ARMv6 optimised pix_sum

Originally committed as revision 21705 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/arm/dsputil_armv6.S

@@ -594,3 +594,28 @@ function ff_pix_norm1_armv6, export=1
         mov             r0,  lr
         pop             {r4-r6, pc}
 .endfunc
+
+function ff_pix_sum_armv6, export=1
+        push            {r4-r7, lr}
+        mov             r12, #16
+        mov             r2,  #0
+        mov             r3,  #0
+        mov             lr,  #0
+        ldr             r4,  [r0]
+1:
+        subs            r12, r12, #1
+        ldr             r5,  [r0, #4]
+        usada8          r2,  r4,  lr,  r2
+        ldr             r6,  [r0, #8]
+        usada8          r3,  r5,  lr,  r3
+        ldr             r7,  [r0, #12]
+        usada8          r2,  r6,  lr,  r2
+        beq             2f
+        ldr             r4,  [r0, r1]!
+        usada8          r3,  r7,  lr,  r3
+        bgt             1b
+2:
+        usada8          r3,  r7,  lr,  r3
+        add             r0,  r2,  r3
+        pop             {r4-r7, pc}
+.endfunc

libavcodec/arm/dsputil_init_armv6.c

@@ -68,6 +68,7 @@ int ff_sse16_armv6(void *s, uint8_t *blk1, uint8_t *blk2,
                    int line_size, int h);
 
 int ff_pix_norm1_armv6(uint8_t *pix, int line_size);
+int ff_pix_sum_armv6(uint8_t *pix, int line_size);
 
 void av_cold ff_dsputil_init_armv6(DSPContext* c, AVCodecContext *avctx)
 {
@@ -116,4 +117,5 @@ void av_cold ff_dsputil_init_armv6(DSPContext* c, AVCodecContext *avctx)
     c->sse[0] = ff_sse16_armv6;
 
     c->pix_norm1 = ff_pix_norm1_armv6;
+    c->pix_sum   = ff_pix_sum_armv6;
 }