New qpel MC functions conforming to VC-1 standard.

Existing DSPUtil functions cause chroma artifacts on some files.

Originally committed as revision 6139 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/dsputil.c

@@ -2598,6 +2598,10 @@ void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);
 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, uint8_t *src, int stride, int rnd) {
     put_pixels8_c(dst, src, stride, 8);
 }
+
+void ff_put_vc1_qpel_mc00_c(uint8_t *dst, uint8_t *src, int stride, int rnd) {
+    put_pixels8_c(dst, src, stride, 8);
+}
 #endif /* CONFIG_VC1_DECODER||CONFIG_WMV3_DECODER */
 
 static void wmv2_mspel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int w){

libavcodec/dsputil.h

@@ -396,6 +396,7 @@ typedef struct DSPContext {
      * last argument is actually round value instead of height
      */
     op_pixels_func put_vc1_mspel_pixels_tab[16];
+    op_pixels_func put_vc1_qpel_pixels_tab[16];
 } DSPContext;
 
 void dsputil_static_init(void);

libavcodec/vc1.c

@@ -879,13 +879,8 @@ static void vc1_mc_1mv(VC1Context *v, int dir)
     if(s->flags & CODEC_FLAG_GRAY) return;
     /* Chroma MC always uses qpel bilinear */
     uvdxy = ((uvmy & 3) << 2) | (uvmx & 3);
-    if(!v->rnd){
-        dsp->put_qpel_pixels_tab[1][uvdxy](s->dest[1], srcU, s->uvlinesize);
-        dsp->put_qpel_pixels_tab[1][uvdxy](s->dest[2], srcV, s->uvlinesize);
-    }else{
-        dsp->put_no_rnd_qpel_pixels_tab[1][uvdxy](s->dest[1], srcU, s->uvlinesize);
-        dsp->put_no_rnd_qpel_pixels_tab[1][uvdxy](s->dest[2], srcV, s->uvlinesize);
-    }
+    dsp->put_vc1_qpel_pixels_tab[uvdxy](s->dest[1], srcU, s->uvlinesize, v->rnd);
+    dsp->put_vc1_qpel_pixels_tab[uvdxy](s->dest[2], srcV, s->uvlinesize, v->rnd);
 }
 
 /** Do motion compensation for 4-MV macroblock - luminance block
@@ -1082,13 +1077,8 @@ static void vc1_mc_4mv_chroma(VC1Context *v)
 
     /* Chroma MC always uses qpel bilinear */
     uvdxy = ((uvmy & 3) << 2) | (uvmx & 3);
-    if(!v->rnd){
-        dsp->put_qpel_pixels_tab[1][uvdxy](s->dest[1], srcU, s->uvlinesize);
-        dsp->put_qpel_pixels_tab[1][uvdxy](s->dest[2], srcV, s->uvlinesize);
-    }else{
-        dsp->put_no_rnd_qpel_pixels_tab[1][uvdxy](s->dest[1], srcU, s->uvlinesize);
-        dsp->put_no_rnd_qpel_pixels_tab[1][uvdxy](s->dest[2], srcV, s->uvlinesize);
-    }
+    dsp->put_vc1_qpel_pixels_tab[uvdxy](s->dest[1], srcU, s->uvlinesize, v->rnd);
+    dsp->put_vc1_qpel_pixels_tab[uvdxy](s->dest[2], srcV, s->uvlinesize, v->rnd);
 }
 
 /**

libavcodec/vc1dsp.c

@@ -424,6 +424,169 @@ static void ff_put_vc1_mspel_mc33_c(uint8_t *dst, const uint8_t *src, int stride
     vc1_mspel_mc(dst, src, stride, 0xF, rnd);
 }
 
+/** Filter used to interpolate fractional pel values
+ * except for half-pel cases for _mcXY:
+ *   A = (4-X)*(4-Y)
+ *   B =    X *(4-Y)
+ *   C = (4-X)*   Y
+ *   D =    X *   Y
+ */
+#define VC1_QPEL_FILTER(src, i, stride, rnd, A, B, C, D) \
+    clip_uint8((A*src[i] + B*src[i+1] + C*src[i+stride] + D*src[i+stride+1] + 8 - rnd) >> 4)
+
+/* this one is defined in dsputil.c */
+void ff_put_vc1_qpel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd);
+
+static void ff_put_vc1_qpel_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 12, 4, 0, 0);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = clip_uint8((src[i] + src[i + 1] + 1 - rnd) >> 1);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc30_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 4, 12, 0, 0);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 12, 0, 4, 0);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 9, 3, 3, 1);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 6, 6, 2, 2);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc31_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 3, 9, 1, 3);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = clip_uint8((src[i] + src[i + stride] + 1 - rnd) >> 1);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 6, 2, 6, 2);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = clip_uint8((src[i] + src[i + 1] + src[i + stride] + src[i + stride + 1] + 2 - rnd) >> 2);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc32_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 2, 6, 2, 6);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc03_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 4, 0, 12, 0);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc13_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 3, 1, 9, 3);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc23_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 2, 2, 6, 6);
+        dst += stride;
+        src += stride;
+    }
+}
+
+static void ff_put_vc1_qpel_mc33_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
+    int i, j;
+    for(j = 0; j < 8; j++) {
+        for(i = 0; i < 8; i++)
+            dst[i] = VC1_QPEL_FILTER(src, i, stride, rnd, 1, 3, 3, 9);
+        dst += stride;
+        src += stride;
+    }
+}
+
 void ff_vc1dsp_init(DSPContext* dsp, AVCodecContext *avctx) {
     dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_c;
     dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c;
@@ -448,4 +611,21 @@ void ff_vc1dsp_init(DSPContext* dsp, AVCodecContext *avctx) {
     dsp->put_vc1_mspel_pixels_tab[13] = ff_put_vc1_mspel_mc13_c;
     dsp->put_vc1_mspel_pixels_tab[14] = ff_put_vc1_mspel_mc23_c;
     dsp->put_vc1_mspel_pixels_tab[15] = ff_put_vc1_mspel_mc33_c;
+
+    dsp->put_vc1_qpel_pixels_tab[ 0] = ff_put_vc1_qpel_mc00_c;
+    dsp->put_vc1_qpel_pixels_tab[ 1] = ff_put_vc1_qpel_mc10_c;
+    dsp->put_vc1_qpel_pixels_tab[ 2] = ff_put_vc1_qpel_mc20_c;
+    dsp->put_vc1_qpel_pixels_tab[ 3] = ff_put_vc1_qpel_mc30_c;
+    dsp->put_vc1_qpel_pixels_tab[ 4] = ff_put_vc1_qpel_mc01_c;
+    dsp->put_vc1_qpel_pixels_tab[ 5] = ff_put_vc1_qpel_mc11_c;
+    dsp->put_vc1_qpel_pixels_tab[ 6] = ff_put_vc1_qpel_mc21_c;
+    dsp->put_vc1_qpel_pixels_tab[ 7] = ff_put_vc1_qpel_mc31_c;
+    dsp->put_vc1_qpel_pixels_tab[ 8] = ff_put_vc1_qpel_mc02_c;
+    dsp->put_vc1_qpel_pixels_tab[ 9] = ff_put_vc1_qpel_mc12_c;
+    dsp->put_vc1_qpel_pixels_tab[10] = ff_put_vc1_qpel_mc22_c;
+    dsp->put_vc1_qpel_pixels_tab[11] = ff_put_vc1_qpel_mc32_c;
+    dsp->put_vc1_qpel_pixels_tab[12] = ff_put_vc1_qpel_mc03_c;
+    dsp->put_vc1_qpel_pixels_tab[13] = ff_put_vc1_qpel_mc13_c;
+    dsp->put_vc1_qpel_pixels_tab[14] = ff_put_vc1_qpel_mc23_c;
+    dsp->put_vc1_qpel_pixels_tab[15] = ff_put_vc1_qpel_mc33_c;
 }