On Mon, Jan 30, 2012 at 12:24:50AM +0100, Christophe Gisquet wrote: > 2012/1/16 Christophe Gisquet <[email protected]>: > > And the last point I had forgotten, replacing RET by REP_RET > > Here's a correct patch file (hopefully).
I'll queue it tomorrow, I need sleep now. Some comments/nits below. > From 3ce1d044ec27d185131c0213bb704347409dd4c3 Mon Sep 17 00:00:00 2001 > From: Christophe GISQUET <[email protected]> > Date: Thu, 12 Jan 2012 00:11:15 +0100 > Subject: [PATCH 1/4] rv34: x86 SIMD for biweight > > Provide MMX, SSE2 and SSSE3 versions, with a fast-path when the weights are > multiple of 512 (which is often the case when the values roundup nicely). multipleS / a multiple > --- /dev/null > +++ b/libavcodec/x86/rv40dsp_yasm.asm > @@ -0,0 +1,207 @@ > + > +align 16 > +shift_round: times 8 dw 1<<(16-6) > +cextern pw_16 I'm not familiar with yasm syntax, but can we have some spaces around operators here (and in some other places) or will that break? > + ; The weights are FP0.14 notation of fractions depending on pts. > + ; For timebases without rounding error (ie PAL), the fractions can i.e. > + ; be simplified, and several operations can be avoided. > + ; Therefore, we check here whether they are multiple of 2^9 for > + ; those simplifications to occur. > + > + ; Weights are multiple of 512, which allows some shortcuts See my comment from above about multiple(s). Diego _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
