Closed dicepd closed 6 years ago
Ok i'm correcting the function, it's ok now. I push the code tonight after my job
I found a realy cool link https://web.archive.org/web/20161219143458/http://http.developer.nvidia.com/Cg/index_stdlib.html for implement Step, Saturate, SmoothStep, FaceForward....
It seems that fastmath can reduce precision, should we reflect this in the tests by ifdef USE_FASTMATH and put two epsilon values to pass at the available precision for the mode. It will at least be some sort of reference/documentation to where precision is reduced through fastmath.
Yes Fastmath reduce precision we do approximation. A correct minimum precision for me wil be 1e-4 with a range error max of 0.0003. Put two epsilon values it will a good thing for having a correct evaluation.
As above