Command line tools for building cuda kernels?

Accelerated Computing CUDA CUDA Programming and Performance
1

Are there any command line tools for building cuda Kernels?

I have a very lengthy build chain and would like to automate the compilation and deployment of cuda kernel code.

I cant even seem to build a static lib from the command line without tons of internal errors.

2

are you talking about windows?

3

As stated, your requirements are possibly underspecified, especially if your use case involves online compilation. For classical off-line builds, nvcc and makefiles and possibly some scripts are all you need for complex builds and automation, just like in the case of CPU-only build environments. You can find a simple example of how to create a static library of CUDA code in these forums, and may want to spend some quality time with the nvcc documentation.

[Later:] No need to search the forums, NVIDIA provides an example of how to build a static library, it seems:

[url]CUDA Samples :: CUDA Toolkit Documentation

Here is a relevant forum thread:

[url]https://devtalk.nvidia.com/default/topic/526645/how-to-create-a-static-lib-using-cuda-5-0-6-5-and-vs2010-problem-solved-and-bug-found-/[/url]

4

Windows64 for sure.

The static lib is what I am trying to build.

Cant seem to get it right out side of the IDE.

Looks like I was able to generate a kernel.cu.obj

This batch file :

setlocal  

CALL "%VS140COMNTOOLS%..\..\vc\vcvarsall.bat" x86_amd64

SET PATH=%PATH%;"C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\bin\x86_amd64\"

"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\bin\nvcc.exe" -ccbin "C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\bin\x86_amd64" -x cu  -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\include" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\include"     --keep-dir x64\Release -maxrregcount=0  --machine 64 --compile      -DWIN32 -DWIN64 -DNDEBUG -D_CONSOLE -D_MBCS -Xcompiler "/EHsc /W3 /nologo /O2 /FS /Zi  /MD " -o x64\Release\kernel.cu.obj %1 -clean 

"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\bin\nvcc.exe" -gencode=arch=compute_30,code=\"sm_30,compute_30\" --use-local-env -ccbin "C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\bin\x86_amd64" -x cu  -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\include" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\include"     --keep-dir x64\Release -maxrregcount=0  --machine 64 --compile -cudart static     -DWIN32 -DWIN64 -DNDEBUG -D_CONSOLE -D_MBCS -Xcompiler "/EHsc /W3 /nologo /O2 /FS /Zi  /MD " -o x64\Release\kernel.cu.obj %1

endlocal

Whats the majic to create a lib?

5

No magic should be required, though it probably helps to have knowledge of how static libraries are built on Windows when building regular CPU code.

I would suggest starting the other way around: First look at the forum thread I linked above and make sure you can replicate the example I gave using the step-by-step process given there. Once that works, start replacing the files used in the example with your own files.

6

The lib can be created simply:

LIB.EXE /OUT:MYLIB.LIB FILE1.OBJ FILE2.OBJ

Or a dll:

LINK.EXE /DLL /OUT:MYLIB.DLL FILE3.OBJ FILE4.OBJ

7

However Online compiling and deployment would be a huge benefit.

I cant provide people with an app that adapts its core functionality as per their wishes, without requiring them to first install the SDK, Visual Studio…

I was hoping to sell stuff to people who aren’t engineers.