aboutsummaryrefslogtreecommitdiff
path: root/vendor/gmp-6.3.0/mpn/x86/pentium/mmx/lshift.asm
blob: 04b0ddcc8f09e986e394d462acd965e612f64655 (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
dnl  Intel P5 mpn_lshift -- mpn left shift.

dnl  Copyright 2000-2002 Free Software Foundation, Inc.

dnl  This file is part of the GNU MP Library.
dnl
dnl  The GNU MP Library is free software; you can redistribute it and/or modify
dnl  it under the terms of either:
dnl
dnl    * the GNU Lesser General Public License as published by the Free
dnl      Software Foundation; either version 3 of the License, or (at your
dnl      option) any later version.
dnl
dnl  or
dnl
dnl    * the GNU General Public License as published by the Free Software
dnl      Foundation; either version 2 of the License, or (at your option) any
dnl      later version.
dnl
dnl  or both in parallel, as here.
dnl
dnl  The GNU MP Library is distributed in the hope that it will be useful, but
dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
dnl  for more details.
dnl
dnl  You should have received copies of the GNU General Public License and the
dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
dnl  see https://www.gnu.org/licenses/.

include(`../config.m4')


C P5: 1.75 cycles/limb.


C mp_limb_t mpn_lshift (mp_ptr dst, mp_srcptr src, mp_size_t size,
C                       unsigned shift);
C
C Shift src,size left by shift many bits and store the result in dst,size.
C Zeros are shifted in at the right.  Return the bits shifted out at the
C left.
C
C The comments in mpn_rshift apply here too.

defframe(PARAM_SHIFT,16)
defframe(PARAM_SIZE, 12)
defframe(PARAM_SRC,  8)
defframe(PARAM_DST,  4)
deflit(`FRAME',0)

dnl  minimum 5, because the unrolled loop can't handle less
deflit(UNROLL_THRESHOLD, 5)

	TEXT
	ALIGN(8)

PROLOGUE(mpn_lshift)

	pushl	%ebx
	pushl	%edi
deflit(`FRAME',8)

	movl	PARAM_SIZE, %eax
	movl	PARAM_DST, %edx

	movl	PARAM_SRC, %ebx
	movl	PARAM_SHIFT, %ecx

	cmp	$UNROLL_THRESHOLD, %eax
	jae	L(unroll)

	movl	-4(%ebx,%eax,4), %edi	C src high limb
	decl	%eax

	jnz	L(simple)

	shldl(	%cl, %edi, %eax)	C eax was decremented to zero

	shll	%cl, %edi

	movl	%edi, (%edx)		C dst low limb
	popl	%edi			C risk of data cache bank clash

	popl	%ebx

	ret


C -----------------------------------------------------------------------------
L(simple):
	C eax	size-1
	C ebx	src
	C ecx	shift
	C edx	dst
	C esi
	C edi
	C ebp
deflit(`FRAME',8)

	movd	(%ebx,%eax,4), %mm5	C src high limb

	movd	%ecx, %mm6		C lshift
	negl	%ecx

	psllq	%mm6, %mm5
	addl	$32, %ecx

	movd	%ecx, %mm7
	psrlq	$32, %mm5		C retval


L(simple_top):
	C eax	counter, limbs, negative
	C ebx	src
	C ecx
	C edx	dst
	C esi
	C edi
	C
	C mm0	scratch
	C mm5	return value
	C mm6	shift
	C mm7	32-shift

	movq	-4(%ebx,%eax,4), %mm0
	decl	%eax

	psrlq	%mm7, %mm0

	C

	movd	%mm0, 4(%edx,%eax,4)
	jnz	L(simple_top)


	movd	(%ebx), %mm0

	movd	%mm5, %eax
	psllq	%mm6, %mm0

	popl	%edi
	popl	%ebx

	movd	%mm0, (%edx)

	emms

	ret


C -----------------------------------------------------------------------------
	ALIGN(8)
L(unroll):
	C eax	size
	C ebx	src
	C ecx	shift
	C edx	dst
	C esi
	C edi
	C ebp
deflit(`FRAME',8)

	movd	-4(%ebx,%eax,4), %mm5	C src high limb
	leal	(%ebx,%eax,4), %edi

	movd	%ecx, %mm6		C lshift
	andl	$4, %edi

	psllq	%mm6, %mm5
	jz	L(start_src_aligned)


	C src isn't aligned, process high limb separately (marked xxx) to
	C make it so.
	C
	C  source     -8(ebx,%eax,4)
	C                  |
	C  +-------+-------+-------+--
	C  |               |
	C  +-------+-------+-------+--
	C        0mod8   4mod8   0mod8
	C
	C  dest
	C     -4(edx,%eax,4)
	C          |
	C  +-------+-------+--
	C  |  xxx  |       |
	C  +-------+-------+--

	movq	-8(%ebx,%eax,4), %mm0	C unaligned load

	psllq	%mm6, %mm0
	decl	%eax

	psrlq	$32, %mm0

	C

	movd	%mm0, (%edx,%eax,4)
L(start_src_aligned):

	movq	-8(%ebx,%eax,4), %mm1	C src high qword
	leal	(%edx,%eax,4), %edi

	andl	$4, %edi
	psrlq	$32, %mm5		C return value

	movq	-16(%ebx,%eax,4), %mm3	C src second highest qword
	jz	L(start_dst_aligned)

	C dst isn't aligned, subtract 4 to make it so, and pretend the shift
	C is 32 bits extra.  High limb of dst (marked xxx) handled here
	C separately.
	C
	C  source     -8(ebx,%eax,4)
	C                  |
	C  +-------+-------+--
	C  |      mm1      |
	C  +-------+-------+--
	C                0mod8   4mod8
	C
	C  dest
	C     -4(edx,%eax,4)
	C          |
	C  +-------+-------+-------+--
	C  |  xxx  |               |
	C  +-------+-------+-------+--
	C        0mod8   4mod8   0mod8

	movq	%mm1, %mm0
	addl	$32, %ecx		C new shift

	psllq	%mm6, %mm0

	movd	%ecx, %mm6
	psrlq	$32, %mm0

	C wasted cycle here waiting for %mm0

	movd	%mm0, -4(%edx,%eax,4)
	subl	$4, %edx
L(start_dst_aligned):


	psllq	%mm6, %mm1
	negl	%ecx			C -shift

	addl	$64, %ecx		C 64-shift
	movq	%mm3, %mm2

	movd	%ecx, %mm7
	subl	$8, %eax		C size-8

	psrlq	%mm7, %mm3

	por	%mm1, %mm3		C mm3 ready to store
	jc	L(finish)


	C The comments in mpn_rshift apply here too.

	ALIGN(8)
L(unroll_loop):
	C eax	counter, limbs
	C ebx	src
	C ecx
	C edx	dst
	C esi
	C edi
	C
	C mm0
	C mm1
	C mm2	src qword from 16(%ebx,%eax,4)
	C mm3	dst qword ready to store to 24(%edx,%eax,4)
	C
	C mm5	return value
	C mm6	lshift
	C mm7	rshift

	movq	8(%ebx,%eax,4), %mm0
	psllq	%mm6, %mm2

	movq	%mm0, %mm1
	psrlq	%mm7, %mm0

	movq	%mm3, 24(%edx,%eax,4)	C prev
	por	%mm2, %mm0

	movq	(%ebx,%eax,4), %mm3	C
	psllq	%mm6, %mm1		C

	movq	%mm0, 16(%edx,%eax,4)
	movq	%mm3, %mm2		C

	psrlq	%mm7, %mm3		C
	subl	$4, %eax

	por	%mm1, %mm3		C
	jnc	L(unroll_loop)



L(finish):
	C eax	-4 to -1 representing respectively 0 to 3 limbs remaining

	testb	$2, %al

	jz	L(finish_no_two)

	movq	8(%ebx,%eax,4), %mm0
	psllq	%mm6, %mm2

	movq	%mm0, %mm1
	psrlq	%mm7, %mm0

	movq	%mm3, 24(%edx,%eax,4)	C prev
	por	%mm2, %mm0

	movq	%mm1, %mm2
	movq	%mm0, %mm3

	subl	$2, %eax
L(finish_no_two):


	C eax	-4 or -3 representing respectively 0 or 1 limbs remaining
	C
	C mm2	src prev qword, from 16(%ebx,%eax,4)
	C mm3	dst qword, for 24(%edx,%eax,4)

	testb	$1, %al
	movd	%mm5, %eax	C retval

	popl	%edi
	jz	L(finish_zero)


	C One extra src limb, destination was aligned.
	C
	C                 source                  ebx
	C                 --+---------------+-------+
	C                   |      mm2      |       |
	C                 --+---------------+-------+
	C
	C dest         edx+12           edx+4     edx
	C --+---------------+---------------+-------+
	C   |      mm3      |               |       |
	C --+---------------+---------------+-------+
	C
	C mm6 = shift
	C mm7 = ecx = 64-shift


	C One extra src limb, destination was unaligned.
	C
	C                 source                  ebx
	C                 --+---------------+-------+
	C                   |      mm2      |       |
	C                 --+---------------+-------+
	C
	C         dest         edx+12           edx+4
	C         --+---------------+---------------+
	C           |      mm3      |               |
	C         --+---------------+---------------+
	C
	C mm6 = shift+32
	C mm7 = ecx = 64-(shift+32)


	C In both cases there's one extra limb of src to fetch and combine
	C with mm2 to make a qword at 4(%edx), and in the aligned case
	C there's an extra limb of dst to be formed from that extra src limb
	C left shifted.


	movd	(%ebx), %mm0
	psllq	%mm6, %mm2

	movq	%mm3, 12(%edx)
	psllq	$32, %mm0

	movq	%mm0, %mm1
	psrlq	%mm7, %mm0

	por	%mm2, %mm0
	psllq	%mm6, %mm1

	movq	%mm0, 4(%edx)
	psrlq	$32, %mm1

	andl	$32, %ecx
	popl	%ebx

	jz	L(finish_one_unaligned)

	movd	%mm1, (%edx)
L(finish_one_unaligned):

	emms

	ret


L(finish_zero):

	C No extra src limbs, destination was aligned.
	C
	C                 source          ebx
	C                 --+---------------+
	C                   |      mm2      |
	C                 --+---------------+
	C
	C dest          edx+8             edx
	C --+---------------+---------------+
	C   |      mm3      |               |
	C --+---------------+---------------+
	C
	C mm6 = shift
	C mm7 = ecx = 64-shift


	C No extra src limbs, destination was unaligned.
	C
	C               source            ebx
	C                 --+---------------+
	C                   |      mm2      |
	C                 --+---------------+
	C
	C         dest          edx+8   edx+4
	C         --+---------------+-------+
	C           |      mm3      |       |
	C         --+---------------+-------+
	C
	C mm6 = shift+32
	C mm7 = ecx = 64-(shift+32)


	C The movd for the unaligned case writes the same data to 4(%edx)
	C that the movq does for the aligned case.


	movq	%mm3, 8(%edx)
	andl	$32, %ecx

	psllq	%mm6, %mm2
	jz	L(finish_zero_unaligned)

	movq	%mm2, (%edx)
L(finish_zero_unaligned):

	psrlq	$32, %mm2
	popl	%ebx

	movd	%mm5, %eax	C retval

	movd	%mm2, 4(%edx)

	emms

	ret

EPILOGUE()