Hey!
I am trying to profile a basic OptiX7.7 program (with Cuda Toolkit 12.0, as per the Release Note requirements), using NSight Compute 2025.2, and cannot seem to get the source code side of things working. I was hoping to use this to identify which parts of my program are taking up the most time, and focus any optimisations accordingly.
I am following the instructions in the programming guide (https://raytracing-docs.nvidia.com/optix7/guide/index.html#program_pipeline_creation#7017):
- I am compiling a .ptx file using cmake, with
--generate-line-info --relocatable-device-code=true --use_fast_math. - This is then read during the setup of the OptiX application, in which I am setting the
OptixModuleCompileOptions.debugLeveltoOPTIX_COMPILE_DEBUG_LEVEL_MODERATE, andOptixModuleCompileOptions.optLeveltoOPTIX_COMPILE_OPTIMIZATION_LEVEL_3. - I am using Nsight Compute’s Profile activity, with
OPTIX_FORCE_DEPRECATED_LAUNCHER=1in the Environment field, with Import Source set to Yes, and the location of the .cu file (and .ptx file, just to be safe) provided as the Source Folders.
I am able to see the NVIDIA internal (optixLaunch) call’s, and their duration etc., but when I go to the Source tab, I just see “No source available.”.
I have tried a number of other options, with no success, including:
- A variety of different
optLevelanddebugLevelconfigurations - Using a .optixir file instead, with extra debug info using
-G
Any suggestions would be much appreciated!
Thanks!