| 1 |
#!/usr/bin/env perl |
| 2 |
# |
| 3 |
# ==================================================================== |
| 4 |
# Written by Andy Polyakov <[email protected]> for the OpenSSL |
| 5 |
# project. The module is, however, dual licensed under OpenSSL and |
| 6 |
# CRYPTOGAMS licenses depending on where you obtain it. For further |
| 7 |
# details see http://www.openssl.org/~appro/cryptogams/. |
| 8 |
# ==================================================================== |
| 9 |
# |
| 10 |
# January 2015 |
| 11 |
# |
| 12 |
# ChaCha20 for x86. |
| 13 |
# |
| 14 |
# Performance in cycles per byte out of large buffer. |
| 15 |
# |
| 16 |
# 1xIALU/gcc 4xSSSE3 |
| 17 |
# Pentium 17.5/+80% |
| 18 |
# PIII 14.2/+60% |
| 19 |
# P4 18.6/+84% |
| 20 |
# Core2 9.56/+89% 4.83 |
| 21 |
# Westmere 9.50/+45% 3.35 |
| 22 |
# Sandy Bridge 10.5/+47% 3.20 |
| 23 |
# Haswell 8.15/+50% 2.83 |
| 24 |
# Silvermont 17.4/+36% 8.35 |
| 25 |
# Sledgehammer 10.2/+54% |
| 26 |
# Bulldozer 13.4/+50% 4.38(*) |
| 27 |
# |
| 28 |
# (*) Bulldozer actually executes 4xXOP code path that delivers 3.55; |
| 29 |
# |
| 30 |
# Modified from upstream OpenSSL to remove the XOP code. |
| 31 |
|
| 32 |
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; |
| 33 |
push(@INC,"${dir}","${dir}../../perlasm"); |
| 34 |
require "x86asm.pl"; |
| 35 |
|
| 36 |
&asm_init($ARGV[0],"chacha-x86.pl",$ARGV[$#ARGV] eq "386"); |
| 37 |
|
| 38 |
$xmm=$ymm=0; |
| 39 |
for (@ARGV) { $xmm=1 if (/-DOPENSSL_IA32_SSE2/); } |
| 40 |
|
| 41 |
$ymm=$xmm; |
| 42 |
|
| 43 |
$a="eax"; |
| 44 |
($b,$b_)=("ebx","ebp"); |
| 45 |
($c,$c_)=("ecx","esi"); |
| 46 |
($d,$d_)=("edx","edi"); |
| 47 |
|
| 48 |
sub QUARTERROUND { |
| 49 |
my ($ai,$bi,$ci,$di,$i)=@_; |
| 50 |
my ($an,$bn,$cn,$dn)=map(($_&~3)+(($_+1)&3),($ai,$bi,$ci,$di)); # next |
| 51 |
my ($ap,$bp,$cp,$dp)=map(($_&~3)+(($_-1)&3),($ai,$bi,$ci,$di)); # previous |
| 52 |
|
| 53 |
# a b c d |
| 54 |
# |
| 55 |
# 0 4 8 12 < even round |
| 56 |
# 1 5 9 13 |
| 57 |
# 2 6 10 14 |
| 58 |
# 3 7 11 15 |
| 59 |
# 0 5 10 15 < odd round |
| 60 |
# 1 6 11 12 |
| 61 |
# 2 7 8 13 |
| 62 |
# 3 4 9 14 |
| 63 |
|
| 64 |
if ($i==0) { |
| 65 |
my $j=4; |
| 66 |
($ap,$bp,$cp,$dp)=map(($_&~3)+(($_-$j--)&3),($ap,$bp,$cp,$dp)); |
| 67 |
} elsif ($i==3) { |
| 68 |
my $j=0; |
| 69 |
($an,$bn,$cn,$dn)=map(($_&~3)+(($_+$j++)&3),($an,$bn,$cn,$dn)); |
| 70 |
} elsif ($i==4) { |
| 71 |
my $j=4; |
| 72 |
($ap,$bp,$cp,$dp)=map(($_&~3)+(($_+$j--)&3),($ap,$bp,$cp,$dp)); |
| 73 |
} elsif ($i==7) { |
| 74 |
my $j=0; |
| 75 |
($an,$bn,$cn,$dn)=map(($_&~3)+(($_-$j++)&3),($an,$bn,$cn,$dn)); |
| 76 |
} |
| 77 |
|
| 78 |
#&add ($a,$b); # see elsewhere |
| 79 |
&xor ($d,$a); |
| 80 |
&mov (&DWP(4*$cp,"esp"),$c_) if ($ai>0 && $ai<3); |
| 81 |
&rol ($d,16); |
| 82 |
&mov (&DWP(4*$bp,"esp"),$b_) if ($i!=0); |
| 83 |
&add ($c,$d); |
| 84 |
&mov ($c_,&DWP(4*$cn,"esp")) if ($ai>0 && $ai<3); |
| 85 |
&xor ($b,$c); |
| 86 |
&mov ($d_,&DWP(4*$dn,"esp")) if ($di!=$dn); |
| 87 |
&rol ($b,12); |
| 88 |
&mov ($b_,&DWP(4*$bn,"esp")) if ($i<7); |
| 89 |
&mov ($b_,&DWP(128,"esp")) if ($i==7); # loop counter |
| 90 |
&add ($a,$b); |
| 91 |
&xor ($d,$a); |
| 92 |
&mov (&DWP(4*$ai,"esp"),$a); |
| 93 |
&rol ($d,8); |
| 94 |
&mov ($a,&DWP(4*$an,"esp")); |
| 95 |
&add ($c,$d); |
| 96 |
&mov (&DWP(4*$di,"esp"),$d) if ($di!=$dn); |
| 97 |
&mov ($d_,$d) if ($di==$dn); |
| 98 |
&xor ($b,$c); |
| 99 |
&add ($a,$b_) if ($i<7); # elsewhere |
| 100 |
&rol ($b,7); |
| 101 |
|
| 102 |
($b,$b_)=($b_,$b); |
| 103 |
($c,$c_)=($c_,$c); |
| 104 |
($d,$d_)=($d_,$d); |
| 105 |
} |
| 106 |
|
| 107 |
&static_label("ssse3_shortcut"); |
| 108 |
&static_label("ssse3_data"); |
| 109 |
&static_label("pic_point"); |
| 110 |
|
| 111 |
&function_begin("ChaCha20_ctr32"); |
| 112 |
&xor ("eax","eax"); |
| 113 |
&cmp ("eax",&wparam(2)); # len==0? |
| 114 |
&je (&label("no_data")); |
| 115 |
if ($xmm) { |
| 116 |
&call (&label("pic_point")); |
| 117 |
&set_label("pic_point"); |
| 118 |
&blindpop("eax"); |
| 119 |
&picmeup("ebp","OPENSSL_ia32cap_P","eax",&label("pic_point")); |
| 120 |
&test (&DWP(0,"ebp"),1<<24); # test FXSR bit |
| 121 |
&jz (&label("x86")); |
| 122 |
&test (&DWP(4,"ebp"),1<<9); # test SSSE3 bit |
| 123 |
&jz (&label("x86")); |
| 124 |
&jmp (&label("ssse3_shortcut")); |
| 125 |
&set_label("x86"); |
| 126 |
} |
| 127 |
&mov ("esi",&wparam(3)); # key |
| 128 |
&mov ("edi",&wparam(4)); # counter and nonce |
| 129 |
|
| 130 |
&stack_push(33); |
| 131 |
|
| 132 |
&mov ("eax",&DWP(4*0,"esi")); # copy key |
| 133 |
&mov ("ebx",&DWP(4*1,"esi")); |
| 134 |
&mov ("ecx",&DWP(4*2,"esi")); |
| 135 |
&mov ("edx",&DWP(4*3,"esi")); |
| 136 |
&mov (&DWP(64+4*4,"esp"),"eax"); |
| 137 |
&mov (&DWP(64+4*5,"esp"),"ebx"); |
| 138 |
&mov (&DWP(64+4*6,"esp"),"ecx"); |
| 139 |
&mov (&DWP(64+4*7,"esp"),"edx"); |
| 140 |
&mov ("eax",&DWP(4*4,"esi")); |
| 141 |
&mov ("ebx",&DWP(4*5,"esi")); |
| 142 |
&mov ("ecx",&DWP(4*6,"esi")); |
| 143 |
&mov ("edx",&DWP(4*7,"esi")); |
| 144 |
&mov (&DWP(64+4*8,"esp"),"eax"); |
| 145 |
&mov (&DWP(64+4*9,"esp"),"ebx"); |
| 146 |
&mov (&DWP(64+4*10,"esp"),"ecx"); |
| 147 |
&mov (&DWP(64+4*11,"esp"),"edx"); |
| 148 |
&mov ("eax",&DWP(4*0,"edi")); # copy counter and nonce |
| 149 |
&mov ("ebx",&DWP(4*1,"edi")); |
| 150 |
&mov ("ecx",&DWP(4*2,"edi")); |
| 151 |
&mov ("edx",&DWP(4*3,"edi")); |
| 152 |
&sub ("eax",1); |
| 153 |
&mov (&DWP(64+4*12,"esp"),"eax"); |
| 154 |
&mov (&DWP(64+4*13,"esp"),"ebx"); |
| 155 |
&mov (&DWP(64+4*14,"esp"),"ecx"); |
| 156 |
&mov (&DWP(64+4*15,"esp"),"edx"); |
| 157 |
&jmp (&label("entry")); |
| 158 |
|
| 159 |
&set_label("outer_loop",16); |
| 160 |
&mov (&wparam(1),$b); # save input |
| 161 |
&mov (&wparam(0),$a); # save output |
| 162 |
&mov (&wparam(2),$c); # save len |
| 163 |
&set_label("entry"); |
| 164 |
&mov ($a,0x61707865); |
| 165 |
&mov (&DWP(4*1,"esp"),0x3320646e); |
| 166 |
&mov (&DWP(4*2,"esp"),0x79622d32); |
| 167 |
&mov (&DWP(4*3,"esp"),0x6b206574); |
| 168 |
|
| 169 |
&mov ($b, &DWP(64+4*5,"esp")); # copy key material |
| 170 |
&mov ($b_,&DWP(64+4*6,"esp")); |
| 171 |
&mov ($c, &DWP(64+4*10,"esp")); |
| 172 |
&mov ($c_,&DWP(64+4*11,"esp")); |
| 173 |
&mov ($d, &DWP(64+4*13,"esp")); |
| 174 |
&mov ($d_,&DWP(64+4*14,"esp")); |
| 175 |
&mov (&DWP(4*5,"esp"),$b); |
| 176 |
&mov (&DWP(4*6,"esp"),$b_); |
| 177 |
&mov (&DWP(4*10,"esp"),$c); |
| 178 |
&mov (&DWP(4*11,"esp"),$c_); |
| 179 |
&mov (&DWP(4*13,"esp"),$d); |
| 180 |
&mov (&DWP(4*14,"esp"),$d_); |
| 181 |
|
| 182 |
&mov ($b, &DWP(64+4*7,"esp")); |
| 183 |
&mov ($d_,&DWP(64+4*15,"esp")); |
| 184 |
&mov ($d, &DWP(64+4*12,"esp")); |
| 185 |
&mov ($b_,&DWP(64+4*4,"esp")); |
| 186 |
&mov ($c, &DWP(64+4*8,"esp")); |
| 187 |
&mov ($c_,&DWP(64+4*9,"esp")); |
| 188 |
&add ($d,1); # counter value |
| 189 |
&mov (&DWP(4*7,"esp"),$b); |
| 190 |
&mov (&DWP(4*15,"esp"),$d_); |
| 191 |
&mov (&DWP(64+4*12,"esp"),$d); # save counter value |
| 192 |
|
| 193 |
&mov ($b,10); # loop counter |
| 194 |
&jmp (&label("loop")); |
| 195 |
|
| 196 |
&set_label("loop",16); |
| 197 |
&add ($a,$b_); # elsewhere |
| 198 |
&mov (&DWP(128,"esp"),$b); # save loop counter |
| 199 |
&mov ($b,$b_); |
| 200 |
&QUARTERROUND(0, 4, 8, 12, 0); |
| 201 |
&QUARTERROUND(1, 5, 9, 13, 1); |
| 202 |
&QUARTERROUND(2, 6,10, 14, 2); |
| 203 |
&QUARTERROUND(3, 7,11, 15, 3); |
| 204 |
&QUARTERROUND(0, 5,10, 15, 4); |
| 205 |
&QUARTERROUND(1, 6,11, 12, 5); |
| 206 |
&QUARTERROUND(2, 7, 8, 13, 6); |
| 207 |
&QUARTERROUND(3, 4, 9, 14, 7); |
| 208 |
&dec ($b); |
| 209 |
&jnz (&label("loop")); |
| 210 |
|
| 211 |
&mov ($b,&wparam(2)); # load len |
| 212 |
|
| 213 |
&add ($a,0x61707865); # accumulate key material |
| 214 |
&add ($b_,&DWP(64+4*4,"esp")); |
| 215 |
&add ($c, &DWP(64+4*8,"esp")); |
| 216 |
&add ($c_,&DWP(64+4*9,"esp")); |
| 217 |
|
| 218 |
&cmp ($b,64); |
| 219 |
&jb (&label("tail")); |
| 220 |
|
| 221 |
&mov ($b,&wparam(1)); # load input pointer |
| 222 |
&add ($d, &DWP(64+4*12,"esp")); |
| 223 |
&add ($d_,&DWP(64+4*14,"esp")); |
| 224 |
|
| 225 |
&xor ($a, &DWP(4*0,$b)); # xor with input |
| 226 |
&xor ($b_,&DWP(4*4,$b)); |
| 227 |
&mov (&DWP(4*0,"esp"),$a); |
| 228 |
&mov ($a,&wparam(0)); # load output pointer |
| 229 |
&xor ($c, &DWP(4*8,$b)); |
| 230 |
&xor ($c_,&DWP(4*9,$b)); |
| 231 |
&xor ($d, &DWP(4*12,$b)); |
| 232 |
&xor ($d_,&DWP(4*14,$b)); |
| 233 |
&mov (&DWP(4*4,$a),$b_); # write output |
| 234 |
&mov (&DWP(4*8,$a),$c); |
| 235 |
&mov (&DWP(4*9,$a),$c_); |
| 236 |
&mov (&DWP(4*12,$a),$d); |
| 237 |
&mov (&DWP(4*14,$a),$d_); |
| 238 |
|
| 239 |
&mov ($b_,&DWP(4*1,"esp")); |
| 240 |
&mov ($c, &DWP(4*2,"esp")); |
| 241 |
&mov ($c_,&DWP(4*3,"esp")); |
| 242 |
&mov ($d, &DWP(4*5,"esp")); |
| 243 |
&mov ($d_,&DWP(4*6,"esp")); |
| 244 |
&add ($b_,0x3320646e); # accumulate key material |
| 245 |
&add ($c, 0x79622d32); |
| 246 |
&add ($c_,0x6b206574); |
| 247 |
&add ($d, &DWP(64+4*5,"esp")); |
| 248 |
&add ($d_,&DWP(64+4*6,"esp")); |
| 249 |
&xor ($b_,&DWP(4*1,$b)); |
| 250 |
&xor ($c, &DWP(4*2,$b)); |
| 251 |
&xor ($c_,&DWP(4*3,$b)); |
| 252 |
&xor ($d, &DWP(4*5,$b)); |
| 253 |
&xor ($d_,&DWP(4*6,$b)); |
| 254 |
&mov (&DWP(4*1,$a),$b_); |
| 255 |
&mov (&DWP(4*2,$a),$c); |
| 256 |
&mov (&DWP(4*3,$a),$c_); |
| 257 |
&mov (&DWP(4*5,$a),$d); |
| 258 |
&mov (&DWP(4*6,$a),$d_); |
| 259 |
|
| 260 |
&mov ($b_,&DWP(4*7,"esp")); |
| 261 |
&mov ($c, &DWP(4*10,"esp")); |
| 262 |
&mov ($c_,&DWP(4*11,"esp")); |
| 263 |
&mov ($d, &DWP(4*13,"esp")); |
| 264 |
&mov ($d_,&DWP(4*15,"esp")); |
| 265 |
&add ($b_,&DWP(64+4*7,"esp")); |
| 266 |
&add ($c, &DWP(64+4*10,"esp")); |
| 267 |
&add ($c_,&DWP(64+4*11,"esp")); |
| 268 |
&add ($d, &DWP(64+4*13,"esp")); |
| 269 |
&add ($d_,&DWP(64+4*15,"esp")); |
| 270 |
&xor ($b_,&DWP(4*7,$b)); |
| 271 |
&xor ($c, &DWP(4*10,$b)); |
| 272 |
&xor ($c_,&DWP(4*11,$b)); |
| 273 |
&xor ($d, &DWP(4*13,$b)); |
| 274 |
&xor ($d_,&DWP(4*15,$b)); |
| 275 |
&lea ($b,&DWP(4*16,$b)); |
| 276 |
&mov (&DWP(4*7,$a),$b_); |
| 277 |
&mov ($b_,&DWP(4*0,"esp")); |
| 278 |
&mov (&DWP(4*10,$a),$c); |
| 279 |
&mov ($c,&wparam(2)); # len |
| 280 |
&mov (&DWP(4*11,$a),$c_); |
| 281 |
&mov (&DWP(4*13,$a),$d); |
| 282 |
&mov (&DWP(4*15,$a),$d_); |
| 283 |
&mov (&DWP(4*0,$a),$b_); |
| 284 |
&lea ($a,&DWP(4*16,$a)); |
| 285 |
&sub ($c,64); |
| 286 |
&jnz (&label("outer_loop")); |
| 287 |
|
| 288 |
&jmp (&label("done")); |
| 289 |
|
| 290 |
&set_label("tail"); |
| 291 |
&add ($d, &DWP(64+4*12,"esp")); |
| 292 |
&add ($d_,&DWP(64+4*14,"esp")); |
| 293 |
&mov (&DWP(4*0,"esp"),$a); |
| 294 |
&mov (&DWP(4*4,"esp"),$b_); |
| 295 |
&mov (&DWP(4*8,"esp"),$c); |
| 296 |
&mov (&DWP(4*9,"esp"),$c_); |
| 297 |
&mov (&DWP(4*12,"esp"),$d); |
| 298 |
&mov (&DWP(4*14,"esp"),$d_); |
| 299 |
|
| 300 |
&mov ($b_,&DWP(4*1,"esp")); |
| 301 |
&mov ($c, &DWP(4*2,"esp")); |
| 302 |
&mov ($c_,&DWP(4*3,"esp")); |
| 303 |
&mov ($d, &DWP(4*5,"esp")); |
| 304 |
&mov ($d_,&DWP(4*6,"esp")); |
| 305 |
&add ($b_,0x3320646e); # accumulate key material |
| 306 |
&add ($c, 0x79622d32); |
| 307 |
&add ($c_,0x6b206574); |
| 308 |
&add ($d, &DWP(64+4*5,"esp")); |
| 309 |
&add ($d_,&DWP(64+4*6,"esp")); |
| 310 |
&mov (&DWP(4*1,"esp"),$b_); |
| 311 |
&mov (&DWP(4*2,"esp"),$c); |
| 312 |
&mov (&DWP(4*3,"esp"),$c_); |
| 313 |
&mov (&DWP(4*5,"esp"),$d); |
| 314 |
&mov (&DWP(4*6,"esp"),$d_); |
| 315 |
|
| 316 |
&mov ($b_,&DWP(4*7,"esp")); |
| 317 |
&mov ($c, &DWP(4*10,"esp")); |
| 318 |
&mov ($c_,&DWP(4*11,"esp")); |
| 319 |
&mov ($d, &DWP(4*13,"esp")); |
| 320 |
&mov ($d_,&DWP(4*15,"esp")); |
| 321 |
&add ($b_,&DWP(64+4*7,"esp")); |
| 322 |
&add ($c, &DWP(64+4*10,"esp")); |
| 323 |
&add ($c_,&DWP(64+4*11,"esp")); |
| 324 |
&add ($d, &DWP(64+4*13,"esp")); |
| 325 |
&add ($d_,&DWP(64+4*15,"esp")); |
| 326 |
&mov (&DWP(4*7,"esp"),$b_); |
| 327 |
&mov ($b_,&wparam(1)); # load input |
| 328 |
&mov (&DWP(4*10,"esp"),$c); |
| 329 |
&mov ($c,&wparam(0)); # load output |
| 330 |
&mov (&DWP(4*11,"esp"),$c_); |
| 331 |
&xor ($c_,$c_); |
| 332 |
&mov (&DWP(4*13,"esp"),$d); |
| 333 |
&mov (&DWP(4*15,"esp"),$d_); |
| 334 |
|
| 335 |
&xor ("eax","eax"); |
| 336 |
&xor ("edx","edx"); |
| 337 |
&set_label("tail_loop"); |
| 338 |
&movb ("al",&BP(0,$c_,$b_)); |
| 339 |
&movb ("dl",&BP(0,"esp",$c_)); |
| 340 |
&lea ($c_,&DWP(1,$c_)); |
| 341 |
&xor ("al","dl"); |
| 342 |
&mov (&BP(-1,$c,$c_),"al"); |
| 343 |
&dec ($b); |
| 344 |
&jnz (&label("tail_loop")); |
| 345 |
|
| 346 |
&set_label("done"); |
| 347 |
&stack_pop(33); |
| 348 |
&set_label("no_data"); |
| 349 |
&function_end("ChaCha20_ctr32"); |
| 350 |
|
| 351 |
if ($xmm) { |
| 352 |
my ($xa,$xa_,$xb,$xb_,$xc,$xc_,$xd,$xd_)=map("xmm$_",(0..7)); |
| 353 |
my ($out,$inp,$len)=("edi","esi","ecx"); |
| 354 |
|
| 355 |
sub QUARTERROUND_SSSE3 { |
| 356 |
my ($ai,$bi,$ci,$di,$i)=@_; |
| 357 |
my ($an,$bn,$cn,$dn)=map(($_&~3)+(($_+1)&3),($ai,$bi,$ci,$di)); # next |
| 358 |
my ($ap,$bp,$cp,$dp)=map(($_&~3)+(($_-1)&3),($ai,$bi,$ci,$di)); # previous |
| 359 |
|
| 360 |
# a b c d |
| 361 |
# |
| 362 |
# 0 4 8 12 < even round |
| 363 |
# 1 5 9 13 |
| 364 |
# 2 6 10 14 |
| 365 |
# 3 7 11 15 |
| 366 |
# 0 5 10 15 < odd round |
| 367 |
# 1 6 11 12 |
| 368 |
# 2 7 8 13 |
| 369 |
# 3 4 9 14 |
| 370 |
|
| 371 |
if ($i==0) { |
| 372 |
my $j=4; |
| 373 |
($ap,$bp,$cp,$dp)=map(($_&~3)+(($_-$j--)&3),($ap,$bp,$cp,$dp)); |
| 374 |
} elsif ($i==3) { |
| 375 |
my $j=0; |
| 376 |
($an,$bn,$cn,$dn)=map(($_&~3)+(($_+$j++)&3),($an,$bn,$cn,$dn)); |
| 377 |
} elsif ($i==4) { |
| 378 |
my $j=4; |
| 379 |
($ap,$bp,$cp,$dp)=map(($_&~3)+(($_+$j--)&3),($ap,$bp,$cp,$dp)); |
| 380 |
} elsif ($i==7) { |
| 381 |
my $j=0; |
| 382 |
($an,$bn,$cn,$dn)=map(($_&~3)+(($_-$j++)&3),($an,$bn,$cn,$dn)); |
| 383 |
} |
| 384 |
|
| 385 |
#&paddd ($xa,$xb); # see elsewhere |
| 386 |
#&pxor ($xd,$xa); # see elsewhere |
| 387 |
&movdqa(&QWP(16*$cp-128,"ebx"),$xc_) if ($ai>0 && $ai<3); |
| 388 |
&pshufb ($xd,&QWP(0,"eax")); # rot16 |
| 389 |
&movdqa(&QWP(16*$bp-128,"ebx"),$xb_) if ($i!=0); |
| 390 |
&paddd ($xc,$xd); |
| 391 |
&movdqa($xc_,&QWP(16*$cn-128,"ebx")) if ($ai>0 && $ai<3); |
| 392 |
&pxor ($xb,$xc); |
| 393 |
&movdqa($xb_,&QWP(16*$bn-128,"ebx")) if ($i<7); |
| 394 |
&movdqa ($xa_,$xb); # borrow as temporary |
| 395 |
&pslld ($xb,12); |
| 396 |
&psrld ($xa_,20); |
| 397 |
&por ($xb,$xa_); |
| 398 |
&movdqa($xa_,&QWP(16*$an-128,"ebx")); |
| 399 |
&paddd ($xa,$xb); |
| 400 |
&movdqa($xd_,&QWP(16*$dn-128,"ebx")) if ($di!=$dn); |
| 401 |
&pxor ($xd,$xa); |
| 402 |
&movdqa (&QWP(16*$ai-128,"ebx"),$xa); |
| 403 |
&pshufb ($xd,&QWP(16,"eax")); # rot8 |
| 404 |
&paddd ($xc,$xd); |
| 405 |
&movdqa (&QWP(16*$di-128,"ebx"),$xd) if ($di!=$dn); |
| 406 |
&movdqa ($xd_,$xd) if ($di==$dn); |
| 407 |
&pxor ($xb,$xc); |
| 408 |
&paddd ($xa_,$xb_) if ($i<7); # elsewhere |
| 409 |
&movdqa ($xa,$xb); # borrow as temporary |
| 410 |
&pslld ($xb,7); |
| 411 |
&psrld ($xa,25); |
| 412 |
&pxor ($xd_,$xa_) if ($i<7); # elsewhere |
| 413 |
&por ($xb,$xa); |
| 414 |
|
| 415 |
($xa,$xa_)=($xa_,$xa); |
| 416 |
($xb,$xb_)=($xb_,$xb); |
| 417 |
($xc,$xc_)=($xc_,$xc); |
| 418 |
($xd,$xd_)=($xd_,$xd); |
| 419 |
} |
| 420 |
|
| 421 |
&function_begin("ChaCha20_ssse3"); |
| 422 |
&set_label("ssse3_shortcut"); |
| 423 |
&mov ($out,&wparam(0)); |
| 424 |
&mov ($inp,&wparam(1)); |
| 425 |
&mov ($len,&wparam(2)); |
| 426 |
&mov ("edx",&wparam(3)); # key |
| 427 |
&mov ("ebx",&wparam(4)); # counter and nonce |
| 428 |
|
| 429 |
&mov ("ebp","esp"); |
| 430 |
&stack_push (131); |
| 431 |
&and ("esp",-64); |
| 432 |
&mov (&DWP(512,"esp"),"ebp"); |
| 433 |
|
| 434 |
&lea ("eax",&DWP(&label("ssse3_data")."-". |
| 435 |
&label("pic_point"),"eax")); |
| 436 |
&movdqu ("xmm3",&QWP(0,"ebx")); # counter and nonce |
| 437 |
|
| 438 |
&cmp ($len,64*4); |
| 439 |
&jb (&label("1x")); |
| 440 |
|
| 441 |
&mov (&DWP(512+4,"esp"),"edx"); # offload pointers |
| 442 |
&mov (&DWP(512+8,"esp"),"ebx"); |
| 443 |
&sub ($len,64*4); # bias len |
| 444 |
&lea ("ebp",&DWP(256+128,"esp")); # size optimization |
| 445 |
|
| 446 |
&movdqu ("xmm7",&QWP(0,"edx")); # key |
| 447 |
&pshufd ("xmm0","xmm3",0x00); |
| 448 |
&pshufd ("xmm1","xmm3",0x55); |
| 449 |
&pshufd ("xmm2","xmm3",0xaa); |
| 450 |
&pshufd ("xmm3","xmm3",0xff); |
| 451 |
&paddd ("xmm0",&QWP(16*3,"eax")); # fix counters |
| 452 |
&pshufd ("xmm4","xmm7",0x00); |
| 453 |
&pshufd ("xmm5","xmm7",0x55); |
| 454 |
&psubd ("xmm0",&QWP(16*4,"eax")); |
| 455 |
&pshufd ("xmm6","xmm7",0xaa); |
| 456 |
&pshufd ("xmm7","xmm7",0xff); |
| 457 |
&movdqa (&QWP(16*12-128,"ebp"),"xmm0"); |
| 458 |
&movdqa (&QWP(16*13-128,"ebp"),"xmm1"); |
| 459 |
&movdqa (&QWP(16*14-128,"ebp"),"xmm2"); |
| 460 |
&movdqa (&QWP(16*15-128,"ebp"),"xmm3"); |
| 461 |
&movdqu ("xmm3",&QWP(16,"edx")); # key |
| 462 |
&movdqa (&QWP(16*4-128,"ebp"),"xmm4"); |
| 463 |
&movdqa (&QWP(16*5-128,"ebp"),"xmm5"); |
| 464 |
&movdqa (&QWP(16*6-128,"ebp"),"xmm6"); |
| 465 |
&movdqa (&QWP(16*7-128,"ebp"),"xmm7"); |
| 466 |
&movdqa ("xmm7",&QWP(16*2,"eax")); # sigma |
| 467 |
&lea ("ebx",&DWP(128,"esp")); # size optimization |
| 468 |
|
| 469 |
&pshufd ("xmm0","xmm3",0x00); |
| 470 |
&pshufd ("xmm1","xmm3",0x55); |
| 471 |
&pshufd ("xmm2","xmm3",0xaa); |
| 472 |
&pshufd ("xmm3","xmm3",0xff); |
| 473 |
&pshufd ("xmm4","xmm7",0x00); |
| 474 |
&pshufd ("xmm5","xmm7",0x55); |
| 475 |
&pshufd ("xmm6","xmm7",0xaa); |
| 476 |
&pshufd ("xmm7","xmm7",0xff); |
| 477 |
&movdqa (&QWP(16*8-128,"ebp"),"xmm0"); |
| 478 |
&movdqa (&QWP(16*9-128,"ebp"),"xmm1"); |
| 479 |
&movdqa (&QWP(16*10-128,"ebp"),"xmm2"); |
| 480 |
&movdqa (&QWP(16*11-128,"ebp"),"xmm3"); |
| 481 |
&movdqa (&QWP(16*0-128,"ebp"),"xmm4"); |
| 482 |
&movdqa (&QWP(16*1-128,"ebp"),"xmm5"); |
| 483 |
&movdqa (&QWP(16*2-128,"ebp"),"xmm6"); |
| 484 |
&movdqa (&QWP(16*3-128,"ebp"),"xmm7"); |
| 485 |
|
| 486 |
&lea ($inp,&DWP(128,$inp)); # size optimization |
| 487 |
&lea ($out,&DWP(128,$out)); # size optimization |
| 488 |
&jmp (&label("outer_loop")); |
| 489 |
|
| 490 |
&set_label("outer_loop",16); |
| 491 |
#&movdqa ("xmm0",&QWP(16*0-128,"ebp")); # copy key material |
| 492 |
&movdqa ("xmm1",&QWP(16*1-128,"ebp")); |
| 493 |
&movdqa ("xmm2",&QWP(16*2-128,"ebp")); |
| 494 |
&movdqa ("xmm3",&QWP(16*3-128,"ebp")); |
| 495 |
#&movdqa ("xmm4",&QWP(16*4-128,"ebp")); |
| 496 |
&movdqa ("xmm5",&QWP(16*5-128,"ebp")); |
| 497 |
&movdqa ("xmm6",&QWP(16*6-128,"ebp")); |
| 498 |
&movdqa ("xmm7",&QWP(16*7-128,"ebp")); |
| 499 |
#&movdqa (&QWP(16*0-128,"ebx"),"xmm0"); |
| 500 |
&movdqa (&QWP(16*1-128,"ebx"),"xmm1"); |
| 501 |
&movdqa (&QWP(16*2-128,"ebx"),"xmm2"); |
| 502 |
&movdqa (&QWP(16*3-128,"ebx"),"xmm3"); |
| 503 |
#&movdqa (&QWP(16*4-128,"ebx"),"xmm4"); |
| 504 |
&movdqa (&QWP(16*5-128,"ebx"),"xmm5"); |
| 505 |
&movdqa (&QWP(16*6-128,"ebx"),"xmm6"); |
| 506 |
&movdqa (&QWP(16*7-128,"ebx"),"xmm7"); |
| 507 |
#&movdqa ("xmm0",&QWP(16*8-128,"ebp")); |
| 508 |
#&movdqa ("xmm1",&QWP(16*9-128,"ebp")); |
| 509 |
&movdqa ("xmm2",&QWP(16*10-128,"ebp")); |
| 510 |
&movdqa ("xmm3",&QWP(16*11-128,"ebp")); |
| 511 |
&movdqa ("xmm4",&QWP(16*12-128,"ebp")); |
| 512 |
&movdqa ("xmm5",&QWP(16*13-128,"ebp")); |
| 513 |
&movdqa ("xmm6",&QWP(16*14-128,"ebp")); |
| 514 |
&movdqa ("xmm7",&QWP(16*15-128,"ebp")); |
| 515 |
&paddd ("xmm4",&QWP(16*4,"eax")); # counter value |
| 516 |
#&movdqa (&QWP(16*8-128,"ebx"),"xmm0"); |
| 517 |
#&movdqa (&QWP(16*9-128,"ebx"),"xmm1"); |
| 518 |
&movdqa (&QWP(16*10-128,"ebx"),"xmm2"); |
| 519 |
&movdqa (&QWP(16*11-128,"ebx"),"xmm3"); |
| 520 |
&movdqa (&QWP(16*12-128,"ebx"),"xmm4"); |
| 521 |
&movdqa (&QWP(16*13-128,"ebx"),"xmm5"); |
| 522 |
&movdqa (&QWP(16*14-128,"ebx"),"xmm6"); |
| 523 |
&movdqa (&QWP(16*15-128,"ebx"),"xmm7"); |
| 524 |
&movdqa (&QWP(16*12-128,"ebp"),"xmm4"); # save counter value |
| 525 |
|
| 526 |
&movdqa ($xa, &QWP(16*0-128,"ebp")); |
| 527 |
&movdqa ($xd, "xmm4"); |
| 528 |
&movdqa ($xb_,&QWP(16*4-128,"ebp")); |
| 529 |
&movdqa ($xc, &QWP(16*8-128,"ebp")); |
| 530 |
&movdqa ($xc_,&QWP(16*9-128,"ebp")); |
| 531 |
|
| 532 |
&mov ("edx",10); # loop counter |
| 533 |
&nop (); |
| 534 |
|
| 535 |
&set_label("loop",16); |
| 536 |
&paddd ($xa,$xb_); # elsewhere |
| 537 |
&movdqa ($xb,$xb_); |
| 538 |
&pxor ($xd,$xa); # elsewhere |
| 539 |
&QUARTERROUND_SSSE3(0, 4, 8, 12, 0); |
| 540 |
&QUARTERROUND_SSSE3(1, 5, 9, 13, 1); |
| 541 |
&QUARTERROUND_SSSE3(2, 6,10, 14, 2); |
| 542 |
&QUARTERROUND_SSSE3(3, 7,11, 15, 3); |
| 543 |
&QUARTERROUND_SSSE3(0, 5,10, 15, 4); |
| 544 |
&QUARTERROUND_SSSE3(1, 6,11, 12, 5); |
| 545 |
&QUARTERROUND_SSSE3(2, 7, 8, 13, 6); |
| 546 |
&QUARTERROUND_SSSE3(3, 4, 9, 14, 7); |
| 547 |
&dec ("edx"); |
| 548 |
&jnz (&label("loop")); |
| 549 |
|
| 550 |
&movdqa (&QWP(16*4-128,"ebx"),$xb_); |
| 551 |
&movdqa (&QWP(16*8-128,"ebx"),$xc); |
| 552 |
&movdqa (&QWP(16*9-128,"ebx"),$xc_); |
| 553 |
&movdqa (&QWP(16*12-128,"ebx"),$xd); |
| 554 |
&movdqa (&QWP(16*14-128,"ebx"),$xd_); |
| 555 |
|
| 556 |
my ($xa0,$xa1,$xa2,$xa3,$xt0,$xt1,$xt2,$xt3)=map("xmm$_",(0..7)); |
| 557 |
|
| 558 |
#&movdqa ($xa0,&QWP(16*0-128,"ebx")); # it's there |
| 559 |
&movdqa ($xa1,&QWP(16*1-128,"ebx")); |
| 560 |
&movdqa ($xa2,&QWP(16*2-128,"ebx")); |
| 561 |
&movdqa ($xa3,&QWP(16*3-128,"ebx")); |
| 562 |
|
| 563 |
for($i=0;$i<256;$i+=64) { |
| 564 |
&paddd ($xa0,&QWP($i+16*0-128,"ebp")); # accumulate key material |
| 565 |
&paddd ($xa1,&QWP($i+16*1-128,"ebp")); |
| 566 |
&paddd ($xa2,&QWP($i+16*2-128,"ebp")); |
| 567 |
&paddd ($xa3,&QWP($i+16*3-128,"ebp")); |
| 568 |
|
| 569 |
&movdqa ($xt2,$xa0); # "de-interlace" data |
| 570 |
&punpckldq ($xa0,$xa1); |
| 571 |
&movdqa ($xt3,$xa2); |
| 572 |
&punpckldq ($xa2,$xa3); |
| 573 |
&punpckhdq ($xt2,$xa1); |
| 574 |
&punpckhdq ($xt3,$xa3); |
| 575 |
&movdqa ($xa1,$xa0); |
| 576 |
&punpcklqdq ($xa0,$xa2); # "a0" |
| 577 |
&movdqa ($xa3,$xt2); |
| 578 |
&punpcklqdq ($xt2,$xt3); # "a2" |
| 579 |
&punpckhqdq ($xa1,$xa2); # "a1" |
| 580 |
&punpckhqdq ($xa3,$xt3); # "a3" |
| 581 |
|
| 582 |
#($xa2,$xt2)=($xt2,$xa2); |
| 583 |
|
| 584 |
&movdqu ($xt0,&QWP(64*0-128,$inp)); # load input |
| 585 |
&movdqu ($xt1,&QWP(64*1-128,$inp)); |
| 586 |
&movdqu ($xa2,&QWP(64*2-128,$inp)); |
| 587 |
&movdqu ($xt3,&QWP(64*3-128,$inp)); |
| 588 |
&lea ($inp,&QWP($i<192?16:(64*4-16*3),$inp)); |
| 589 |
&pxor ($xt0,$xa0); |
| 590 |
&movdqa ($xa0,&QWP($i+16*4-128,"ebx")) if ($i<192); |
| 591 |
&pxor ($xt1,$xa1); |
| 592 |
&movdqa ($xa1,&QWP($i+16*5-128,"ebx")) if ($i<192); |
| 593 |
&pxor ($xt2,$xa2); |
| 594 |
&movdqa ($xa2,&QWP($i+16*6-128,"ebx")) if ($i<192); |
| 595 |
&pxor ($xt3,$xa3); |
| 596 |
&movdqa ($xa3,&QWP($i+16*7-128,"ebx")) if ($i<192); |
| 597 |
&movdqu (&QWP(64*0-128,$out),$xt0); # store output |
| 598 |
&movdqu (&QWP(64*1-128,$out),$xt1); |
| 599 |
&movdqu (&QWP(64*2-128,$out),$xt2); |
| 600 |
&movdqu (&QWP(64*3-128,$out),$xt3); |
| 601 |
&lea ($out,&QWP($i<192?16:(64*4-16*3),$out)); |
| 602 |
} |
| 603 |
&sub ($len,64*4); |
| 604 |
&jnc (&label("outer_loop")); |
| 605 |
|
| 606 |
&add ($len,64*4); |
| 607 |
&jz (&label("done")); |
| 608 |
|
| 609 |
&mov ("ebx",&DWP(512+8,"esp")); # restore pointers |
| 610 |
&lea ($inp,&DWP(-128,$inp)); |
| 611 |
&mov ("edx",&DWP(512+4,"esp")); |
| 612 |
&lea ($out,&DWP(-128,$out)); |
| 613 |
|
| 614 |
&movd ("xmm2",&DWP(16*12-128,"ebp")); # counter value |
| 615 |
&movdqu ("xmm3",&QWP(0,"ebx")); |
| 616 |
&paddd ("xmm2",&QWP(16*6,"eax")); # +four |
| 617 |
&pand ("xmm3",&QWP(16*7,"eax")); |
| 618 |
&por ("xmm3","xmm2"); # counter value |
| 619 |
{ |
| 620 |
my ($a,$b,$c,$d,$t,$t1,$rot16,$rot24)=map("xmm$_",(0..7)); |
| 621 |
|
| 622 |
sub SSSE3ROUND { # critical path is 20 "SIMD ticks" per round |
| 623 |
&paddd ($a,$b); |
| 624 |
&pxor ($d,$a); |
| 625 |
&pshufb ($d,$rot16); |
| 626 |
|
| 627 |
&paddd ($c,$d); |
| 628 |
&pxor ($b,$c); |
| 629 |
&movdqa ($t,$b); |
| 630 |
&psrld ($b,20); |
| 631 |
&pslld ($t,12); |
| 632 |
&por ($b,$t); |
| 633 |
|
| 634 |
&paddd ($a,$b); |
| 635 |
&pxor ($d,$a); |
| 636 |
&pshufb ($d,$rot24); |
| 637 |
|
| 638 |
&paddd ($c,$d); |
| 639 |
&pxor ($b,$c); |
| 640 |
&movdqa ($t,$b); |
| 641 |
&psrld ($b,25); |
| 642 |
&pslld ($t,7); |
| 643 |
&por ($b,$t); |
| 644 |
} |
| 645 |
|
| 646 |
&set_label("1x"); |
| 647 |
&movdqa ($a,&QWP(16*2,"eax")); # sigma |
| 648 |
&movdqu ($b,&QWP(0,"edx")); |
| 649 |
&movdqu ($c,&QWP(16,"edx")); |
| 650 |
#&movdqu ($d,&QWP(0,"ebx")); # already loaded |
| 651 |
&movdqa ($rot16,&QWP(0,"eax")); |
| 652 |
&movdqa ($rot24,&QWP(16,"eax")); |
| 653 |
&mov (&DWP(16*3,"esp"),"ebp"); |
| 654 |
|
| 655 |
&movdqa (&QWP(16*0,"esp"),$a); |
| 656 |
&movdqa (&QWP(16*1,"esp"),$b); |
| 657 |
&movdqa (&QWP(16*2,"esp"),$c); |
| 658 |
&movdqa (&QWP(16*3,"esp"),$d); |
| 659 |
&mov ("edx",10); |
| 660 |
&jmp (&label("loop1x")); |
| 661 |
|
| 662 |
&set_label("outer1x",16); |
| 663 |
&movdqa ($d,&QWP(16*5,"eax")); # one |
| 664 |
&movdqa ($a,&QWP(16*0,"esp")); |
| 665 |
&movdqa ($b,&QWP(16*1,"esp")); |
| 666 |
&movdqa ($c,&QWP(16*2,"esp")); |
| 667 |
&paddd ($d,&QWP(16*3,"esp")); |
| 668 |
&mov ("edx",10); |
| 669 |
&movdqa (&QWP(16*3,"esp"),$d); |
| 670 |
&jmp (&label("loop1x")); |
| 671 |
|
| 672 |
&set_label("loop1x",16); |
| 673 |
&SSSE3ROUND(); |
| 674 |
&pshufd ($c,$c,0b01001110); |
| 675 |
&pshufd ($b,$b,0b00111001); |
| 676 |
&pshufd ($d,$d,0b10010011); |
| 677 |
&nop (); |
| 678 |
|
| 679 |
&SSSE3ROUND(); |
| 680 |
&pshufd ($c,$c,0b01001110); |
| 681 |
&pshufd ($b,$b,0b10010011); |
| 682 |
&pshufd ($d,$d,0b00111001); |
| 683 |
|
| 684 |
&dec ("edx"); |
| 685 |
&jnz (&label("loop1x")); |
| 686 |
|
| 687 |
&paddd ($a,&QWP(16*0,"esp")); |
| 688 |
&paddd ($b,&QWP(16*1,"esp")); |
| 689 |
&paddd ($c,&QWP(16*2,"esp")); |
| 690 |
&paddd ($d,&QWP(16*3,"esp")); |
| 691 |
|
| 692 |
&cmp ($len,64); |
| 693 |
&jb (&label("tail")); |
| 694 |
|
| 695 |
&movdqu ($t,&QWP(16*0,$inp)); |
| 696 |
&movdqu ($t1,&QWP(16*1,$inp)); |
| 697 |
&pxor ($a,$t); # xor with input |
| 698 |
&movdqu ($t,&QWP(16*2,$inp)); |
| 699 |
&pxor ($b,$t1); |
| 700 |
&movdqu ($t1,&QWP(16*3,$inp)); |
| 701 |
&pxor ($c,$t); |
| 702 |
&pxor ($d,$t1); |
| 703 |
&lea ($inp,&DWP(16*4,$inp)); # inp+=64 |
| 704 |
|
| 705 |
&movdqu (&QWP(16*0,$out),$a); # write output |
| 706 |
&movdqu (&QWP(16*1,$out),$b); |
| 707 |
&movdqu (&QWP(16*2,$out),$c); |
| 708 |
&movdqu (&QWP(16*3,$out),$d); |
| 709 |
&lea ($out,&DWP(16*4,$out)); # inp+=64 |
| 710 |
|
| 711 |
&sub ($len,64); |
| 712 |
&jnz (&label("outer1x")); |
| 713 |
|
| 714 |
&jmp (&label("done")); |
| 715 |
|
| 716 |
&set_label("tail"); |
| 717 |
&movdqa (&QWP(16*0,"esp"),$a); |
| 718 |
&movdqa (&QWP(16*1,"esp"),$b); |
| 719 |
&movdqa (&QWP(16*2,"esp"),$c); |
| 720 |
&movdqa (&QWP(16*3,"esp"),$d); |
| 721 |
|
| 722 |
&xor ("eax","eax"); |
| 723 |
&xor ("edx","edx"); |
| 724 |
&xor ("ebp","ebp"); |
| 725 |
|
| 726 |
&set_label("tail_loop"); |
| 727 |
&movb ("al",&BP(0,"esp","ebp")); |
| 728 |
&movb ("dl",&BP(0,$inp,"ebp")); |
| 729 |
&lea ("ebp",&DWP(1,"ebp")); |
| 730 |
&xor ("al","dl"); |
| 731 |
&movb (&BP(-1,$out,"ebp"),"al"); |
| 732 |
&dec ($len); |
| 733 |
&jnz (&label("tail_loop")); |
| 734 |
} |
| 735 |
&set_label("done"); |
| 736 |
&mov ("esp",&DWP(512,"esp")); |
| 737 |
&function_end("ChaCha20_ssse3"); |
| 738 |
|
| 739 |
&align (64); |
| 740 |
&set_label("ssse3_data"); |
| 741 |
&data_byte(0x2,0x3,0x0,0x1, 0x6,0x7,0x4,0x5, 0xa,0xb,0x8,0x9, 0xe,0xf,0xc,0xd); |
| 742 |
&data_byte(0x3,0x0,0x1,0x2, 0x7,0x4,0x5,0x6, 0xb,0x8,0x9,0xa, 0xf,0xc,0xd,0xe); |
| 743 |
&data_word(0x61707865,0x3320646e,0x79622d32,0x6b206574); |
| 744 |
&data_word(0,1,2,3); |
| 745 |
&data_word(4,4,4,4); |
| 746 |
&data_word(1,0,0,0); |
| 747 |
&data_word(4,0,0,0); |
| 748 |
&data_word(0,-1,-1,-1); |
| 749 |
&align (64); |
| 750 |
} |
| 751 |
&asciz ("ChaCha20 for x86, CRYPTOGAMS by <appro\@openssl.org>"); |
| 752 |
|
| 753 |
&asm_finish(); |