On Tuesday 20 January 2009 10:53:53 Cactus wrote: > In configuring the Windows build of the experimental branch in SVN, I > have to decide what to do about the new __gmpn_divide_by_3 and > __gmpn_divide_by_3c assembler code (I offer both on Windows). > > In GMP the function __gmpn_divide_by_3c is used as the primitive and > there is both a function and a macro based on this for > __gmpn_divide_by_3 > > Our new assembler code on Windows is a lot faster and is fully > compatible when the division is exact. But it returns a different > division result if the division is not exact. Since the Toom3 code is > always calls for an an exact division we can use our new code > internally without any problems so the remaining question is that of > whether or not we should expose our new function to MPIR users and, if > so, how. > > If we replace the GMP functions, we offer much faster code but the > results will not match GMP when the division is not exact. Does this > matter? Should anyone be relying on the GMP result in when the > division is inexact anyway? Is it sufficient to simply document the > fact that people using MPIR should not rely on this function to give > the same results as GMP if the division is not exact? > > Or should we simply define new functions alongside the existing ones? > > Or should we not expose the new code externally? > > I don't think we want different results for MPIR on Windows so we need > a common policy on this. > > What do people think is the right approach here? > > Brian >
To answer the more general question I like the way GCC does things. New features are introduced , a few releases pass , old features are obsoleted and a few releases down the line are removed. > --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "mpir-devel" group. To post to this group, send email to mpir-devel@googlegroups.com To unsubscribe from this group, send email to mpir-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/mpir-devel?hl=en -~----------~----~----~----~------~----~------~--~---
