CUBLAS: nrm2 support for StridedCuArray with length requiring Int64
tipfom opened this issue · comments
Currently, the CUDA.CUBLAS.nrm2 wrapper function fails, when the length of the array is greater than supported by Int32.
This also causes the LinearAlgebra.norm implementation to fail causing an error with the following signature: "InexactError: trunc(Int32, 4294967296)".
For me, appending the nrm2 wrapper definition to include the 64 bit variants of the nrm2 functions, namely cublasDnrm2_v2_64, cublasSnrm2_v2_64, cublasDznrm2_v2_64, cublasScnrm2_v2_64, presents the best solution.
The PR #2269 provides a suggested fix :)