Nsight Compute Command Line Profiler for Optix kernels

[michael.blazej]

I’d like to identify which commands in closest_hit are taking the longest but ncu only provides statistics on raygen no matter what command line options I provide. From the release notes, it looks like ncu doesn’t provide any further detail past the raygen kernel. Is this true and will it be changing in the future?

Thank you.

[droettger]

I’m only using the Nsight Compute UI version but since the CLI version dumps a report you can open with the UI version, the rest should be the same.

it looks like ncu doesn’t provide any further detail past the raygen kernel. Is this true and will it be changing in the future?

That’s not what this means.

The ray generation function inside the OptiX device code is the entry point of the kernel. There can only be one raygen program inside the shader binding table and that is what Nsight Compute just uses to assign a name to the kernel inside the API trace so that you can identify it when profiling.

All OptiX functions you provided for any program domain are just function names inside that profiled kernel and you will be able to see them all inside Nsight Compute when looking at the list of functions inside the source view, described here:
https://docs.nvidia.com/nsight-compute/NsightCompute/index.html#navigation

This all implies you compiled your fully optimized OptiX release device code with line information and used the proper OptixModuleCompileOptions debugLevel (moderate).
https://raytracing-docs.nvidia.com/optix8/guide/index.html#program_pipeline_creation#program-input