Move VC1 loop filter to DSPContext

Originally committed as revision 18520 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/dsputil.h

@@ -482,6 +482,7 @@ typedef struct DSPContext {
     void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
     void (*vc1_v_overlap)(uint8_t* src, int stride);
     void (*vc1_h_overlap)(uint8_t* src, int stride);
+    void (*vc1_loop_filter)(uint8_t *src, int step, int stride, int len, int pq);
     /* put 8x8 block with bicubic interpolation and quarterpel precision
      * last argument is actually round value instead of height
      */

libavcodec/vc1dsp.c

@@ -78,6 +78,73 @@ static void vc1_h_overlap_c(uint8_t* src, int stride)
     }
 }
 
+/**
+ * VC-1 in-loop deblocking filter for one line
+ * @param src source block type
+ * @param stride block stride
+ * @param pq block quantizer
+ * @return whether other 3 pairs should be filtered or not
+ * @see 8.6
+ */
+static av_always_inline int vc1_filter_line(uint8_t* src, int stride, int pq){
+    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
+
+    int a0 = (2*(src[-2*stride] - src[ 1*stride]) - 5*(src[-1*stride] - src[ 0*stride]) + 4) >> 3;
+    int a0_sign = a0 >> 31;        /* Store sign */
+    a0 = (a0 ^ a0_sign) - a0_sign; /* a0 = FFABS(a0); */
+    if(a0 < pq){
+        int a1 = FFABS((2*(src[-4*stride] - src[-1*stride]) - 5*(src[-3*stride] - src[-2*stride]) + 4) >> 3);
+        int a2 = FFABS((2*(src[ 0*stride] - src[ 3*stride]) - 5*(src[ 1*stride] - src[ 2*stride]) + 4) >> 3);
+        if(a1 < a0 || a2 < a0){
+            int clip = src[-1*stride] - src[ 0*stride];
+            int clip_sign = clip >> 31;
+            clip = ((clip ^ clip_sign) - clip_sign)>>1;
+            if(clip){
+                int a3 = FFMIN(a1, a2);
+                int d = 5 * (a3 - a0);
+                int d_sign = (d >> 31);
+                d = ((d ^ d_sign) - d_sign) >> 3;
+                d_sign ^= a0_sign;
+
+                if( d_sign ^ clip_sign )
+                    d = 0;
+                else{
+                    d = FFMIN(d, clip);
+                    d = (d ^ d_sign) - d_sign;          /* Restore sign */
+                    src[-1*stride] = cm[src[-1*stride] - d];
+                    src[ 0*stride] = cm[src[ 0*stride] + d];
+                }
+                return 1;
+            }
+        }
+    }
+    return 0;
+}
+
+/**
+ * VC-1 in-loop deblocking filter
+ * @param src source block type
+ * @param step distance between horizontally adjacent elements
+ * @param stride distance between vertically adjacent elements
+ * @param len edge length to filter (4 or 8 pixels)
+ * @param pq block quantizer
+ * @see 8.6
+ */
+static void vc1_loop_filter(uint8_t* src, int step, int stride, int len, int pq)
+{
+    int i;
+    int filt3;
+
+    for(i = 0; i < len; i += 4){
+        filt3 = vc1_filter_line(src + 2*step, stride, pq);
+        if(filt3){
+            vc1_filter_line(src + 0*step, stride, pq);
+            vc1_filter_line(src + 1*step, stride, pq);
+            vc1_filter_line(src + 3*step, stride, pq);
+        }
+        src += step * 4;
+    }
+}
 
 /** Do inverse transform on 8x8 block
 */
@@ -450,6 +517,7 @@ void ff_vc1dsp_init(DSPContext* dsp, AVCodecContext *avctx) {
     dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c;
     dsp->vc1_h_overlap = vc1_h_overlap_c;
     dsp->vc1_v_overlap = vc1_v_overlap_c;
+    dsp->vc1_loop_filter = vc1_loop_filter;
 
     dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_c;
     dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_c;