[xiph-commits] r16031 - branches/theora-thusnelda/lib/enc/x86

[giles at svn.xiph.org]

Author: giles
Date: 2009-05-25 17:37:53 -0700 (Mon, 25 May 2009)
New Revision: 16031

Modified:
   branches/theora-thusnelda/lib/enc/x86/mmxencfrag.c
Log:
Use a single-digit local label in oc_int_frag_satd_thresh_mmxext instead
of generating a unique multidigit one to maintain compatibility with the
GNU assembler version shipped on MacOS 10.4.

In theory the unique multidigit label is safer because it can't collide 
with any compiler-generated local labels which happen to span the asm
block. We don't believe gcc currently does this.

The newer multidigit integer labels are supported in MacOS 10.5
(XCode 3) and later.


Modified: branches/theora-thusnelda/lib/enc/x86/mmxencfrag.c
===================================================================
--- branches/theora-thusnelda/lib/enc/x86/mmxencfrag.c	2009-05-25 08:58:39 UTC (rev 16030)
+++ branches/theora-thusnelda/lib/enc/x86/mmxencfrag.c	2009-05-26 00:37:53 UTC (rev 16031)
@@ -501,7 +501,7 @@
     "movq 0x40(%[buf]),%%mm0\n\t"
     "cmp %[ret2],%[ret]\n\t"
     "movq 0x48(%[buf]),%%mm4\n\t"
-    "jae %=1f\n\t"
+    "jae 1f\n\t"
     OC_HADAMARD_ABS_ACCUM_8x4("0x68","0x78")
     "pmaddwd %%mm7,%%mm0\n\t"
     /*There isn't much to stick in here to hide the latency this time, but the
@@ -514,7 +514,7 @@
     "movd %%mm4,%[ret2]\n\t"
     "lea (%[ret],%[ret2],2),%[ret]\n\t"
     ".p2align 4,,15\n\t"
-    "%=1:\n\t"
+    "1:\n\t"
     /*Although it looks like we're using 7 registers here, gcc can alias %[ret]
        and %[ret2] with some of the inputs, since for once we don't write to
        them until after we're done using everything but %[buf] (which is also