mirror of
https://github.com/jart/cosmopolitan.git
synced 2025-01-31 19:43:32 +00:00
e75ffde09e
You can now build Cosmopolitan with Clang:
make -j8 MODE=llvm
o/llvm/examples/hello.com
The assembler and linker code is now friendly to LLVM too.
So it's not needed to configure Clang to use binutils under
the hood. If you love LLVM then you can now use pure LLVM.
407 lines
21 KiB
ArmAsm
407 lines
21 KiB
ArmAsm
/*-*- mode:unix-assembly; indent-tabs-mode:t; tab-width:8; coding:utf-8 -*-│
|
||
│vi: set et ft=asm ts=8 sw=8 fenc=utf-8 :vi│
|
||
╞══════════════════════════════════════════════════════════════════════════════╡
|
||
│ Copyright 2020 Justine Alexandra Roberts Tunney │
|
||
│ │
|
||
│ Permission to use, copy, modify, and/or distribute this software for │
|
||
│ any purpose with or without fee is hereby granted, provided that the │
|
||
│ above copyright notice and this permission notice appear in all copies. │
|
||
│ │
|
||
│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL │
|
||
│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED │
|
||
│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE │
|
||
│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL │
|
||
│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR │
|
||
│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER │
|
||
│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR │
|
||
│ PERFORMANCE OF THIS SOFTWARE. │
|
||
╚─────────────────────────────────────────────────────────────────────────────*/
|
||
#include "libc/nexgen32e/x86feature.h"
|
||
#include "libc/nexgen32e/macros.h"
|
||
#include "libc/macros.h"
|
||
.source __FILE__
|
||
|
||
// Returns length of NUL-terminated string w/ security blankets.
|
||
//
|
||
// This is like strnlen() except it'll return 0 if (1) RDI is NULL
|
||
// or (2) a NUL-terminator wasn't found in the first RSI bytes.
|
||
//
|
||
// @param rdi is a nullable NUL-terminated string pointer
|
||
// @param rsi is the maximum number of bytes to consider
|
||
// @return rax is the number of bytes, excluding the NUL
|
||
strnlen_s:
|
||
.leafprologue
|
||
.profilable
|
||
xor %eax,%eax
|
||
xor %r10d,%r10d
|
||
test %rdi,%rdi
|
||
jnz 0f
|
||
.leafepilogue
|
||
0: xor %edx,%edx
|
||
mov %rdi,%r8
|
||
// 𝑠𝑙𝑖𝑑𝑒
|
||
.endfn strnlen_s,globl
|
||
|
||
// Swiss army knife of string character scanning.
|
||
// Used to be fourteen fast functions in one.
|
||
//
|
||
// @param rdi is non-null string memory
|
||
// @param rsi is max number of bytes to consider
|
||
// @param dl is search character #1
|
||
// @param dh is search character #2
|
||
// @param r8 is subtracted from result (for length vs. pointer)
|
||
// @param r9 masks result if DH is found (for NUL vs. NULL)
|
||
// @param r10 masks result on bytes exhausted (for length v. NULL)
|
||
// @return rax end pointer after r8/r9/r10 modifications
|
||
strsak: lea -1(%rdi),%rax
|
||
1: add $1,%rax
|
||
sub $1,%rsi
|
||
jb .Lend
|
||
test $31,%al
|
||
jz .Lfast
|
||
.Lbyte: mov (%rax),%cl
|
||
cmp %cl,%dl
|
||
je .Ldone
|
||
cmp %cl,%dh
|
||
je .Lnul
|
||
jmp 1b
|
||
.Ldone: sub %r8,%rax
|
||
jmp .Lret
|
||
.Lend: mov %r10,%r9
|
||
.Lnul: sub %r8,%rax
|
||
and %r9,%rax
|
||
.Lret: .leafepilogue
|
||
.Lslow: add $32,%rsi
|
||
jmp .Lbyte
|
||
.Lfast: movzbl %dl,%ecx
|
||
movd %ecx,%xmm0
|
||
movzbl %dh,%ecx
|
||
movd %ecx,%xmm1
|
||
sub $32,%rax
|
||
#if !X86_NEED(AVX2)
|
||
testb X86_HAVE(AVX2)+kCpuids(%rip)
|
||
jz .Lsse2
|
||
#endif
|
||
vpbroadcastb %xmm0,%ymm0
|
||
vpbroadcastb %xmm1,%ymm1
|
||
1: add $32,%rax
|
||
sub $32,%rsi
|
||
9: jb .Lslow
|
||
vmovdqa (%rax),%ymm2
|
||
vpcmpeqb %ymm0,%ymm2,%ymm3
|
||
vpcmpeqb %ymm1,%ymm2,%ymm2
|
||
vpor %ymm3,%ymm2,%ymm2
|
||
vpmovmskb %ymm2,%ecx
|
||
bsf %ecx,%ecx
|
||
je 1b
|
||
vzeroupper
|
||
2: add %rcx,%rax
|
||
jmp .Lbyte
|
||
#if !X86_NEED(AVX2)
|
||
.Lsse2: pbroadcastb %xmm0
|
||
pbroadcastb %xmm1
|
||
1: add $32,%rax
|
||
sub $32,%rsi
|
||
jb 9b
|
||
movdqa (%rax),%xmm2
|
||
movdqa 16(%rax),%xmm3
|
||
movdqa %xmm3,%xmm4
|
||
pcmpeqb %xmm0,%xmm3
|
||
pcmpeqb %xmm1,%xmm4
|
||
por %xmm4,%xmm3
|
||
pmovmskb %xmm3,%ecx
|
||
shl $16,%ecx
|
||
movdqa %xmm2,%xmm4
|
||
pcmpeqb %xmm0,%xmm2
|
||
pcmpeqb %xmm1,%xmm4
|
||
por %xmm4,%xmm2
|
||
pmovmskb %xmm2,%r11d
|
||
or %r11d,%ecx
|
||
bsf %ecx,%ecx
|
||
je 1b
|
||
jmp 2b
|
||
#endif
|
||
.endfn strsak,globl,hidden
|
||
|
||
/* benchmarked on intel core i7-6700 @ 3.40GHz (skylake)
|
||
includes function call overhead (unless marked otherwise)
|
||
|
||
your strlen, &c (strsak+avx2) for #c per n where c ≈ 0.293ns
|
||
N x1 x8 x64 mBps
|
||
------------------------------------------------------------
|
||
1 47.000 36.375 35.141 99
|
||
1 35.000 34.625 36.234 96
|
||
2 31.500 18.812 18.992 184
|
||
3 19.667 13.042 13.182 265
|
||
4 30.750 10.281 10.285 339
|
||
7 15.857 8.946 7.551 462
|
||
8 12.125 9.203 7.119 490
|
||
15 10.467 5.475 4.601 758
|
||
16 6.812 5.523 4.798 727
|
||
31 5.387 4.327 3.517 992
|
||
32 4.719 1.645 1.532 2278
|
||
63 5.000 2.403 2.034 1715
|
||
64 2.047 0.779 0.788 4427
|
||
127 2.134 1.194 1.027 3399
|
||
128 1.742 0.444 0.419 8327
|
||
255 0.945 0.594 0.554 6295
|
||
256 0.574 0.271 0.264 13226
|
||
511 0.785 0.362 0.307 11384
|
||
512 0.326 0.178 0.151 23134
|
||
1023 0.288 0.242 0.185 18862
|
||
1024 0.208 0.114 0.107 32565
|
||
2047 0.235 0.127 0.123 28430
|
||
2048 0.127 0.090 0.084 41413
|
||
4095 0.119 0.106 0.099 35116
|
||
4096 0.100 0.081 0.079 44372
|
||
8191 0.092 0.082 0.081 43176
|
||
8192 0.081 0.072 0.071 49419
|
||
16383 0.076 0.072 0.071 48847
|
||
16384 0.071 0.068 0.067 52381
|
||
32767 0.072 0.069 0.068 51154
|
||
32768 0.068 0.066 0.065 53409
|
||
|
||
your tinystrlen()
|
||
N x1 x8 x64 mBps
|
||
------------------------------------------------------------
|
||
1 53.000 33.625 33.672 97
|
||
1 33.000 32.125 32.234 101
|
||
2 24.500 19.438 17.711 184
|
||
3 23.667 12.875 11.911 273
|
||
4 13.750 9.281 9.238 352
|
||
7 11.000 6.125 5.801 560
|
||
8 7.625 5.609 5.232 621
|
||
15 11.800 3.825 3.364 966
|
||
16 4.562 3.648 3.173 1024 « optimal
|
||
31 3.710 2.851 2.298 1414
|
||
32 3.031 2.254 2.159 1506 « dropoff
|
||
63 2.683 1.827 1.691 1922
|
||
64 2.078 1.932 1.689 1924
|
||
127 1.630 1.647 1.622 2004
|
||
128 1.727 1.671 1.652 1968
|
||
255 1.392 1.450 1.435 2265
|
||
256 1.473 1.427 1.437 2262
|
||
511 1.325 1.353 1.337 2431
|
||
512 1.408 1.343 1.337 2431
|
||
1023 1.289 1.281 1.287 2525
|
||
1024 1.269 1.295 1.297 2506
|
||
2047 1.269 1.274 1.269 2561
|
||
2048 1.280 1.263 1.281 2538
|
||
4095 1.262 1.270 1.266 2568
|
||
4096 1.270 1.264 1.265 2570
|
||
8191 1.253 1.254 1.254 2592
|
||
8192 1.219 1.224 1.225 2653
|
||
16383 1.225 1.222 1.220 2663
|
||
16384 1.226 1.221 1.222 2659
|
||
32767 1.227 1.224 1.223 2658
|
||
32768 1.220 1.221 1.222 2659
|
||
|
||
glibc strlen for #c per n where c ≈ 0.273ns
|
||
N x1 x8 x64 mBps
|
||
------------------------------------------------------------
|
||
1 3497.000 53.125 42.641 82
|
||
1 69.000 44.875 42.547 82
|
||
2 45.500 24.188 21.852 160
|
||
3 23.000 15.625 14.557 240
|
||
4 22.250 11.406 10.637 328
|
||
7 10.143 6.768 6.230 560
|
||
8 11.125 5.797 5.486 636
|
||
15 5.800 3.142 2.859 1220
|
||
16 7.062 3.070 2.737 1275
|
||
31 2.806 1.585 1.407 2481
|
||
32 3.156 1.574 1.349 2587
|
||
63 2.016 0.895 0.691 5049
|
||
64 1.328 0.744 0.670 5207
|
||
127 1.441 0.521 0.407 8577
|
||
128 0.648 0.454 0.405 8619
|
||
255 0.553 0.286 0.214 16277
|
||
256 0.387 0.235 0.218 15984
|
||
511 0.456 0.151 0.129 27077
|
||
512 0.182 0.134 0.129 27117
|
||
1023 0.171 0.106 0.082 42795
|
||
1024 0.112 0.088 0.082 42741
|
||
2047 0.099 0.069 0.059 59537
|
||
2048 0.072 0.060 0.058 59925
|
||
4095 0.065 0.053 0.047 74122
|
||
4096 0.061 0.048 0.047 74478
|
||
8191 0.048 0.045 0.044 79117
|
||
8192 0.051 0.045 0.044 79181
|
||
16383 0.042 0.040 0.061 57018
|
||
16384 0.069 0.063 0.061 57245
|
||
32767 0.081 0.073 0.068 51426
|
||
32768 0.084 0.072 0.068 51285
|
||
|
||
GCC strlen (-Os REPNZ SCASB) for #c per n where c ≈ 0.293ns
|
||
N x1 x8 x64 mBps
|
||
------------------------------------------------------------
|
||
1 103.000 84.125 88.766 37
|
||
1 81.000 85.125 87.328 37
|
||
2 43.500 44.562 45.508 71
|
||
3 33.000 30.208 30.995 105
|
||
4 24.750 23.156 23.113 141
|
||
7 17.000 13.054 15.355 212
|
||
8 13.375 14.047 13.982 232
|
||
15 9.533 9.258 55.111 59
|
||
16 6.312 6.352 6.364 511
|
||
31 4.032 4.141 4.141 785
|
||
32 3.969 4.059 4.048 803
|
||
63 2.937 2.970 2.995 1086
|
||
64 2.922 2.939 2.956 1100
|
||
127 2.386 2.408 2.403 1353
|
||
128 2.383 2.403 2.401 1354
|
||
255 2.129 2.118 2.124 1530
|
||
256 2.137 2.133 2.130 1526
|
||
511 1.982 1.986 3.351 970
|
||
512 1.982 1.990 1.986 1637
|
||
1023 1.915 1.916 2.587 1257
|
||
1024 1.868 1.867 1.866 1742
|
||
2047 1.835 1.833 1.832 1775
|
||
2048 1.830 1.831 1.832 1775
|
||
4095 1.814 1.814 1.815 1791
|
||
4096 1.810 1.815 1.815 1791
|
||
8191 1.805 1.807 1.806 1800
|
||
8192 1.805 1.806 1.806 1800
|
||
16383 1.803 1.756 1.756 1851
|
||
16384 1.758 1.756 1.756 1851
|
||
32767 1.756 1.754 1.754 1853
|
||
32768 1.756 1.754 1.754 1853
|
||
|
||
Intel Optimz. Manual (SSE4.2) for #c per n where c ≈ 0.273ns
|
||
N x1 x8 x64 mBps
|
||
------------------------------------------------------------
|
||
1 37.000 43.125 34.078 102
|
||
1 33.000 33.875 34.016 103
|
||
2 39.500 17.188 17.555 199
|
||
3 18.333 12.208 12.036 290
|
||
4 30.250 9.344 9.137 382
|
||
7 14.429 5.732 5.766 605
|
||
8 7.875 6.797 5.354 652
|
||
15 10.733 5.825 3.516 993
|
||
16 3.812 2.383 2.325 1501
|
||
31 4.097 2.609 2.079 1678
|
||
32 3.031 1.395 1.349 2587
|
||
63 2.937 1.558 1.079 3235
|
||
64 2.016 0.893 0.690 5056
|
||
127 1.929 0.721 0.607 5745
|
||
128 0.617 0.483 0.428 8147
|
||
255 1.275 0.404 0.411 8486
|
||
256 0.480 0.319 0.299 11681
|
||
511 0.479 0.307 0.288 12127
|
||
512 0.322 0.244 0.232 15013
|
||
1023 0.324 0.224 0.225 15512
|
||
1024 0.245 0.240 0.223 15651
|
||
2047 0.222 0.213 0.206 16938
|
||
2048 0.204 0.194 0.192 18140
|
||
4095 0.204 0.188 0.185 18888
|
||
4096 0.183 0.179 0.179 19446
|
||
8191 0.179 0.176 0.174 20000
|
||
8192 0.174 0.172 0.171 20383
|
||
16383 0.171 0.170 0.169 20604
|
||
16384 0.169 0.169 0.168 20808
|
||
32767 0.213 0.225 0.267 13064
|
||
32768 0.231 0.215 0.220 15852
|
||
|
||
musl libc strlen for #c per n where c ≈ 0.273ns
|
||
N x1 x8 x64 mBps
|
||
------------------------------------------------------------
|
||
1 65.000 36.125 37.984 92
|
||
1 39.000 37.625 37.422 93
|
||
2 41.500 21.938 20.695 169
|
||
3 22.333 17.625 15.859 220
|
||
4 21.250 13.656 12.105 288
|
||
7 22.143 9.018 7.609 459
|
||
8 31.125 7.234 7.346 475
|
||
15 11.267 5.025 4.709 741
|
||
16 9.438 4.039 3.849 907
|
||
31 4.871 3.133 2.488 1402
|
||
32 5.219 2.246 2.039 1712
|
||
63 4.302 1.462 1.407 2479
|
||
64 2.109 1.428 1.155 3023
|
||
127 1.551 1.078 0.879 3971
|
||
128 1.742 0.903 0.760 4591
|
||
255 0.922 0.558 0.605 5764
|
||
256 0.934 0.575 0.537 6495
|
||
511 0.550 0.493 0.455 7674
|
||
512 0.646 0.490 0.426 8183
|
||
1023 0.550 0.439 0.425 8203
|
||
1024 0.472 0.421 0.408 8549
|
||
2047 0.507 0.334 0.373 9360
|
||
2048 0.403 0.426 0.409 8540
|
||
4095 0.391 0.240 0.236 14799
|
||
4096 0.238 0.222 0.221 15766
|
||
8191 0.225 0.223 0.221 15779
|
||
8192 0.225 0.214 0.215 16250
|
||
16383 0.212 0.212 0.210 16595
|
||
16384 0.209 0.210 0.211 16535
|
||
32767 0.214 0.208 0.205 17001
|
||
32768 0.207 0.207 0.291 12002
|
||
|
||
newlib strlen for #c per n where c ≈ 0.273ns
|
||
N x1 x8 x64 mBps
|
||
------------------------------------------------------------
|
||
1 33.000 34.625 34.141 102
|
||
1 33.000 34.125 33.984 103
|
||
2 58.500 18.562 17.508 199
|
||
3 16.333 12.792 12.016 290
|
||
4 19.250 9.219 9.215 379
|
||
7 17.571 6.089 5.685 614
|
||
8 16.625 5.078 5.432 642
|
||
15 8.467 4.042 3.207 1088
|
||
16 3.938 2.773 2.733 1277
|
||
31 3.645 1.673 1.598 2183
|
||
32 3.281 1.527 1.493 2338
|
||
63 2.619 1.042 0.895 3901
|
||
64 1.422 0.928 0.813 4294
|
||
127 0.984 0.718 0.561 6222
|
||
128 1.195 0.591 0.532 6558
|
||
255 0.600 0.404 0.397 8785
|
||
256 0.621 0.429 0.376 9280
|
||
511 0.346 0.311 0.306 11421
|
||
512 0.420 0.308 0.296 11776
|
||
1023 0.284 0.285 0.285 12237
|
||
1024 0.321 0.282 0.280 12456
|
||
2047 0.253 0.252 0.252 13864
|
||
2048 0.260 0.249 0.249 14012
|
||
4095 0.236 0.236 0.236 14811
|
||
4096 0.239 0.235 0.234 14906
|
||
8191 0.233 0.228 0.227 15371
|
||
8192 0.230 0.227 0.227 15397
|
||
16383 0.223 0.224 0.223 15638
|
||
16384 0.223 0.224 0.223 15663
|
||
32767 0.224 0.387 0.225 15527
|
||
32768 0.223 0.222 0.222 15724
|
||
|
||
Agner Fog's strlen (SSE2) for #c per n where c ≈ 0.273ns
|
||
N x1 x8 x64 mBps
|
||
------------------------------------------------------------
|
||
1 59.000 38.375 38.453 91
|
||
1 37.000 38.625 38.234 91
|
||
2 18.500 19.062 19.273 181
|
||
3 13.000 12.792 12.859 271
|
||
4 9.250 9.594 9.660 361
|
||
7 5.286 5.554 5.502 634
|
||
8 4.625 4.703 4.791 728
|
||
15 2.600 2.858 2.622 1331
|
||
16 2.438 2.414 2.421 1442
|
||
31 2.161 1.399 1.290 2706
|
||
32 1.219 1.262 1.250 2793
|
||
63 1.508 0.875 0.693 5038
|
||
64 0.641 0.654 0.655 5328
|
||
127 1.205 0.406 0.379 9200
|
||
128 0.367 0.372 0.369 9463
|
||
255 0.467 0.310 0.235 14835
|
||
256 0.230 0.232 0.232 15034
|
||
511 0.272 0.181 0.159 21918
|
||
512 0.174 0.161 0.158 22148
|
||
1023 0.175 0.134 0.120 29043
|
||
1024 0.140 0.122 0.120 29005
|
||
2047 0.128 0.114 0.112 31205
|
||
2048 0.130 0.113 0.112 31242
|
||
4095 0.105 0.098 0.097 35984
|
||
4096 0.105 0.098 0.097 35973
|
||
8191 0.093 0.090 0.090 38953
|
||
8192 0.094 0.090 0.090 38986
|
||
16383 0.088 0.086 0.086 40648
|
||
16384 0.088 0.086 0.086 40652
|
||
32767 0.088 0.086 0.085 40956
|
||
32768 0.087 0.085 0.085 41114 */
|