On Wednesday, 3 August 2016 at 20:53:59 UTC, Ilya Yaroshenko wrote:
Hi all,

There are two first [1] benchmarks for upcoming ndslice.algorithm [2]. Recent LDC alpha based on LLVM 3.8 and recent Mir v0.16.0-alpha3 are required. @fasmath syntax may be changed a little bit and will be simplified anyway.

Dot Product:

       ndReduce vectorized = 3 ms, 314 μs
                  ndReduce = 14 ms, 767 μs

**That's** the difference with or without fastmath??

(awesome work of course!)


Reply via email to