fdct_mmx.c
Go to the documentation of this file.
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 Libav.
17  *
18  * Libav 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  * Libav 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 Libav; 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 "libavutil/x86_cpu.h"
35 #include "libavcodec/dsputil.h"
36 
38 //
39 // constants for the forward DCT
40 // -----------------------------
41 //
42 // Be sure to check that your compiler is aligning all constants to QWORD
43 // (8-byte) memory boundaries! Otherwise the unaligned memory access will
44 // severely stall MMX execution.
45 //
47 
48 #define BITS_FRW_ACC 3 //; 2 or 3 for accuracy
49 #define SHIFT_FRW_COL BITS_FRW_ACC
50 #define SHIFT_FRW_ROW (BITS_FRW_ACC + 17 - 3)
51 #define RND_FRW_ROW (1 << (SHIFT_FRW_ROW-1))
52 //#define RND_FRW_COL (1 << (SHIFT_FRW_COL-1))
53 
54 #define X8(x) x,x,x,x,x,x,x,x
55 
56 //concatenated table, for forward DCT transformation
57 DECLARE_ALIGNED(16, static const int16_t, fdct_tg_all_16)[24] = {
58  X8(13036), // tg * (2<<16) + 0.5
59  X8(27146), // tg * (2<<16) + 0.5
60  X8(-21746) // tg * (2<<16) + 0.5
61 };
62 
63 DECLARE_ALIGNED(16, static const int16_t, ocos_4_16)[8] = {
64  X8(23170) //cos * (2<<15) + 0.5
65 };
66 
67 DECLARE_ALIGNED(16, static const int16_t, fdct_one_corr)[8] = { X8(1) };
68 
69 DECLARE_ALIGNED(8, static const int32_t, fdct_r_row)[2] = {RND_FRW_ROW, RND_FRW_ROW };
70 
71 static struct
72 {
73  DECLARE_ALIGNED(16, const int32_t, fdct_r_row_sse2)[4];
75 {{
76  RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW
77 }};
78 //DECLARE_ALIGNED(16, static const long, fdct_r_row_sse2)[4] = {RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW};
79 
80 DECLARE_ALIGNED(8, static const int16_t, tab_frw_01234567)[] = { // forward_dct coeff table
81  16384, 16384, 22725, 19266,
82  16384, 16384, 12873, 4520,
83  21407, 8867, 19266, -4520,
84  -8867, -21407, -22725, -12873,
85  16384, -16384, 12873, -22725,
86  -16384, 16384, 4520, 19266,
87  8867, -21407, 4520, -12873,
88  21407, -8867, 19266, -22725,
89 
90  22725, 22725, 31521, 26722,
91  22725, 22725, 17855, 6270,
92  29692, 12299, 26722, -6270,
93  -12299, -29692, -31521, -17855,
94  22725, -22725, 17855, -31521,
95  -22725, 22725, 6270, 26722,
96  12299, -29692, 6270, -17855,
97  29692, -12299, 26722, -31521,
98 
99  21407, 21407, 29692, 25172,
100  21407, 21407, 16819, 5906,
101  27969, 11585, 25172, -5906,
102  -11585, -27969, -29692, -16819,
103  21407, -21407, 16819, -29692,
104  -21407, 21407, 5906, 25172,
105  11585, -27969, 5906, -16819,
106  27969, -11585, 25172, -29692,
107 
108  19266, 19266, 26722, 22654,
109  19266, 19266, 15137, 5315,
110  25172, 10426, 22654, -5315,
111  -10426, -25172, -26722, -15137,
112  19266, -19266, 15137, -26722,
113  -19266, 19266, 5315, 22654,
114  10426, -25172, 5315, -15137,
115  25172, -10426, 22654, -26722,
116 
117  16384, 16384, 22725, 19266,
118  16384, 16384, 12873, 4520,
119  21407, 8867, 19266, -4520,
120  -8867, -21407, -22725, -12873,
121  16384, -16384, 12873, -22725,
122  -16384, 16384, 4520, 19266,
123  8867, -21407, 4520, -12873,
124  21407, -8867, 19266, -22725,
125 
126  19266, 19266, 26722, 22654,
127  19266, 19266, 15137, 5315,
128  25172, 10426, 22654, -5315,
129  -10426, -25172, -26722, -15137,
130  19266, -19266, 15137, -26722,
131  -19266, 19266, 5315, 22654,
132  10426, -25172, 5315, -15137,
133  25172, -10426, 22654, -26722,
134 
135  21407, 21407, 29692, 25172,
136  21407, 21407, 16819, 5906,
137  27969, 11585, 25172, -5906,
138  -11585, -27969, -29692, -16819,
139  21407, -21407, 16819, -29692,
140  -21407, 21407, 5906, 25172,
141  11585, -27969, 5906, -16819,
142  27969, -11585, 25172, -29692,
143 
144  22725, 22725, 31521, 26722,
145  22725, 22725, 17855, 6270,
146  29692, 12299, 26722, -6270,
147  -12299, -29692, -31521, -17855,
148  22725, -22725, 17855, -31521,
149  -22725, 22725, 6270, 26722,
150  12299, -29692, 6270, -17855,
151  29692, -12299, 26722, -31521,
152 };
153 
154 static struct
155 {
156  DECLARE_ALIGNED(16, const int16_t, tab_frw_01234567_sse2)[256];
158 {{
159 //DECLARE_ALIGNED(16, static const int16_t, tab_frw_01234567_sse2)[] = { // forward_dct coeff table
160 #define TABLE_SSE2 C4, C4, C1, C3, -C6, -C2, -C1, -C5, \
161  C4, C4, C5, C7, C2, C6, C3, -C7, \
162  -C4, C4, C7, C3, C6, -C2, C7, -C5, \
163  C4, -C4, C5, -C1, C2, -C6, C3, -C1,
164 // c1..c7 * cos(pi/4) * 2^15
165 #define C1 22725
166 #define C2 21407
167 #define C3 19266
168 #define C4 16384
169 #define C5 12873
170 #define C6 8867
171 #define C7 4520
173 
174 #undef C1
175 #undef C2
176 #undef C3
177 #undef C4
178 #undef C5
179 #undef C6
180 #undef C7
181 #define C1 31521
182 #define C2 29692
183 #define C3 26722
184 #define C4 22725
185 #define C5 17855
186 #define C6 12299
187 #define C7 6270
189 
190 #undef C1
191 #undef C2
192 #undef C3
193 #undef C4
194 #undef C5
195 #undef C6
196 #undef C7
197 #define C1 29692
198 #define C2 27969
199 #define C3 25172
200 #define C4 21407
201 #define C5 16819
202 #define C6 11585
203 #define C7 5906
205 
206 #undef C1
207 #undef C2
208 #undef C3
209 #undef C4
210 #undef C5
211 #undef C6
212 #undef C7
213 #define C1 26722
214 #define C2 25172
215 #define C3 22654
216 #define C4 19266
217 #define C5 15137
218 #define C6 10426
219 #define C7 5315
221 
222 #undef C1
223 #undef C2
224 #undef C3
225 #undef C4
226 #undef C5
227 #undef C6
228 #undef C7
229 #define C1 22725
230 #define C2 21407
231 #define C3 19266
232 #define C4 16384
233 #define C5 12873
234 #define C6 8867
235 #define C7 4520
237 
238 #undef C1
239 #undef C2
240 #undef C3
241 #undef C4
242 #undef C5
243 #undef C6
244 #undef C7
245 #define C1 26722
246 #define C2 25172
247 #define C3 22654
248 #define C4 19266
249 #define C5 15137
250 #define C6 10426
251 #define C7 5315
253 
254 #undef C1
255 #undef C2
256 #undef C3
257 #undef C4
258 #undef C5
259 #undef C6
260 #undef C7
261 #define C1 29692
262 #define C2 27969
263 #define C3 25172
264 #define C4 21407
265 #define C5 16819
266 #define C6 11585
267 #define C7 5906
269 
270 #undef C1
271 #undef C2
272 #undef C3
273 #undef C4
274 #undef C5
275 #undef C6
276 #undef C7
277 #define C1 31521
278 #define C2 29692
279 #define C3 26722
280 #define C4 22725
281 #define C5 17855
282 #define C6 12299
283 #define C7 6270
285 }};
286 
287 #define S(s) AV_TOSTRING(s) //AV_STRINGIFY is too long
288 
289 #define FDCT_COL(cpu, mm, mov)\
290 static av_always_inline void fdct_col_##cpu(const int16_t *in, int16_t *out, int offset)\
291 {\
292  __asm__ volatile (\
293  #mov" 16(%0), %%"#mm"0 \n\t" \
294  #mov" 96(%0), %%"#mm"1 \n\t" \
295  #mov" %%"#mm"0, %%"#mm"2 \n\t" \
296  #mov" 32(%0), %%"#mm"3 \n\t" \
297  "paddsw %%"#mm"1, %%"#mm"0 \n\t" \
298  #mov" 80(%0), %%"#mm"4 \n\t" \
299  "psllw $"S(SHIFT_FRW_COL)", %%"#mm"0 \n\t" \
300  #mov" (%0), %%"#mm"5 \n\t" \
301  "paddsw %%"#mm"3, %%"#mm"4 \n\t" \
302  "paddsw 112(%0), %%"#mm"5 \n\t" \
303  "psllw $"S(SHIFT_FRW_COL)", %%"#mm"4 \n\t" \
304  #mov" %%"#mm"0, %%"#mm"6 \n\t" \
305  "psubsw %%"#mm"1, %%"#mm"2 \n\t" \
306  #mov" 16(%1), %%"#mm"1 \n\t" \
307  "psubsw %%"#mm"4, %%"#mm"0 \n\t" \
308  #mov" 48(%0), %%"#mm"7 \n\t" \
309  "pmulhw %%"#mm"0, %%"#mm"1 \n\t" \
310  "paddsw 64(%0), %%"#mm"7 \n\t" \
311  "psllw $"S(SHIFT_FRW_COL)", %%"#mm"5 \n\t" \
312  "paddsw %%"#mm"4, %%"#mm"6 \n\t" \
313  "psllw $"S(SHIFT_FRW_COL)", %%"#mm"7 \n\t" \
314  #mov" %%"#mm"5, %%"#mm"4 \n\t" \
315  "psubsw %%"#mm"7, %%"#mm"5 \n\t" \
316  "paddsw %%"#mm"5, %%"#mm"1 \n\t" \
317  "paddsw %%"#mm"7, %%"#mm"4 \n\t" \
318  "por (%2), %%"#mm"1 \n\t" \
319  "psllw $"S(SHIFT_FRW_COL)"+1, %%"#mm"2 \n\t" \
320  "pmulhw 16(%1), %%"#mm"5 \n\t" \
321  #mov" %%"#mm"4, %%"#mm"7 \n\t" \
322  "psubsw 80(%0), %%"#mm"3 \n\t" \
323  "psubsw %%"#mm"6, %%"#mm"4 \n\t" \
324  #mov" %%"#mm"1, 32(%3) \n\t" \
325  "paddsw %%"#mm"6, %%"#mm"7 \n\t" \
326  #mov" 48(%0), %%"#mm"1 \n\t" \
327  "psllw $"S(SHIFT_FRW_COL)"+1, %%"#mm"3 \n\t" \
328  "psubsw 64(%0), %%"#mm"1 \n\t" \
329  #mov" %%"#mm"2, %%"#mm"6 \n\t" \
330  #mov" %%"#mm"4, 64(%3) \n\t" \
331  "paddsw %%"#mm"3, %%"#mm"2 \n\t" \
332  "pmulhw (%4), %%"#mm"2 \n\t" \
333  "psubsw %%"#mm"3, %%"#mm"6 \n\t" \
334  "pmulhw (%4), %%"#mm"6 \n\t" \
335  "psubsw %%"#mm"0, %%"#mm"5 \n\t" \
336  "por (%2), %%"#mm"5 \n\t" \
337  "psllw $"S(SHIFT_FRW_COL)", %%"#mm"1 \n\t" \
338  "por (%2), %%"#mm"2 \n\t" \
339  #mov" %%"#mm"1, %%"#mm"4 \n\t" \
340  #mov" (%0), %%"#mm"3 \n\t" \
341  "paddsw %%"#mm"6, %%"#mm"1 \n\t" \
342  "psubsw 112(%0), %%"#mm"3 \n\t" \
343  "psubsw %%"#mm"6, %%"#mm"4 \n\t" \
344  #mov" (%1), %%"#mm"0 \n\t" \
345  "psllw $"S(SHIFT_FRW_COL)", %%"#mm"3 \n\t" \
346  #mov" 32(%1), %%"#mm"6 \n\t" \
347  "pmulhw %%"#mm"1, %%"#mm"0 \n\t" \
348  #mov" %%"#mm"7, (%3) \n\t" \
349  "pmulhw %%"#mm"4, %%"#mm"6 \n\t" \
350  #mov" %%"#mm"5, 96(%3) \n\t" \
351  #mov" %%"#mm"3, %%"#mm"7 \n\t" \
352  #mov" 32(%1), %%"#mm"5 \n\t" \
353  "psubsw %%"#mm"2, %%"#mm"7 \n\t" \
354  "paddsw %%"#mm"2, %%"#mm"3 \n\t" \
355  "pmulhw %%"#mm"7, %%"#mm"5 \n\t" \
356  "paddsw %%"#mm"3, %%"#mm"0 \n\t" \
357  "paddsw %%"#mm"4, %%"#mm"6 \n\t" \
358  "pmulhw (%1), %%"#mm"3 \n\t" \
359  "por (%2), %%"#mm"0 \n\t" \
360  "paddsw %%"#mm"7, %%"#mm"5 \n\t" \
361  "psubsw %%"#mm"6, %%"#mm"7 \n\t" \
362  #mov" %%"#mm"0, 16(%3) \n\t" \
363  "paddsw %%"#mm"4, %%"#mm"5 \n\t" \
364  #mov" %%"#mm"7, 48(%3) \n\t" \
365  "psubsw %%"#mm"1, %%"#mm"3 \n\t" \
366  #mov" %%"#mm"5, 80(%3) \n\t" \
367  #mov" %%"#mm"3, 112(%3) \n\t" \
368  : \
369  : "r" (in + offset), "r" (fdct_tg_all_16), "r" (fdct_one_corr), \
370  "r" (out + offset), "r" (ocos_4_16)); \
371 }
372 
373 FDCT_COL(mmx, mm, movq)
374 FDCT_COL(sse2, xmm, movdqa)
375 
376 static av_always_inline void fdct_row_sse2(const int16_t *in, int16_t *out)
377 {
378  __asm__ volatile(
379 #define FDCT_ROW_SSE2_H1(i,t) \
380  "movq " #i "(%0), %%xmm2 \n\t" \
381  "movq " #i "+8(%0), %%xmm0 \n\t" \
382  "movdqa " #t "+32(%1), %%xmm3 \n\t" \
383  "movdqa " #t "+48(%1), %%xmm7 \n\t" \
384  "movdqa " #t "(%1), %%xmm4 \n\t" \
385  "movdqa " #t "+16(%1), %%xmm5 \n\t"
386 
387 #define FDCT_ROW_SSE2_H2(i,t) \
388  "movq " #i "(%0), %%xmm2 \n\t" \
389  "movq " #i "+8(%0), %%xmm0 \n\t" \
390  "movdqa " #t "+32(%1), %%xmm3 \n\t" \
391  "movdqa " #t "+48(%1), %%xmm7 \n\t"
392 
393 #define FDCT_ROW_SSE2(i) \
394  "movq %%xmm2, %%xmm1 \n\t" \
395  "pshuflw $27, %%xmm0, %%xmm0 \n\t" \
396  "paddsw %%xmm0, %%xmm1 \n\t" \
397  "psubsw %%xmm0, %%xmm2 \n\t" \
398  "punpckldq %%xmm2, %%xmm1 \n\t" \
399  "pshufd $78, %%xmm1, %%xmm2 \n\t" \
400  "pmaddwd %%xmm2, %%xmm3 \n\t" \
401  "pmaddwd %%xmm1, %%xmm7 \n\t" \
402  "pmaddwd %%xmm5, %%xmm2 \n\t" \
403  "pmaddwd %%xmm4, %%xmm1 \n\t" \
404  "paddd %%xmm7, %%xmm3 \n\t" \
405  "paddd %%xmm2, %%xmm1 \n\t" \
406  "paddd %%xmm6, %%xmm3 \n\t" \
407  "paddd %%xmm6, %%xmm1 \n\t" \
408  "psrad %3, %%xmm3 \n\t" \
409  "psrad %3, %%xmm1 \n\t" \
410  "packssdw %%xmm3, %%xmm1 \n\t" \
411  "movdqa %%xmm1, " #i "(%4) \n\t"
412 
413  "movdqa (%2), %%xmm6 \n\t"
414  FDCT_ROW_SSE2_H1(0,0)
415  FDCT_ROW_SSE2(0)
416  FDCT_ROW_SSE2_H2(64,0)
417  FDCT_ROW_SSE2(64)
418 
419  FDCT_ROW_SSE2_H1(16,64)
420  FDCT_ROW_SSE2(16)
421  FDCT_ROW_SSE2_H2(112,64)
422  FDCT_ROW_SSE2(112)
423 
424  FDCT_ROW_SSE2_H1(32,128)
425  FDCT_ROW_SSE2(32)
426  FDCT_ROW_SSE2_H2(96,128)
427  FDCT_ROW_SSE2(96)
428 
429  FDCT_ROW_SSE2_H1(48,192)
430  FDCT_ROW_SSE2(48)
431  FDCT_ROW_SSE2_H2(80,192)
432  FDCT_ROW_SSE2(80)
433  :
434  : "r" (in), "r" (tab_frw_01234567_sse2.tab_frw_01234567_sse2),
435  "r" (fdct_r_row_sse2.fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out)
436  XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
437  "%xmm4", "%xmm5", "%xmm6", "%xmm7")
438  );
439 }
440 
441 static av_always_inline void fdct_row_mmx2(const int16_t *in, int16_t *out, const int16_t *table)
442 {
443  __asm__ volatile (
444  "pshufw $0x1B, 8(%0), %%mm5 \n\t"
445  "movq (%0), %%mm0 \n\t"
446  "movq %%mm0, %%mm1 \n\t"
447  "paddsw %%mm5, %%mm0 \n\t"
448  "psubsw %%mm5, %%mm1 \n\t"
449  "movq %%mm0, %%mm2 \n\t"
450  "punpckldq %%mm1, %%mm0 \n\t"
451  "punpckhdq %%mm1, %%mm2 \n\t"
452  "movq (%1), %%mm1 \n\t"
453  "movq 8(%1), %%mm3 \n\t"
454  "movq 16(%1), %%mm4 \n\t"
455  "movq 24(%1), %%mm5 \n\t"
456  "movq 32(%1), %%mm6 \n\t"
457  "movq 40(%1), %%mm7 \n\t"
458  "pmaddwd %%mm0, %%mm1 \n\t"
459  "pmaddwd %%mm2, %%mm3 \n\t"
460  "pmaddwd %%mm0, %%mm4 \n\t"
461  "pmaddwd %%mm2, %%mm5 \n\t"
462  "pmaddwd %%mm0, %%mm6 \n\t"
463  "pmaddwd %%mm2, %%mm7 \n\t"
464  "pmaddwd 48(%1), %%mm0 \n\t"
465  "pmaddwd 56(%1), %%mm2 \n\t"
466  "paddd %%mm1, %%mm3 \n\t"
467  "paddd %%mm4, %%mm5 \n\t"
468  "paddd %%mm6, %%mm7 \n\t"
469  "paddd %%mm0, %%mm2 \n\t"
470  "movq (%2), %%mm0 \n\t"
471  "paddd %%mm0, %%mm3 \n\t"
472  "paddd %%mm0, %%mm5 \n\t"
473  "paddd %%mm0, %%mm7 \n\t"
474  "paddd %%mm0, %%mm2 \n\t"
475  "psrad $"S(SHIFT_FRW_ROW)", %%mm3 \n\t"
476  "psrad $"S(SHIFT_FRW_ROW)", %%mm5 \n\t"
477  "psrad $"S(SHIFT_FRW_ROW)", %%mm7 \n\t"
478  "psrad $"S(SHIFT_FRW_ROW)", %%mm2 \n\t"
479  "packssdw %%mm5, %%mm3 \n\t"
480  "packssdw %%mm2, %%mm7 \n\t"
481  "movq %%mm3, (%3) \n\t"
482  "movq %%mm7, 8(%3) \n\t"
483  :
484  : "r" (in), "r" (table), "r" (fdct_r_row), "r" (out));
485 }
486 
487 static av_always_inline void fdct_row_mmx(const int16_t *in, int16_t *out, const int16_t *table)
488 {
489  //FIXME reorder (I do not have an old MMX-only CPU here to benchmark ...)
490  __asm__ volatile(
491  "movd 12(%0), %%mm1 \n\t"
492  "punpcklwd 8(%0), %%mm1 \n\t"
493  "movq %%mm1, %%mm2 \n\t"
494  "psrlq $0x20, %%mm1 \n\t"
495  "movq 0(%0), %%mm0 \n\t"
496  "punpcklwd %%mm2, %%mm1 \n\t"
497  "movq %%mm0, %%mm5 \n\t"
498  "paddsw %%mm1, %%mm0 \n\t"
499  "psubsw %%mm1, %%mm5 \n\t"
500  "movq %%mm0, %%mm2 \n\t"
501  "punpckldq %%mm5, %%mm0 \n\t"
502  "punpckhdq %%mm5, %%mm2 \n\t"
503  "movq 0(%1), %%mm1 \n\t"
504  "movq 8(%1), %%mm3 \n\t"
505  "movq 16(%1), %%mm4 \n\t"
506  "movq 24(%1), %%mm5 \n\t"
507  "movq 32(%1), %%mm6 \n\t"
508  "movq 40(%1), %%mm7 \n\t"
509  "pmaddwd %%mm0, %%mm1 \n\t"
510  "pmaddwd %%mm2, %%mm3 \n\t"
511  "pmaddwd %%mm0, %%mm4 \n\t"
512  "pmaddwd %%mm2, %%mm5 \n\t"
513  "pmaddwd %%mm0, %%mm6 \n\t"
514  "pmaddwd %%mm2, %%mm7 \n\t"
515  "pmaddwd 48(%1), %%mm0 \n\t"
516  "pmaddwd 56(%1), %%mm2 \n\t"
517  "paddd %%mm1, %%mm3 \n\t"
518  "paddd %%mm4, %%mm5 \n\t"
519  "paddd %%mm6, %%mm7 \n\t"
520  "paddd %%mm0, %%mm2 \n\t"
521  "movq (%2), %%mm0 \n\t"
522  "paddd %%mm0, %%mm3 \n\t"
523  "paddd %%mm0, %%mm5 \n\t"
524  "paddd %%mm0, %%mm7 \n\t"
525  "paddd %%mm0, %%mm2 \n\t"
526  "psrad $"S(SHIFT_FRW_ROW)", %%mm3 \n\t"
527  "psrad $"S(SHIFT_FRW_ROW)", %%mm5 \n\t"
528  "psrad $"S(SHIFT_FRW_ROW)", %%mm7 \n\t"
529  "psrad $"S(SHIFT_FRW_ROW)", %%mm2 \n\t"
530  "packssdw %%mm5, %%mm3 \n\t"
531  "packssdw %%mm2, %%mm7 \n\t"
532  "movq %%mm3, 0(%3) \n\t"
533  "movq %%mm7, 8(%3) \n\t"
534  :
535  : "r" (in), "r" (table), "r" (fdct_r_row), "r" (out));
536 }
537 
538 void ff_fdct_mmx(int16_t *block)
539 {
540  DECLARE_ALIGNED(8, int64_t, align_tmp)[16];
541  int16_t * block1= (int16_t*)align_tmp;
542  const int16_t *table= tab_frw_01234567;
543  int i;
544 
545  fdct_col_mmx(block, block1, 0);
546  fdct_col_mmx(block, block1, 4);
547 
548  for(i=8;i>0;i--) {
549  fdct_row_mmx(block1, block, table);
550  block1 += 8;
551  table += 32;
552  block += 8;
553  }
554 }
555 
556 void ff_fdct_mmx2(int16_t *block)
557 {
558  DECLARE_ALIGNED(8, int64_t, align_tmp)[16];
559  int16_t *block1= (int16_t*)align_tmp;
560  const int16_t *table= tab_frw_01234567;
561  int i;
562 
563  fdct_col_mmx(block, block1, 0);
564  fdct_col_mmx(block, block1, 4);
565 
566  for(i=8;i>0;i--) {
567  fdct_row_mmx2(block1, block, table);
568  block1 += 8;
569  table += 32;
570  block += 8;
571  }
572 }
573 
574 void ff_fdct_sse2(int16_t *block)
575 {
576  DECLARE_ALIGNED(16, int64_t, align_tmp)[16];
577  int16_t * const block1= (int16_t*)align_tmp;
578 
579  fdct_col_sse2(block, block1, 0);
580  fdct_row_sse2(block1, block);
581 }
582