Hello, dear developers and users!
My question is about float (single precision) type support in cusolverRf sublibrary of cusolver library.
First of all, i’m interested in batch versions, but it seems entire sublibrary lacks type-specific functions. There are no functions for float type (or complex), just functions for double input arrays and variables. Also there is no type identification in function names like for example in cusolverDN and cusolverSp sublibraries (S,D,C letters in function names). Does it mean that cusolverRF can be used only with doubles or it’s just temporal thing. If so, when we could expect to have support for other types (for me, float is primary interest).
Thank you in advance.
Hello!
I’v watched through cuda 9 rc documentation here
and discovered that still only double versions of cusolverRf functions exist. I kind of hoped that in new version this problem would be treated but it seems everything remains unchanged. Does anybody know what is the problem with making float versions of functions? Could they appear at least in some further cuda toolkits?
cusolverRF only supports double:
[url]cuSOLVER :: CUDA Toolkit Documentation
(there is no change in this respect in CUDA 9)
If you would like to see a feature enhancement in CUDA, the best suggestion is to file a bug at developer.nvidia.com
Use your registered developer credentials. Indicate clearly in the subject line that this is an enhancement request for cusolver.
Historically, just about any feature in the CUDA environment has been driven by customer requests. That does not mean that all customer requests will eventually be satisfied (at any given time, there are an order of magnitude more than could possibly be implemented), but it pretty much means that if a formal feature request is not recorded that desired enhancement is exceedingly unlikely to happen.
The tool to capture, collate, and prioritize customer requests is the bug database, where customers can file RFEs (requests for enhancements). As txbob recommended, file a bug report and prefix the bug report synopsis with “RFE:” to mark it as an enhancement request rather than a functional bug.