Statistics
| Branch: | Revision:

ffmpeg / libavcodec / ppc / dsputil_h264_template_altivec.c @ 11c19637

History | View | Annotate | Download (27.5 KB)

1
/*
2
 * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
3
 *
4
 * This library is free software; you can redistribute it and/or
5
 * modify it under the terms of the GNU Lesser General Public
6
 * License as published by the Free Software Foundation; either
7
 * version 2 of the License, or (at your option) any later version.
8
 *
9
 * This library is distributed in the hope that it will be useful,
10
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
12
 * Lesser General Public License for more details.
13
 *
14
 * You should have received a copy of the GNU Lesser General Public
15
 * License along with this library; if not, write to the Free Software
16
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
 */
18

    
19
/* this code assume that stride % 16 == 0 */
20
void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
21
  POWERPC_PERF_DECLARE(PREFIX_h264_chroma_mc8_num, 1);
22
  POWERPC_PERF_START_COUNT(PREFIX_h264_chroma_mc8_num, 1);
23
    signed int ABCD[4] __attribute__((aligned(16)));
24
    register int i;
25
    ABCD[0] = ((8 - x) * (8 - y));
26
    ABCD[1] = ((x) * (8 - y));
27
    ABCD[2] = ((8 - x) * (y));
28
    ABCD[3] = ((x) * (y));
29
    const vector signed int vABCD = vec_ld(0, ABCD);
30
    const vector signed short vA = vec_splat((vector signed short)vABCD, 1);
31
    const vector signed short vB = vec_splat((vector signed short)vABCD, 3);
32
    const vector signed short vC = vec_splat((vector signed short)vABCD, 5);
33
    const vector signed short vD = vec_splat((vector signed short)vABCD, 7);
34
    const vector signed int vzero = vec_splat_s32(0);
35
    const vector signed short v32ss = vec_sl(vec_splat_s16(1),vec_splat_u16(5));
36
    const vector unsigned short v6us = vec_splat_u16(6);
37

    
38
    vector unsigned char fperm;
39

    
40
    if (((unsigned long)dst) % 16 == 0) {
41
      fperm = (vector unsigned char)AVV(0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
42
                                        0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F);
43
    } else {
44
      fperm = (vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
45
                                        0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
46
    }
47

    
48
    register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
49
    register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
50

    
51
    vector unsigned char vsrcAuc;
52
    vector unsigned char vsrcBuc;
53
    vector unsigned char vsrcperm0;
54
    vector unsigned char vsrcperm1;
55
    vsrcAuc = vec_ld(0, src);
56
    if (loadSecond)
57
      vsrcBuc = vec_ld(16, src);
58
    vsrcperm0 = vec_lvsl(0, src);
59
    vsrcperm1 = vec_lvsl(1, src);
60

    
61
    vector unsigned char vsrc0uc;
62
    vector unsigned char vsrc1uc;
63
    vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0);
64
    if (reallyBadAlign)
65
      vsrc1uc = vsrcBuc;
66
    else
67
      vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);
68

    
69
    vector signed short vsrc0ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc0uc);
70
    vector signed short vsrc1ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc1uc);
71

    
72
    if (!loadSecond) {// -> !reallyBadAlign
73
      for (i = 0 ; i < h ; i++) {
74
        vector unsigned char vsrcCuc;
75
        vsrcCuc = vec_ld(stride + 0, src);
76

    
77
        vector unsigned char vsrc2uc;
78
        vector unsigned char vsrc3uc;
79
        vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
80
        vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);
81

    
82
        vector signed short vsrc2ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc2uc);
83
        vector signed short vsrc3ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc3uc);
84

    
85
        vector signed short psum;
86

    
87
        psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
88
        psum = vec_mladd(vB, vsrc1ssH, psum);
89
        psum = vec_mladd(vC, vsrc2ssH, psum);
90
        psum = vec_mladd(vD, vsrc3ssH, psum);
91
        psum = vec_add(v32ss, psum);
92
        psum = vec_sra(psum, v6us);
93

    
94
        vector unsigned char vdst = vec_ld(0, dst);
95
        vector unsigned char ppsum = (vector unsigned char)vec_packsu(psum, psum);
96

    
97
        vector unsigned char vfdst = vec_perm(vdst, ppsum, fperm);
98
        vector unsigned char fsum;
99

    
100
        OP_U8_ALTIVEC(fsum, vfdst, vdst);
101

    
102
        vec_st(fsum, 0, dst);
103

    
104
        vsrc0ssH = vsrc2ssH;
105
        vsrc1ssH = vsrc3ssH;
106

    
107
        dst += stride;
108
        src += stride;
109
      }
110
    } else {
111
      for (i = 0 ; i < h ; i++) {
112
        vector unsigned char vsrcCuc;
113
        vector unsigned char vsrcDuc;
114
        vsrcCuc = vec_ld(stride + 0, src);
115
        vsrcDuc = vec_ld(stride + 16, src);
116

    
117
        vector unsigned char vsrc2uc;
118
        vector unsigned char vsrc3uc;
119
        vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0);
120
        if (reallyBadAlign)
121
          vsrc3uc = vsrcDuc;
122
        else
123
          vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);
124

    
125
        vector signed short vsrc2ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc2uc);
126
        vector signed short vsrc3ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc3uc);
127

    
128
        vector signed short psum;
129

    
130
        psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
131
        psum = vec_mladd(vB, vsrc1ssH, psum);
132
        psum = vec_mladd(vC, vsrc2ssH, psum);
133
        psum = vec_mladd(vD, vsrc3ssH, psum);
134
        psum = vec_add(v32ss, psum);
135
        psum = vec_sr(psum, v6us);
136

    
137
        vector unsigned char vdst = vec_ld(0, dst);
138
        vector unsigned char ppsum = (vector unsigned char)vec_pack(psum, psum);
139

    
140
        vector unsigned char vfdst = vec_perm(vdst, ppsum, fperm);
141
        vector unsigned char fsum;
142

    
143
        OP_U8_ALTIVEC(fsum, vfdst, vdst);
144

    
145
        vec_st(fsum, 0, dst);
146

    
147
        vsrc0ssH = vsrc2ssH;
148
        vsrc1ssH = vsrc3ssH;
149

    
150
        dst += stride;
151
        src += stride;
152
      }
153
    }
154
    POWERPC_PERF_STOP_COUNT(PREFIX_h264_chroma_mc8_num, 1);
155
}
156

    
157
/* this code assume stride % 16 == 0 */
158
static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
159
  POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_h_lowpass_num, 1);
160
  POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);
161
  register int i;
162

    
163
  const vector signed int vzero = vec_splat_s32(0);
164
  const vector unsigned char permM2 = vec_lvsl(-2, src);
165
  const vector unsigned char permM1 = vec_lvsl(-1, src);
166
  const vector unsigned char permP0 = vec_lvsl(+0, src);
167
  const vector unsigned char permP1 = vec_lvsl(+1, src);
168
  const vector unsigned char permP2 = vec_lvsl(+2, src);
169
  const vector unsigned char permP3 = vec_lvsl(+3, src);
170
  const vector signed short v5ss = vec_splat_s16(5);
171
  const vector unsigned short v5us = vec_splat_u16(5);
172
  const vector signed short v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
173
  const vector signed short v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4));
174
  const vector unsigned char dstperm = vec_lvsr(0, dst);
175
  const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
176
  const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
177

    
178
  register int align = ((((unsigned long)src) - 2) % 16);
179

    
180
  for (i = 0 ; i < 16 ; i ++) {
181
    vector unsigned char srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
182
    vector unsigned char srcR1 = vec_ld(-2, src);
183
    vector unsigned char srcR2 = vec_ld(14, src);
184

    
185
    switch (align) {
186
    default: {
187
      srcM2 = vec_perm(srcR1, srcR2, permM2);
188
      srcM1 = vec_perm(srcR1, srcR2, permM1);
189
      srcP0 = vec_perm(srcR1, srcR2, permP0);
190
      srcP1 = vec_perm(srcR1, srcR2, permP1);
191
      srcP2 = vec_perm(srcR1, srcR2, permP2);
192
      srcP3 = vec_perm(srcR1, srcR2, permP3);
193
    } break;
194
    case 11: {
195
      srcM2 = vec_perm(srcR1, srcR2, permM2);
196
      srcM1 = vec_perm(srcR1, srcR2, permM1);
197
      srcP0 = vec_perm(srcR1, srcR2, permP0);
198
      srcP1 = vec_perm(srcR1, srcR2, permP1);
199
      srcP2 = vec_perm(srcR1, srcR2, permP2);
200
      srcP3 = srcR2;
201
    } break;
202
    case 12: {
203
      vector unsigned char srcR3 = vec_ld(30, src);
204
      srcM2 = vec_perm(srcR1, srcR2, permM2);
205
      srcM1 = vec_perm(srcR1, srcR2, permM1);
206
      srcP0 = vec_perm(srcR1, srcR2, permP0);
207
      srcP1 = vec_perm(srcR1, srcR2, permP1);
208
      srcP2 = srcR2;
209
      srcP3 = vec_perm(srcR2, srcR3, permP3);
210
    } break;
211
    case 13: {
212
      vector unsigned char srcR3 = vec_ld(30, src);
213
      srcM2 = vec_perm(srcR1, srcR2, permM2);
214
      srcM1 = vec_perm(srcR1, srcR2, permM1);
215
      srcP0 = vec_perm(srcR1, srcR2, permP0);
216
      srcP1 = srcR2;
217
      srcP2 = vec_perm(srcR2, srcR3, permP2);
218
      srcP3 = vec_perm(srcR2, srcR3, permP3);
219
    } break;
220
    case 14: {
221
      vector unsigned char srcR3 = vec_ld(30, src);
222
      srcM2 = vec_perm(srcR1, srcR2, permM2);
223
      srcM1 = vec_perm(srcR1, srcR2, permM1);
224
      srcP0 = srcR2;
225
      srcP1 = vec_perm(srcR2, srcR3, permP1);
226
      srcP2 = vec_perm(srcR2, srcR3, permP2);
227
      srcP3 = vec_perm(srcR2, srcR3, permP3);
228
    } break;
229
    case 15: {
230
      vector unsigned char srcR3 = vec_ld(30, src);
231
      srcM2 = vec_perm(srcR1, srcR2, permM2);
232
      srcM1 = srcR2;
233
      srcP0 = vec_perm(srcR2, srcR3, permP0);
234
      srcP1 = vec_perm(srcR2, srcR3, permP1);
235
      srcP2 = vec_perm(srcR2, srcR3, permP2);
236
      srcP3 = vec_perm(srcR2, srcR3, permP3);
237
    } break;
238
    }
239

    
240
    const vector signed short srcP0A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
241
    const vector signed short srcP0B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
242
    const vector signed short srcP1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
243
    const vector signed short srcP1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
244

    
245
    const vector signed short srcP2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
246
    const vector signed short srcP2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
247
    const vector signed short srcP3A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
248
    const vector signed short srcP3B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
249

    
250
    const vector signed short srcM1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
251
    const vector signed short srcM1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
252
    const vector signed short srcM2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
253
    const vector signed short srcM2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
254

    
255
    const vector signed short sum1A = vec_adds(srcP0A, srcP1A);
256
    const vector signed short sum1B = vec_adds(srcP0B, srcP1B);
257
    const vector signed short sum2A = vec_adds(srcM1A, srcP2A);
258
    const vector signed short sum2B = vec_adds(srcM1B, srcP2B);
259
    const vector signed short sum3A = vec_adds(srcM2A, srcP3A);
260
    const vector signed short sum3B = vec_adds(srcM2B, srcP3B);
261

    
262
    const vector signed short pp1A = vec_mladd(sum1A, v20ss, v16ss);
263
    const vector signed short pp1B = vec_mladd(sum1B, v20ss, v16ss);
264

    
265
    const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
266
    const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
267

    
268
    const vector signed short pp3A = vec_add(sum3A, pp1A);
269
    const vector signed short pp3B = vec_add(sum3B, pp1B);
270

    
271
    const vector signed short psumA = vec_sub(pp3A, pp2A);
272
    const vector signed short psumB = vec_sub(pp3B, pp2B);
273

    
274
    const vector signed short sumA = vec_sra(psumA, v5us);
275
    const vector signed short sumB = vec_sra(psumB, v5us);
276

    
277
    const vector unsigned char sum = vec_packsu(sumA, sumB);
278

    
279
    const vector unsigned char dst1 = vec_ld(0, dst);
280
    const vector unsigned char dst2 = vec_ld(16, dst);
281
    const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
282

    
283
    vector unsigned char fsum;
284
    OP_U8_ALTIVEC(fsum, sum, vdst);
285

    
286
    const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
287
    const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
288
    const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
289

    
290
    vec_st(fdst1, 0, dst);
291
    vec_st(fdst2, 16, dst);
292

    
293
    src += srcStride;
294
    dst += dstStride;
295
  }
296
POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);
297
}
298

    
299
/* this code assume stride % 16 == 0 */
300
static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
301
  POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_v_lowpass_num, 1);
302
  POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1);
303

    
304
  register int i;
305

    
306
  const vector signed int vzero = vec_splat_s32(0);
307
  const vector unsigned char perm = vec_lvsl(0, src);
308
  const vector signed short v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
309
  const vector unsigned short v5us = vec_splat_u16(5);
310
  const vector signed short v5ss = vec_splat_s16(5);
311
  const vector signed short v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4));
312
  const vector unsigned char dstperm = vec_lvsr(0, dst);
313
  const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
314
  const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
315

    
316
  uint8_t *srcbis = src - (srcStride * 2);
317

    
318
  const vector unsigned char srcM2a = vec_ld(0, srcbis);
319
  const vector unsigned char srcM2b = vec_ld(16, srcbis);
320
  const vector unsigned char srcM2 = vec_perm(srcM2a, srcM2b, perm);
321
  srcbis += srcStride;
322
  const vector unsigned char srcM1a = vec_ld(0, srcbis);
323
  const vector unsigned char srcM1b = vec_ld(16, srcbis);
324
  const vector unsigned char srcM1 = vec_perm(srcM1a, srcM1b, perm);
325
  srcbis += srcStride;
326
  const vector unsigned char srcP0a = vec_ld(0, srcbis);
327
  const vector unsigned char srcP0b = vec_ld(16, srcbis);
328
  const vector unsigned char srcP0 = vec_perm(srcP0a, srcP0b, perm);
329
  srcbis += srcStride;
330
  const vector unsigned char srcP1a = vec_ld(0, srcbis);
331
  const vector unsigned char srcP1b = vec_ld(16, srcbis);
332
  const vector unsigned char srcP1 = vec_perm(srcP1a, srcP1b, perm);
333
  srcbis += srcStride;
334
  const vector unsigned char srcP2a = vec_ld(0, srcbis);
335
  const vector unsigned char srcP2b = vec_ld(16, srcbis);
336
  const vector unsigned char srcP2 = vec_perm(srcP2a, srcP2b, perm);
337
  srcbis += srcStride;
338

    
339
  vector signed short srcM2ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
340
  vector signed short srcM2ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
341
  vector signed short srcM1ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
342
  vector signed short srcM1ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
343
  vector signed short srcP0ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
344
  vector signed short srcP0ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
345
  vector signed short srcP1ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
346
  vector signed short srcP1ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
347
  vector signed short srcP2ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
348
  vector signed short srcP2ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
349

    
350
  for (i = 0 ; i < 16 ; i++) {
351
    const vector unsigned char srcP3a = vec_ld(0, srcbis);
352
    const vector unsigned char srcP3b = vec_ld(16, srcbis);
353
    const vector unsigned char srcP3 = vec_perm(srcP3a, srcP3b, perm);
354
    const vector signed short srcP3ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
355
    const vector signed short srcP3ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
356
    srcbis += srcStride;
357

    
358
    const vector signed short sum1A = vec_adds(srcP0ssA, srcP1ssA);
359
    const vector signed short sum1B = vec_adds(srcP0ssB, srcP1ssB);
360
    const vector signed short sum2A = vec_adds(srcM1ssA, srcP2ssA);
361
    const vector signed short sum2B = vec_adds(srcM1ssB, srcP2ssB);
362
    const vector signed short sum3A = vec_adds(srcM2ssA, srcP3ssA);
363
    const vector signed short sum3B = vec_adds(srcM2ssB, srcP3ssB);
364

    
365
    srcM2ssA = srcM1ssA;
366
    srcM2ssB = srcM1ssB;
367
    srcM1ssA = srcP0ssA;
368
    srcM1ssB = srcP0ssB;
369
    srcP0ssA = srcP1ssA;
370
    srcP0ssB = srcP1ssB;
371
    srcP1ssA = srcP2ssA;
372
    srcP1ssB = srcP2ssB;
373
    srcP2ssA = srcP3ssA;
374
    srcP2ssB = srcP3ssB;
375

    
376
    const vector signed short pp1A = vec_mladd(sum1A, v20ss, v16ss);
377
    const vector signed short pp1B = vec_mladd(sum1B, v20ss, v16ss);
378

    
379
    const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
380
    const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
381

    
382
    const vector signed short pp3A = vec_add(sum3A, pp1A);
383
    const vector signed short pp3B = vec_add(sum3B, pp1B);
384

    
385
    const vector signed short psumA = vec_sub(pp3A, pp2A);
386
    const vector signed short psumB = vec_sub(pp3B, pp2B);
387

    
388
    const vector signed short sumA = vec_sra(psumA, v5us);
389
    const vector signed short sumB = vec_sra(psumB, v5us);
390

    
391
    const vector unsigned char sum = vec_packsu(sumA, sumB);
392

    
393
    const vector unsigned char dst1 = vec_ld(0, dst);
394
    const vector unsigned char dst2 = vec_ld(16, dst);
395
    const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
396

    
397
    vector unsigned char fsum;
398
    OP_U8_ALTIVEC(fsum, sum, vdst);
399

    
400
    const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
401
    const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
402
    const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
403

    
404
    vec_st(fdst1, 0, dst);
405
    vec_st(fdst2, 16, dst);
406

    
407
    dst += dstStride;
408
  }
409
  POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1);
410
}
411

    
412
/* this code assume stride % 16 == 0 *and* tmp is properly aligned */
413
static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp, uint8_t * src, int dstStride, int tmpStride, int srcStride) {
414
  POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_hv_lowpass_num, 1);
415
  POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1);
416
  register int i;
417
  const vector signed int vzero = vec_splat_s32(0);
418
  const vector unsigned char permM2 = vec_lvsl(-2, src);
419
  const vector unsigned char permM1 = vec_lvsl(-1, src);
420
  const vector unsigned char permP0 = vec_lvsl(+0, src);
421
  const vector unsigned char permP1 = vec_lvsl(+1, src);
422
  const vector unsigned char permP2 = vec_lvsl(+2, src);
423
  const vector unsigned char permP3 = vec_lvsl(+3, src);
424
  const vector signed short v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
425
  const vector unsigned int v10ui = vec_splat_u32(10);
426
  const vector signed short v5ss = vec_splat_s16(5);
427
  const vector signed short v1ss = vec_splat_s16(1);
428
  const vector signed int v512si = vec_sl(vec_splat_s32(1),vec_splat_u32(9));
429
  const vector unsigned int v16ui = vec_sl(vec_splat_u32(1),vec_splat_u32(4));
430

    
431
  register int align = ((((unsigned long)src) - 2) % 16);
432

    
433
  src -= (2 * srcStride);
434

    
435
  for (i = 0 ; i < 21 ; i ++) {
436
    vector unsigned char srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
437
    vector unsigned char srcR1 = vec_ld(-2, src);
438
    vector unsigned char srcR2 = vec_ld(14, src);
439

    
440
    switch (align) {
441
    default: {
442
      srcM2 = vec_perm(srcR1, srcR2, permM2);
443
      srcM1 = vec_perm(srcR1, srcR2, permM1);
444
      srcP0 = vec_perm(srcR1, srcR2, permP0);
445
      srcP1 = vec_perm(srcR1, srcR2, permP1);
446
      srcP2 = vec_perm(srcR1, srcR2, permP2);
447
      srcP3 = vec_perm(srcR1, srcR2, permP3);
448
    } break;
449
    case 11: {
450
      srcM2 = vec_perm(srcR1, srcR2, permM2);
451
      srcM1 = vec_perm(srcR1, srcR2, permM1);
452
      srcP0 = vec_perm(srcR1, srcR2, permP0);
453
      srcP1 = vec_perm(srcR1, srcR2, permP1);
454
      srcP2 = vec_perm(srcR1, srcR2, permP2);
455
      srcP3 = srcR2;
456
    } break;
457
    case 12: {
458
      vector unsigned char srcR3 = vec_ld(30, src);
459
      srcM2 = vec_perm(srcR1, srcR2, permM2);
460
      srcM1 = vec_perm(srcR1, srcR2, permM1);
461
      srcP0 = vec_perm(srcR1, srcR2, permP0);
462
      srcP1 = vec_perm(srcR1, srcR2, permP1);
463
      srcP2 = srcR2;
464
      srcP3 = vec_perm(srcR2, srcR3, permP3);
465
    } break;
466
    case 13: {
467
      vector unsigned char srcR3 = vec_ld(30, src);
468
      srcM2 = vec_perm(srcR1, srcR2, permM2);
469
      srcM1 = vec_perm(srcR1, srcR2, permM1);
470
      srcP0 = vec_perm(srcR1, srcR2, permP0);
471
      srcP1 = srcR2;
472
      srcP2 = vec_perm(srcR2, srcR3, permP2);
473
      srcP3 = vec_perm(srcR2, srcR3, permP3);
474
    } break;
475
    case 14: {
476
      vector unsigned char srcR3 = vec_ld(30, src);
477
      srcM2 = vec_perm(srcR1, srcR2, permM2);
478
      srcM1 = vec_perm(srcR1, srcR2, permM1);
479
      srcP0 = srcR2;
480
      srcP1 = vec_perm(srcR2, srcR3, permP1);
481
      srcP2 = vec_perm(srcR2, srcR3, permP2);
482
      srcP3 = vec_perm(srcR2, srcR3, permP3);
483
    } break;
484
    case 15: {
485
      vector unsigned char srcR3 = vec_ld(30, src);
486
      srcM2 = vec_perm(srcR1, srcR2, permM2);
487
      srcM1 = srcR2;
488
      srcP0 = vec_perm(srcR2, srcR3, permP0);
489
      srcP1 = vec_perm(srcR2, srcR3, permP1);
490
      srcP2 = vec_perm(srcR2, srcR3, permP2);
491
      srcP3 = vec_perm(srcR2, srcR3, permP3);
492
    } break;
493
    }
494

    
495
    const vector signed short srcP0A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
496
    const vector signed short srcP0B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
497
    const vector signed short srcP1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
498
    const vector signed short srcP1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
499

    
500
    const vector signed short srcP2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
501
    const vector signed short srcP2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
502
    const vector signed short srcP3A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
503
    const vector signed short srcP3B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
504

    
505
    const vector signed short srcM1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
506
    const vector signed short srcM1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
507
    const vector signed short srcM2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
508
    const vector signed short srcM2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
509

    
510
    const vector signed short sum1A = vec_adds(srcP0A, srcP1A);
511
    const vector signed short sum1B = vec_adds(srcP0B, srcP1B);
512
    const vector signed short sum2A = vec_adds(srcM1A, srcP2A);
513
    const vector signed short sum2B = vec_adds(srcM1B, srcP2B);
514
    const vector signed short sum3A = vec_adds(srcM2A, srcP3A);
515
    const vector signed short sum3B = vec_adds(srcM2B, srcP3B);
516

    
517
    const vector signed short pp1A = vec_mladd(sum1A, v20ss, sum3A);
518
    const vector signed short pp1B = vec_mladd(sum1B, v20ss, sum3B);
519

    
520
    const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
521
    const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
522

    
523
    const vector signed short psumA = vec_sub(pp1A, pp2A);
524
    const vector signed short psumB = vec_sub(pp1B, pp2B);
525

    
526
    vec_st(psumA, 0, tmp);
527
    vec_st(psumB, 16, tmp);
528

    
529
    src += srcStride;
530
    tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */
531
  }
532

    
533
  const vector unsigned char dstperm = vec_lvsr(0, dst);
534
  const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
535
  const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
536
  const vector unsigned char mperm = (const vector unsigned char)
537
    AVV(0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B,
538
        0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F);
539

    
540
  int16_t *tmpbis = tmp - (tmpStride * 21);
541

    
542
  vector signed short tmpM2ssA = vec_ld(0, tmpbis);
543
  vector signed short tmpM2ssB = vec_ld(16, tmpbis);
544
  tmpbis += tmpStride;
545
  vector signed short tmpM1ssA = vec_ld(0, tmpbis);
546
  vector signed short tmpM1ssB = vec_ld(16, tmpbis);
547
  tmpbis += tmpStride;
548
  vector signed short tmpP0ssA = vec_ld(0, tmpbis);
549
  vector signed short tmpP0ssB = vec_ld(16, tmpbis);
550
  tmpbis += tmpStride;
551
  vector signed short tmpP1ssA = vec_ld(0, tmpbis);
552
  vector signed short tmpP1ssB = vec_ld(16, tmpbis);
553
  tmpbis += tmpStride;
554
  vector signed short tmpP2ssA = vec_ld(0, tmpbis);
555
  vector signed short tmpP2ssB = vec_ld(16, tmpbis);
556
  tmpbis += tmpStride;
557

    
558
  for (i = 0 ; i < 16 ; i++) {
559
    const vector signed short tmpP3ssA = vec_ld(0, tmpbis);
560
    const vector signed short tmpP3ssB = vec_ld(16, tmpbis);
561
    tmpbis += tmpStride;
562

    
563
    const vector signed short sum1A = vec_adds(tmpP0ssA, tmpP1ssA);
564
    const vector signed short sum1B = vec_adds(tmpP0ssB, tmpP1ssB);
565
    const vector signed short sum2A = vec_adds(tmpM1ssA, tmpP2ssA);
566
    const vector signed short sum2B = vec_adds(tmpM1ssB, tmpP2ssB);
567
    const vector signed short sum3A = vec_adds(tmpM2ssA, tmpP3ssA);
568
    const vector signed short sum3B = vec_adds(tmpM2ssB, tmpP3ssB);
569

    
570
    tmpM2ssA = tmpM1ssA;
571
    tmpM2ssB = tmpM1ssB;
572
    tmpM1ssA = tmpP0ssA;
573
    tmpM1ssB = tmpP0ssB;
574
    tmpP0ssA = tmpP1ssA;
575
    tmpP0ssB = tmpP1ssB;
576
    tmpP1ssA = tmpP2ssA;
577
    tmpP1ssB = tmpP2ssB;
578
    tmpP2ssA = tmpP3ssA;
579
    tmpP2ssB = tmpP3ssB;
580

    
581
    const vector signed int pp1Ae = vec_mule(sum1A, v20ss);
582
    const vector signed int pp1Ao = vec_mulo(sum1A, v20ss);
583
    const vector signed int pp1Be = vec_mule(sum1B, v20ss);
584
    const vector signed int pp1Bo = vec_mulo(sum1B, v20ss);
585

    
586
    const vector signed int pp2Ae = vec_mule(sum2A, v5ss);
587
    const vector signed int pp2Ao = vec_mulo(sum2A, v5ss);
588
    const vector signed int pp2Be = vec_mule(sum2B, v5ss);
589
    const vector signed int pp2Bo = vec_mulo(sum2B, v5ss);
590

    
591
    const vector signed int pp3Ae = vec_sra((vector signed int)sum3A, v16ui);
592
    const vector signed int pp3Ao = vec_mulo(sum3A, v1ss);
593
    const vector signed int pp3Be = vec_sra((vector signed int)sum3B, v16ui);
594
    const vector signed int pp3Bo = vec_mulo(sum3B, v1ss);
595

    
596
    const vector signed int pp1cAe = vec_add(pp1Ae, v512si);
597
    const vector signed int pp1cAo = vec_add(pp1Ao, v512si);
598
    const vector signed int pp1cBe = vec_add(pp1Be, v512si);
599
    const vector signed int pp1cBo = vec_add(pp1Bo, v512si);
600

    
601
    const vector signed int pp32Ae = vec_sub(pp3Ae, pp2Ae);
602
    const vector signed int pp32Ao = vec_sub(pp3Ao, pp2Ao);
603
    const vector signed int pp32Be = vec_sub(pp3Be, pp2Be);
604
    const vector signed int pp32Bo = vec_sub(pp3Bo, pp2Bo);
605

    
606
    const vector signed int sumAe = vec_add(pp1cAe, pp32Ae);
607
    const vector signed int sumAo = vec_add(pp1cAo, pp32Ao);
608
    const vector signed int sumBe = vec_add(pp1cBe, pp32Be);
609
    const vector signed int sumBo = vec_add(pp1cBo, pp32Bo);
610

    
611
    const vector signed int ssumAe = vec_sra(sumAe, v10ui);
612
    const vector signed int ssumAo = vec_sra(sumAo, v10ui);
613
    const vector signed int ssumBe = vec_sra(sumBe, v10ui);
614
    const vector signed int ssumBo = vec_sra(sumBo, v10ui);
615

    
616
    const vector signed short ssume = vec_packs(ssumAe, ssumBe);
617
    const vector signed short ssumo = vec_packs(ssumAo, ssumBo);
618

    
619
    const vector unsigned char sumv = vec_packsu(ssume, ssumo);
620
    const vector unsigned char sum = vec_perm(sumv, sumv, mperm);
621

    
622
    const vector unsigned char dst1 = vec_ld(0, dst);
623
    const vector unsigned char dst2 = vec_ld(16, dst);
624
    const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
625

    
626
    vector unsigned char fsum;
627
    OP_U8_ALTIVEC(fsum, sum, vdst);
628

    
629
    const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
630
    const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
631
    const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
632

    
633
    vec_st(fdst1, 0, dst);
634
    vec_st(fdst2, 16, dst);
635

    
636
    dst += dstStride;
637
  }
638
  POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1);
639
}