[MPlayer-cvslog] r33712 - trunk/configure

[Reimar Döffinger]

On Thu, Jun 23, 2011 at 12:50:28PM +0200, Diego Biurrun wrote:
> On Thu, Jun 23, 2011 at 08:17:27AM +0200, Reimar Döffinger wrote:
> > On 22 Jun 2011, at 23:35, Diego Biurrun <diego at biurrun.de> wrote:
> > > On Wed, Jun 22, 2011 at 09:02:21PM +0200, reimar wrote:
> > >> 
> > >> Log:
> > >> Do not force SSE disabled for gcc 2, it seems to compile
> > >> and work fine.
> > > 
> > > You're seriously spending time on supporting gcc 2.95?
> > > We should just drop support, 2.95 was outdated a decade ago...
> > 
> > So far the only significant time spent was fixing actual bugs it uncovered.
> > While the SSE issues should mostly be aligned stack issue, the reason
> > I came over it is that disabling SSE is broken: it does not disable SSE2 and it
> > does disable MMX2 on some systems.
> 
> Elaborate...

I guess only the MMX2 part is unclear?
The below patch is my proposed fix, the problem is that if _sse is forced to no
it will stay that, and thus the "SSE support implies MMX2 support" logic breaks.
Index: configure
===================================================================
--- configure   (revision 33713)
+++ configure   (working copy)
@@ -1777,14 +1777,13 @@
 
   pparam=$(echo $exts | sed -e s/k6_mtrr/mtrr/ -e s/cyrix_arr/mtrr/ -e s/centaur_mcr/mtrr/ \
                             -e s/xmm/sse/ -e s/kni/sse/)
+  # SSE implies MMX2, but not all SSE processors report the mmxext CPU flag.
+  pparam=$(echo $pparam | sed -e s/sse/sse mmxext/)
 
   for ext in $pparam ; do
     eval test \"\$_$ext\" = auto 2>/dev/null && eval _$ext=kernel_check
   done
 
-  # SSE implies MMX2, but not all SSE processors report the mmxext CPU flag.
-  test $_sse = kernel_check && _mmxext=kernel_check
-
   echocheck "CPU vendor"
   echores "$pvendor ($pfamily:$pmodel:$pstepping)"