| 1 | #! /usr/bin/env perl
|
|---|
| 2 | # Copyright 2005-2020 The OpenSSL Project Authors. All Rights Reserved.
|
|---|
| 3 | #
|
|---|
| 4 | # Licensed under the Apache License 2.0 (the "License"). You may not use
|
|---|
| 5 | # this file except in compliance with the License. You can obtain a copy
|
|---|
| 6 | # in the file LICENSE in the source distribution or at
|
|---|
| 7 | # https://www.openssl.org/source/license.html
|
|---|
| 8 |
|
|---|
| 9 |
|
|---|
| 10 | # ====================================================================
|
|---|
| 11 | # Written by Andy Polyakov <[email protected]> for the OpenSSL
|
|---|
| 12 | # project. The module is, however, dual licensed under OpenSSL and
|
|---|
| 13 | # CRYPTOGAMS licenses depending on where you obtain it. For further
|
|---|
| 14 | # details see http://www.openssl.org/~appro/cryptogams/.
|
|---|
| 15 | # ====================================================================
|
|---|
| 16 |
|
|---|
| 17 | # October 2005
|
|---|
| 18 | #
|
|---|
| 19 | # This is a "teaser" code, as it can be improved in several ways...
|
|---|
| 20 | # First of all non-SSE2 path should be implemented (yes, for now it
|
|---|
| 21 | # performs Montgomery multiplication/convolution only on SSE2-capable
|
|---|
| 22 | # CPUs such as P4, others fall down to original code). Then inner loop
|
|---|
| 23 | # can be unrolled and modulo-scheduled to improve ILP and possibly
|
|---|
| 24 | # moved to 128-bit XMM register bank (though it would require input
|
|---|
| 25 | # rearrangement and/or increase bus bandwidth utilization). Dedicated
|
|---|
| 26 | # squaring procedure should give further performance improvement...
|
|---|
| 27 | # Yet, for being draft, the code improves rsa512 *sign* benchmark by
|
|---|
| 28 | # 110%(!), rsa1024 one - by 70% and rsa4096 - by 20%:-)
|
|---|
| 29 |
|
|---|
| 30 | # December 2006
|
|---|
| 31 | #
|
|---|
| 32 | # Modulo-scheduling SSE2 loops results in further 15-20% improvement.
|
|---|
| 33 | # Integer-only code [being equipped with dedicated squaring procedure]
|
|---|
| 34 | # gives ~40% on rsa512 sign benchmark...
|
|---|
| 35 |
|
|---|
| 36 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
|
|---|
| 37 | push(@INC,"${dir}","${dir}../../perlasm");
|
|---|
| 38 | require "x86asm.pl";
|
|---|
| 39 |
|
|---|
| 40 | $output = pop and open STDOUT,">$output";
|
|---|
| 41 |
|
|---|
| 42 | &asm_init($ARGV[0]);
|
|---|
| 43 |
|
|---|
| 44 | $sse2=0;
|
|---|
| 45 | for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
|
|---|
| 46 |
|
|---|
| 47 | &external_label("OPENSSL_ia32cap_P") if ($sse2);
|
|---|
| 48 |
|
|---|
| 49 | &function_begin("bn_mul_mont");
|
|---|
| 50 |
|
|---|
| 51 | $i="edx";
|
|---|
| 52 | $j="ecx";
|
|---|
| 53 | $ap="esi"; $tp="esi"; # overlapping variables!!!
|
|---|
| 54 | $rp="edi"; $bp="edi"; # overlapping variables!!!
|
|---|
| 55 | $np="ebp";
|
|---|
| 56 | $num="ebx";
|
|---|
| 57 |
|
|---|
| 58 | $_num=&DWP(4*0,"esp"); # stack top layout
|
|---|
| 59 | $_rp=&DWP(4*1,"esp");
|
|---|
| 60 | $_ap=&DWP(4*2,"esp");
|
|---|
| 61 | $_bp=&DWP(4*3,"esp");
|
|---|
| 62 | $_np=&DWP(4*4,"esp");
|
|---|
| 63 | $_n0=&DWP(4*5,"esp"); $_n0q=&QWP(4*5,"esp");
|
|---|
| 64 | $_sp=&DWP(4*6,"esp");
|
|---|
| 65 | $_bpend=&DWP(4*7,"esp");
|
|---|
| 66 | $frame=32; # size of above frame rounded up to 16n
|
|---|
| 67 |
|
|---|
| 68 | &xor ("eax","eax");
|
|---|
| 69 | &mov ("edi",&wparam(5)); # int num
|
|---|
| 70 | &cmp ("edi",4);
|
|---|
| 71 | &jl (&label("just_leave"));
|
|---|
| 72 |
|
|---|
| 73 | &lea ("esi",&wparam(0)); # put aside pointer to argument block
|
|---|
| 74 | &lea ("edx",&wparam(1)); # load ap
|
|---|
| 75 | &add ("edi",2); # extra two words on top of tp
|
|---|
| 76 | &neg ("edi");
|
|---|
| 77 | &lea ("ebp",&DWP(-$frame,"esp","edi",4)); # future alloca($frame+4*(num+2))
|
|---|
| 78 | &neg ("edi");
|
|---|
| 79 |
|
|---|
| 80 | # minimize cache contention by arranging 2K window between stack
|
|---|
| 81 | # pointer and ap argument [np is also position sensitive vector,
|
|---|
| 82 | # but it's assumed to be near ap, as it's allocated at ~same
|
|---|
| 83 | # time].
|
|---|
| 84 | &mov ("eax","ebp");
|
|---|
| 85 | &sub ("eax","edx");
|
|---|
| 86 | &and ("eax",2047);
|
|---|
| 87 | &sub ("ebp","eax"); # this aligns sp and ap modulo 2048
|
|---|
| 88 |
|
|---|
| 89 | &xor ("edx","ebp");
|
|---|
| 90 | &and ("edx",2048);
|
|---|
| 91 | &xor ("edx",2048);
|
|---|
| 92 | &sub ("ebp","edx"); # this splits them apart modulo 4096
|
|---|
| 93 |
|
|---|
| 94 | &and ("ebp",-64); # align to cache line
|
|---|
| 95 |
|
|---|
| 96 | # An OS-agnostic version of __chkstk.
|
|---|
| 97 | #
|
|---|
| 98 | # Some OSes (Windows) insist on stack being "wired" to
|
|---|
| 99 | # physical memory in strictly sequential manner, i.e. if stack
|
|---|
| 100 | # allocation spans two pages, then reference to farmost one can
|
|---|
| 101 | # be punishable by SEGV. But page walking can do good even on
|
|---|
| 102 | # other OSes, because it guarantees that villain thread hits
|
|---|
| 103 | # the guard page before it can make damage to innocent one...
|
|---|
| 104 | &mov ("eax","esp");
|
|---|
| 105 | &sub ("eax","ebp");
|
|---|
| 106 | &and ("eax",-4096);
|
|---|
| 107 | &mov ("edx","esp"); # saved stack pointer!
|
|---|
| 108 | &lea ("esp",&DWP(0,"ebp","eax"));
|
|---|
| 109 | &mov ("eax",&DWP(0,"esp"));
|
|---|
| 110 | &cmp ("esp","ebp");
|
|---|
| 111 | &ja (&label("page_walk"));
|
|---|
| 112 | &jmp (&label("page_walk_done"));
|
|---|
| 113 |
|
|---|
| 114 | &set_label("page_walk",16);
|
|---|
| 115 | &lea ("esp",&DWP(-4096,"esp"));
|
|---|
| 116 | &mov ("eax",&DWP(0,"esp"));
|
|---|
| 117 | &cmp ("esp","ebp");
|
|---|
| 118 | &ja (&label("page_walk"));
|
|---|
| 119 | &set_label("page_walk_done");
|
|---|
| 120 |
|
|---|
| 121 | ################################# load argument block...
|
|---|
| 122 | &mov ("eax",&DWP(0*4,"esi"));# BN_ULONG *rp
|
|---|
| 123 | &mov ("ebx",&DWP(1*4,"esi"));# const BN_ULONG *ap
|
|---|
| 124 | &mov ("ecx",&DWP(2*4,"esi"));# const BN_ULONG *bp
|
|---|
| 125 | &mov ("ebp",&DWP(3*4,"esi"));# const BN_ULONG *np
|
|---|
| 126 | &mov ("esi",&DWP(4*4,"esi"));# const BN_ULONG *n0
|
|---|
| 127 | #&mov ("edi",&DWP(5*4,"esi"));# int num
|
|---|
| 128 |
|
|---|
| 129 | &mov ("esi",&DWP(0,"esi")); # pull n0[0]
|
|---|
| 130 | &mov ($_rp,"eax"); # ... save a copy of argument block
|
|---|
| 131 | &mov ($_ap,"ebx");
|
|---|
| 132 | &mov ($_bp,"ecx");
|
|---|
| 133 | &mov ($_np,"ebp");
|
|---|
| 134 | &mov ($_n0,"esi");
|
|---|
| 135 | &lea ($num,&DWP(-3,"edi")); # num=num-1 to assist modulo-scheduling
|
|---|
| 136 | #&mov ($_num,$num); # redundant as $num is not reused
|
|---|
| 137 | &mov ($_sp,"edx"); # saved stack pointer!
|
|---|
| 138 | |
|---|
| 139 |
|
|---|
| 140 | if($sse2) {
|
|---|
| 141 | $acc0="mm0"; # mmx register bank layout
|
|---|
| 142 | $acc1="mm1";
|
|---|
| 143 | $car0="mm2";
|
|---|
| 144 | $car1="mm3";
|
|---|
| 145 | $mul0="mm4";
|
|---|
| 146 | $mul1="mm5";
|
|---|
| 147 | $temp="mm6";
|
|---|
| 148 | $mask="mm7";
|
|---|
| 149 |
|
|---|
| 150 | &picmeup("eax","OPENSSL_ia32cap_P");
|
|---|
| 151 | &bt (&DWP(0,"eax"),26);
|
|---|
| 152 | &jnc (&label("non_sse2"));
|
|---|
| 153 |
|
|---|
| 154 | &mov ("eax",-1);
|
|---|
| 155 | &movd ($mask,"eax"); # mask 32 lower bits
|
|---|
| 156 |
|
|---|
| 157 | &mov ($ap,$_ap); # load input pointers
|
|---|
| 158 | &mov ($bp,$_bp);
|
|---|
| 159 | &mov ($np,$_np);
|
|---|
| 160 |
|
|---|
| 161 | &xor ($i,$i); # i=0
|
|---|
| 162 | &xor ($j,$j); # j=0
|
|---|
| 163 |
|
|---|
| 164 | &movd ($mul0,&DWP(0,$bp)); # bp[0]
|
|---|
| 165 | &movd ($mul1,&DWP(0,$ap)); # ap[0]
|
|---|
| 166 | &movd ($car1,&DWP(0,$np)); # np[0]
|
|---|
| 167 |
|
|---|
| 168 | &pmuludq($mul1,$mul0); # ap[0]*bp[0]
|
|---|
| 169 | &movq ($car0,$mul1);
|
|---|
| 170 | &movq ($acc0,$mul1); # I wish movd worked for
|
|---|
| 171 | &pand ($acc0,$mask); # inter-register transfers
|
|---|
| 172 |
|
|---|
| 173 | &pmuludq($mul1,$_n0q); # *=n0
|
|---|
| 174 |
|
|---|
| 175 | &pmuludq($car1,$mul1); # "t[0]"*np[0]*n0
|
|---|
| 176 | &paddq ($car1,$acc0);
|
|---|
| 177 |
|
|---|
| 178 | &movd ($acc1,&DWP(4,$np)); # np[1]
|
|---|
| 179 | &movd ($acc0,&DWP(4,$ap)); # ap[1]
|
|---|
| 180 |
|
|---|
| 181 | &psrlq ($car0,32);
|
|---|
| 182 | &psrlq ($car1,32);
|
|---|
| 183 |
|
|---|
| 184 | &inc ($j); # j++
|
|---|
| 185 | &set_label("1st",16);
|
|---|
| 186 | &pmuludq($acc0,$mul0); # ap[j]*bp[0]
|
|---|
| 187 | &pmuludq($acc1,$mul1); # np[j]*m1
|
|---|
| 188 | &paddq ($car0,$acc0); # +=c0
|
|---|
| 189 | &paddq ($car1,$acc1); # +=c1
|
|---|
| 190 |
|
|---|
| 191 | &movq ($acc0,$car0);
|
|---|
| 192 | &pand ($acc0,$mask);
|
|---|
| 193 | &movd ($acc1,&DWP(4,$np,$j,4)); # np[j+1]
|
|---|
| 194 | &paddq ($car1,$acc0); # +=ap[j]*bp[0];
|
|---|
| 195 | &movd ($acc0,&DWP(4,$ap,$j,4)); # ap[j+1]
|
|---|
| 196 | &psrlq ($car0,32);
|
|---|
| 197 | &movd (&DWP($frame-4,"esp",$j,4),$car1); # tp[j-1]=
|
|---|
| 198 | &psrlq ($car1,32);
|
|---|
| 199 |
|
|---|
| 200 | &lea ($j,&DWP(1,$j));
|
|---|
| 201 | &cmp ($j,$num);
|
|---|
| 202 | &jl (&label("1st"));
|
|---|
| 203 |
|
|---|
| 204 | &pmuludq($acc0,$mul0); # ap[num-1]*bp[0]
|
|---|
| 205 | &pmuludq($acc1,$mul1); # np[num-1]*m1
|
|---|
| 206 | &paddq ($car0,$acc0); # +=c0
|
|---|
| 207 | &paddq ($car1,$acc1); # +=c1
|
|---|
| 208 |
|
|---|
| 209 | &movq ($acc0,$car0);
|
|---|
| 210 | &pand ($acc0,$mask);
|
|---|
| 211 | &paddq ($car1,$acc0); # +=ap[num-1]*bp[0];
|
|---|
| 212 | &movd (&DWP($frame-4,"esp",$j,4),$car1); # tp[num-2]=
|
|---|
| 213 |
|
|---|
| 214 | &psrlq ($car0,32);
|
|---|
| 215 | &psrlq ($car1,32);
|
|---|
| 216 |
|
|---|
| 217 | &paddq ($car1,$car0);
|
|---|
| 218 | &movq (&QWP($frame,"esp",$num,4),$car1); # tp[num].tp[num-1]
|
|---|
| 219 | |
|---|
| 220 |
|
|---|
| 221 | &inc ($i); # i++
|
|---|
| 222 | &set_label("outer");
|
|---|
| 223 | &xor ($j,$j); # j=0
|
|---|
| 224 |
|
|---|
| 225 | &movd ($mul0,&DWP(0,$bp,$i,4)); # bp[i]
|
|---|
| 226 | &movd ($mul1,&DWP(0,$ap)); # ap[0]
|
|---|
| 227 | &movd ($temp,&DWP($frame,"esp")); # tp[0]
|
|---|
| 228 | &movd ($car1,&DWP(0,$np)); # np[0]
|
|---|
| 229 | &pmuludq($mul1,$mul0); # ap[0]*bp[i]
|
|---|
| 230 |
|
|---|
| 231 | &paddq ($mul1,$temp); # +=tp[0]
|
|---|
| 232 | &movq ($acc0,$mul1);
|
|---|
| 233 | &movq ($car0,$mul1);
|
|---|
| 234 | &pand ($acc0,$mask);
|
|---|
| 235 |
|
|---|
| 236 | &pmuludq($mul1,$_n0q); # *=n0
|
|---|
| 237 |
|
|---|
| 238 | &pmuludq($car1,$mul1);
|
|---|
| 239 | &paddq ($car1,$acc0);
|
|---|
| 240 |
|
|---|
| 241 | &movd ($temp,&DWP($frame+4,"esp")); # tp[1]
|
|---|
| 242 | &movd ($acc1,&DWP(4,$np)); # np[1]
|
|---|
| 243 | &movd ($acc0,&DWP(4,$ap)); # ap[1]
|
|---|
| 244 |
|
|---|
| 245 | &psrlq ($car0,32);
|
|---|
| 246 | &psrlq ($car1,32);
|
|---|
| 247 | &paddq ($car0,$temp); # +=tp[1]
|
|---|
| 248 |
|
|---|
| 249 | &inc ($j); # j++
|
|---|
| 250 | &dec ($num);
|
|---|
| 251 | &set_label("inner");
|
|---|
| 252 | &pmuludq($acc0,$mul0); # ap[j]*bp[i]
|
|---|
| 253 | &pmuludq($acc1,$mul1); # np[j]*m1
|
|---|
| 254 | &paddq ($car0,$acc0); # +=c0
|
|---|
| 255 | &paddq ($car1,$acc1); # +=c1
|
|---|
| 256 |
|
|---|
| 257 | &movq ($acc0,$car0);
|
|---|
| 258 | &movd ($temp,&DWP($frame+4,"esp",$j,4));# tp[j+1]
|
|---|
| 259 | &pand ($acc0,$mask);
|
|---|
| 260 | &movd ($acc1,&DWP(4,$np,$j,4)); # np[j+1]
|
|---|
| 261 | &paddq ($car1,$acc0); # +=ap[j]*bp[i]+tp[j]
|
|---|
| 262 | &movd ($acc0,&DWP(4,$ap,$j,4)); # ap[j+1]
|
|---|
| 263 | &psrlq ($car0,32);
|
|---|
| 264 | &movd (&DWP($frame-4,"esp",$j,4),$car1);# tp[j-1]=
|
|---|
| 265 | &psrlq ($car1,32);
|
|---|
| 266 | &paddq ($car0,$temp); # +=tp[j+1]
|
|---|
| 267 |
|
|---|
| 268 | &dec ($num);
|
|---|
| 269 | &lea ($j,&DWP(1,$j)); # j++
|
|---|
| 270 | &jnz (&label("inner"));
|
|---|
| 271 |
|
|---|
| 272 | &mov ($num,$j);
|
|---|
| 273 | &pmuludq($acc0,$mul0); # ap[num-1]*bp[i]
|
|---|
| 274 | &pmuludq($acc1,$mul1); # np[num-1]*m1
|
|---|
| 275 | &paddq ($car0,$acc0); # +=c0
|
|---|
| 276 | &paddq ($car1,$acc1); # +=c1
|
|---|
| 277 |
|
|---|
| 278 | &movq ($acc0,$car0);
|
|---|
| 279 | &pand ($acc0,$mask);
|
|---|
| 280 | &paddq ($car1,$acc0); # +=ap[num-1]*bp[i]+tp[num-1]
|
|---|
| 281 | &movd (&DWP($frame-4,"esp",$j,4),$car1); # tp[num-2]=
|
|---|
| 282 | &psrlq ($car0,32);
|
|---|
| 283 | &psrlq ($car1,32);
|
|---|
| 284 |
|
|---|
| 285 | &movd ($temp,&DWP($frame+4,"esp",$num,4)); # += tp[num]
|
|---|
| 286 | &paddq ($car1,$car0);
|
|---|
| 287 | &paddq ($car1,$temp);
|
|---|
| 288 | &movq (&QWP($frame,"esp",$num,4),$car1); # tp[num].tp[num-1]
|
|---|
| 289 |
|
|---|
| 290 | &lea ($i,&DWP(1,$i)); # i++
|
|---|
| 291 | &cmp ($i,$num);
|
|---|
| 292 | &jle (&label("outer"));
|
|---|
| 293 |
|
|---|
| 294 | &emms (); # done with mmx bank
|
|---|
| 295 | &jmp (&label("common_tail"));
|
|---|
| 296 |
|
|---|
| 297 | &set_label("non_sse2",16);
|
|---|
| 298 | }
|
|---|
| 299 | |
|---|
| 300 |
|
|---|
| 301 | if (0) {
|
|---|
| 302 | &mov ("esp",$_sp);
|
|---|
| 303 | &xor ("eax","eax"); # signal "not fast enough [yet]"
|
|---|
| 304 | &jmp (&label("just_leave"));
|
|---|
| 305 | # While the below code provides competitive performance for
|
|---|
| 306 | # all key lengths on modern Intel cores, it's still more
|
|---|
| 307 | # than 10% slower for 4096-bit key elsewhere:-( "Competitive"
|
|---|
| 308 | # means compared to the original integer-only assembler.
|
|---|
| 309 | # 512-bit RSA sign is better by ~40%, but that's about all
|
|---|
| 310 | # one can say about all CPUs...
|
|---|
| 311 | } else {
|
|---|
| 312 | $inp="esi"; # integer path uses these registers differently
|
|---|
| 313 | $word="edi";
|
|---|
| 314 | $carry="ebp";
|
|---|
| 315 |
|
|---|
| 316 | &mov ($inp,$_ap);
|
|---|
| 317 | &lea ($carry,&DWP(1,$num));
|
|---|
| 318 | &mov ($word,$_bp);
|
|---|
| 319 | &xor ($j,$j); # j=0
|
|---|
| 320 | &mov ("edx",$inp);
|
|---|
| 321 | &and ($carry,1); # see if num is even
|
|---|
| 322 | &sub ("edx",$word); # see if ap==bp
|
|---|
| 323 | &lea ("eax",&DWP(4,$word,$num,4)); # &bp[num]
|
|---|
| 324 | &or ($carry,"edx");
|
|---|
| 325 | &mov ($word,&DWP(0,$word)); # bp[0]
|
|---|
| 326 | &jz (&label("bn_sqr_mont"));
|
|---|
| 327 | &mov ($_bpend,"eax");
|
|---|
| 328 | &mov ("eax",&DWP(0,$inp));
|
|---|
| 329 | &xor ("edx","edx");
|
|---|
| 330 |
|
|---|
| 331 | &set_label("mull",16);
|
|---|
| 332 | &mov ($carry,"edx");
|
|---|
| 333 | &mul ($word); # ap[j]*bp[0]
|
|---|
| 334 | &add ($carry,"eax");
|
|---|
| 335 | &lea ($j,&DWP(1,$j));
|
|---|
| 336 | &adc ("edx",0);
|
|---|
| 337 | &mov ("eax",&DWP(0,$inp,$j,4)); # ap[j+1]
|
|---|
| 338 | &cmp ($j,$num);
|
|---|
| 339 | &mov (&DWP($frame-4,"esp",$j,4),$carry); # tp[j]=
|
|---|
| 340 | &jl (&label("mull"));
|
|---|
| 341 |
|
|---|
| 342 | &mov ($carry,"edx");
|
|---|
| 343 | &mul ($word); # ap[num-1]*bp[0]
|
|---|
| 344 | &mov ($word,$_n0);
|
|---|
| 345 | &add ("eax",$carry);
|
|---|
| 346 | &mov ($inp,$_np);
|
|---|
| 347 | &adc ("edx",0);
|
|---|
| 348 | &imul ($word,&DWP($frame,"esp")); # n0*tp[0]
|
|---|
| 349 |
|
|---|
| 350 | &mov (&DWP($frame,"esp",$num,4),"eax"); # tp[num-1]=
|
|---|
| 351 | &xor ($j,$j);
|
|---|
| 352 | &mov (&DWP($frame+4,"esp",$num,4),"edx"); # tp[num]=
|
|---|
| 353 | &mov (&DWP($frame+8,"esp",$num,4),$j); # tp[num+1]=
|
|---|
| 354 |
|
|---|
| 355 | &mov ("eax",&DWP(0,$inp)); # np[0]
|
|---|
| 356 | &mul ($word); # np[0]*m
|
|---|
| 357 | &add ("eax",&DWP($frame,"esp")); # +=tp[0]
|
|---|
| 358 | &mov ("eax",&DWP(4,$inp)); # np[1]
|
|---|
| 359 | &adc ("edx",0);
|
|---|
| 360 | &inc ($j);
|
|---|
| 361 |
|
|---|
| 362 | &jmp (&label("2ndmadd"));
|
|---|
| 363 | |
|---|
| 364 | |
|---|
| 365 |
|
|---|
| 366 | &set_label("1stmadd",16);
|
|---|
| 367 | &mov ($carry,"edx");
|
|---|
| 368 | &mul ($word); # ap[j]*bp[i]
|
|---|
| 369 | &add ($carry,&DWP($frame,"esp",$j,4)); # +=tp[j]
|
|---|
| 370 | &lea ($j,&DWP(1,$j));
|
|---|
| 371 | &adc ("edx",0);
|
|---|
| 372 | &add ($carry,"eax");
|
|---|
| 373 | &mov ("eax",&DWP(0,$inp,$j,4)); # ap[j+1]
|
|---|
| 374 | &adc ("edx",0);
|
|---|
| 375 | &cmp ($j,$num);
|
|---|
| 376 | &mov (&DWP($frame-4,"esp",$j,4),$carry); # tp[j]=
|
|---|
| 377 | &jl (&label("1stmadd"));
|
|---|
| 378 |
|
|---|
| 379 | &mov ($carry,"edx");
|
|---|
| 380 | &mul ($word); # ap[num-1]*bp[i]
|
|---|
| 381 | &add ("eax",&DWP($frame,"esp",$num,4)); # +=tp[num-1]
|
|---|
| 382 | &mov ($word,$_n0);
|
|---|
| 383 | &adc ("edx",0);
|
|---|
| 384 | &mov ($inp,$_np);
|
|---|
| 385 | &add ($carry,"eax");
|
|---|
| 386 | &adc ("edx",0);
|
|---|
| 387 | &imul ($word,&DWP($frame,"esp")); # n0*tp[0]
|
|---|
| 388 |
|
|---|
| 389 | &xor ($j,$j);
|
|---|
| 390 | &add ("edx",&DWP($frame+4,"esp",$num,4)); # carry+=tp[num]
|
|---|
| 391 | &mov (&DWP($frame,"esp",$num,4),$carry); # tp[num-1]=
|
|---|
| 392 | &adc ($j,0);
|
|---|
| 393 | &mov ("eax",&DWP(0,$inp)); # np[0]
|
|---|
| 394 | &mov (&DWP($frame+4,"esp",$num,4),"edx"); # tp[num]=
|
|---|
| 395 | &mov (&DWP($frame+8,"esp",$num,4),$j); # tp[num+1]=
|
|---|
| 396 |
|
|---|
| 397 | &mul ($word); # np[0]*m
|
|---|
| 398 | &add ("eax",&DWP($frame,"esp")); # +=tp[0]
|
|---|
| 399 | &mov ("eax",&DWP(4,$inp)); # np[1]
|
|---|
| 400 | &adc ("edx",0);
|
|---|
| 401 | &mov ($j,1);
|
|---|
| 402 | |
|---|
| 403 |
|
|---|
| 404 | &set_label("2ndmadd",16);
|
|---|
| 405 | &mov ($carry,"edx");
|
|---|
| 406 | &mul ($word); # np[j]*m
|
|---|
| 407 | &add ($carry,&DWP($frame,"esp",$j,4)); # +=tp[j]
|
|---|
| 408 | &lea ($j,&DWP(1,$j));
|
|---|
| 409 | &adc ("edx",0);
|
|---|
| 410 | &add ($carry,"eax");
|
|---|
| 411 | &mov ("eax",&DWP(0,$inp,$j,4)); # np[j+1]
|
|---|
| 412 | &adc ("edx",0);
|
|---|
| 413 | &cmp ($j,$num);
|
|---|
| 414 | &mov (&DWP($frame-8,"esp",$j,4),$carry); # tp[j-1]=
|
|---|
| 415 | &jl (&label("2ndmadd"));
|
|---|
| 416 |
|
|---|
| 417 | &mov ($carry,"edx");
|
|---|
| 418 | &mul ($word); # np[j]*m
|
|---|
| 419 | &add ($carry,&DWP($frame,"esp",$num,4)); # +=tp[num-1]
|
|---|
| 420 | &adc ("edx",0);
|
|---|
| 421 | &add ($carry,"eax");
|
|---|
| 422 | &adc ("edx",0);
|
|---|
| 423 | &mov (&DWP($frame-4,"esp",$num,4),$carry); # tp[num-2]=
|
|---|
| 424 |
|
|---|
| 425 | &xor ("eax","eax");
|
|---|
| 426 | &mov ($j,$_bp); # &bp[i]
|
|---|
| 427 | &add ("edx",&DWP($frame+4,"esp",$num,4)); # carry+=tp[num]
|
|---|
| 428 | &adc ("eax",&DWP($frame+8,"esp",$num,4)); # +=tp[num+1]
|
|---|
| 429 | &lea ($j,&DWP(4,$j));
|
|---|
| 430 | &mov (&DWP($frame,"esp",$num,4),"edx"); # tp[num-1]=
|
|---|
| 431 | &cmp ($j,$_bpend);
|
|---|
| 432 | &mov (&DWP($frame+4,"esp",$num,4),"eax"); # tp[num]=
|
|---|
| 433 | &je (&label("common_tail"));
|
|---|
| 434 |
|
|---|
| 435 | &mov ($word,&DWP(0,$j)); # bp[i+1]
|
|---|
| 436 | &mov ($inp,$_ap);
|
|---|
| 437 | &mov ($_bp,$j); # &bp[++i]
|
|---|
| 438 | &xor ($j,$j);
|
|---|
| 439 | &xor ("edx","edx");
|
|---|
| 440 | &mov ("eax",&DWP(0,$inp));
|
|---|
| 441 | &jmp (&label("1stmadd"));
|
|---|
| 442 | |
|---|
| 443 |
|
|---|
| 444 | &set_label("bn_sqr_mont",16);
|
|---|
| 445 | $sbit=$num;
|
|---|
| 446 | &mov ($_num,$num);
|
|---|
| 447 | &mov ($_bp,$j); # i=0
|
|---|
| 448 |
|
|---|
| 449 | &mov ("eax",$word); # ap[0]
|
|---|
| 450 | &mul ($word); # ap[0]*ap[0]
|
|---|
| 451 | &mov (&DWP($frame,"esp"),"eax"); # tp[0]=
|
|---|
| 452 | &mov ($sbit,"edx");
|
|---|
| 453 | &shr ("edx",1);
|
|---|
| 454 | &and ($sbit,1);
|
|---|
| 455 | &inc ($j);
|
|---|
| 456 | &set_label("sqr",16);
|
|---|
| 457 | &mov ("eax",&DWP(0,$inp,$j,4)); # ap[j]
|
|---|
| 458 | &mov ($carry,"edx");
|
|---|
| 459 | &mul ($word); # ap[j]*ap[0]
|
|---|
| 460 | &add ("eax",$carry);
|
|---|
| 461 | &lea ($j,&DWP(1,$j));
|
|---|
| 462 | &adc ("edx",0);
|
|---|
| 463 | &lea ($carry,&DWP(0,$sbit,"eax",2));
|
|---|
| 464 | &shr ("eax",31);
|
|---|
| 465 | &cmp ($j,$_num);
|
|---|
| 466 | &mov ($sbit,"eax");
|
|---|
| 467 | &mov (&DWP($frame-4,"esp",$j,4),$carry); # tp[j]=
|
|---|
| 468 | &jl (&label("sqr"));
|
|---|
| 469 |
|
|---|
| 470 | &mov ("eax",&DWP(0,$inp,$j,4)); # ap[num-1]
|
|---|
| 471 | &mov ($carry,"edx");
|
|---|
| 472 | &mul ($word); # ap[num-1]*ap[0]
|
|---|
| 473 | &add ("eax",$carry);
|
|---|
| 474 | &mov ($word,$_n0);
|
|---|
| 475 | &adc ("edx",0);
|
|---|
| 476 | &mov ($inp,$_np);
|
|---|
| 477 | &lea ($carry,&DWP(0,$sbit,"eax",2));
|
|---|
| 478 | &imul ($word,&DWP($frame,"esp")); # n0*tp[0]
|
|---|
| 479 | &shr ("eax",31);
|
|---|
| 480 | &mov (&DWP($frame,"esp",$j,4),$carry); # tp[num-1]=
|
|---|
| 481 |
|
|---|
| 482 | &lea ($carry,&DWP(0,"eax","edx",2));
|
|---|
| 483 | &mov ("eax",&DWP(0,$inp)); # np[0]
|
|---|
| 484 | &shr ("edx",31);
|
|---|
| 485 | &mov (&DWP($frame+4,"esp",$j,4),$carry); # tp[num]=
|
|---|
| 486 | &mov (&DWP($frame+8,"esp",$j,4),"edx"); # tp[num+1]=
|
|---|
| 487 |
|
|---|
| 488 | &mul ($word); # np[0]*m
|
|---|
| 489 | &add ("eax",&DWP($frame,"esp")); # +=tp[0]
|
|---|
| 490 | &mov ($num,$j);
|
|---|
| 491 | &adc ("edx",0);
|
|---|
| 492 | &mov ("eax",&DWP(4,$inp)); # np[1]
|
|---|
| 493 | &mov ($j,1);
|
|---|
| 494 | |
|---|
| 495 | |
|---|
| 496 |
|
|---|
| 497 | &set_label("3rdmadd",16);
|
|---|
| 498 | &mov ($carry,"edx");
|
|---|
| 499 | &mul ($word); # np[j]*m
|
|---|
| 500 | &add ($carry,&DWP($frame,"esp",$j,4)); # +=tp[j]
|
|---|
| 501 | &adc ("edx",0);
|
|---|
| 502 | &add ($carry,"eax");
|
|---|
| 503 | &mov ("eax",&DWP(4,$inp,$j,4)); # np[j+1]
|
|---|
| 504 | &adc ("edx",0);
|
|---|
| 505 | &mov (&DWP($frame-4,"esp",$j,4),$carry); # tp[j-1]=
|
|---|
| 506 |
|
|---|
| 507 | &mov ($carry,"edx");
|
|---|
| 508 | &mul ($word); # np[j+1]*m
|
|---|
| 509 | &add ($carry,&DWP($frame+4,"esp",$j,4)); # +=tp[j+1]
|
|---|
| 510 | &lea ($j,&DWP(2,$j));
|
|---|
| 511 | &adc ("edx",0);
|
|---|
| 512 | &add ($carry,"eax");
|
|---|
| 513 | &mov ("eax",&DWP(0,$inp,$j,4)); # np[j+2]
|
|---|
| 514 | &adc ("edx",0);
|
|---|
| 515 | &cmp ($j,$num);
|
|---|
| 516 | &mov (&DWP($frame-8,"esp",$j,4),$carry); # tp[j]=
|
|---|
| 517 | &jl (&label("3rdmadd"));
|
|---|
| 518 |
|
|---|
| 519 | &mov ($carry,"edx");
|
|---|
| 520 | &mul ($word); # np[j]*m
|
|---|
| 521 | &add ($carry,&DWP($frame,"esp",$num,4)); # +=tp[num-1]
|
|---|
| 522 | &adc ("edx",0);
|
|---|
| 523 | &add ($carry,"eax");
|
|---|
| 524 | &adc ("edx",0);
|
|---|
| 525 | &mov (&DWP($frame-4,"esp",$num,4),$carry); # tp[num-2]=
|
|---|
| 526 |
|
|---|
| 527 | &mov ($j,$_bp); # i
|
|---|
| 528 | &xor ("eax","eax");
|
|---|
| 529 | &mov ($inp,$_ap);
|
|---|
| 530 | &add ("edx",&DWP($frame+4,"esp",$num,4)); # carry+=tp[num]
|
|---|
| 531 | &adc ("eax",&DWP($frame+8,"esp",$num,4)); # +=tp[num+1]
|
|---|
| 532 | &mov (&DWP($frame,"esp",$num,4),"edx"); # tp[num-1]=
|
|---|
| 533 | &cmp ($j,$num);
|
|---|
| 534 | &mov (&DWP($frame+4,"esp",$num,4),"eax"); # tp[num]=
|
|---|
| 535 | &je (&label("common_tail"));
|
|---|
| 536 | |
|---|
| 537 |
|
|---|
| 538 | &mov ($word,&DWP(4,$inp,$j,4)); # ap[i]
|
|---|
| 539 | &lea ($j,&DWP(1,$j));
|
|---|
| 540 | &mov ("eax",$word);
|
|---|
| 541 | &mov ($_bp,$j); # ++i
|
|---|
| 542 | &mul ($word); # ap[i]*ap[i]
|
|---|
| 543 | &add ("eax",&DWP($frame,"esp",$j,4)); # +=tp[i]
|
|---|
| 544 | &adc ("edx",0);
|
|---|
| 545 | &mov (&DWP($frame,"esp",$j,4),"eax"); # tp[i]=
|
|---|
| 546 | &xor ($carry,$carry);
|
|---|
| 547 | &cmp ($j,$num);
|
|---|
| 548 | &lea ($j,&DWP(1,$j));
|
|---|
| 549 | &je (&label("sqrlast"));
|
|---|
| 550 |
|
|---|
| 551 | &mov ($sbit,"edx"); # zaps $num
|
|---|
| 552 | &shr ("edx",1);
|
|---|
| 553 | &and ($sbit,1);
|
|---|
| 554 | &set_label("sqradd",16);
|
|---|
| 555 | &mov ("eax",&DWP(0,$inp,$j,4)); # ap[j]
|
|---|
| 556 | &mov ($carry,"edx");
|
|---|
| 557 | &mul ($word); # ap[j]*ap[i]
|
|---|
| 558 | &add ("eax",$carry);
|
|---|
| 559 | &lea ($carry,&DWP(0,"eax","eax"));
|
|---|
| 560 | &adc ("edx",0);
|
|---|
| 561 | &shr ("eax",31);
|
|---|
| 562 | &add ($carry,&DWP($frame,"esp",$j,4)); # +=tp[j]
|
|---|
| 563 | &lea ($j,&DWP(1,$j));
|
|---|
| 564 | &adc ("eax",0);
|
|---|
| 565 | &add ($carry,$sbit);
|
|---|
| 566 | &adc ("eax",0);
|
|---|
| 567 | &cmp ($j,$_num);
|
|---|
| 568 | &mov (&DWP($frame-4,"esp",$j,4),$carry); # tp[j]=
|
|---|
| 569 | &mov ($sbit,"eax");
|
|---|
| 570 | &jle (&label("sqradd"));
|
|---|
| 571 |
|
|---|
| 572 | &mov ($carry,"edx");
|
|---|
| 573 | &add ("edx","edx");
|
|---|
| 574 | &shr ($carry,31);
|
|---|
| 575 | &add ("edx",$sbit);
|
|---|
| 576 | &adc ($carry,0);
|
|---|
| 577 | &set_label("sqrlast");
|
|---|
| 578 | &mov ($word,$_n0);
|
|---|
| 579 | &mov ($inp,$_np);
|
|---|
| 580 | &imul ($word,&DWP($frame,"esp")); # n0*tp[0]
|
|---|
| 581 |
|
|---|
| 582 | &add ("edx",&DWP($frame,"esp",$j,4)); # +=tp[num]
|
|---|
| 583 | &mov ("eax",&DWP(0,$inp)); # np[0]
|
|---|
| 584 | &adc ($carry,0);
|
|---|
| 585 | &mov (&DWP($frame,"esp",$j,4),"edx"); # tp[num]=
|
|---|
| 586 | &mov (&DWP($frame+4,"esp",$j,4),$carry); # tp[num+1]=
|
|---|
| 587 |
|
|---|
| 588 | &mul ($word); # np[0]*m
|
|---|
| 589 | &add ("eax",&DWP($frame,"esp")); # +=tp[0]
|
|---|
| 590 | &lea ($num,&DWP(-1,$j));
|
|---|
| 591 | &adc ("edx",0);
|
|---|
| 592 | &mov ($j,1);
|
|---|
| 593 | &mov ("eax",&DWP(4,$inp)); # np[1]
|
|---|
| 594 |
|
|---|
| 595 | &jmp (&label("3rdmadd"));
|
|---|
| 596 | }
|
|---|
| 597 | |
|---|
| 598 |
|
|---|
| 599 | &set_label("common_tail",16);
|
|---|
| 600 | &mov ($np,$_np); # load modulus pointer
|
|---|
| 601 | &mov ($rp,$_rp); # load result pointer
|
|---|
| 602 | &lea ($tp,&DWP($frame,"esp")); # [$ap and $bp are zapped]
|
|---|
| 603 |
|
|---|
| 604 | &mov ("eax",&DWP(0,$tp)); # tp[0]
|
|---|
| 605 | &mov ($j,$num); # j=num-1
|
|---|
| 606 | &xor ($i,$i); # i=0 and clear CF!
|
|---|
| 607 |
|
|---|
| 608 | &set_label("sub",16);
|
|---|
| 609 | &sbb ("eax",&DWP(0,$np,$i,4));
|
|---|
| 610 | &mov (&DWP(0,$rp,$i,4),"eax"); # rp[i]=tp[i]-np[i]
|
|---|
| 611 | &dec ($j); # doesn't affect CF!
|
|---|
| 612 | &mov ("eax",&DWP(4,$tp,$i,4)); # tp[i+1]
|
|---|
| 613 | &lea ($i,&DWP(1,$i)); # i++
|
|---|
| 614 | &jge (&label("sub"));
|
|---|
| 615 |
|
|---|
| 616 | &sbb ("eax",0); # handle upmost overflow bit
|
|---|
| 617 | &mov ("edx",-1);
|
|---|
| 618 | &xor ("edx","eax");
|
|---|
| 619 | &jmp (&label("copy"));
|
|---|
| 620 |
|
|---|
| 621 | &set_label("copy",16); # conditional copy
|
|---|
| 622 | &mov ($tp,&DWP($frame,"esp",$num,4));
|
|---|
| 623 | &mov ($np,&DWP(0,$rp,$num,4));
|
|---|
| 624 | &mov (&DWP($frame,"esp",$num,4),$j); # zap temporary vector
|
|---|
| 625 | &and ($tp,"eax");
|
|---|
| 626 | &and ($np,"edx");
|
|---|
| 627 | &or ($np,$tp);
|
|---|
| 628 | &mov (&DWP(0,$rp,$num,4),$np);
|
|---|
| 629 | &dec ($num);
|
|---|
| 630 | &jge (&label("copy"));
|
|---|
| 631 |
|
|---|
| 632 | &mov ("esp",$_sp); # pull saved stack pointer
|
|---|
| 633 | &mov ("eax",1);
|
|---|
| 634 | &set_label("just_leave");
|
|---|
| 635 | &function_end("bn_mul_mont");
|
|---|
| 636 |
|
|---|
| 637 | &asciz("Montgomery Multiplication for x86, CRYPTOGAMS by <appro\@openssl.org>");
|
|---|
| 638 |
|
|---|
| 639 | &asm_finish();
|
|---|
| 640 |
|
|---|
| 641 | close STDOUT or die "error closing STDOUT: $!";
|
|---|
