Star1330
1447
代码
代码
Issues118
Pull Requests254
Actions
项目讨论
Wiki
分析
Star1330
1447
已合并
Square算子AscendC实现贡献 #291
liwen创建于 2025年11月24日
Square算子AscendC实现贡献 #291
已合并
liwen创建于 2025年11月24日
从liwenkkklll/ops-math-squaremaster合入到cann/ops-mathmaster
共 18 个文件变更+1308-0
Aexperimental/math/square_v2/CMakeLists.txt+12-0
@@ -0,0 +1,12 @@
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_all_modules_sources(OPTYPE square_v2 ACLNNTYPE aclnn)
Aexperimental/math/square_v2/examples/test_aclnn_square_v2.cpp+170-0
@@ -0,0 +1,170 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+ 
21+ #include <iostream>
22+ #include <vector>
23+ #include "acl/acl.h"
24+ #include "aclnn_square_v2.h"
25+
26+ #define CHECK_RET(cond, return_expr) \
27+ do { \
28+ if (!(cond)) { \
29+ return_expr; \
30+ } \
31+ } while (0)
32+
33+ #define LOG_PRINT(message, ...) \
34+ do { \
35+ printf(message, ##__VA_ARGS__); \
36+ } while (0)
37+
38+ int64_t GetShapeSize(const std::vector<int64_t>& shape)
39+ {
40+ int64_t shapeSize = 1;
41+ for (auto i : shape) {
42+ shapeSize *= i;
43+ }
44+ return shapeSize;
45+ }
46+
47+ void PrintOutResult(std::vector<int64_t>& shape, void** deviceAddr)
48+ {
49+ auto size = GetShapeSize(shape);
50+ std::vector<float> resultData(size, 0);
51+ auto ret = aclrtMemcpy(
52+ resultData.data(), resultData.size() * sizeof(resultData[0]), *deviceAddr, size * sizeof(resultData[0]),
53+ ACL_MEMCPY_DEVICE_TO_HOST);
54+ CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return);
55+ for (int64_t i = 0; i < size; i++) {
56+ LOG_PRINT("mean result[%ld] is: %f\n", i, resultData[i]);
57+ }
58+ }
59+
60+ int Init(int32_t deviceId, aclrtStream* stream)
61+ {
62+ // 固定写法，初始化
63+ auto ret = aclInit(nullptr);
64+ CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclInit failed. ERROR: %d\n", ret); return ret);
65+ ret = aclrtSetDevice(deviceId);
66+ CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret); return ret);
67+ ret = aclrtCreateStream(stream);
68+ CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateStream failed. ERROR: %d\n", ret); return ret);
69+ return 0;
70+ }
71+
72+ template <typename T>
73+ int CreateAclTensor(
74+ const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr, aclDataType dataType,
75+ aclTensor** tensor)
76+ {
77+ auto size = GetShapeSize(shape) * sizeof(T);
78+ // 2. 申请device侧内存
79+ auto ret = aclrtMalloc(deviceAddr, size, ACL_MEM_MALLOC_HUGE_FIRST);
80+ CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMalloc failed. ERROR: %d\n", ret); return ret);
81+ // 3. 调用aclrtMemcpy将host侧数据拷贝到device侧内存上
82+ ret = aclrtMemcpy(*deviceAddr, size, hostData.data(), size, ACL_MEMCPY_HOST_TO_DEVICE);
83+ CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMemcpy failed. ERROR: %d\n", ret); return ret);
84+
85+ // 计算连续tensor的strides
86+ std::vector<int64_t> strides(shape.size(), 1);
87+ for (int64_t i = shape.size() - 2; i >= 0; i--) {
88+ strides[i] = shape[i + 1] * strides[i + 1];
89+ }
90+
91+ // 调用aclCreateTensor接口创建aclTensor
92+ *tensor = aclCreateTensor(
93+ shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(),
94+ *deviceAddr);
95+ return 0;
96+ }
97+
98+ int main()
99+ {
100+ // 1. 调用acl进行device/stream初始化
101+ int32_t deviceId = 0;
102+ aclrtStream stream;
103+ auto ret = Init(deviceId, &stream);
104+ CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);
105+
106+ // 2. 构造输入与输出，需要根据API的接口自定义构造
107+ aclTensor* selfX = nullptr;
108+ void* selfXDeviceAddr = nullptr;
109+ std::vector<int64_t> selfXShape = {9};
110+ std::vector<float> selfXHostData(9, 2.5f);
111+ ret = CreateAclTensor(selfXHostData, selfXShape, &selfXDeviceAddr, aclDataType::ACL_FLOAT, &selfX);
112+ CHECK_RET(ret == ACL_SUCCESS, return ret);
113+
114+ aclTensor* out = nullptr;
115+ void* outDeviceAddr = nullptr;
116+ std::vector<int64_t> outShape = {9};
117+ std::vector<float> outHostData(9, 0);
118+ ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out);
119+ CHECK_RET(ret == ACL_SUCCESS, return ret);
120+
121+ // 3. 调用CANN算子库API，需要修改为具体的Api名称
122+ uint64_t workspaceSize = 0;
123+ aclOpExecutor* executor;
124+
125+ // 4. 调用aclnnSquareV2第一段接口
126+ ret = aclnnSquareV2GetWorkspaceSize(selfX, out, &workspaceSize, &executor);
127+ // CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnSquareV2GetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
128+ if (ret != ACL_SUCCESS) {
129+ const char* errMsg = aclGetRecentErrMsg(); // ← 关键一行
130+ LOG_PRINT("[ERROR] aclnnSquareV2GetWorkspaceSize failed, ret = %d\n", ret);
131+ LOG_PRINT(" recent AscendCL msg: %s\n",
132+ errMsg ? errMsg : "no additional message");
133+ return ret;
134+ }
135+ // 根据第一段接口计算出的workspaceSize申请device内存
136+ void* workspaceAddr = nullptr;
137+ if (workspaceSize > static_cast<uint64_t>(0)) {
138+ ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
139+ CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
140+ }
141+
142+ // 5. 调用aclnnSquareV2第二段接口
143+ ret = aclnnSquareV2(workspaceAddr, workspaceSize, executor, stream);
144+ CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnSquareV2 failed. ERROR: %d\n", ret); return ret);
145+
146+ // 6. （固定写法）同步等待任务执行结束
147+ ret = aclrtSynchronizeStream(stream);
148+ CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);
149+
150+ // 5. 获取输出的值，将device侧内存上的结果拷贝至host侧，需要根据具体API的接口定义修改
151+ PrintOutResult(outShape, &outDeviceAddr);
152+
153+ // 7. 释放aclTensor，需要根据具体API的接口定义修改
154+ aclDestroyTensor(selfX);
155+ aclDestroyTensor(out);
156+
157+ // 8. 释放device资源
158+ aclrtFree(selfXDeviceAddr);
159+ aclrtFree(outDeviceAddr);
160+ if (workspaceSize > static_cast<uint64_t>(0)) {
161+ aclrtFree(workspaceAddr);
162+ }
163+ aclrtDestroyStream(stream);
164+ aclrtResetDevice(deviceId);
165+
166+ // 9. acl去初始化
167+ aclFinalize();
168+
169+ return 0;
170+ }
Aexperimental/math/square_v2/examples/test_geir_square_v2.cpp+325-0
@@ -0,0 +1,325 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+#include <iostream>
21+#include <fstream>
22+#include <string.h>
23+#include <stdint.h>
24+#include <vector>
25+#include <string>
26+#include <map>
27+#include "assert.h"
28+ 
29+#include "graph.h"
30+#include "types.h"
31+#include "tensor.h"
32+#include "ge_error_codes.h"
33+#include "ge_api_types.h"
34+#include "ge_api.h"
35+#include "array_ops.h"
36+#include "ge_ir_build.h"
37+ 
38+#include "experiment_ops.h"
39+#include "nn_other.h"
40+#include "../op_graph/square_v2_proto.h"
41+ 
42+#define FAILED -1
43+#define SUCCESS 0
44+ 
45+using namespace ge;
46+using std::map;
47+using std::string;
48+using std::vector;
49+#define ADD_INPUT(intputIndex, intputName, intputDtype, inputShape,value) \
50+ vector<int64_t> placeholder##intputIndex##_shape = inputShape; \
51+ auto placeholder##intputIndex = op::Data("placeholder" + intputIndex).set_attr_index(0); \
52+ TensorDesc placeholder##intputIndex##_desc = \
53+ TensorDesc(ge::Shape(placeholder##intputIndex##_shape), FORMAT_ND, intputDtype); \
54+ placeholder##intputIndex##_desc.SetPlacement(ge::kPlacementHost); \
55+ placeholder##intputIndex##_desc.SetFormat(FORMAT_ND); \
56+ Tensor tensor_placeholder##intputIndex; \
57+ if (intputDtype == DT_FLOAT || intputDtype == DT_FLOAT16) { \
58+ ret = GenOnesDataFloat32(placeholder##intputIndex##_shape, tensor_placeholder##intputIndex, \
59+ placeholder##intputIndex##_desc, static_cast<float>(value)); \
60+ } else if(intputDtype == DT_INT32 || intputDtype == DT_INT16){ \
61+ ret = GenOnesData(placeholder##intputIndex##_shape, tensor_placeholder##intputIndex, \
62+ placeholder##intputIndex##_desc, intputDtype, static_cast<int>(value)); \
63+ } \
64+ if (ret != SUCCESS) { \
65+ printf("%s - ERROR - [XIR]: Generate input data failed\n", GetTime().c_str()); \
66+ return FAILED; \
67+ } \
68+ placeholder##intputIndex.update_input_desc_x(placeholder##intputIndex##_desc); \
69+ input.push_back(tensor_placeholder##intputIndex); \
70+ graph.AddOp(placeholder##intputIndex); \
71+ add1.set_input_##intputName(placeholder##intputIndex); \
72+ inputs.push_back(placeholder##intputIndex);
73+ 
74+#define ADD_CONST_INPUT(intputIndex, intputName, intputDtype, inputShape,value) \
75+ vector<int64_t> placeholder##intputIndex##_shape = inputShape; \
76+ auto placeholder##intputIndex = op::Const("placeholder" + intputIndex); \
77+ TensorDesc placeholder##intputIndex##_desc = \
78+ TensorDesc(ge::Shape(placeholder##intputIndex##_shape), FORMAT_ND, intputDtype); \
79+ placeholder##intputIndex##_desc.SetPlacement(ge::kPlacementHost); \
80+ placeholder##intputIndex##_desc.SetFormat(FORMAT_ND); \
81+ Tensor tensor_placeholder##intputIndex; \
82+ if (intputDtype == DT_FLOAT || intputDtype == DT_FLOAT16) { \
83+ ret = GenOnesDataFloat32(placeholder##intputIndex##_shape, tensor_placeholder##intputIndex, \
84+ placeholder##intputIndex##_desc, static_cast<float>(value)); \
85+ } else if(intputDtype == DT_INT32 || intputDtype == DT_INT16){ \
86+ ret = GenOnesData(placeholder##intputIndex##_shape, tensor_placeholder##intputIndex, \
87+ placeholder##intputIndex##_desc, intputDtype, static_cast<int>(value)); \
88+ } \
89+ if (ret != SUCCESS) { \
90+ printf("%s - ERROR - [XIR]: Generate input data failed\n", GetTime().c_str()); \
91+ return FAILED; \
92+ } \
93+ placeholder##intputIndex.SetAttr("value", tensor_placeholder##intputIndex); \
94+ placeholder##intputIndex.update_output_desc_y(placeholder##intputIndex##_desc); \
95+ graph.AddOp(placeholder##intputIndex); \
96+ add1.set_input_##intputName(placeholder##intputIndex); \
97+ add1.update_input_desc_##intputName(placeholder##intputIndex##_desc); \
98+ inputs.push_back(placeholder##intputIndex);
99+ 
100+#define ADD_OUTPUT(outputIndex, outputName, outputDtype, outputShape) \
101+ TensorDesc outputName##outputIndex##_desc = TensorDesc(ge::Shape(outputShape), FORMAT_ND, outputDtype); \
102+ add1.update_output_desc_##outputName(outputName##outputIndex##_desc);
103+ //新加
104+ #define LOG_PRINT(message, ...) \
105+ do { \
106+ printf(message, ##__VA_ARGS__); \
107+ } while (0)
108+ 
109+string GetTime()
110+{
111+ time_t timep;
112+ time(&timep);
113+ char tmp[64];
114+ strftime(tmp, sizeof(tmp), "%Y-%m-%d %H:%M:%S,000", localtime(&timep));
115+ return tmp;
116+}
117+ 
118+uint32_t GetDataTypeSize(DataType dt)
119+{
120+ uint32_t dilation = 1;
121+ uint32_t oneByte = 1;
122+ uint32_t twoByte = 2;
123+ uint32_t fourByte = 4;
124+ uint32_t eightByte = 8;
125+ 
126+ if (dt == ge::DT_FLOAT) {
127+ dilation = fourByte;
128+ } else if (dt == ge::DT_FLOAT16) {
129+ dilation = twoByte;
130+ } else if (dt == ge::DT_BF16) {
131+ dilation = twoByte;
132+ } else if (dt == ge::DT_INT16) {
133+ dilation = twoByte;
134+ } else if (dt == ge::DT_UINT16) {
135+ dilation = twoByte;
136+ } else if (dt == ge::DT_INT32) {
137+ dilation = fourByte;
138+ } else if (dt == ge::DT_UINT32) {
139+ dilation = fourByte;
140+ } else if (dt == ge::DT_INT64) {
141+ dilation = eightByte;
142+ } else if (dt == ge::DT_UINT64) {
143+ dilation = eightByte;
144+ } else if (dt == ge::DT_INT8) {
145+ dilation = oneByte;
146+ }
147+ return dilation;
148+}
149+ 
150+int32_t GenOnesDataFloat32(vector<int64_t> shapes, Tensor& input_tensor, TensorDesc& input_tensor_desc, float value)
151+{
152+ input_tensor_desc.SetRealDimCnt(shapes.size());
153+ size_t size = 1;
154+ for (uint32_t i = 0; i < shapes.size(); i++) {
155+ size *= shapes[i];
156+ }
157+ uint32_t byteSizeFloat32 = 4;
158+ uint32_t data_len = size * byteSizeFloat32;
159+ float* pData = new (std::nothrow) float[size];
160+ 
161+ for (size_t i = 0; i < size; ++i) {
162+ *(pData + i) = value;
163+ }
164+ input_tensor = Tensor(input_tensor_desc, (uint8_t*)pData, data_len);
165+ return SUCCESS;
166+}
167+ 
168+int32_t GenOnesData(
169+ vector<int64_t> shapes, Tensor& input_tensor, TensorDesc& input_tensor_desc, DataType data_type, int value)
170+{
171+ input_tensor_desc.SetRealDimCnt(shapes.size());
172+ size_t size = 1;
173+ for (uint32_t i = 0; i < shapes.size(); i++) {
174+ size *= shapes[i];
175+ }
176+ uint32_t data_len = size * GetDataTypeSize(data_type);
177+ int32_t* pData = new (std::nothrow) int32_t[size];
178+ for (uint32_t i = 0; i < size; ++i) {
179+ *(pData + i) = value;
180+ }
181+ input_tensor = Tensor(input_tensor_desc, reinterpret_cast<uint8_t*>(pData), data_len);
182+ return SUCCESS;
183+}
184+ 
185+int32_t WriteDataToFile(string bin_file, uint64_t data_size, uint8_t* inputData)
186+{
187+ FILE* fp;
188+ fp = fopen(bin_file.c_str(), "w");
189+ fwrite(inputData, sizeof(uint8_t), data_size, fp);
190+ fclose(fp);
191+ return SUCCESS;
192+}
193+ 
194+int CreateOppInGraph(
195+ DataType inDtype, std::vector<ge::Tensor>& input, std::vector<Operator>& inputs, std::vector<Operator>& outputs,
196+ Graph& graph)
197+{
198+ Status ret = SUCCESS;
199+ // 自定义代码：添加单算子定义到图中
200+ auto add1 = op::SquareV2("add1");
201+ std::vector<int64_t> xShape = {1, 4, 4, 4};
202+ ADD_INPUT(1, x, inDtype, xShape,6.1f);
203+ 
204+ ADD_OUTPUT(1, z, inDtype, xShape);
205+ 
206+ outputs.push_back(add1);
207+ // 添加完毕
208+ return SUCCESS;
209+}
210+ 
211+int main(int argc, char* argv[])
212+{
213+ const char* graph_name = "tc_ge_irrun_test";
214+ Graph graph(graph_name);
215+ std::vector<ge::Tensor> input;
216+ 
217+ printf("%s - INFO - [XIR]: Start to initialize ge using ge global options\n", GetTime().c_str());
218+ std::map<AscendString, AscendString> global_options = {{"ge.exec.deviceId", "0"}, {"ge.graphRunMode", "1"}};
219+ Status ret = ge::GEInitialize(global_options);
220+ if (ret != SUCCESS) {
221+ printf("%s - INFO - [XIR]: Initialize ge using ge global options failed\n", GetTime().c_str());
222+ return FAILED;
223+ }
224+ printf("%s - INFO - [XIR]: Initialize ge using ge global options success\n", GetTime().c_str());
225+ 
226+ std::vector<Operator> inputs{};
227+ std::vector<Operator> outputs{};
228+ 
229+ std::cout << argv[1] << std::endl;
230+ char* endptr;
231+ //修改类型
232+ // DataType inDtype = DT_INT32;
233+ DataType inDtype = DT_FLOAT;
234+ 
235+ std::cout << inDtype << std::endl;
236+ 
237+ ret = CreateOppInGraph(inDtype, input, inputs, outputs, graph);
238+ if (ret != SUCCESS) {
239+ printf("%s - ERROR - [XIR]: Create ir session using build options failed\n", GetTime().c_str());
240+ return FAILED;
241+ }
242+ 
243+ if (!inputs.empty() && !outputs.empty()) {
244+ graph.SetInputs(inputs).SetOutputs(outputs);
245+ }
246+ 
247+ std::map<AscendString, AscendString> build_options = {
248+ 
249+ };
250+ printf("%s - INFO - [XIR]: Start to create ir session using build options\n", GetTime().c_str());
251+ ge::Session* session = new Session(build_options);
252+ 
253+ if (session == nullptr) {
254+ printf("%s - ERROR - [XIR]: Create ir session using build options failed\n", GetTime().c_str());
255+ return FAILED;
256+ }
257+ printf("%s - INFO - [XIR]: Create ir session using build options success\n", GetTime().c_str());
258+ printf("%s - INFO - [XIR]: Start to add compute graph to ir session\n", GetTime().c_str());
259+ 
260+ std::map<AscendString, AscendString> graph_options = {
261+ 
262+ };
263+ uint32_t graph_id = 0;
264+ ret = session->AddGraph(graph_id, graph, graph_options);
265+ 
266+ printf("%s - INFO - [XIR]: Session add ir compute graph to ir session success\n", GetTime().c_str());
267+ printf("%s - INFO - [XIR]: dump graph to txt\n", GetTime().c_str());
268+ std::string file_path = "./dump";
269+ aclgrphDumpGraph(graph, file_path.c_str(), file_path.length());
270+ printf("%s - INFO - [XIR]: Start to run ir compute graph\n", GetTime().c_str());
271+ std::vector<ge::Tensor> output;
272+ ret = session->RunGraph(graph_id, input, output);
273+ if (ret != SUCCESS) {
274+ printf("%s - INFO - [XIR]: Run graph failed\n", GetTime().c_str());
275+ delete session;
276+ GEFinalize();
277+ return FAILED;
278+ }
279+ printf("%s - INFO - [XIR]: Session run ir compute graph success\n", GetTime().c_str());
280+ 
281+ int input_num = input.size();
282+ for (int i = 0; i < input_num; i++) {
283+ std::cout << "input " << i << " dtype : " << input[i].GetTensorDesc().GetDataType() << std::endl;
284+ string input_file = "./tc_ge_irrun_test_0008_npu_input_" + std::to_string(i) + ".bin";
285+ uint8_t* input_data_i = input[i].GetData();
286+ int64_t input_shape = input[i].GetTensorDesc().GetShape().GetShapeSize();
287+ std::cout << "this is " << i << "th input, input shape size =" << input_shape << std::endl;
288+ uint32_t data_size = input_shape * GetDataTypeSize(input[i].GetTensorDesc().GetDataType());
289+ WriteDataToFile((const char*)input_file.c_str(), data_size, input_data_i);
290+ }
291+ 
292+ int output_num = output.size();
293+ for (int i = 0; i < output_num; i++) {
294+ std::cout << "output " << i << " dtype : " << output[i].GetTensorDesc().GetDataType() << std::endl;
295+ string output_file = "./tc_ge_irrun_test_0008_npu_output_" + std::to_string(i) + ".bin";
296+ uint8_t* output_data_i = output[i].GetData();
297+ int64_t output_shape = output[i].GetTensorDesc().GetShape().GetShapeSize();
298+ std::cout << "this is " << i << "th output, output shape size =" << output_shape << std::endl;
299+ uint32_t data_size = output_shape * GetDataTypeSize(output[i].GetTensorDesc().GetDataType());
300+ WriteDataToFile((const char*)output_file.c_str(), data_size, output_data_i);
301+ //新加3行打印
302+ float* result = (float*)output_data_i;
303+ // int32_t* result = (int32_t*)output_data_i;
304+ for (int64_t j = 0; j < 8; j++) {
305+ LOG_PRINT("result[%ld] is: %f\n", j, result[j]);
306+ // LOG_PRINT("result[%ld] is: %d\n", j, result[j]);
307+ }
308+ }
309+ 
310+ ge::AscendString error_msg = ge::GEGetErrorMsgV2();
311+ std::string error_str(error_msg.GetString());
312+ std::cout << "Error message: " << error_str << std::endl;
313+ ge::AscendString warning_msg = ge::GEGetWarningMsgV2();
314+ std::string warning_str(warning_msg.GetString());
315+ std::cout << "Warning message: " << warning_str << std::endl;
316+ printf("%s - INFO - [XIR]: Precision is ok\n", GetTime().c_str());
317+ printf("%s - INFO - [XIR]: Start to finalize ir graph session\n", GetTime().c_str());
318+ ret = ge::GEFinalize();
319+ if (ret != SUCCESS) {
320+ printf("%s - INFO - [XIR]: Finalize ir graph session failed\n", GetTime().c_str());
321+ return FAILED;
322+ }
323+ printf("%s - INFO - [XIR]: Finalize ir graph session success\n", GetTime().c_str());
324+ return SUCCESS;
325+}
Aexperimental/math/square_v2/op_graph/CMakeLists.txt+12-0
@@ -0,0 +1,12 @@
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_graph_plugin_sources()
Aexperimental/math/square_v2/op_graph/fusion_pass/CMakeLists.txt+10-0
@@ -0,0 +1,10 @@
1+# This program is free software, you can redistribute it and/or modify it.
2+# Copyright (c) 2025 Huawei Technologies Co., Ltd.
3+# This file is a part of the CANN Open Software.
4+# Licensed under 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, INCLUDING
7+# 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+ 
Aexperimental/math/square_v2/op_graph/square_v2_graph_infer.cpp+46-0
@@ -0,0 +1,46 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+ 
21+/*!
22+ * \file square_v2_graph_infer.cpp
23+ * \brief square_v2 operater graph infer resource
24+ */
25+#include "register/op_impl_registry.h"
26+#include "log/log.h"
27+ 
28+namespace ops {
29+using namespace ge;
30+ 
31+static constexpr int64_t IDX_0 = 0;
32+ 
33+static ge::graphStatus InferDataTypeSquareV2(gert::InferDataTypeContext* context)
34+{
35+ OP_LOGD(context->GetNodeName(), "Begin to do InferDataTypeSquareV2");
36+ 
37+ ge::DataType xDtype = context->GetInputDataType(IDX_0);
38+ context->SetOutputDataType(IDX_0, xDtype);
39+ 
40+ OP_LOGD(context->GetNodeName(), "End to do InferDataTypeSquareV2");
41+ return GRAPH_SUCCESS;
42+}
43+ 
44+IMPL_OP(SquareV2).InferDataType(InferDataTypeSquareV2);
45+ 
46+}; // namespace ops
Aexperimental/math/square_v2/op_graph/square_v2_proto.h+50-0
@@ -0,0 +1,50 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+ 
21+/*!
22+ * \file square_v2_proto.h
23+ * \brief
24+*/
25+#ifndef OPS_OP_PROTO_INC_DIV_H_
26+#define OPS_OP_PROTO_INC_DIV_H_
27+ 
28+#include "graph/operator_reg.h"
29+#include "graph/types.h"
30+ 
31+namespace ge {
32+/**
33+*@brief Returns element-wise square of "x".
34+ 
35+*@par Inputs:
36+x: A ND Tensor of type float16 or float32. \n
37+*@par Outputs:
38+*z: A ND Tensor. Has the same dtype as "x".
39+*@par Third-party framework compatibility
40+*Compatible with the TensorFlow operator Square.
41+*/
42+ 
43+REG_OP(SquareV2)
44+ .INPUT(x, TensorType({DT_FLOAT, DT_FLOAT16}))
45+ .OUTPUT(z, TensorType({DT_FLOAT, DT_FLOAT16}))
46+ .OP_END_FACTORY_REG(SquareV2)
47+ 
48+} // namespace ge
49+ 
50+#endif // OPS_OP_PROTO_INC_SquareV2_H_
Aexperimental/math/square_v2/op_host/config/ascend910b/square_v2_binary.json+67-0
@@ -0,0 +1,67 @@
1+{
2+ "op_type": "SquareV2",
3+ "op_list": [
4+ {
5+ "bin_filename": "SquareV2_a1532827238e1555db7b997c7bce2928",
6+ "inputs": [
7+ {
8+ "name": "x",
9+ "index": 0,
10+ "dtype": "float32",
11+ "format": "ND",
12+ "paramType": "required",
13+ "shape": [
14+ -2
15+ ],
16+ "format_match_mode": "FormatAgnostic"
17+ }
18+ ],
19+ "attrs": [
20+ ],
21+ "outputs": [
22+ {
23+ "name": "z",
24+ "index": 0,
25+ "dtype": "float32",
26+ "format": "ND",
27+ "paramType": "required",
28+ "shape": [
29+ -2
30+ ],
31+ "format_match_mode": "FormatAgnostic"
32+ }
33+ ]
34+ },
35+ {
36+ "bin_filename": "SquareV2_11132827238e1555db7b997c7bce2928",
37+ "inputs": [
38+ {
39+ "name": "x",
40+ "index": 0,
41+ "dtype": "float16",
42+ "format": "ND",
43+ "paramType": "required",
44+ "shape": [
45+ -2
46+ ],
47+ "format_match_mode": "FormatAgnostic"
48+ }
49+ ],
50+ "attrs": [
51+ ],
52+ "outputs": [
53+ {
54+ "name": "z",
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+ }
66+ ]
67+}
Aexperimental/math/square_v2/op_host/config/ascend910b/square_v2_simplified_key.ini+13-0
@@ -0,0 +1,13 @@
1+; 该文件主要影响 opc 工具 编译二进制kernel时， --simplified_key_mode 选项中填写的值，格式如下所示：
2+; [某算子]
3+; default=xx
4+; ascendxx=xx
5+; 其中，default为默认mode，ascendxx为可选mode，如果不同芯片有差异化要求时，需要配置；
6+; 1)如果没有配置：非ascendC算子继续按空处理，即opc编译命令中不添加 --simplified_key_mode 选项，AscendC算子按照 simplified_key_mode=0 处理
7+; 2)如果仅有default配置：各个版本按default配置
8+; 3)如果仅有某些平台的配置，没有default配置：对应平台的按照配置的值传递，非对应平台的：非AscendC算子继续按空处理，AscendC算子按照 simplified_key_mode=0 处理
9+; 4)如果default配置和平台配置都有：对应平台的使用平台的配置，非对应的平台的以default值配置。
10+; 5)对于自定义simplified key的情况，需要在binary_simplified_key_mode.ini 文件中显式配置为None，不传入 --simplified_key_mode 选项，由opc工具和FE框架自行判断使用何种模式
11+; 6)是否是AscendC算子，由 ops/build-in/tbe/op_info_cfg/parser/ascendc_config.json 中配置的算子名字和对于的平台决定
12+[SquareV2]
13+default=0
Aexperimental/math/square_v2/op_host/square_v2_def.cpp+58-0
@@ -0,0 +1,58 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+ 
21+/*!
22+ * \file square_v2.cpp
23+ * \brief
24+ */
25+#include "register/op_def_registry.h"
26+ 
27+namespace ops {
28+ class SquareV2 : public OpDef {
29+ public:
30+ explicit SquareV2(const char* name) : OpDef(name)
31+ {
32+ this->Input("x") // 输入x定义
33+ .ParamType(REQUIRED) // 必选输入
34+ .DataType({ge::DT_FLOAT,ge::DT_FLOAT16}) // 支持数据类型
35+ .Format({ge::FORMAT_ND,ge::FORMAT_ND}) // 支持format格式
36+ .UnknownShapeFormat({ge::FORMAT_ND, ge::FORMAT_ND}) // 未确定大小shape对应format格式
37+ .AutoContiguous(); // 内存自动连续化
38+
39+ this->Output("z")
40+ .ParamType(REQUIRED)
41+ .DataType({ge::DT_FLOAT, ge::DT_FLOAT16})
42+ .Format({ge::FORMAT_ND, ge::FORMAT_ND})
43+ .UnknownShapeFormat({ge::FORMAT_ND, ge::FORMAT_ND})
44+ .AutoContiguous();
45+
46+ OpAICoreConfig aicoreConfig;
47+ aicoreConfig.DynamicCompileStaticFlag(true)
48+ .DynamicFormatFlag(false)
49+ .DynamicRankSupportFlag(true)
50+ .DynamicShapeSupportFlag(true)
51+ .NeedCheckSupportFlag(false)
52+ .PrecisionReduceFlag(true)
53+ .ExtendCfgInfo("opFile.value", "square_v2"); // 这里制定的值会对应到kernel入口文件名.cpp
54+ this->AICore().AddConfig("ascend910b", aicoreConfig); // 其他的soc版本补充部分配置项
55+ }
56+ };
57+ OP_ADD(SquareV2); // 添加算子信息库
58+ } // namespace ops
Aexperimental/math/square_v2/op_host/square_v2_infershape.cpp+52-0
@@ -0,0 +1,52 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+ 
21+/*!
22+ * \file square_v2_infer.cpp
23+ * \brief
24+ */
25+#include "register/op_impl_registry.h"
26+#include "log/log.h"
27+ 
28+using namespace ge;
29+ 
30+namespace ops {
31+ static constexpr int64_t IDX_0 = 0;
32+
33+ static ge::graphStatus InferShapeSquareV2(gert::InferShapeContext* context)
34+ {
35+ OP_LOGD(context->GetNodeName(), "Begin to do InferShapeSquareV2");
36+
37+ // get input shapes
38+ const gert::Shape* xShape = context->GetInputShape(IDX_0);
39+ OP_CHECK_NULL_WITH_CONTEXT(context, xShape);
40+
41+ // get output shapes
42+ gert::Shape* zShape = context->GetOutputShape(IDX_0);
43+ OP_CHECK_NULL_WITH_CONTEXT(context, zShape);
44+
45+ *zShape = *xShape;
46+
47+ OP_LOGD(context->GetNodeName(), "End to do InferShapeSquareV2");
48+ return GRAPH_SUCCESS;
49+ }
50+
51+ IMPL_OP_INFERSHAPE(SquareV2).InferShape(InferShapeSquareV2);
52+ } // namespace ops
Aexperimental/math/square_v2/op_host/square_v2_tiling.cpp+227-0
@@ -0,0 +1,227 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+ 
21+/*!
22+ * \file square_v2_tiling.cpp
23+ * \brief
24+ */
25+ 
26+#include "log/log.h"
27+#include "util/math_util.h"
28+#include "tiling_base/tiling_util.h"
29+#include "tiling_base/tiling_templates_registry.h"
30+#include "../op_kernel/square_v2_tiling_data.h"
31+#include "../op_kernel/square_v2_tiling_key.h"
32+ 
33+namespace optiling {
34+ 
35+ using namespace Ops::Math::OpTiling;
36+ const uint32_t BLOCK_SIZE = 32;
37+ const uint32_t BUFFER_NUM = 2;
38+ const uint32_t WS_SYS_SIZE = 16U * 1024U * 1024U;
39+
40+ struct SquareV2CompileInfo {};
41+
42+ // 获取平台信息如ubSize, coreNum
43+ static ge::graphStatus GetPlatformInfo(gert::TilingContext* context, uint64_t& ubSize, int64_t& coreNum)
44+ {
45+ // 获取ubsize coreNum
46+ fe::PlatFormInfos* platformInfoPtr = context->GetPlatformInfo();
47+ OP_CHECK_NULL_WITH_CONTEXT(context, platformInfoPtr);
48+ auto ascendcPlatform = platform_ascendc::PlatformAscendC(platformInfoPtr);
49+ coreNum = ascendcPlatform.GetCoreNumAiv();
50+ OP_CHECK_IF(coreNum == 0, OP_LOGE(context, "coreNum is 0"), return ge::GRAPH_FAILED);
51+ ascendcPlatform.GetCoreMemSize(platform_ascendc::CoreMemType::UB, ubSize);
52+ OP_CHECK_IF(ubSize == 0, OP_LOGE(context, "ubSize is 0"), return ge::GRAPH_FAILED);
53+ return ge::GRAPH_SUCCESS;
54+ }
55+
56+ // 获取属性，shape信息
57+ static ge::graphStatus GetShapeAttrsInfo(gert::TilingContext* context, int64_t& totalIdx, ge::DataType& dataType)
58+ {
59+ // 获取输入shape信息
60+ auto inputX = context->GetInputShape(0);
61+ OP_CHECK_NULL_WITH_CONTEXT(context, inputX);
62+ // 如果输入shape 是标量 转换为{1}，否则保持原 shape 不变
63+ auto inputShapeX = EnsureNotScalar(inputX->GetStorageShape());
64+
65+ auto outZ = context->GetOutputShape(0);
66+ OP_CHECK_NULL_WITH_CONTEXT(context, outZ);
67+ auto outShapeZ = EnsureNotScalar(outZ->GetStorageShape());
68+
69+ // shape校验
70+ bool shapeMatch = true;
71+ // 校验维度数一致
72+ if (inputShapeX.GetDimNum() != outShapeZ.GetDimNum()) {
73+ shapeMatch = false;
74+ } else {
75+ // 校验每个维度的大小一致
76+ size_t dimNum = inputShapeX.GetDimNum();
77+ for (size_t i = 0; i < dimNum; i++) {
78+ if (inputShapeX.GetDim(i) != outShapeZ.GetDim(i)) {
79+ shapeMatch = false;
80+ break;
81+ }
82+ }
83+ }
84+
85+ // 形状不匹配则报错
86+ OP_CHECK_IF(
87+ !shapeMatch,
88+ OP_LOGE(
89+ context, "SquareV2: inputx,outputz shape not match! dim num: x=%zu, z=%zu",
90+ inputShapeX.GetDimNum(), outShapeZ.GetDimNum()),
91+ return ge::GRAPH_FAILED);
92+
93+ totalIdx = inputX->GetOriginShape().GetShapeSize();
94+ // dtype校验
95+ const std::set<ge::DataType> supportedDtype = {ge::DT_FLOAT, ge::DT_FLOAT16};
96+ auto inputDesc = context->GetInputDesc(0);
97+ OP_CHECK_NULL_WITH_CONTEXT(context, inputDesc);
98+ dataType = inputDesc->GetDataType();
99+ if (supportedDtype.count(dataType) == 0) {
100+ OP_LOGE(context, "invalid dtype");
101+ return ge::GRAPH_FAILED;
102+ }
103+ return ge::GRAPH_SUCCESS;
104+ }
105+
106+ static ge::graphStatus GetWorkspaceSize(gert::TilingContext* context)
107+ {
108+ size_t* currentWorkspace = context->GetWorkspaceSizes(1);
109+ OP_CHECK_NULL_WITH_CONTEXT(context, currentWorkspace);
110+ currentWorkspace[0] = WS_SYS_SIZE;
111+ return ge::GRAPH_SUCCESS;
112+ }
113+
114+ // tiling 分发入口
115+ // 可直接替换你的 SquareV2TilingFunc 内部实现（保留函数签名）
116+ static ge::graphStatus SquareV2TilingFunc(gert::TilingContext* context)
117+ {
118+ // 1. platform
119+ uint64_t ubSize = 0;
120+ int64_t coreNum = 0;
121+ OP_CHECK_IF(GetPlatformInfo(context, ubSize, coreNum) != ge::GRAPH_SUCCESS,
122+ OP_LOGE(context, "GetPlatformInfo error"), return ge::GRAPH_FAILED);
123+
124+ // 2. shapes & dtype
125+ int64_t totalIdx = 0;
126+ ge::DataType dataType;
127+ OP_CHECK_IF(GetShapeAttrsInfo(context, totalIdx, dataType) != ge::GRAPH_SUCCESS,
128+ OP_LOGE(context, "GetShapeAttrsInfo error"), return ge::GRAPH_FAILED);
129+
130+ // 3. workspace
131+ OP_CHECK_IF(GetWorkspaceSize(context) != ge::GRAPH_SUCCESS,
132+ OP_LOGE(context, "GetWorkspaceSize error"), return ge::GRAPH_FAILED);
133+
134+ SquareV2TilingData* tiling = context->GetTilingData<SquareV2TilingData>();
135+ OP_CHECK_NULL_WITH_CONTEXT(context, tiling);
136+ OP_CHECK_IF(memset_s(tiling, sizeof(SquareV2TilingData), 0, sizeof(SquareV2TilingData)) != EOK,
137+ OP_LOGE(context, "set tiling data error"), return ge::GRAPH_FAILED);
138+
139+ // --- safer numeric types ---
140+ uint32_t typeLength = 0;
141+ ge::TypeUtils::GetDataTypeLength(context->GetInputDesc(0)->GetDataType(), typeLength);
142+ if (typeLength == 0) {
143+ OP_LOGE(context, "typeLength is 0");
144+ return ge::GRAPH_FAILED;
145+ }
146+ uint64_t inputBytes = static_cast<uint64_t>(typeLength);
147+ uint64_t inputLengthBytes = static_cast<uint64_t>(totalIdx) * inputBytes;
148+
149+ // ub-based tileBlockNum guard (避免为0)
150+ uint32_t ubDataNumber = (inputBytes == 1ULL) ? 4U : 2U;
151+ uint64_t tmp = (ubSize / BLOCK_SIZE / BUFFER_NUM);
152+ uint32_t tileBlockNum = 1U;
153+ if (tmp > 0) {
154+ uint64_t tb = tmp / ubDataNumber;
155+ tileBlockNum = (tb == 0) ? 1U : static_cast<uint32_t>(tb);
156+ }
157+
158+ // 每个 tile 包含的元素数（至少 1）
159+ uint32_t tileDataNum = static_cast<uint32_t>((static_cast<uint64_t>(tileBlockNum) * BLOCK_SIZE) / inputBytes);
160+ if (tileDataNum == 0U) tileDataNum = 1U;
161+
162+ // 总 block 数（向上取整）
163+ uint64_t blocksTotal = (inputLengthBytes + BLOCK_SIZE - 1ULL) / BLOCK_SIZE;
164+ uint64_t coreNum64 = static_cast<uint64_t>(coreNum);
165+ if (coreNum64 > blocksTotal){
166+ coreNum64 = blocksTotal;
167+ }
168+ if (coreNum64 == 0ULL) coreNum64 = 1ULL; // 最少 1 core
169+ uint32_t finalCoreNum = static_cast<uint32_t>(coreNum64);
170+
171+ uint64_t everyCoreInputBlockNum = blocksTotal / coreNum64; // 基本块数
172+ uint32_t tailBlockNum = static_cast<uint32_t>(blocksTotal % coreNum64); // 前 tailBlockNum 个核是 big-core
173+
174+ // small-core 数量（元素）
175+ uint64_t smallCoreDataNum_u = everyCoreInputBlockNum * BLOCK_SIZE / inputBytes;
176+ uint32_t smallCoreDataNum = static_cast<uint32_t>(smallCoreDataNum_u);
177+
178+ uint32_t smallTileNum = static_cast<uint32_t>(everyCoreInputBlockNum / static_cast<uint64_t>(tileBlockNum));
179+ uint32_t finalSmallTileNum = ((everyCoreInputBlockNum % tileBlockNum) == 0) ? smallTileNum : (smallTileNum + 1);
180+ int64_t smallTailDataNum_i = static_cast<int64_t>(smallCoreDataNum) - static_cast<int64_t>(tileDataNum) * static_cast<int64_t>(smallTileNum);
181+ uint32_t smallTailDataNum = (smallTailDataNum_i <= 0) ? tileDataNum : static_cast<uint32_t>(smallTailDataNum_i);
182+
183+ // big-core（每个多一个 block）
184+ uint64_t bigEveryCoreBlockNum = everyCoreInputBlockNum + 1ULL;
185+ uint64_t bigCoreDataNum_u = bigEveryCoreBlockNum * BLOCK_SIZE / inputBytes;
186+ uint32_t bigCoreDataNum = static_cast<uint32_t>(bigCoreDataNum_u);
187+ uint32_t bigTileNum = static_cast<uint32_t>(bigEveryCoreBlockNum / static_cast<uint64_t>(tileBlockNum));
188+ uint32_t finalBigTileNum = ((bigEveryCoreBlockNum % tileBlockNum) == 0) ? bigTileNum : (bigTileNum + 1);
189+ int64_t bigTailDataNum_i = static_cast<int64_t>(bigCoreDataNum) - static_cast<int64_t>(tileDataNum) * static_cast<int64_t>(bigTileNum);
190+ uint32_t bigTailDataNum = (bigTailDataNum_i <= 0) ? tileDataNum : static_cast<uint32_t>(bigTailDataNum_i);
191+
192+ // write back
193+ tiling->smallCoreDataNum = static_cast<int64_t>(smallCoreDataNum);
194+ tiling->bigCoreDataNum = static_cast<int64_t>(bigCoreDataNum);
195+ tiling->tileDataNum = static_cast<int64_t>(tileDataNum);
196+ tiling->smallTailDataNum = static_cast<int64_t>(smallTailDataNum);
197+ tiling->bigTailDataNum = static_cast<int64_t>(bigTailDataNum);
198+ tiling->finalSmallTileNum = static_cast<int64_t>(finalSmallTileNum);
199+ tiling->finalBigTileNum = static_cast<int64_t>(finalBigTileNum);
200+ tiling->tailBlockNum = static_cast<int64_t>(tailBlockNum);
201+
202+ context->SetBlockDim(finalCoreNum);
203+
204+ uint64_t tilingKey = 0;
205+ if (dataType == ge::DT_FLOAT) {
206+ tilingKey = GET_TPL_TILING_KEY(ELEMENTWISE_TPL_SCH_MODE_0);
207+ context->SetTilingKey(tilingKey);
208+ } else if (dataType == ge::DT_FLOAT16) {
209+ tilingKey = GET_TPL_TILING_KEY(ELEMENTWISE_TPL_SCH_MODE_1);
210+ context->SetTilingKey(tilingKey);
211+ } else {
212+ OP_LOGE(context, "get dtype error");
213+ return ge::GRAPH_FAILED;
214+ }
215+ return ge::GRAPH_SUCCESS;
216+ }
217+
218+
219+ static ge::graphStatus TilingParseForSquareV2([[maybe_unused]] gert::TilingParseContext* context)
220+ {
221+ return ge::GRAPH_SUCCESS;
222+ }
223+
224+ // tiling注册入口.
225+ IMPL_OP_OPTILING(SquareV2).Tiling(SquareV2TilingFunc).TilingParse<SquareV2CompileInfo>(TilingParseForSquareV2);
226+ } // namespace optiling
227+
Aexperimental/math/square_v2/op_kernel/square_v2.cpp+50-0
@@ -0,0 +1,50 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+ 
21+/*!
22+ * \file square_v2.cpp
23+ * \brief
24+ */
25+ 
26+#include "square_v2.h"
27+ 
28+enum class SquareV2TilingKey : uint32_t
29+{
30+ TILING_KEY_EXAMPLE_FLOAT = 0,
31+ TILING_KEY_EXAMPLE_HALF = 1,
32+};
33+ 
34+template <uint32_t schMode>
35+__global__ __aicore__ void square_v2(GM_ADDR x, GM_ADDR z, GM_ADDR workspace, GM_ADDR tiling)
36+{
37+ REGISTER_TILING_DEFAULT(SquareV2TilingData);
38+ GET_TILING_DATA_WITH_STRUCT(SquareV2TilingData, tilingData, tiling);
39+ if constexpr (schMode == static_cast<uint32_t>(SquareV2TilingKey::TILING_KEY_EXAMPLE_FLOAT)) {
40+ NsSquareV2::SquareV2<float> op; // 算子kernel实例获取
41+ op.Init(x, z, &tilingData); // 算子kernel实例初始化
42+ op.Process(); // 算子kernel实例执行
43+ }
44+
45+ else if constexpr (schMode == static_cast<uint32_t>(SquareV2TilingKey::TILING_KEY_EXAMPLE_HALF)) {
46+ NsSquareV2::SquareV2<half> op; // 算子kernel实例获取
47+ op.Init(x, z, &tilingData); // 算子kernel实例初始化
48+ op.Process(); // 算子kernel实例执行
49+ }
50+}
Aexperimental/math/square_v2/op_kernel/square_v2.h+132-0
@@ -0,0 +1,132 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+ 
21+/*!
22+ * \file square_v2.h
23+ * \brief
24+ */
25+#ifndef __SQUARE_V2_H__
26+#define __SQUARE_V2_H__
27+ 
28+#include "kernel_operator.h"
29+#include "kernel_tiling/kernel_tiling.h"
30+#include "square_v2_tiling_data.h"
31+#include "square_v2_tiling_key.h"
32+ 
33+namespace NsSquareV2 {
34+ 
35+ using namespace AscendC;
36+ 
37+ constexpr int32_t BUFFER_NUM = 2;
38+
39+ template <typename T>
40+ class SquareV2 {
41+ public:
42+ __aicore__ inline SquareV2(){};
43+
44+ __aicore__ inline void Init(GM_ADDR x, GM_ADDR z, const SquareV2TilingData* tilingData);
45+ __aicore__ inline void Process();
46+
47+ private:
48+ __aicore__ inline void CopyIn(int32_t progress);
49+ __aicore__ inline void CopyOut(int32_t progress);
50+ __aicore__ inline void Compute(int32_t progress);
51+
52+ private:
53+ TPipe pipe;
54+ TQue<QuePosition::VECIN, BUFFER_NUM> inputQueueX;
55+ TQue<QuePosition::VECOUT, BUFFER_NUM> outputQueueZ;
56+ GlobalTensor<T> inputGMX;
57+ GlobalTensor<T> outputGMZ;
58+
59+ uint32_t coreDataNum;
60+ uint32_t tileNum;
61+ uint32_t tileDataNum;
62+ uint32_t tailDataNum;
63+ uint32_t processDataNum;
64+ };
65+
66+ template <typename T>
67+ __aicore__ inline void SquareV2<T>::Init(GM_ADDR x, GM_ADDR z, const SquareV2TilingData* tilingData)
68+ {
69+ ASSERT(AscendC::GetBlockNum() != 0 && "block dim can not be zero!");
70+ uint32_t coreNum = AscendC::GetBlockIdx();
71+ uint32_t globalBufferIndex = tilingData->bigCoreDataNum * AscendC::GetBlockIdx();
72+ this->tileDataNum = tilingData->tileDataNum;
73+ if (coreNum < tilingData->tailBlockNum) {
74+ this->coreDataNum = tilingData->bigCoreDataNum;
75+ this->tileNum = tilingData->finalBigTileNum;
76+ this->tailDataNum = tilingData->bigTailDataNum;
77+ }
78+ else {
79+ this->coreDataNum = tilingData->smallCoreDataNum;
80+ this->tileNum = tilingData->finalSmallTileNum;
81+ this->tailDataNum = tilingData->smallTailDataNum;
82+ globalBufferIndex -= (tilingData->bigCoreDataNum - tilingData->smallCoreDataNum) * (AscendC::GetBlockIdx() - tilingData->tailBlockNum);
83+ }
84+ inputGMX.SetGlobalBuffer((__gm__ T*)x + globalBufferIndex, this->coreDataNum);
85+ outputGMZ.SetGlobalBuffer((__gm__ T*)z + globalBufferIndex, this->coreDataNum);
86+ pipe.InitBuffer(inputQueueX, BUFFER_NUM, this->tileDataNum * sizeof(T));
87+ pipe.InitBuffer(outputQueueZ, BUFFER_NUM, this->tileDataNum * sizeof(T));
88+ }
89+
90+ template <typename T>
91+ __aicore__ inline void SquareV2<T>::CopyIn(int32_t progress)
92+ {
93+ AscendC::LocalTensor<T> xLocal = inputQueueX.AllocTensor<T>();
94+ AscendC::DataCopy(xLocal, inputGMX[progress * this->tileDataNum], this->processDataNum);
95+ inputQueueX.EnQue(xLocal);
96+ }
97+
98+ template <typename T>
99+ __aicore__ inline void SquareV2<T>::CopyOut(int32_t progress)
100+ {
101+ AscendC::LocalTensor<T> zLocal = outputQueueZ.DeQue<T>();
102+ AscendC::DataCopy(outputGMZ[progress * this->tileDataNum], zLocal, this->processDataNum);
103+ outputQueueZ.FreeTensor(zLocal);
104+ }
105+
106+ template <typename T>
107+ __aicore__ inline void SquareV2<T>::Compute(int32_t progress)
108+ {
109+ AscendC::LocalTensor<T> xLocal = inputQueueX.DeQue<T>();
110+ AscendC::LocalTensor<T> zLocal = outputQueueZ.AllocTensor<T>();
111+ AscendC::Mul(zLocal, xLocal,xLocal, this->processDataNum);
112+ outputQueueZ.EnQue<T>(zLocal);
113+ inputQueueX.FreeTensor(xLocal);
114+ }
115+
116+ template <typename T>
117+ __aicore__ inline void SquareV2<T>::Process()
118+ {
119+ int32_t loopCount = this->tileNum;
120+ this->processDataNum = this->tileDataNum;
121+ for (int32_t i = 0; i < loopCount; i++) {
122+ if (i == this->tileNum - 1) {
123+ this->processDataNum = this->tailDataNum;
124+ }
125+ CopyIn(i);
126+ Compute(i);
127+ CopyOut(i);
128+ }
129+ }
130+ 
131+} // namespace NsSquareV2
132+#endif // SQUARE_V2_H
Aexperimental/math/square_v2/op_kernel/square_v2_tiling_data.h+39-0
@@ -0,0 +1,39 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+ 
21+/*!
22+ * \file square_v2_tiling_data.h
23+ * \brief tiling data struct
24+ */
25+ 
26+#ifndef __SQUARE_V2_TILLING_DATA_H__
27+#define __SQUARE_V2_TILLING_DATA_H__
28+ 
29+struct SquareV2TilingData {
30+ int64_t smallCoreDataNum;
31+ int64_t bigCoreDataNum;
32+ int64_t finalBigTileNum;
33+ int64_t finalSmallTileNum;
34+ int64_t tileDataNum;
35+ int64_t smallTailDataNum;
36+ int64_t bigTailDataNum;
37+ int64_t tailBlockNum;
38+};
39+#endif
Aexperimental/math/square_v2/op_kernel/square_v2_tiling_key.h+45-0
@@ -0,0 +1,45 @@
1+/**
2+ * This file is part of the OpenBOAT project at Harbin Institute of Technology (HIT)
3+ * and is contributed to the CANN Open Software.
4+ *
5+ * Copyright (c) 2025 AISS Group, Harbin Institute of Technology (HIT).
6+ * All Rights Reserved.
7+ *
8+ * Authors (accounts):
9+ * - Li Wen <@liwenkkklll>
10+ * - Su Tonghua <@sutonghua>
11+ *
12+ * This program is free software: you can redistribute it and/or modify it.
13+ * Licensed under the CANN Open Software License Agreement Version 2.0 (the "License").
14+ * You may not use this file except in compliance with the License.
15+ * See the LICENSE file at the root of the repository for the full text of the License.
16+ *
17+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
18+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
19+ */
20+ 
21+/*!
22+ * \file square_v2_tiling_key.h
23+ * \brief square_v2 tiling key declare
24+ */
25+ 
26+#ifndef __SQUARE_V2_TILING_KEY_H__
27+#define __SQUARE_V2_TILING_KEY_H__
28+ 
29+#include "ascendc/host_api/tiling/template_argument.h"
30+ 
31+/* Mode场景定义 */
32+#define ELEMENTWISE_TPL_SCH_MODE_0 0
33+#define ELEMENTWISE_TPL_SCH_MODE_1 1
34+/* 继续定义其他Mode场景... */
35+ 
36+/* 模板参数 */
37+ASCENDC_TPL_ARGS_DECL(
38+ SquareV2,
39+ ASCENDC_TPL_UINT_DECL(schMode, 1, ASCENDC_TPL_UI_LIST, ELEMENTWISE_TPL_SCH_MODE_0, ELEMENTWISE_TPL_SCH_MODE_1));
40+ 
41+/* 模板参数组合 */
42+ASCENDC_TPL_SEL(ASCENDC_TPL_ARGS_SEL(
43+ ASCENDC_TPL_UINT_SEL(schMode, ASCENDC_TPL_UI_LIST, ELEMENTWISE_TPL_SCH_MODE_0, ELEMENTWISE_TPL_SCH_MODE_1)));
44+ 
45+#endif