pixel.c 38.5 KB
Newer Older
Laurent Aimar's avatar
Laurent Aimar committed
1
/*****************************************************************************
Fiona Glaser's avatar
Fiona Glaser committed
2
 * pixel.c: pixel metrics
Laurent Aimar's avatar
Laurent Aimar committed
3
 *****************************************************************************
Fiona Glaser's avatar
Fiona Glaser committed
4
 * Copyright (C) 2003-2010 x264 project
Laurent Aimar's avatar
Laurent Aimar committed
5
 *
6
7
 * Authors: Loren Merritt <lorenm@u.washington.edu>
 *          Laurent Aimar <fenrir@via.ecp.fr>
Fiona Glaser's avatar
Fiona Glaser committed
8
 *          Fiona Glaser <fiona@x264.com>
Laurent Aimar's avatar
Laurent Aimar committed
9
10
11
12
13
14
15
16
17
18
19
20
21
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program 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 General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
Fiona Glaser's avatar
Fiona Glaser committed
23
24
25
 *
 * This program is also available under a commercial proprietary license.
 * For more information, contact us at licensing@x264.com.
Laurent Aimar's avatar
Laurent Aimar committed
26
27
 *****************************************************************************/

Loren Merritt's avatar
Loren Merritt committed
28
#include "common.h"
Laurent Aimar's avatar
Laurent Aimar committed
29

Steven Walters's avatar
Steven Walters committed
30
#if HAVE_MMX
31
#   include "x86/pixel.h"
Laurent Aimar's avatar
Laurent Aimar committed
32
#endif
Steven Walters's avatar
Steven Walters committed
33
#if ARCH_PPC
Laurent Aimar's avatar
Laurent Aimar committed
34
35
#   include "ppc/pixel.h"
#endif
Steven Walters's avatar
Steven Walters committed
36
#if ARCH_ARM
37
38
#   include "arm/pixel.h"
#endif
Sean McGovern's avatar
Sean McGovern committed
39
#if ARCH_UltraSPARC
40
41
#   include "sparc/pixel.h"
#endif
Laurent Aimar's avatar
Laurent Aimar committed
42
43
44
45
46
47


/****************************************************************************
 * pixel_sad_WxH
 ****************************************************************************/
#define PIXEL_SAD_C( name, lx, ly ) \
48
49
static int name( pixel *pix1, int i_stride_pix1,  \
                 pixel *pix2, int i_stride_pix2 ) \
Laurent Aimar's avatar
Laurent Aimar committed
50
51
{                                                   \
    int i_sum = 0;                                  \
52
    for( int y = 0; y < ly; y++ )                   \
Laurent Aimar's avatar
Laurent Aimar committed
53
    {                                               \
54
        for( int x = 0; x < lx; x++ )               \
Laurent Aimar's avatar
Laurent Aimar committed
55
56
57
58
59
60
61
62
63
64
        {                                           \
            i_sum += abs( pix1[x] - pix2[x] );      \
        }                                           \
        pix1 += i_stride_pix1;                      \
        pix2 += i_stride_pix2;                      \
    }                                               \
    return i_sum;                                   \
}


65
66
67
68
69
70
71
PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
Laurent Aimar's avatar
Laurent Aimar committed
72

73
74
75
76
77

/****************************************************************************
 * pixel_ssd_WxH
 ****************************************************************************/
#define PIXEL_SSD_C( name, lx, ly ) \
78
79
static int name( pixel *pix1, int i_stride_pix1,  \
                 pixel *pix2, int i_stride_pix2 ) \
80
81
{                                                   \
    int i_sum = 0;                                  \
82
    for( int y = 0; y < ly; y++ )                   \
83
    {                                               \
84
        for( int x = 0; x < lx; x++ )               \
85
86
87
88
89
90
91
92
93
94
        {                                           \
            int d = pix1[x] - pix2[x];              \
            i_sum += d*d;                           \
        }                                           \
        pix1 += i_stride_pix1;                      \
        pix2 += i_stride_pix2;                      \
    }                                               \
    return i_sum;                                   \
}

95
96
97
98
99
100
101
PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
102

103
uint64_t x264_pixel_ssd_wxh( x264_pixel_function_t *pf, pixel *pix1, int i_pix1, pixel *pix2, int i_pix2, int i_width, int i_height )
Loren Merritt's avatar
Loren Merritt committed
104
{
105
    uint64_t i_ssd = 0;
106
    int y;
Anton Mitrofanov's avatar
Anton Mitrofanov committed
107
    int align = !(((intptr_t)pix1 | (intptr_t)pix2 | i_pix1 | i_pix2) & 15);
Loren Merritt's avatar
Loren Merritt committed
108
109
110
111
112

#define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
                                          pix2 + y*i_pix2 + x, i_pix2 );
    for( y = 0; y < i_height-15; y += 16 )
    {
113
        int x = 0;
Loren Merritt's avatar
Loren Merritt committed
114
115
116
117
        if( align )
            for( ; x < i_width-15; x += 16 )
                SSD(PIXEL_16x16);
        for( ; x < i_width-7; x += 8 )
Loren Merritt's avatar
Loren Merritt committed
118
119
120
            SSD(PIXEL_8x16);
    }
    if( y < i_height-7 )
121
        for( int x = 0; x < i_width-7; x += 8 )
Loren Merritt's avatar
Loren Merritt committed
122
123
124
125
            SSD(PIXEL_8x8);
#undef SSD

#define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
126
    if( i_width & 7 )
Loren Merritt's avatar
Loren Merritt committed
127
128
    {
        for( y = 0; y < (i_height & ~7); y++ )
129
            for( int x = i_width & ~7; x < i_width; x++ )
Loren Merritt's avatar
Loren Merritt committed
130
131
                SSD1;
    }
132
    if( i_height & 7 )
Loren Merritt's avatar
Loren Merritt committed
133
134
    {
        for( y = i_height & ~7; y < i_height; y++ )
135
            for( int x = 0; x < i_width; x++ )
Loren Merritt's avatar
Loren Merritt committed
136
137
138
139
140
141
142
                SSD1;
    }
#undef SSD1

    return i_ssd;
}

143
static void pixel_ssd_nv12_core( pixel *pixuv1, int stride1, pixel *pixuv2, int stride2, int width, int height, uint64_t *ssd_u, uint64_t *ssd_v )
144
{
145
    *ssd_u = 0, *ssd_v = 0;
146
147
148
149
150
    for( int y = 0; y < height; y++, pixuv1+=stride1, pixuv2+=stride2 )
        for( int x = 0; x < width; x++ )
        {
            int du = pixuv1[2*x]   - pixuv2[2*x];
            int dv = pixuv1[2*x+1] - pixuv2[2*x+1];
151
152
            *ssd_u += du*du;
            *ssd_v += dv*dv;
153
154
155
        }
}

156
void x264_pixel_ssd_nv12( x264_pixel_function_t *pf, pixel *pix1, int i_pix1, pixel *pix2, int i_pix2, int i_width, int i_height, uint64_t *ssd_u, uint64_t *ssd_v )
157
{
158
    pf->ssd_nv12_core( pix1, i_pix1, pix2, i_pix2, i_width&~7, i_height, ssd_u, ssd_v );
159
    if( i_width&7 )
160
161
162
163
164
165
    {
        uint64_t tmp[2];
        pixel_ssd_nv12_core( pix1+(i_width&~7), i_pix1, pix2+(i_width&~7), i_pix2, i_width&7, i_height, &tmp[0], &tmp[1] );
        *ssd_u += tmp[0];
        *ssd_v += tmp[1];
    }
166
}
167

168
169
170
/****************************************************************************
 * pixel_var_wxh
 ****************************************************************************/
Fiona Glaser's avatar
Fiona Glaser committed
171
#define PIXEL_VAR_C( name, w ) \
172
static uint64_t name( pixel *pix, int i_stride ) \
173
{                                             \
Fiona Glaser's avatar
Fiona Glaser committed
174
    uint32_t sum = 0, sqr = 0;                \
175
    for( int y = 0; y < w; y++ )              \
176
    {                                         \
177
        for( int x = 0; x < w; x++ )          \
178
179
180
181
182
183
        {                                     \
            sum += pix[x];                    \
            sqr += pix[x] * pix[x];           \
        }                                     \
        pix += i_stride;                      \
    }                                         \
Fiona Glaser's avatar
Fiona Glaser committed
184
    return sum + ((uint64_t)sqr << 32);       \
185
186
}

Fiona Glaser's avatar
Fiona Glaser committed
187
188
PIXEL_VAR_C( x264_pixel_var_16x16, 16 )
PIXEL_VAR_C( x264_pixel_var_8x8,    8 )
189

190
191
192
/****************************************************************************
 * pixel_var2_wxh
 ****************************************************************************/
193
static int pixel_var2_8x8( pixel *pix1, int i_stride1, pixel *pix2, int i_stride2, int *ssd )
194
195
{
    uint32_t var = 0, sum = 0, sqr = 0;
196
    for( int y = 0; y < 8; y++ )
197
    {
198
        for( int x = 0; x < 8; x++ )
199
200
201
202
203
204
205
206
207
        {
            int diff = pix1[x] - pix2[x];
            sum += diff;
            sqr += diff * diff;
        }
        pix1 += i_stride1;
        pix2 += i_stride2;
    }
    sum = abs(sum);
208
    var = sqr - ((uint64_t)sum * sum >> 6);
209
210
211
212
    *ssd = sqr;
    return var;
}

213
214
215
216
217
218
219
220
#if BIT_DEPTH > 8
    typedef uint32_t sum_t;
    typedef uint64_t sum2_t;
#else
    typedef uint16_t sum_t;
    typedef uint32_t sum2_t;
#endif
#define BITS_PER_SUM (8 * sizeof(sum_t))
221

222
#define HADAMARD4(d0, d1, d2, d3, s0, s1, s2, s3) {\
223
224
225
226
    sum2_t t0 = s0 + s1;\
    sum2_t t1 = s0 - s1;\
    sum2_t t2 = s2 + s3;\
    sum2_t t3 = s2 - s3;\
Loren Merritt's avatar
Loren Merritt committed
227
228
229
230
231
    d0 = t0 + t2;\
    d2 = t0 - t2;\
    d1 = t1 + t3;\
    d3 = t1 - t3;\
}
232

233
234
// in: a pseudo-simd number of the form x+(y<<16)
// return: abs(x)+(abs(y)<<16)
235
static ALWAYS_INLINE sum2_t abs2( sum2_t a )
236
{
237
    sum2_t s = ((a>>(BITS_PER_SUM-1))&(((sum2_t)1<<BITS_PER_SUM)+1))*((sum_t)-1);
238
239
240
    return (a+s)^s;
}

241
242
243
/****************************************************************************
 * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 ****************************************************************************/
Laurent Aimar's avatar
Laurent Aimar committed
244

245
static NOINLINE int x264_pixel_satd_4x4( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
246
{
247
248
249
    sum2_t tmp[4][2];
    sum2_t a0, a1, a2, a3, b0, b1;
    sum2_t sum = 0;
250
    for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
Laurent Aimar's avatar
Laurent Aimar committed
251
    {
252
253
        a0 = pix1[0] - pix2[0];
        a1 = pix1[1] - pix2[1];
254
        b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
255
256
        a2 = pix1[2] - pix2[2];
        a3 = pix1[3] - pix2[3];
257
        b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
258
259
        tmp[i][0] = b0 + b1;
        tmp[i][1] = b0 - b1;
Laurent Aimar's avatar
Laurent Aimar committed
260
    }
261
    for( int i = 0; i < 2; i++ )
262
    {
263
        HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
264
        a0 = abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
265
        sum += ((sum_t)a0) + (a0>>BITS_PER_SUM);
266
267
    }
    return sum >> 1;
Laurent Aimar's avatar
Laurent Aimar committed
268
269
}

270
static NOINLINE int x264_pixel_satd_8x4( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
271
{
272
273
274
    sum2_t tmp[4][4];
    sum2_t a0, a1, a2, a3;
    sum2_t sum = 0;
275
    for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
276
    {
277
278
279
280
        a0 = (pix1[0] - pix2[0]) + ((sum2_t)(pix1[4] - pix2[4]) << BITS_PER_SUM);
        a1 = (pix1[1] - pix2[1]) + ((sum2_t)(pix1[5] - pix2[5]) << BITS_PER_SUM);
        a2 = (pix1[2] - pix2[2]) + ((sum2_t)(pix1[6] - pix2[6]) << BITS_PER_SUM);
        a3 = (pix1[3] - pix2[3]) + ((sum2_t)(pix1[7] - pix2[7]) << BITS_PER_SUM);
281
282
        HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
    }
283
    for( int i = 0; i < 4; i++ )
284
    {
285
        HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
286
287
        sum += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
    }
288
    return (((sum_t)sum) + (sum>>BITS_PER_SUM)) >> 1;
289
}
Laurent Aimar's avatar
Laurent Aimar committed
290

291
#define PIXEL_SATD_C( w, h, sub )\
292
static int x264_pixel_satd_##w##x##h( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )\
293
294
295
296
297
298
299
300
301
302
303
304
305
{\
    int sum = sub( pix1, i_pix1, pix2, i_pix2 )\
            + sub( pix1+4*i_pix1, i_pix1, pix2+4*i_pix2, i_pix2 );\
    if( w==16 )\
        sum+= sub( pix1+8, i_pix1, pix2+8, i_pix2 )\
            + sub( pix1+8+4*i_pix1, i_pix1, pix2+8+4*i_pix2, i_pix2 );\
    if( h==16 )\
        sum+= sub( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )\
            + sub( pix1+12*i_pix1, i_pix1, pix2+12*i_pix2, i_pix2 );\
    if( w==16 && h==16 )\
        sum+= sub( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 )\
            + sub( pix1+8+12*i_pix1, i_pix1, pix2+8+12*i_pix2, i_pix2 );\
    return sum;\
306
}
307
308
309
310
311
PIXEL_SATD_C( 16, 16, x264_pixel_satd_8x4 )
PIXEL_SATD_C( 16, 8,  x264_pixel_satd_8x4 )
PIXEL_SATD_C( 8,  16, x264_pixel_satd_8x4 )
PIXEL_SATD_C( 8,  8,  x264_pixel_satd_8x4 )
PIXEL_SATD_C( 4,  8,  x264_pixel_satd_4x4 )
312
313


314
static NOINLINE int sa8d_8x8( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
315
{
316
317
318
    sum2_t tmp[8][4];
    sum2_t a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3;
    sum2_t sum = 0;
319
    for( int i = 0; i < 8; i++, pix1 += i_pix1, pix2 += i_pix2 )
320
    {
321
322
        a0 = pix1[0] - pix2[0];
        a1 = pix1[1] - pix2[1];
323
        b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
324
325
        a2 = pix1[2] - pix2[2];
        a3 = pix1[3] - pix2[3];
326
        b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
327
328
        a4 = pix1[4] - pix2[4];
        a5 = pix1[5] - pix2[5];
329
        b2 = (a4+a5) + ((a4-a5)<<BITS_PER_SUM);
330
331
        a6 = pix1[6] - pix2[6];
        a7 = pix1[7] - pix2[7];
332
        b3 = (a6+a7) + ((a6-a7)<<BITS_PER_SUM);
333
        HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], b0,b1,b2,b3 );
334
    }
335
    for( int i = 0; i < 4; i++ )
336
    {
337
338
        HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
        HADAMARD4( a4, a5, a6, a7, tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i] );
339
340
341
342
        b0  = abs2(a0+a4) + abs2(a0-a4);
        b0 += abs2(a1+a5) + abs2(a1-a5);
        b0 += abs2(a2+a6) + abs2(a2-a6);
        b0 += abs2(a3+a7) + abs2(a3-a7);
343
        sum += (sum_t)b0 + (b0>>BITS_PER_SUM);
344
345
346
    }
    return sum;
}
347

348
static int x264_pixel_sa8d_8x8( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
349
350
351
{
    int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 );
    return (sum+2)>>2;
352
353
}

354
static int x264_pixel_sa8d_16x16( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
355
356
357
358
359
360
{
    int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 )
            + sa8d_8x8( pix1+8, i_pix1, pix2+8, i_pix2 )
            + sa8d_8x8( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )
            + sa8d_8x8( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 );
    return (sum+2)>>2;
361
362
}

Loren Merritt's avatar
Loren Merritt committed
363

364
static NOINLINE uint64_t pixel_hadamard_ac( pixel *pix, int stride )
Loren Merritt's avatar
Loren Merritt committed
365
{
366
367
368
    sum2_t tmp[32];
    sum2_t a0, a1, a2, a3, dc;
    sum2_t sum4 = 0, sum8 = 0;
369
    for( int i = 0; i < 8; i++, pix+=stride )
Loren Merritt's avatar
Loren Merritt committed
370
    {
371
372
373
        sum2_t *t = tmp + (i&3) + (i&4)*4;
        a0 = (pix[0]+pix[1]) + ((sum2_t)(pix[0]-pix[1])<<BITS_PER_SUM);
        a1 = (pix[2]+pix[3]) + ((sum2_t)(pix[2]-pix[3])<<BITS_PER_SUM);
374
375
        t[0] = a0 + a1;
        t[4] = a0 - a1;
376
377
        a2 = (pix[4]+pix[5]) + ((sum2_t)(pix[4]-pix[5])<<BITS_PER_SUM);
        a3 = (pix[6]+pix[7]) + ((sum2_t)(pix[6]-pix[7])<<BITS_PER_SUM);
378
379
        t[8] = a2 + a3;
        t[12] = a2 - a3;
Loren Merritt's avatar
Loren Merritt committed
380
    }
381
    for( int i = 0; i < 8; i++ )
Loren Merritt's avatar
Loren Merritt committed
382
    {
383
        HADAMARD4( a0, a1, a2, a3, tmp[i*4+0], tmp[i*4+1], tmp[i*4+2], tmp[i*4+3] );
384
385
386
387
388
        tmp[i*4+0] = a0;
        tmp[i*4+1] = a1;
        tmp[i*4+2] = a2;
        tmp[i*4+3] = a3;
        sum4 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
Loren Merritt's avatar
Loren Merritt committed
389
    }
390
    for( int i = 0; i < 8; i++ )
Loren Merritt's avatar
Loren Merritt committed
391
    {
392
393
        HADAMARD4( a0,a1,a2,a3, tmp[i], tmp[8+i], tmp[16+i], tmp[24+i] );
        sum8 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
Loren Merritt's avatar
Loren Merritt committed
394
    }
395
396
397
    dc = (sum_t)(tmp[0] + tmp[8] + tmp[16] + tmp[24]);
    sum4 = (sum_t)sum4 + (sum4>>BITS_PER_SUM) - dc;
    sum8 = (sum_t)sum8 + (sum8>>BITS_PER_SUM) - dc;
Loren Merritt's avatar
Loren Merritt committed
398
399
400
401
    return ((uint64_t)sum8<<32) + sum4;
}

#define HADAMARD_AC(w,h) \
402
static uint64_t x264_pixel_hadamard_ac_##w##x##h( pixel *pix, int stride )\
Loren Merritt's avatar
Loren Merritt committed
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
{\
    uint64_t sum = pixel_hadamard_ac( pix, stride );\
    if( w==16 )\
        sum += pixel_hadamard_ac( pix+8, stride );\
    if( h==16 )\
        sum += pixel_hadamard_ac( pix+8*stride, stride );\
    if( w==16 && h==16 )\
        sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
    return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
}
HADAMARD_AC( 16, 16 )
HADAMARD_AC( 16, 8 )
HADAMARD_AC( 8, 16 )
HADAMARD_AC( 8, 8 )


Loren Merritt's avatar
Loren Merritt committed
419
420
421
/****************************************************************************
 * pixel_sad_x4
 ****************************************************************************/
422
#define SAD_X( size ) \
423
static void x264_pixel_sad_x3_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, int i_stride, int scores[3] )\
424
425
426
427
428
{\
    scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
    scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
    scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
}\
429
static void x264_pixel_sad_x4_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3, int i_stride, int scores[4] )\
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
{\
    scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
    scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
    scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
    scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
}

SAD_X( 16x16 )
SAD_X( 16x8 )
SAD_X( 8x16 )
SAD_X( 8x8 )
SAD_X( 8x4 )
SAD_X( 4x8 )
SAD_X( 4x4 )

445
#if !HIGH_BIT_DEPTH
Sean McGovern's avatar
Sean McGovern committed
446
#if ARCH_UltraSPARC
447
448
449
450
451
SAD_X( 16x16_vis )
SAD_X( 16x8_vis )
SAD_X( 8x16_vis )
SAD_X( 8x8_vis )
#endif
452
#endif // !HIGH_BIT_DEPTH
453

454
455
456
457
458
459
/****************************************************************************
 * pixel_satd_x4
 * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 ****************************************************************************/

#define SATD_X( size, cpu ) \
460
static void x264_pixel_satd_x3_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, int i_stride, int scores[3] )\
461
462
463
464
465
{\
    scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
    scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
    scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
}\
466
static void x264_pixel_satd_x4_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3, int i_stride, int scores[4] )\
467
468
469
470
471
472
{\
    scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
    scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
    scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
    scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
}
473
#define SATD_X_DECL6( cpu )\
474
475
476
477
SATD_X( 16x16, cpu )\
SATD_X( 16x8, cpu )\
SATD_X( 8x16, cpu )\
SATD_X( 8x8, cpu )\
478
479
SATD_X( 8x4, cpu )\
SATD_X( 4x8, cpu )
480
#define SATD_X_DECL7( cpu )\
481
SATD_X_DECL6( cpu )\
482
483
484
SATD_X( 4x4, cpu )

SATD_X_DECL7()
Steven Walters's avatar
Steven Walters committed
485
#if HAVE_MMX
486
SATD_X_DECL7( _mmxext )
487
#if !HIGH_BIT_DEPTH
488
SATD_X_DECL6( _sse2 )
Fiona Glaser's avatar
Fiona Glaser committed
489
SATD_X_DECL7( _ssse3 )
490
SATD_X_DECL7( _sse4 )
491
#endif // !HIGH_BIT_DEPTH
492
493
#endif

494
#if !HIGH_BIT_DEPTH
Steven Walters's avatar
Steven Walters committed
495
#if HAVE_ARMV6
496
497
SATD_X_DECL7( _neon )
#endif
498
#endif // !HIGH_BIT_DEPTH
499

500
501
#define INTRA_MBCMP_8x8( mbcmp, cpu )\
void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[33], int res[3] )\
502
{\
503
    ALIGNED_ARRAY_16( pixel, pix, [8*FDEC_STRIDE] );\
504
    x264_predict_8x8_v_c( pix, edge );\
505
    res[0] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
506
    x264_predict_8x8_h_c( pix, edge );\
507
    res[1] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
508
    x264_predict_8x8_dc_c( pix, edge );\
509
    res[2] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
510
511
}

512
513
INTRA_MBCMP_8x8( sad, )
INTRA_MBCMP_8x8(sa8d, )
514
#if HIGH_BIT_DEPTH && HAVE_MMX
515
516
517
518
519
INTRA_MBCMP_8x8( sad, _mmxext)
INTRA_MBCMP_8x8( sad, _sse2  )
INTRA_MBCMP_8x8( sad, _ssse3 )
INTRA_MBCMP_8x8(sa8d, _sse2  )
#endif
520

521
522
#define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma, cpu )\
void x264_intra_##mbcmp##_x3_##size##x##size##chroma##cpu( pixel *fenc, pixel *fdec, int res[3] )\
523
524
{\
    x264_predict_##size##x##size##chroma##_##pred1##_c( fdec );\
525
    res[0] = x264_pixel_##mbcmp##_##size##x##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
526
    x264_predict_##size##x##size##chroma##_##pred2##_c( fdec );\
527
    res[1] = x264_pixel_##mbcmp##_##size##x##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
528
    x264_predict_##size##x##size##chroma##_##pred3##_c( fdec );\
529
    res[2] = x264_pixel_##mbcmp##_##size##x##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
530
531
}

532
533
534
535
536
537
538
INTRA_MBCMP( sad,  4,  v, h, dc,  , )
INTRA_MBCMP(satd,  4,  v, h, dc,  , )
INTRA_MBCMP( sad,  8, dc, h,  v, c, )
INTRA_MBCMP(satd,  8, dc, h,  v, c, )
INTRA_MBCMP( sad, 16,  v, h, dc,  , )
INTRA_MBCMP(satd, 16,  v, h, dc,  , )

539
#if HIGH_BIT_DEPTH && HAVE_MMX
540
541
542
543
544
545
546
547
548
549
550
INTRA_MBCMP( sad,  4,  v, h, dc,  , _mmxext)
INTRA_MBCMP(satd,  4,  v, h, dc,  , _mmxext)
INTRA_MBCMP( sad,  8, dc, h,  v, c, _mmxext)
INTRA_MBCMP(satd,  8, dc, h,  v, c, _mmxext)
INTRA_MBCMP( sad, 16,  v, h, dc,  , _mmxext)
INTRA_MBCMP(satd, 16,  v, h, dc,  , _mmxext)
INTRA_MBCMP( sad,  8, dc, h,  v, c, _sse2  )
INTRA_MBCMP( sad, 16,  v, h, dc,  , _sse2  )
INTRA_MBCMP( sad,  8, dc, h,  v, c, _ssse3 )
INTRA_MBCMP( sad, 16,  v, h, dc,  , _ssse3 )
#endif
551

Loren Merritt's avatar
Loren Merritt committed
552
553
554
/****************************************************************************
 * structural similarity metric
 ****************************************************************************/
555
556
static void ssim_4x4x2_core( const pixel *pix1, int stride1,
                             const pixel *pix2, int stride2,
557
558
                             int sums[2][4])
{
559
    for( int z = 0; z < 2; z++ )
560
    {
561
562
563
        uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
        for( int y = 0; y < 4; y++ )
            for( int x = 0; x < 4; x++ )
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
            {
                int a = pix1[x+y*stride1];
                int b = pix2[x+y*stride2];
                s1  += a;
                s2  += b;
                ss  += a*a;
                ss  += b*b;
                s12 += a*b;
            }
        sums[z][0] = s1;
        sums[z][1] = s2;
        sums[z][2] = ss;
        sums[z][3] = s12;
        pix1 += 4;
        pix2 += 4;
    }
}

static float ssim_end1( int s1, int s2, int ss, int s12 )
{
584
585
586
587
588
589
590
591
592
/* Maximum value for 10-bit is: ss*64 = (2^10-1)^2*16*4*64 = 4286582784, which will overflow in some cases.
 * s1*s1, s2*s2, and s1*s2 also obtain this value for edge cases: ((2^10-1)*16*4)^2 = 4286582784.
 * Maximum value for 9-bit is: ss*64 = (2^9-1)^2*16*4*64 = 1069551616, which will not overflow. */
#if BIT_DEPTH > 9
#define type float
    static const float ssim_c1 = .01*.01*PIXEL_MAX*PIXEL_MAX*64;
    static const float ssim_c2 = .03*.03*PIXEL_MAX*PIXEL_MAX*64*63;
#else
#define type int
593
594
    static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
    static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
595
596
597
598
599
600
601
602
603
604
#endif
    type fs1 = s1;
    type fs2 = s2;
    type fss = ss;
    type fs12 = s12;
    type vars = fss*64 - fs1*fs1 - fs2*fs2;
    type covar = fs12*64 - fs1*fs2;
    return (float)(2*fs1*fs2 + ssim_c1) * (float)(2*covar + ssim_c2)
         / ((float)(fs1*fs1 + fs2*fs2 + ssim_c1) * (float)(vars + ssim_c2));
#undef type
605
606
607
608
609
}

static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
{
    float ssim = 0.0;
610
    for( int i = 0; i < width; i++ )
611
612
613
614
615
616
617
618
        ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
                           sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
                           sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
                           sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
    return ssim;
}

float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
619
620
                           pixel *pix1, int stride1,
                           pixel *pix2, int stride2,
621
                           int width, int height, void *buf )
622
{
623
    int z = 0;
624
    float ssim = 0.0;
625
    int (*sum0)[4] = buf;
626
    int (*sum1)[4] = sum0 + (width >> 2) + 3;
627
628
    width >>= 2;
    height >>= 2;
629
    for( int y = 1; y < height; y++ )
630
631
632
633
    {
        for( ; z <= y; z++ )
        {
            XCHG( void*, sum0, sum1 );
634
            for( int x = 0; x < width; x+=2 )
635
636
                pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
        }
637
        for( int x = 0; x < width-1; x += 4 )
638
639
            ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
    }
640
    return ssim;
641
642
643
}


Loren Merritt's avatar
Loren Merritt committed
644
645
646
/****************************************************************************
 * successive elimination
 ****************************************************************************/
647
648
static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
                            uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
Loren Merritt's avatar
Loren Merritt committed
649
{
650
651
    int nmv = 0;
    for( int i = 0; i < width; i++, sums++ )
652
653
654
655
656
657
658
659
660
661
    {
        int ads = abs( enc_dc[0] - sums[0] )
                + abs( enc_dc[1] - sums[8] )
                + abs( enc_dc[2] - sums[delta] )
                + abs( enc_dc[3] - sums[delta+8] )
                + cost_mvx[i];
        if( ads < thresh )
            mvs[nmv++] = i;
    }
    return nmv;
Loren Merritt's avatar
Loren Merritt committed
662
663
}

664
665
static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
                            uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
Loren Merritt's avatar
Loren Merritt committed
666
{
667
668
    int nmv = 0;
    for( int i = 0; i < width; i++, sums++ )
669
670
671
672
673
674
675
676
    {
        int ads = abs( enc_dc[0] - sums[0] )
                + abs( enc_dc[1] - sums[delta] )
                + cost_mvx[i];
        if( ads < thresh )
            mvs[nmv++] = i;
    }
    return nmv;
Loren Merritt's avatar
Loren Merritt committed
677
678
}

679
680
static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
                            uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
Loren Merritt's avatar
Loren Merritt committed
681
{
682
683
    int nmv = 0;
    for( int i = 0; i<width; i++, sums++ )
684
685
686
687
688
689
690
    {
        int ads = abs( enc_dc[0] - sums[0] )
                + cost_mvx[i];
        if( ads < thresh )
            mvs[nmv++] = i;
    }
    return nmv;
Loren Merritt's avatar
Loren Merritt committed
691
692
693
}


Laurent Aimar's avatar
Laurent Aimar committed
694
695
696
697
698
/****************************************************************************
 * x264_pixel_init:
 ****************************************************************************/
void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
{
699
700
    memset( pixf, 0, sizeof(*pixf) );

701
702
703
704
705
706
707
708
709
710
#define INIT2_NAME( name1, name2, cpu ) \
    pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
    pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
#define INIT4_NAME( name1, name2, cpu ) \
    INIT2_NAME( name1, name2, cpu ) \
    pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
    pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
#define INIT5_NAME( name1, name2, cpu ) \
    INIT4_NAME( name1, name2, cpu ) \
    pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
711
#define INIT6_NAME( name1, name2, cpu ) \
712
    INIT5_NAME( name1, name2, cpu ) \
713
714
715
    pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
#define INIT7_NAME( name1, name2, cpu ) \
    INIT6_NAME( name1, name2, cpu ) \
716
717
718
719
    pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
#define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
#define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
#define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
720
#define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
721
#define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
722

723
724
725
726
727
#define INIT_ADS( cpu ) \
    pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
    pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
    pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;

728
    INIT7( sad, );
729
    INIT7_NAME( sad_aligned, sad, );
730
731
732
733
    INIT7( sad_x3, );
    INIT7( sad_x4, );
    INIT7( ssd, );
    INIT7( satd, );
734
735
    INIT7( satd_x3, );
    INIT7( satd_x4, );
Loren Merritt's avatar
Loren Merritt committed
736
    INIT4( hadamard_ac, );
737
    INIT_ADS( );
738

739
740
    pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
    pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
741
742
743
    pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
    pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;

744
    pixf->ssd_nv12_core = pixel_ssd_nv12_core;
745
746
    pixf->ssim_4x4x2_core = ssim_4x4x2_core;
    pixf->ssim_end4 = ssim_end4;
747
    pixf->var2_8x8 = pixel_var2_8x8;
748

749
750
751
752
753
754
755
756
757
    pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4;
    pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4;
    pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8;
    pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8;
    pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c;
    pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c;
    pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16;
    pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16;

758
#if HIGH_BIT_DEPTH
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
#if HAVE_MMX
    if( cpu&X264_CPU_MMXEXT )
    {
        INIT7( sad, _mmxext );
        INIT7( sad_x3, _mmxext );
        INIT7( sad_x4, _mmxext );
        INIT7( satd, _mmxext );
        INIT7( satd_x3, _mmxext );
        INIT7( satd_x4, _mmxext );
        INIT4( hadamard_ac, _mmxext );
        INIT7( ssd, _mmxext );
        INIT_ADS( _mmxext );

        pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_mmxext;
        pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmxext;
        pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmxext;
        pixf->var2_8x8 = x264_pixel_var2_8x8_mmxext;

        pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmxext;
        pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
        pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmxext;
        pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmxext;
        pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
        pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmxext;
        pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
    }
    if( cpu&X264_CPU_SSE2 )
    {
        INIT4_NAME( sad_aligned, sad, _sse2_aligned );
        INIT5( ssd, _sse2 );

        pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
        pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
#if ARCH_X86_64
        pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
#endif
        pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_sse2;
        pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
        pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
        pixf->var2_8x8 = x264_pixel_var2_8x8_sse2;
    }
    if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
    {
        INIT5( sad, _sse2 );
        INIT2( sad_x3, _sse2 );
        INIT2( sad_x4, _sse2 );
        INIT_ADS( _sse2 );

        if( !(cpu&X264_CPU_STACK_MOD4) )
        {
            INIT4( hadamard_ac, _sse2 );
        }

        pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_sse2;
        pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_sse2;
        pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_sse2;
    }
    if( cpu&X264_CPU_SSE2_IS_FAST )
    {
        pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
        pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
        pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_sse2;
        pixf->sad_x3[PIXEL_8x4]  = x264_pixel_sad_x3_8x4_sse2;
        pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
        pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_sse2;
        pixf->sad_x4[PIXEL_8x4]  = x264_pixel_sad_x4_8x4_sse2;
    }
    if( cpu&X264_CPU_SSSE3 )
    {
        INIT7( sad, _ssse3 );
        INIT7( sad_x3, _ssse3 );
        INIT7( sad_x4, _ssse3 );
        INIT_ADS( _ssse3 );

        if( !(cpu&X264_CPU_STACK_MOD4) )
        {
            INIT4( hadamard_ac, _ssse3 );
        }

        pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
        pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
        pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_ssse3;
        pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
        pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
    }
#endif // HAVE_MMX
845
#else // !HIGH_BIT_DEPTH
Steven Walters's avatar
Steven Walters committed
846
#if HAVE_MMX
847
848
    if( cpu&X264_CPU_MMX )
    {
849
        INIT7( ssd, _mmx );
850
851
    }

Laurent Aimar's avatar
Laurent Aimar committed
852
853
    if( cpu&X264_CPU_MMXEXT )
    {
854
        INIT7( sad, _mmxext );
855
        INIT7_NAME( sad_aligned, sad, _mmxext );
856
857
858
        INIT7( sad_x3, _mmxext );
        INIT7( sad_x4, _mmxext );
        INIT7( satd, _mmxext );
859
860
        INIT7( satd_x3, _mmxext );
        INIT7( satd_x4, _mmxext );
Loren Merritt's avatar
Loren Merritt committed
861
        INIT4( hadamard_ac, _mmxext );
862
        INIT_ADS( _mmxext );
863
864
        pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmxext;
        pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmxext;
865
        pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_mmxext;
Steven Walters's avatar
Steven Walters committed
866
#if ARCH_X86
867
868
        pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmxext;
        pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmxext;
869
        pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmxext;
870
        pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_mmxext;
871
        pixf->var2_8x8 = x264_pixel_var2_8x8_mmxext;
872

873
        if( cpu&X264_CPU_CACHELINE_32 )
874
        {
875
876
877
878
879
880
881
882
883
            INIT5( sad, _cache32_mmxext );
            INIT4( sad_x3, _cache32_mmxext );
            INIT4( sad_x4, _cache32_mmxext );
        }
        else if( cpu&X264_CPU_CACHELINE_64 )
        {
            INIT5( sad, _cache64_mmxext );
            INIT4( sad_x3, _cache64_mmxext );
            INIT4( sad_x4, _cache64_mmxext );
884
885
        }
#else
886
        if( cpu&X264_CPU_CACHELINE_64 )
887
888
889
890
891
892
893
894
895
        {
            pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmxext;
            pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmxext;
            pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmxext;
            pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmxext;
            pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmxext;
            pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmxext;
            pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmxext;
        }
896
#endif
897
        pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
Fiona Glaser's avatar
Fiona Glaser committed
898
        pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmxext;
899
        pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
Fiona Glaser's avatar
Fiona Glaser committed
900
        pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmxext;
Fiona Glaser's avatar
Fiona Glaser committed
901
        pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmxext;
902
        pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
Fiona Glaser's avatar
Fiona Glaser committed
903
        pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmxext;
Laurent Aimar's avatar
Laurent Aimar committed
904
    }
905

906
907
908
909
910
    if( cpu&X264_CPU_SSE2 )
    {
        INIT5( ssd, _sse2slow );
        INIT2_NAME( sad_aligned, sad, _sse2_aligned );
        pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
911
        pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_sse2;
912
913
914
915
        pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
        pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
        pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
        pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
Steven Walters's avatar
Steven Walters committed
916
#if ARCH_X86_64
917
918
        pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
#endif
919
        pixf->var2_8x8 = x264_pixel_var2_8x8_sse2;
920
921
    }

922
    if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
923
    {
924
925
926
        INIT2( sad, _sse2 );
        INIT2( sad_x3, _sse2 );
        INIT2( sad_x4, _sse2 );
927
928
929
        INIT6( satd, _sse2 );
        INIT6( satd_x3, _sse2 );
        INIT6( satd_x4, _sse2 );
930
931
932
933
        if( !(cpu&X264_CPU_STACK_MOD4) )
        {
            INIT4( hadamard_ac, _sse2 );
        }
934
        INIT_ADS( _sse2 );
935
        pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
936
        pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
937
        if( cpu&X264_CPU_CACHELINE_64 )
938
        {
939
            INIT2( ssd, _sse2); /* faster for width 16 on p4 */
Steven Walters's avatar
Steven Walters committed
940
#if ARCH_X86
941
942
943
944
            INIT2( sad, _cache64_sse2 );
            INIT2( sad_x3, _cache64_sse2 );
            INIT2( sad_x4, _cache64_sse2 );
#endif
945
946
947
948
949
950
951
           if( cpu&X264_CPU_SSE2_IS_FAST )
           {
               pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
               pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
           }
        }

Fiona Glaser's avatar
Fiona Glaser committed
952
953
954
955
956
        if( cpu&X264_CPU_SSE_MISALIGN )
        {
            INIT2( sad_x3, _sse2_misalign );
            INIT2( sad_x4, _sse2_misalign );
        }
957
    }
958

959
960
961
962
963
964
965
966
967
968
969
970
    if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
    {
        pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
        pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
        pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
        pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
        pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
        pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
        pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
        pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
    }

971
    if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
972
973
974
975
976
977
    {
        INIT2( sad, _sse3 );
        INIT2( sad_x3, _sse3 );
        INIT2( sad_x4, _sse3 );
    }

978
979
    if( cpu&X264_CPU_SSSE3 )
    {
980
981
982
983
        if( !(cpu&X264_CPU_STACK_MOD4) )
        {
            INIT4( hadamard_ac, _ssse3 );
        }
984
        INIT_ADS( _ssse3 );
985
986
987
988
989
990
991
992
993
        if( !(cpu&X264_CPU_SLOW_ATOM) )
        {
            INIT7( ssd, _ssse3 );
            pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
            pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
            INIT7( satd, _ssse3 );
            INIT7( satd_x3, _ssse3 );
            INIT7( satd_x4, _ssse3 );
        }
994
        pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
995
        pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
996
        pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
Fiona Glaser's avatar
Fiona Glaser committed
997
        pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
998
        pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_ssse3;
Steven Walters's avatar
Steven Walters committed
999
#if ARCH_X86_64
1000
        pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_ssse3;
Loren Merritt's avatar
Loren Merritt committed
1001
#endif
1002
        pixf->var2_8x8 = x264_pixel_var2_8x8_ssse3;
1003
        if( cpu&X264_CPU_CACHELINE_64 )
1004
1005
1006
1007
1008
        {
            INIT2( sad, _cache64_ssse3 );
            INIT2( sad_x3, _cache64_ssse3 );
            INIT2( sad_x4, _cache64_ssse3 );
        }
1009
        if( cpu&X264_CPU_SLOW_ATOM || !(cpu&X264_CPU_SHUFFLE_IS_FAST) )
1010
        {
1011
            INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
1012
        }
1013
    }
1014
1015
1016

    if( cpu&X264_CPU_SSE4 )
    {
1017
1018
1019
1020
1021
1022
1023
1024
1025
        INIT7( satd, _sse4 );
        INIT7( satd_x3, _sse4 );
        INIT7( satd_x4, _sse4 );
        if( !(cpu&X264_CPU_STACK_MOD4) )
        {
            INIT4( hadamard_ac, _sse4 );
        }
        pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
        pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
1026
1027
1028
        pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_sse4;
        /* Slower on Conroe, so only enable under SSE4 */
        pixf->intra_sad_x3_8x8  = x264_intra_sad_x3_8x8_ssse3;
1029
    }
1030
#endif //HAVE_MMX
1031

Steven Walters's avatar
Steven Walters committed
1032
#if HAVE_ARMV6
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
    if( cpu&X264_CPU_ARMV6 )
    {
        pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
        pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
        pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
        pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
    }
    if( cpu&X264_CPU_NEON )
    {
        INIT5( sad, _neon );
        INIT5( sad_aligned, _neon );
        INIT7( sad_x3, _neon );
        INIT7( sad_x4, _neon );
        INIT7( ssd, _neon );
        INIT7( satd, _neon );
        INIT7( satd_x3, _neon );
        INIT7( satd_x4, _neon );
        INIT4( hadamard_ac, _neon );
        pixf->sa8d[PIXEL_8x8]