Hi,
I am trying to solve dense rectangular, linear systems A*x = b in batch.
I have seen it is possible to solve the non-batched version using a combination of “geqrf”, “ormqr” and “trsm”, as shown in the following example:
[url]https://docs.nvidia.com/cuda/cusolver/index.html#ormqr-example1[/url]
I have found batched versions of “geqrf” and “trsm”:
[url]https://docs.nvidia.com/cuda/cublas/index.html#cublas-lt-t-gt-geqrfbatched[/url]
and
[url]https://docs.nvidia.com/cuda/cublas/index.html#cublas-lt-t-gt-trsmbatched[/url]
respectively.
However, I am not able to find a batched “ormqr” function and can therefore not complete the solve.
i Is there another way to use the batched “geqrf” and “trsm” to complete the solve? (is there a reason that there’s no batched “ormqr”?)[/i]
As another possibility, I found that batched “gels” can solve an overdetermined linear system:
[url]https://docs.nvidia.com/cuda/cublas/index.html#cublas-lt-t-gt-gelsBatched[/url]
However, it seems batched “gels” lacks the possibility to solve underdetermined systems for the time being.
i Does anyone know if this limitation will be removed sometime soon?
(3) Does anyone know if there is any other possible way to solve a dense rectangular, linear system in batch?[/i]
[Note: I found a post dealing with a similar question, which is about a year old:
[url]https://devtalk.nvidia.com/default/topic/1021856/?comment=5200231[/url]
However, the answer there only points to non-batched solutions.
Therefore, I thought I should try my luck and ask the question again.]
Best regards,
David