1/*
2 * MMX optimized forward DCT
3 * The gcc porting is Copyright (c) 2001 Fabrice Bellard.
4 * cleanup/optimizations are Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5 * SSE2 optimization is Copyright (c) 2004 Denes Balatoni.
6 *
7 * from  fdctam32.c - AP922 MMX(3D-Now) forward-DCT
8 *
9 *  Intel Application Note AP-922 - fast, precise implementation of DCT
10 *        http://developer.intel.com/vtune/cbts/appnotes.htm
11 *
12 * Also of inspiration:
13 * a page about fdct at http://www.geocities.com/ssavekar/dct.htm
14 * Skal's fdct at http://skal.planet-d.net/coding/dct.html
15 *
16 * This file is part of FFmpeg.
17 *
18 * FFmpeg is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU Lesser General Public
20 * License as published by the Free Software Foundation; either
21 * version 2.1 of the License, or (at your option) any later version.
22 *
23 * FFmpeg is distributed in the hope that it will be useful,
24 * but WITHOUT ANY WARRANTY; without even the implied warranty of
25 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
26 * Lesser General Public License for more details.
27 *
28 * You should have received a copy of the GNU Lesser General Public
29 * License along with FFmpeg; if not, write to the Free Software
30 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
31 */
32
33#include "libavutil/common.h"
34#include "libavcodec/dsputil.h"
35
36#define ATTR_ALIGN(align) __attribute__ ((__aligned__ (align)))
37
38//////////////////////////////////////////////////////////////////////
39//
40// constants for the forward DCT
41// -----------------------------
42//
43// Be sure to check that your compiler is aligning all constants to QWORD
44// (8-byte) memory boundaries!  Otherwise the unaligned memory access will
45// severely stall MMX execution.
46//
47//////////////////////////////////////////////////////////////////////
48
49#define BITS_FRW_ACC   3 //; 2 or 3 for accuracy
50#define SHIFT_FRW_COL  BITS_FRW_ACC
51#define SHIFT_FRW_ROW  (BITS_FRW_ACC + 17 - 3)
52#define RND_FRW_ROW    (1 << (SHIFT_FRW_ROW-1))
53//#define RND_FRW_COL    (1 << (SHIFT_FRW_COL-1))
54
55#define X8(x) x,x,x,x,x,x,x,x
56
57//concatenated table, for forward DCT transformation
58static const int16_t fdct_tg_all_16[24] ATTR_ALIGN(16) = {
59    X8(13036),  // tg * (2<<16) + 0.5
60    X8(27146),  // tg * (2<<16) + 0.5
61    X8(-21746)  // tg * (2<<16) + 0.5
62};
63
64static const int16_t ocos_4_16[8] ATTR_ALIGN(16) = {
65    X8(23170)   //cos * (2<<15) + 0.5
66};
67
68static const int16_t fdct_one_corr[8] ATTR_ALIGN(16) = { X8(1) };
69
70static const int32_t fdct_r_row[2] ATTR_ALIGN(8) = {RND_FRW_ROW, RND_FRW_ROW };
71
72static struct
73{
74 const int32_t fdct_r_row_sse2[4] ATTR_ALIGN(16);
75} fdct_r_row_sse2 ATTR_ALIGN(16)=
76{{
77 RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW
78}};
79//static const long fdct_r_row_sse2[4] ATTR_ALIGN(16) = {RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW};
80
81static const int16_t tab_frw_01234567[] ATTR_ALIGN(8) = {  // forward_dct coeff table
82  16384,   16384,   22725,   19266,
83  16384,   16384,   12873,    4520,
84  21407,    8867,   19266,   -4520,
85  -8867,  -21407,  -22725,  -12873,
86  16384,  -16384,   12873,  -22725,
87 -16384,   16384,    4520,   19266,
88   8867,  -21407,    4520,  -12873,
89  21407,   -8867,   19266,  -22725,
90
91  22725,   22725,   31521,   26722,
92  22725,   22725,   17855,    6270,
93  29692,   12299,   26722,   -6270,
94 -12299,  -29692,  -31521,  -17855,
95  22725,  -22725,   17855,  -31521,
96 -22725,   22725,    6270,   26722,
97  12299,  -29692,    6270,  -17855,
98  29692,  -12299,   26722,  -31521,
99
100  21407,   21407,   29692,   25172,
101  21407,   21407,   16819,    5906,
102  27969,   11585,   25172,   -5906,
103 -11585,  -27969,  -29692,  -16819,
104  21407,  -21407,   16819,  -29692,
105 -21407,   21407,    5906,   25172,
106  11585,  -27969,    5906,  -16819,
107  27969,  -11585,   25172,  -29692,
108
109  19266,   19266,   26722,   22654,
110  19266,   19266,   15137,    5315,
111  25172,   10426,   22654,   -5315,
112 -10426,  -25172,  -26722,  -15137,
113  19266,  -19266,   15137,  -26722,
114 -19266,   19266,    5315,   22654,
115  10426,  -25172,    5315,  -15137,
116  25172,  -10426,   22654,  -26722,
117
118  16384,   16384,   22725,   19266,
119  16384,   16384,   12873,    4520,
120  21407,    8867,   19266,   -4520,
121  -8867,  -21407,  -22725,  -12873,
122  16384,  -16384,   12873,  -22725,
123 -16384,   16384,    4520,   19266,
124   8867,  -21407,    4520,  -12873,
125  21407,   -8867,   19266,  -22725,
126
127  19266,   19266,   26722,   22654,
128  19266,   19266,   15137,    5315,
129  25172,   10426,   22654,   -5315,
130 -10426,  -25172,  -26722,  -15137,
131  19266,  -19266,   15137,  -26722,
132 -19266,   19266,    5315,   22654,
133  10426,  -25172,    5315,  -15137,
134  25172,  -10426,   22654,  -26722,
135
136  21407,   21407,   29692,   25172,
137  21407,   21407,   16819,    5906,
138  27969,   11585,   25172,   -5906,
139 -11585,  -27969,  -29692,  -16819,
140  21407,  -21407,   16819,  -29692,
141 -21407,   21407,    5906,   25172,
142  11585,  -27969,    5906,  -16819,
143  27969,  -11585,   25172,  -29692,
144
145  22725,   22725,   31521,   26722,
146  22725,   22725,   17855,    6270,
147  29692,   12299,   26722,   -6270,
148 -12299,  -29692,  -31521,  -17855,
149  22725,  -22725,   17855,  -31521,
150 -22725,   22725,    6270,   26722,
151  12299,  -29692,    6270,  -17855,
152  29692,  -12299,   26722,  -31521,
153};
154
155static struct
156{
157 const int16_t tab_frw_01234567_sse2[256] ATTR_ALIGN(16);
158} tab_frw_01234567_sse2 ATTR_ALIGN(16) =
159{{
160//static const int16_t tab_frw_01234567_sse2[] ATTR_ALIGN(16) = {  // forward_dct coeff table
161#define TABLE_SSE2 C4,  C4,  C1,  C3, -C6, -C2, -C1, -C5, \
162                   C4,  C4,  C5,  C7,  C2,  C6,  C3, -C7, \
163                  -C4,  C4,  C7,  C3,  C6, -C2,  C7, -C5, \
164                   C4, -C4,  C5, -C1,  C2, -C6,  C3, -C1,
165// c1..c7 * cos(pi/4) * 2^15
166#define C1 22725
167#define C2 21407
168#define C3 19266
169#define C4 16384
170#define C5 12873
171#define C6 8867
172#define C7 4520
173TABLE_SSE2
174
175#undef C1
176#undef C2
177#undef C3
178#undef C4
179#undef C5
180#undef C6
181#undef C7
182#define C1 31521
183#define C2 29692
184#define C3 26722
185#define C4 22725
186#define C5 17855
187#define C6 12299
188#define C7 6270
189TABLE_SSE2
190
191#undef C1
192#undef C2
193#undef C3
194#undef C4
195#undef C5
196#undef C6
197#undef C7
198#define C1 29692
199#define C2 27969
200#define C3 25172
201#define C4 21407
202#define C5 16819
203#define C6 11585
204#define C7 5906
205TABLE_SSE2
206
207#undef C1
208#undef C2
209#undef C3
210#undef C4
211#undef C5
212#undef C6
213#undef C7
214#define C1 26722
215#define C2 25172
216#define C3 22654
217#define C4 19266
218#define C5 15137
219#define C6 10426
220#define C7 5315
221TABLE_SSE2
222
223#undef C1
224#undef C2
225#undef C3
226#undef C4
227#undef C5
228#undef C6
229#undef C7
230#define C1 22725
231#define C2 21407
232#define C3 19266
233#define C4 16384
234#define C5 12873
235#define C6 8867
236#define C7 4520
237TABLE_SSE2
238
239#undef C1
240#undef C2
241#undef C3
242#undef C4
243#undef C5
244#undef C6
245#undef C7
246#define C1 26722
247#define C2 25172
248#define C3 22654
249#define C4 19266
250#define C5 15137
251#define C6 10426
252#define C7 5315
253TABLE_SSE2
254
255#undef C1
256#undef C2
257#undef C3
258#undef C4
259#undef C5
260#undef C6
261#undef C7
262#define C1 29692
263#define C2 27969
264#define C3 25172
265#define C4 21407
266#define C5 16819
267#define C6 11585
268#define C7 5906
269TABLE_SSE2
270
271#undef C1
272#undef C2
273#undef C3
274#undef C4
275#undef C5
276#undef C6
277#undef C7
278#define C1 31521
279#define C2 29692
280#define C3 26722
281#define C4 22725
282#define C5 17855
283#define C6 12299
284#define C7 6270
285TABLE_SSE2
286}};
287
288#define S(s) AV_TOSTRING(s) //AV_STRINGIFY is too long
289
290#define FDCT_COL(cpu, mm, mov)\
291static av_always_inline void fdct_col_##cpu(const int16_t *in, int16_t *out, int offset)\
292{\
293    __asm__ volatile (\
294        #mov"      16(%0),  %%"#mm"0 \n\t" \
295        #mov"      96(%0),  %%"#mm"1 \n\t" \
296        #mov"    %%"#mm"0,  %%"#mm"2 \n\t" \
297        #mov"      32(%0),  %%"#mm"3 \n\t" \
298        "paddsw  %%"#mm"1,  %%"#mm"0 \n\t" \
299        #mov"      80(%0),  %%"#mm"4 \n\t" \
300        "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"0 \n\t" \
301        #mov"        (%0),  %%"#mm"5 \n\t" \
302        "paddsw  %%"#mm"3,  %%"#mm"4 \n\t" \
303        "paddsw   112(%0),  %%"#mm"5 \n\t" \
304        "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"4 \n\t" \
305        #mov"    %%"#mm"0,  %%"#mm"6 \n\t" \
306        "psubsw  %%"#mm"1,  %%"#mm"2 \n\t" \
307        #mov"      16(%1),  %%"#mm"1 \n\t" \
308        "psubsw  %%"#mm"4,  %%"#mm"0 \n\t" \
309        #mov"      48(%0),  %%"#mm"7 \n\t" \
310        "pmulhw  %%"#mm"0,  %%"#mm"1 \n\t" \
311        "paddsw    64(%0),  %%"#mm"7 \n\t" \
312        "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"5 \n\t" \
313        "paddsw  %%"#mm"4,  %%"#mm"6 \n\t" \
314        "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"7 \n\t" \
315        #mov"    %%"#mm"5,  %%"#mm"4 \n\t" \
316        "psubsw  %%"#mm"7,  %%"#mm"5 \n\t" \
317        "paddsw  %%"#mm"5,  %%"#mm"1 \n\t" \
318        "paddsw  %%"#mm"7,  %%"#mm"4 \n\t" \
319        "por         (%2),  %%"#mm"1 \n\t" \
320        "psllw  $"S(SHIFT_FRW_COL)"+1, %%"#mm"2 \n\t" \
321        "pmulhw    16(%1),  %%"#mm"5 \n\t" \
322        #mov"    %%"#mm"4,  %%"#mm"7 \n\t" \
323        "psubsw    80(%0),  %%"#mm"3 \n\t" \
324        "psubsw  %%"#mm"6,  %%"#mm"4 \n\t" \
325        #mov"    %%"#mm"1,    32(%3) \n\t" \
326        "paddsw  %%"#mm"6,  %%"#mm"7 \n\t" \
327        #mov"      48(%0),  %%"#mm"1 \n\t" \
328        "psllw  $"S(SHIFT_FRW_COL)"+1, %%"#mm"3 \n\t" \
329        "psubsw    64(%0),  %%"#mm"1 \n\t" \
330        #mov"    %%"#mm"2,  %%"#mm"6 \n\t" \
331        #mov"    %%"#mm"4,    64(%3) \n\t" \
332        "paddsw  %%"#mm"3,  %%"#mm"2 \n\t" \
333        "pmulhw      (%4),  %%"#mm"2 \n\t" \
334        "psubsw  %%"#mm"3,  %%"#mm"6 \n\t" \
335        "pmulhw      (%4),  %%"#mm"6 \n\t" \
336        "psubsw  %%"#mm"0,  %%"#mm"5 \n\t" \
337        "por         (%2),  %%"#mm"5 \n\t" \
338        "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"1 \n\t" \
339        "por         (%2),  %%"#mm"2 \n\t" \
340        #mov"    %%"#mm"1,  %%"#mm"4 \n\t" \
341        #mov"        (%0),  %%"#mm"3 \n\t" \
342        "paddsw  %%"#mm"6,  %%"#mm"1 \n\t" \
343        "psubsw   112(%0),  %%"#mm"3 \n\t" \
344        "psubsw  %%"#mm"6,  %%"#mm"4 \n\t" \
345        #mov"        (%1),  %%"#mm"0 \n\t" \
346        "psllw  $"S(SHIFT_FRW_COL)", %%"#mm"3 \n\t" \
347        #mov"      32(%1),  %%"#mm"6 \n\t" \
348        "pmulhw  %%"#mm"1,  %%"#mm"0 \n\t" \
349        #mov"    %%"#mm"7,      (%3) \n\t" \
350        "pmulhw  %%"#mm"4,  %%"#mm"6 \n\t" \
351        #mov"    %%"#mm"5,    96(%3) \n\t" \
352        #mov"    %%"#mm"3,  %%"#mm"7 \n\t" \
353        #mov"      32(%1),  %%"#mm"5 \n\t" \
354        "psubsw  %%"#mm"2,  %%"#mm"7 \n\t" \
355        "paddsw  %%"#mm"2,  %%"#mm"3 \n\t" \
356        "pmulhw  %%"#mm"7,  %%"#mm"5 \n\t" \
357        "paddsw  %%"#mm"3,  %%"#mm"0 \n\t" \
358        "paddsw  %%"#mm"4,  %%"#mm"6 \n\t" \
359        "pmulhw      (%1),  %%"#mm"3 \n\t" \
360        "por         (%2),  %%"#mm"0 \n\t" \
361        "paddsw  %%"#mm"7,  %%"#mm"5 \n\t" \
362        "psubsw  %%"#mm"6,  %%"#mm"7 \n\t" \
363        #mov"    %%"#mm"0,    16(%3) \n\t" \
364        "paddsw  %%"#mm"4,  %%"#mm"5 \n\t" \
365        #mov"    %%"#mm"7,    48(%3) \n\t" \
366        "psubsw  %%"#mm"1,  %%"#mm"3 \n\t" \
367        #mov"    %%"#mm"5,    80(%3) \n\t" \
368        #mov"    %%"#mm"3,   112(%3) \n\t" \
369        : \
370        : "r" (in  + offset), "r" (fdct_tg_all_16), "r" (fdct_one_corr), \
371          "r" (out + offset), "r" (ocos_4_16)); \
372}
373
374FDCT_COL(mmx, mm, movq)
375FDCT_COL(sse2, xmm, movdqa)
376
377static av_always_inline void fdct_row_sse2(const int16_t *in, int16_t *out)
378{
379    __asm__ volatile(
380#define FDCT_ROW_SSE2_H1(i,t)                    \
381        "movq      " #i "(%0), %%xmm2      \n\t" \
382        "movq      " #i "+8(%0), %%xmm0    \n\t" \
383        "movdqa    " #t "+32(%1), %%xmm3   \n\t" \
384        "movdqa    " #t "+48(%1), %%xmm7   \n\t" \
385        "movdqa    " #t "(%1), %%xmm4      \n\t" \
386        "movdqa    " #t "+16(%1), %%xmm5   \n\t"
387
388#define FDCT_ROW_SSE2_H2(i,t)                    \
389        "movq      " #i "(%0), %%xmm2      \n\t" \
390        "movq      " #i "+8(%0), %%xmm0    \n\t" \
391        "movdqa    " #t "+32(%1), %%xmm3   \n\t" \
392        "movdqa    " #t "+48(%1), %%xmm7   \n\t"
393
394#define FDCT_ROW_SSE2(i)                      \
395        "movq      %%xmm2, %%xmm1       \n\t" \
396        "pshuflw   $27, %%xmm0, %%xmm0  \n\t" \
397        "paddsw    %%xmm0, %%xmm1       \n\t" \
398        "psubsw    %%xmm0, %%xmm2       \n\t" \
399        "punpckldq %%xmm2, %%xmm1       \n\t" \
400        "pshufd    $78, %%xmm1, %%xmm2  \n\t" \
401        "pmaddwd   %%xmm2, %%xmm3       \n\t" \
402        "pmaddwd   %%xmm1, %%xmm7       \n\t" \
403        "pmaddwd   %%xmm5, %%xmm2       \n\t" \
404        "pmaddwd   %%xmm4, %%xmm1       \n\t" \
405        "paddd     %%xmm7, %%xmm3       \n\t" \
406        "paddd     %%xmm2, %%xmm1       \n\t" \
407        "paddd     %%xmm6, %%xmm3       \n\t" \
408        "paddd     %%xmm6, %%xmm1       \n\t" \
409        "psrad     %3, %%xmm3           \n\t" \
410        "psrad     %3, %%xmm1           \n\t" \
411        "packssdw  %%xmm3, %%xmm1       \n\t" \
412        "movdqa    %%xmm1, " #i "(%4)   \n\t"
413
414        "movdqa    (%2), %%xmm6         \n\t"
415        FDCT_ROW_SSE2_H1(0,0)
416        FDCT_ROW_SSE2(0)
417        FDCT_ROW_SSE2_H2(64,0)
418        FDCT_ROW_SSE2(64)
419
420        FDCT_ROW_SSE2_H1(16,64)
421        FDCT_ROW_SSE2(16)
422        FDCT_ROW_SSE2_H2(112,64)
423        FDCT_ROW_SSE2(112)
424
425        FDCT_ROW_SSE2_H1(32,128)
426        FDCT_ROW_SSE2(32)
427        FDCT_ROW_SSE2_H2(96,128)
428        FDCT_ROW_SSE2(96)
429
430        FDCT_ROW_SSE2_H1(48,192)
431        FDCT_ROW_SSE2(48)
432        FDCT_ROW_SSE2_H2(80,192)
433        FDCT_ROW_SSE2(80)
434        :
435        : "r" (in), "r" (tab_frw_01234567_sse2.tab_frw_01234567_sse2), "r" (fdct_r_row_sse2.fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out)
436    );
437}
438
439static av_always_inline void fdct_row_mmx2(const int16_t *in, int16_t *out, const int16_t *table)
440{
441    __asm__ volatile (
442        "pshufw    $0x1B, 8(%0), %%mm5 \n\t"
443        "movq       (%0), %%mm0 \n\t"
444        "movq      %%mm0, %%mm1 \n\t"
445        "paddsw    %%mm5, %%mm0 \n\t"
446        "psubsw    %%mm5, %%mm1 \n\t"
447        "movq      %%mm0, %%mm2 \n\t"
448        "punpckldq %%mm1, %%mm0 \n\t"
449        "punpckhdq %%mm1, %%mm2 \n\t"
450        "movq       (%1), %%mm1 \n\t"
451        "movq      8(%1), %%mm3 \n\t"
452        "movq     16(%1), %%mm4 \n\t"
453        "movq     24(%1), %%mm5 \n\t"
454        "movq     32(%1), %%mm6 \n\t"
455        "movq     40(%1), %%mm7 \n\t"
456        "pmaddwd   %%mm0, %%mm1 \n\t"
457        "pmaddwd   %%mm2, %%mm3 \n\t"
458        "pmaddwd   %%mm0, %%mm4 \n\t"
459        "pmaddwd   %%mm2, %%mm5 \n\t"
460        "pmaddwd   %%mm0, %%mm6 \n\t"
461        "pmaddwd   %%mm2, %%mm7 \n\t"
462        "pmaddwd  48(%1), %%mm0 \n\t"
463        "pmaddwd  56(%1), %%mm2 \n\t"
464        "paddd     %%mm1, %%mm3 \n\t"
465        "paddd     %%mm4, %%mm5 \n\t"
466        "paddd     %%mm6, %%mm7 \n\t"
467        "paddd     %%mm0, %%mm2 \n\t"
468        "movq       (%2), %%mm0 \n\t"
469        "paddd     %%mm0, %%mm3 \n\t"
470        "paddd     %%mm0, %%mm5 \n\t"
471        "paddd     %%mm0, %%mm7 \n\t"
472        "paddd     %%mm0, %%mm2 \n\t"
473        "psrad $"S(SHIFT_FRW_ROW)", %%mm3 \n\t"
474        "psrad $"S(SHIFT_FRW_ROW)", %%mm5 \n\t"
475        "psrad $"S(SHIFT_FRW_ROW)", %%mm7 \n\t"
476        "psrad $"S(SHIFT_FRW_ROW)", %%mm2 \n\t"
477        "packssdw  %%mm5, %%mm3 \n\t"
478        "packssdw  %%mm2, %%mm7 \n\t"
479        "movq      %%mm3,  (%3) \n\t"
480        "movq      %%mm7, 8(%3) \n\t"
481        :
482        : "r" (in), "r" (table), "r" (fdct_r_row), "r" (out));
483}
484
485static av_always_inline void fdct_row_mmx(const int16_t *in, int16_t *out, const int16_t *table)
486{
487    //FIXME reorder (I do not have an old MMX-only CPU here to benchmark ...)
488    __asm__ volatile(
489        "movd     12(%0), %%mm1 \n\t"
490        "punpcklwd 8(%0), %%mm1 \n\t"
491        "movq      %%mm1, %%mm2 \n\t"
492        "psrlq     $0x20, %%mm1 \n\t"
493        "movq      0(%0), %%mm0 \n\t"
494        "punpcklwd %%mm2, %%mm1 \n\t"
495        "movq      %%mm0, %%mm5 \n\t"
496        "paddsw    %%mm1, %%mm0 \n\t"
497        "psubsw    %%mm1, %%mm5 \n\t"
498        "movq      %%mm0, %%mm2 \n\t"
499        "punpckldq %%mm5, %%mm0 \n\t"
500        "punpckhdq %%mm5, %%mm2 \n\t"
501        "movq      0(%1), %%mm1 \n\t"
502        "movq      8(%1), %%mm3 \n\t"
503        "movq     16(%1), %%mm4 \n\t"
504        "movq     24(%1), %%mm5 \n\t"
505        "movq     32(%1), %%mm6 \n\t"
506        "movq     40(%1), %%mm7 \n\t"
507        "pmaddwd   %%mm0, %%mm1 \n\t"
508        "pmaddwd   %%mm2, %%mm3 \n\t"
509        "pmaddwd   %%mm0, %%mm4 \n\t"
510        "pmaddwd   %%mm2, %%mm5 \n\t"
511        "pmaddwd   %%mm0, %%mm6 \n\t"
512        "pmaddwd   %%mm2, %%mm7 \n\t"
513        "pmaddwd  48(%1), %%mm0 \n\t"
514        "pmaddwd  56(%1), %%mm2 \n\t"
515        "paddd     %%mm1, %%mm3 \n\t"
516        "paddd     %%mm4, %%mm5 \n\t"
517        "paddd     %%mm6, %%mm7 \n\t"
518        "paddd     %%mm0, %%mm2 \n\t"
519        "movq       (%2), %%mm0 \n\t"
520        "paddd     %%mm0, %%mm3 \n\t"
521        "paddd     %%mm0, %%mm5 \n\t"
522        "paddd     %%mm0, %%mm7 \n\t"
523        "paddd     %%mm0, %%mm2 \n\t"
524        "psrad $"S(SHIFT_FRW_ROW)", %%mm3 \n\t"
525        "psrad $"S(SHIFT_FRW_ROW)", %%mm5 \n\t"
526        "psrad $"S(SHIFT_FRW_ROW)", %%mm7 \n\t"
527        "psrad $"S(SHIFT_FRW_ROW)", %%mm2 \n\t"
528        "packssdw  %%mm5, %%mm3 \n\t"
529        "packssdw  %%mm2, %%mm7 \n\t"
530        "movq      %%mm3, 0(%3) \n\t"
531        "movq      %%mm7, 8(%3) \n\t"
532        :
533        : "r" (in), "r" (table), "r" (fdct_r_row), "r" (out));
534}
535
536void ff_fdct_mmx(int16_t *block)
537{
538    int64_t align_tmp[16] ATTR_ALIGN(8);
539    int16_t * block1= (int16_t*)align_tmp;
540    const int16_t *table= tab_frw_01234567;
541    int i;
542
543    fdct_col_mmx(block, block1, 0);
544    fdct_col_mmx(block, block1, 4);
545
546    for(i=8;i>0;i--) {
547        fdct_row_mmx(block1, block, table);
548        block1 += 8;
549        table += 32;
550        block += 8;
551    }
552}
553
554void ff_fdct_mmx2(int16_t *block)
555{
556    int64_t align_tmp[16] ATTR_ALIGN(8);
557    int16_t *block1= (int16_t*)align_tmp;
558    const int16_t *table= tab_frw_01234567;
559    int i;
560
561    fdct_col_mmx(block, block1, 0);
562    fdct_col_mmx(block, block1, 4);
563
564    for(i=8;i>0;i--) {
565        fdct_row_mmx2(block1, block, table);
566        block1 += 8;
567        table += 32;
568        block += 8;
569    }
570}
571
572void ff_fdct_sse2(int16_t *block)
573{
574    int64_t align_tmp[16] ATTR_ALIGN(16);
575    int16_t * const block1= (int16_t*)align_tmp;
576
577    fdct_col_sse2(block, block1, 0);
578    fdct_row_sse2(block1, block);
579}
580
581