[PATCH] ASM-optimized byteswap functions

--TYecfFk8j8mZq+dy
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

Hi,

this stuff has been sitting in the MPlayer tree forever.

Please review / apply.

Diego

--TYecfFk8j8mZq+dy
Content-Type: text/plain; charset=us-ascii
Content-Disposition: attachment; filename="libdvdread_asm.patch"

Index: dvdread/bswap.h
===================================================================
RCS file: /cvsroot/ogle/libdvdread/dvdread/bswap.h,v
retrieving revision 1.13
diff -u -r1.13 bswap.h
--- dvdread/bswap.h	19 Sep 2005 14:14:34 -0000	1.13
+++ dvdread/bswap.h	18 Jan 2006 23:54:49 -0000
@@ -76,6 +76,44 @@
 #define B2N_32(x) x = be32toh(x)
 #define B2N_64(x) x = be64toh(x)
 
+#elif defined(ARCH_X86)
+inline static unsigned short bswap_16(unsigned short x)
+{
+  __asm("xchgb %b0,%h0" :
+        "=q" (x)        :
+        "0" (x));
+    return x;
+}
+#define B2N_16(x) x = bswap_16(x)
+
+inline static unsigned int bswap_32(unsigned int x)
+{
+ __asm(
+#if __CPU__ > 386
+      "bswap   %0":
+      "=r" (x)     :
+#else
+      "xchgb   %b0,%h0\n"
+      " rorl    $16,%0\n"
+      " xchgb   %b0,%h0":
+      "=q" (x)          :
+#endif
+      "0" (x));
+  return x;
+}
+#define B2N_32(x) x = bswap_32(x)
+
+inline static unsigned long long int bswap_64(unsigned long long int x)
+{
+  register union { __extension__ uint64_t __ll;
+          uint32_t __l[2]; } __x;
+  asm("xchgl    %0,%1":
+      "=r"(__x.__l[0]),"=r"(__x.__l[1]):
+      "0"(bswap_32((unsigned long)x)),"1"(bswap_32((unsigned long)(x>>32))));
+  return __x.__ll;
+}
+#define B2N_64(x) x = bswap_64(x)
+
 /* This is a slow but portable implementation, it has multiple evaluation 
  * problems so beware.
  * Old FreeBSD's and Solaris don't have <byteswap.h> or any other such 

--TYecfFk8j8mZq+dy--