2015-10-08 8:22 GMT+02:00 Christophe Gisquet <christophe.gisq...@gmail.com>: > +SIC_WRAP(12) > + > +static const struct algo idct_tab12[] = { > + { "SIMPLE12-C", simple_idct12_put_wrap, FF_IDCT_PERM_NONE }, > + { 0 } > +};
Strange results are generated for it, and I don't know why (the C version shouldn't overflow?). If I replace the current coeffs by their x86 coeffs * 2 (like it is done for the 10 bits version to have a bit-exact version), the issue no longer appears. I haven't verified if the compiler plays a trick here by rescaling coefficients because there's an undefined behaviour somewhere in the C. I also don't know if there are codecs, beside jpeg, that would use this 12bits version, and if there's an associated fate test: fate-video, fate-vcodec and any test having jp in its name pass. -- Christophe _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel