Should I use the short vector types for C++ AMP?

Question

Hello, is there any performance reason to use short vector types (as outlined in http://blogs.msdn.com/b/nativeconcurrency/archive/2012/04/03/short-vector-types-in-c-amp.aspx)
in my C++ AMP kernels that will be confined to execution only on the GPU? Or is this simply to clean up the syntax a little?

Answer 1

Hello LKeene.
This would depend on the GPU architectures that you are running on. For example, older AMD hardware (pre-GCN i.e. 5xxx and 6xxx series) was VLIW at the ALU level, so it might have benefited from the explicit vector usage as the compiler wouldn't have had
to work as hard to find elements to pack. That being said, more modern hardware would not directly benefit. If you think that the syntax ends up being cleaner, go for it (that's still a win!). I hope this helps. Cheers!