Blackfin optimized uyvytoyv12 routine. the implementation includes chroma averaging.

This is 10x faster than the cooresponding C function.



git-svn-id: file:///var/local/repositories/mplayer/trunk/libswscale@23345 b3059339-0415-0410-9bf9-f77b7e298cf2

internal_bfin.S

@@ -452,3 +452,82 @@ DEFUN(yuv2rgb24_line,MEM,
         unlink;
         rts;
 DEFUN_END(yuv2rgb888_line)
+
+
+
+#define ARG_vdst        20
+#define ARG_width       24
+#define ARG_height      28
+#define ARG_lumStride   32
+#define ARG_chromStride 36
+#define ARG_srcStride   40
+
+DEFUN(uyvytoyv12, mL3,  (const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+                         long width, long height,
+                         long lumStride, long chromStride, long srcStride)):
+        link 0;
+        [--sp] = (r7:4,p5:4);
+
+        p0 = r1;       // Y top even
+
+        i2 = r2; // *u
+        r2 = [fp + ARG_vdst];
+        i3 = r2; // *v
+
+        r1 = [fp + ARG_srcStride];
+        r2 = r0 + r1;
+        r1 += -16;
+        m0 = r1;
+
+        i0 = r0;  // uyvy_T even
+        i1 = r2;  // uyvy_B odd
+
+        r2 = [fp + ARG_chromStride];
+        m1 = r2;
+
+        p2 = [fp + ARG_lumStride];
+        p1 = p0 + p2;  // Y bot odd
+        p2 += -4;
+
+        p5 = [fp + ARG_width];
+        p4 = [fp + ARG_height];
+        p4 = p4 >> 1;
+        p5 = p5 >> 2;
+
+        r6.l = w[i2--];  r6.l = w[i2];
+        r6.h = w[i3--];  r6.h = w[i3];
+        /*   I0,I1 - src input line pointers
+         *   p0,p1 - luma output line pointers
+         *   I2    - dstU
+         *   I3    - dstV
+         */
+
+        lsetup (0f, 1f) lc0 = p4;
+
+0:        lsetup (2f, 3f) lc1 = p5;
+            r0 = [i0++] || r2 = [i1++];
+            r1 = [i0++] || r3 = [i1++];
+2:          r4 = byteop1p(r1:0, r3:2)     ||   w[i2++] = r6.l;
+            r5 = byteop1p(r1:0, r3:2) (r) ||   w[i3++] = r6.h;
+            r0 = r0 >> 8(v);
+            r1 = r1 >> 8(v);
+            r2 = r2 >> 8(v);
+            r3 = r3 >> 8(v);
+            r0 = bytepack(r0, r1);
+            r2 = bytepack(r2, r3)         ||  [p0++] = r0;
+            r6 = pack(r5.l, r4.l)         ||  [p1++] = r2;
+            r7 = pack(r5.h, r4.h)         ||  r0 = [i0++]   || r2 = [i1++];
+3:          r6 = bytepack(r6, r7)         ||  r1 = [i0++]   || r3 = [i1++];
+
+          i0 += m0;
+          i1 += m0;
+          p0 = p0 + p2;
+1:        p1 = p1 + p2;
+
+        w[i2++] = r6.l;
+        w[i3++] = r6.h;
+
+        (r7:4,p5:4) = [sp++];
+        unlink;
+        rts;
+DEFUN_END(uyvytoyv12)