Sycl-crowd-simulation

oneAPI for NVIDIA GPUs
1

I am trying to build this project and run it on an Nvidia RTX 4000 GPU GitHub - codeplaysoftware/sycl-crowd-simulation: A crowd simulation algorithm implemented using C++ and SYCL.

  1. Please note that the code supports only double, so will not work out of the box for Intel dGPUs that does not support double…

All you have to do is edit external/Heatmap.cpp and change the double to float…
(this is how it should look like (lines 32-33)
float scaledH = sycl::fmod(h / 60.0f, 6.0f);
float x = c * (1 - sycl::fabs(sycl::fmod(scaledH, 2.0f) - 1));

  1. when targeting nVidia RTX 4000, I’m getting segmentation fault…
2

Hi Guy,
It sounds like you need to check a few things.
Firstly, are you compiling for the right target?

Secondly, are you using the correct device selector for your machine? You will need to specifically choose the Nvidia device

If these are looking correct you’ll need to get a backtrace to try and get some clues on where it is failing as a segfault could be many things.
You can do this using the cuda-gdb debugger using the bt command to get the backtrace information