Hi, On Tue, May 15, 2012 at 6:20 AM, Christophe Gisquet <[email protected]> wrote: > 2012/5/11 Ronald S. Bultje <[email protected]>: >> I agree a cpuflags-based SPLATW implementation would be wonderful. Do >> you have time to go through all current users of SPLATW to ensure we >> don't break anything (i.e. make it use mmx code where mmx2 was >> intended because it doesn't use cpuflags yet)? If not, I can do that >> over the weekend, hopefully. > > The issue is that the full intent of SPLATW can't be handled by MMX. > The most common use, and the easiest to deal with, is the case where > 3rd argument is 0. I see 2 ways (surely there are more to deal with > that: > 1) Is it possible, and is it desirable to have SPLATW macro cleanly > cause an error when called from code invoking it in a MMX context? > Then most of the code could stay in, at the cost of maybe a semantic > weirdness for MMX
mmx version: movaifnidn %1, %2 %if %3 & 2 punpckhwd %1, %1 %else punpcklwd %1, %1 %endif %if %3 & 1 punpckhdq %1, %1 %else punpckldq %1, %1 %endif I agree this is ugly, but it's the best I can think of. Ronald _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
