how to use cudaLaunch?

short version first:
cudaLaunch lacks any documentation, and i don’t have any clue of what to pass to it…

long version:
i tried to wrap the cudalaunch interface into a C++ object, to allow high-level launches of kernels…
my idea was to use strings as kernel names to have an higher flexibilty, but i couldn’t find any example of the string’s format…
-passing name didn’t work
-passing the name + “__entry” didn’t work
-passing the extended bin name didn’t work…

I found anyway many examples for cuLaunchGrid, but i don’t think it’s good habit to mix “high” and driver-level APIs…

Thanks in advance for any clues! :thumbup:


what is the content of the cudaError_t after the launch failed? It can be one of

cudaSuccess, cudaErrorInvalidDeviceFunction, cudaErrorInvalidConfiguration



Ah right i forgot to mention it… it is cudaErrorInvalidDeviceFunction.

The function works fine if called with the <<<>>> syntax.

Any ideas? :blink:

I bet it’s name mangling! Add extern “C” around your kernel prototype and watch it magically work…

It did magically work :thumbup:

I already got it to work without it, but looking in the .cubin and using names like “_Z21rasterizePointsKernelP6uchar35uint2P6float3Pf” wasn’t the same thing.