Main Page
Related Pages
Modules
Data Structures
Files
Examples
File List
Globals
libavcodec
x86
mpegvideo_mmx.c
Go to the documentation of this file.
1
/*
2
* The simplest mpeg encoder (well, it was the simplest!)
3
* Copyright (c) 2000,2001 Fabrice Bellard
4
*
5
* Optimized for ia32 CPUs by Nick Kurshev <nickols_k@mail.ru>
6
* h263, mpeg1, mpeg2 dequantizer & draw_edges by Michael Niedermayer <michaelni@gmx.at>
7
*
8
* This file is part of Libav.
9
*
10
* Libav is free software; you can redistribute it and/or
11
* modify it under the terms of the GNU Lesser General Public
12
* License as published by the Free Software Foundation; either
13
* version 2.1 of the License, or (at your option) any later version.
14
*
15
* Libav is distributed in the hope that it will be useful,
16
* but WITHOUT ANY WARRANTY; without even the implied warranty of
17
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18
* Lesser General Public License for more details.
19
*
20
* You should have received a copy of the GNU Lesser General Public
21
* License along with Libav; if not, write to the Free Software
22
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23
*/
24
25
#include "
libavutil/cpu.h
"
26
#include "
libavutil/x86_cpu.h
"
27
#include "
libavcodec/avcodec.h
"
28
#include "
libavcodec/dsputil.h
"
29
#include "
libavcodec/mpegvideo.h
"
30
#include "
dsputil_mmx.h
"
31
32
extern
uint16_t
inv_zigzag_direct16
[64];
33
34
35
static
void
dct_unquantize_h263_intra_mmx
(
MpegEncContext
*s,
36
DCTELEM
*
block
,
int
n,
int
qscale)
37
{
38
x86_reg
level
, qmul, qadd, nCoeffs;
39
40
qmul = qscale << 1;
41
42
assert(s->
block_last_index
[n]>=0 || s->
h263_aic
);
43
44
if
(!s->
h263_aic
) {
45
if
(n < 4)
46
level = block[0] * s->
y_dc_scale
;
47
else
48
level = block[0] * s->
c_dc_scale
;
49
qadd = (qscale - 1) | 1;
50
}
else
{
51
qadd = 0;
52
level= block[0];
53
}
54
if
(s->
ac_pred
)
55
nCoeffs=63;
56
else
57
nCoeffs= s->
inter_scantable
.
raster_end
[ s->
block_last_index
[n] ];
58
//printf("%d %d ", qmul, qadd);
59
__asm__
volatile
(
60
"movd %1, %%mm6 \n\t"
//qmul
61
"packssdw %%mm6, %%mm6 \n\t"
62
"packssdw %%mm6, %%mm6 \n\t"
63
"movd %2, %%mm5 \n\t"
//qadd
64
"pxor %%mm7, %%mm7 \n\t"
65
"packssdw %%mm5, %%mm5 \n\t"
66
"packssdw %%mm5, %%mm5 \n\t"
67
"psubw %%mm5, %%mm7 \n\t"
68
"pxor %%mm4, %%mm4 \n\t"
69
".p2align 4 \n\t"
70
"1: \n\t"
71
"movq (%0, %3), %%mm0 \n\t"
72
"movq 8(%0, %3), %%mm1 \n\t"
73
74
"pmullw %%mm6, %%mm0 \n\t"
75
"pmullw %%mm6, %%mm1 \n\t"
76
77
"movq (%0, %3), %%mm2 \n\t"
78
"movq 8(%0, %3), %%mm3 \n\t"
79
80
"pcmpgtw %%mm4, %%mm2 \n\t"
// block[i] < 0 ? -1 : 0
81
"pcmpgtw %%mm4, %%mm3 \n\t"
// block[i] < 0 ? -1 : 0
82
83
"pxor %%mm2, %%mm0 \n\t"
84
"pxor %%mm3, %%mm1 \n\t"
85
86
"paddw %%mm7, %%mm0 \n\t"
87
"paddw %%mm7, %%mm1 \n\t"
88
89
"pxor %%mm0, %%mm2 \n\t"
90
"pxor %%mm1, %%mm3 \n\t"
91
92
"pcmpeqw %%mm7, %%mm0 \n\t"
// block[i] == 0 ? -1 : 0
93
"pcmpeqw %%mm7, %%mm1 \n\t"
// block[i] == 0 ? -1 : 0
94
95
"pandn %%mm2, %%mm0 \n\t"
96
"pandn %%mm3, %%mm1 \n\t"
97
98
"movq %%mm0, (%0, %3) \n\t"
99
"movq %%mm1, 8(%0, %3) \n\t"
100
101
"add $16, %3 \n\t"
102
"jng 1b \n\t"
103
::
"r"
(block+nCoeffs),
"rm"
(qmul),
"rm"
(qadd),
"r"
(2*(-nCoeffs))
104
:
"memory"
105
);
106
block[0]=
level
;
107
}
108
109
110
static
void
dct_unquantize_h263_inter_mmx
(
MpegEncContext
*s,
111
DCTELEM
*
block
,
int
n,
int
qscale)
112
{
113
x86_reg
qmul, qadd, nCoeffs;
114
115
qmul = qscale << 1;
116
qadd = (qscale - 1) | 1;
117
118
assert(s->
block_last_index
[n]>=0 || s->
h263_aic
);
119
120
nCoeffs= s->
inter_scantable
.
raster_end
[ s->
block_last_index
[n] ];
121
//printf("%d %d ", qmul, qadd);
122
__asm__
volatile
(
123
"movd %1, %%mm6 \n\t"
//qmul
124
"packssdw %%mm6, %%mm6 \n\t"
125
"packssdw %%mm6, %%mm6 \n\t"
126
"movd %2, %%mm5 \n\t"
//qadd
127
"pxor %%mm7, %%mm7 \n\t"
128
"packssdw %%mm5, %%mm5 \n\t"
129
"packssdw %%mm5, %%mm5 \n\t"
130
"psubw %%mm5, %%mm7 \n\t"
131
"pxor %%mm4, %%mm4 \n\t"
132
".p2align 4 \n\t"
133
"1: \n\t"
134
"movq (%0, %3), %%mm0 \n\t"
135
"movq 8(%0, %3), %%mm1 \n\t"
136
137
"pmullw %%mm6, %%mm0 \n\t"
138
"pmullw %%mm6, %%mm1 \n\t"
139
140
"movq (%0, %3), %%mm2 \n\t"
141
"movq 8(%0, %3), %%mm3 \n\t"
142
143
"pcmpgtw %%mm4, %%mm2 \n\t"
// block[i] < 0 ? -1 : 0
144
"pcmpgtw %%mm4, %%mm3 \n\t"
// block[i] < 0 ? -1 : 0
145
146
"pxor %%mm2, %%mm0 \n\t"
147
"pxor %%mm3, %%mm1 \n\t"
148
149
"paddw %%mm7, %%mm0 \n\t"
150
"paddw %%mm7, %%mm1 \n\t"
151
152
"pxor %%mm0, %%mm2 \n\t"
153
"pxor %%mm1, %%mm3 \n\t"
154
155
"pcmpeqw %%mm7, %%mm0 \n\t"
// block[i] == 0 ? -1 : 0
156
"pcmpeqw %%mm7, %%mm1 \n\t"
// block[i] == 0 ? -1 : 0
157
158
"pandn %%mm2, %%mm0 \n\t"
159
"pandn %%mm3, %%mm1 \n\t"
160
161
"movq %%mm0, (%0, %3) \n\t"
162
"movq %%mm1, 8(%0, %3) \n\t"
163
164
"add $16, %3 \n\t"
165
"jng 1b \n\t"
166
::
"r"
(block+nCoeffs),
"rm"
(qmul),
"rm"
(qadd),
"r"
(2*(-nCoeffs))
167
:
"memory"
168
);
169
}
170
171
172
/*
173
NK:
174
Note: looking at PARANOID:
175
"enable all paranoid tests for rounding, overflows, etc..."
176
177
#ifdef PARANOID
178
if (level < -2048 || level > 2047)
179
fprintf(stderr, "unquant error %d %d\n", i, level);
180
#endif
181
We can suppose that result of two multiplications can't be greater than 0xFFFF
182
i.e. is 16-bit, so we use here only PMULLW instruction and can avoid
183
a complex multiplication.
184
=====================================================
185
Full formula for multiplication of 2 integer numbers
186
which are represent as high:low words:
187
input: value1 = high1:low1
188
value2 = high2:low2
189
output: value3 = value1*value2
190
value3=high3:low3 (on overflow: modulus 2^32 wrap-around)
191
this mean that for 0x123456 * 0x123456 correct result is 0x766cb0ce4
192
but this algorithm will compute only 0x66cb0ce4
193
this limited by 16-bit size of operands
194
---------------------------------
195
tlow1 = high1*low2
196
tlow2 = high2*low1
197
tlow1 = tlow1 + tlow2
198
high3:low3 = low1*low2
199
high3 += tlow1
200
*/
201
static
void
dct_unquantize_mpeg1_intra_mmx
(
MpegEncContext
*s,
202
DCTELEM
*
block
,
int
n,
int
qscale)
203
{
204
x86_reg
nCoeffs;
205
const
uint16_t *quant_matrix;
206
int
block0;
207
208
assert(s->
block_last_index
[n]>=0);
209
210
nCoeffs= s->
intra_scantable
.
raster_end
[ s->
block_last_index
[n] ]+1;
211
212
if
(n < 4)
213
block0 = block[0] * s->
y_dc_scale
;
214
else
215
block0 = block[0] * s->
c_dc_scale
;
216
/* XXX: only mpeg1 */
217
quant_matrix = s->
intra_matrix
;
218
__asm__
volatile
(
219
"pcmpeqw %%mm7, %%mm7 \n\t"
220
"psrlw $15, %%mm7 \n\t"
221
"movd %2, %%mm6 \n\t"
222
"packssdw %%mm6, %%mm6 \n\t"
223
"packssdw %%mm6, %%mm6 \n\t"
224
"mov %3, %%"
REG_a
" \n\t"
225
".p2align 4 \n\t"
226
"1: \n\t"
227
"movq (%0, %%"
REG_a
"), %%mm0 \n\t"
228
"movq 8(%0, %%"
REG_a
"), %%mm1 \n\t"
229
"movq (%1, %%"
REG_a
"), %%mm4 \n\t"
230
"movq 8(%1, %%"
REG_a
"), %%mm5 \n\t"
231
"pmullw %%mm6, %%mm4 \n\t"
// q=qscale*quant_matrix[i]
232
"pmullw %%mm6, %%mm5 \n\t"
// q=qscale*quant_matrix[i]
233
"pxor %%mm2, %%mm2 \n\t"
234
"pxor %%mm3, %%mm3 \n\t"
235
"pcmpgtw %%mm0, %%mm2 \n\t"
// block[i] < 0 ? -1 : 0
236
"pcmpgtw %%mm1, %%mm3 \n\t"
// block[i] < 0 ? -1 : 0
237
"pxor %%mm2, %%mm0 \n\t"
238
"pxor %%mm3, %%mm1 \n\t"
239
"psubw %%mm2, %%mm0 \n\t"
// abs(block[i])
240
"psubw %%mm3, %%mm1 \n\t"
// abs(block[i])
241
"pmullw %%mm4, %%mm0 \n\t"
// abs(block[i])*q
242
"pmullw %%mm5, %%mm1 \n\t"
// abs(block[i])*q
243
"pxor %%mm4, %%mm4 \n\t"
244
"pxor %%mm5, %%mm5 \n\t"
// FIXME slow
245
"pcmpeqw (%0, %%"
REG_a
"), %%mm4 \n\t"
// block[i] == 0 ? -1 : 0
246
"pcmpeqw 8(%0, %%"
REG_a
"), %%mm5\n\t"
// block[i] == 0 ? -1 : 0
247
"psraw $3, %%mm0 \n\t"
248
"psraw $3, %%mm1 \n\t"
249
"psubw %%mm7, %%mm0 \n\t"
250
"psubw %%mm7, %%mm1 \n\t"
251
"por %%mm7, %%mm0 \n\t"
252
"por %%mm7, %%mm1 \n\t"
253
"pxor %%mm2, %%mm0 \n\t"
254
"pxor %%mm3, %%mm1 \n\t"
255
"psubw %%mm2, %%mm0 \n\t"
256
"psubw %%mm3, %%mm1 \n\t"
257
"pandn %%mm0, %%mm4 \n\t"
258
"pandn %%mm1, %%mm5 \n\t"
259
"movq %%mm4, (%0, %%"
REG_a
") \n\t"
260
"movq %%mm5, 8(%0, %%"
REG_a
") \n\t"
261
262
"add $16, %%"
REG_a
" \n\t"
263
"js 1b \n\t"
264
::
"r"
(block+nCoeffs),
"r"
(quant_matrix+nCoeffs),
"rm"
(qscale),
"g"
(-2*nCoeffs)
265
:
"%"
REG_a,
"memory"
266
);
267
block[0]= block0;
268
}
269
270
static
void
dct_unquantize_mpeg1_inter_mmx
(
MpegEncContext
*s,
271
DCTELEM
*
block
,
int
n,
int
qscale)
272
{
273
x86_reg
nCoeffs;
274
const
uint16_t *quant_matrix;
275
276
assert(s->
block_last_index
[n]>=0);
277
278
nCoeffs= s->
intra_scantable
.
raster_end
[ s->
block_last_index
[n] ]+1;
279
280
quant_matrix = s->
inter_matrix
;
281
__asm__
volatile
(
282
"pcmpeqw %%mm7, %%mm7 \n\t"
283
"psrlw $15, %%mm7 \n\t"
284
"movd %2, %%mm6 \n\t"
285
"packssdw %%mm6, %%mm6 \n\t"
286
"packssdw %%mm6, %%mm6 \n\t"
287
"mov %3, %%"
REG_a
" \n\t"
288
".p2align 4 \n\t"
289
"1: \n\t"
290
"movq (%0, %%"
REG_a
"), %%mm0 \n\t"
291
"movq 8(%0, %%"
REG_a
"), %%mm1 \n\t"
292
"movq (%1, %%"
REG_a
"), %%mm4 \n\t"
293
"movq 8(%1, %%"
REG_a
"), %%mm5 \n\t"
294
"pmullw %%mm6, %%mm4 \n\t"
// q=qscale*quant_matrix[i]
295
"pmullw %%mm6, %%mm5 \n\t"
// q=qscale*quant_matrix[i]
296
"pxor %%mm2, %%mm2 \n\t"
297
"pxor %%mm3, %%mm3 \n\t"
298
"pcmpgtw %%mm0, %%mm2 \n\t"
// block[i] < 0 ? -1 : 0
299
"pcmpgtw %%mm1, %%mm3 \n\t"
// block[i] < 0 ? -1 : 0
300
"pxor %%mm2, %%mm0 \n\t"
301
"pxor %%mm3, %%mm1 \n\t"
302
"psubw %%mm2, %%mm0 \n\t"
// abs(block[i])
303
"psubw %%mm3, %%mm1 \n\t"
// abs(block[i])
304
"paddw %%mm0, %%mm0 \n\t"
// abs(block[i])*2
305
"paddw %%mm1, %%mm1 \n\t"
// abs(block[i])*2
306
"paddw %%mm7, %%mm0 \n\t"
// abs(block[i])*2 + 1
307
"paddw %%mm7, %%mm1 \n\t"
// abs(block[i])*2 + 1
308
"pmullw %%mm4, %%mm0 \n\t"
// (abs(block[i])*2 + 1)*q
309
"pmullw %%mm5, %%mm1 \n\t"
// (abs(block[i])*2 + 1)*q
310
"pxor %%mm4, %%mm4 \n\t"
311
"pxor %%mm5, %%mm5 \n\t"
// FIXME slow
312
"pcmpeqw (%0, %%"
REG_a
"), %%mm4 \n\t"
// block[i] == 0 ? -1 : 0
313
"pcmpeqw 8(%0, %%"
REG_a
"), %%mm5\n\t"
// block[i] == 0 ? -1 : 0
314
"psraw $4, %%mm0 \n\t"
315
"psraw $4, %%mm1 \n\t"
316
"psubw %%mm7, %%mm0 \n\t"
317
"psubw %%mm7, %%mm1 \n\t"
318
"por %%mm7, %%mm0 \n\t"
319
"por %%mm7, %%mm1 \n\t"
320
"pxor %%mm2, %%mm0 \n\t"
321
"pxor %%mm3, %%mm1 \n\t"
322
"psubw %%mm2, %%mm0 \n\t"
323
"psubw %%mm3, %%mm1 \n\t"
324
"pandn %%mm0, %%mm4 \n\t"
325
"pandn %%mm1, %%mm5 \n\t"
326
"movq %%mm4, (%0, %%"
REG_a
") \n\t"
327
"movq %%mm5, 8(%0, %%"
REG_a
") \n\t"
328
329
"add $16, %%"
REG_a
" \n\t"
330
"js 1b \n\t"
331
::
"r"
(block+nCoeffs),
"r"
(quant_matrix+nCoeffs),
"rm"
(qscale),
"g"
(-2*nCoeffs)
332
:
"%"
REG_a,
"memory"
333
);
334
}
335
336
static
void
dct_unquantize_mpeg2_intra_mmx
(
MpegEncContext
*s,
337
DCTELEM
*
block
,
int
n,
int
qscale)
338
{
339
x86_reg
nCoeffs;
340
const
uint16_t *quant_matrix;
341
int
block0;
342
343
assert(s->
block_last_index
[n]>=0);
344
345
if
(s->
alternate_scan
) nCoeffs= 63;
//FIXME
346
else
nCoeffs= s->
intra_scantable
.
raster_end
[ s->
block_last_index
[n] ];
347
348
if
(n < 4)
349
block0 = block[0] * s->
y_dc_scale
;
350
else
351
block0 = block[0] * s->
c_dc_scale
;
352
quant_matrix = s->
intra_matrix
;
353
__asm__
volatile
(
354
"pcmpeqw %%mm7, %%mm7 \n\t"
355
"psrlw $15, %%mm7 \n\t"
356
"movd %2, %%mm6 \n\t"
357
"packssdw %%mm6, %%mm6 \n\t"
358
"packssdw %%mm6, %%mm6 \n\t"
359
"mov %3, %%"
REG_a
" \n\t"
360
".p2align 4 \n\t"
361
"1: \n\t"
362
"movq (%0, %%"
REG_a
"), %%mm0 \n\t"
363
"movq 8(%0, %%"
REG_a
"), %%mm1 \n\t"
364
"movq (%1, %%"
REG_a
"), %%mm4 \n\t"
365
"movq 8(%1, %%"
REG_a
"), %%mm5 \n\t"
366
"pmullw %%mm6, %%mm4 \n\t"
// q=qscale*quant_matrix[i]
367
"pmullw %%mm6, %%mm5 \n\t"
// q=qscale*quant_matrix[i]
368
"pxor %%mm2, %%mm2 \n\t"
369
"pxor %%mm3, %%mm3 \n\t"
370
"pcmpgtw %%mm0, %%mm2 \n\t"
// block[i] < 0 ? -1 : 0
371
"pcmpgtw %%mm1, %%mm3 \n\t"
// block[i] < 0 ? -1 : 0
372
"pxor %%mm2, %%mm0 \n\t"
373
"pxor %%mm3, %%mm1 \n\t"
374
"psubw %%mm2, %%mm0 \n\t"
// abs(block[i])
375
"psubw %%mm3, %%mm1 \n\t"
// abs(block[i])
376
"pmullw %%mm4, %%mm0 \n\t"
// abs(block[i])*q
377
"pmullw %%mm5, %%mm1 \n\t"
// abs(block[i])*q
378
"pxor %%mm4, %%mm4 \n\t"
379
"pxor %%mm5, %%mm5 \n\t"
// FIXME slow
380
"pcmpeqw (%0, %%"
REG_a
"), %%mm4 \n\t"
// block[i] == 0 ? -1 : 0
381
"pcmpeqw 8(%0, %%"
REG_a
"), %%mm5\n\t"
// block[i] == 0 ? -1 : 0
382
"psraw $3, %%mm0 \n\t"
383
"psraw $3, %%mm1 \n\t"
384
"pxor %%mm2, %%mm0 \n\t"
385
"pxor %%mm3, %%mm1 \n\t"
386
"psubw %%mm2, %%mm0 \n\t"
387
"psubw %%mm3, %%mm1 \n\t"
388
"pandn %%mm0, %%mm4 \n\t"
389
"pandn %%mm1, %%mm5 \n\t"
390
"movq %%mm4, (%0, %%"
REG_a
") \n\t"
391
"movq %%mm5, 8(%0, %%"
REG_a
") \n\t"
392
393
"add $16, %%"
REG_a
" \n\t"
394
"jng 1b \n\t"
395
::
"r"
(block+nCoeffs),
"r"
(quant_matrix+nCoeffs),
"rm"
(qscale),
"g"
(-2*nCoeffs)
396
:
"%"
REG_a,
"memory"
397
);
398
block[0]= block0;
399
//Note, we do not do mismatch control for intra as errors cannot accumulate
400
}
401
402
static
void
dct_unquantize_mpeg2_inter_mmx
(
MpegEncContext
*s,
403
DCTELEM
*
block
,
int
n,
int
qscale)
404
{
405
x86_reg
nCoeffs;
406
const
uint16_t *quant_matrix;
407
408
assert(s->
block_last_index
[n]>=0);
409
410
if
(s->
alternate_scan
) nCoeffs= 63;
//FIXME
411
else
nCoeffs= s->
intra_scantable
.
raster_end
[ s->
block_last_index
[n] ];
412
413
quant_matrix = s->
inter_matrix
;
414
__asm__
volatile
(
415
"pcmpeqw %%mm7, %%mm7 \n\t"
416
"psrlq $48, %%mm7 \n\t"
417
"movd %2, %%mm6 \n\t"
418
"packssdw %%mm6, %%mm6 \n\t"
419
"packssdw %%mm6, %%mm6 \n\t"
420
"mov %3, %%"
REG_a
" \n\t"
421
".p2align 4 \n\t"
422
"1: \n\t"
423
"movq (%0, %%"
REG_a
"), %%mm0 \n\t"
424
"movq 8(%0, %%"
REG_a
"), %%mm1 \n\t"
425
"movq (%1, %%"
REG_a
"), %%mm4 \n\t"
426
"movq 8(%1, %%"
REG_a
"), %%mm5 \n\t"
427
"pmullw %%mm6, %%mm4 \n\t"
// q=qscale*quant_matrix[i]
428
"pmullw %%mm6, %%mm5 \n\t"
// q=qscale*quant_matrix[i]
429
"pxor %%mm2, %%mm2 \n\t"
430
"pxor %%mm3, %%mm3 \n\t"
431
"pcmpgtw %%mm0, %%mm2 \n\t"
// block[i] < 0 ? -1 : 0
432
"pcmpgtw %%mm1, %%mm3 \n\t"
// block[i] < 0 ? -1 : 0
433
"pxor %%mm2, %%mm0 \n\t"
434
"pxor %%mm3, %%mm1 \n\t"
435
"psubw %%mm2, %%mm0 \n\t"
// abs(block[i])
436
"psubw %%mm3, %%mm1 \n\t"
// abs(block[i])
437
"paddw %%mm0, %%mm0 \n\t"
// abs(block[i])*2
438
"paddw %%mm1, %%mm1 \n\t"
// abs(block[i])*2
439
"pmullw %%mm4, %%mm0 \n\t"
// abs(block[i])*2*q
440
"pmullw %%mm5, %%mm1 \n\t"
// abs(block[i])*2*q
441
"paddw %%mm4, %%mm0 \n\t"
// (abs(block[i])*2 + 1)*q
442
"paddw %%mm5, %%mm1 \n\t"
// (abs(block[i])*2 + 1)*q
443
"pxor %%mm4, %%mm4 \n\t"
444
"pxor %%mm5, %%mm5 \n\t"
// FIXME slow
445
"pcmpeqw (%0, %%"
REG_a
"), %%mm4 \n\t"
// block[i] == 0 ? -1 : 0
446
"pcmpeqw 8(%0, %%"
REG_a
"), %%mm5\n\t"
// block[i] == 0 ? -1 : 0
447
"psrlw $4, %%mm0 \n\t"
448
"psrlw $4, %%mm1 \n\t"
449
"pxor %%mm2, %%mm0 \n\t"
450
"pxor %%mm3, %%mm1 \n\t"
451
"psubw %%mm2, %%mm0 \n\t"
452
"psubw %%mm3, %%mm1 \n\t"
453
"pandn %%mm0, %%mm4 \n\t"
454
"pandn %%mm1, %%mm5 \n\t"
455
"pxor %%mm4, %%mm7 \n\t"
456
"pxor %%mm5, %%mm7 \n\t"
457
"movq %%mm4, (%0, %%"
REG_a
") \n\t"
458
"movq %%mm5, 8(%0, %%"
REG_a
") \n\t"
459
460
"add $16, %%"
REG_a
" \n\t"
461
"jng 1b \n\t"
462
"movd 124(%0, %3), %%mm0 \n\t"
463
"movq %%mm7, %%mm6 \n\t"
464
"psrlq $32, %%mm7 \n\t"
465
"pxor %%mm6, %%mm7 \n\t"
466
"movq %%mm7, %%mm6 \n\t"
467
"psrlq $16, %%mm7 \n\t"
468
"pxor %%mm6, %%mm7 \n\t"
469
"pslld $31, %%mm7 \n\t"
470
"psrlq $15, %%mm7 \n\t"
471
"pxor %%mm7, %%mm0 \n\t"
472
"movd %%mm0, 124(%0, %3) \n\t"
473
474
::
"r"
(block+nCoeffs),
"r"
(quant_matrix+nCoeffs),
"rm"
(qscale),
"r"
(-2*nCoeffs)
475
:
"%"
REG_a,
"memory"
476
);
477
}
478
479
static
void
denoise_dct_mmx
(
MpegEncContext
*s,
DCTELEM
*
block
){
480
const
int
intra= s->
mb_intra
;
481
int
*sum= s->
dct_error_sum
[intra];
482
uint16_t *offset= s->
dct_offset
[intra];
483
484
s->
dct_count
[intra]++;
485
486
__asm__
volatile
(
487
"pxor %%mm7, %%mm7 \n\t"
488
"1: \n\t"
489
"pxor %%mm0, %%mm0 \n\t"
490
"pxor %%mm1, %%mm1 \n\t"
491
"movq (%0), %%mm2 \n\t"
492
"movq 8(%0), %%mm3 \n\t"
493
"pcmpgtw %%mm2, %%mm0 \n\t"
494
"pcmpgtw %%mm3, %%mm1 \n\t"
495
"pxor %%mm0, %%mm2 \n\t"
496
"pxor %%mm1, %%mm3 \n\t"
497
"psubw %%mm0, %%mm2 \n\t"
498
"psubw %%mm1, %%mm3 \n\t"
499
"movq %%mm2, %%mm4 \n\t"
500
"movq %%mm3, %%mm5 \n\t"
501
"psubusw (%2), %%mm2 \n\t"
502
"psubusw 8(%2), %%mm3 \n\t"
503
"pxor %%mm0, %%mm2 \n\t"
504
"pxor %%mm1, %%mm3 \n\t"
505
"psubw %%mm0, %%mm2 \n\t"
506
"psubw %%mm1, %%mm3 \n\t"
507
"movq %%mm2, (%0) \n\t"
508
"movq %%mm3, 8(%0) \n\t"
509
"movq %%mm4, %%mm2 \n\t"
510
"movq %%mm5, %%mm3 \n\t"
511
"punpcklwd %%mm7, %%mm4 \n\t"
512
"punpckhwd %%mm7, %%mm2 \n\t"
513
"punpcklwd %%mm7, %%mm5 \n\t"
514
"punpckhwd %%mm7, %%mm3 \n\t"
515
"paddd (%1), %%mm4 \n\t"
516
"paddd 8(%1), %%mm2 \n\t"
517
"paddd 16(%1), %%mm5 \n\t"
518
"paddd 24(%1), %%mm3 \n\t"
519
"movq %%mm4, (%1) \n\t"
520
"movq %%mm2, 8(%1) \n\t"
521
"movq %%mm5, 16(%1) \n\t"
522
"movq %%mm3, 24(%1) \n\t"
523
"add $16, %0 \n\t"
524
"add $32, %1 \n\t"
525
"add $16, %2 \n\t"
526
"cmp %3, %0 \n\t"
527
" jb 1b \n\t"
528
:
"+r"
(
block
),
"+r"
(sum),
"+r"
(offset)
529
:
"r"
(block+64)
530
);
531
}
532
533
static
void
denoise_dct_sse2
(
MpegEncContext
*s,
DCTELEM
*
block
){
534
const
int
intra= s->
mb_intra
;
535
int
*sum= s->
dct_error_sum
[intra];
536
uint16_t *offset= s->
dct_offset
[intra];
537
538
s->
dct_count
[intra]++;
539
540
__asm__
volatile
(
541
"pxor %%xmm7, %%xmm7 \n\t"
542
"1: \n\t"
543
"pxor %%xmm0, %%xmm0 \n\t"
544
"pxor %%xmm1, %%xmm1 \n\t"
545
"movdqa (%0), %%xmm2 \n\t"
546
"movdqa 16(%0), %%xmm3 \n\t"
547
"pcmpgtw %%xmm2, %%xmm0 \n\t"
548
"pcmpgtw %%xmm3, %%xmm1 \n\t"
549
"pxor %%xmm0, %%xmm2 \n\t"
550
"pxor %%xmm1, %%xmm3 \n\t"
551
"psubw %%xmm0, %%xmm2 \n\t"
552
"psubw %%xmm1, %%xmm3 \n\t"
553
"movdqa %%xmm2, %%xmm4 \n\t"
554
"movdqa %%xmm3, %%xmm5 \n\t"
555
"psubusw (%2), %%xmm2 \n\t"
556
"psubusw 16(%2), %%xmm3 \n\t"
557
"pxor %%xmm0, %%xmm2 \n\t"
558
"pxor %%xmm1, %%xmm3 \n\t"
559
"psubw %%xmm0, %%xmm2 \n\t"
560
"psubw %%xmm1, %%xmm3 \n\t"
561
"movdqa %%xmm2, (%0) \n\t"
562
"movdqa %%xmm3, 16(%0) \n\t"
563
"movdqa %%xmm4, %%xmm6 \n\t"
564
"movdqa %%xmm5, %%xmm0 \n\t"
565
"punpcklwd %%xmm7, %%xmm4 \n\t"
566
"punpckhwd %%xmm7, %%xmm6 \n\t"
567
"punpcklwd %%xmm7, %%xmm5 \n\t"
568
"punpckhwd %%xmm7, %%xmm0 \n\t"
569
"paddd (%1), %%xmm4 \n\t"
570
"paddd 16(%1), %%xmm6 \n\t"
571
"paddd 32(%1), %%xmm5 \n\t"
572
"paddd 48(%1), %%xmm0 \n\t"
573
"movdqa %%xmm4, (%1) \n\t"
574
"movdqa %%xmm6, 16(%1) \n\t"
575
"movdqa %%xmm5, 32(%1) \n\t"
576
"movdqa %%xmm0, 48(%1) \n\t"
577
"add $32, %0 \n\t"
578
"add $64, %1 \n\t"
579
"add $32, %2 \n\t"
580
"cmp %3, %0 \n\t"
581
" jb 1b \n\t"
582
:
"+r"
(
block
),
"+r"
(sum),
"+r"
(offset)
583
:
"r"
(block+64)
584
XMM_CLOBBERS_ONLY
(
"%xmm0"
,
"%xmm1"
,
"%xmm2"
,
"%xmm3"
,
585
"%xmm4"
,
"%xmm5"
,
"%xmm6"
,
"%xmm7"
)
586
);
587
}
588
589
#if HAVE_SSSE3
590
#define HAVE_SSSE3_BAK
591
#endif
592
#undef HAVE_SSSE3
593
#define HAVE_SSSE3 0
594
595
#undef HAVE_SSE2
596
#undef HAVE_MMX2
597
#define HAVE_SSE2 0
598
#define HAVE_MMX2 0
599
#define RENAME(a) a ## _MMX
600
#define RENAMEl(a) a ## _mmx
601
#include "
mpegvideo_mmx_template.c
"
602
603
#undef HAVE_MMX2
604
#define HAVE_MMX2 1
605
#undef RENAME
606
#undef RENAMEl
607
#define RENAME(a) a ## _MMX2
608
#define RENAMEl(a) a ## _mmx2
609
#include "
mpegvideo_mmx_template.c
"
610
611
#undef HAVE_SSE2
612
#define HAVE_SSE2 1
613
#undef RENAME
614
#undef RENAMEl
615
#define RENAME(a) a ## _SSE2
616
#define RENAMEl(a) a ## _sse2
617
#include "
mpegvideo_mmx_template.c
"
618
619
#ifdef HAVE_SSSE3_BAK
620
#undef HAVE_SSSE3
621
#define HAVE_SSSE3 1
622
#undef RENAME
623
#undef RENAMEl
624
#define RENAME(a) a ## _SSSE3
625
#define RENAMEl(a) a ## _sse2
626
#include "
mpegvideo_mmx_template.c
"
627
#endif
628
629
void
MPV_common_init_mmx
(
MpegEncContext
*s)
630
{
631
int
mm_flags =
av_get_cpu_flags
();
632
633
if
(mm_flags &
AV_CPU_FLAG_MMX
) {
634
const
int
dct_algo = s->
avctx
->
dct_algo
;
635
636
s->
dct_unquantize_h263_intra
=
dct_unquantize_h263_intra_mmx
;
637
s->
dct_unquantize_h263_inter
=
dct_unquantize_h263_inter_mmx
;
638
s->
dct_unquantize_mpeg1_intra
=
dct_unquantize_mpeg1_intra_mmx
;
639
s->
dct_unquantize_mpeg1_inter
=
dct_unquantize_mpeg1_inter_mmx
;
640
if
(!(s->
flags
&
CODEC_FLAG_BITEXACT
))
641
s->
dct_unquantize_mpeg2_intra
=
dct_unquantize_mpeg2_intra_mmx
;
642
s->
dct_unquantize_mpeg2_inter
=
dct_unquantize_mpeg2_inter_mmx
;
643
644
if
(mm_flags &
AV_CPU_FLAG_SSE2
) {
645
s->
denoise_dct
=
denoise_dct_sse2
;
646
}
else
{
647
s->
denoise_dct
=
denoise_dct_mmx
;
648
}
649
650
if
(dct_algo==
FF_DCT_AUTO
|| dct_algo==
FF_DCT_MMX
){
651
#if HAVE_SSSE3
652
if
(mm_flags &
AV_CPU_FLAG_SSSE3
){
653
s->
dct_quantize
= dct_quantize_SSSE3;
654
}
else
655
#endif
656
if
(mm_flags & AV_CPU_FLAG_SSE2){
657
s->
dct_quantize
= dct_quantize_SSE2;
658
}
else
if
(mm_flags &
AV_CPU_FLAG_MMX2
){
659
s->
dct_quantize
= dct_quantize_MMX2;
660
}
else
{
661
s->
dct_quantize
= dct_quantize_MMX;
662
}
663
}
664
}
665
}