Debian Codegraph @

libmp3lame/i386/fft3dn.nas

raw
1 ; from a new GOGO-no-coda (1999/09)
2 ; Copyright (C) 1999 shigeo
3 ; special thanks to Keiichi SAKAI, URURI
4 ; hacked and back-ported to LAME
5 ; by Takehiro TOMINAGA Nov 2000
6
7 %include "nasm.h"
8
9 globaldef fht_3DN
10
11 segment_data
12 align 16
13 costab dd 0x80000000, 0
14 dd 1.414213562,1.414213562
15 dd 9.238795283293805e-01, 9.238795283293805e-01
16 dd 3.826834424611044e-01, 3.826834424611044e-01
17 dd 9.951847264044178e-01, 9.951847264044178e-01
18 dd 9.801714304836734e-02, 9.801714304836734e-02
19 dd 9.996988186794428e-01, 9.996988186794428e-01
20 dd 2.454122920569705e-02, 2.454122920569705e-02
21 dd 9.999811752815535e-01, 9.999811752815535e-01
22 dd 6.135884819898878e-03, 6.135884819898878e-03
23 D_1_0_0_0 dd 0.0 , 1.0
24
25 segment_code
26
27 PIC_OFFSETTABLE
28
29
30 ;void fht_3DN(float *fz, int nn);
31
32 proc fht_3DN
33
34 pushd ebp, ebx, esi, edi
35
36 sub esp, 20
37
38 call get_pc.bp
39 add ebp, PIC_BASE()
40
41 mov r0, [esp+40] ;fi
42 mov r1, [esp+44] ;r1 = nn
43 lea r3, [PIC_EBP_REL(costab)] ;tri = costab
44 lea r4, [r0+r1*8] ;r4 = fn = &fz[n]
45 mov [esp+16], r4
46 mov r4, 8 ;kx = k1/2
47
48 pmov mm7, [r3]
49
50 loopalign 16
51 .do1
52 lea r3, [r3+16] ;tri += 2;
53 pmov mm6, [PIC_EBP_REL(costab+8)]
54 lea r2, [r4+r4*2] ;k3*fsize/2
55 mov r5, 4 ;i = 1*fsize
56
57 loopalign 16
58 .do2:
59 lea r1, [r0+r4] ;gi = fi + kx
60 ;f
61 pmov mm0, [r0] ;fi0
62 pmov mm1, [r0+r4*2] ;fi1
63 pmov mm2, [r0+r2*2] ;fi3
64 pmov mm3, [r0+r4*4] ;fi2
65
66 pupldq mm0, mm0 ;fi0 | fi0
67 pupldq mm1, mm1 ;fi1 | fi1
68 pupldq mm2, mm2 ;fi2 | fi2
69 pupldq mm3, mm3 ;fi3 | fi3
70
71 pxor mm1, mm7 ;fi1 | -fi1
72 pxor mm3, mm7 ;fi3 | -fi3
73
74 pfsub mm0, mm1 ;f1 | f0
75 pfsub mm2, mm3 ;f3 | f2
76
77 pmov mm4, mm0
78 pfadd mm0, mm2 ;f1+f3|f0+f2 = fi1 | fi0
79 pfsub mm4, mm2 ;f1-f3|f0-f2 = fi3 | fi2
80
81 pmovd [r0], mm0 ;fi[0]
82 puphdq mm0, mm0
83 pmovd [r0+r4*4], mm4 ;fi[k2]
84 puphdq mm4, mm4
85
86 pmovd [r0+r4*2], mm4 ;fi[k1]
87 pmovd [r0+r2*2], mm0 ;fi[k3]
88 lea r0, [r0+r4*8]
89
90 ;g
91 pmov mm0, [r1] ;gi0
92 pmov mm1, [r1+r4*2] ;gi1
93 pmov mm2, [r1+r4*4] ;gi2
94 pmov mm3, [r1+r2*2] ;gi3
95
96 pupldq mm1, mm1
97 pupldq mm0, mm0 ;gi0 | gi0
98 pupldq mm2, mm3 ;gi3 | gi2
99
100 pxor mm1, mm7 ;gi1 | -gi1
101
102 pfsub mm0, mm1 ;gi0-gi1|gi0+gi1 = g1 | g0
103 pfmul mm2, mm6 ;gi3*SQRT2|gi2*SQRT2 = g3 | g2
104
105 pmov mm4, mm0
106 pfadd mm0, mm2 ;g1+g3|g0+g2 = gi1 | gi0
107 pfsub mm4, mm2 ;g1-g3|g0-g2 = gi3 | gi2
108
109 pmovd [r1], mm0 ;gi[0]
110 puphdq mm0, mm0
111 pmovd [r1+r4*4], mm4 ;gi[k2]
112 puphdq mm4, mm4
113
114 cmp r0, [esp + 16]
115 pmovd [r1+r4*2], mm0 ;gi[k1]
116 pmovd [r1+r2*2], mm4 ;gi[k3]
117
118 jb near .do2
119
120 pmov mm6, [r3+r5] ; this is not aligned address!!
121
122 loopalign 16
123 .for:
124 ;
125 ; mm6 = c1 | s1
126 ; mm7 = 0x800000000 | 0
127 ;
128 pmov mm1, mm6
129 mov r0, [esp+40] ; fz
130 puphdq mm1, mm1 ; c1 | c1
131 lea r1, [r0+r4*2]
132 pfadd mm1, mm1 ; c1+c1 | c1+c1
133 pfmul mm1, mm6 ; 2*c1*c1 | 2*c1*s1
134 pfsub mm1, [PIC_EBP_REL(D_1_0_0_0)] ; 2*c1*c1-1.0 | 2*c1*s1 = -c2 | s2
135
136 pmov mm0, mm1
137 pxor mm7, mm6 ; c1 | -s1
138
139 pupldq mm2, mm0
140 pupldq mm3, mm6 ; ** | c1
141 puphdq mm0, mm2 ; s2 | c2
142 puphdq mm6, mm3 ;-s1 | c1
143
144 pxor mm0, [PIC_EBP_REL(costab)] ; c2 | -s2
145
146 ; mm0 = s2| c2
147 ; mm1 = -c2| s2
148 ; mm6 = c1| s1
149 ; mm7 = s1|-c1 (we use the opposite sign. from GOGO here)
150
151 pmov [esp], mm0
152 pmov [esp+8], mm1
153
154 sub r1, r5 ;r1 = gi
155 add r0, r5 ;r0 = fi
156
157 loopalign 16
158 .do3:
159 pmov mm2, [r0+r4*2] ; fi[k1]
160 pmov mm4, [r1+r4*2] ; gi[k1]
161 pmov mm3, [r0+r2*2] ; fi[k3]
162 pmov mm5, [r1+r2*2] ; gi[k3]
163
164 pupldq mm2, mm2 ; fi1 | fi1
165 pupldq mm4, mm4 ; gi1 | gi1
166 pupldq mm3, mm3 ; fi3 | fi3
167 pupldq mm5, mm5 ; gi3 | gi3
168
169 pfmul mm2, mm0 ; s2 * fi1 | c2 * fi1
170 pfmul mm4, mm1 ;-c2 * gi1 | s2 * gi1
171 pfmul mm3, mm0 ; s2 * fi3 | c2 * fi3
172 pfmul mm5, mm1 ;-c2 * gi3 | s2 * gi3
173
174 pfadd mm2, mm4 ;b | a
175 pfadd mm3, mm5 ;d | c
176
177 pmov mm0, [r0]
178 pmov mm4, [r1]
179 pmov mm1, [r0+r4*4]
180 pmov mm5, [r1+r4*4]
181
182 pupldq mm0, mm4 ;gi0 | fi0
183 pupldq mm1, mm5 ;gi2 | fi2
184
185 pmov mm4, mm2
186 pmov mm5, mm3
187
188 pfadd mm2, mm0 ;g0 | f0
189 pfadd mm3, mm1 ;g2 | f2
190
191 pfsub mm0, mm4 ;g1 | f1
192 pfsub mm1, mm5 ;g3 | f3
193
194 pmov mm4, mm3
195 pmov mm5, mm1
196
197 pupldq mm4, mm4 ;f2 | f2
198 puphdq mm5, mm5 ;g3 | g3
199 puphdq mm3, mm3 ;g2 | g2
200 pupldq mm1, mm1 ;f3 | f3
201
202 pfmul mm4, mm6 ;f2 * c1 | f2 * s1
203 pfmul mm5, mm7 ;g3 * s1 | g3 *-c1
204 pfmul mm3, mm6 ;g2 * c1 | g2 * s1
205 pfmul mm1, mm7 ;f3 * s1 | f3 *-c1
206
207 pfadd mm4, mm5 ;a | b
208 pfsub mm3, mm1 ;d | c
209
210 pmov mm5, mm2
211 pmov mm1, mm0
212
213 pupldq mm2, mm2 ;f0 | f0
214 pupldq mm0, mm0 ;f1 | f1
215
216 puphdq mm1, mm2 ;f0 | g1
217 puphdq mm5, mm0 ;f1 | g0
218
219 pmov mm2, mm4
220 pmov mm0, mm3
221
222 pfadd mm4, mm1 ;fi0 | gi1
223 pfadd mm3, mm5 ;fi1 | gi0
224 pfsub mm1, mm2 ;fi2 | gi3
225 pfsub mm5, mm0 ;fi3 | gi2
226
227 pmovd [r1+r4*2], mm4 ;gi[k1]
228 puphdq mm4, mm4
229 pmovd [r1], mm3 ;gi[0]
230 puphdq mm3, mm3
231 pmovd [r1+r2*2], mm1 ;gi[k3]
232 puphdq mm1, mm1
233 pmovd [r1+r4*4], mm5 ;gi[k2]
234 puphdq mm5, mm5
235
236 pmovd [r0], mm4 ;fi[0]
237 pmovd [r0+r4*2], mm3 ;fi[k1]
238 pmovd [r0+r4*4], mm1 ;fi[k2]
239 pmovd [r0+r2*2], mm5 ;fi[k3]
240
241 lea r0, [r0+r4*8]
242 lea r1, [r1+r4*8]
243 cmp r0, [esp + 16]
244 pmov mm0, [esp]
245 pmov mm1, [esp+8]
246
247 jb near .do3
248
249 add r5, 4
250 ; mm6 = c1| s1
251 ; mm7 = s1|-c1 (we use the opposite sign. from GOGO here)
252 pfmul mm6, [r3] ; c1*a | s1*a
253 pfmul mm7, [r3+8] ; s1*b |-c1*b
254 cmp r5, r4
255
256 pfsub mm6, mm7 ; c1*a-s1*b | s1*a+c1*b
257 pupldq mm7,mm6
258 puphdq mm6,mm7
259 pmov mm7, [PIC_EBP_REL(costab)]
260 jb near .for
261
262 mov r0, [esp+40] ;fi
263 cmp r4, [esp+40+4]
264 lea r4, [r4*4] ;kx *= 4
265
266 jb near .do1
267 .exitttt
268 femms
269 add esp,20
270 popd ebp, ebx, esi, edi
271 endproc
272
273
274 ;void fht_E3DN(float *fz, int nn);
275
276 proc fht_E3DN
277
278 pushd ebp, ebx, esi, edi
279
280 sub esp, 20
281
282 call get_pc.bp
283 add ebp, PIC_BASE()
284
285 mov r0, [esp+40] ;fi
286 mov r1, [esp+44] ;r1 = nn
287 lea r3, [PIC_EBP_REL(costab)] ;tri = costab
288 lea r4, [r0+r1*8] ;r4 = fn = &fz[n]
289 mov [esp+16], r4
290 mov r4, 8 ;kx = k1/2
291
292 pmov mm7, [r3]
293
294 loopalign 16
295 .do1
296 lea r3, [r3+16] ;tri += 2;
297 pmov mm6, [PIC_EBP_REL(costab+8)]
298 lea r2, [r4+r4*2] ;k3*fsize/2
299 mov r5, 4 ;i = 1*fsize
300
301 loopalign 16
302 .do2:
303 lea r1, [r0+r4] ;gi = fi + kx
304 ;f
305 pmov mm0, [r0] ; X | fi0
306 pmov mm1, [r0+r4*4] ; X | fi2
307 pupldq mm0, [r0+r4*2] ;fi1 | fi0
308 pupldq mm1, [r0+r2*2] ;fi3 | fi2
309 pfpnacc mm0, mm0 ;fi0+fi1 | fi0-fi1 = f0|f1
310 pfpnacc mm1, mm1 ;fi2+fi3 | fi2-fi3 = f2|f3
311
312 pmov mm2, mm0
313 pfadd mm0, mm1 ;f0+f2|f1+f3 = fi0 | fi1
314 pfsub mm2, mm1 ;f0-f2|f1-f3 = fi2 | fi3
315
316 pmovd [r0+r4*2], mm0 ;fi[k1]
317 pmovd [r0+r2*2], mm2 ;fi[k3]
318
319 puphdq mm0, mm0
320 puphdq mm2, mm2
321 pmovd [r0], mm0 ;fi[0]
322 pmovd [r0+r4*4], mm2 ;fi[k2]
323
324 lea r0, [r0+r4*8]
325 ;g
326 pmov mm3, [r1] ; gi0
327 pmov mm4, [r1+r2*2] ; gi3
328 pupldq mm3, [r1+r4*2] ;gi1|gi0
329 pupldq mm4, [r1+r4*4] ;gi2|gi3
330
331 pfpnacc mm3, mm3 ;gi0+gi1 |gi0-gi1 = f0|f1
332 pfmul mm4, mm6 ;gi2*SQRT2|gi3*SQRT2 = f2|f3
333
334 pmov mm5, mm3
335 pfadd mm3, mm4 ;f0+f2|f1+f3
336 pfsub mm5, mm4 ;f0-f2|f1-f3
337
338 cmp r0, [esp + 16]
339 pmovd [r1+r4*2], mm3 ;gi[k1]
340 pmovd [r1+r2*2], mm5 ;gi[k3]
341 puphdq mm3, mm3
342 puphdq mm5, mm5
343 pmovd [r1], mm3 ;gi[0]
344 pmovd [r1+r4*4], mm5 ;gi[k2]
345
346 jb near .do2
347
348 pmov mm6, [r3+r5] ; this is not aligned address!!
349
350 loopalign 16
351 .for:
352 ;
353 ; mm6 = c1 | s1
354 ; mm7 = 0x800000000 | 0
355 ;
356 pmov mm5, mm6
357 mov r0, [esp+40] ; fz
358 puphdq mm5, mm5 ; c1 | c1
359 lea r1, [r0+r4*2]
360 pfadd mm5, mm5 ; c1+c1 | c1+c1
361 pfmul mm5, mm6 ; 2*c1*c1 | 2*c1*s1
362 pfsub mm5, [PIC_EBP_REL(D_1_0_0_0)] ; 2*c1*c1-1.0 | 2*c1*s1 = -c2 | s2
363
364 pswapd mm4, mm5 ; s2 |-c2
365 pxor mm4, mm7 ; s2 | c2
366 pxor mm7, mm6 ; c1 |-s1
367 pswapd mm6, mm6 ; s1 | c1
368
369 ; mm4 = s2| c2
370 ; mm5 = -c2| s2
371 ; mm6 = c1| s1
372 ; mm7 = s1|-c1 (we use the opposite sign. from GOGO here)
373
374 pmov [esp], mm4
375 pmov [esp+8], mm5
376
377 sub r1, r5 ;r1 = gi
378 add r0, r5 ;r0 = fi
379
380 loopalign 16
381 .do3:
382 pmov mm0, [r0+r2*2] ; fi[k1]
383 pmov mm2, [r1+r2*2] ; gi[k1]
384 pmov mm1, [r0+r4*2] ; fi[k3]
385 pmov mm3, [r1+r4*2] ; gi[k3]
386
387 pupldq mm0, mm0
388 pupldq mm2, mm2
389 pupldq mm1, mm1
390 pupldq mm3, mm3
391
392 pfmul mm0, mm4
393 pfmul mm2, mm5
394 pfmul mm1, mm4
395 pfmul mm3, mm5
396
397 pfadd mm0, mm2 ;d | c
398 pfadd mm1, mm3 ;b | a
399
400 pmov mm2, [r0+r4*4] ;fi2
401 pupldq mm3, [r1+r4*4] ;gi2 | -
402 pmov mm4, [r0] ;fi0
403 pupldq mm5, [r1] ;gi0 | -
404
405 pupldq mm2, mm0 ;c | fi2
406 puphdq mm3, mm0 ;d | gi2
407 pupldq mm4, mm1 ;a | fi0
408 puphdq mm5, mm1 ;b | gi0
409
410 pfpnacc mm2, mm2 ;f2 | f3
411 pfpnacc mm3, mm3 ;g2 | g3
412 pfpnacc mm4, mm4 ;f0 | f1
413 pfpnacc mm5, mm5 ;g0 | g1
414
415 pmov mm0, mm2
416 pmov mm1, mm3
417 pupldq mm2, mm2 ;f3 | f3
418 pupldq mm3, mm3 ;g3 | g3
419 puphdq mm0, mm0 ;f2 | f2
420 puphdq mm1, mm1 ;g2 | g2
421
422 pswapd mm4, mm4 ;f1 | f0
423 pswapd mm5, mm5 ;g1 | g0
424
425 pfmul mm0, mm7 ;f2 * s1 | f2 *-c1
426 pfmul mm3, mm6 ;g3 * c1 | g3 * s1
427 pfmul mm1, mm6 ;g2 * c1 | g2 * s1
428 pfmul mm2, mm7 ;f3 * s1 | f3 *-c1
429
430 pfsub mm0, mm3 ; b |-a
431 pfsub mm1, mm2 ; d | c
432
433 pmov mm2, mm5
434 pmov mm3, mm4
435 pupldq mm4, mm0 ;-a | f0
436 pupldq mm5, mm1 ; c | g0
437 puphdq mm2, mm0 ; b | g1
438 puphdq mm3, mm1 ; d | f1
439
440 pfpnacc mm4, mm4 ;fi2 | fi0
441 pfpnacc mm5, mm5 ;gi0 | gi2
442 pfpnacc mm2, mm2 ;gi1 | gi3
443 pfpnacc mm3, mm3 ;fi1 | fi3
444
445 pmovd [r0], mm4 ;fi[0]
446 pmovd [r1+r4*4], mm5 ;gi[k2]
447 pmovd [r1+r2*2], mm2 ;gi[k3]
448 pmovd [r0+r2*2], mm3 ;fi[k3]
449
450 puphdq mm4, mm4
451 puphdq mm5, mm5
452 puphdq mm2, mm2
453 puphdq mm3, mm3
454 pmovd [r0+r4*4], mm4 ;fi[k2]
455 pmovd [r1], mm5 ;gi[0]
456 pmovd [r1+r4*2], mm2 ;gi[k1]
457 pmovd [r0+r4*2], mm3 ;fi[k1]
458
459 lea r0, [r0+r4*8]
460 lea r1, [r1+r4*8]
461 cmp r0, [esp + 16]
462 pmov mm4, [esp]
463 pmov mm5, [esp+8]
464
465 jb near .do3
466
467 add r5, 4
468 ; mm6 = c1| s1
469 ; mm7 = s1|-c1 (we use the opposite sign. from GOGO here)
470 pfmul mm6, [r3] ; c1*a | s1*a
471 pfmul mm7, [r3+8] ; s1*b |-c1*b
472 cmp r5, r4
473
474 pfsub mm6, mm7 ; c1*a-s1*b | s1*a+c1*b
475 pswapd mm6, mm6 ; ??? ; s1*a+c1*b | c1*a-s1*b
476 pmov mm7, [PIC_EBP_REL(costab)]
477 jb near .for
478
479 mov r0, [esp+40] ;fi
480 cmp r4, [esp+40+4]
481 lea r4, [r4*4] ;kx *= 4
482
483 jb near .do1
484 .exitttt
485 femms
486 add esp,20
487 popd ebp, ebx, esi, edi
488 endproc