On Wed, Mar 5, 2014 at 6:11 PM, Anton Khirnov <[email protected]> wrote: > > On Wed, 05 Mar 2014 10:59:20 -0500, Justin Ruggles <[email protected]> > wrote: >> On 03/05/2014 09:25 AM, Anton Khirnov wrote: >> > On Wed, 05 Mar 2014 08:56:52 -0500, Justin Ruggles >> > <[email protected]> wrote: >> >> On 03/05/2014 04:43 AM, Luca Barbato wrote: >> >>> diff --git a/libavfilter/af_volume.c b/libavfilter/af_volume.c >> >>> index 12d496e..2cdfe23 100644 >> >>> --- a/libavfilter/af_volume.c >> >>> +++ b/libavfilter/af_volume.c >> >>> @@ -262,6 +262,7 @@ static int filter_frame(AVFilterLink *inlink, >> >>> AVFrame *buf) >> >>> vol->scale_samples(out_buf->extended_data[p], >> >>> buf->extended_data[p], >> >>> plane_samples, >> >>> vol->volume_i); >> >>> + emms_c(); >> >>> } >> >>> } else if (av_get_packed_sample_fmt(vol->sample_fmt) == >> >>> AV_SAMPLE_FMT_FLT) { >> >>> for (p = 0; p < vol->planes; p++) { >> >> Neither of the dsp functions used in af_volume utilize MMX registers. >> >> Adding emms_c() is completely pointless. >> >> >> > That requires the caller to know this. >> > >> > And we might add such functions in the future (unlikely, but still >> > possible) >> > >> >> What caller are you referring to, and why would they have to know >> anything? Littering the code with unnecessary emms seems pretty silly to >> me. If someone happens to add more MMX/3DNow functions (they better have >> a very good reason...) it's up to them to add the required emms either >> in their function or in the appropriate places where that function is used. >> > > I mean the caller of float_dsp does not know what kind of functions are used > in > it. Or we should document it that functions requiring emms are not used and > ban them forever. >
Is it very likely to get MMX code written these days, and not at least SSE/SSE2? Maybe put a notice in float_dsp to warn people. :p - Hendrik _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
