[FFmpeg-devel] [PATCH] fix for roundup issue 2127

[Reimar Döffinger]

On 30 dec 2010, at 19:14, Daniel Kang <daniel.d.kang at gmail.com> wrote:

> I am a Google Code-In student, and as part of a task, I have written a
> fix for: https://roundup.ffmpeg.org/issue2127. The patch compiles with
> 32/64-bit and --disable-optimizations. It passes "make fate" with and
> without --disable-optimizations.
> 
> I have ran several timings with and without my patch. The
> START/STOP_TIMER was placed around the actual "asm" block in the
> transpose4x4 function.
> With my patch: 21.32 +/- 0.04 clocks
> Without my patch: 21.58 +/- 0.04 clocks
> 
> Are there any comments on my patch? If not, can this be committed?

There must be something wrong, the #else code will never be compiled.
Also, the new code seems to require 6 registers, IIRC PIC + frame-pointer only leaves available 5.