Mercurial > libavcodec.hg

comparison x86/fft_sse.c @ 12100:db9ef48dc0e4 libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Move SSE optimized 32-point DCT to its own file. Should fix breakage with YASM disabled.
author vitor
date Tue, 06 Jul 2010 17:48:23 +0000
parents 1bf322283429
children 020540442072
comparison
equal deleted inserted replaced
12099:1bf322283429 12100:db9ef48dc0e4
18 * License along with FFmpeg; if not, write to the Free Software 18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */ 20 */
21 21
22 #include "libavutil/x86_cpu.h" 22 #include "libavutil/x86_cpu.h"
23 #include "libavutil/common.h"
24 #include "libavcodec/dsputil.h" 23 #include "libavcodec/dsputil.h"
25 #include "fft.h" 24 #include "fft.h"
26 25
27 DECLARE_ALIGNED(16, static const int, m1m1m1m1)[4] = 26 DECLARE_ALIGNED(16, static const int, m1m1m1m1)[4] =
28 { 1 << 31, 1 << 31, 1 << 31, 1 << 31 }; 27 { 1 << 31, 1 << 31, 1 << 31, 1 << 31 };
200 :"r"(output+n4), "r"(output+n4*3), 199 :"r"(output+n4), "r"(output+n4*3),
201 "m"(*m1m1m1m1) 200 "m"(*m1m1m1m1)
202 ); 201 );
203 } 202 }
204 203
205 DECLARE_ALIGNED(16, static const float, b1)[] = {
206 0.500603, 0.505471, 0.515447, 0.531043,
207 0.553104, 0.582935, 0.622504, 0.674808,
208 -1.169440, -0.972568, -0.839350, -0.744536,
209 -10.190008, -3.407609, -2.057781, -1.484165,
210 0.502419, 0.522499, 0.566944, 0.646822,
211 0.788155, 1.060678, 1.722447, 5.101149,
212 0.509796, 0.601345, 0.899976, 2.562916,
213 1.000000, 1.000000, 1.306563, 0.541196,
214 1.000000, 0.707107, 1.000000, -0.707107
215 };
216
217 DECLARE_ALIGNED(16, static const int32_t, smask)[4] = {
218 0, 0, 0x80000000, 0x80000000
219 };
220
221 /* butterfly operator */
222 #define BUTTERFLY(a,b,c,tmp) \
223 "movaps %%" #a ", %%" #tmp " \n\t" \
224 "subps %%" #b ", %%" #a " \n\t" \
225 "addps %%" #tmp ", %%" #b " \n\t" \
226 "mulps " #c ", %%" #a " \n\t"
227
228 ///* Same as BUTTERFLY when vectors a and b overlap */
229 #define BUTTERFLY0(val, mask, cos, tmp, shuf) \
230 "movaps %%" #val ", %%" #tmp " \n\t" \
231 "shufps " #shuf ", %%" #val ",%%" #val " \n\t" \
232 "xorps %%" #mask ", %%" #tmp " \n\t" /* flip signs */ \
233 "addps %%" #tmp ", %%" #val " \n\t" \
234 "mulps %%" #cos ", %%" #val " \n\t"
235
236 #define BUTTERFLY2(val, mask, cos, tmp) BUTTERFLY0(val, mask, cos, tmp, $0x1b)
237 #define BUTTERFLY3(val, mask, cos, tmp) BUTTERFLY0(val, mask, cos, tmp, $0xb1)
238
239 void ff_dct32_float_sse(FFTSample *out, const FFTSample *in)
240 {
241 int32_t tmp1 = 0;
242 __asm__ volatile(
243 /* pass 1 */
244
245 "movaps (%4), %%xmm0 \n\t"
246 "movaps 112(%4), %%xmm1 \n\t"
247 "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
248 BUTTERFLY(xmm0, xmm1, (%2), xmm3)
249
250 "movaps 64(%4), %%xmm7 \n\t"
251 "movaps 48(%4), %%xmm4 \n\t"
252 "shufps $0x1b, %%xmm4, %%xmm4 \n\t"
253 BUTTERFLY(xmm7, xmm4, 48(%2), xmm3)
254
255
256 /* pass 2 */
257 "movaps 64(%2), %%xmm2 \n\t"
258 BUTTERFLY(xmm1, xmm4, %%xmm2, xmm3)
259 "movaps %%xmm1, 48(%1) \n\t"
260 "movaps %%xmm4, (%1) \n\t"
261
262 /* pass 1 */
263 "movaps 16(%4), %%xmm1 \n\t"
264 "movaps 96(%4), %%xmm6 \n\t"
265 "shufps $0x1b, %%xmm6, %%xmm6 \n\t"
266 BUTTERFLY(xmm1, xmm6, 16(%2), xmm3)
267
268 "movaps 80(%4), %%xmm4 \n\t"
269 "movaps 32(%4), %%xmm5 \n\t"
270 "shufps $0x1b, %%xmm5, %%xmm5 \n\t"
271 BUTTERFLY(xmm4, xmm5, 32(%2), xmm3)
272
273 /* pass 2 */
274 BUTTERFLY(xmm0, xmm7, %%xmm2, xmm3)
275
276 "movaps 80(%2), %%xmm2 \n\t"
277 BUTTERFLY(xmm6, xmm5, %%xmm2, xmm3)
278
279 BUTTERFLY(xmm1, xmm4, %%xmm2, xmm3)
280
281 /* pass 3 */
282 "movaps 96(%2), %%xmm2 \n\t"
283 "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
284 BUTTERFLY(xmm0, xmm1, %%xmm2, xmm3)
285 "movaps %%xmm0, 112(%1) \n\t"
286 "movaps %%xmm1, 96(%1) \n\t"
287
288 "movaps 0(%1), %%xmm0 \n\t"
289 "shufps $0x1b, %%xmm5, %%xmm5 \n\t"
290 BUTTERFLY(xmm0, xmm5, %%xmm2, xmm3)
291
292 "movaps 48(%1), %%xmm1 \n\t"
293 "shufps $0x1b, %%xmm6, %%xmm6 \n\t"
294 BUTTERFLY(xmm1, xmm6, %%xmm2, xmm3)
295 "movaps %%xmm1, 48(%1) \n\t"
296
297 "shufps $0x1b, %%xmm4, %%xmm4 \n\t"
298 BUTTERFLY(xmm7, xmm4, %%xmm2, xmm3)
299
300 /* pass 4 */
301 "movaps (%3), %%xmm3 \n\t"
302 "movaps 112(%2), %%xmm2 \n\t"
303
304 BUTTERFLY2(xmm5, xmm3, xmm2, xmm1)
305
306 BUTTERFLY2(xmm0, xmm3, xmm2, xmm1)
307 "movaps %%xmm0, 16(%1) \n\t"
308
309 BUTTERFLY2(xmm6, xmm3, xmm2, xmm1)
310 "movaps %%xmm6, 32(%1) \n\t"
311
312 "movaps 48(%1), %%xmm0 \n\t"
313 BUTTERFLY2(xmm0, xmm3, xmm2, xmm1)
314 "movaps %%xmm0, 48(%1) \n\t"
315
316 BUTTERFLY2(xmm4, xmm3, xmm2, xmm1)
317
318 BUTTERFLY2(xmm7, xmm3, xmm2, xmm1)
319
320 "movaps 96(%1), %%xmm6 \n\t"
321 BUTTERFLY2(xmm6, xmm3, xmm2, xmm1)
322
323 "movaps 112(%1), %%xmm0 \n\t"
324 BUTTERFLY2(xmm0, xmm3, xmm2, xmm1)
325
326 /* pass 5 */
327 "movaps 128(%2), %%xmm2 \n\t"
328 "shufps $0xCC, %%xmm3,%%xmm3 \n\t"
329
330 BUTTERFLY3(xmm5, xmm3, xmm2, xmm1)
331 "movaps %%xmm5, (%1) \n\t"
332
333 "movaps 16(%1), %%xmm1 \n\t"
334 BUTTERFLY3(xmm1, xmm3, xmm2, xmm5)
335 "movaps %%xmm1, 16(%1) \n\t"
336
337 BUTTERFLY3(xmm4, xmm3, xmm2, xmm5)
338 "movaps %%xmm4, 64(%1) \n\t"
339
340 BUTTERFLY3(xmm7, xmm3, xmm2, xmm5)
341 "movaps %%xmm7, 80(%1) \n\t"
342
343 "movaps 32(%1), %%xmm5 \n\t"
344 BUTTERFLY3(xmm5, xmm3, xmm2, xmm7)
345 "movaps %%xmm5, 32(%1) \n\t"
346
347 "movaps 48(%1), %%xmm4 \n\t"
348 BUTTERFLY3(xmm4, xmm3, xmm2, xmm7)
349 "movaps %%xmm4, 48(%1) \n\t"
350
351 BUTTERFLY3(xmm6, xmm3, xmm2, xmm7)
352 "movaps %%xmm6, 96(%1) \n\t"
353
354 BUTTERFLY3(xmm0, xmm3, xmm2, xmm7)
355 "movaps %%xmm0, 112(%1) \n\t"
356
357
358 /* pass 6, no SIMD... */
359 "movss 56(%1), %%xmm3 \n\t"
360 "movl 4(%1), %0 \n\t"
361 "addss 60(%1), %%xmm3 \n\t"
362 "movss 72(%1), %%xmm7 \n\t"
363 "addss %%xmm3, %%xmm4 \n\t"
364 "movss 52(%1), %%xmm2 \n\t"
365 "addss %%xmm3, %%xmm2 \n\t"
366 "movss 24(%1), %%xmm3 \n\t"
367 "addss 28(%1), %%xmm3 \n\t"
368 "addss 76(%1), %%xmm7 \n\t"
369 "addss %%xmm3, %%xmm1 \n\t"
370 "addss %%xmm4, %%xmm5 \n\t"
371 "movss %%xmm1, 16(%1) \n\t"
372 "movss 20(%1), %%xmm1 \n\t"
373 "addss %%xmm3, %%xmm1 \n\t"
374 "movss 40(%1), %%xmm3 \n\t"
375 "movss %%xmm1, 48(%1) \n\t"
376 "addss 44(%1), %%xmm3 \n\t"
377 "movss 20(%1), %%xmm1 \n\t"
378 "addss %%xmm3, %%xmm4 \n\t"
379 "addss %%xmm2, %%xmm3 \n\t"
380 "addss 28(%1), %%xmm1 \n\t"
381 "movss %%xmm3, 40(%1) \n\t"
382 "addss 36(%1), %%xmm2 \n\t"
383 "movss 8(%1), %%xmm3 \n\t"
384 "movss %%xmm2, 56(%1) \n\t"
385 "addss 12(%1), %%xmm3 \n\t"
386 "movss %%xmm5, 8(%1) \n\t"
387 "movss %%xmm3, 32(%1) \n\t"
388 "movss 52(%1), %%xmm2 \n\t"
389 "movss 80(%1), %%xmm3 \n\t"
390 "movss 120(%1), %%xmm5 \n\t"
391 "movss %%xmm1, 80(%1) \n\t"
392 "movss %%xmm4, 24(%1) \n\t"
393 "addss 124(%1), %%xmm5 \n\t"
394 "movss 64(%1), %%xmm1 \n\t"
395 "addss 60(%1), %%xmm2 \n\t"
396 "addss %%xmm5, %%xmm0 \n\t"
397 "addss 116(%1), %%xmm5 \n\t"
398 "movl %0, 64(%1) \n\t"
399 "addss %%xmm0, %%xmm6 \n\t"
400 "addss %%xmm6, %%xmm1 \n\t"
401 "movl 12(%1), %0 \n\t"
402 "movss %%xmm1, 4(%1) \n\t"
403 "movss 88(%1), %%xmm1 \n\t"
404 "movl %0, 96(%1) \n\t"
405 "addss 92(%1), %%xmm1 \n\t"
406 "movss 104(%1), %%xmm4 \n\t"
407 "movl 28(%1), %0 \n\t"
408 "addss 108(%1), %%xmm4 \n\t"
409 "addss %%xmm4, %%xmm0 \n\t"
410 "addss %%xmm1, %%xmm3 \n\t"
411 "addss 84(%1), %%xmm1 \n\t"
412 "addss %%xmm5, %%xmm4 \n\t"
413 "addss %%xmm3, %%xmm6 \n\t"
414 "addss %%xmm0, %%xmm3 \n\t"
415 "addss %%xmm7, %%xmm0 \n\t"
416 "addss 100(%1), %%xmm5 \n\t"
417 "addss %%xmm4, %%xmm7 \n\t"
418 "movl %0, 112(%1) \n\t"
419 "movss %%xmm0, 28(%1) \n\t"
420 "movss 36(%1), %%xmm0 \n\t"
421 "movss %%xmm7, 36(%1) \n\t"
422 "addss %%xmm1, %%xmm4 \n\t"
423 "movss 116(%1), %%xmm7 \n\t"
424 "addss %%xmm2, %%xmm0 \n\t"
425 "addss 124(%1), %%xmm7 \n\t"
426 "movss %%xmm0, 72(%1) \n\t"
427 "movss 44(%1), %%xmm0 \n\t"
428 "movss %%xmm6, 12(%1) \n\t"
429 "movss %%xmm3, 20(%1) \n\t"
430 "addss %%xmm0, %%xmm2 \n\t"
431 "movss %%xmm4, 44(%1) \n\t"
432 "movss %%xmm2, 88(%1) \n\t"
433 "addss 60(%1), %%xmm0 \n\t"
434 "movl 60(%1), %0 \n\t"
435 "movl %0, 120(%1) \n\t"
436 "movss %%xmm0, 104(%1) \n\t"
437 "addss %%xmm5, %%xmm1 \n\t"
438 "addss 68(%1), %%xmm5 \n\t"
439 "movss %%xmm1, 52(%1) \n\t"
440 "movss %%xmm5, 60(%1) \n\t"
441 "movss 68(%1), %%xmm1 \n\t"
442 "movss 100(%1), %%xmm5 \n\t"
443 "addss %%xmm7, %%xmm5 \n\t"
444 "addss 108(%1), %%xmm7 \n\t"
445 "addss %%xmm5, %%xmm1 \n\t"
446 "movss 84(%1), %%xmm2 \n\t"
447 "addss 92(%1), %%xmm2 \n\t"
448 "addss %%xmm2, %%xmm5 \n\t"
449 "movss %%xmm1, 68(%1) \n\t"
450 "addss %%xmm7, %%xmm2 \n\t"
451 "movss 76(%1), %%xmm1 \n\t"
452 "movss %%xmm2, 84(%1) \n\t"
453 "movss %%xmm5, 76(%1) \n\t"
454 "movss 108(%1), %%xmm2 \n\t"
455 "addss %%xmm1, %%xmm7 \n\t"
456 "addss 124(%1), %%xmm2 \n\t"
457 "addss %%xmm2, %%xmm1 \n\t"
458 "addss 92(%1), %%xmm2 \n\t"
459 "movss %%xmm1, 100(%1) \n\t"
460 "movss %%xmm2, 108(%1) \n\t"
461 "movss 92(%1), %%xmm2 \n\t"
462 "movss %%xmm7, 92(%1) \n\t"
463 "addss 124(%1), %%xmm2 \n\t"
464 "movss %%xmm2, 116(%1) \n\t"
465 :"+&r"(tmp1)
466 :"r"(out), "r"(b1), "r"(smask), "r"(in)
467 :"memory"
468 );
469 }