--- trunk/xvidcore/src/motion/x86_asm/sad_mmx.asm	2004/03/22 22:36:25	1382
+++ trunk/xvidcore/src/motion/x86_asm/sad_mmx.asm	2004/04/12 15:49:56	1424
@@ -20,7 +20,7 @@
 ; *  along with this program; if not, write to the Free Software
 ; *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
 ; *
-; * $Id: sad_mmx.asm,v 1.12 2004-03-22 22:36:24 edgomez Exp $
+; * $Id: sad_mmx.asm,v 1.13 2004-04-12 15:49:56 edgomez Exp $
 ; *
 ; ***************************************************************************/
 
@@ -270,6 +270,7 @@
 cglobal sad8bi_mmx
 cglobal dev16_mmx
 cglobal sse8_16bit_mmx
+cglobal sse8_8bit_mmx
 	
 ;-----------------------------------------------------------------------------
 ;
@@ -630,7 +631,7 @@
 ;
 ;-----------------------------------------------------------------------------
 
-%macro ROW_SSE_MMX 2
+%macro ROW_SSE_16bit_MMX 2
   movq mm0, [%1]
   movq mm1, [%1+8]
   psubw mm0, [%2]
@@ -639,7 +640,7 @@
   pmaddwd mm1, mm1
   paddd mm2, mm0
   paddd mm2, mm1
-%endmacro	
+%endmacro
 	
 sse8_16bit_mmx:
   push esi
@@ -654,30 +655,11 @@
   pxor mm2, mm2
 
   ;; Let's go
-  ROW_SSE_MMX esi, edi
-  lea esi, [esi+edx]
-  lea edi, [edi+edx]
-  ROW_SSE_MMX esi, edi
-  lea esi, [esi+edx]
-  lea edi, [edi+edx]
-  ROW_SSE_MMX esi, edi
-  lea esi, [esi+edx]
-  lea edi, [edi+edx]
-  ROW_SSE_MMX esi, edi
-  lea esi, [esi+edx]
-  lea edi, [edi+edx]
-  ROW_SSE_MMX esi, edi
-  lea esi, [esi+edx]
-  lea edi, [edi+edx]
-  ROW_SSE_MMX esi, edi
-  lea esi, [esi+edx]
-  lea edi, [edi+edx]
-  ROW_SSE_MMX esi, edi
-  lea esi, [esi+edx]
-  lea edi, [edi+edx]
-  ROW_SSE_MMX esi, edi
+%rep 8
+  ROW_SSE_16bit_MMX esi, edi
   lea esi, [esi+edx]
   lea edi, [edi+edx]
+%endrep
 
   ;; Finish adding each dword of the accumulator
   movq mm3, mm2
@@ -689,3 +671,67 @@
   pop edi
   pop esi
   ret
+
+;-----------------------------------------------------------------------------
+;
+; uint32_t sse8_8bit_mmx(const int8_t *b1,
+;                        const int8_t *b2,
+;                        const uint32_t stride);
+;
+;-----------------------------------------------------------------------------
+
+%macro ROW_SSE_8bit_MMX 2
+  movq mm0, [%1] ; load a row
+  movq mm2, [%2] ; load a row
+
+  movq mm1, mm0  ; copy row
+  movq mm3, mm2  ; copy row
+
+  punpcklbw mm0, mm7 ; turn the 4low elements into 16bit
+  punpckhbw mm1, mm7 ; turn the 4high elements into 16bit
+
+  punpcklbw mm2, mm7 ; turn the 4low elements into 16bit
+  punpckhbw mm3, mm7 ; turn the 4high elements into 16bit
+
+  psubw mm0, mm2 ; low  part of src-dst
+  psubw mm1, mm3 ; high part of src-dst
+
+  pmaddwd mm0, mm0 ; compute the square sum
+  pmaddwd mm1, mm1 ; compute the square sum
+
+  paddd mm6, mm0 ; add to the accumulator
+  paddd mm6, mm1 ; add to the accumulator
+%endmacro
+
+sse8_8bit_mmx:
+  push esi
+  push edi
+
+  ;; Load the function params
+  mov esi, [esp+8+4]
+  mov edi, [esp+8+8]
+  mov edx, [esp+8+12]
+
+  ;; Reset the sse accumulator
+  pxor mm6, mm6
+
+  ;; Used to interleave 8bit data with 0x00 values
+  pxor mm7, mm7
+
+  ;; Let's go
+%rep 8
+  ROW_SSE_8bit_MMX esi, edi
+  lea esi, [esi+edx]
+  lea edi, [edi+edx]
+%endrep
+
+  ;; Finish adding each dword of the accumulator
+  movq mm7, mm6
+  psrlq mm6, 32
+  paddd mm6, mm7
+  movd eax, mm6
+
+  ;; All done
+  pop edi
+  pop esi
+  ret