AnsweredAssumed Answered

Underlying rounding in matmmlt_fr16

Question asked by DanaGood on Aug 11, 2009

Question about the rounding process in matmmlt_fr16.  What type of rounding is done in this routine when 32-bit negative numbers must be rounded to 16 (toward zero, or away from zero)?  If rounding is done toward zero, is there a way to modify this, either through choice of libraries or through compiler options?

 

Thanks,

 

Dana

Outcomes