Where is cusolverDnXsytrf?

Accelerated Computing GPU-Accelerated Libraries
,
1

Where is cusolverDnXsytrf()? The docs for cusolverDnXsytrs() refer to the factorization from cusolverDnXsytrf() however this function does not seem to exist in the docs or cusolverDn.h

1 Like
2

That Generic API doesn’t exist yet. You have to use the legacy API cusolverdn<t>sytrf

3

but… the legacy version of sytrf uses a 32-bit pivot vector, whereas the generic version of sytrs wants a 64-bit pivot vector. so it’s difficult to chain the two together.

2 Likes
4

The documentation for cusolverDnXsytrs explicitly states:

A contains the factorization from cusolverDnXsytrf()

However cusolver64_12.dll (installed from CUDA toolkit 13.0) still does not export that function. I’ve looked for it via dumpbin /exports to no avail. Very disappointing that this is still an issue years later with the documentation being misleading!