The restriction that one needs to use generic functions (and as such, can't use explicit BLAS calls) isn't clearly stated in ForwardDiff.jl's documentation; I just made an issue to track documenting it <https://github.com/JuliaDiff/ForwardDiff.jl/issues/72>. Feel free to open up an issue on ForwardDiff.jl's repository if you run into further errors or unexpected performance bugs.
Best, Jarrett On Tuesday, December 1, 2015 at 8:07:28 AM UTC-5, Patrick Kofod Mogensen wrote: > > I know help is much easier to provide with a MWE, but it must be the > explicit BLAS-calls, I'm sure. I will re-write the function using Julia > functions only. If it doesn't work, I'll have to try to reduce my mapping > to a simpler version I can show here. Thanks for taking your time so far. > > On Tuesday, December 1, 2015 at 1:23:16 PM UTC+1, Kristoffer Carlsson > wrote: >> >> Do you have explicit blas calls? That won't work. Instead you should just >> use the more high level functions (either * or the Ax_mul_Bx versions) so >> that Julias multiple dispatch can do its thing and run the generic versions >> (julia implementations) of the functions when it is called with "graident >> numbers". >> >> Easiest would be if you could post a small runnable example of something >> that models what you actually want to do. >> >
