summaryrefslogblamecommitdiff
path: root/kaldi_io/src/tools/ATLAS/include/contrib/SSE3Dnow.h
blob: a783749ebf96dd52efd26470363929bfdb77a173 (plain) (tree)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709




































































































































































































































































































































































































































































































































































































































































































































                                                                                                                          
#if !defined(ATL_GAS_x8632) &&  !defined(ATL_GAS_x8664)
   #error "This kernel requires gas x86 assembler!"
#endif
#ifndef Mstr   /* Added by RCW to make multiline macros work */
   #define Mstr2(m) # m
   #define Mstr(m) Mstr2(m)
#endif
/*  The mening of the defined macros is as follows:
 *  VECLEN:         The length of a singleprecision vector register
 *  vec_add:        Add to single precision vectors. 
 *  vec_mul:        Multiply to single precision vectors.
 *  vec_mov:        Moves data around
 *  vec_mov1:       Load one element in a vector and zero all other entries!
 *  vec_splat:      Load one element relpicated in all positions in the vector.
 *  vec_load_apart: Load elements from different memory positions into a register. 
 *  vec_sum:        Sums a register.
 *  vec_store_one:  Stores lowest element in vector to memory, no zero-extend!
 * Meaning of suffixes is as follows:
 *    mr means memory to register
 *    rr means register to register 
 *    rm means register to memory
 *    a means that instruction needs aligned data
 *    1 means that the instructions only operates on the lowest element of the
 *      vector.
 * 
 * The _1 instructions work under one important assumption: That you never mix
 * them with regular instructions, e.g. loading into a register with a normal
 * mov, and then using add_rr_1 will not work under 3dnow! since it is in
 * reality a normal add.  However, if using a mov_1 first, the upper part of
 * the register will be zeroed, and it will therefore work. The _1 system is
 * more robust under SSE, but other architectures might be implemented the
 * same way as 3dnow!
 *
 * RCW: I added the following functionality for SSE only (note that vw may
 *      be overwritten with intermediate results, but is not used as input,
 *      and that all input array may be overwritten wt intermediate results.
 *      VL : vector length -1):
 *   vec_red(vd, vw) : vd[0] = sum(vd[0:VL])
 *   vec_red2(v1, v2, vw) : v1[0] = sum(v1[0:VL]); v1[1] = sum(v2[0:VL])
 *   vec_red4(v0, v1, v2, v3 vw1, vw2) : 
 *      v0[0] = sum(v0[0:VL]); v0[1] = sum(v1[0:VL])
 *      if type = double:
 *         v2[0] = sum(v2[0:VL]); v2[1] = sum(v3[0:VL])
 *      else 
 *         v0[2] = sum(v2[0:VL]); v0[3] = sum(v3[0:VL])
 *    vec_zero(vd) : vd[0:VL] = 0.0
 */


/* Things to try:
 *    Non-temporal stores 
 *    Sequences of instructions instead of movups
 *
 *
 *
 *
 */



#define gen_vec_rr(op,reg1,reg2) \
        __asm__ __volatile__ (#op " " #reg1 ", " #reg2 \
                              :  /* nothing */ \
                              : /* nothing */)


#define w(p) p

#define nop()             __asm__ __volatile__ ("nop")

#define rep()             __asm__ __volatile__ ("rep")

#define align()           __asm__ __volatile__ (".align 16")


#ifdef x87double

#define st0 %%st(0)
#define st1 %%st(1)
#define st2 %%st(2)
#define st3 %%st(3)
#define st4 %%st(4)
#define st5 %%st(5)
#define st6 %%st(6)
#define st7 %%st(7)


#define gen_stack_rt(op,reg) \
        __asm__ __volatile__ (#op " " #reg \
                              :  /* nothing */ \
                              : /* nothing */)

#define gen_stack_tr(op,reg) \
        __asm__ __volatile__ (#op " %%st(0)," #reg \
                              :  \
                              : )


#define gen_stack_rr(op,reg1,reg2) \
        __asm__ __volatile__ (#op " " #reg1 ", " #reg2 \
                              :  /* nothing */ \
                              : /* nothing */)

#define gen_stack_t(op)  \
        __asm__ __volatile__ (#op \
                              :  /* nothing */ \
                              : /* nothing */)


#define gen_stack_tm(op,mem) \
        __asm__ __volatile__ (#op " %0" \
                              : "=m" (((mem)[0])) \
                              :  )

#define gen_stack_mt(op,mem) \
        __asm__ __volatile__ (#op " %0" \
                              :  \
                              : "m" (((mem)[0])))


#define stack_mov_mt_push(mem)  gen_stack_mt(fldl,mem)

#define stack_add_tr_pop(reg)   gen_stack_tr(faddp,reg)
#define stack_add_mt(mem)       gen_stack_mt(faddl,mem)

#define stack_mul_tr(reg)       gen_stack_tr(fmul,reg)
#define stack_mul_tr_pop(reg)   gen_stack_tr(fmulp,reg)
#define stack_mul_mt(mem)       gen_stack_mt(fmul,mem)

#define stack_mov_tm_pop(mem)   gen_stack_tm(fstpl,mem)

#define stack_zero_push()       gen_stack_t(fldz)

#endif /* x87double */

#ifdef SSE

/* Peculiarities of SSE: Alignment is good, but not mandatory. It is possible to
 * load/store from misaligned adresses using movups at a cost of some cycles. Loading
 * using mul/add must always be aligned. Alignment is 16 bytes.
 * No muladd.
 */



#define gen_vec_mr(op,mem,reg) \
        __asm__ __volatile__ (#op " %0, " #reg \
                              :  /* nothing */ \
                              : "m" (((mem)[0])), "m" (((mem)[1])), "m" (((mem)[2])), "m" (((mem)[3])))


#define gen_vec_rm(op,reg,mem) \
        __asm__ __volatile__ (#op " " #reg ", %0" \
                              : "=m" (((mem)[0])), "=m" (((mem)[1])), "=m" (((mem)[2])), "=m" (((mem)[3])) \
                              :  /* nothing */ )                          




#define VECLEN 4

#define reg0 %%xmm0
#define reg1 %%xmm1
#define reg2 %%xmm2
#define reg3 %%xmm3
#define reg4 %%xmm4
#define reg5 %%xmm5
#define reg6 %%xmm6
#define reg7 %%xmm7
#ifdef ATL_GAS_x8664
   #define reg8 %%xmm8
   #define reg9 %%xmm9
   #define reg10 %%xmm10
   #define reg11 %%xmm11
   #define reg12 %%xmm12
   #define reg13 %%xmm13
   #define reg14 %%xmm14
   #define reg15 %%xmm15
#endif

#define vec_mov_mr(mem,reg)     gen_vec_mr(movups,mem,reg)
#define vec_mov_rm(reg,mem)     gen_vec_rm(movups,reg,mem)
#define vec_mov_mr_a(mem,reg)   gen_vec_mr(movaps,mem,reg)
#define vec_mov_rm_a(reg,mem)   gen_vec_rm(movaps,reg,mem)
#define vec_mov_rr(reg1,reg2)   gen_vec_rr(movaps,reg1,reg2)

#define vec_add_mr_a(mem,reg)   gen_vec_mr(addps,mem,reg)
#define vec_mul_mr_a(mem,reg)   gen_vec_mr(mulps,mem,reg)

#define vec_add_rr(mem,reg)     gen_vec_rr(addps,mem,reg)
#define vec_mul_rr(mem,reg)     gen_vec_rr(mulps,mem,reg)

#define vec_mov_mr_1(mem,reg)   gen_vec_mr(movss,mem,reg)
#define vec_mov_rm_1(reg,mem)   gen_vec_rm(movss,reg,mem)
#define vec_mov_rr_1(reg1,reg2) gen_vec_rr(movss,reg1,reg2)

#define vec_add_mr_1(mem,reg)   gen_vec_mr(addss,mem,reg)
#define vec_add_rr_1(reg1,reg2) gen_vec_rr(addss,reg1,reg2)

#define vec_mul_mr_1(mem,reg)   gen_vec_mr(mulss,mem,reg)
#define vec_mul_rr_1(reg1,reg2) gen_vec_rr(mulss,reg1,reg2)

#define vec_unpack_low(reg1,reg2)  gen_vec_rr(unpcklps,reg1,reg2)
#define vec_unpack_high(reg1,reg2) gen_vec_rr(unpckhps,reg1,reg2)
#define vec_shuffle(mode,reg1,reg2) vec_shuffle_wrap(mode,reg1,reg2)
#define vec_shuffle_wrap(mode,reg1,reg2) \
        __asm__ __volatile__ ("shufps " #mode ", " #reg1 ", " #reg2 \
			      : /* nothing */\
			      : /* nothing */) 

/* Hack! */
/* To use this instruction be sure that register 7 is not in use!!! */
/* It must be possible to reduce this sequence to only four instructions.
 * please tell me how! */
#define vec_sum(reg) vec_sum_wrap(reg)
#define vec_sum_wrap(reg) \
        __asm__ __volatile__ ("movhlps " #reg ", %%xmm7\n"\
    			      "addps " #reg ", %%xmm7\n"\
    			      "movaps %%xmm7, " #reg "\n"\
                              "shufps $1, " #reg ", %%xmm7\n"\
    			      "addss %%xmm7, " #reg "\n"\
			      : /* nothing */\
			      : /* nothing */) 

/* RCW: added to safely replace vec_sum (vec reduce), and use SSE3 when avail */
#define vec_zero(vd) __asm__ __volatile__("xorps " Mstr(vd) ", " Mstr(vd) ::)
#ifdef ATL_SSE3
   #define vec_red(vr, vwrk) \
     __asm__ __volatile__("haddps " Mstr(vr) ", " Mstr(vr) "\n"\
                          "haddps " Mstr(vr) ", " Mstr(vr) "\n" ::)
/*
 * haddps v1 v0         # v0 = {v1cd, v1ab, v0cd, v0ab}
 * haddps v0 v0         # v0 = {v1abcd, v0abcd, v1abcd, v0abcd}
 */
   #define vec_red2(v0, v1, vwork) \
     __asm__ __volatile__("haddps " Mstr(v1) ", " Mstr(v0) "\n"\
                          "haddps " Mstr(v0) ", " Mstr(v0) "\n" ::)
/*
 * haddps   v1, v0     # v0 = {v1cd,v1ab,v0cd,v0ab}
 * haddps   v3, v2     # v2 = {v3cd,v3ab,v2cd,v2ab}
 * haddps   v2, v0     # v0 = {v3abcd,v2abcd,v1abcd, v0abcd}
 */
   #define vec_red4(v0, v1, v2, v3, w0, w1) \
     __asm__ __volatile__("haddps " Mstr(v1) ", " Mstr(v0) "\n"\
                          "haddps " Mstr(v3) ", " Mstr(v2) "\n"\
                          "haddps " Mstr(v2) ", " Mstr(v0) "\n" ::)
#elif defined(ATL_SSE2)
   #define vec_red(vr, vwrk) \
      __asm__ __volatile__ ("pshufd $0xEE, " Mstr(vr) ", " Mstr(vwrk) "\n"\
                            "addps " Mstr(vwrk) ", " Mstr(vr) "\n"\
                            "pshufd $0xE5, " Mstr(vr) ", " Mstr(vwrk) "\n"\
                            "addss " Mstr(vwrk) ", " Mstr(vr) "\n"\
                            ::)
#else
   #define vec_red(vr, vwrk) \
      __asm__ __volatile__ ("movhlps " Mstr(vr) ", " Mstr(vwrk) "\n"\
                            "addps " Mstr(vwrk) ", " Mstr(vr) "\n"\
                            "movaps " Mstr(vr) ", " Mstr(vwrk) "\n"\
                            "shufps $0xE5, " Mstr(vr) ", " Mstr(vr) "\n"\
                            "addss " Mstr(vwrk) ", " Mstr(vr) "\n"\
                            ::)
#endif
#ifndef ATL_SSE3  /* codes that are the same for SSE2 and SSE1 */
/*
                                # v0 = {v0d,v0c,v0b,v0a}
                                # v1 = {v1d,v1c,v1b,v1a}
       movaps   v0, vw          # vw = {v0d,v0c,v0b,v0a}
       unpacklps v1, v0         # v0 = {v1b,v0b,v1a,v0a}
       unpackhps v1, vw         # vw = {v1d,v0d,v1c,v0c}
       addps    vw, v0          # v0 = {v1bd,v0bd,v1ac,v0ac}
       movhlps  v0, vw          # vw = {X   ,   X,v1bd,v0bd}
       addps    vw, v0          # v0 = {X   ,   X,v1abcd,v0abcd}
*/
   #define vec_red2(v0, v1, vw) \
     __asm__ __volatile__ ("movaps " Mstr(v0) ", " Mstr(vw) "\n"\
                           "unpcklps " Mstr(v1) ", " Mstr(v0) "\n"\
                           "unpckhps " Mstr(v1) ", " Mstr(vw) "\n"\
                           "addps " Mstr(vw) ", " Mstr(v0) "\n"\
                           "movhlps " Mstr(v0) ", " Mstr(vw) "\n"\
                           "addps " Mstr(vw) ", " Mstr(v0) "\n"\
                           ::)
/* 
 * movaps  v0, w0       # w0 = {v0d, v0c, v0b, v0a}
 * unpcklps v1, v0      # v0 = {v1b, v0b, v1a, v0a}
 * movaps  v2, w1       # w1 = {v2d, v2c, v2b, v2a}
 * unpckhps v1, w0      # w0 = {v1d, v0d, v1c, v0c}
 * unpcklps v3, v2      # v2 = {v3b, v2b, v3a, v2a}
 * addps    w0, v0      # v0 = {v1bd, v0bd, v1ac, v0ac}
 * unpckhps v3, w1      # w1 = {v3d, v2d, v3c, v2c}
 * movaps   v0, w0      # w0 = {v1bd, v0bd, v1ac, v0ac}
 * addps    w1, v2      # v2 = {v3bd, v2bd, v3ac, v2ac}
 * shufps   $0x44,v2,v0 # v0 = {v3ac, v2ac, v1ac, v0ac}
 * shufps   $0xEE,v2,w0 # w0 = {v3bd, v2bd, v1bd, v0bd}
 * addps    w0, v0      # v0 = {v3abcd, v2abcd, v1abcd, v0abcd}
 */
   #define vec_red4(v0, v1, v2, v3, w0, w1) \
     __asm__ __volatile__ ("movaps " Mstr(v0) ", " Mstr(w0) "\n"\
                          "unpcklps " Mstr(v1) ", " Mstr(v0) "\n"\
                          "movaps " Mstr(v2) ", " Mstr(w1) "\n"\
                          "unpckhps " Mstr(v1) ", " Mstr(w0) "\n"\
                          "unpcklps " Mstr(v3) ", " Mstr(v2) "\n"\
                          "addps " Mstr(w0) ", " Mstr(v0) "\n"\
                          "unpckhps " Mstr(v3) ", " Mstr(w1) "\n"\
                          "movaps " Mstr(v0) ", " Mstr(w0) "\n"\
                          "addps " Mstr(w1) ", " Mstr(v2) "\n"\
                          "shufps $0x44, " Mstr(v2) ", " Mstr(v0) "\n"\
                          "shufps $0xEE, " Mstr(v2) ", " Mstr(w0) "\n"\
                          "addps " Mstr(w0) ", " Mstr(v0) "\n"\
                           ::)
#endif
     
#define vec_splat(mem,reg)      vec_splat_wrap(mem,reg)
#define vec_splat_wrap(mem,reg) \
        __asm__ __volatile__ ("movss %0, " #reg "\n"\
			      "unpcklps " #reg ", " #reg "\n"\
			      "movlhps " #reg ", " #reg "\n"\
			      : /* nothing */ \
                              : "m" ((mem)[0]))


/* This instruction sequence appears courtesy of Camm Maguire. */
#define vec_sum_full(reg0,reg1,reg2,reg3,regout,empty0,empty1) vec_sum_full_wrap(reg0,reg1,reg2,reg3,regout,empty0,empty1)
#define vec_sum_full_wrap(reg0,reg1,reg2,reg3,regout,empty0,empty1) \
      __asm__ __volatile__ ("movaps " #reg0 "," #empty0 "\n"\
			    "unpcklps " #reg1 "," #reg0 "\n"\
			    "movaps " #reg2 "," #empty1 "\n"\
			    "unpckhps " #reg1 "," #empty0 "\n"\
			    "unpcklps " #reg3 "," #reg2 "\n"\
			    "addps  " #empty0 "," #reg0 "\n"\
			    "unpckhps " #reg3 "," #empty1 "\n"\
			    "movaps " #reg0 "," #regout "\n"\
			    "addps  " #empty1 "," #reg2 "\n"\
			    "shufps $0x44," #reg2 "," #reg0 "\n"\
			    "shufps $0xee," #reg2 "," #regout "\n"\
			    "addps  " #reg0 "," #regout "\n"\
			    : /* nothing */  \
			    : /* nothing */)			    



typedef float vector[VECLEN];

#endif  /* end ifdef SSE */


#ifdef SSE2

/* Peculiarities of SSE: Alignment is good, but not mandatory. It is possible to
 * load/store from misaligned adresses using movups at a cost of some cycles. Loading
 * using mul/add must always be aligned. Alignment is 16 bytes.
 * No muladd.
 */



#define gen_vec_mr(op,mem,reg) \
        __asm__ __volatile__ (#op " %0, " #reg \
                              :  /* nothing */ \
                              : "m" (((mem)[0])), "m" (((mem)[1])))


#define gen_vec_rm(op,reg,mem) \
        __asm__ __volatile__ (#op " " #reg ", %0" \
                              : "=m" (((mem)[0])), "=m" (((mem)[1])) \
                              :  /* nothing */ )                          




#define VECLEN 2

#define reg0 %%xmm0
#define reg1 %%xmm1
#define reg2 %%xmm2
#define reg3 %%xmm3
#define reg4 %%xmm4
#define reg5 %%xmm5
#define reg6 %%xmm6
#define reg7 %%xmm7
#ifdef ATL_GAS_x8664
   #define reg8 %%xmm8
   #define reg9 %%xmm9
   #define reg10 %%xmm10
   #define reg11 %%xmm11
   #define reg12 %%xmm12
   #define reg13 %%xmm13
   #define reg14 %%xmm14
   #define reg15 %%xmm15
#endif


#define vec_mov_mr(mem,reg)     gen_vec_mr(movupd,mem,reg)
#define vec_mov_rm(reg,mem)     gen_vec_rm(movupd,reg,mem)
#define vec_mov_mr_a(mem,reg)   gen_vec_mr(movapd,mem,reg)
#define vec_mov_rm_a(reg,mem)   gen_vec_rm(movapd,reg,mem)
#define vec_mov_rr(reg1,reg2)   gen_vec_rr(movapd,reg1,reg2)

#define vec_add_mr_a(mem,reg)   gen_vec_mr(addpd,mem,reg)
#define vec_mul_mr_a(mem,reg)   gen_vec_mr(mulpd,mem,reg)

#define vec_add_rr(mem,reg)     gen_vec_rr(addpd,mem,reg)
#define vec_mul_rr(mem,reg)     gen_vec_rr(mulpd,mem,reg)

#define vec_mov_mr_1(mem,reg)   gen_vec_mr(movsd,mem,reg)
#define vec_mov_rm_1(reg,mem)   gen_vec_rm(movsd,reg,mem)
#define vec_mov_rr_1(reg1,reg2) gen_vec_rr(movsd,reg1,reg2)

#define vec_add_mr_1(mem,reg)   gen_vec_mr(addsd,mem,reg)
#define vec_add_rr_1(reg1,reg2) gen_vec_rr(addsd,reg1,reg2)

#define vec_mul_mr_1(mem,reg)   gen_vec_mr(mulsd,mem,reg)
#define vec_mul_rr_1(reg1,reg2) gen_vec_rr(mulsd,reg1,reg2)

#define vec_splat(mem,reg)      vec_splat_wrap(mem,reg)
#define vec_splat_wrap(mem,reg) \
        __asm__ __volatile__ ("movsd %0, " #reg "\n"\
			      "unpcklpd " #reg ", " #reg \
			      : /* nothing */ \
                              : "m" ((mem)[0]))

/* Hack! */
/* To use this instruction be sure that register 7 is not in use!!! */
#define vec_sum(reg) vec_sum_wrap(reg)
#define vec_sum_wrap(reg) \
        __asm__ __volatile__ ("movhlps " #reg ", %%xmm7\n"\
    			      "addpd %%xmm7, " #reg "\n"\
			      : /* nothing */\
			      : /* nothing */) 
/* 
 * Added by RCW to improve performance and avoid xmm7 hack (replace vec_sum)
 */
#define vec_zero(vd) __asm__ __volatile__("xorps " Mstr(vd) ", " Mstr(vd) ::)
#ifdef ATL_SSE3
   #define vec_red(vr, vwrk) \
      __asm__ __volatile__("haddpd " Mstr(vr) ", " Mstr(vr) "\n" ::)
   #define vec_red2(v0, v1, vw) \
      __asm__ __volatile__("haddpd " Mstr(v1) ", " Mstr(v0) "\n" ::)
   #define vec_red4(v0, v1, v2, v3, w0, w1) \
      __asm__ __volatile__("haddpd " Mstr(v1) ", " Mstr(v0) "\n"\
                           "haddpd " Mstr(v3) ", " Mstr(v2) "\n"\
                           ::)
#else
   #define vec_red(vr, vwrk) \
      __asm__ __volatile__ ("pshufd $0xEE, " Mstr(vr) ", " Mstr(vwrk) "\n"\
                            "addsd " Mstr(vwrk) ", " Mstr(vr) "\n" ::)
/*
 * movapd v0, vw        # vw = {v0b, v0a}
 * unpcklpd v1,v0      # v0 = {v1a, v0a}
 * unpckhpd v1, vw     # vw = {v1b, v0b}
 * addpd     vw, v0     # v0 = {v1ab,v0ab}
 */
   #define vec_red2(v0, v1, vw) \
      __asm__ __volatile__("movapd " Mstr(v0) ", " Mstr(vw) "\n"\
                           "unpcklpd " Mstr(v1) ", " Mstr(v0) "\n"\
                           "unpckhpd " Mstr(v1) ", " Mstr(vw) "\n"\
                           "addpd " Mstr(vw) ", " Mstr(v0) "\n"\
                           ::)
/*
 * movapd   v0, w0      # w0 = {v0b, v0a}
 * movapd   v2, w1      # w1 = {v2b, v2a}
 * unpcklpd v1, v0      # v0 = {v1a, v0a}
 * unpcklpd v3, v2      # v2 = {v3a, v2a}
 * unpckhpd v1, w0      # w0 = {v1b, v0b}
 * unpckhpd v3, w1      # w1 = {v3b, v2b}
 * addpd    w0, v0      # v0 = {v1ab, v0ab}
 * addpd    w1, v2      # v2 = {v3ab, v2ab}
 */
   #define vec_red4(v0, v1, v2, v3, w0, w1) \
      __asm__ __volatile__("movapd " Mstr(v0) ", " Mstr(w0) "\n"\
                           "movapd " Mstr(v2) ", " Mstr(w1) "\n"\
                           "unpcklpd " Mstr(v1) ", " Mstr(v0) "\n"\
                           "unpcklpd " Mstr(v3) ", " Mstr(v2) "\n"\
                           "unpckhpd " Mstr(v1) ", " Mstr(w0) "\n"\
                           "unpckhpd " Mstr(v3) ", " Mstr(w1) "\n"\
                           "addpd " Mstr(w0) ", " Mstr(v0) "\n"\
                           "addpd " Mstr(w1) ", " Mstr(v2) "\n"\
                           ::)
#endif

#define vec_sum_full(reg1,reg2,empty1) vec_sum_full_wrap(reg1,reg2,empty1)
#define vec_sum_full_wrap(reg1,reg2,empty1) \
        __asm__ __volatile__ ("movhlps " #reg2 ", " #empty1 "\n"\
                              "movlhps " #reg2 ", " #empty1 "\n"\
    			      "addpd " #empty1 ", " #reg1 "\n"\
			      : /* nothing */\
			      : /* nothing */) 

     
typedef double vector[VECLEN];

#endif  /* end ifdef SSE2 */


#ifdef THREEDNOW

/* Peculiarities of 3DNOW. Alignment is not an issue,
 * all alignments are legal, however alignment gives a speed increase.
 * The vec_acc instruction can be used to sum to registers at once more efficiently
 * than a series of vec_sum and vec_store_one
 * No muladd.
 */


#define gen_vec_mr(op,mem,reg) \
        __asm__ __volatile__ (#op " %0, " #reg \
                              :  /* nothing */ \
                              : "m" (((mem)[0])), "m" (((mem)[1])))

#define gen_vec_rm(op,reg,mem) \
        __asm__ __volatile__ (#op " " #reg ", %0" \
                              : "=m" (((mem)[0])), "=m" (((mem)[1])) \
			      :  /* nothing */ )                            




#define VECLEN 2

#define reg0 %%mm0
#define reg1 %%mm1
#define reg2 %%mm2
#define reg3 %%mm3
#define reg4 %%mm4
#define reg5 %%mm5
#define reg6 %%mm6
#define reg7 %%mm7

#define vec_add_mr(mem,reg)     gen_vec_mr(pfadd,mem,reg)
#define vec_mul_mr(mem,reg)     gen_vec_mr(pfmul,mem,reg)
#define vec_mov_mr(mem,reg)     gen_vec_mr(movq,mem,reg)
#define vec_mov_rm(reg,mem)     gen_vec_rm(movq,reg,mem)
#define vec_add_rr(reg1,reg2)   gen_vec_rr(pfadd,reg1,reg2)
#define vec_mul_rr(reg1,reg2)   gen_vec_rr(pfmul,reg1,reg2)
#define vec_acc_rr(reg1,reg2)   gen_vec_rr(pfacc,reg1,reg2)
#define vec_mov_rr(reg1,reg2)   gen_vec_rr(movq,reg1,reg2)

#define vec_sum(reg)            gen_vec_rr(pfacc,reg,reg)
#define vec_sum_full(reg1,reg2)  gen_vec_rr(pfacc,reg1,reg2)

#define vec_mov_mr_1(mem,reg)   gen_vec_mr(movd,mem,reg)
#define vec_mov_rm_1(reg,mem)   gen_vec_rm(movd,reg,mem)
#define vec_mov_rr_1(reg1,reg2) gen_vec_rr(movd,reg1,reg2)

#define vec_add_rr_1(reg1,reg2) gen_vec_rr(pfadd,reg1,reg2)
#define vec_mul_rr_1(reg1,reg2) gen_vec_rr(pfmul,reg1,reg2)


#define vec_splat(mem,reg)      vec_splat_wrap(mem,reg)
#define vec_splat_wrap(mem,reg) \
        __asm__ __volatile__ ("movd %0, " #reg "\n"\
			      "punpckldq " #reg ", " #reg \
			      : /* nothing */ \
                              : "m" ((mem)[0]))


#define vec_load_apart(mem1,mem2,reg) vec_load_apart_wrap(mem1,mem2,reg)
#define vec_load_apart_wrap(mem1,mem2,reg) \
        __asm__ __volatile__ ("movd %0, " #reg "\n"\
			      "punpckldq %1, " #reg \
			      : /* nothing */ \
                              : "m" ((mem1)[0]), "m" (((mem2)[0])))


#define vec_zero(reg)           gen_vec_rr(pxor,reg,reg)     

#define vec_enter()             __asm__ __volatile__ ("femms")
#define vec_exit()              __asm__ __volatile__ ("femms")

#define align()                 __asm__ __volatile__ (".align 16")


typedef float vector[VECLEN];

#endif





#ifdef ALTIVEC

#define VECLEN 4

#define reg0 %%vr0
#define reg1 %%vr1
#define reg2 %%vr2
#define reg3 %%vr3
#define reg4 %%vr4
#define reg5 %%vr5
#define reg6 %%vr6
#define reg7 %%vr7
#define reg8 %%vr8
#define reg9 %%vr9
#define reg10 %%vr10
#define reg11 %%vr11
#define reg12 %%vr12
#define reg13 %%vr13
#define reg14 %%vr14
#define reg15 %%vr15
#define reg16 %%vr16
#define reg17 %%vr17
#define reg18 %%vr18
#define reg19 %%vr19
#define reg20 %%vr20
#define reg21 %%vr21
#define reg22 %%vr22
#define reg23 %%vr23
#define reg24 %%vr24
#define reg25 %%vr25
#define reg26 %%vr26
#define reg27 %%vr27
#define reg28 %%vr28
#define reg29 %%vr29
#define reg30 %%vr30
#define reg31 %%vr31

#define gen_vec_mr(op,mem,reg) \
        __asm__ __volatile__ (#op " %0, " #reg \
                              :  /* nothing */ \
                              : "m" (((mem)[0])), "m" (((mem)[1])), "m" (((mem)[2])), "m" (((mem)[3])))


#define gen_vec_rm(op,reg,mem) \
        __asm__ __volatile__ (#op " " #reg ", %0" \
                              : "=m" (((mem)[0])), "=m" (((mem)[1])), "=m" (((mem)[2])), "=m" (((mem)[3])) \
                              :  /* nothing */ )                          


#define gen_alti3(op,reg1,reg2,regout) \
        __asm__ __volatile__ (#op " " #reg1 ", " #reg2 ", " #regout \
                              :  /* nothing */ \
                              : /* nothing */)

#define gen_alti_muladd(op,reg1,reg2,regout) \
        __asm__ __volatile__ (#op " " #reg1 ", " #reg2 ", " #regout ", " #regout \
                              :  /* nothing */ \
                              : /* nothing */)



#define vec_mov_mr_a(mem,reg) gen_vec_mr(lvx,mem,reg)
#define vec_mov_rm_a(reg,mem) gen_vec_rm(svx,reg,mem)
#define vec_muladd(reg1,reg2,regout) gen_alti3(vmaddfp,reg1,reg2,regout)

#define vec_zero(reg) gen_alti3(vxor,reg,reg,reg)


typedef float vector[VECLEN];

#endif


#ifdef ALTIVEC_C

/* These macros have been written by, or greatly inspired by,
 *  Nicholas A. Coult . Thanks.
 */

/* assumes that last four registers are not in use! */
#define transpose(x0,x1,x2,x3) \
reg28 = vec_mergeh(x0,x2); \
reg29 = vec_mergeh(x1,x3); \
reg30 = vec_mergel(x0,x2); \
reg31 = vec_mergel(x1,x3); \
x0 = vec_mergeh(reg28,reg29); \
x1 = vec_mergel(reg28,reg29); \
x2 = vec_mergeh(reg30,reg31); \
x3 = vec_mergel(reg30,reg31)

#define vec_mov_rm(v, where) \
low = vec_ld(0, (where)); \
high = vec_ld(16, (where)); \
p_vector = vec_lvsr(0, (int *)(where)); \
mask  = vec_perm((vector unsigned char)(0), (vector unsigned char)(-1), p_vector); \
v = vec_perm(v, v, p_vector); \
low = vec_sel(low,  v, mask); \
high = vec_sel(v, high, mask); \
vec_st(low,  0, (where)); \
vec_st(high, 16, (where))

#define vec_mov_mr_a(mem,reg)  reg = vec_ld(0, mem) 

#define vec_mov_mr(u,v) \
p_vector = (vector unsigned char)vec_lvsl(0, (int*)(v)); \
low = (vector unsigned char)vec_ld(0, (v)); \
high = (vector unsigned char)vec_ld(16, (v)); \
u=(vector float)vec_perm(low, high, p_vector)

#define vec_muladd(reg1,reg2,regout) regout = vec_madd(reg1,reg2,regout)
#define vec_add_rr(reg1,reg2)        reg2 = vec_add(reg1,reg2)

#define vec_zero(reg)                reg = vec_xor(reg,reg)

#define vec_sum_full(reg0,reg1,reg2,reg3,regout,empty0,empty1) \
transpose(reg0, reg1,reg2,reg3,regout,empty0,empty1); \
empty0 = vec_add(reg0,reg1); \
empty1 = vec_add(reg2,reg3); \
regout = vec_add(empty0,empty1)


#endif /* ALTIVEC_C */