aboutsummaryrefslogtreecommitdiffstats
path: root/src/bin/e_mmx.h
blob: 3569fd067aaeb3f106bd87927ad91eafb292354a (plain) (blame)
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
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
/*	mmx.h

	MultiMedia eXtensions GCC interface library for IA32.

	To use this library, simply include this header file
	and compile with GCC.  You MUST have inlining enabled
	in order for mmx_ok() to work; this can be done by
	simply using -O on the GCC command line.

	Compiling with -DMMX_TRACE will cause detailed trace
	output to be sent to stderr for each mmx operation.
	This adds lots of code, and obviously slows execution to
	a crawl, but can be very useful for debugging.

	THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY
	EXPRESS OR IMPLIED WARRANTIES, INCLUDING, WITHOUT
	LIMITATION, THE IMPLIED WARRANTIES OF MERCHANTABILITY
	AND FITNESS FOR ANY PARTICULAR PURPOSE.

	1997-98 by H. Dietz and R. Fisher

 History:
	97-98*	R.Fisher	Early versions
	980501	R.Fisher	Original Release
	980611*	H.Dietz		Rewrite, correctly implementing inlines, and
		R.Fisher	 including direct register accesses.
	980616	R.Fisher	Release of 980611 as 980616.
	980714	R.Fisher	Minor corrections to Makefile, etc.
	980715	R.Fisher	mmx_ok() now prevents optimizer from using
				 clobbered values.
				mmx_ok() now checks if cpuid instruction is
				 available before trying to use it.
	980726*	R.Fisher	mm_support() searches for AMD 3DNow, Cyrix
				 Extended MMX, and standard MMX.  It returns a
				 value which is positive if any of these are
				 supported, and can be masked with constants to
				 see which.  mmx_ok() is now a call to this
	980726*	R.Fisher	Added i2r support for shift functions
	980919	R.Fisher	Fixed AMD extended feature recognition bug.
	980921	R.Fisher	Added definition/check for _MMX_H.
				Added "float s[2]" to mmx_t for use with
				  3DNow and EMMX.  So same mmx_t can be used.
	981013	R.Fisher	Fixed cpuid function 1 bug (looked at wrong reg)
				Fixed psllq_i2r error in mmxtest.c

	* Unreleased (internal or interim) versions

 Notes:
	It appears that the latest gas has the pand problem fixed, therefore
	  I'll undefine BROKEN_PAND by default.
	String compares may be quicker than the multiple test/jumps in vendor
	  test sequence in mmx_ok(), but I'm not concerned with that right now.

 Acknowledgments:
	Jussi Laako for pointing out the errors ultimately found to be
	  connected to the failure to notify the optimizer of clobbered values.
	Roger Hardiman for reminding us that CPUID isn't everywhere, and that
	  someone may actually try to use this on a machine without CPUID.
	  Also for suggesting code for checking this.
	Robert Dale for pointing out the AMD recognition bug.
	Jimmy Mayfield and Carl Witty for pointing out the Intel recognition
	  bug.
	Carl Witty for pointing out the psllq_i2r test bug.
*/

#ifndef _MMX_H
#define _MMX_H

/*	Warning:  at this writing, the version of GAS packaged
	with most Linux distributions does not handle the
	parallel AND operation mnemonic correctly.  If the
	symbol BROKEN_PAND is defined, a slower alternative
	coding will be used.  If execution of mmxtest results
	in an illegal instruction fault, define this symbol.
*/
#undef	BROKEN_PAND


/*	The type of an value that fits in an MMX register
	(note that long long constant values MUST be suffixed
	 by LL and unsigned long long values by ULL, lest
	 they be truncated by the compiler)
*/
typedef	union {
	long long		q;	/* Quadword (64-bit) value */
	unsigned long long	uq;	/* Unsigned Quadword */
	int			d[2];	/* 2 Doubleword (32-bit) values */
	unsigned int		ud[2];	/* 2 Unsigned Doubleword */
	short			w[4];	/* 4 Word (16-bit) values */
	unsigned short		uw[4];	/* 4 Unsigned Word */
	char			b[8];	/* 8 Byte (8-bit) values */
	unsigned char		ub[8];	/* 8 Unsigned Byte */
	float			s[2];	/* Single-precision (32-bit) value */
}  __attribute__ ((aligned (8))) mmx_t;

/*	Helper functions for the instruction macros that follow...
	(note that memory-to-register, m2r, instructions are nearly
	 as efficient as register-to-register, r2r, instructions;
	 however, memory-to-memory instructions are really simulated
	 as a convenience, and are only 1/3 as efficient)
*/

/*	These macros are a lot simpler without the tracing...
*/

#define	mmx_i2r(op, imm, reg) \
	__asm__ __volatile__ (#op " $" #imm ", %%" #reg \
			      : /* nothing */ \
			      : /* nothing */);

#define	mmx_m2r(op, mem, reg) \
	__asm__ __volatile__ (#op " %0, %%" #reg \
			      : /* nothing */ \
			      : "m" (mem))

#define	mmx_r2m(op, reg, mem) \
	__asm__ __volatile__ (#op " %%" #reg ", %0" \
			      : "=m" (mem) \
			      : /* nothing */ )

#define	mmx_a2r(op, mem, reg) \
	__asm__ __volatile__ (#op " %0, %%" #reg \
			      : /* nothing */ \
			      : "m" (mem))

#define	mmx_r2a(op, reg, mem) \
	__asm__ __volatile__ (#op " %%" #reg ", %0" \
			      : "=m" (mem) \
			      : /* nothing */ )

#define	mmx_r2r(op, regs, regd) \
	__asm__ __volatile__ (#op " %" #regs ", %" #regd)

#define	mmx_m2m(op, mems, memd) \
	__asm__ __volatile__ ("movq %0, %%mm0\n\t" \
			      #op " %1, %%mm0\n\t" \
			      "movq %%mm0, %0" \
			      : "=X" (memd) \
			      : "X" (mems))

/*	1x64 MOVE Quadword
	(this is both a load and a store...
	 in fact, it is the only way to store)
*/
#define	movq_m2r(var, reg)	mmx_m2r(movq, var, reg)
#define	movq_r2m(reg, var)	mmx_r2m(movq, reg, var)
#define	movq_r2r(regs, regd)	mmx_r2r(movq, regs, regd)
#define	movq(vars, vard) \
	__asm__ __volatile__ ("movq %1, %%mm0\n\t" \
			      "movq %%mm0, %0" \
			      : "=X" (vard) \
			      : "X" (vars))
#define	movntq_r2m(reg, var)   mmx_r2m(movntq, reg, var)


/*	1x32 MOVE Doubleword
	(like movq, this is both load and store...
	 but is most useful for moving things between
	 mmx registers and ordinary registers)
*/
#define	movd_m2r(var, reg)	mmx_a2r(movd, var, reg)
#define	movd_r2m(reg, var)	mmx_r2a(movd, reg, var)
#define	movd_r2r(regs, regd)	mmx_r2r(movd, regs, regd)
#define	movd(vars, vard) \
	__asm__ __volatile__ ("movd %1, %%mm0\n\t" \
			      "movd %%mm0, %0" \
			      : "=X" (vard) \
			      : "X" (vars))


/*	2x32, 4x16, and 8x8 Parallel ADDs
*/
#define	paddd_m2r(var, reg)	mmx_m2r(paddd, var, reg)
#define	paddd_r2r(regs, regd)	mmx_r2r(paddd, regs, regd)
#define	paddd(vars, vard)	mmx_m2m(paddd, vars, vard)

#define	paddw_m2r(var, reg)	mmx_m2r(paddw, var, reg)
#define	paddw_r2r(regs, regd)	mmx_r2r(paddw, regs, regd)
#define	paddw(vars, vard)	mmx_m2m(paddw, vars, vard)

#define	paddb_m2r(var, reg)	mmx_m2r(paddb, var, reg)
#define	paddb_r2r(regs, regd)	mmx_r2r(paddb, regs, regd)
#define	paddb(vars, vard)	mmx_m2m(paddb, vars, vard)


/*	4x16 and 8x8 Parallel ADDs using Saturation arithmetic
*/
#define	paddsw_m2r(var, reg)	mmx_m2r(paddsw, var, reg)
#define	paddsw_r2r(regs, regd)	mmx_r2r(paddsw, regs, regd)
#define	paddsw(vars, vard)	mmx_m2m(paddsw, vars, vard)

#define	paddsb_m2r(var, reg)	mmx_m2r(paddsb, var, reg)
#define	paddsb_r2r(regs, regd)	mmx_r2r(paddsb, regs, regd)
#define	paddsb(vars, vard)	mmx_m2m(paddsb, vars, vard)


/*	4x16 and 8x8 Parallel ADDs using Unsigned Saturation arithmetic
*/
#define	paddusw_m2r(var, reg)	mmx_m2r(paddusw, var, reg)
#define	paddusw_r2r(regs, regd)	mmx_r2r(paddusw, regs, regd)
#define	paddusw(vars, vard)	mmx_m2m(paddusw, vars, vard)

#define	paddusb_m2r(var, reg)	mmx_m2r(paddusb, var, reg)
#define	paddusb_r2r(regs, regd)	mmx_r2r(paddusb, regs, regd)
#define	paddusb(vars, vard)	mmx_m2m(paddusb, vars, vard)


/*	2x32, 4x16, and 8x8 Parallel SUBs
*/
#define	psubd_m2r(var, reg)	mmx_m2r(psubd, var, reg)
#define	psubd_r2r(regs, regd)	mmx_r2r(psubd, regs, regd)
#define	psubd(vars, vard)	mmx_m2m(psubd, vars, vard)

#define	psubw_m2r(var, reg)	mmx_m2r(psubw, var, reg)
#define	psubw_r2r(regs, regd)	mmx_r2r(psubw, regs, regd)
#define	psubw(vars, vard)	mmx_m2m(psubw, vars, vard)

#define	psubb_m2r(var, reg)	mmx_m2r(psubb, var, reg)
#define	psubb_r2r(regs, regd)	mmx_r2r(psubb, regs, regd)
#define	psubb(vars, vard)	mmx_m2m(psubb, vars, vard)


/*	4x16 and 8x8 Parallel SUBs using Saturation arithmetic
*/
#define	psubsw_m2r(var, reg)	mmx_m2r(psubsw, var, reg)
#define	psubsw_r2r(regs, regd)	mmx_r2r(psubsw, regs, regd)
#define	psubsw(vars, vard)	mmx_m2m(psubsw, vars, vard)

#define	psubsb_m2r(var, reg)	mmx_m2r(psubsb, var, reg)
#define	psubsb_r2r(regs, regd)	mmx_r2r(psubsb, regs, regd)
#define	psubsb(vars, vard)	mmx_m2m(psubsb, vars, vard)


/*	4x16 and 8x8 Parallel SUBs using Unsigned Saturation arithmetic
*/
#define	psubusw_m2r(var, reg)	mmx_m2r(psubusw, var, reg)
#define	psubusw_r2r(regs, regd)	mmx_r2r(psubusw, regs, regd)
#define	psubusw(vars, vard)	mmx_m2m(psubusw, vars, vard)

#define	psubusb_m2r(var, reg)	mmx_m2r(psubusb, var, reg)
#define	psubusb_r2r(regs, regd)	mmx_r2r(psubusb, regs, regd)
#define	psubusb(vars, vard)	mmx_m2m(psubusb, vars, vard)


/*	4x16 Parallel MULs giving Low 4x16 portions of results
*/
#define	pmullw_m2r(var, reg)	mmx_m2r(pmullw, var, reg)
#define	pmullw_r2r(regs, regd)	mmx_r2r(pmullw, regs, regd)
#define	pmullw(vars, vard)	mmx_m2m(pmullw, vars, vard)


/*	4x16 Parallel MULs giving High 4x16 portions of results
*/
#define	pmulhw_m2r(var, reg)	mmx_m2r(pmulhw, var, reg)
#define	pmulhw_r2r(regs, regd)	mmx_r2r(pmulhw, regs, regd)
#define	pmulhw(vars, vard)	mmx_m2m(pmulhw, vars, vard)


/*	4x16->2x32 Parallel Mul-ADD
	(muls like pmullw, then adds adjacent 16-bit fields
	 in the multiply result to make the final 2x32 result)
*/
#define	pmaddwd_m2r(var, reg)	mmx_m2r(pmaddwd, var, reg)
#define	pmaddwd_r2r(regs, regd)	mmx_r2r(pmaddwd, regs, regd)
#define	pmaddwd(vars, vard)	mmx_m2m(pmaddwd, vars, vard)


/*	1x64 bitwise AND
*/
#ifdef	BROKEN_PAND
#define	pand_m2r(var, reg) \
	{ \
		mmx_m2r(pandn, (mmx_t) -1LL, reg); \
		mmx_m2r(pandn, var, reg); \
	}
#define	pand_r2r(regs, regd) \
	{ \
		mmx_m2r(pandn, (mmx_t) -1LL, regd); \
		mmx_r2r(pandn, regs, regd) \
	}
#define	pand(vars, vard) \
	{ \
		movq_m2r(vard, mm0); \
		mmx_m2r(pandn, (mmx_t) -1LL, mm0); \
		mmx_m2r(pandn, vars, mm0); \
		movq_r2m(mm0, vard); \
	}
#else
#define	pand_m2r(var, reg)	mmx_m2r(pand, var, reg)
#define	pand_r2r(regs, regd)	mmx_r2r(pand, regs, regd)
#define	pand(vars, vard)	mmx_m2m(pand, vars, vard)
#endif


/*	1x64 bitwise AND with Not the destination
*/
#define	pandn_m2r(var, reg)	mmx_m2r(pandn, var, reg)
#define	pandn_r2r(regs, regd)	mmx_r2r(pandn, regs, regd)
#define	pandn(vars, vard)	mmx_m2m(pandn, vars, vard)


/*	1x64 bitwise OR
*/
#define	por_m2r(var, reg)	mmx_m2r(por, var, reg)
#define	por_r2r(regs, regd)	mmx_r2r(por, regs, regd)
#define	por(vars, vard)	mmx_m2m(por, vars, vard)


/*	1x64 bitwise eXclusive OR
*/
#define	pxor_m2r(var, reg)	mmx_m2r(pxor, var, reg)
#define	pxor_r2r(regs, regd)	mmx_r2r(pxor, regs, regd)
#define	pxor(vars, vard)	mmx_m2m(pxor, vars, vard)


/*	2x32, 4x16, and 8x8 Parallel CoMPare for EQuality
	(resulting fields are either 0 or -1)
*/
#define	pcmpeqd_m2r(var, reg)	mmx_m2r(pcmpeqd, var, reg)
#define	pcmpeqd_r2r(regs, regd)	mmx_r2r(pcmpeqd, regs, regd)
#define	pcmpeqd(vars, vard)	mmx_m2m(pcmpeqd, vars, vard)

#define	pcmpeqw_m2r(var, reg)	mmx_m2r(pcmpeqw, var, reg)
#define	pcmpeqw_r2r(regs, regd)	mmx_r2r(pcmpeqw, regs, regd)
#define	pcmpeqw(vars, vard)	mmx_m2m(pcmpeqw, vars, vard)

#define	pcmpeqb_m2r(var, reg)	mmx_m2r(pcmpeqb, var, reg)
#define	pcmpeqb_r2r(regs, regd)	mmx_r2r(pcmpeqb, regs, regd)
#define	pcmpeqb(vars, vard)	mmx_m2m(pcmpeqb, vars, vard)


/*	2x32, 4x16, and 8x8 Parallel CoMPare for Greater Than
	(resulting fields are either 0 or -1)
*/
#define	pcmpgtd_m2r(var, reg)	mmx_m2r(pcmpgtd, var, reg)
#define	pcmpgtd_r2r(regs, regd)	mmx_r2r(pcmpgtd, regs, regd)
#define	pcmpgtd(vars, vard)	mmx_m2m(pcmpgtd, vars, vard)

#define	pcmpgtw_m2r(var, reg)	mmx_m2r(pcmpgtw, var, reg)
#define	pcmpgtw_r2r(regs, regd)	mmx_r2r(pcmpgtw, regs, regd)
#define	pcmpgtw(vars, vard)	mmx_m2m(pcmpgtw, vars, vard)

#define	pcmpgtb_m2r(var, reg)	mmx_m2r(pcmpgtb, var, reg)
#define	pcmpgtb_r2r(regs, regd)	mmx_r2r(pcmpgtb, regs, regd)
#define	pcmpgtb(vars, vard)	mmx_m2m(pcmpgtb, vars, vard)


/*	1x64, 2x32, and 4x16 Parallel Shift Left Logical
*/
#define	psllq_i2r(imm, reg)	mmx_i2r(psllq, imm, reg)
#define	psllq_m2r(var, reg)	mmx_m2r(psllq, var, reg)
#define	psllq_r2r(regs, regd)	mmx_r2r(psllq, regs, regd)
#define	psllq(vars, vard)	mmx_m2m(psllq, vars, vard)

#define	pslld_i2r(imm, reg)	mmx_i2r(pslld, imm, reg)
#define	pslld_m2r(var, reg)	mmx_m2r(pslld, var, reg)
#define	pslld_r2r(regs, regd)	mmx_r2r(pslld, regs, regd)
#define	pslld(vars, vard)	mmx_m2m(pslld, vars, vard)

#define	psllw_i2r(imm, reg)	mmx_i2r(psllw, imm, reg)
#define	psllw_m2r(var, reg)	mmx_m2r(psllw, var, reg)
#define	psllw_r2r(regs, regd)	mmx_r2r(psllw, regs, regd)
#define	psllw(vars, vard)	mmx_m2m(psllw, vars, vard)


/*	1x64, 2x32, and 4x16 Parallel Shift Right Logical
*/
#define	psrlq_i2r(imm, reg)	mmx_i2r(psrlq, imm, reg)
#define	psrlq_m2r(var, reg)	mmx_m2r(psrlq, var, reg)
#define	psrlq_r2r(regs, regd)	mmx_r2r(psrlq, regs, regd)
#define	psrlq(vars, vard)	mmx_m2m(psrlq, vars, vard)

#define	psrld_i2r(imm, reg)	mmx_i2r(psrld, imm, reg)
#define	psrld_m2r(var, reg)	mmx_m2r(psrld, var, reg)
#define	psrld_r2r(regs, regd)	mmx_r2r(psrld, regs, regd)
#define	psrld(vars, vard)	mmx_m2m(psrld, vars, vard)

#define	psrlw_i2r(imm, reg)	mmx_i2r(psrlw, imm, reg)
#define	psrlw_m2r(var, reg)	mmx_m2r(psrlw, var, reg)
#define	psrlw_r2r(regs, regd)	mmx_r2r(psrlw, regs, regd)
#define	psrlw(vars, vard)	mmx_m2m(psrlw, vars, vard)


/*	2x32 and 4x16 Parallel Shift Right Arithmetic
*/
#define	psrad_i2r(imm, reg)	mmx_i2r(psrad, imm, reg)
#define	psrad_m2r(var, reg)	mmx_m2r(psrad, var, reg)
#define	psrad_r2r(regs, regd)	mmx_r2r(psrad, regs, regd)
#define	psrad(vars, vard)	mmx_m2m(psrad, vars, vard)

#define	psraw_i2r(imm, reg)	mmx_i2r(psraw, imm, reg)
#define	psraw_m2r(var, reg)	mmx_m2r(psraw, var, reg)
#define	psraw_r2r(regs, regd)	mmx_r2r(psraw, regs, regd)
#define	psraw(vars, vard)	mmx_m2m(psraw, vars, vard)


/*	2x32->4x16 and 4x16->8x8 PACK and Signed Saturate
	(packs source and dest fields into dest in that order)
*/
#define	packssdw_m2r(var, reg)	mmx_m2r(packssdw, var, reg)
#define	packssdw_r2r(regs, regd) mmx_r2r(packssdw, regs, regd)
#define	packssdw(vars, vard)	mmx_m2m(packssdw, vars, vard)

#define	packsswb_m2r(var, reg)	mmx_m2r(packsswb, var, reg)
#define	packsswb_r2r(regs, regd) mmx_r2r(packsswb, regs, regd)
#define	packsswb(vars, vard)	mmx_m2m(packsswb, vars, vard)


/*	4x16->8x8 PACK and Unsigned Saturate
	(packs source and dest fields into dest in that order)
*/
#define	packuswb_m2r(var, reg)	mmx_m2r(packuswb, var, reg)
#define	packuswb_r2r(regs, regd) mmx_r2r(packuswb, regs, regd)
#define	packuswb(vars, vard)	mmx_m2m(packuswb, vars, vard)


/*	2x32->1x64, 4x16->2x32, and 8x8->4x16 UNPaCK Low
	(interleaves low half of dest with low half of source
	 as padding in each result field)
*/
#define	punpckldq_m2r(var, reg)	mmx_m2r(punpckldq, var, reg)
#define	punpckldq_r2r(regs, regd) mmx_r2r(punpckldq, regs, regd)
#define	punpckldq(vars, vard)	mmx_m2m(punpckldq, vars, vard)

#define	punpcklwd_m2r(var, reg)	mmx_m2r(punpcklwd, var, reg)
#define	punpcklwd_r2r(regs, regd) mmx_r2r(punpcklwd, regs, regd)
#define	punpcklwd(vars, vard)	mmx_m2m(punpcklwd, vars, vard)

#define	punpcklbw_m2r(var, reg)	mmx_m2r(punpcklbw, var, reg)
#define	punpcklbw_r2r(regs, regd) mmx_r2r(punpcklbw, regs, regd)
#define	punpcklbw(vars, vard)	mmx_m2m(punpcklbw, vars, vard)


/*	2x32->1x64, 4x16->2x32, and 8x8->4x16 UNPaCK High
	(interleaves high half of dest with high half of source
	 as padding in each result field)
*/
#define	punpckhdq_m2r(var, reg)	mmx_m2r(punpckhdq, var, reg)
#define	punpckhdq_r2r(regs, regd) mmx_r2r(punpckhdq, regs, regd)
#define	punpckhdq(vars, vard)	mmx_m2m(punpckhdq, vars, vard)

#define	punpckhwd_m2r(var, reg)	mmx_m2r(punpckhwd, var, reg)
#define	punpckhwd_r2r(regs, regd) mmx_r2r(punpckhwd, regs, regd)
#define	punpckhwd(vars, vard)	mmx_m2m(punpckhwd, vars, vard)

#define	punpckhbw_m2r(var, reg)	mmx_m2r(punpckhbw, var, reg)
#define	punpckhbw_r2r(regs, regd) mmx_r2r(punpckhbw, regs, regd)
#define	punpckhbw(vars, vard)	mmx_m2m(punpckhbw, vars, vard)

#define MOVE_8DWORDS_MMX(src,dst) \
	   __asm__ ( \
		"movq (%1), %%mm0 \n" \
		"movq 0x8(%1), %%mm1 \n" \
	 	"movq 0x10(%1), %%mm2 \n" \
		"movq 0x18(%1), %%mm3 \n" \
		"movq %%mm0, (%0) \n" \
		"movq %%mm1, 0x8(%0) \n" \
		"movq %%mm2, 0x10(%0) \n" \
		"movq %%mm3, 0x18(%0) \n" \
		: \
		: "q" (dst), "r" (src) \
		: "memory",  "st");

#define MOVE_10DWORDS_MMX(src,dst) \
	   __asm__ ( \
		"movq (%1), %%mm0 \n" \
		"movq 0x8(%1), %%mm1 \n" \
	 	"movq 0x10(%1), %%mm2 \n" \
		"movq 0x18(%1), %%mm3 \n" \
		"movq 0x20(%1), %%mm4 \n" \
		"movq %%mm0, (%0) \n" \
		"movq %%mm1, 0x8(%0) \n" \
		"movq %%mm2, 0x10(%0) \n" \
		"movq %%mm3, 0x18(%0) \n" \
		"movq %%mm4, 0x20(%0) \n" \
		: \
		: "q" (dst), "r" (src) \
		: "memory",  "st");

#define MOVE_16DWORDS_MMX(src,dst) \
	   __asm__ ( \
		"movq (%1), %%mm0 \n" \
		"movq 0x8(%1), %%mm1 \n" \
	 	"movq 0x10(%1), %%mm2 \n" \
		"movq 0x18(%1), %%mm3 \n" \
		"movq 0x20(%1), %%mm4 \n" \
		"movq 0x28(%1), %%mm5 \n" \
		"movq 0x30(%1), %%mm6 \n" \
		"movq 0x38(%1), %%mm7 \n" \
		"movq %%mm0, (%0) \n" \
		"movq %%mm1, 0x8(%0) \n" \
		"movq %%mm2, 0x10(%0) \n" \
		"movq %%mm3, 0x18(%0) \n" \
		"movq %%mm4, 0x20(%0) \n" \
		"movq %%mm5, 0x28(%0) \n" \
		"movq %%mm6, 0x30(%0) \n" \
		"movq %%mm7, 0x38(%0) \n" \
		: \
		: "q" (dst), "r" (src) \
		: "memory",  "st");

#define MOVE_16DWORDS_MMX2(src,dst) \
	   __asm__ ( \
		"movq (%1), %%mm0 \n" \
		"movq 0x8(%1), %%mm1 \n" \
	 	"movq 0x10(%1), %%mm2 \n" \
		"movq 0x18(%1), %%mm3 \n" \
		"movq 0x20(%1), %%mm4 \n" \
		"movq 0x28(%1), %%mm5 \n" \
		"movq 0x30(%1), %%mm6 \n" \
		"movq 0x38(%1), %%mm7 \n" \
		"movntq %%mm0, (%0) \n" \
		"movntq %%mm1, 0x8(%0) \n" \
		"movntq %%mm2, 0x10(%0) \n" \
		"movntq %%mm3, 0x18(%0) \n" \
		"movntq %%mm4, 0x20(%0) \n" \
		"movntq %%mm5, 0x28(%0) \n" \
		"movntq %%mm6, 0x30(%0) \n" \
		"movntq %%mm7, 0x38(%0) \n" \
		: \
		: "q" (dst), "r" (src) \
		: "memory",  "st");

#define MOVE_32DWORDS_SSE2(src,dst) \
	   __asm__ ( \
		"movdqu (%1), %%xmm0 \n" \
		"movdqu 0x10(%1), %%xmm1 \n" \
	 	"movdqu 0x20(%1), %%xmm2 \n" \
		"movdqu 0x30(%1), %%xmm3 \n" \
		"movdqu 0x40(%1), %%xmm4 \n" \
		"movdqu 0x50(%1), %%xmm5 \n" \
		"movdqu 0x60(%1), %%xmm6 \n" \
		"movdqu 0x70(%1), %%xmm7 \n" \
		"movntdq %%xmm0, (%0) \n" \
		"movntdq %%xmm1, 0x10(%0) \n" \
		"movntdq %%xmm2, 0x20(%0) \n" \
		"movntdq %%xmm3, 0x30(%0) \n" \
		"movntdq %%xmm4, 0x40(%0) \n" \
		"movntdq %%xmm5, 0x50(%0) \n" \
		"movntdq %%xmm6, 0x60(%0) \n" \
		"movntdq %%xmm7, 0x70(%0) \n" \
		: \
		: "q" (dst), "r" (src) \
		: "memory",  "st");

#define MOVE_32DWORDS_ALIGNED_SSE2(src,dst) \
	   __asm__ ( \
		"movdqa (%1), %%xmm0 \n" \
		"movdqa 0x10(%1), %%xmm1 \n" \
	 	"movdqa 0x20(%1), %%xmm2 \n" \
		"movdqa 0x30(%1), %%xmm3 \n" \
		"movdqa 0x40(%1), %%xmm4 \n" \
		"movdqa 0x50(%1), %%xmm5 \n" \
		"movdqa 0x60(%1), %%xmm6 \n" \
		"movdqa 0x70(%1), %%xmm7 \n" \
		"movntdq %%xmm0, (%0) \n" \
		"movntdq %%xmm1, 0x10(%0) \n" \
		"movntdq %%xmm2, 0x20(%0) \n" \
		"movntdq %%xmm3, 0x30(%0) \n" \
		"movntdq %%xmm4, 0x40(%0) \n" \
		"movntdq %%xmm5, 0x50(%0) \n" \
		"movntdq %%xmm6, 0x60(%0) \n" \
		"movntdq %%xmm7, 0x70(%0) \n" \
		: \
		: "q" (dst), "r" (src) \
		: "memory",  "st");

/*	Empty MMx State
	(used to clean-up when going from mmx to float use
	 of the registers that are shared by both; note that
	 there is no float-to-mmx operation needed, because
	 only the float tag word info is corruptible)
*/

#define	emms()			__asm__ __volatile__ ("emms":::"memory")
#define	sfence()		__asm__ __volatile__ ("sfence":::"memory")

/* additions to detect mmx - */
/* Raster <raster@rasterman.com> */

#ifndef CPUID_MMX
# define CPUID_MMX  (1 << 23) /* flags: mmx */
#endif
#ifndef CPUID_SSE
# define CPUID_SSE  (1 << 25) /* flags: xmm */
#endif
#ifndef CPUID_SSE2
# define CPUID_SSE2 (1 << 26) /* flags: ? */
#endif

#ifdef __amd64
#define have_cpuid(cpuid_ret) \
         __asm__ __volatile__ ( \
                                  ".align 32               \n" \
                                  "  pushq %%rbx           \n" \
                                  "  pushfq                \n" \
                                  "  popq %%rax            \n" \
                                  "  movq %%rax, %%rbx     \n" \
                                  "  xorq $0x200000, %%rax \n" \
                                  "  pushq %%rax           \n" \
                                  "  popfq                 \n" \
                                  "  pushfq                \n" \
                                  "  popq %%rax            \n" \
                                  "  cmpq %%rax, %%rbx     \n" \
                                  "  je 1f                 \n" \
                                  "  movl $1, %0           \n" \
                                  "  jmp 2f                \n" \
                                  "1:                      \n" \
                                  "  movl $0, %0           \n" \
                                  "2:                      \n" \
                                  "  popq %%rbx            \n" \
				   : "=m" (cpuid_ret)           \
                                  );

#define get_cpuid(cpuid_ret) \
         __asm__ __volatile__ ( \
                                  ".align 32               \n" \
                                  "  pushq %%rax           \n" \
                                  "  movl $1, %%eax        \n" \
                                  "  cpuid                 \n" \
                                  "  test $0x00800000, %%edx\n" \
                                  "1:                      \n" \
                                  "  movl %%edx, %0        \n" \
                                  "  jmp 2f                \n" \
                                  "2:                      \n" \
                                  "  movl $0, %0           \n" \
                                  "  popq %%rax            \n" \
				   : "=m" (cpuid_ret)           \
                                  );
#else
#define have_cpuid(cpuid_ret) \
	 __asm__ __volatile__ ( \
				  ".align 32               \n" \
				  "  pushl %%ebx           \n" \
				  "  pushfl                \n" \
				  "  popl %%eax            \n" \
				  "  movl %%eax, %%ebx     \n" \
				  "  xorl $0x200000, %%eax \n" \
				  "  pushl %%eax           \n" \
				  "  popfl                 \n" \
				  "  pushfl                \n" \
				  "  popl %%eax            \n" \
				  "  cmpl %%eax, %%ebx     \n" \
				  "  je 1f                 \n" \
				  "  movl $1, %0           \n" \
				  "  jmp 2f                \n" \
				  "1:                      \n" \
				  "  movl $0, %0           \n" \
				  "2:                      \n" \
				  "  popl %%ebx            \n" \
				  : "=m" (cpuid_ret)           \
				  );

#define get_cpuid(cpuid_ret) \
	 __asm__ __volatile__ ( \
				  ".align 32               \n" \
				  "  pushl %%eax           \n" \
				  "  movl $1, %%eax        \n" \
				  "  cpuid                 \n" \
				  "  test $0x00800000, %%edx\n" \
				  "1:                      \n" \
				  "  movl %%edx, %0        \n" \
				  "  jmp 2f                \n" \
				  "2:                      \n" \
				  "  movl $0, %0           \n" \
				  "  popl %%eax            \n" \
				  : "=m" (cpuid_ret)           \
				  );
#endif
#define prefetch(var) \
	__asm__ __volatile__ ( \
				 "prefetchnta (%0) \n" \
				 : \
				 : "r" (var) \
				 );
#define prefetch0(var) \
	__asm__ __volatile__ ( \
				 "prefetcht0 (%0) \n" \
				 : \
				 : "r" (var) \
				 );
#define prefetch1(var) \
	__asm__ __volatile__ ( \
				 "prefetcht1 (%0) \n" \
				 : \
				 : "r" (var) \
				 );
#define prefetch2(var) \
	__asm__ __volatile__ ( \
				 "prefetcht2 (%0) \n" \
				 : \
				 : "r" (var) \
				 );
#define pshufw(r1, r2, imm) \
	__asm__ __volatile__ ( \
				 "pshufw $" #imm ", %" #r1 ", %" #r2 " \n" \
				 );

#define pshufhw(r1, r2, imm) \
	__asm__ __volatile__ ( \
				 "pshufhw $" #imm ", %" #r1 ", %" #r2 " \n" \
				 );

#define pshuflw(r1, r2, imm) \
	__asm__ __volatile__ ( \
				 "pshuflw $" #imm ", %" #r1 ", %" #r2 " \n" \
				 );
#define pshufd(r1, r2, imm) \
	__asm__ __volatile__ ( \
				 "pshufd $" #imm ", %" #r1 ", %" #r2 " \n" \
				 );

/*	1x238 MOVE Doouble Quadword
	(this is both a load and a store...
	 in fact, it is the only way to store)
*/
#define	movdqu_m2r(var, reg)	mmx_m2r(movdqu, var, reg)
#define	movdqu_r2m(reg, var)	mmx_r2m(movdqu, reg, var)
#define	movdqu_r2r(regs, regd)	mmx_r2r(movdqu, regs, regd)
#define	movdqu(vars, vard) \
	__asm__ __volatile__ ("movdqu %1, %%xmm0\n\t" \
			      "movdqu %%xmm0, %0" \
			      : "=X" (vard) \
			      : "X" (vars))
#define	movdqa_m2r(var, reg)	mmx_m2r(movdqa, var, reg)
#define	movdqa_r2m(reg, var)	mmx_r2m(movdqa, reg, var)
#define	movdqa_r2r(regs, regd)	mmx_r2r(movdqa, regs, regd)
#define	movdqa(vars, vard) \
	__asm__ __volatile__ ("movdqa %1, %%xmm0\n\t" \
			      "movdqa %%xmm0, %0" \
			      : "=X" (vard) \
			      : "X" (vars))
#define	movntdq_r2m(reg, var)   mmx_r2m(movntdq, reg, var)


/* end additions */

#endif