CODEWITHSUNDEEP
parallel-processingcuda
vydesaster
vydesaster

Reputation: 253

Parallelisation settings for cuSolver functions on host and device

I am using QR and LU decomposition and solution functions for sparse matrices from cuSolver CUDA library. According to the documentation, the LU function is only available on the host path whereas QR functions are also availabe on the device.
I was wondering if there is any setting or similar to control the parallelisation settings for host and device functions. How is the number of threads controlled for both types of functions?

Upvotes: 1

Views: 94

Answers (1)

Robert Crovella
Robert Crovella

Reputation: 151859

There aren't any parallelization settings.

The number of threads cannot be controlled directly, and there is no public description of how the function works at the level you are asking.

The library is opaque. The number of threads associated with any kernel call that is launched by the library call will be determined by a heuristic that the library uses which is not described.

Upvotes: 3

Related Questions