--- xvid.c	2002/07/18 13:47:46	1.31
+++ xvid.c	2002/09/04 22:07:07	1.34
@@ -3,6 +3,8 @@
  *  XVID MPEG-4 VIDEO CODEC
  *  - Native API implementation  -
  *
+ *  Copyright(C) 2001-2002 Peter Ross <pross@cs.rmit.edu.au>
+ *
  *  This program is an implementation of a part of one or more MPEG-4
  *  Video tools as specified in ISO/IEC 14496-2 standard.  Those intending
  *  to use this software module in hardware or software products are
@@ -28,19 +30,6 @@
  *
  ****************************************************************************/
 
-/*****************************************************************************
- *
- *  History
- *
- *	- 23.06.2002	added XVID_CPU_CHKONLY
- *  - 17.03.2002	Added interpolate8x8_halfpel_hv_xmm
- *  - 22.12.2001  API change: added xvid_init() - Isibaar
- *  - 16.12.2001	inital version; (c)2001 peter ross <pross@cs.rmit.edu.au>
- *
- *  $Id: xvid.c,v 1.31 2002/07/18 13:47:46 suxen_drol Exp $
- *
- ****************************************************************************/
-
 #include "xvid.h"
 #include "decoder.h"
 #include "encoder.h"
@@ -81,12 +70,12 @@
 
 
 /*
-calls the funcptr, and returns whether SIGILL (illegal instruction) was signalled
-return values:
--1 : could not determine
-0  : SIGILL was *not* signalled
-1  : SIGILL was signalled
-*/
+ * Calls the funcptr, and returns whether SIGILL (illegal instruction) was signalled
+ * Return values:
+ * -1 : could not determine
+ * 0  : SIGILL was *not* signalled
+ * 1  : SIGILL was signalled
+ */
 
 int
 sigill_check(void (*func)())
@@ -245,9 +234,11 @@
 	/* Functions used in motion estimation algorithms */
 	calc_cbp = calc_cbp_c;
 	sad16    = sad16_c;
-	sad16bi  = sad16bi_c;
 	sad8     = sad8_c;
+	sad16bi  = sad16bi_c;
+	sad8bi   = sad8bi_c;
 	dev16    = dev16_c;
+	
 	Halfpel8_Refine = Halfpel8_Refine_c;
 
 #ifdef ARCH_X86
@@ -302,10 +293,21 @@
 		calc_cbp = calc_cbp_mmx;
 		sad16    = sad16_mmx;
 		sad8     = sad8_mmx;
+		sad16bi = sad16bi_mmx;
+		sad8bi  = sad8bi_mmx;
 		dev16    = dev16_mmx;
 
 	}
 
+	/* these 3dnow functions are faster than mmx, but slower than xmm. */
+	if ((cpu_flags & XVID_CPU_3DNOW) > 0) {
+
+		/* ME functions */
+		sad16bi = sad16bi_3dn;
+		sad8bi  = sad8bi_3dn;
+	}
+
+
 	if ((cpu_flags & XVID_CPU_MMXEXT) > 0) {
 
 		/* Inverse DCT */
@@ -328,8 +330,9 @@
 
 		/* ME functions */
 		sad16 = sad16_xmm;
-		sad16bi = sad16bi_xmm;
 		sad8  = sad8_xmm;
+		sad16bi = sad16bi_xmm;
+		sad8bi  = sad8bi_xmm;
 		dev16 = dev16_xmm;
 
 	}