Recently I have been in the mix of developing a almost complete FORTRAN wrapper for the cuda runtime api host code using the FORTRAN 2003 ISO C bind intrinsics. But as some of you may know that the pointers in FORTRAN are just for memory addressing , you cannot perform arithmetic operations on them for dynamic addressing. Hence, Now I am stuck on how to replicate the cuda streams feature equivalent to the one shown in programming guide 2.3 page 34 (section 188.8.131.52) . Does anyone know how to go about replicating that via a wrapper in FORTRAN ?
I looked at the pinned memory example on Nvidia’s website; it doesn’t tell anything about streams as it just calls CUBLAS.
Anyone from Nvidia/here at forums help me out here please ? I plan to release this whole wrapper as a library for all but am stuck rite now :( because of this problem…
Hi, I don’t have an immediate answer. I am interested in the subject however, and it will be interesting to see what answers you get. I am sure there is someone at NVIDIA who can help, but… What Fortran compiler are you using? NVIDIA may have abdicated their responsibility in the Fortran area to PGI. It might be worth contacting them.
I have already developed library with the most frequently used CUDA host runtime function and data types. But I need this stream capability for a project am working on. I use the Intel Fortran Compiler Ver 11.1, its very good as far as my experience goes.
Thanks, I will contact PGI but I dont see why would they help me given that they have a product which does the same but it doesnt hurt to ask External Image
… but whom to ask ?
Also please someone at NVidia if can help me out here ? or anyone else has bright ideas ?