Quoting Diego Biurrun (2016-08-01 13:48:47) > On Sat, Jul 30, 2016 at 10:02:47PM +0200, Anton Khirnov wrote: > > --- a/libavcodec/x86/vp9dsp_init.c > > +++ b/libavcodec/x86/vp9dsp_init.c > > @@ -95,7 +98,9 @@ ff_ ## avg ## _8tap_1d_ ## dir ## _ ## sz ## _ ## > > opt(uint8_t *dst, \ > > mc_rep_func(put, sz, hsz, v, ssse3); \ > > mc_rep_func(avg, sz, hsz, v, ssse3) > > > > +#if ARCH_X86_32 > > mc_rep_funcs(16, 8); > > +#endif > > mc_rep_funcs(32, 16); > > mc_rep_funcs(64, 32); > > This hunk is odd.
The macro mc_rep_func emulates a function for blocksize N with two calls to the function for blocksize N/2. Since the new function is 64bit only, this emulation needs to be done only for 32bit. -- Anton Khirnov _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
