--- xvid.c	2003/07/10 17:39:51	1.45.2.9
+++ xvid.c	2003/10/01 23:23:01	1.45.2.15
@@ -19,7 +19,7 @@
  *  along with this program ; if not, write to the Free Software
  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  *
- * $Id: xvid.c,v 1.45.2.9 2003/07/10 17:39:51 edgomez Exp $
+ * $Id: xvid.c,v 1.45.2.15 2003/10/01 23:23:01 edgomez Exp $
  *
  ****************************************************************************/
 
@@ -46,6 +46,7 @@
 #include "utils/emms.h"
 #include "utils/timer.h"
 #include "bitstream/mbcoding.h"
+#include "image/qpel.h"
 
 #if defined(_DEBUG)
 unsigned int xvid_debug = 0; /* xvid debug mask */
@@ -150,7 +151,7 @@
  * to the CPU features forced by the library client or autodetected (depending
  * on the XVID_CPU_FORCE flag). It also initializes vlc coding tables and all
  * image colorspace transformation tables.
- * 
+ *
  * Returned value : XVID_ERR_OK
  *                  + API_VERSION in the input XVID_INIT_PARAM structure
  *                  + core build  "   "    "       "               "
@@ -158,12 +159,12 @@
  ****************************************************************************/
 
 
-static 
+static
 int xvid_gbl_init(xvid_gbl_init_t * init)
 {
 	unsigned int cpu_flags;
 
-	if (XVID_MAJOR(init->version) != 1) /* v1.x.x */
+	if (XVID_VERSION_MAJOR(init->version) != 1) /* v1.x.x */
 		return XVID_ERR_VERSION;
 
 	cpu_flags = (init->cpu_flags & XVID_CPU_FORCE) ? init->cpu_flags : detect_cpu_flags();
@@ -182,6 +183,11 @@
 	/* Restore FPU context : emms_c is a nop functions */
 	emms = emms_c;
 
+	/* Qpel stuff */
+	xvid_QP_Funcs = &xvid_QP_Funcs_C;
+	xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_C;
+	xvid_Init_QP();
+
 	/* Quantization functions */
 	quant_intra   = quant_intra_c;
 	dequant_intra = dequant_intra_c;
@@ -265,7 +271,7 @@
 	yv12_to_rgba    = yv12_to_rgba_c;
 	yv12_to_yuyv    = yv12_to_yuyv_c;
 	yv12_to_uyvy    = yv12_to_uyvy_c;
- 
+
 	yv12_to_rgb555i = yv12_to_rgb555i_c;
 	yv12_to_rgb565i = yv12_to_rgb565i_c;
 	yv12_to_bgri    = yv12_to_bgri_c;
@@ -283,7 +289,7 @@
 	sad8bi   = sad8bi_c;
 	dev16    = dev16_c;
 	sad16v	 = sad16v_c;
-	
+
 /*	Halfpel8_Refine = Halfpel8_Refine_c; */
 
 #if defined(ARCH_IS_IA32)
@@ -307,6 +313,10 @@
 		fdct = fdct_mmx;
 		idct = idct_mmx;
 
+		/* Qpel stuff */
+		xvid_QP_Funcs = &xvid_QP_Funcs_mmx;
+		xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_mmx;
+
 		/* Quantization related functions */
 		quant_intra   = quant_intra_mmx;
 		dequant_intra = dequant_intra_mmx;
@@ -360,7 +370,7 @@
 		yv12_to_bgra  = yv12_to_bgra_mmx;
 		yv12_to_yuyv  = yv12_to_yuyv_mmx;
 		yv12_to_uyvy  = yv12_to_uyvy_mmx;
-		
+
 		yv12_to_yuyvi = yv12_to_yuyvi_mmx;
 		yv12_to_uyvyi = yv12_to_uyvyi_mmx;
 
@@ -470,7 +480,8 @@
 		interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne;
 	}
 
-
+#if defined(EXPERIMENTAL_SSE2_CODE) /* mark the whole SSE2 stuff as experimental. At least on
+									   my P4, it crashes... */
 	if ((cpu_flags & XVID_CPU_SSE2)) {
 
 		calc_cbp = calc_cbp_sse2;
@@ -481,11 +492,9 @@
 		quant_inter   = quant_inter_sse2;
 		dequant_inter = dequant_inter_sse2;
 
-#if defined(EXPERIMENTAL_SSE2_CODE)
 		/* ME; slower than xmm */
 		sad16    = sad16_sse2;
 		dev16    = dev16_sse2;
-#endif
 		/* Forward and Inverse DCT */
 #if 0 /* Both function are known to be unprecise, better keep them deactivated */
 		idct  = idct_sse2;
@@ -493,6 +502,7 @@
 #endif
 	}
 #endif
+#endif
 
 #if defined(ARCH_IS_IA64)
 	if ((cpu_flags & XVID_CPU_ASM)) { /* use assembler routines? */
@@ -518,7 +528,7 @@
 	  transfer_16to8add = transfer_16to8add_ia64;
 	  transfer8x8_copy = transfer8x8_copy_ia64;
 	}
-#endif 
+#endif
 
 #if defined(ARCH_IS_PPC)
 	if ((cpu_flags & XVID_CPU_ASM))
@@ -549,7 +559,7 @@
 static int
 xvid_gbl_info(xvid_gbl_info_t * info)
 {
-	if (XVID_MAJOR(info->version) != 1) /* v1.x.x */
+	if (XVID_VERSION_MAJOR(info->version) != 1) /* v1.x.x */
 		return XVID_ERR_VERSION;
 
 	info->actual_version = XVID_VERSION;
@@ -575,7 +585,7 @@
 	int height2;
 	IMAGE img;
 
-	if (XVID_MAJOR(convert->version) != 1)   /* v1.x.x */
+	if (XVID_VERSION_MAJOR(convert->version) != 1)   /* v1.x.x */
 	      return XVID_ERR_VERSION;
 
 #if 0
@@ -590,7 +600,7 @@
 	{
 		case XVID_CSP_YV12 :
 			img.y = convert->input.plane[0];
-			img.v = (uint8_t*)convert->input.plane[0] + convert->input.stride[0]*height; 
+			img.v = (uint8_t*)convert->input.plane[0] + convert->input.stride[0]*height;
 			img.u = (uint8_t*)convert->input.plane[0] + convert->input.stride[0]*height + (convert->input.stride[0]/2)*height2;
 			image_output(&img, width, height, width,
 						(uint8_t**)convert->output.plane, convert->output.stride,
@@ -676,7 +686,7 @@
 	DECLARE_ALIGNED_MATRIX(arrayB, 1, 64, int16_t, CACHE_LINE);
 	int min, max;
 	int count = 0;
-	
+
 	int tmp;
 	int min_error = 0x10000*64;
 	int max_error = 0;
@@ -818,7 +828,7 @@
 	if (errors>0)
 		printf("\t(%i errors out of %i)", errors, count);
 	printf("\n");
-	
+
 	return 0;
 }
 
@@ -980,7 +990,7 @@
  * to the CPU features forced by the library client or autodetected (depending
  * on the XVID_CPU_FORCE flag). It also initializes vlc coding tables and all
  * image colorspace transformation tables.
- * 
+ *
  ****************************************************************************/