CIS 565 Performance Lab

This is the second in series of 2 labs for CUDA. In this lab, we will discuss how NVIDIA's Nsight tools can aid in profiling CUDA programs.

The performance lab will be held in class. Please bring your Windows CUDA-Capable laptop with the code downloaded and built. If you do not have a Windows CUDA-Capable laptop, please find a partner to work with. You will be able to follow the steps with Linux as well, but we will show this with Windows in the class.

TODO Before Performance Lab

Clone, Build and Run

The source code consists of a CMake build structure similar to Project 0. Clone this repository from Github and then run the instructions below.

1. From the Performance-Lab directory, run:
   • Windows: cmake -B build -S . -G "Visual Studio 17 2022" -A x64 to generate the Visual Studio project. You may choose a different Visual Studio version.
   • Linux: cmake -B build -S . -G "Unix Makefiles" -DCMAKE_BUILD_TYPE=Debug to generate the makefiles.
2. Open the generated PerformanceLab.sln project from the build directory.
3. Build the project in Visual Studio. (Note that there are Debug and Release configuration options.)
4. Run. Make sure you run the transpose target (not ALL_BUILD) by right-clicking it and selecting "Set as StartUp Project".

Note:

• If you start running using F5, the command prompt will open and close.
  • The F5 shortcut is Start Debugging, which means Visual Studio will monitoring your application and it will not run at full performance.
  • Use F5 only when you are debugging.
• Instead, use Ctrl+F5 when you want to run without debugging. This will run the application at full performance as well as keep the command prompt open after the application ends.

Please ask me or the TAs ahead of time if you have trouble compiling the code. We want to be ready to go at the start of the lab.

Complete Copy Kernel, Naive Transpose and Shared Memory Transpose

• In the interest of time, we will not be writing code from scratch of these during the lab. So it is in your best interest to complete the code before the lab.
  • This will let you review how transpose with shared memory works. So when we write more advance kernel that build on these, you will know what changes we are making.
• The sections you need to work on are marked by TODO: COMPLETE THIS.
  • There are 3 kernels you need to write, and 3 blocks/grids configurations you need to set up.
• Once completed, your output should look like this:

Third Party Code

This repository includes code from termcolor licensed under the BSD 3 Clause.