Huawei Technologies Co., Ltd.
3-# This program is free software, you can redistribute it and/or modify it under the terms and conditions of 3+# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4# CANN Open Software License Agreement Version 2.0 (the "License").4# CANN Open Software License Agreement Version 2.0 (the "License").
5# Please refer to the License for details. You may not use this file except in compliance with the License.5# Please refer to the License for details. You may not use this file except in compliance with the License.
6-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, 6+# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7-# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. 7+# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9-# ---------------------------------------------------------------------------------------------------------9+# ----------------------------------------------------------------------------
10 10 
11-add_all_modules_sources(OPTYPE right_shift ACLNNTYPE aclnn_exclude)11+# 设置算子定义时支持的芯片类型
12-add_subdirectory(framework)12+set(SUPPORT_COMPUTE_UNIT "ascend950")
13+# 设置每种芯片类型对应的tiling文件目录，即采用op_host目录下哪个文件夹下的tiling文件编译
14+set(SUPPORT_TILING_DIR "arch35")
15+add_all_modules_sources(OPTYPE right_shift ACLNNTYPE aclnn_exclude COMPUTE_UNIT ${SUPPORT_COMPUTE_UNIT} TILING_DIR ${SUPPORT_TILING_DIR} DISABLE_IN_OPP TRUE)
16+add_subdirectory(framework)
Mmath/right_shift/framework/CMakeLists.txt+11-11
@@ -1,12 +1,12 @@
1-# ----------------------------------------------------------------------------1+# ----------------------------------------------------------------------------
2-# This program is free software, you can redistribute it and/or modify.2+# This program is free software, you can redistribute it and/or modify.
3-# Copyright (c) 2025 Huawei Technologies Co., Ltd.3+# Copyright (c) 2025 Huawei Technologies Co., Ltd.
4-# This file is a part of the CANN Open Software.4+# This file is a part of the CANN Open Software.
5-# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").5+# Licensed under CANN Open Software License Agreement Version 2.0 (the "License").
6-# Please refer to the License for details. You may not use this file except in compliance with the License.6+# Please refer to the License for details. You may not use this file except in compliance with the License.
7-# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING7+# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8-# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. See LICENSE in the root of8+# BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. See LICENSE in the root of
9-# the software repository for the full text of the License.9+# the software repository for the full text of the License.
10-# ----------------------------------------------------------------------------10+# ----------------------------------------------------------------------------
11- 11+ 
12add_onnx_plugin_sources()12add_onnx_plugin_sources()
Amath/right_shift/op_graph/right_shift_graph_infer.cpp+31-0
@@ -0,0 +1,31 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file right_shift_graph_infer.cpp
13+ * \brief right_shift graph infer
14+ */
15+ 
16+#include "register/op_impl_registry.h"
17+#include "log/log.h"
18+ 
19+using namespace ge;
20+namespace ops {
21+static constexpr size_t kInputIndex0 = 0U;
22+static constexpr size_t kOutputIndex0 = 0U;
23+static ge::graphStatus InferDataTypeForRightShift(gert::InferDataTypeContext* context)
24+{
25+ OP_LOGI(context->GetNodeName(), "Begin to do InferDtypeForRightShift");
26+ const auto x_data_type = context->GetInputDataType(kInputIndex0);
27+ return context->SetOutputDataType(kOutputIndex0, x_data_type);
28+}
29+ 
30+IMPL_OP(RightShift).InferDataType(InferDataTypeForRightShift);
31+} // namespace ops
Mmath/right_shift/op_graph/right_shift_proto.h+6-9
@@ -7,12 +7,12 @@
7 * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING7 * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
8 * BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. See LICENSE in the root of8 * BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. See LICENSE in the root of
9 * the software repository for the full text of the License.9 * the software repository for the full text of the License.
10- 10+ 
11 * The code snippet comes from Huawei's open-source Mindspore project.11 * The code snippet comes from Huawei's open-source Mindspore project.
12 * Copyright 2019-2020 Huawei Technologies Co., Ltd.12 * Copyright 2019-2020 Huawei Technologies Co., Ltd.
13 * Licensed under the Apache License, Version 2.0 (the "License");13 * Licensed under the Apache License, Version 2.0 (the "License");
14 * You may obtain a copy of the License at14 * You may obtain a copy of the License at
15- * 15+ *
16 * http://www.apache.org/licenses/LICENSE-2.016 * http://www.apache.org/licenses/LICENSE-2.0
17 */17 */
18 18 
@@ -44,13 +44,10 @@ int64, uint8, uint16, uint32, uint64.
44*/44*/
45 45 
46REG_OP(RightShift)46REG_OP(RightShift)
47- .INPUT(x, TensorType({DT_INT8, DT_INT16, DT_INT32, DT_INT64, \47+ .INPUT(x, TensorType({DT_INT8, DT_INT16, DT_INT32, DT_INT64, DT_UINT8, DT_UINT16, DT_UINT32, DT_UINT64}))
48- DT_UINT8, DT_UINT16, DT_UINT32, DT_UINT64}))48+ .INPUT(y, TensorType({DT_INT8, DT_INT16, DT_INT32, DT_INT64, DT_UINT8, DT_UINT16, DT_UINT32, DT_UINT64}))
49- .INPUT(y, TensorType({DT_INT8, DT_INT16, DT_INT32, DT_INT64, \49+ .OUTPUT(z, TensorType({DT_INT8, DT_INT16, DT_INT32, DT_INT64, DT_UINT8, DT_UINT16, DT_UINT32, DT_UINT64}))
50- DT_UINT8, DT_UINT16, DT_UINT32, DT_UINT64}))
51- .OUTPUT(z, TensorType({DT_INT8, DT_INT16, DT_INT32, DT_INT64, \
52- DT_UINT8, DT_UINT16, DT_UINT32, DT_UINT64}))
53 .OP_END_FACTORY_REG(RightShift)50 .OP_END_FACTORY_REG(RightShift)
54-}51+} // namespace ge
55 52 
56#endif53#endif
Amath/right_shift/op_host/arch35/right_shift_tiling_arch35.cpp+175-0
@@ -0,0 +1,175 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file right_shift_tiling_arch35.cpp
13+ * \brief right_shift_tiling source file
14+ */
15+ 
16+#include "right_shift_tiling_arch35.h"
17+#include <graph/utils/type_utils.h>
18+#include "op_host/tiling_templates_registry.h"
19+#include "atvoss/broadcast/broadcast_tiling.h"
20+#include "register/op_impl_registry.h"
21+#include "log/log.h"
22+#include "util/math_util.h"
23+#include "platform/platform_info.h"
24+#include "math/right_shift/op_kernel/arch35/right_shift_dag.h"
25+#include "math/right_shift/op_kernel/arch35/right_shift_struct.h"
26+ 
27+using namespace ge;
28+using namespace RightShiftOp;
29+using namespace Ops::Base;
30+ 
31+namespace optiling {
32+ 
33+static constexpr uint64_t RIGHT_SHIFT_COMMON_TILING_PRIORITY = 0;
34+ 
35+bool RightShiftTiling::CheckDtype(
36+ const ge::DataType& xDtype, const ge::DataType& yDtype, const ge::DataType& zDtype) const
37+{
38+ if (xDtype != yDtype || xDtype != zDtype) {
39+ OP_LOGE(
40+ context_->GetNodeName(), "Dtype of x[%s] should be equal to dtype of y[%s] and z[%s].",
41+ ge::TypeUtils::DataTypeToSerialString(xDtype).c_str(),
42+ ge::TypeUtils::DataTypeToSerialString(yDtype).c_str(),
43+ ge::TypeUtils::DataTypeToSerialString(zDtype).c_str());
44+ return false;
45+ }
46+ return true;
47+}
48+ 
49+bool RightShiftTiling::IsCapable()
50+{
51+ return true;
52+}
53+ 
54+ge::graphStatus RightShiftTiling::GetPlatformInfo()
55+{
56+ return ge::GRAPH_SUCCESS;
57+}
58+ 
59+ge::graphStatus RightShiftTiling::GetShapeAttrsInfo()
60+{
61+ return ge::GRAPH_SUCCESS;
62+}
63+ 
64+ge::graphStatus RightShiftTiling::DoOpTiling()
65+{
66+ OP_LOGD(context_->GetNodeName(), "RightShift DoOpTiling start.");
67+ auto xDesc = context_->GetInputDesc(0);
68+ OP_CHECK_NULL_WITH_CONTEXT(context_, xDesc);
69+ ge::DataType xDType = xDesc->GetDataType();
70+ 
71+ auto yDesc = context_->GetInputDesc(1);
72+ OP_CHECK_NULL_WITH_CONTEXT(context_, yDesc);
73+ ge::DataType yDType = yDesc->GetDataType();
74+ 
75+ auto zDesc = context_->GetOutputDesc(0);
76+ OP_CHECK_NULL_WITH_CONTEXT(context_, zDesc);
77+ ge::DataType zDType = zDesc->GetDataType();
78+ if (!CheckDtype(xDType, yDType, zDType)) {
79+ return ge::GRAPH_FAILED;
80+ }
81+ 
82+ ge::graphStatus baseTilingResult = ge::GRAPH_FAILED;
83+ 
84+ OP_LOGD(context_->GetNodeName(), "Current data type is %s.", ge::TypeUtils::DataTypeToSerialString(xDType).c_str());
85+ if (xDType == ge::DT_INT8) {
86+ BroadcastBaseTiling<RightShiftDag8<int8_t>::OpDag> brcBaseTiling(context_);
87+ baseTilingResult = brcBaseTiling.DoTiling();
88+ tilingKey = GET_TPL_TILING_KEY(brcBaseTiling.GetSchMode(), TPL_INT8);
89+ } else if (xDType == ge::DT_UINT8) {
90+ BroadcastBaseTiling<RightShiftDag8<uint8_t>::OpDag> brcBaseTiling(context_);
91+ baseTilingResult = brcBaseTiling.DoTiling();
92+ tilingKey = GET_TPL_TILING_KEY(brcBaseTiling.GetSchMode(), TPL_UINT8);
93+ } else if (xDType == ge::DT_INT16) {
94+ BroadcastBaseTiling<RightShiftDag16<int16_t>::OpDag> brcBaseTiling(context_);
95+ baseTilingResult = brcBaseTiling.DoTiling();
96+ tilingKey = GET_TPL_TILING_KEY(brcBaseTiling.GetSchMode(), TPL_INT16);
97+ } else if (xDType == ge::DT_UINT16) {
98+ BroadcastBaseTiling<RightShiftDag16<uint16_t>::OpDag> brcBaseTiling(context_);
99+ baseTilingResult = brcBaseTiling.DoTiling();
100+ tilingKey = GET_TPL_TILING_KEY(brcBaseTiling.GetSchMode(), TPL_UINT16);
101+ } else if (xDType == ge::DT_INT32) {
102+ BroadcastBaseTiling<RightShiftDag32<int32_t>::OpDag> brcBaseTiling(context_);
103+ baseTilingResult = brcBaseTiling.DoTiling();
104+ tilingKey = GET_TPL_TILING_KEY(brcBaseTiling.GetSchMode(), TPL_INT32);
105+ } else if (xDType == ge::DT_UINT32) {
106+ BroadcastBaseTiling<RightShiftDag32<uint32_t>::OpDag> brcBaseTiling(context_);
107+ baseTilingResult = brcBaseTiling.DoTiling();
108+ tilingKey = GET_TPL_TILING_KEY(brcBaseTiling.GetSchMode(), TPL_UINT32);
109+ } else if (xDType == ge::DT_INT64) {
110+ BroadcastBaseTiling<RightShiftDag64<int64_t>::OpDag> brcBaseTiling(context_);
111+ baseTilingResult = brcBaseTiling.DoTiling();
112+ tilingKey = GET_TPL_TILING_KEY(brcBaseTiling.GetSchMode(), TPL_INT64);
113+ } else if (xDType == ge::DT_UINT64) {
114+ BroadcastBaseTiling<RightShiftDag64<uint64_t>::OpDag> brcBaseTiling(context_);
115+ baseTilingResult = brcBaseTiling.DoTiling();
116+ tilingKey = GET_TPL_TILING_KEY(brcBaseTiling.GetSchMode(), TPL_UINT64);
117+ } else {
118+ OP_LOGE(
119+ context_->GetNodeName(),
120+ "Input dtype is only support int8, int16, int32, int64, uint8, uint16, uint32, uint64, "
121+ "while got %s!",
122+ ge::TypeUtils::DataTypeToSerialString(xDType).c_str());
123+ return ge::GRAPH_FAILED;
124+ }
125+ OP_CHECK_IF(
126+ baseTilingResult == ge::GRAPH_FAILED, OP_LOGE(context_->GetNodeName(), "broadcastBaseTiling doTiling failed."),
127+ return ge::GRAPH_FAILED);
128+ OP_LOGD(context_->GetNodeName(), "[RightShiftTilingData] : tilingKey=%lu", tilingKey);
129+ return baseTilingResult;
130+}
131+ 
132+ge::graphStatus RightShiftTiling::DoLibApiTiling()
133+{
134+ return ge::GRAPH_SUCCESS;
135+}
136+ 
137+uint64_t RightShiftTiling::GetTilingKey() const
138+{
139+ return tilingKey;
140+}
141+ 
142+ge::graphStatus RightShiftTiling::GetWorkspaceSize()
143+{
144+ return ge::GRAPH_SUCCESS;
145+}
146+ 
147+ge::graphStatus RightShiftTiling::PostTiling()
148+{
149+ return ge::GRAPH_SUCCESS;
150+}
151+ 
152+ge::graphStatus Tiling4RightShift(gert::TilingContext* context)
153+{
154+ OP_LOGD("RightShiftTiling", "Enter Tiling4RightShift");
155+ if (context == nullptr) {
156+ OP_LOGE("Tiling4RightShift", "Tiling context is nullptr");
157+ return ge::GRAPH_FAILED;
158+ }
159+ 
160+ auto compileInfo = reinterpret_cast<const BroadcastCompileInfo*>(context->GetCompileInfo());
161+ OP_CHECK_NULL_WITH_CONTEXT(context, compileInfo);
162+ 
163+ OP_LOGD(context, "Enter ascendc Tiling4RightShift");
164+ RightShiftTiling tiling(context);
165+ return tiling.DoTiling();
166+}
167+ 
168+static ge::graphStatus TilingPrepareForRightShift([[maybe_unused]] gert::TilingParseContext* context)
169+{
170+ return ge::GRAPH_SUCCESS;
171+}
172+ 
173+IMPL_OP_OPTILING(RightShift).Tiling(Tiling4RightShift).TilingParse<BroadcastCompileInfo>(TilingPrepareForRightShift);
174+ 
175+} // namespace optiling
Amath/right_shift/op_host/arch35/right_shift_tiling_arch35.h+46-0
@@ -0,0 +1,46 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file right_shift_tiling_arch35.h
13+ * \brief right_shift_tiling head file
14+ */
15+#ifndef OPS_MATH_RIGHT_SHIFT_TILING_ARCH35_H
16+#define OPS_MATH_RIGHT_SHIFT_TILING_ARCH35_H
17+ 
18+#include "register/tilingdata_base.h"
19+#include "atvoss/elewise/elewise_tiling.h"
20+#include "op_host/tiling_base.h"
21+#include "platform/platform_ascendc.h"
22+ 
23+namespace optiling {
24+using namespace Ops::Math::OpTiling;
25+ 
26+class RightShiftTiling : public TilingBaseClass {
27+public:
28+ explicit RightShiftTiling(gert::TilingContext* context) : TilingBaseClass(context)
29+ {}
30+ 
31+protected:
32+ bool IsCapable() override;
33+ ge::graphStatus GetPlatformInfo() override;
34+ ge::graphStatus GetShapeAttrsInfo() override;
35+ ge::graphStatus DoOpTiling() override;
36+ ge::graphStatus DoLibApiTiling() override;
37+ uint64_t GetTilingKey() const override;
38+ ge::graphStatus GetWorkspaceSize() override;
39+ ge::graphStatus PostTiling() override;
40+ 
41+private:
42+ uint64_t tilingKey = 0;
43+ bool CheckDtype(const ge::DataType& xDtype, const ge::DataType& yDtype, const ge::DataType& zDtype) const;
44+};
45+} // namespace optiling
46+#endif // OPS_MATH_RIGHT_SHIFT_TILING_ARCH35_H
Amath/right_shift/op_host/config/ascend950/right_shift_binary.json+325-0
@@ -0,0 +1,325 @@
1+{
2+ "op_type": "RightShift",
3+ "op_list": [
4+ {
5+ "bin_filename": "RightShift_int8",
6+ "inputs": [
7+ {
8+ "name": "x",
9+ "index": 0,
10+ "dtype": "int8",
11+ "format": "ND",
12+ "paramType": "required",
13+ "shape": [
14+ -2
15+ ],
16+ "format_match_mode": "FormatAgnostic"
17+ },
18+ {
19+ "name": "y",
20+ "index": 1,
21+ "dtype": "int8",
22+ "format": "ND",
23+ "paramType": "required",
24+ "shape": [
25+ -2
26+ ],
27+ "format_match_mode": "FormatAgnostic"
28+ }
29+ ],
30+ "outputs": [
31+ {
32+ "name": "z",
33+ "index": 0,
34+ "dtype": "int8",
35+ "format": "ND",
36+ "paramType": "required",
37+ "shape": [
38+ -2
39+ ],
40+ "format_match_mode": "FormatAgnostic"
41+ }
42+ ]
43+ },
44+ {
45+ "bin_filename": "RightShift_uint8",
46+ "inputs": [
47+ {
48+ "name": "x",
49+ "index": 0,
50+ "dtype": "uint8",
51+ "format": "ND",
52+ "paramType": "required",
53+ "shape": [
54+ -2
55+ ],
56+ "format_match_mode": "FormatAgnostic"
57+ },
58+ {
59+ "name": "y",
60+ "index": 1,
61+ "dtype": "uint8",
62+ "format": "ND",
63+ "paramType": "required",
64+ "shape": [
65+ -2
66+ ],
67+ "format_match_mode": "FormatAgnostic"
68+ }
69+ ],
70+ "outputs": [
71+ {
72+ "name": "z",
73+ "index": 0,
74+ "dtype": "uint8",
75+ "format": "ND",
76+ "paramType": "required",
77+ "shape": [
78+ -2
79+ ],
80+ "format_match_mode": "FormatAgnostic"
81+ }
82+ ]
83+ },
84+ {
85+ "bin_filename": "RightShift_int16",
86+ "inputs": [
87+ {
88+ "name": "x",
89+ "index": 0,
90+ "dtype": "int16",
91+ "format": "ND",
92+ "paramType": "required",
93+ "shape": [
94+ -2
95+ ],
96+ "format_match_mode": "FormatAgnostic"
97+ },
98+ {
99+ "name": "y",
100+ "index": 1,
101+ "dtype": "int16",
102+ "format": "ND",
103+ "paramType": "required",
104+ "shape": [
105+ -2
106+ ],
107+ "format_match_mode": "FormatAgnostic"
108+ }
109+ ],
110+ "outputs": [
111+ {
112+ "name": "z",
113+ "index": 0,
114+ "dtype": "int16",
115+ "format": "ND",
116+ "paramType": "required",
117+ "shape": [
118+ -2
119+ ],
120+ "format_match_mode": "FormatAgnostic"
121+ }
122+ ]
123+ },
124+ {
125+ "bin_filename": "RightShift_uint16",
126+ "inputs": [
127+ {
128+ "name": "x",
129+ "index": 0,
130+ "dtype": "uint16",
131+ "format": "ND",
132+ "paramType": "required",
133+ "shape": [
134+ -2
135+ ],
136+ "format_match_mode": "FormatAgnostic"
137+ },
138+ {
139+ "name": "y",
140+ "index": 1,
141+ "dtype": "uint16",
142+ "format": "ND",
143+ "paramType": "required",
144+ "shape": [
145+ -2
146+ ],
147+ "format_match_mode": "FormatAgnostic"
148+ }
149+ ],
150+ "outputs": [
151+ {
152+ "name": "z",
153+ "index": 0,
154+ "dtype": "uint16",
155+ "format": "ND",
156+ "paramType": "required",
157+ "shape": [
158+ -2
159+ ],
160+ "format_match_mode": "FormatAgnostic"
161+ }
162+ ]
163+ },
164+ {
165+ "bin_filename": "RightShift_int32",
166+ "inputs": [
167+ {
168+ "name": "x",
169+ "index": 0,
170+ "dtype": "int32",
171+ "format": "ND",
172+ "paramType": "required",
173+ "shape": [
174+ -2
175+ ],
176+ "format_match_mode": "FormatAgnostic"
177+ },
178+ {
179+ "name": "y",
180+ "index": 1,
181+ "dtype": "int32",
182+ "format": "ND",
183+ "paramType": "required",
184+ "shape": [
185+ -2
186+ ],
187+ "format_match_mode": "FormatAgnostic"
188+ }
189+ ],
190+ "outputs": [
191+ {
192+ "name": "z",
193+ "index": 0,
194+ "dtype": "int32",
195+ "format": "ND",
196+ "paramType": "required",
197+ "shape": [
198+ -2
199+ ],
200+ "format_match_mode": "FormatAgnostic"
201+ }
202+ ]
203+ },
204+ {
205+ "bin_filename": "RightShift_uint32",
206+ "inputs": [
207+ {
208+ "name": "x",
209+ "index": 0,
210+ "dtype": "uint32",
211+ "format": "ND",
212+ "paramType": "required",
213+ "shape": [
214+ -2
215+ ],
216+ "format_match_mode": "FormatAgnostic"
217+ },
218+ {
219+ "name": "y",
220+ "index": 1,
221+ "dtype": "uint32",
222+ "format": "ND",
223+ "paramType": "required",
224+ "shape": [
225+ -2
226+ ],
227+ "format_match_mode": "FormatAgnostic"
228+ }
229+ ],
230+ "outputs": [
231+ {
232+ "name": "z",
233+ "index": 0,
234+ "dtype": "uint32",
235+ "format": "ND",
236+ "paramType": "required",
237+ "shape": [
238+ -2
239+ ],
240+ "format_match_mode": "FormatAgnostic"
241+ }
242+ ]
243+ },
244+ {
245+ "bin_filename": "RightShift_int64",
246+ "inputs": [
247+ {
248+ "name": "x",
249+ "index": 0,
250+ "dtype": "int64",
251+ "format": "ND",
252+ "paramType": "required",
253+ "shape": [
254+ -2
255+ ],
256+ "format_match_mode": "FormatAgnostic"
257+ },
258+ {
259+ "name": "y",
260+ "index": 1,
261+ "dtype": "int64",
262+ "format": "ND",
263+ "paramType": "required",
264+ "shape": [
265+ -2
266+ ],
267+ "format_match_mode": "FormatAgnostic"
268+ }
269+ ],
270+ "outputs": [
271+ {
272+ "name": "z",
273+ "index": 0,
274+ "dtype": "int64",
275+ "format": "ND",
276+ "paramType": "required",
277+ "shape": [
278+ -2
279+ ],
280+ "format_match_mode": "FormatAgnostic"
281+ }
282+ ]
283+ },
284+ {
285+ "bin_filename": "RightShift_uint64",
286+ "inputs": [
287+ {
288+ "name": "x",
289+ "index": 0,
290+ "dtype": "uint64",
291+ "format": "ND",
292+ "paramType": "required",
293+ "shape": [
294+ -2
295+ ],
296+ "format_match_mode": "FormatAgnostic"
297+ },
298+ {
299+ "name": "y",
300+ "index": 1,
301+ "dtype": "uint64",
302+ "format": "ND",
303+ "paramType": "required",
304+ "shape": [
305+ -2
306+ ],
307+ "format_match_mode": "FormatAgnostic"
308+ }
309+ ],
310+ "outputs": [
311+ {
312+ "name": "z",
313+ "index": 0,
314+ "dtype": "uint64",
315+ "format": "ND",
316+ "paramType": "required",
317+ "shape": [
318+ -2
319+ ],
320+ "format_match_mode": "FormatAgnostic"
321+ }
322+ ]
323+ }
324+ ]
325+}
Amath/right_shift/op_host/config/ascend950/right_shift_simplified_key.ini+2-0
@@ -0,0 +1,2 @@
1+[RightShift]
2+default=0
Dmath/right_shift/op_host/op_api/aclnn_right_shift.h+0-78
@@ -1,78 +0,0 @@
1-/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4- * CANN Open Software License Agreement Version 2.0 (the "License").
5- * Please refer to the License for details. You may not use this file except in compliance with the License.
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8- * See LICENSE in the root of the software repository for the full text of the License.
9- */
10- 
11-#ifndef OP_API_INC_LEVEL2_ACLNN_RIGHT_SHIFT_H_
12-#define OP_API_INC_LEVEL2_ACLNN_RIGHT_SHIFT_H_
13- 
14-#include "aclnn/aclnn_base.h"
15-#include "aclnn_util.h"
16- 
17-#ifdef __cplusplus
18-extern "C" {
19-#endif
20-/**
21- * @brief aclnnRightShift的第一段接口，根据具体的计算流程，计算workspace大小。
22- * 功能描述：输入张量input中每个元素，根据shiftBits对应位置的参数，按位进行右移。
23- * 计算公式：
24- * output_{i} = input_{i}>>shiftBits_{i}
25- * @domain aclnn_math
26- * 参数描述：
27- * @param [in] input
28- * 输入Tensor，数据类型支持 INT8, INT16, INT32, INT64, UINT8, UINT16, UINT32, UINT64。
29- * 数据类型需要与shiftBits构成互相推导关系，input需要与shiftBits满足broadcast关系。
30- * 支持非连续的Tensor，数据格式支持ND。
31- * @param [in] shiftBits
32- * 输入Tensor，数据类型支持 INT8, INT16, INT32, INT64, UINT8, UINT16, UINT32, UINT64。
33- * 数据类型需要与input构成互相推导关系，input需要与shiftBits满足broadcast关系。
34- * 支持非连续的Tensor，数据格式支持ND。
35- * @param [out] out
36- * 输出Tensor，数据类型支持 INT8, INT16, INT32, INT64, UINT8, UINT16, UINT32, UINT64。
37- * shape需要与input和shiftBits做broadcast后的shape一致。
38- * 支持非连续的Tensor，数据格式支持ND。
39- * @param [out] workspaceSize 返回用户需要在npu device侧申请的workspace大小。
40- * @param [out] executor 返回op执行器，包含了算子计算流程。
41- * @return aclnnStatus: 返回状态码。
42- */
43-ACLNN_API aclnnStatus aclnnRightShiftGetWorkspaceSize(const aclTensor *input, const aclTensor *shiftBits,
44- aclTensor *out, uint64_t *workspaceSize, aclOpExecutor **executor);
45- 
46-/**
47- * @brief aclnnRightShift的第二段接口，用于执行计算。
48- * 功能描述：输入张量input中每个元素，根据shiftBits对应位置的参数，按位进行右移。
49- * 计算公式：
50- * output_{i} = input_{i}>>shiftBits_{i}
51- * @domain aclnn_math
52- * 实现说明：
53- * api计算的基本路径：
54- * ```mermaid
55- * flowchart LR
56- * input[(input)]-->l0op::Contiguous_input([l0op::Contiguous])
57- * l0op::Contiguous_input([l0op::Contiguous])-->l0op::Cast_input([l0op::Cast])
58- * l0op::Cast_input([l0op::Cast])-->l0op::rightshift([l0op::rightshift])
59- * shiftBits[(shiftBits)]-->l0op::Contiguous_shiftBits([l0op::Contiguous])
60- * -->l0op::Cast_shiftBits([l0op::Cast])-->l0op::rightshift([l0op::RightShift])
61- * l0op::rightshift([l0op::RightShift])-->l0op::Cast_out([l0op::Cast])
62- * l0op::Cast_out([l0op::Cast])-->l0op::ViewCopy_out([l0op::ViewCopy])-->out[(out)]
63- * ```
64- * @param [in] workspace: 在npu device侧申请的workspace内存起址。
65- * @param [in] workspaceSize: 在npu device侧申请的workspace大小，由第一段接口aclnnRightShiftGetWorkspaceSize获取。
66- * @param [in] executor: op执行器，包含了算子计算流程。
67- * @param [in] stream: acl stream流。
68- * @return aclnnStatus: 返回状态码。
69- */
70-ACLNN_API aclnnStatus aclnnRightShift(void *workspace, uint64_t workspaceSize,
71- aclOpExecutor *executor, aclrtStream stream);
72- 
73- 
74-#ifdef __cplusplus
75-}
76-#endif
77- 
78-#endif // OP_API_INC_LEVEL2_ACLNN_RIGHT_SHIFT_H_
Amath/right_shift/op_host/right_shift_def.cpp+70-0
@@ -0,0 +1,70 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file right_shift_def.cpp
13+ * \brief RightShift def
14+ */
15+ 
16+#include <cstdint>
17+#include "register/op_def_registry.h"
18+ 
19+namespace ops {
20+class RightShift : public OpDef {
21+public:
22+ explicit RightShift(const char* name) : OpDef(name)
23+ {
24+ this->Input("x")
25+ .ParamType(REQUIRED)
26+ .DataType(
27+ {ge::DT_INT8, ge::DT_UINT8, ge::DT_INT16, ge::DT_UINT16, ge::DT_INT32, ge::DT_UINT32, ge::DT_INT64,
28+ ge::DT_UINT64})
29+ .Format(
30+ {ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
31+ ge::FORMAT_ND, ge::FORMAT_ND})
32+ .UnknownShapeFormat(
33+ {ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
34+ ge::FORMAT_ND, ge::FORMAT_ND});
35+ this->Input("y")
36+ .ParamType(REQUIRED)
37+ .DataType(
38+ {ge::DT_INT8, ge::DT_UINT8, ge::DT_INT16, ge::DT_UINT16, ge::DT_INT32, ge::DT_UINT32, ge::DT_INT64,
39+ ge::DT_UINT64})
40+ .Format(
41+ {ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
42+ ge::FORMAT_ND, ge::FORMAT_ND})
43+ .UnknownShapeFormat(
44+ {ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
45+ ge::FORMAT_ND, ge::FORMAT_ND});
46+ this->Output("z")
47+ .ParamType(REQUIRED)
48+ .DataType(
49+ {ge::DT_INT8, ge::DT_UINT8, ge::DT_INT16, ge::DT_UINT16, ge::DT_INT32, ge::DT_UINT32, ge::DT_INT64,
50+ ge::DT_UINT64})
51+ .Format(
52+ {ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
53+ ge::FORMAT_ND, ge::FORMAT_ND})
54+ .UnknownShapeFormat(
55+ {ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
56+ ge::FORMAT_ND, ge::FORMAT_ND});
57+ 
58+ OpAICoreConfig aicoreConfig;
59+ aicoreConfig.DynamicCompileStaticFlag(true)
60+ .DynamicFormatFlag(false)
61+ .DynamicRankSupportFlag(true)
62+ .DynamicShapeSupportFlag(true)
63+ .NeedCheckSupportFlag(false)
64+ .PrecisionReduceFlag(true)
65+ .ExtendCfgInfo("opFile.value", "right_shift_apt");
66+ this->AICore().AddConfig("ascend950", aicoreConfig);
67+ }
68+};
69+OP_ADD(RightShift);
70+} // namespace ops
Amath/right_shift/op_host/right_shift_infershape.cpp+43-0
@@ -0,0 +1,43 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file right_shift_infershape.cpp
13+ * \brief RightShift infershape
14+ */
15+ 
16+#include "register/op_impl_registry.h"
17+#include "op_host/infershape_broadcast_util.h"
18+#include "log/log.h"
19+ 
20+using namespace ge;
21+using namespace Ops::Base;
22+ 
23+namespace ops {
24+static ge::graphStatus InferShapeForRightShift(gert::InferShapeContext* context)
25+{
26+ OP_LOGI(context->GetNodeName(), "Begin to do InferShapeForRightShift");
27+ const gert::Shape* x_shape = context->GetInputShape(0);
28+ OP_CHECK_NULL_WITH_CONTEXT(context, x_shape);
29+ const gert::Shape* y_shape = context->GetInputShape(1);
30+ OP_CHECK_NULL_WITH_CONTEXT(context, y_shape);
31+ gert::Shape* z_shape = context->GetOutputShape(0);
32+ OP_CHECK_NULL_WITH_CONTEXT(context, z_shape);
33+ OP_CHECK_IF(
34+ !BroadcastShape(x_shape, y_shape, z_shape),
35+ OP_LOGE(
36+ context->GetNodeName(), "shape %s and %s cannot broadcast!", ToString(*x_shape).c_str(),
37+ ToString(*y_shape).c_str()),
38+ return ge::GRAPH_FAILED);
39+ return GRAPH_SUCCESS;
40+}
41+ 
42+IMPL_OP_INFERSHAPE(RightShift).InferShape(InferShapeForRightShift);
43+} // namespace ops
Amath/right_shift/op_kernel/arch35/right_shift_dag.h+302-0
@@ -0,0 +1,302 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file right_shift_dag.h
13+ * \brief right_shift dag
14+ */
15+ 
16+#ifndef OPS_MATH_RIGHT_SHIFT_DAG_H
17+#define OPS_MATH_RIGHT_SHIFT_DAG_H
18+ 
19+#include "atvoss/util/dag.h"
20+#include "atvoss/util/vec.h"
21+#include "atvoss/util/placeholder.h"
22+ 
23+namespace RightShiftOp {
24+using namespace Ops::Base;
25+ 
26+template <typename T>
27+struct RightShiftCustom8 : public Vec::ElemwiseBinaryOP<T, T, T> {
28+ __aicore__ inline RightShiftCustom8(LocalTensor<T>& dst, LocalTensor<T>& src1, LocalTensor<T>& src2, uint32_t count)
29+ {
30+#ifdef __CCE_AICORE__
31+ uint32_t dtypeSize = sizeof(T);
32+ int8_t digitZero = 0;
33+ int8_t rightShifts = 7;
34+ constexpr uint64_t VECTOR_REG_WIDTH = 256UL;
35+ uint32_t vl = VECTOR_REG_WIDTH / dtypeSize;
36+ uint16_t loopNum = (count + vl - 1) / vl;
37+ uint32_t vlSize = vl;
38+ 
39+ __ubuf__ T* src1Addr = (__ubuf__ T*)src1.GetPhyAddr();
40+ __ubuf__ T* src2Addr = (__ubuf__ T*)src2.GetPhyAddr();
41+ __ubuf__ T* dstAddr = (__ubuf__ T*)dst.GetPhyAddr();
42+ 
43+ MicroAPI::RegTensor<T> xReg;
44+ MicroAPI::RegTensor<T> yReg;
45+ MicroAPI::RegTensor<T> zReg;
46+ MicroAPI::RegTensor<T> zeroReg;
47+ MicroAPI::RegTensor<T> sizeReg;
48+ 
49+ MicroAPI::MaskReg mask;
50+ MicroAPI::MaskReg calcMask;
51+ 
52+ __VEC_SCOPE__
53+ {
54+ MicroAPI::MaskReg scalarMaskReg = MicroAPI::CreateMask<T, MicroAPI::MaskPattern::ALL>();
55+ MicroAPI::Duplicate(zeroReg, digitZero, scalarMaskReg);
56+ MicroAPI::Duplicate(sizeReg, rightShifts, scalarMaskReg);
57+ 
58+ for (uint16_t loopIdx = 0; loopIdx < loopNum; loopIdx++) {
59+ mask = MicroAPI::UpdateMask<T, MicroAPI::RegTraitNumOne>(count);
60+ MicroAPI::DataCopy(xReg, (__ubuf__ T*)(src1Addr + loopIdx * vlSize));
61+ MicroAPI::DataCopy(yReg, (__ubuf__ T*)(src2Addr + loopIdx * vlSize));
62+ 
63+ if constexpr (std::is_same_v<T, int8_t>) {
64+ MicroAPI::Compare<T, CMPMODE::GE>(calcMask, yReg, zeroReg, mask);
65+ MicroAPI::Select(yReg, yReg, zeroReg, calcMask);
66+ MicroAPI::ShiftRight(zReg, xReg, (MicroAPI::RegTensor<T>&)yReg, mask);
67+ } else if constexpr (std::is_same_v<T, uint8_t>) {
68+ MicroAPI::Compare<T, CMPMODE::LE>(calcMask, yReg, sizeReg, mask);
69+ MicroAPI::Select(yReg, yReg, sizeReg, calcMask);
70+ MicroAPI::ShiftRight(zReg, xReg, (MicroAPI::RegTensor<int8_t>&)yReg, mask);
71+ }
72+ MicroAPI::DataCopy((__ubuf__ T*)(dstAddr + loopIdx * vlSize), zReg, mask);
73+ }
74+ }
75+#endif
76+ }
77+};
78+ 
79+template <typename T>
80+struct RightShiftCustom16 : public Vec::ElemwiseBinaryOP<T, T, T> {
81+ __aicore__ inline RightShiftCustom16(
82+ LocalTensor<T>& dst, LocalTensor<T>& src1, LocalTensor<T>& src2, uint32_t count)
83+ {
84+#ifdef __CCE_AICORE__
85+ uint32_t dtypeSize = sizeof(T);
86+ int8_t digitZero = 0;
87+ int8_t rightShifts = 15;
88+ constexpr uint64_t VECTOR_REG_WIDTH = 256UL;
89+ uint32_t vl = VECTOR_REG_WIDTH / dtypeSize;
90+ uint16_t loopNum = (count + vl - 1) / vl;
91+ uint32_t vlSize = vl;
92+ 
93+ __ubuf__ T* src1Addr = (__ubuf__ T*)src1.GetPhyAddr();
94+ __ubuf__ T* src2Addr = (__ubuf__ T*)src2.GetPhyAddr();
95+ __ubuf__ T* dstAddr = (__ubuf__ T*)dst.GetPhyAddr();
96+ 
97+ MicroAPI::RegTensor<T> xReg;
98+ MicroAPI::RegTensor<T> yReg;
99+ MicroAPI::RegTensor<T> zReg;
100+ MicroAPI::RegTensor<T> zeroReg;
101+ MicroAPI::RegTensor<T> sizeReg;
102+ 
103+ MicroAPI::MaskReg mask;
104+ MicroAPI::MaskReg calcMask;
105+ 
106+ __VEC_SCOPE__
107+ {
108+ MicroAPI::MaskReg scalarMaskReg = MicroAPI::CreateMask<T, MicroAPI::MaskPattern::ALL>();
109+ MicroAPI::Duplicate(zeroReg, digitZero, scalarMaskReg);
110+ MicroAPI::Duplicate(sizeReg, rightShifts, scalarMaskReg);
111+ 
112+ for (uint16_t loopIdx = 0; loopIdx < loopNum; loopIdx++) {
113+ mask = MicroAPI::UpdateMask<T, MicroAPI::RegTraitNumOne>(count);
114+ MicroAPI::DataCopy(xReg, (__ubuf__ T*)(src1Addr + loopIdx * vlSize));
115+ MicroAPI::DataCopy(yReg, (__ubuf__ T*)(src2Addr + loopIdx * vlSize));
116+ 
117+ if constexpr (std::is_same_v<T, int16_t>) {
118+ MicroAPI::Compare<T, CMPMODE::GE>(calcMask, yReg, zeroReg, mask);
119+ MicroAPI::Select(yReg, yReg, zeroReg, calcMask);
120+ MicroAPI::ShiftRight(zReg, xReg, (MicroAPI::RegTensor<T>&)yReg, mask);
121+ } else if constexpr (std::is_same_v<T, uint16_t>) {
122+ MicroAPI::Compare<T, CMPMODE::LE>(calcMask, yReg, sizeReg, mask);
123+ MicroAPI::Select(yReg, yReg, sizeReg, calcMask);
124+ MicroAPI::ShiftRight(zReg, xReg, (MicroAPI::RegTensor<int16_t>&)yReg, mask);
125+ }
126+ MicroAPI::DataCopy((__ubuf__ T*)(dstAddr + loopIdx * vlSize), zReg, mask);
127+ }
128+ }
129+#endif
130+ }
131+};
132+ 
133+template <typename T>
134+struct RightShiftCustom32 : public Vec::ElemwiseBinaryOP<T, T, T> {
135+ __aicore__ inline RightShiftCustom32(
136+ LocalTensor<T>& dst, LocalTensor<T>& src1, LocalTensor<T>& src2, uint32_t count)
137+ {
138+#ifdef __CCE_AICORE__
139+ uint32_t dtypeSize = sizeof(T);
140+ int8_t digitZero = 0;
141+ int8_t rightShifts = 31;
142+ constexpr uint64_t VECTOR_REG_WIDTH = 256UL;
143+ uint32_t vl = VECTOR_REG_WIDTH / dtypeSize;
144+ uint16_t loopNum = (count + vl - 1) / vl;
145+ uint32_t vlSize = vl;
146+ 
147+ __ubuf__ T* src1Addr = (__ubuf__ T*)src1.GetPhyAddr();
148+ __ubuf__ T* src2Addr = (__ubuf__ T*)src2.GetPhyAddr();
149+ __ubuf__ T* dstAddr = (__ubuf__ T*)dst.GetPhyAddr();
150+ 
151+ MicroAPI::RegTensor<T> xReg;
152+ MicroAPI::RegTensor<T> yReg;
153+ MicroAPI::RegTensor<T> zReg;
154+ MicroAPI::RegTensor<T> zeroReg;
155+ MicroAPI::RegTensor<T> sizeReg;
156+ 
157+ MicroAPI::MaskReg mask;
158+ MicroAPI::MaskReg calcMask;
159+ 
160+ __VEC_SCOPE__
161+ {
162+ MicroAPI::MaskReg scalarMaskReg = MicroAPI::CreateMask<T, MicroAPI::MaskPattern::ALL>();
163+ MicroAPI::Duplicate(zeroReg, digitZero, scalarMaskReg);
164+ MicroAPI::Duplicate(sizeReg, rightShifts, scalarMaskReg);
165+ 
166+ for (uint16_t loopIdx = 0; loopIdx < loopNum; loopIdx++) {
167+ mask = MicroAPI::UpdateMask<T, MicroAPI::RegTraitNumOne>(count);
168+ MicroAPI::DataCopy(xReg, (__ubuf__ T*)(src1Addr + loopIdx * vlSize));
169+ MicroAPI::DataCopy(yReg, (__ubuf__ T*)(src2Addr + loopIdx * vlSize));
170+ 
171+ if constexpr (std::is_same_v<T, int32_t>) {
172+ MicroAPI::Compare<T, CMPMODE::GE>(calcMask, yReg, zeroReg, mask);
173+ MicroAPI::Select(yReg, yReg, zeroReg, calcMask);
174+ MicroAPI::ShiftRight(zReg, xReg, (MicroAPI::RegTensor<T>&)yReg, mask);
175+ } else if constexpr (std::is_same_v<T, uint32_t>) {
176+ MicroAPI::Compare<T, CMPMODE::LE>(calcMask, yReg, sizeReg, mask);
177+ MicroAPI::Select(yReg, yReg, sizeReg, calcMask);
178+ MicroAPI::ShiftRight(zReg, xReg, (MicroAPI::RegTensor<int32_t>&)yReg, mask);
179+ }
180+ MicroAPI::DataCopy((__ubuf__ T*)(dstAddr + loopIdx * vlSize), zReg, mask);
181+ }
182+ }
183+#endif
184+ }
185+};
186+ 
187+template <typename T>
188+struct RightShiftCustom64 : public Vec::ElemwiseBinaryOP<T, T, T> {
189+ __aicore__ inline RightShiftCustom64(
190+ LocalTensor<T>& dst, LocalTensor<T>& src1, LocalTensor<T>& src2, uint32_t count)
191+ {
192+#ifdef __CCE_AICORE__
193+ uint32_t dtypeSize = sizeof(T);
194+ int8_t digitZero = 0;
195+ int8_t rightShifts = 63;
196+ constexpr uint64_t VECTOR_REG_WIDTH = 256UL;
197+ uint32_t vl = VECTOR_REG_WIDTH / dtypeSize;
198+ uint16_t loopNum = (count + vl - 1) / vl;
199+ uint32_t vlSize = vl;
200+ 
201+ __ubuf__ T* src1Addr = (__ubuf__ T*)src1.GetPhyAddr();
202+ __ubuf__ T* src2Addr = (__ubuf__ T*)src2.GetPhyAddr();
203+ __ubuf__ T* dstAddr = (__ubuf__ T*)dst.GetPhyAddr();
204+ 
205+ MicroAPI::RegTensor<T> xReg;
206+ MicroAPI::RegTensor<T> yReg;
207+ MicroAPI::RegTensor<T> zReg;
208+ MicroAPI::RegTensor<T> zeroReg;
209+ MicroAPI::RegTensor<T> sizeReg;
210+ 
211+ MicroAPI::MaskReg mask;
212+ MicroAPI::MaskReg calcMask;
213+ 
214+ __VEC_SCOPE__
215+ {
216+ MicroAPI::MaskReg scalarMaskReg = MicroAPI::CreateMask<T, MicroAPI::MaskPattern::ALL>();
217+ MicroAPI::Duplicate(zeroReg, digitZero, scalarMaskReg);
218+ MicroAPI::Duplicate(sizeReg, rightShifts, scalarMaskReg);
219+ 
220+ for (uint16_t loopIdx = 0; loopIdx < loopNum; loopIdx++) {
221+ mask = MicroAPI::UpdateMask<T, MicroAPI::RegTraitNumOne>(count);
222+ MicroAPI::DataCopy(xReg, (__ubuf__ T*)(src1Addr + loopIdx * vlSize));
223+ MicroAPI::DataCopy(yReg, (__ubuf__ T*)(src2Addr + loopIdx * vlSize));
224+ 
225+ if constexpr (std::is_same_v<T, int64_t>) {
226+ MicroAPI::Compare<T, CMPMODE::GE>(calcMask, yReg, zeroReg, mask);
227+ MicroAPI::Select(yReg, yReg, zeroReg, calcMask);
228+ MicroAPI::ShiftRight(zReg, xReg, (MicroAPI::RegTensor<T>&)yReg, mask);
229+ } else if constexpr (std::is_same_v<T, uint64_t>) {
230+ MicroAPI::Compare<T, CMPMODE::LE>(calcMask, yReg, sizeReg, mask);
231+ MicroAPI::Select(yReg, yReg, sizeReg, calcMask);
232+ MicroAPI::ShiftRight(zReg, xReg, (MicroAPI::RegTensor<int64_t>&)yReg, mask);
233+ }
234+ MicroAPI::DataCopy((__ubuf__ T*)(dstAddr + loopIdx * vlSize), zReg, mask);
235+ }
236+ }
237+#endif
238+ }
239+};
240+ 
241+template <typename T>
242+struct RightShiftDag8 {
243+ using InputX = Bind<Vec::CopyInBrc<T>, Placeholder::In0<T>>;
244+ using InputY = Bind<Vec::CopyInBrc<T>, Placeholder::In1<T>>;
245+ 
246+ using OpResult = Bind<RightShiftCustom8<T>, InputX, InputY>;
247+ 
248+ using OpCopyOut = Bind<Vec::CopyOut<T>, Placeholder::Out0<T>, OpResult>;
249+ 
250+ // 指定输出节点
251+ using Outputs = Elems<OpCopyOut>;
252+ using MemCfg = MemOptCfg<MemLevel::LEVEL_2>;
253+ using OpDag = DAGSch<Outputs, void, MemCfg>;
254+};
255+ 
256+template <typename T>
257+struct RightShiftDag16 {
258+ using InputX = Bind<Vec::CopyInBrc<T>, Placeholder::In0<T>>;
259+ using InputY = Bind<Vec::CopyInBrc<T>, Placeholder::In1<T>>;
260+ 
261+ using OpResult = Bind<RightShiftCustom16<T>, InputX, InputY>;
262+ 
263+ using OpCopyOut = Bind<Vec::CopyOut<T>, Placeholder::Out0<T>, OpResult>;
264+ 
265+ // 指定输出节点
266+ using Outputs = Elems<OpCopyOut>;
267+ using MemCfg = MemOptCfg<MemLevel::LEVEL_2>;
268+ using OpDag = DAGSch<Outputs, void, MemCfg>;
269+};
270+ 
271+template <typename T>
272+struct RightShiftDag32 {
273+ using InputX = Bind<Vec::CopyInBrc<T>, Placeholder::In0<T>>;
274+ using InputY = Bind<Vec::CopyInBrc<T>, Placeholder::In1<T>>;
275+ 
276+ using OpResult = Bind<RightShiftCustom32<T>, InputX, InputY>;
277+ 
278+ using OpCopyOut = Bind<Vec::CopyOut<T>, Placeholder::Out0<T>, OpResult>;
279+ 
280+ // 指定输出节点
281+ using Outputs = Elems<OpCopyOut>;
282+ using MemCfg = MemOptCfg<MemLevel::LEVEL_2>;
283+ using OpDag = DAGSch<Outputs, void, MemCfg>;
284+};
285+ 
286+template <typename T>
287+struct RightShiftDag64 {
288+ using InputX = Bind<Vec::CopyInBrc<T>, Placeholder::In0<T>>;
289+ using InputY = Bind<Vec::CopyInBrc<T>, Placeholder::In1<T>>;
290+ 
291+ using OpResult = Bind<RightShiftCustom64<T>, InputX, InputY>;
292+ 
293+ using OpCopyOut = Bind<Vec::CopyOut<T>, Placeholder::Out0<T>, OpResult>;
294+ 
295+ // 指定输出节点
296+ using Outputs = Elems<OpCopyOut>;
297+ using MemCfg = MemOptCfg<MemLevel::LEVEL_2>;
298+ using OpDag = DAGSch<Outputs, void, MemCfg>;
299+};
300+} // namespace RightShiftOp
301+ 
302+#endif // OPS_MATH_RIGHT_SHIFT_DAG_H
Amath/right_shift/op_kernel/arch35/right_shift_struct.h+51-0
@@ -0,0 +1,51 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file right_shift_struct.h
13+ * \brief right_shift_struct
14+ */
15+ 
16+#ifndef OPS_MATH_RIGHT_SHIFT_STRUCT_H
17+#define OPS_MATH_RIGHT_SHIFT_STRUCT_H
18+ 
19+#include "ascendc/host_api/tiling/template_argument.h"
20+#include "atvoss/broadcast/broadcast_base_struct.h"
21+ 
22+namespace RightShiftOp {
23+#define TPL_INT8 1
24+#define TPL_UINT8 2
25+#define TPL_INT16 3
26+#define TPL_UINT16 4
27+#define TPL_INT32 5
28+#define TPL_UINT32 6
29+#define TPL_INT64 7
30+#define TPL_UINT64 8
31+ 
32+#define TPL_SCH_MODE_0 0
33+#define TPL_SCH_MODE_1 1
34+ 
35+ASCENDC_TPL_ARGS_DECL(
36+ RightShift, BRC_TEMP_SCH_MODE_KEY_DECL(schMode),
37+ ASCENDC_TPL_DTYPE_DECL(
38+ dType, TPL_INT8, TPL_UINT8, TPL_INT16, TPL_UINT16, TPL_INT32, TPL_UINT32, TPL_INT64, TPL_UINT64));
39+ 
40+ASCENDC_TPL_SEL(
41+ ASCENDC_TPL_ARGS_SEL(BRC_TEMP_SCH_MODE_KEY_SEL(schMode), ASCENDC_TPL_DTYPE_SEL(dType, TPL_INT8)),
42+ ASCENDC_TPL_ARGS_SEL(BRC_TEMP_SCH_MODE_KEY_SEL(schMode), ASCENDC_TPL_DTYPE_SEL(dType, TPL_UINT8)),
43+ ASCENDC_TPL_ARGS_SEL(BRC_TEMP_SCH_MODE_KEY_SEL(schMode), ASCENDC_TPL_DTYPE_SEL(dType, TPL_INT16)),
44+ ASCENDC_TPL_ARGS_SEL(BRC_TEMP_SCH_MODE_KEY_SEL(schMode), ASCENDC_TPL_DTYPE_SEL(dType, TPL_UINT16)),
45+ ASCENDC_TPL_ARGS_SEL(BRC_TEMP_SCH_MODE_KEY_SEL(schMode), ASCENDC_TPL_DTYPE_SEL(dType, TPL_INT32)),
46+ ASCENDC_TPL_ARGS_SEL(BRC_TEMP_SCH_MODE_KEY_SEL(schMode), ASCENDC_TPL_DTYPE_SEL(dType, TPL_UINT32)),
47+ ASCENDC_TPL_ARGS_SEL(BRC_TEMP_SCH_MODE_KEY_SEL(schMode), ASCENDC_TPL_DTYPE_SEL(dType, TPL_INT64)),
48+ ASCENDC_TPL_ARGS_SEL(BRC_TEMP_SCH_MODE_KEY_SEL(schMode), ASCENDC_TPL_DTYPE_SEL(dType, TPL_UINT64)));
49+} // namespace RightShiftOp
50+ 
51+#endif // OPS_MATH_RIGHT_SHIFT_STRUCT_H
Amath/right_shift/op_kernel/right_shift_apt.cpp+61-0
@@ -0,0 +1,61 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+/*!
12+ * \file right_shift_apt.cpp
13+ * \brief right_shift_apt
14+ */
15+ 
16+#include "kernel_operator.h"
17+#include "arch35/right_shift_dag.h"
18+#include "arch35/right_shift_struct.h"
19+#include "atvoss/broadcast/broadcast_sch.h"
20+ 
21+using namespace AscendC;
22+using namespace RightShiftOp;
23+ 
24+template <uint64_t schMode, uint64_t dType>
25+__global__ __aicore__ void right_shift(GM_ADDR x, GM_ADDR y, GM_ADDR z, GM_ADDR workspace, GM_ADDR tiling)
26+{
27+ KERNEL_TASK_TYPE_DEFAULT(KERNEL_TYPE_AIV_ONLY);
28+ if constexpr (dType == TPL_INT8) {
29+ using OpDag = RightShiftDag8<int8_t>::OpDag;
30+ BroadcastSch<schMode, OpDag> sch(tiling);
31+ sch.Process(x, y, z);
32+ } else if constexpr (dType == TPL_UINT8) {
33+ using OpDag = RightShiftDag8<uint8_t>::OpDag;
34+ BroadcastSch<schMode, OpDag> sch(tiling);
35+ sch.Process(x, y, z);
36+ } else if constexpr (dType == TPL_INT16) {
37+ using OpDag = RightShiftDag16<int16_t>::OpDag;
38+ BroadcastSch<schMode, OpDag> sch(tiling);
39+ sch.Process(x, y, z);
40+ } else if constexpr (dType == TPL_UINT16) {
41+ using OpDag = RightShiftDag16<uint16_t>::OpDag;
42+ BroadcastSch<schMode, OpDag> sch(tiling);
43+ sch.Process(x, y, z);
44+ } else if constexpr (dType == TPL_INT32) {
45+ using OpDag = RightShiftDag32<int32_t>::OpDag;
46+ BroadcastSch<schMode, OpDag> sch(tiling);
47+ sch.Process(x, y, z);
48+ } else if constexpr (dType == TPL_UINT32) {
49+ using OpDag = RightShiftDag32<uint32_t>::OpDag;
50+ BroadcastSch<schMode, OpDag> sch(tiling);
51+ sch.Process(x, y, z);
52+ } else if constexpr (dType == TPL_INT64) {
53+ using OpDag = RightShiftDag64<int64_t>::OpDag;
54+ BroadcastSch<schMode, OpDag> sch(tiling);
55+ sch.Process(x, y, z);
56+ } else if constexpr (dType == TPL_UINT64) {
57+ using OpDag = RightShiftDag64<uint64_t>::OpDag;
58+ BroadcastSch<schMode, OpDag> sch(tiling);
59+ sch.Process(x, y, z);
60+ }
61+}
Rmath/right_shift/tests/ut/op_host/op_api/test_right_shift.cpp→math/right_shift/tests/ut/op_api/test_aclnn_right_shift_l0.cpp+48-45
@@ -1,5 +1,5 @@
1/**1/**
2- * Copyright (c) 2025 Huawei Technologies Co., Ltd.2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3 * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3 * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4 * CANN Open Software License Agreement Version 2.0 (the "License").4 * CANN Open Software License Agreement Version 2.0 (the "License").
5 * Please refer to the License for details. You may not use this file except in compliance with the License.5 * Please refer to the License for details. You may not use this file except in compliance with the License.
@@ -9,11 +9,8 @@
9 */9 */
10 10 
11#include <gtest/gtest.h>11#include <gtest/gtest.h>
12-#include <iostream>
13- 
14#include "opdev/make_op_executor.h"12#include "opdev/make_op_executor.h"
15-#include "level0/right_shift.h"13+#include "math/right_shift/op_api/right_shift.h"
16-#include "ut_stub.h"
17 14 
18using namespace op;15using namespace op;
19using namespace std;16using namespace std;
@@ -22,58 +19,64 @@ const int64_t DATA_SIZE = 24;
22 19 
23class RightShiftTest : public ::testing::Test {20class RightShiftTest : public ::testing::Test {
24public:21public:
25- RightShiftTest() : exe(nullptr) {22+ RightShiftTest() : exe(nullptr)
26- }23+ {}
27 24 
28- aclTensor* CreateAclTensor(std::vector<int64_t> shape, aclDataType dtype) {25+ aclTensor* CreateAclTensor(std::vector<int64_t> shape, aclDataType dtype)
29- return aclCreateTensor(shape.data(), shape.size(), dtype, nullptr, 0, ACL_FORMAT_ND, shape.data(), shape.size(),26+ {
30- data);27+ return aclCreateTensor(
31- }28+ shape.data(), shape.size(), dtype, nullptr, 0, ACL_FORMAT_ND, shape.data(), shape.size(), data);
29+ }
32 30 
33- void Clear() {31+ void Clear()
34- exe->kernelLaunchObjList_.clear();32+ {}
35- }
36 33 
37- void SetUp() override {34+ void SetUp() override
38- auto executor = &exe;35+ {
39- auto unique_executor = CREATE_EXECUTOR();36+ auto executor = &exe;
40- unique_executor.ReleaseTo(executor);37+ auto unique_executor = CREATE_EXECUTOR();
41- }38+ unique_executor.ReleaseTo(executor);
39+ }
42 40 
43- void TearDown() override {41+ void TearDown() override
44- delete exe;42+ {
45- }43+ delete exe;
44+ }
46 45 
47public:46public:
48- aclOpExecutor* exe;47+ aclOpExecutor* exe;
49- int64_t data[DATA_SIZE] = {1};48+ int64_t data[DATA_SIZE] = {1};
50};49};
51 50 
52-TEST_F(RightShiftTest, RightShiftTest_SUCC) {51+TEST_F(RightShiftTest, RightShiftTest_SUCC)
53- auto x = CreateAclTensor({16}, ACL_INT32);52+{
54- auto y = CreateAclTensor({1}, ACL_INT32);53+ auto x = CreateAclTensor({16}, ACL_INT32);
55- auto z = l0op::RightShift(x, y, exe);54+ auto y = CreateAclTensor({1}, ACL_INT32);
56- ASSERT_NE(y, nullptr);55+ auto z = l0op::RightShift(x, y, exe);
56+ ASSERT_NE(y, nullptr);
57}57}
58 58 
59-TEST_F(RightShiftTest, RightShiftTest_FAILED_1) {59+TEST_F(RightShiftTest, RightShiftTest_FAILED_1)
60- auto x = CreateAclTensor({16}, ACL_FLOAT);60+{
61- auto y = CreateAclTensor({1}, ACL_INT32);61+ auto x = CreateAclTensor({16}, ACL_FLOAT);
62- auto z = l0op::RightShift(x, y, exe);62+ auto y = CreateAclTensor({1}, ACL_INT32);
63- ASSERT_NE(y, nullptr);63+ auto z = l0op::RightShift(x, y, exe);
64+ ASSERT_NE(y, nullptr);
64}65}
65 66 
66-TEST_F(RightShiftTest, RightShiftTest_FAILED_2) {67+TEST_F(RightShiftTest, RightShiftTest_FAILED_2)
67- auto x = CreateAclTensor({16}, ACL_INT32);68+{
68- auto y = CreateAclTensor({1}, ACL_FLOAT);69+ auto x = CreateAclTensor({16}, ACL_INT32);
69- auto z = l0op::RightShift(x, y, exe);70+ auto y = CreateAclTensor({1}, ACL_FLOAT);
70- ASSERT_NE(y, nullptr);71+ auto z = l0op::RightShift(x, y, exe);
72+ ASSERT_NE(y, nullptr);
71}73}
72 74 
73-TEST_F(RightShiftTest, RightShiftTest_SUCC_950) {75+TEST_F(RightShiftTest, RightShiftTest_SUCC_950)
74- UtMock::GetInstance().SetSocVersion("ASCEND950PR_9589");76+{
75- auto x = CreateAclTensor({16}, ACL_INT32);77+ SetPlatformSocVersion(SocVersion::ASCEND950);
76- auto y = CreateAclTensor({1}, ACL_INT32);78+ auto x = CreateAclTensor({16}, ACL_INT32);
77- auto z = l0op::RightShift(x, y, exe);79+ auto y = CreateAclTensor({1}, ACL_INT32);
78- ASSERT_NE(y, nullptr);80+ auto z = l0op::RightShift(x, y, exe);
81+ ASSERT_NE(y, nullptr);
79}82}
Amath/right_shift/tests/ut/op_api/test_aclnn_right_shift_l2.cpp+212-0
@@ -0,0 +1,212 @@
1+/**
2+ * Copyright (c) 2025 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include "gtest/gtest.h"
12+#include "math/right_shift/op_api/aclnn_right_shift.h"
13+#include "op_api_ut_common/tensor_desc.h"
14+#include "op_api_ut_common/op_api_ut.h"
15+ 
16+using namespace std;
17+using namespace op;
18+ 
19+class l2_right_shift_test : public testing::Test {
20+protected:
21+ static void SetUpTestCase()
22+ {
23+ cout << "tensor_right_shift_test SetUp" << endl;
24+ }
25+ 
26+ static void TearDownTestCase()
27+ {
28+ cout << "tensor_right_shift_test TearDown" << endl;
29+ }
30+};
31+ 
32+// 正常调用流程
33+TEST_F(l2_right_shift_test, normal_success)
34+{
35+ auto inputDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40});
36+ auto shiftBitsDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1, 2, 3, 4});
37+ auto outDesc = TensorDesc(inputDesc);
38+ uint64_t workspace_size = 0;
39+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
40+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
41+ EXPECT_EQ(aclRet, ACL_SUCCESS);
42+}
43+ 
44+// 输入input为空指针
45+TEST_F(l2_right_shift_test, nullptr_input)
46+{
47+ auto inputDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40});
48+ auto shiftBitsDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1, 2, 3, 4});
49+ auto outDesc = TensorDesc(inputDesc);
50+ uint64_t workspace_size = 0;
51+ auto ut = OP_API_UT(aclnnRightShift, INPUT(nullptr, shiftBitsDesc), OUTPUT(outDesc));
52+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
53+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_NULLPTR);
54+}
55+ 
56+// 输入shiftBits为空指针
57+TEST_F(l2_right_shift_test, nullptr_shiftBits)
58+{
59+ auto inputDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40});
60+ auto shiftBitsDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1, 2, 3, 4});
61+ auto outDesc = TensorDesc(inputDesc);
62+ uint64_t workspace_size = 0;
63+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, nullptr), OUTPUT(outDesc));
64+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
65+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_NULLPTR);
66+}
67+ 
68+// 输入out为空指针
69+TEST_F(l2_right_shift_test, nullptr_out)
70+{
71+ auto inputDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40});
72+ auto shiftBitsDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1, 2, 3, 4});
73+ auto outDesc = TensorDesc(inputDesc);
74+ uint64_t workspace_size = 0;
75+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(nullptr));
76+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
77+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_NULLPTR);
78+}
79+ 
80+// 输入类型不支持
81+TEST_F(l2_right_shift_test, type_unsupport)
82+{
83+ auto inputDesc = TensorDesc({1, 4}, ACL_FLOAT, ACL_FORMAT_ND).Value(vector<float>{10, 20, 30, 40});
84+ auto shiftBitsDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1, 2, 3, 4});
85+ auto outDesc = TensorDesc(inputDesc);
86+ uint64_t workspace_size = 0;
87+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
88+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
89+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
90+}
91+ 
92+// 输入类型可以正常类型推导
93+TEST_F(l2_right_shift_test, type_promote_normal)
94+{
95+ auto inputDesc = TensorDesc({1, 4}, ACL_INT64, ACL_FORMAT_ND).Value(vector<int64_t>{10, 20, 30, 40});
96+ auto shiftBitsDesc = TensorDesc({1, 4}, ACL_INT32, ACL_FORMAT_ND).Value(vector<int32_t>{1, 2, 3, 4});
97+ auto outDesc = TensorDesc(inputDesc);
98+ uint64_t workspace_size = 0;
99+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
100+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
101+ EXPECT_EQ(aclRet, ACL_SUCCESS);
102+}
103+ 
104+// 输入类型不支持类型推导
105+TEST_F(l2_right_shift_test, type_promote_unsupport)
106+{
107+ auto inputDesc = TensorDesc({1, 4}, ACL_UINT16, ACL_FORMAT_ND).Value(vector<uint16_t>{10, 20, 30, 40});
108+ auto shiftBitsDesc = TensorDesc({1, 4}, ACL_UINT32, ACL_FORMAT_ND).Value(vector<uint32_t>{1, 2, 3, 4});
109+ auto outDesc = TensorDesc(inputDesc);
110+ uint64_t workspace_size = 0;
111+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
112+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
113+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
114+}
115+ 
116+// 输入类型推导结果无法转换为out类型-不存在，整型可以向任何类型转换，结果应该为true
117+TEST_F(l2_right_shift_test, type_promote_can_not_cast_to_out)
118+{
119+ auto inputDesc = TensorDesc({1, 4}, ACL_INT32, ACL_FORMAT_ND).Value(vector<int32_t>{10, 20, 30, 40});
120+ auto shiftBitsDesc = TensorDesc({1, 4}, ACL_INT64, ACL_FORMAT_ND).Value(vector<int64_t>{1, 2, 3, 4});
121+ auto outDesc = TensorDesc({1, 4}, ACL_UINT16, ACL_FORMAT_ND).Value(vector<int64_t>{0, 0, 0, 0});
122+ uint64_t workspace_size = 0;
123+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
124+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
125+ EXPECT_EQ(aclRet, ACL_SUCCESS);
126+}
127+ 
128+// 输入shape为空
129+TEST_F(l2_right_shift_test, shape_empty_input)
130+{
131+ auto inputDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40});
132+ auto shiftBitsDesc = TensorDesc({}, ACL_INT8, ACL_FORMAT_ND);
133+ auto outDesc = TensorDesc(inputDesc);
134+ uint64_t workspace_size = 0;
135+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
136+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
137+ EXPECT_EQ(aclRet, ACL_SUCCESS);
138+}
139+ 
140+// 输入input shape维度超过8
141+TEST_F(l2_right_shift_test, input_shape_dim_over_8)
142+{
143+ auto inputDesc =
144+ TensorDesc({1, 1, 1, 1, 1, 1, 1, 1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40});
145+ auto shiftBitsDesc = TensorDesc({1}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1});
146+ auto outDesc = TensorDesc(inputDesc);
147+ uint64_t workspace_size = 0;
148+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
149+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
150+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
151+}
152+ 
153+// 输入shiftbits shape维度超过8
154+TEST_F(l2_right_shift_test, shiftBits_shape_dim_over_8)
155+{
156+ auto inputDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40});
157+ auto shiftBitsDesc =
158+ TensorDesc({1, 1, 1, 1, 1, 1, 1, 1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1, 1, 1, 1});
159+ auto outDesc = TensorDesc(inputDesc);
160+ uint64_t workspace_size = 0;
161+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
162+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
163+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
164+}
165+ 
166+// 输入shape与输出shape不一致
167+TEST_F(l2_right_shift_test, shape_input_noeq_out)
168+{
169+ auto inputDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40});
170+ auto shiftBitsDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1, 1, 1, 1});
171+ auto outDesc = TensorDesc({4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40});
172+ uint64_t workspace_size = 0;
173+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
174+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
175+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
176+}
177+ 
178+// 输入shape支持广播
179+TEST_F(l2_right_shift_test, shape_broadcast_normal)
180+{
181+ auto inputDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40});
182+ auto shiftBitsDesc = TensorDesc({1}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1});
183+ auto outDesc = TensorDesc(inputDesc);
184+ uint64_t workspace_size = 0;
185+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
186+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
187+ EXPECT_EQ(aclRet, ACL_SUCCESS);
188+}
189+ 
190+// 输入shape不支持广播-维度数一致，大小不一致，且没有等于1的维度
191+TEST_F(l2_right_shift_test, shape_broadcast_unsupport)
192+{
193+ auto inputDesc = TensorDesc({2, 3}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10, 20, 30, 40, 50, 60});
194+ auto shiftBitsDesc = TensorDesc({3, 2}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1, 2, 3, 4, 1, 2});
195+ auto outDesc = TensorDesc(inputDesc);
196+ uint64_t workspace_size = 0;
197+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
198+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
199+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
200+}
201+ 
202+// 输入shape广播后结果与out的shape不一致
203+TEST_F(l2_right_shift_test, shape_broadcast_not_equal_with_out)
204+{
205+ auto inputDesc = TensorDesc({1}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{10});
206+ auto shiftBitsDesc = TensorDesc({1, 4}, ACL_INT8, ACL_FORMAT_ND).Value(vector<int8_t>{1, 1, 1, 1});
207+ auto outDesc = TensorDesc(inputDesc);
208+ uint64_t workspace_size = 0;
209+ auto ut = OP_API_UT(aclnnRightShift, INPUT(inputDesc, shiftBitsDesc), OUTPUT(outDesc));
210+ aclnnStatus aclRet = ut.TestGetWorkspaceSize(&workspace_size);
211+ EXPECT_EQ(aclRet, ACLNN_ERR_PARAM_INVALID);
212+}
Amath/right_shift/tests/ut/op_host/arch35/test_right_shift_tiling.cpp+238-0
@@ -0,0 +1,238 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <gtest/gtest.h>
12+#include <iostream>
13+#include "math/right_shift/op_host/arch35/right_shift_tiling_arch35.h"
14+#include "tiling_context_faker.h"
15+#include "tiling_case_executor.h"
16+#include "atvoss/broadcast/broadcast_tiling.h"
17+ 
18+using namespace std;
19+ 
20+class RightShiftTilingTest : public testing::Test {
21+protected:
22+ static void SetUpTestCase()
23+ {
24+ std::cout << "RightShiftTilingTest SetUp" << std::endl;
25+ }
26+ 
27+ static void TearDownTestCase()
28+ {
29+ std::cout << "RightShiftTilingTest TearDown" << std::endl;
30+ }
31+};
32+ 
33+TEST_F(RightShiftTilingTest, right_shift_test_tiling_int8)
34+{
35+ Ops::Base::BroadcastCompileInfo compileInfo;
36+ compileInfo.coreNum = 64;
37+ compileInfo.ubSize = 253952;
38+ 
39+ gert::TilingContextPara tilingContextPara(
40+ "RightShift",
41+ {
42+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT8, ge::FORMAT_ND},
43+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT8, ge::FORMAT_ND},
44+ },
45+ {
46+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT8, ge::FORMAT_ND},
47+ },
48+ {}, &compileInfo);
49+ 
50+ uint64_t expectTilingKey = 65544;
51+ string expectTilingData = "8192 34359738880 ";
52+ std::vector<size_t> expectWorkspaces = {16777216};
53+ ExecuteTestCase(tilingContextPara, ge::GRAPH_SUCCESS, expectTilingKey, expectTilingData, expectWorkspaces);
54+}
55+ 
56+TEST_F(RightShiftTilingTest, right_shift_test_tiling_uint8)
57+{
58+ Ops::Base::BroadcastCompileInfo compileInfo;
59+ compileInfo.coreNum = 64;
60+ compileInfo.ubSize = 253952;
61+ 
62+ gert::TilingContextPara tilingContextPara(
63+ "RightShift",
64+ {
65+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT8, ge::FORMAT_ND},
66+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT8, ge::FORMAT_ND},
67+ },
68+ {
69+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT8, ge::FORMAT_ND},
70+ },
71+ {}, &compileInfo);
72+ 
73+ uint64_t expectTilingKey = 131080;
74+ string expectTilingData = "8192 34359738880 ";
75+ std::vector<size_t> expectWorkspaces = {16777216};
76+ ExecuteTestCase(tilingContextPara, ge::GRAPH_SUCCESS, expectTilingKey, expectTilingData, expectWorkspaces);
77+}
78+ 
79+TEST_F(RightShiftTilingTest, right_shift_test_tiling_int16)
80+{
81+ Ops::Base::BroadcastCompileInfo compileInfo;
82+ compileInfo.coreNum = 64;
83+ compileInfo.ubSize = 253952;
84+ 
85+ gert::TilingContextPara tilingContextPara(
86+ "RightShift",
87+ {
88+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT16, ge::FORMAT_ND},
89+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT16, ge::FORMAT_ND},
90+ },
91+ {
92+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT16, ge::FORMAT_ND},
93+ },
94+ {}, &compileInfo);
95+ 
96+ uint64_t expectTilingKey = 196616;
97+ string expectTilingData = "8192 68719476992 ";
98+ std::vector<size_t> expectWorkspaces = {16777216};
99+ ExecuteTestCase(tilingContextPara, ge::GRAPH_SUCCESS, expectTilingKey, expectTilingData, expectWorkspaces);
100+}
101+ 
102+TEST_F(RightShiftTilingTest, right_shift_test_tiling_uint16)
103+{
104+ Ops::Base::BroadcastCompileInfo compileInfo;
105+ compileInfo.coreNum = 64;
106+ compileInfo.ubSize = 253952;
107+ 
108+ gert::TilingContextPara tilingContextPara(
109+ "RightShift",
110+ {
111+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT16, ge::FORMAT_ND},
112+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT16, ge::FORMAT_ND},
113+ },
114+ {
115+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT16, ge::FORMAT_ND},
116+ },
117+ {}, &compileInfo);
118+ 
119+ uint64_t expectTilingKey = 262152;
120+ string expectTilingData = "8192 68719476992 ";
121+ std::vector<size_t> expectWorkspaces = {16777216};
122+ ExecuteTestCase(tilingContextPara, ge::GRAPH_SUCCESS, expectTilingKey, expectTilingData, expectWorkspaces);
123+}
124+ 
125+TEST_F(RightShiftTilingTest, right_shift_test_tiling_int32)
126+{
127+ Ops::Base::BroadcastCompileInfo compileInfo;
128+ compileInfo.coreNum = 64;
129+ compileInfo.ubSize = 253952;
130+ 
131+ gert::TilingContextPara tilingContextPara(
132+ "RightShift",
133+ {
134+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT32, ge::FORMAT_ND},
135+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT32, ge::FORMAT_ND},
136+ },
137+ {
138+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT32, ge::FORMAT_ND},
139+ },
140+ {}, &compileInfo);
141+ 
142+ uint64_t expectTilingKey = 327688;
143+ string expectTilingData = "8192 137438953600 ";
144+ std::vector<size_t> expectWorkspaces = {16777216};
145+ ExecuteTestCase(tilingContextPara, ge::GRAPH_SUCCESS, expectTilingKey, expectTilingData, expectWorkspaces);
146+}
147+ 
148+TEST_F(RightShiftTilingTest, right_shift_test_tiling_uint32)
149+{
150+ Ops::Base::BroadcastCompileInfo compileInfo;
151+ compileInfo.coreNum = 64;
152+ compileInfo.ubSize = 253952;
153+ 
154+ gert::TilingContextPara tilingContextPara(
155+ "RightShift",
156+ {
157+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT32, ge::FORMAT_ND},
158+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT32, ge::FORMAT_ND},
159+ },
160+ {
161+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT32, ge::FORMAT_ND},
162+ },
163+ {}, &compileInfo);
164+ 
165+ uint64_t expectTilingKey = 393224;
166+ string expectTilingData = "8192 137438953600 ";
167+ std::vector<size_t> expectWorkspaces = {16777216};
168+ ExecuteTestCase(tilingContextPara, ge::GRAPH_SUCCESS, expectTilingKey, expectTilingData, expectWorkspaces);
169+}
170+ 
171+TEST_F(RightShiftTilingTest, right_shift_test_tiling_int64)
172+{
173+ Ops::Base::BroadcastCompileInfo compileInfo;
174+ compileInfo.coreNum = 64;
175+ compileInfo.ubSize = 253952;
176+ 
177+ gert::TilingContextPara tilingContextPara(
178+ "RightShift",
179+ {
180+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT64, ge::FORMAT_ND},
181+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT64, ge::FORMAT_ND},
182+ },
183+ {
184+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT64, ge::FORMAT_ND},
185+ },
186+ {}, &compileInfo);
187+ 
188+ uint64_t expectTilingKey = 458760;
189+ string expectTilingData = "8192 137438953600 ";
190+ std::vector<size_t> expectWorkspaces = {16777216};
191+ ExecuteTestCase(tilingContextPara, ge::GRAPH_SUCCESS, expectTilingKey, expectTilingData, expectWorkspaces);
192+}
193+ 
194+TEST_F(RightShiftTilingTest, right_shift_test_tiling_uint64)
195+{
196+ Ops::Base::BroadcastCompileInfo compileInfo;
197+ compileInfo.coreNum = 64;
198+ compileInfo.ubSize = 253952;
199+ 
200+ gert::TilingContextPara tilingContextPara(
201+ "RightShift",
202+ {
203+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT64, ge::FORMAT_ND},
204+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT64, ge::FORMAT_ND},
205+ },
206+ {
207+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_UINT64, ge::FORMAT_ND},
208+ },
209+ {}, &compileInfo);
210+ 
211+ uint64_t expectTilingKey = 524296;
212+ string expectTilingData = "8192 137438953600 ";
213+ std::vector<size_t> expectWorkspaces = {16777216};
214+ ExecuteTestCase(tilingContextPara, ge::GRAPH_SUCCESS, expectTilingKey, expectTilingData, expectWorkspaces);
215+}
216+ 
217+TEST_F(RightShiftTilingTest, right_shift_test_tiling_invalid_dtype)
218+{
219+ Ops::Base::BroadcastCompileInfo compileInfo;
220+ compileInfo.coreNum = 64;
221+ compileInfo.ubSize = 253952;
222+ 
223+ gert::TilingContextPara tilingContextPara(
224+ "RightShift",
225+ {
226+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_FLOAT16, ge::FORMAT_ND},
227+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT8, ge::FORMAT_ND},
228+ },
229+ {
230+ {{{1, 64, 2, 64}, {1, 64, 2, 64}}, ge::DT_INT8, ge::FORMAT_ND},
231+ },
232+ {}, &compileInfo);
233+ 
234+ uint64_t expectTilingKey = 65536;
235+ string expectTilingData = "";
236+ std::vector<size_t> expectWorkspaces = {16777216};
237+ ExecuteTestCase(tilingContextPara, ge::GRAPH_FAILED, expectTilingKey, expectTilingData, expectWorkspaces);
238+}
Amath/right_shift/tests/ut/op_host/test_right_shift_infershape.cpp+59-0
@@ -0,0 +1,59 @@
1+/**
2+ * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+ * CANN Open Software License Agreement Version 2.0 (the "License").
5+ * Please refer to the License for details. You may not use this file except in compliance with the License.
6+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+ * See LICENSE in the root of the software repository for the full text of the License.
9+ */
10+ 
11+#include <gtest/gtest.h>
12+#include <iostream>
13+#include "infershape_context_faker.h"
14+#include "infershape_case_executor.h"
15+ 
16+using namespace ge;
17+ 
18+class RightShiftInferShape : public testing::Test {
19+protected:
20+ static void SetUpTestCase()
21+ {
22+ std::cout << "RightShift InferShape SetUp" << std::endl;
23+ }
24+ 
25+ static void TearDownTestCase()
26+ {
27+ std::cout << "RightShift InferShape TearDown" << std::endl;
28+ }
29+};
30+ 
31+TEST_F(RightShiftInferShape, right_shift_infershape_test)
32+{
33+ gert::InfershapeContextPara infershapeContextPara(
34+ "RightShift",
35+ {
36+ {{{4, 3, 4}, {4, 3, 4}}, ge::DT_INT8, ge::FORMAT_ND},
37+ {{{4, 3, 4}, {4, 3, 4}}, ge::DT_INT8, ge::FORMAT_ND},
38+ },
39+ {
40+ {{{}, {}}, ge::DT_INT8, ge::FORMAT_ND},
41+ });
42+ std::vector<std::vector<int64_t>> expectOutputShape = {{4, 3, 4}};
43+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
44+}
45+ 
46+TEST_F(RightShiftInferShape, right_shift_infershape_dynamic_test)
47+{
48+ gert::InfershapeContextPara infershapeContextPara(
49+ "RightShift",
50+ {
51+ {{{-1}, {-1}}, ge::DT_INT8, ge::FORMAT_ND},
52+ {{{-1}, {-1}}, ge::DT_INT8, ge::FORMAT_ND},
53+ },
54+ {
55+ {{{}, {}}, ge::DT_INT8, ge::FORMAT_ND},
56+ });
57+ std::vector<std::vector<int64_t>> expectOutputShape = {{-1}};
58+ ExecuteTestCase(infershapeContextPara, ge::GRAPH_SUCCESS, expectOutputShape);
59+}
Mscripts/ci/ascend950/ops_math_operator_list.yaml+5-0
@@ -24,6 +24,7 @@ operator_group_1:
24 - arg_min_with_value24 - arg_min_with_value
25 - assign_add25 - assign_add
26 - assign_sub26 - assign_sub
27+ - atan
27 - axpy28 - axpy
28 - axpy_v229 - axpy_v2
29 - bias_add30 - bias_add
@@ -48,8 +49,10 @@ operator_group_1:
48 - dynamic_partition49 - dynamic_partition
49 - dynamic_stitch50 - dynamic_stitch
50 - equal51 - equal
52+ - erf
51 - exp53 - exp
52 - expand54 - expand
55+ - expm1
53 - exp_segsum_grad56 - exp_segsum_grad
54 - eye57 - eye
55 - floor58 - floor
@@ -89,6 +92,7 @@ operator_group_2:
89 - mul92 - mul
90 - muls93 - muls
91 - mul_addn94 - mul_addn
95+ - nan_to_num
92 - neg96 - neg
93 - non_finite_check97 - non_finite_check
94 - not_equal98 - not_equal
@@ -113,6 +117,7 @@ operator_group_3:
113 - reduce_log_sum_exp117 - reduce_log_sum_exp
114operator_group_4:118operator_group_4:
115 - rfft1_d119 - rfft1_d
120+ - right_shift
116 - round121 - round
117 - rsqrt122 - rsqrt
118 - rsqrt_grad123 - rsqrt_grad