rohezal
Advanced Member
- Joined
- Oct 18, 2009
- Messages
- 1,712
Some people from intel vectorizing scene are posting on source forge vectorizing mailinglist. They are talking about how to auto detect parallel loops and build data structures which support vectorization on the fly. This would be very nice to have for the Pandora, if people would use intel compilers ( I think gcc is starting to support this too). It can give you a very big speed boost, since it uses SSE4 and multiple cores without much manual work. This isn't as fast as hand coded neon code, but for the average Joe coder, it can give you lots of speed without much knowledge. This could maybe mean less power consumption but I am not sure, how 1 SSE instruction performs against several normal FPU instructions.
Last edited by a moderator: