[FFmpeg-devel] Fw: [foms] Paper submissions to LCA

[Carl Eugen Hoyos]

Jason Garrett-Glaser <darkshikari <at> gmail.com> writes:

> > the failing functions for icc 11.x 64 bit with optimisations are:
> > h263.c: ? ?h263_pred_motion()
> > wmv2dec.c: wmv2_pred_motion()
> > snow.c: ? ?pred_mv()
> > h264.c: ? ?pred_motion()
> > vc1dec.c: vc1_pred_dc()

I should add that vc1dec is different from the others, because several functions
fail and one of them also fails for icc 10.1.

> > Is there some pattern visible?
> 
> All except the last use mid_pred().

I was more thinking about the similar semantics, however, using the other
implementation for mid_pred (without attributes) did not change anything.
And in wmv2_pred_motion(), the problem seems to be the calculation of mot_val,
A[], B[] and C[] (according to old printf tests of mine) which does not use
mid_pred().

Carl Eugen