Compiling DLL for running Fiji on Workstation GPU

Good day all,

We just recently got a workstation for processing SPIM data, and I am trying to build a dynamic library (from files downloaded in the link below) to run the bead-based registration on Fiji using the PC’s GPU. I should do that using Nsight in Microsoft Visual Studio… as stated in the link below.

I have tried my hands on this and its getting really confusing. I do not really have a good knowledge base to run this through. I guess the instructions in the link assumes some level of related IT knowledge. so I was hoping someone here must have done this before and will be willing to give me some help…maybe a more detailed guide/documentation of how to do set this up.

Thanks for your help

Gideon

1 Like

@StephanPreibisch @hoerldavid @haesleinhuepf

3 Likes

still hoping for help on this please

@StephanPreibisch @hoerldavid @haesleinhuepf

Hey @gideon! Stephan actually provided a pre-compiled DLL, see Release Release Windows Binary · StephanPreibisch/SeparableConvolutionCUDALib · GitHub – did you try that one by chance?

1 Like

Hi @skalarproduktraum

Thanks for your response. i can give that a go. Though the CUDA 6.5 available is for Win 8 and 7.

I have windows 10 on the work station. I was thinking more futuristic on how these will work on our PC over time. But I’ll give it a go

Thanks

Hey @gideon! The newer CUDA versions should be backwards compatible, so you can go with CUDA 10 or 11 :+1:

Try the Visual Studio 2019 project for CUDA 10.2 at https://www.lfd.uci.edu/~gohlke/download/SeparableConvolutionCUDA.zip

2 Likes

will do, and give feedback here. Thanks

Hello people. Thanks for your help. So, I tried two things:

  1. The precompiled version for windows 7 (the one provided by @StephanPreibisch) with the vcredist_x64.exe. Ran it on windows 10 with CUDA 10.2 and 11.2. It appeared to engage the GPU, but could not load the view to register. I have attached the error log for this (named Log_1).

  2. Also as suggested by @cgohlke, I tried Visio 2019 and CUDA 10.2 with the library here: https://www.lfd.uci.edu/~gohlke/download/SeparableConvolutionCUDA.zip

It appeared to aslo engage the GPU but couldn’t load the views to be registered. Please see the error log for this (named Log_2)
Please I will appreciate if anyone can help with ideas on how to move forward on this?
@schmiedc @skalarproduktraum @StephanPreibisch @cgohlke @haesleinhuepf @hoerldavid

Thank you.
Log_1.txt (2.0 KB) Log_2.txt (2.1 KB)