Revision 563c72da libavcodec/h264dsp.c

View differences:

libavcodec/h264dsp.c
29 29
#include "avcodec.h"
30 30
#include "h264dsp.h"
31 31

  
32
#define op_scale1(x)  block[x] = av_clip_uint8( (block[x]*weight + offset) >> log2_denom )
33
#define op_scale2(x)  dst[x] = av_clip_uint8( (src[x]*weights + dst[x]*weightd + offset) >> (log2_denom+1))
34
#define H264_WEIGHT(W,H) \
35
static void weight_h264_pixels ## W ## x ## H ## _c(uint8_t *block, int stride, int log2_denom, int weight, int offset){ \
36
    int y; \
37
    offset <<= log2_denom; \
38
    if(log2_denom) offset += 1<<(log2_denom-1); \
39
    for(y=0; y<H; y++, block += stride){ \
40
        op_scale1(0); \
41
        op_scale1(1); \
42
        if(W==2) continue; \
43
        op_scale1(2); \
44
        op_scale1(3); \
45
        if(W==4) continue; \
46
        op_scale1(4); \
47
        op_scale1(5); \
48
        op_scale1(6); \
49
        op_scale1(7); \
50
        if(W==8) continue; \
51
        op_scale1(8); \
52
        op_scale1(9); \
53
        op_scale1(10); \
54
        op_scale1(11); \
55
        op_scale1(12); \
56
        op_scale1(13); \
57
        op_scale1(14); \
58
        op_scale1(15); \
59
    } \
60
} \
61
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){ \
62
    int y; \
63
    offset = ((offset + 1) | 1) << log2_denom; \
64
    for(y=0; y<H; y++, dst += stride, src += stride){ \
65
        op_scale2(0); \
66
        op_scale2(1); \
67
        if(W==2) continue; \
68
        op_scale2(2); \
69
        op_scale2(3); \
70
        if(W==4) continue; \
71
        op_scale2(4); \
72
        op_scale2(5); \
73
        op_scale2(6); \
74
        op_scale2(7); \
75
        if(W==8) continue; \
76
        op_scale2(8); \
77
        op_scale2(9); \
78
        op_scale2(10); \
79
        op_scale2(11); \
80
        op_scale2(12); \
81
        op_scale2(13); \
82
        op_scale2(14); \
83
        op_scale2(15); \
84
    } \
85
}
86

  
87
H264_WEIGHT(16,16)
88
H264_WEIGHT(16,8)
89
H264_WEIGHT(8,16)
90
H264_WEIGHT(8,8)
91
H264_WEIGHT(8,4)
92
H264_WEIGHT(4,8)
93
H264_WEIGHT(4,4)
94
H264_WEIGHT(4,2)
95
H264_WEIGHT(2,4)
96
H264_WEIGHT(2,2)
97

  
98
#undef op_scale1
99
#undef op_scale2
100
#undef H264_WEIGHT
101

  
102
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)
103
{
104
    int i, d;
105
    for( i = 0; i < 4; i++ ) {
106
        if( tc0[i] < 0 ) {
107
            pix += inner_iters*ystride;
108
            continue;
109
        }
110
        for( d = 0; d < inner_iters; d++ ) {
111
            const int p0 = pix[-1*xstride];
112
            const int p1 = pix[-2*xstride];
113
            const int p2 = pix[-3*xstride];
114
            const int q0 = pix[0];
115
            const int q1 = pix[1*xstride];
116
            const int q2 = pix[2*xstride];
117

  
118
            if( FFABS( p0 - q0 ) < alpha &&
119
                FFABS( p1 - p0 ) < beta &&
120
                FFABS( q1 - q0 ) < beta ) {
121

  
122
                int tc = tc0[i];
123
                int i_delta;
124

  
125
                if( FFABS( p2 - p0 ) < beta ) {
126
                    if(tc0[i])
127
                    pix[-2*xstride] = p1 + av_clip( (( p2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - p1, -tc0[i], tc0[i] );
128
                    tc++;
129
                }
130
                if( FFABS( q2 - q0 ) < beta ) {
131
                    if(tc0[i])
132
                    pix[   xstride] = q1 + av_clip( (( q2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - q1, -tc0[i], tc0[i] );
133
                    tc++;
134
                }
135

  
136
                i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
137
                pix[-xstride] = av_clip_uint8( p0 + i_delta );    /* p0' */
138
                pix[0]        = av_clip_uint8( q0 - i_delta );    /* q0' */
139
            }
140
            pix += ystride;
141
        }
142
    }
143
}
144
static void h264_v_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
145
{
146
    h264_loop_filter_luma_c(pix, stride, 1, 4, alpha, beta, tc0);
147
}
148
static void h264_h_loop_filter_luma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
149
{
150
    h264_loop_filter_luma_c(pix, 1, stride, 4, alpha, beta, tc0);
151
}
152
static void h264_h_loop_filter_luma_mbaff_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
153
{
154
    h264_loop_filter_luma_c(pix, 1, stride, 2, alpha, beta, tc0);
155
}
156

  
157
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)
158
{
159
    int d;
160
    for( d = 0; d < 4 * inner_iters; d++ ) {
161
        const int p2 = pix[-3*xstride];
162
        const int p1 = pix[-2*xstride];
163
        const int p0 = pix[-1*xstride];
164

  
165
        const int q0 = pix[ 0*xstride];
166
        const int q1 = pix[ 1*xstride];
167
        const int q2 = pix[ 2*xstride];
168

  
169
        if( FFABS( p0 - q0 ) < alpha &&
170
            FFABS( p1 - p0 ) < beta &&
171
            FFABS( q1 - q0 ) < beta ) {
172

  
173
            if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
174
                if( FFABS( p2 - p0 ) < beta)
175
                {
176
                    const int p3 = pix[-4*xstride];
177
                    /* p0', p1', p2' */
178
                    pix[-1*xstride] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
179
                    pix[-2*xstride] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
180
                    pix[-3*xstride] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
181
                } else {
182
                    /* p0' */
183
                    pix[-1*xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
184
                }
185
                if( FFABS( q2 - q0 ) < beta)
186
                {
187
                    const int q3 = pix[3*xstride];
188
                    /* q0', q1', q2' */
189
                    pix[0*xstride] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
190
                    pix[1*xstride] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
191
                    pix[2*xstride] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
192
                } else {
193
                    /* q0' */
194
                    pix[0*xstride] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
195
                }
196
            }else{
197
                /* p0', q0' */
198
                pix[-1*xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
199
                pix[ 0*xstride] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
200
            }
201
        }
202
        pix += ystride;
203
    }
204
}
205
static void h264_v_loop_filter_luma_intra_c(uint8_t *pix, int stride, int alpha, int beta)
206
{
207
    h264_loop_filter_luma_intra_c(pix, stride, 1, 4, alpha, beta);
208
}
209
static void h264_h_loop_filter_luma_intra_c(uint8_t *pix, int stride, int alpha, int beta)
210
{
211
    h264_loop_filter_luma_intra_c(pix, 1, stride, 4, alpha, beta);
212
}
213
static void h264_h_loop_filter_luma_mbaff_intra_c(uint8_t *pix, int stride, int alpha, int beta)
214
{
215
    h264_loop_filter_luma_intra_c(pix, 1, stride, 2, alpha, beta);
216
}
217

  
218
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)
219
{
220
    int i, d;
221
    for( i = 0; i < 4; i++ ) {
222
        const int tc = tc0[i];
223
        if( tc <= 0 ) {
224
            pix += inner_iters*ystride;
225
            continue;
226
        }
227
        for( d = 0; d < inner_iters; d++ ) {
228
            const int p0 = pix[-1*xstride];
229
            const int p1 = pix[-2*xstride];
230
            const int q0 = pix[0];
231
            const int q1 = pix[1*xstride];
232

  
233
            if( FFABS( p0 - q0 ) < alpha &&
234
                FFABS( p1 - p0 ) < beta &&
235
                FFABS( q1 - q0 ) < beta ) {
236

  
237
                int delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
238

  
239
                pix[-xstride] = av_clip_uint8( p0 + delta );    /* p0' */
240
                pix[0]        = av_clip_uint8( q0 - delta );    /* q0' */
241
            }
242
            pix += ystride;
243
        }
244
    }
245
}
246
static void h264_v_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
247
{
248
    h264_loop_filter_chroma_c(pix, stride, 1, 2, alpha, beta, tc0);
249
}
250
static void h264_h_loop_filter_chroma_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
251
{
252
    h264_loop_filter_chroma_c(pix, 1, stride, 2, alpha, beta, tc0);
253
}
254
static void h264_h_loop_filter_chroma_mbaff_c(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
255
{
256
    h264_loop_filter_chroma_c(pix, 1, stride, 1, alpha, beta, tc0);
257
}
258

  
259
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)
260
{
261
    int d;
262
    for( d = 0; d < 4 * inner_iters; d++ ) {
263
        const int p0 = pix[-1*xstride];
264
        const int p1 = pix[-2*xstride];
265
        const int q0 = pix[0];
266
        const int q1 = pix[1*xstride];
267

  
268
        if( FFABS( p0 - q0 ) < alpha &&
269
            FFABS( p1 - p0 ) < beta &&
270
            FFABS( q1 - q0 ) < beta ) {
271

  
272
            pix[-xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2;   /* p0' */
273
            pix[0]        = ( 2*q1 + q0 + p1 + 2 ) >> 2;   /* q0' */
274
        }
275
        pix += ystride;
276
    }
277
}
278
static void h264_v_loop_filter_chroma_intra_c(uint8_t *pix, int stride, int alpha, int beta)
279
{
280
    h264_loop_filter_chroma_intra_c(pix, stride, 1, 2, alpha, beta);
281
}
282
static void h264_h_loop_filter_chroma_intra_c(uint8_t *pix, int stride, int alpha, int beta)
283
{
284
    h264_loop_filter_chroma_intra_c(pix, 1, stride, 2, alpha, beta);
285
}
286
static void h264_h_loop_filter_chroma_mbaff_intra_c(uint8_t *pix, int stride, int alpha, int beta)
287
{
288
    h264_loop_filter_chroma_intra_c(pix, 1, stride, 1, alpha, beta);
289
}
32
#include "h264dsp_template.c"
290 33

  
291 34
void ff_h264dsp_init(H264DSPContext *c)
292 35
{

Also available in: Unified diff