On Mon, 13 Jun 2016, Diego Biurrun wrote:
On Mon, Jun 13, 2016 at 12:08:15AM +0300, Martin Storsjö wrote:On Sat, 11 Jun 2016, Martin Storsjö wrote: > --- /dev/null > +++ b/tests/checkasm/h264dsp.c > @@ -0,0 +1,268 @@ > + for (dc = 0; dc <= 1; dc++) { > + void (*idct)(uint8_t *, int16_t *, int); > + switch ((sz << 1) | dc) { > + case (4 << 1) | 0: idct = h.h264_idct_add; break; > + case (4 << 1) | 1: idct = h.h264_idct_dc_add; break; > + case (8 << 1) | 0: idct = h.h264_idct8_add; break; > + case (8 << 1) | 1: idct = h.h264_idct8_dc_add; break; > + } > + if (check_func(idct, "h264_idct%d_add%s_%dbpp", sz, dc ? "_dc" : "", bit_depth)) { > + if (dc) { > + memset(subcoef0, 0, sz * sz * SIZEOF_COEF); > + subcoef0[0] = coef[0];Changed this locally into memcpy(subcoef0, coef, SIZEOF_COEF); in order to properly copy the dc component for the >8bpp tests.Can you send your updated patch? I'd like to play around with it.
It's just literally replacing the subcoef[0] = coef[0] with that line. If you want to test it, the last posted version will work just fine as well, although it's less correct.
// Martin _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
