Sample Code

aclrtContext context_;
aclrtStream stream_;
 
// 1. Initialize AscendCL.
 
// 2. Allocate runtime resources, including setting the compute device and creating a context and a stream.
aclError aclRet = ACL_ERROR_NONE;
aclRet = aclprofStart(config);
if (aclRet != ACL_ERROR_NONE) {
    ERROR_LOG("acl prof start failed");
    return FAILED;
}

ret = aclrtSetDevice(0);
if (ret != ACL_ERROR_NONE) {
    ERROR_LOG("aclrtSetDevice faile");
    return FAILED;
}
ret = aclrtCreateContext(&context_, 0);
if (ret != ACL_ERROR_NONE) {
    ERROR_LOG("acl create context failed");
    return FAILED;
}
aclError ret = aclrtCreateStream(&stream_);
if (ret != ACL_ERROR_NONE) {
    ERROR_LOG("acl create stream failed");
    return FAILED;
}
....
 
// 3. Initialize profiling.
// Set the data flush path.
const char *aclProfPath = "...";
aclprofInit(aclProfPath, strlen(aclProfPath));
 
// 4. Configure profiling.
uint32_t deviceIdList[1] = {0}; // Set this parameter based on the device ID in the actual environment.
// Create a configuration struct.
aclprofConfig *config = aclprofCreateConfig(deviceIdList, 1, ACL_AICORE_ARITHMETIC_UTILIZATION,
nullptr,ACL_PROF_ACL_API | ACL_PROF_TASK_TIME | ACL_PROF_MSPROFTX);
 
// 5. Enable profiling.
ret = aclprofStart(config);
....
 
 
// 6. Add dots, for example, before and after model execution.
mstxRangeId id = mstxRangeStartA("model execute", nullptr); // Set the second input parameter to nullptr, which records only the time consumed by the range on the host (applicable to the code snippet on the host). A valid stream is set, and the time consumed by the host and the corresponding device is recorded (applicable to the code snippet for delivering computation or communication).
ret = aclmdlExecute(modelId, input, output);
mstxRangeEnd(id);
 
// 7. Other operations
 
// 8. Stop profiling.
aclprofStop(config);
aclprofDestroyConfig(config);
aclprofFinalize();
 
// 9. Destroy runtime allocations.
 
// 10. Deinitialize AscendCL.