h264: DSP'ize MBAFF loopfilter.
[ffmpeg.git] / libavcodec / h264dsp.c
index 4f18afa..1ef6a26 100644 (file)
 /*
- * H.264/MPEG-4 Part 10 (Base profile) encoder.
+ * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
+ * Copyright (c) 2003-2010 Michael Niedermayer <michaelni@gmx.at>
  *
- * DSP functions
+ * This file is part of Libav.
  *
- * Copyright (c) 2006 Expertisecentrum Digitale Media, UHasselt
- *
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
  * version 2.1 of the License, or (at your option) any later version.
  *
- * FFmpeg is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 /**
- * @file h264dsp.c
- * H.264 encoder related DSP utils
- *
+ * @file
+ * H.264 / AVC / MPEG4 part10 DSP functions.
+ * @author Michael Niedermayer <michaelni@gmx.at>
  */
 
-#include "dsputil.h"
-
-extern const uint8_t ff_div6[52];
-extern const uint8_t ff_rem6[52];
-
-#define  H264_DCT_PART1(X) \
-         a = block[0][X]+block[3][X]; \
-         c = block[0][X]-block[3][X]; \
-         b = block[1][X]+block[2][X]; \
-         d = block[1][X]-block[2][X]; \
-         pieces[0][X] = a+b; \
-         pieces[2][X] = a-b; \
-         pieces[1][X] = (c<<1)+d; \
-         pieces[3][X] = c-(d<<1);
-
-#define  H264_DCT_PART2(X) \
-         a = pieces[X][0]+pieces[X][3]; \
-         c = pieces[X][0]-pieces[X][3]; \
-         b = pieces[X][1]+pieces[X][2]; \
-         d = pieces[X][1]-pieces[X][2]; \
-         block[0][X] = a+b; \
-         block[2][X] = a-b; \
-         block[1][X] = (c<<1)+d; \
-         block[3][X] = c-(d<<1);
+#include <stdint.h>
+#include "avcodec.h"
+#include "h264dsp.h"
 
-/**
- * Transform the provided matrix using the H.264 modified DCT.
- * @note
- * we'll always work with transposed input blocks, to avoid having to make a
- * distinction between C and mmx implementations.
- *
- * @param block transposed input block
- */
-static void h264_dct_c(DCTELEM block[4][4])
+#define op_scale1(x)  block[x] = av_clip_uint8( (block[x]*weight + offset) >> log2_denom )
+#define op_scale2(x)  dst[x] = av_clip_uint8( (src[x]*weights + dst[x]*weightd + offset) >> (log2_denom+1))
+#define H264_WEIGHT(W,H) \
+static void weight_h264_pixels ## W ## x ## H ## _c(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \
+    int y; \
+    offset <<= log2_denom; \
+    if(log2_denom) offset += 1<<(log2_denom-1); \
+    for(y=0; y<H; y++, block += stride){ \
+        op_scale1(0); \
+        op_scale1(1); \
+        if(W==2) continue; \
+        op_scale1(2); \
+        op_scale1(3); \
+        if(W==4) continue; \
+        op_scale1(4); \
+        op_scale1(5); \
+        op_scale1(6); \
+        op_scale1(7); \
+        if(W==8) continue; \
+        op_scale1(8); \
+        op_scale1(9); \
+        op_scale1(10); \
+        op_scale1(11); \
+        op_scale1(12); \
+        op_scale1(13); \
+        op_scale1(14); \
+        op_scale1(15); \
+    } \
+} \
+static void biweight_h264_pixels ## W ## x ## H ## _c(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \
+    int y; \
+    offset = ((offset + 1) | 1) << log2_denom; \
+    for(y=0; y<H; y++, dst += stride, src += stride){ \
+        op_scale2(0); \
+        op_scale2(1); \
+        if(W==2) continue; \
+        op_scale2(2); \
+        op_scale2(3); \
+        if(W==4) continue; \
+        op_scale2(4); \
+        op_scale2(5); \
+        op_scale2(6); \
+        op_scale2(7); \
+        if(W==8) continue; \
+        op_scale2(8); \
+        op_scale2(9); \
+        op_scale2(10); \
+        op_scale2(11); \
+        op_scale2(12); \
+        op_scale2(13); \
+        op_scale2(14); \
+        op_scale2(15); \
+    } \
+}
+
+H264_WEIGHT(16,16)
+H264_WEIGHT(16,8)
+H264_WEIGHT(8,16)
+H264_WEIGHT(8,8)
+H264_WEIGHT(8,4)
+H264_WEIGHT(4,8)
+H264_WEIGHT(4,4)
+H264_WEIGHT(4,2)
+H264_WEIGHT(2,4)
+H264_WEIGHT(2,2)
+
+#undef op_scale1
+#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 inner_iters, int alpha, int beta, int8_t *tc0)
 {
-    DCTELEM pieces[4][4];
-    DCTELEM a, b, c, d;
+    int i, d;
+    for( i = 0; i < 4; i++ ) {
+        if( tc0[i] < 0 ) {
+            pix += inner_iters*ystride;
+            continue;
+        }
+        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];
+            const int q0 = pix[0];
+            const int q1 = pix[1*xstride];
+            const int q2 = pix[2*xstride];
+
+            if( FFABS( p0 - q0 ) < alpha &&
+                FFABS( p1 - p0 ) < beta &&
+                FFABS( q1 - q0 ) < beta ) {
+
+                int tc = tc0[i];
+                int i_delta;
 
-    H264_DCT_PART1(0);
-    H264_DCT_PART1(1);
-    H264_DCT_PART1(2);
-    H264_DCT_PART1(3);
-    H264_DCT_PART2(0);
-    H264_DCT_PART2(1);
-    H264_DCT_PART2(2);
-    H264_DCT_PART2(3);
+                if( FFABS( p2 - p0 ) < beta ) {
+                    if(tc0[i])
+                    pix[-2*xstride] = p1 + av_clip( (( p2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - p1, -tc0[i], tc0[i] );
+                    tc++;
+                }
+                if( FFABS( q2 - q0 ) < beta ) {
+                    if(tc0[i])
+                    pix[   xstride] = q1 + av_clip( (( q2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - q1, -tc0[i], tc0[i] );
+                    tc++;
+                }
+
+                i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
+                pix[-xstride] = av_clip_uint8( p0 + i_delta );    /* p0' */
+                pix[0]        = av_clip_uint8( q0 - i_delta );    /* q0' */
+            }
+            pix += ystride;
+        }
+    }
+}
+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, 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, 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 inner_iters, int alpha, int beta)
+{
+    int 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];
+
+        const int q0 = pix[ 0*xstride];
+        const int q1 = pix[ 1*xstride];
+        const int q2 = pix[ 2*xstride];
 
-void ff_h264dsp_init(DSPContext* c, AVCodecContext *avctx)
+        if( FFABS( p0 - q0 ) < alpha &&
+            FFABS( p1 - p0 ) < beta &&
+            FFABS( q1 - q0 ) < beta ) {
+
+            if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
+                if( FFABS( p2 - p0 ) < beta)
+                {
+                    const int p3 = pix[-4*xstride];
+                    /* p0', p1', p2' */
+                    pix[-1*xstride] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
+                    pix[-2*xstride] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
+                    pix[-3*xstride] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
+                } else {
+                    /* p0' */
+                    pix[-1*xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
+                }
+                if( FFABS( q2 - q0 ) < beta)
+                {
+                    const int q3 = pix[3*xstride];
+                    /* q0', q1', q2' */
+                    pix[0*xstride] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
+                    pix[1*xstride] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
+                    pix[2*xstride] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
+                } else {
+                    /* q0' */
+                    pix[0*xstride] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
+                }
+            }else{
+                /* p0', q0' */
+                pix[-1*xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
+                pix[ 0*xstride] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
+            }
+        }
+        pix += ystride;
+    }
+}
+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, 4, alpha, beta);
+}
+static void h264_h_loop_filter_luma_intra_c(uint8_t *pix, int stride, int alpha, int beta)
 {
-    c->h264_dct = h264_dct_c;
+    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 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 += inner_iters*ystride;
+            continue;
+        }
+        for( d = 0; d < inner_iters; d++ ) {
+            const int p0 = pix[-1*xstride];
+            const int p1 = pix[-2*xstride];
+            const int q0 = pix[0];
+            const int q1 = pix[1*xstride];
+
+            if( FFABS( p0 - q0 ) < alpha &&
+                FFABS( p1 - p0 ) < beta &&
+                FFABS( q1 - q0 ) < beta ) {
+
+                int delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
+
+                pix[-xstride] = av_clip_uint8( p0 + delta );    /* p0' */
+                pix[0]        = av_clip_uint8( q0 - delta );    /* q0' */
+            }
+            pix += ystride;
+        }
+    }
+}
+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, 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, 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 inner_iters, int alpha, int beta)
+{
+    int 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];
+        const int q1 = pix[1*xstride];
+
+        if( FFABS( p0 - q0 ) < alpha &&
+            FFABS( p1 - p0 ) < beta &&
+            FFABS( q1 - q0 ) < beta ) {
+
+            pix[-xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2;   /* p0' */
+            pix[0]        = ( 2*q1 + q0 + p1 + 2 ) >> 2;   /* q0' */
+        }
+        pix += ystride;
+    }
+}
+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, 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, 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)
+{
+    c->h264_idct_add= ff_h264_idct_add_c;
+    c->h264_idct8_add= ff_h264_idct8_add_c;
+    c->h264_idct_dc_add= ff_h264_idct_dc_add_c;
+    c->h264_idct8_dc_add= ff_h264_idct8_dc_add_c;
+    c->h264_idct_add16     = ff_h264_idct_add16_c;
+    c->h264_idct8_add4     = ff_h264_idct8_add4_c;
+    c->h264_idct_add8      = ff_h264_idct_add8_c;
+    c->h264_idct_add16intra= ff_h264_idct_add16intra_c;
+    c->h264_luma_dc_dequant_idct= ff_h264_luma_dc_dequant_idct_c;
+
+    c->weight_h264_pixels_tab[0]= weight_h264_pixels16x16_c;
+    c->weight_h264_pixels_tab[1]= weight_h264_pixels16x8_c;
+    c->weight_h264_pixels_tab[2]= weight_h264_pixels8x16_c;
+    c->weight_h264_pixels_tab[3]= weight_h264_pixels8x8_c;
+    c->weight_h264_pixels_tab[4]= weight_h264_pixels8x4_c;
+    c->weight_h264_pixels_tab[5]= weight_h264_pixels4x8_c;
+    c->weight_h264_pixels_tab[6]= weight_h264_pixels4x4_c;
+    c->weight_h264_pixels_tab[7]= weight_h264_pixels4x2_c;
+    c->weight_h264_pixels_tab[8]= weight_h264_pixels2x4_c;
+    c->weight_h264_pixels_tab[9]= weight_h264_pixels2x2_c;
+    c->biweight_h264_pixels_tab[0]= biweight_h264_pixels16x16_c;
+    c->biweight_h264_pixels_tab[1]= biweight_h264_pixels16x8_c;
+    c->biweight_h264_pixels_tab[2]= biweight_h264_pixels8x16_c;
+    c->biweight_h264_pixels_tab[3]= biweight_h264_pixels8x8_c;
+    c->biweight_h264_pixels_tab[4]= biweight_h264_pixels8x4_c;
+    c->biweight_h264_pixels_tab[5]= biweight_h264_pixels4x8_c;
+    c->biweight_h264_pixels_tab[6]= biweight_h264_pixels4x4_c;
+    c->biweight_h264_pixels_tab[7]= biweight_h264_pixels4x2_c;
+    c->biweight_h264_pixels_tab[8]= biweight_h264_pixels2x4_c;
+    c->biweight_h264_pixels_tab[9]= biweight_h264_pixels2x2_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);
+    if (HAVE_ALTIVEC) ff_h264dsp_init_ppc(c);
+    if (HAVE_MMX) ff_h264dsp_init_x86(c);
+}