باستخدام أدوات تطوير كودا تنفيذ المهام ادناه

بايثون C++ Programming بايثون C++ Programming

Using CUDA Development Tools 1. Build the project in Release mode and run the nvvp profiler. 2. Find and record the running times of the cudaMemcpy functions. What is the speed of data transfer? Does it match the PCI version of the computer? 3. Find and record in the report the operating time of the kernel function, the number of used registers, the theoretical occupancy of the multiprocessor. Are you combining multiple blocks to run on a single multiprocessor? How many blocks are combined? 4. Run the calculation of the following metrics in the profiler and enter the calculated values in the report: 1. Bandwidth read / write global memory 2. The effectiveness of these operations 3. Issued IPC - the average number of instructions executed by the warp per 1 processor cycle 4. Executed IPC - the average number of instructions executed by warp per 1 processor clock cycle, excluding repeated instructions executed due to lack of data in the cache or conflicts of access to shared or constant memory, or register spilling 5. Achieved Occupancy 6. The frequency of successful calls to L1 and L2 cache (Hit Rate). 5. Think about the values of the indicators. Do they meet your expectations? Is it possible to further speed up the program? 6. If the values of the indicators say that the program can be accelerated by more than 2 times, do it. Attach the old and the new version of the program to the answer and write how many times it has accelerated.