[FFmpeg-devel] [PATCH] fix compiling VIS assembly on SPARC
Tue Sep 18 02:23:32 CEST 2007
On Mon, Sep 17, 2007 at 12:18:00PM +0200, Diego Biurrun wrote:
> On Sun, Sep 16, 2007 at 12:14:01AM -0500, Derek E. Lewis wrote:
> > On Fri, 14 Sep 2007, Balatoni Denes wrote:
> > > Thanks Derek, that did clear things up for me - because I had no idea what
> > > these obsolote v8plus registers were.
> > Actually, part of what I said was wrong, but the concept is still the
> > same: by default, GCC (as of 3.x) builds for SPARCv8. 'v8plus' is simply
> > 32-bit code, but uses extra registers that SPARCv9 provides over SPARCv8.
> > This means a binary built for 'v8plus' requires an UltraSPARC or another
> > SPARCv9-compatible system to execute. It looks like GCC as of version 4.x
> > defaults to 'v8plus'. I think it was gcc-2.x that built for SPARCv7 by
> > default. It is probably to our advantage not to assume what architecture
> > gcc will default to when compiling, and to define some level of SPARC
> > compliance we expect. A reasonable level would by SPARCv8, IMO, as there
> > are still some users out there using sun4m systems, which are SPARCv8. Of
> > course, this has no relevance on your inline VIS code.
> It would be nice if somebody could provide us some VIS detection code
> for configure. It should be trivial to do, but it's better to have a
> machine for testing it on...
Somebody with some SPARC-fu please fix the attached patch. I just took
the 23rd assembler line from libavcodec/sparc/simple_idct_vis.c, it
looked as good as any of them ...
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 1183 bytes
Desc: not available
More information about the ffmpeg-devel