Hi, I had to learn the "hard way" that --opt=fast-math of ISPC does not imply "flush to zero" nor "denorm are zero".
FTZ and DAZ can lead to 10X or more performance boost when dealing with unfortunate data... I think these modes could go well with less accurate reciprocal etc. I had to wrap this into an extern "C" function and call it at the beginning of the ISPC function instead _MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON); It's kind of cumbersome as I have to drag along another object file just for that. It would be better if ISPC would set this automatically in the function prelude... What's your opinion on this? Or is there a way to specify FTZ/DAZ via ISPC? Thanks Diego -- You received this message because you are subscribed to the Google Groups "Intel SPMD Program Compiler Users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
