cuda template CY Visual

starter kit for GPU programming. This repo is made primarily for windows / VScode but can be used anywhere with some tweaking This template can replace the old one using freeglut.

• uses opengl / GLFW
• has the same functionnalities than the previous one
• support for window resizing added
• added ImGui - docking branch

Requierement

Compiler

While this repo is made for VSCode, it is highly recommended (and probably mandatory) to install VSCommunity with c/c++ desktop dev before installing cuda, because the nvcc compiler relies on MSVC toolchains which is installed by Microsoft Visual Studio.

CUDA toolkit 12.5

Download and execute the version matching your OS (may take a few minutes): https://developer.nvidia.com/cuda-12-5-0-download-archive

At this point you can easily check that your cuda install is working by cloning the cuda samples repo.

FIXME (replace with exact string) then in VSCommunity > open solution > select cuda samples .sln (may take a few minutes) on any project : right clic > set as start up project Project to check :

• param
• bandwidth
• fluid

Recommended VS Code extension

• C/C++ : for intellisense
• Nsight Visual Studio Code extension : for advanced debugging. for more details see https://www.youtube.com/watch?v=gN3XeFwZ4ng

GLFW

This repo already contains /lib/glfw3_mt.lib which is a pre-compilled glfw runtime binaries for windows 10 / Visual C++ 2022.

if you're using a different OS, you need to replace it with the expected binaries :

• glfw binaries : https://www.glfw.org/download.html

Compiling

Instruction

1. add a few things to path. The location that you need to add should look like this :

C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.41.34120\bin\Hostx64\x64

2. create /exe/ diretory

3. If you're using VSCode You can simply run the provided task (ctrl+shift+p > Run Task > build nvcc)

It will simply run the following command :

nvcc template.cu ./imgui/*.cpp -o ./exe/template.exe -I./include/ -L./lib/ -lglfw3_mt -lopengl32 -luser32 -lgdi32 -lshell32

The wildcard *.cpp might not expand, in that case you'll need to use a script or list file individually

nvcc template.cu (Get-ChildItem -Path .\\imgui\\ -Filter *.cpp | ForEach-Object { $_.FullName }) -o ./exe/template.exe -I./include/ -L./lib/ -lglfw3_mt -lopengl32 -luser32 -lgdi32 -lshell32

nvcc template.cu ./imgui/imgui_draw.cpp ./imgui/imgui_impl_glfw.cpp ./imgui/imgui_impl_opengl3.cpp ./imgui/imgui_tables.cpp ./imgui/imgui_widgets.cpp ./imgui/imgui.cpp -o ./exe/template.exe -I./include/ -L./lib/ -lglfw3_mt -lopengl32 -luser32 -lgdi32 -lshell32

Static vs dynamic compilation for GLFW

The VScode task will compile statically. (as it's obviously recommended). If for some reason you want to compile dynamically, check the readme of glfw pre-compiled binairies At the very least you'll need to add glfw3.dll to your .exe directory (or path) and link with glfw3dll.lib. Note that you can get rid of -luser32, -lgdi32, -lshell32 if you're compiling dynamically

Using ImGui

You can learn more about ImGui here : Original repo

The easiest way to learn about ImGui is by running the example file and fooling arround. Then just ctrl+f in imgui_demo.cpp to find out how things are done

FIXME (commmand to be confirmed) FIXME (add the option to remove debug compilation option for imgui)

If you MSYS2 - mingw64 then you can install GLFW in mingw64 shell with

pacman -S mingw-w64-x86_64-glfw

And compile imgui_example.cpp with :

g++ imgui_example.cpp ./imgui/*.cpp -o ./exe/example.exe -lopengl32 -lglfw3

Otherwise, you need to to precise include path to g++. If you're using windows, you'll also need to make glfw3.dll visible from you're OS. either add glfw3.dll in the same directory than your executable or add glfw-3.4.bin.WIN64\lib-vc2022 to PATH environnement variable

g++ imgui_example.cpp ./imgui/*.cpp -o ./exe/example.exe -I./include/ -L./lib/ -lopengl32 -lglfw3

But please DO NOT manually add dll to /system32/