aboutsummaryrefslogtreecommitdiff
path: root/vendor/gmp-6.3.0/mpn/x86/k7/mmx/lshift.asm
blob: b3383cf2c30011548d0635f6ebe580d51c50099c (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
dnl  AMD K7 mpn_lshift -- mpn left shift.

dnl  Copyright 1999-2002 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 K7: 1.21 cycles/limb (at 16 limbs/loop).



dnl  K7: UNROLL_COUNT cycles/limb
dnl           4           1.51
dnl           8           1.26
dnl          16           1.21
dnl          32           1.2
dnl  Maximum possible with the current code is 64.

deflit(UNROLL_COUNT, 16)


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.  The bits shifted out at the left are
C the return value.
C
C The comments in mpn_rshift apply here too.

ifdef(`PIC',`
deflit(UNROLL_THRESHOLD, 10)
',`
deflit(UNROLL_THRESHOLD, 10)
')

defframe(PARAM_SHIFT,16)
defframe(PARAM_SIZE, 12)
defframe(PARAM_SRC,  8)
defframe(PARAM_DST,  4)

defframe(SAVE_EDI, -4)
defframe(SAVE_ESI, -8)
defframe(SAVE_EBX, -12)
deflit(SAVE_SIZE, 12)

	TEXT
	ALIGN(32)

PROLOGUE(mpn_lshift)
deflit(`FRAME',0)

	movl	PARAM_SIZE, %eax
	movl	PARAM_SRC, %edx
	subl	$SAVE_SIZE, %esp
deflit(`FRAME',SAVE_SIZE)

	movl	PARAM_SHIFT, %ecx
	movl	%edi, SAVE_EDI

	movl	PARAM_DST, %edi
	decl	%eax
	jnz	L(more_than_one_limb)

	movl	(%edx), %edx

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

	shll	%cl, %edx

	movl	%edx, (%edi)
	movl	SAVE_EDI, %edi
	addl	$SAVE_SIZE, %esp

	ret


C -----------------------------------------------------------------------------
L(more_than_one_limb):
	C eax	size-1
	C ebx
	C ecx	shift
	C edx	src
	C esi
	C edi	dst
	C ebp

	movd	PARAM_SHIFT, %mm6
	movd	(%edx,%eax,4), %mm5	C src high limb
	cmp	$UNROLL_THRESHOLD-1, %eax

	jae	L(unroll)
	negl	%ecx
	movd	(%edx), %mm4		C src low limb

	addl	$32, %ecx

	movd	%ecx, %mm7

L(simple_top):
	C eax	loop counter, limbs
	C ebx
	C ecx
	C edx	src
	C esi
	C edi	dst
	C ebp
	C
	C mm0	scratch
	C mm4	src low limb
	C mm5	src high limb
	C mm6	shift
	C mm7	32-shift

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

	psrlq	%mm7, %mm0

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


	psllq	%mm6, %mm5
	psllq	%mm6, %mm4

	psrlq	$32, %mm5
	movd	%mm4, (%edi)		C dst low limb

	movd	%mm5, %eax		C return value

	movl	SAVE_EDI, %edi
	addl	$SAVE_SIZE, %esp
	emms

	ret


C -----------------------------------------------------------------------------
	ALIGN(16)
L(unroll):
	C eax	size-1
	C ebx	(saved)
	C ecx	shift
	C edx	src
	C esi
	C edi	dst
	C ebp
	C
	C mm5	src high limb, for return value
	C mm6	lshift

	movl	%esi, SAVE_ESI
	movl	%ebx, SAVE_EBX
	leal	-4(%edx,%eax,4), %edx   C &src[size-2]

	testb	$4, %dl
	movq	(%edx), %mm1		C src high qword

	jz	L(start_src_aligned)


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

	psllq	%mm6, %mm1
	subl	$4, %edx
	movl	%eax, PARAM_SIZE	C size-1

	psrlq	$32, %mm1
	decl	%eax			C size-2 is new size-1

	movd	%mm1, 4(%edi,%eax,4)
	movq	(%edx), %mm1		C new src high qword
L(start_src_aligned):


	leal	-4(%edi,%eax,4), %edi   C &dst[size-2]
	psllq	%mm6, %mm5

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

	jz	L(start_dst_aligned)


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

	movq	%mm1, %mm0
	psllq	%mm6, %mm1
	addl	$32, %ecx		C shift+32

	psrlq	$32, %mm1

	movd	%mm1, 4(%edi)
	movq	%mm0, %mm1
	subl	$4, %edi

	movd	%ecx, %mm6		C new lshift
L(start_dst_aligned):

	decl	%eax			C size-2, two last limbs handled at end
	movq	%mm1, %mm2		C copy of src high qword
	negl	%ecx

	andl	$-2, %eax		C round size down to even
	addl	$64, %ecx

	movl	%eax, %ebx
	negl	%eax

	andl	$UNROLL_MASK, %eax
	decl	%ebx

	shll	%eax

	movd	%ecx, %mm7		C rshift = 64-lshift

ifdef(`PIC',`
	call	L(pic_calc)
L(here):
',`
	leal	L(entry) (%eax,%eax,4), %esi
')
	shrl	$UNROLL_LOG2, %ebx	C loop counter

	leal	ifelse(UNROLL_BYTES,256,128) -8(%edx,%eax,2), %edx
	leal	ifelse(UNROLL_BYTES,256,128) (%edi,%eax,2), %edi
	movl	PARAM_SIZE, %eax	C for use at end
	jmp	*%esi


ifdef(`PIC',`
L(pic_calc):
	C See mpn/x86/README about old gas bugs
	leal	(%eax,%eax,4), %esi
	addl	$L(entry)-L(here), %esi
	addl	(%esp), %esi

	ret_internal
')


C -----------------------------------------------------------------------------
	ALIGN(32)
L(top):
	C eax	size (for use at end)
	C ebx	loop counter
	C ecx	rshift
	C edx	src
	C esi	computed jump
	C edi	dst
	C ebp
	C
	C mm0	scratch
	C mm1	\ carry (alternating, mm2 first)
	C mm2	/
	C mm6	lshift
	C mm7	rshift
	C
	C 10 code bytes/limb
	C
	C The two chunks differ in whether mm1 or mm2 hold the carry.
	C The computed jump puts the initial carry in both mm1 and mm2.

L(entry):
deflit(CHUNK_COUNT, 4)
forloop(i, 0, UNROLL_COUNT/CHUNK_COUNT-1, `
	deflit(`disp0', eval(-i*CHUNK_COUNT*4 ifelse(UNROLL_BYTES,256,-128)))
	deflit(`disp1', eval(disp0 - 8))

Zdisp(	movq,	disp0,(%edx), %mm0)
	psllq	%mm6, %mm2

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

	por	%mm2, %mm0
Zdisp(	movq,	%mm0, disp0,(%edi))


Zdisp(	movq,	disp1,(%edx), %mm0)
	psllq	%mm6, %mm1

	movq	%mm0, %mm2
	psrlq	%mm7, %mm0

	por	%mm1, %mm0
Zdisp(	movq,	%mm0, disp1,(%edi))
')

	subl	$UNROLL_BYTES, %edx
	subl	$UNROLL_BYTES, %edi
	decl	%ebx

	jns	L(top)



define(`disp', `m4_empty_if_zero(eval($1 ifelse(UNROLL_BYTES,256,-128)))')

L(end):
	testb	$1, %al
	movl	SAVE_EBX, %ebx
	psllq	%mm6, %mm2	C wanted left shifted in all cases below

	movd	%mm5, %eax

	movl	SAVE_ESI, %esi
	jz	L(end_even)


L(end_odd):

	C Size odd, destination was aligned.
	C
	C                 source        edx+8   edx+4
	C                 --+---------------+-------+
	C                   |      mm2      |       |
	C                 --+---------------+-------+
	C
	C dest                            edi
	C --+---------------+---------------+-------+
	C   |   written     |               |       |
	C --+---------------+---------------+-------+
	C
	C mm6 = shift
	C mm7 = ecx = 64-shift


	C Size odd, destination was unaligned.
	C
	C                 source        edx+8   edx+4
	C                 --+---------------+-------+
	C                   |      mm2      |       |
	C                 --+---------------+-------+
	C
	C         dest                            edi
	C         --+---------------+---------------+
	C           |   written     |               |
	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 (%edi), 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	disp(4) (%edx), %mm0
	testb	$32, %cl

	movq	%mm0, %mm1
	psllq	$32, %mm0

	psrlq	%mm7, %mm0
	psllq	%mm6, %mm1

	por	%mm2, %mm0

	movq	%mm0, disp(0) (%edi)
	jz	L(end_odd_unaligned)
	movd	%mm1, disp(-4) (%edi)
L(end_odd_unaligned):

	movl	SAVE_EDI, %edi
	addl	$SAVE_SIZE, %esp
	emms

	ret


L(end_even):

	C Size even, destination was aligned.
	C
	C                 source        edx+8
	C                 --+---------------+
	C                   |      mm2      |
	C                 --+---------------+
	C
	C dest                            edi
	C --+---------------+---------------+
	C   |   written     |               |
	C --+---------------+---------------+
	C
	C mm6 = shift
	C mm7 = ecx = 64-shift


	C Size even, destination was unaligned.
	C
	C               source          edx+8
	C                 --+---------------+
	C                   |      mm2      |
	C                 --+---------------+
	C
	C         dest                  edi+4
	C         --+---------------+-------+
	C           |    written    |       |
	C         --+---------------+-------+
	C
	C mm6 = shift+32
	C mm7 = ecx = 64-(shift+32)


	C The movq for the aligned case overwrites the movd for the
	C unaligned case.

	movq	%mm2, %mm0
	psrlq	$32, %mm2

	testb	$32, %cl
	movd	%mm2, disp(4) (%edi)

	jz	L(end_even_unaligned)
	movq	%mm0, disp(0) (%edi)
L(end_even_unaligned):

	movl	SAVE_EDI, %edi
	addl	$SAVE_SIZE, %esp
	emms

	ret

EPILOGUE()