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
|
/*
* Copyright (c) 2010-2011, The Linux Foundation. All rights reserved.
*
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 and
* only version 2 as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
* 02110-1301, USA.
*/
/*
* Description
*
* library function for memcpy where length bytes are copied from
* ptr_in to ptr_out. ptr_out is returned unchanged.
* Allows any combination of alignment on input and output pointers
* and length from 0 to 2^32-1
*
* Restrictions
* The arrays should not overlap, the program will produce undefined output
* if they do.
* For blocks less than 16 bytes a byte by byte copy is performed. For
* 8byte alignments, and length multiples, a dword copy is performed up to
* 96bytes
* History
*
* DJH 5/15/09 Initial version 1.0
* DJH 6/ 1/09 Version 1.1 modified ABI to inlcude R16-R19
* DJH 7/12/09 Version 1.2 optimized codesize down to 760 was 840
* DJH 10/14/09 Version 1.3 added special loop for aligned case, was
* overreading bloated codesize back up to 892
* DJH 4/20/10 Version 1.4 fixed Ldword_loop_epilog loop to prevent loads
* occuring if only 1 left outstanding, fixes bug
* # 3888, corrected for all alignments. Peeled off
* 1 32byte chunk from kernel loop and extended 8byte
* loop at end to solve all combinations and prevent
* over read. Fixed Ldword_loop_prolog to prevent
* overread for blocks less than 48bytes. Reduced
* codesize to 752 bytes
* DJH 4/21/10 version 1.5 1.4 fix broke code for input block ends not
* aligned to dword boundaries,underwriting by 1
* byte, added detection for this and fixed. A
* little bloat.
* DJH 4/23/10 version 1.6 corrected stack error, R20 was not being restored
* always, fixed the error of R20 being modified
* before it was being saved
* Natural c model
* ===============
* void * memcpy(char * ptr_out, char * ptr_in, int length) {
* int i;
* if(length) for(i=0; i < length; i++) { ptr_out[i] = ptr_in[i]; }
* return(ptr_out);
* }
*
* Optimized memcpy function
* =========================
* void * memcpy(char * ptr_out, char * ptr_in, int len) {
* int i, prolog, kernel, epilog, mask;
* u8 offset;
* s64 data0, dataF8, data70;
*
* s64 * ptr8_in;
* s64 * ptr8_out;
* s32 * ptr4;
* s16 * ptr2;
*
* offset = ((int) ptr_in) & 7;
* ptr8_in = (s64 *) &ptr_in[-offset]; //read in the aligned pointers
*
* data70 = *ptr8_in++;
* dataF8 = *ptr8_in++;
*
* data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
*
* prolog = 32 - ((int) ptr_out);
* mask = 0x7fffffff >> HEXAGON_R_cl0_R(len);
* prolog = prolog & mask;
* kernel = len - prolog;
* epilog = kernel & 0x1F;
* kernel = kernel>>5;
*
* if (prolog & 1) { ptr_out[0] = (u8) data0; data0 >>= 8; ptr_out += 1;}
* ptr2 = (s16 *) &ptr_out[0];
* if (prolog & 2) { ptr2[0] = (u16) data0; data0 >>= 16; ptr_out += 2;}
* ptr4 = (s32 *) &ptr_out[0];
* if (prolog & 4) { ptr4[0] = (u32) data0; data0 >>= 32; ptr_out += 4;}
*
* offset = offset + (prolog & 7);
* if (offset >= 8) {
* data70 = dataF8;
* dataF8 = *ptr8_in++;
* }
* offset = offset & 0x7;
*
* prolog = prolog >> 3;
* if (prolog) for (i=0; i < prolog; i++) {
* data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
* ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
* data70 = dataF8;
* dataF8 = *ptr8_in++;
* }
* if(kernel) { kernel -= 1; epilog += 32; }
* if(kernel) for(i=0; i < kernel; i++) {
* data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
* ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
* data70 = *ptr8_in++;
*
* data0 = HEXAGON_P_valignb_PPp(data70, dataF8, offset);
* ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
* dataF8 = *ptr8_in++;
*
* data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
* ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
* data70 = *ptr8_in++;
*
* data0 = HEXAGON_P_valignb_PPp(data70, dataF8, offset);
* ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
* dataF8 = *ptr8_in++;
* }
* epilogdws = epilog >> 3;
* if (epilogdws) for (i=0; i < epilogdws; i++) {
* data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
* ptr8_out = (s64 *) &ptr_out[0]; *ptr8_out = data0; ptr_out += 8;
* data70 = dataF8;
* dataF8 = *ptr8_in++;
* }
* data0 = HEXAGON_P_valignb_PPp(dataF8, data70, offset);
*
* ptr4 = (s32 *) &ptr_out[0];
* if (epilog & 4) { ptr4[0] = (u32) data0; data0 >>= 32; ptr_out += 4;}
* ptr2 = (s16 *) &ptr_out[0];
* if (epilog & 2) { ptr2[0] = (u16) data0; data0 >>= 16; ptr_out += 2;}
* if (epilog & 1) { *ptr_out++ = (u8) data0; }
*
* return(ptr_out - length);
* }
*
* Codesize : 784 bytes
*/
#define ptr_out R0 /* destination pounter */
#define ptr_in R1 /* source pointer */
#define len R2 /* length of copy in bytes */
#define data70 R13:12 /* lo 8 bytes of non-aligned transfer */
#define dataF8 R11:10 /* hi 8 bytes of non-aligned transfer */
#define ldata0 R7:6 /* even 8 bytes chunks */
#define ldata1 R25:24 /* odd 8 bytes chunks */
#define data1 R7 /* lower 8 bytes of ldata1 */
#define data0 R6 /* lower 8 bytes of ldata0 */
#define ifbyte p0 /* if transfer has bytes in epilog/prolog */
#define ifhword p0 /* if transfer has shorts in epilog/prolog */
#define ifword p0 /* if transfer has words in epilog/prolog */
#define noprolog p0 /* no prolog, xfer starts at 32byte */
#define nokernel p1 /* no 32byte multiple block in the transfer */
#define noepilog p0 /* no epilog, xfer ends on 32byte boundary */
#define align p2 /* alignment of input rel to 8byte boundary */
#define kernel1 p0 /* kernel count == 1 */
#define dalign R25 /* rel alignment of input to output data */
#define star3 R16 /* number bytes in prolog - dwords */
#define rest R8 /* length - prolog bytes */
#define back R7 /* nr bytes > dword boundary in src block */
#define epilog R3 /* bytes in epilog */
#define inc R15:14 /* inc kernel by -1 and defetch ptr by 32 */
#define kernel R4 /* number of 32byte chunks in kernel */
#define ptr_in_p_128 R5 /* pointer for prefetch of input data */
#define mask R8 /* mask used to determine prolog size */
#define shift R8 /* used to work a shifter to extract bytes */
#define shift2 R5 /* in epilog to workshifter to extract bytes */
#define prolog R15 /* bytes in prolog */
#define epilogdws R15 /* number dwords in epilog */
#define shiftb R14 /* used to extract bytes */
#define offset R9 /* same as align in reg */
#define ptr_out_p_32 R17 /* pointer to output dczero */
#define align888 R14 /* if simple dword loop can be used */
#define len8 R9 /* number of dwords in length */
#define over R20 /* nr of bytes > last inp buf dword boundary */
#define ptr_in_p_128kernel R5:4 /* packed fetch pointer & kernel cnt */
.section .text
.p2align 4
.global memcpy
.type memcpy, @function
memcpy:
{
p2 = cmp.eq(len, #0); /* =0 */
align888 = or(ptr_in, ptr_out); /* %8 < 97 */
p0 = cmp.gtu(len, #23); /* %1, <24 */
p1 = cmp.eq(ptr_in, ptr_out); /* attempt to overwrite self */
}
{
p1 = or(p2, p1);
p3 = cmp.gtu(len, #95); /* %8 < 97 */
align888 = or(align888, len); /* %8 < 97 */
len8 = lsr(len, #3); /* %8 < 97 */
}
{
dcfetch(ptr_in); /* zero/ptrin=ptrout causes fetch */
p2 = bitsclr(align888, #7); /* %8 < 97 */
if(p1) jumpr r31; /* =0 */
}
{
p2 = and(p2,!p3); /* %8 < 97 */
if (p2.new) len = add(len, #-8); /* %8 < 97 */
if (p2.new) jump:NT .Ldwordaligned; /* %8 < 97 */
}
{
if(!p0) jump .Lbytes23orless; /* %1, <24 */
mask.l = #LO(0x7fffffff);
/* all bytes before line multiples of data */
prolog = sub(#0, ptr_out);
}
{
/* save r31 on stack, decrement sp by 16 */
allocframe(#24);
mask.h = #HI(0x7fffffff);
ptr_in_p_128 = add(ptr_in, #32);
back = cl0(len);
}
{
memd(sp+#0) = R17:16; /* save r16,r17 on stack6 */
r31.l = #LO(.Lmemcpy_return); /* set up final return pointer */
prolog &= lsr(mask, back);
offset = and(ptr_in, #7);
}
{
memd(sp+#8) = R25:24; /* save r25,r24 on stack */
dalign = sub(ptr_out, ptr_in);
r31.h = #HI(.Lmemcpy_return); /* set up final return pointer */
}
{
/* see if there if input buffer end if aligned */
over = add(len, ptr_in);
back = add(len, offset);
memd(sp+#16) = R21:20; /* save r20,r21 on stack */
}
{
noprolog = bitsclr(prolog, #7);
prolog = and(prolog, #31);
dcfetch(ptr_in_p_128);
ptr_in_p_128 = add(ptr_in_p_128, #32);
}
{
kernel = sub(len, prolog);
shift = asl(prolog, #3);
star3 = and(prolog, #7);
ptr_in = and(ptr_in, #-8);
}
{
prolog = lsr(prolog, #3);
epilog = and(kernel, #31);
ptr_out_p_32 = add(ptr_out, prolog);
over = and(over, #7);
}
{
p3 = cmp.gtu(back, #8);
kernel = lsr(kernel, #5);
dcfetch(ptr_in_p_128);
ptr_in_p_128 = add(ptr_in_p_128, #32);
}
{
p1 = cmp.eq(prolog, #0);
if(!p1.new) prolog = add(prolog, #1);
dcfetch(ptr_in_p_128); /* reserve the line 64bytes on */
ptr_in_p_128 = add(ptr_in_p_128, #32);
}
{
nokernel = cmp.eq(kernel,#0);
dcfetch(ptr_in_p_128); /* reserve the line 64bytes on */
ptr_in_p_128 = add(ptr_in_p_128, #32);
shiftb = and(shift, #8);
}
{
dcfetch(ptr_in_p_128); /* reserve the line 64bytes on */
ptr_in_p_128 = add(ptr_in_p_128, #32);
if(nokernel) jump .Lskip64;
p2 = cmp.eq(kernel, #1); /* skip ovr if kernel == 0 */
}
{
dczeroa(ptr_out_p_32);
/* don't advance pointer */
if(!p2) ptr_out_p_32 = add(ptr_out_p_32, #32);
}
{
dalign = and(dalign, #31);
dczeroa(ptr_out_p_32);
}
.Lskip64:
{
data70 = memd(ptr_in++#16);
if(p3) dataF8 = memd(ptr_in+#8);
if(noprolog) jump .Lnoprolog32;
align = offset;
}
/* upto initial 7 bytes */
{
ldata0 = valignb(dataF8, data70, align);
ifbyte = tstbit(shift,#3);
offset = add(offset, star3);
}
{
if(ifbyte) memb(ptr_out++#1) = data0;
ldata0 = lsr(ldata0, shiftb);
shiftb = and(shift, #16);
ifhword = tstbit(shift,#4);
}
{
if(ifhword) memh(ptr_out++#2) = data0;
|