h264: DSP'ize MBAFF loopfilter.

3 files changed, 64 insertions, 140 deletions

diff --git a/libavcodec/h264_loopfilter.c b/libavcodec/h264_loopfilter.c
index 619ab198e3..35630658b5 100644
--- a/libavcodec/h264_loopfilter.c
+++ b/libavcodec/h264_loopfilter.c
@@ -135,138 +135,38 @@ static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, int16_t
     }
 }
 
-static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
-    int i;
+static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[7], int bsi, int qp ) {
     int index_a = qp + h->slice_alpha_c0_offset;
     int alpha = alpha_table[index_a];
     int beta  = beta_table[qp + h->slice_beta_offset];
-    for( i = 0; i < 8; i++, pix += stride) {
-        const int bS_index = (i >> 1) * bsi;
-
-        if( bS[bS_index] == 0 ) {
-            continue;
-        }
-
-        if( bS[bS_index] < 4 ) {
-            const int tc0 = tc0_table[index_a][bS[bS_index]];
-            const int p0 = pix[-1];
-            const int p1 = pix[-2];
-            const int p2 = pix[-3];
-            const int q0 = pix[0];
-            const int q1 = pix[1];
-            const int q2 = pix[2];
-
-            if( FFABS( p0 - q0 ) < alpha &&
-                FFABS( p1 - p0 ) < beta &&
-                FFABS( q1 - q0 ) < beta ) {
-                int tc = tc0;
-                int i_delta;
-
-                if( FFABS( p2 - p0 ) < beta ) {
-                    if(tc0)
-                    pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
-                    tc++;
-                }
-                if( FFABS( q2 - q0 ) < beta ) {
-                    if(tc0)
-                    pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
-                    tc++;
-                }
-
-                i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
-                pix[-1] = av_clip_uint8( p0 + i_delta );    /* p0' */
-                pix[0]  = av_clip_uint8( q0 - i_delta );    /* q0' */
-                tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
-            }
-        }else{
-            const int p0 = pix[-1];
-            const int p1 = pix[-2];
-            const int p2 = pix[-3];
-
-            const int q0 = pix[0];
-            const int q1 = pix[1];
-            const int q2 = pix[2];
-
-            if( FFABS( p0 - q0 ) < alpha &&
-                FFABS( p1 - p0 ) < beta &&
-                FFABS( q1 - q0 ) < beta ) {
+    if (alpha ==0 || beta == 0) return;
 
-                if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
-                    if( FFABS( p2 - p0 ) < beta)
-                    {
-                        const int p3 = pix[-4];
-                        /* p0', p1', p2' */
-                        pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
-                        pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
-                        pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
-                    } else {
-                        /* p0' */
-                        pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
-                    }
-                    if( FFABS( q2 - q0 ) < beta)
-                    {
-                        const int q3 = pix[3];
-                        /* q0', q1', q2' */
-                        pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
-                        pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
-                        pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
-                    } else {
-                        /* q0' */
-                        pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
-                    }
-                }else{
-                    /* p0', q0' */
-                    pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
-                    pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
-                }
-                tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
-            }
-        }
+    if( bS[0] < 4 ) {
+        int8_t tc[4];
+        tc[0] = tc0_table[index_a][bS[0*bsi]];
+        tc[1] = tc0_table[index_a][bS[1*bsi]];
+        tc[2] = tc0_table[index_a][bS[2*bsi]];
+        tc[3] = tc0_table[index_a][bS[3*bsi]];
+        h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc);
+    } else {
+        h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta);
     }
 }
-static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
-    int i;
+static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[7], int bsi, int qp ) {
     int index_a = qp + h->slice_alpha_c0_offset;
     int alpha = alpha_table[index_a];
     int beta  = beta_table[qp + h->slice_beta_offset];
-    for( i = 0; i < 4; i++, pix += stride) {
-        const int bS_index = i*bsi;
-
-        if( bS[bS_index] == 0 ) {
-            continue;
-        }
+    if (alpha ==0 || beta == 0) return;
 
-        if( bS[bS_index] < 4 ) {
-            const int tc = tc0_table[index_a][bS[bS_index]] + 1;
-            const int p0 = pix[-1];
-            const int p1 = pix[-2];
-            const int q0 = pix[0];
-            const int q1 = pix[1];
-
-            if( FFABS( p0 - q0 ) < alpha &&
-                FFABS( p1 - p0 ) < beta &&
-                FFABS( q1 - q0 ) < beta ) {
-                const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
-
-                pix[-1] = av_clip_uint8( p0 + i_delta );    /* p0' */
-                pix[0]  = av_clip_uint8( q0 - i_delta );    /* q0' */
-                tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
-            }
-        }else{
-            const int p0 = pix[-1];
-            const int p1 = pix[-2];
-            const int q0 = pix[0];
-            const int q1 = pix[1];
-
-            if( FFABS( p0 - q0 ) < alpha &&
-                FFABS( p1 - p0 ) < beta &&
-                FFABS( q1 - q0 ) < beta ) {
-
-                pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;   /* p0' */
-                pix[0]  = ( 2*q1 + q0 + p1 + 2 ) >> 2;   /* q0' */
-                tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
-            }
-        }
+    if( bS[0] < 4 ) {
+        int8_t tc[4];
+        tc[0] = tc0_table[index_a][bS[0*bsi]] + 1;
+        tc[1] = tc0_table[index_a][bS[1*bsi]] + 1;
+        tc[2] = tc0_table[index_a][bS[2*bsi]] + 1;
+        tc[3] = tc0_table[index_a][bS[3*bsi]] + 1;
+        h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc);
+    } else {
+        h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta);
     }
 }
 
diff --git a/libavcodec/h264dsp.c b/libavcodec/h264dsp.c
index 85dbb8f2e5..1ef6a26db4 100644
--- a/libavcodec/h264dsp.c
+++ b/libavcodec/h264dsp.c
@@ -99,15 +99,15 @@ H264_WEIGHT(2,2)
 #undef op_scale2
 #undef H264_WEIGHT
 
-static av_always_inline av_flatten void h264_loop_filter_luma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0)
+static av_always_inline av_flatten void h264_loop_filter_luma_c(uint8_t *pix, int xstride, int ystride, int inner_iters, int alpha, int beta, int8_t *tc0)
 {
     int i, d;
     for( i = 0; i < 4; i++ ) {
         if( tc0[i] < 0 ) {
-            pix += 4*ystride;
+            pix += inner_iters*ystride;
             continue;
         }
-        for( d = 0; d < 4; d++ ) {
+        for( d = 0; d < inner_iters; d++ ) {
             const int p0 = pix[-1*xstride];
             const int p1 = pix[-2*xstride];
             const int p2 = pix[-3*xstride];
@@ -143,17 +143,21 @@ static av_always_inline av_flatten void h264_loop_filter_luma_c(uint8_t *pix, in
 }
 static void h264_v_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 {
-    h264_loop_filter_luma_c(pix, stride, 1, alpha, beta, tc0);
+    h264_loop_filter_luma_c(pix, stride, 1, 4, alpha, beta, tc0);
 }
 static void h264_h_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 {
-    h264_loop_filter_luma_c(pix, 1, stride, alpha, beta, tc0);
+    h264_loop_filter_luma_c(pix, 1, stride, 4, alpha, beta, tc0);
+}
+static void h264_h_loop_filter_luma_mbaff_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
+{
+    h264_loop_filter_luma_c(pix, 1, stride, 2, alpha, beta, tc0);
 }
 
-static av_always_inline av_flatten void h264_loop_filter_luma_intra_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta)
+static av_always_inline av_flatten void h264_loop_filter_luma_intra_c(uint8_t *pix, int xstride, int ystride, int inner_iters, int alpha, int beta)
 {
     int d;
-    for( d = 0; d < 16; d++ ) {
+    for( d = 0; d < 4 * inner_iters; d++ ) {
         const int p2 = pix[-3*xstride];
         const int p1 = pix[-2*xstride];
         const int p0 = pix[-1*xstride];
@@ -200,23 +204,27 @@ static av_always_inline av_flatten void h264_loop_filter_luma_intra_c(uint8_t *p
 }
 static void h264_v_loop_filter_luma_intra_c(uint8_t *pix, int stride, int alpha, int beta)
 {
-    h264_loop_filter_luma_intra_c(pix, stride, 1, alpha, beta);
+    h264_loop_filter_luma_intra_c(pix, stride, 1, 4, alpha, beta);
 }
 static void h264_h_loop_filter_luma_intra_c(uint8_t *pix, int stride, int alpha, int beta)
 {
-    h264_loop_filter_luma_intra_c(pix, 1, stride, alpha, beta);
+    h264_loop_filter_luma_intra_c(pix, 1, stride, 4, alpha, beta);
+}
+static void h264_h_loop_filter_luma_mbaff_intra_c(uint8_t *pix, int stride, int alpha, int beta)
+{
+    h264_loop_filter_luma_intra_c(pix, 1, stride, 2, alpha, beta);
 }
 
-static av_always_inline av_flatten void h264_loop_filter_chroma_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta, int8_t *tc0)
+static av_always_inline av_flatten void h264_loop_filter_chroma_c(uint8_t *pix, int xstride, int ystride, int inner_iters, int alpha, int beta, int8_t *tc0)
 {
     int i, d;
     for( i = 0; i < 4; i++ ) {
         const int tc = tc0[i];
         if( tc <= 0 ) {
-            pix += 2*ystride;
+            pix += inner_iters*ystride;
             continue;
         }
-        for( d = 0; d < 2; d++ ) {
+        for( d = 0; d < inner_iters; d++ ) {
             const int p0 = pix[-1*xstride];
             const int p1 = pix[-2*xstride];
             const int q0 = pix[0];
@@ -237,17 +245,21 @@ static av_always_inline av_flatten void h264_loop_filter_chroma_c(uint8_t *pix,
 }
 static void h264_v_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 {
-    h264_loop_filter_chroma_c(pix, stride, 1, alpha, beta, tc0);
+    h264_loop_filter_chroma_c(pix, stride, 1, 2, alpha, beta, tc0);
 }
 static void h264_h_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 {
-    h264_loop_filter_chroma_c(pix, 1, stride, alpha, beta, tc0);
+    h264_loop_filter_chroma_c(pix, 1, stride, 2, alpha, beta, tc0);
+}
+static void h264_h_loop_filter_chroma_mbaff_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
+{
+    h264_loop_filter_chroma_c(pix, 1, stride, 1, alpha, beta, tc0);
 }
 
-static av_always_inline av_flatten void h264_loop_filter_chroma_intra_c(uint8_t *pix, int xstride, int ystride, int alpha, int beta)
+static av_always_inline av_flatten void h264_loop_filter_chroma_intra_c(uint8_t *pix, int xstride, int ystride, int inner_iters, int alpha, int beta)
 {
     int d;
-    for( d = 0; d < 8; d++ ) {
+    for( d = 0; d < 4 * inner_iters; d++ ) {
         const int p0 = pix[-1*xstride];
         const int p1 = pix[-2*xstride];
         const int q0 = pix[0];
@@ -265,11 +277,15 @@ static av_always_inline av_flatten void h264_loop_filter_chroma_intra_c(uint8_t
 }
 static void h264_v_loop_filter_chroma_intra_c(uint8_t *pix, int stride, int alpha, int beta)
 {
-    h264_loop_filter_chroma_intra_c(pix, stride, 1, alpha, beta);
+    h264_loop_filter_chroma_intra_c(pix, stride, 1, 2, alpha, beta);
 }
 static void h264_h_loop_filter_chroma_intra_c(uint8_t *pix, int stride, int alpha, int beta)
 {
-    h264_loop_filter_chroma_intra_c(pix, 1, stride, alpha, beta);
+    h264_loop_filter_chroma_intra_c(pix, 1, stride, 2, alpha, beta);
+}
+static void h264_h_loop_filter_chroma_mbaff_intra_c(uint8_t *pix, int stride, int alpha, int beta)
+{
+    h264_loop_filter_chroma_intra_c(pix, 1, stride, 1, alpha, beta);
 }
 
 void ff_h264dsp_init(H264DSPContext *c)
@@ -307,12 +323,16 @@ void ff_h264dsp_init(H264DSPContext *c)
 
     c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_c;
     c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_c;
+    c->h264_h_loop_filter_luma_mbaff= h264_h_loop_filter_luma_mbaff_c;
     c->h264_v_loop_filter_luma_intra= h264_v_loop_filter_luma_intra_c;
     c->h264_h_loop_filter_luma_intra= h264_h_loop_filter_luma_intra_c;
+    c->h264_h_loop_filter_luma_mbaff_intra= h264_h_loop_filter_luma_mbaff_intra_c;
     c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_c;
     c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_c;
+    c->h264_h_loop_filter_chroma_mbaff= h264_h_loop_filter_chroma_mbaff_c;
     c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_c;
     c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_c;
+    c->h264_h_loop_filter_chroma_mbaff_intra= h264_h_loop_filter_chroma_mbaff_intra_c;
     c->h264_loop_filter_strength= NULL;
 
     if (ARCH_ARM) ff_h264dsp_init_arm(c);
diff --git a/libavcodec/h264dsp.h b/libavcodec/h264dsp.h
index 6fa1746b89..9db4c13784 100644
--- a/libavcodec/h264dsp.h
+++ b/libavcodec/h264dsp.h
@@ -45,13 +45,17 @@ typedef struct H264DSPContext{
     /* loop filter */
     void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
     void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
+    void (*h264_h_loop_filter_luma_mbaff)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
     /* v/h_loop_filter_luma_intra: align 16 */
     void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
     void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
+    void (*h264_h_loop_filter_luma_mbaff_intra)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta);
     void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
     void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
+    void (*h264_h_loop_filter_chroma_mbaff)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
     void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
+    void (*h264_h_loop_filter_chroma_mbaff_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
     void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
                                       int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);