| 1 | #! /usr/bin/env perl
|
|---|
| 2 | # Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved.
|
|---|
| 3 | #
|
|---|
| 4 | # Licensed under the OpenSSL license (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 | # Montgomery multiplication routine for x86_64. While it gives modest
|
|---|
| 20 | # 9% improvement of rsa4096 sign on Opteron, rsa512 sign runs more
|
|---|
| 21 | # than twice, >2x, as fast. Most common rsa1024 sign is improved by
|
|---|
| 22 | # respectful 50%. It remains to be seen if loop unrolling and
|
|---|
| 23 | # dedicated squaring routine can provide further improvement...
|
|---|
| 24 |
|
|---|
| 25 | # July 2011.
|
|---|
| 26 | #
|
|---|
| 27 | # Add dedicated squaring procedure. Performance improvement varies
|
|---|
| 28 | # from platform to platform, but in average it's ~5%/15%/25%/33%
|
|---|
| 29 | # for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively.
|
|---|
| 30 |
|
|---|
| 31 | # August 2011.
|
|---|
| 32 | #
|
|---|
| 33 | # Unroll and modulo-schedule inner loops in such manner that they
|
|---|
| 34 | # are "fallen through" for input lengths of 8, which is critical for
|
|---|
| 35 | # 1024-bit RSA *sign*. Average performance improvement in comparison
|
|---|
| 36 | # to *initial* version of this module from 2005 is ~0%/30%/40%/45%
|
|---|
| 37 | # for 512-/1024-/2048-/4096-bit RSA *sign* benchmarks respectively.
|
|---|
| 38 |
|
|---|
| 39 | # June 2013.
|
|---|
| 40 | #
|
|---|
| 41 | # Optimize reduction in squaring procedure and improve 1024+-bit RSA
|
|---|
| 42 | # sign performance by 10-16% on Intel Sandy Bridge and later
|
|---|
| 43 | # (virtually same on non-Intel processors).
|
|---|
| 44 |
|
|---|
| 45 | # August 2013.
|
|---|
| 46 | #
|
|---|
| 47 | # Add MULX/ADOX/ADCX code path.
|
|---|
| 48 |
|
|---|
| 49 | $flavour = shift;
|
|---|
| 50 | $output = shift;
|
|---|
| 51 | if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
|
|---|
| 52 |
|
|---|
| 53 | $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
|
|---|
| 54 |
|
|---|
| 55 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
|
|---|
| 56 | ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
|
|---|
| 57 | ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
|
|---|
| 58 | die "can't locate x86_64-xlate.pl";
|
|---|
| 59 |
|
|---|
| 60 | open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
|
|---|
| 61 | *STDOUT=*OUT;
|
|---|
| 62 |
|
|---|
| 63 | if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
|
|---|
| 64 | =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
|
|---|
| 65 | $addx = ($1>=2.23);
|
|---|
| 66 | }
|
|---|
| 67 |
|
|---|
| 68 | if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
|
|---|
| 69 | `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
|
|---|
| 70 | $addx = ($1>=2.10);
|
|---|
| 71 | }
|
|---|
| 72 |
|
|---|
| 73 | if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
|
|---|
| 74 | `ml64 2>&1` =~ /Version ([0-9]+)\./) {
|
|---|
| 75 | $addx = ($1>=12);
|
|---|
| 76 | }
|
|---|
| 77 |
|
|---|
| 78 | if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9])\.([0-9]+)/) {
|
|---|
| 79 | my $ver = $2 + $3/100.0; # 3.1->3.01, 3.10->3.10
|
|---|
| 80 | $addx = ($ver>=3.03);
|
|---|
| 81 | }
|
|---|
| 82 |
|
|---|
| 83 | # int bn_mul_mont(
|
|---|
| 84 | $rp="%rdi"; # BN_ULONG *rp,
|
|---|
| 85 | $ap="%rsi"; # const BN_ULONG *ap,
|
|---|
| 86 | $bp="%rdx"; # const BN_ULONG *bp,
|
|---|
| 87 | $np="%rcx"; # const BN_ULONG *np,
|
|---|
| 88 | $n0="%r8"; # const BN_ULONG *n0,
|
|---|
| 89 | $num="%r9"; # int num);
|
|---|
| 90 | $lo0="%r10";
|
|---|
| 91 | $hi0="%r11";
|
|---|
| 92 | $hi1="%r13";
|
|---|
| 93 | $i="%r14";
|
|---|
| 94 | $j="%r15";
|
|---|
| 95 | $m0="%rbx";
|
|---|
| 96 | $m1="%rbp";
|
|---|
| 97 |
|
|---|
| 98 | $code=<<___;
|
|---|
| 99 | .text
|
|---|
| 100 |
|
|---|
| 101 | .extern OPENSSL_ia32cap_P
|
|---|
| 102 |
|
|---|
| 103 | .globl bn_mul_mont
|
|---|
| 104 | .type bn_mul_mont,\@function,6
|
|---|
| 105 | .align 16
|
|---|
| 106 | bn_mul_mont:
|
|---|
| 107 | mov ${num}d,${num}d
|
|---|
| 108 | mov %rsp,%rax
|
|---|
| 109 | test \$3,${num}d
|
|---|
| 110 | jnz .Lmul_enter
|
|---|
| 111 | cmp \$8,${num}d
|
|---|
| 112 | jb .Lmul_enter
|
|---|
| 113 | ___
|
|---|
| 114 | $code.=<<___ if ($addx);
|
|---|
| 115 | mov OPENSSL_ia32cap_P+8(%rip),%r11d
|
|---|
| 116 | ___
|
|---|
| 117 | $code.=<<___;
|
|---|
| 118 | cmp $ap,$bp
|
|---|
| 119 | jne .Lmul4x_enter
|
|---|
| 120 | test \$7,${num}d
|
|---|
| 121 | jz .Lsqr8x_enter
|
|---|
| 122 | jmp .Lmul4x_enter
|
|---|
| 123 |
|
|---|
| 124 | .align 16
|
|---|
| 125 | .Lmul_enter:
|
|---|
| 126 | push %rbx
|
|---|
| 127 | push %rbp
|
|---|
| 128 | push %r12
|
|---|
| 129 | push %r13
|
|---|
| 130 | push %r14
|
|---|
| 131 | push %r15
|
|---|
| 132 |
|
|---|
| 133 | neg $num
|
|---|
| 134 | mov %rsp,%r11
|
|---|
| 135 | lea -16(%rsp,$num,8),%r10 # future alloca(8*(num+2))
|
|---|
| 136 | neg $num # restore $num
|
|---|
| 137 | and \$-1024,%r10 # minimize TLB usage
|
|---|
| 138 |
|
|---|
| 139 | # An OS-agnostic version of __chkstk.
|
|---|
| 140 | #
|
|---|
| 141 | # Some OSes (Windows) insist on stack being "wired" to
|
|---|
| 142 | # physical memory in strictly sequential manner, i.e. if stack
|
|---|
| 143 | # allocation spans two pages, then reference to farmost one can
|
|---|
| 144 | # be punishable by SEGV. But page walking can do good even on
|
|---|
| 145 | # other OSes, because it guarantees that villain thread hits
|
|---|
| 146 | # the guard page before it can make damage to innocent one...
|
|---|
| 147 | sub %r10,%r11
|
|---|
| 148 | and \$-4096,%r11
|
|---|
| 149 | lea (%r10,%r11),%rsp
|
|---|
| 150 | mov (%rsp),%r11
|
|---|
| 151 | cmp %r10,%rsp
|
|---|
| 152 | ja .Lmul_page_walk
|
|---|
| 153 | jmp .Lmul_page_walk_done
|
|---|
| 154 |
|
|---|
| 155 | .align 16
|
|---|
| 156 | .Lmul_page_walk:
|
|---|
| 157 | lea -4096(%rsp),%rsp
|
|---|
| 158 | mov (%rsp),%r11
|
|---|
| 159 | cmp %r10,%rsp
|
|---|
| 160 | ja .Lmul_page_walk
|
|---|
| 161 | .Lmul_page_walk_done:
|
|---|
| 162 |
|
|---|
| 163 | mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
|
|---|
| 164 | .Lmul_body:
|
|---|
| 165 | mov $bp,%r12 # reassign $bp
|
|---|
| 166 | ___
|
|---|
| 167 | $bp="%r12";
|
|---|
| 168 | $code.=<<___;
|
|---|
| 169 | mov ($n0),$n0 # pull n0[0] value
|
|---|
| 170 | mov ($bp),$m0 # m0=bp[0]
|
|---|
| 171 | mov ($ap),%rax
|
|---|
| 172 |
|
|---|
| 173 | xor $i,$i # i=0
|
|---|
| 174 | xor $j,$j # j=0
|
|---|
| 175 |
|
|---|
| 176 | mov $n0,$m1
|
|---|
| 177 | mulq $m0 # ap[0]*bp[0]
|
|---|
| 178 | mov %rax,$lo0
|
|---|
| 179 | mov ($np),%rax
|
|---|
| 180 |
|
|---|
| 181 | imulq $lo0,$m1 # "tp[0]"*n0
|
|---|
| 182 | mov %rdx,$hi0
|
|---|
| 183 |
|
|---|
| 184 | mulq $m1 # np[0]*m1
|
|---|
| 185 | add %rax,$lo0 # discarded
|
|---|
| 186 | mov 8($ap),%rax
|
|---|
| 187 | adc \$0,%rdx
|
|---|
| 188 | mov %rdx,$hi1
|
|---|
| 189 |
|
|---|
| 190 | lea 1($j),$j # j++
|
|---|
| 191 | jmp .L1st_enter
|
|---|
| 192 |
|
|---|
| 193 | .align 16
|
|---|
| 194 | .L1st:
|
|---|
| 195 | add %rax,$hi1
|
|---|
| 196 | mov ($ap,$j,8),%rax
|
|---|
| 197 | adc \$0,%rdx
|
|---|
| 198 | add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
|
|---|
| 199 | mov $lo0,$hi0
|
|---|
| 200 | adc \$0,%rdx
|
|---|
| 201 | mov $hi1,-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 202 | mov %rdx,$hi1
|
|---|
| 203 |
|
|---|
| 204 | .L1st_enter:
|
|---|
| 205 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 206 | add %rax,$hi0
|
|---|
| 207 | mov ($np,$j,8),%rax
|
|---|
| 208 | adc \$0,%rdx
|
|---|
| 209 | lea 1($j),$j # j++
|
|---|
| 210 | mov %rdx,$lo0
|
|---|
| 211 |
|
|---|
| 212 | mulq $m1 # np[j]*m1
|
|---|
| 213 | cmp $num,$j
|
|---|
| 214 | jne .L1st
|
|---|
| 215 |
|
|---|
| 216 | add %rax,$hi1
|
|---|
| 217 | mov ($ap),%rax # ap[0]
|
|---|
| 218 | adc \$0,%rdx
|
|---|
| 219 | add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
|
|---|
| 220 | adc \$0,%rdx
|
|---|
| 221 | mov $hi1,-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 222 | mov %rdx,$hi1
|
|---|
| 223 | mov $lo0,$hi0
|
|---|
| 224 |
|
|---|
| 225 | xor %rdx,%rdx
|
|---|
| 226 | add $hi0,$hi1
|
|---|
| 227 | adc \$0,%rdx
|
|---|
| 228 | mov $hi1,-8(%rsp,$num,8)
|
|---|
| 229 | mov %rdx,(%rsp,$num,8) # store upmost overflow bit
|
|---|
| 230 |
|
|---|
| 231 | lea 1($i),$i # i++
|
|---|
| 232 | jmp .Louter
|
|---|
| 233 | .align 16
|
|---|
| 234 | .Louter:
|
|---|
| 235 | mov ($bp,$i,8),$m0 # m0=bp[i]
|
|---|
| 236 | xor $j,$j # j=0
|
|---|
| 237 | mov $n0,$m1
|
|---|
| 238 | mov (%rsp),$lo0
|
|---|
| 239 | mulq $m0 # ap[0]*bp[i]
|
|---|
| 240 | add %rax,$lo0 # ap[0]*bp[i]+tp[0]
|
|---|
| 241 | mov ($np),%rax
|
|---|
| 242 | adc \$0,%rdx
|
|---|
| 243 |
|
|---|
| 244 | imulq $lo0,$m1 # tp[0]*n0
|
|---|
| 245 | mov %rdx,$hi0
|
|---|
| 246 |
|
|---|
| 247 | mulq $m1 # np[0]*m1
|
|---|
| 248 | add %rax,$lo0 # discarded
|
|---|
| 249 | mov 8($ap),%rax
|
|---|
| 250 | adc \$0,%rdx
|
|---|
| 251 | mov 8(%rsp),$lo0 # tp[1]
|
|---|
| 252 | mov %rdx,$hi1
|
|---|
| 253 |
|
|---|
| 254 | lea 1($j),$j # j++
|
|---|
| 255 | jmp .Linner_enter
|
|---|
| 256 |
|
|---|
| 257 | .align 16
|
|---|
| 258 | .Linner:
|
|---|
| 259 | add %rax,$hi1
|
|---|
| 260 | mov ($ap,$j,8),%rax
|
|---|
| 261 | adc \$0,%rdx
|
|---|
| 262 | add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
|
|---|
| 263 | mov (%rsp,$j,8),$lo0
|
|---|
| 264 | adc \$0,%rdx
|
|---|
| 265 | mov $hi1,-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 266 | mov %rdx,$hi1
|
|---|
| 267 |
|
|---|
| 268 | .Linner_enter:
|
|---|
| 269 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 270 | add %rax,$hi0
|
|---|
| 271 | mov ($np,$j,8),%rax
|
|---|
| 272 | adc \$0,%rdx
|
|---|
| 273 | add $hi0,$lo0 # ap[j]*bp[i]+tp[j]
|
|---|
| 274 | mov %rdx,$hi0
|
|---|
| 275 | adc \$0,$hi0
|
|---|
| 276 | lea 1($j),$j # j++
|
|---|
| 277 |
|
|---|
| 278 | mulq $m1 # np[j]*m1
|
|---|
| 279 | cmp $num,$j
|
|---|
| 280 | jne .Linner
|
|---|
| 281 |
|
|---|
| 282 | add %rax,$hi1
|
|---|
| 283 | mov ($ap),%rax # ap[0]
|
|---|
| 284 | adc \$0,%rdx
|
|---|
| 285 | add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
|
|---|
| 286 | mov (%rsp,$j,8),$lo0
|
|---|
| 287 | adc \$0,%rdx
|
|---|
| 288 | mov $hi1,-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 289 | mov %rdx,$hi1
|
|---|
| 290 |
|
|---|
| 291 | xor %rdx,%rdx
|
|---|
| 292 | add $hi0,$hi1
|
|---|
| 293 | adc \$0,%rdx
|
|---|
| 294 | add $lo0,$hi1 # pull upmost overflow bit
|
|---|
| 295 | adc \$0,%rdx
|
|---|
| 296 | mov $hi1,-8(%rsp,$num,8)
|
|---|
| 297 | mov %rdx,(%rsp,$num,8) # store upmost overflow bit
|
|---|
| 298 |
|
|---|
| 299 | lea 1($i),$i # i++
|
|---|
| 300 | cmp $num,$i
|
|---|
| 301 | jb .Louter
|
|---|
| 302 |
|
|---|
| 303 | xor $i,$i # i=0 and clear CF!
|
|---|
| 304 | mov (%rsp),%rax # tp[0]
|
|---|
| 305 | lea (%rsp),$ap # borrow ap for tp
|
|---|
| 306 | mov $num,$j # j=num
|
|---|
| 307 | jmp .Lsub
|
|---|
| 308 | .align 16
|
|---|
| 309 | .Lsub: sbb ($np,$i,8),%rax
|
|---|
| 310 | mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 311 | mov 8($ap,$i,8),%rax # tp[i+1]
|
|---|
| 312 | lea 1($i),$i # i++
|
|---|
| 313 | dec $j # doesnn't affect CF!
|
|---|
| 314 | jnz .Lsub
|
|---|
| 315 |
|
|---|
| 316 | sbb \$0,%rax # handle upmost overflow bit
|
|---|
| 317 | xor $i,$i
|
|---|
| 318 | and %rax,$ap
|
|---|
| 319 | not %rax
|
|---|
| 320 | mov $rp,$np
|
|---|
| 321 | and %rax,$np
|
|---|
| 322 | mov $num,$j # j=num
|
|---|
| 323 | or $np,$ap # ap=borrow?tp:rp
|
|---|
| 324 | .align 16
|
|---|
| 325 | .Lcopy: # copy or in-place refresh
|
|---|
| 326 | mov ($ap,$i,8),%rax
|
|---|
| 327 | mov $i,(%rsp,$i,8) # zap temporary vector
|
|---|
| 328 | mov %rax,($rp,$i,8) # rp[i]=tp[i]
|
|---|
| 329 | lea 1($i),$i
|
|---|
| 330 | sub \$1,$j
|
|---|
| 331 | jnz .Lcopy
|
|---|
| 332 |
|
|---|
| 333 | mov 8(%rsp,$num,8),%rsi # restore %rsp
|
|---|
| 334 | mov \$1,%rax
|
|---|
| 335 | mov -48(%rsi),%r15
|
|---|
| 336 | mov -40(%rsi),%r14
|
|---|
| 337 | mov -32(%rsi),%r13
|
|---|
| 338 | mov -24(%rsi),%r12
|
|---|
| 339 | mov -16(%rsi),%rbp
|
|---|
| 340 | mov -8(%rsi),%rbx
|
|---|
| 341 | lea (%rsi),%rsp
|
|---|
| 342 | .Lmul_epilogue:
|
|---|
| 343 | ret
|
|---|
| 344 | .size bn_mul_mont,.-bn_mul_mont
|
|---|
| 345 | ___
|
|---|
| 346 | {{{
|
|---|
| 347 | my @A=("%r10","%r11");
|
|---|
| 348 | my @N=("%r13","%rdi");
|
|---|
| 349 | $code.=<<___;
|
|---|
| 350 | .type bn_mul4x_mont,\@function,6
|
|---|
| 351 | .align 16
|
|---|
| 352 | bn_mul4x_mont:
|
|---|
| 353 | mov ${num}d,${num}d
|
|---|
| 354 | mov %rsp,%rax
|
|---|
| 355 | .Lmul4x_enter:
|
|---|
| 356 | ___
|
|---|
| 357 | $code.=<<___ if ($addx);
|
|---|
| 358 | and \$0x80100,%r11d
|
|---|
| 359 | cmp \$0x80100,%r11d
|
|---|
| 360 | je .Lmulx4x_enter
|
|---|
| 361 | ___
|
|---|
| 362 | $code.=<<___;
|
|---|
| 363 | push %rbx
|
|---|
| 364 | push %rbp
|
|---|
| 365 | push %r12
|
|---|
| 366 | push %r13
|
|---|
| 367 | push %r14
|
|---|
| 368 | push %r15
|
|---|
| 369 |
|
|---|
| 370 | neg $num
|
|---|
| 371 | mov %rsp,%r11
|
|---|
| 372 | lea -32(%rsp,$num,8),%r10 # future alloca(8*(num+4))
|
|---|
| 373 | neg $num # restore
|
|---|
| 374 | and \$-1024,%r10 # minimize TLB usage
|
|---|
| 375 |
|
|---|
| 376 | sub %r10,%r11
|
|---|
| 377 | and \$-4096,%r11
|
|---|
| 378 | lea (%r10,%r11),%rsp
|
|---|
| 379 | mov (%rsp),%r11
|
|---|
| 380 | cmp %r10,%rsp
|
|---|
| 381 | ja .Lmul4x_page_walk
|
|---|
| 382 | jmp .Lmul4x_page_walk_done
|
|---|
| 383 |
|
|---|
| 384 | .Lmul4x_page_walk:
|
|---|
| 385 | lea -4096(%rsp),%rsp
|
|---|
| 386 | mov (%rsp),%r11
|
|---|
| 387 | cmp %r10,%rsp
|
|---|
| 388 | ja .Lmul4x_page_walk
|
|---|
| 389 | .Lmul4x_page_walk_done:
|
|---|
| 390 |
|
|---|
| 391 | mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
|
|---|
| 392 | .Lmul4x_body:
|
|---|
| 393 | mov $rp,16(%rsp,$num,8) # tp[num+2]=$rp
|
|---|
| 394 | mov %rdx,%r12 # reassign $bp
|
|---|
| 395 | ___
|
|---|
| 396 | $bp="%r12";
|
|---|
| 397 | $code.=<<___;
|
|---|
| 398 | mov ($n0),$n0 # pull n0[0] value
|
|---|
| 399 | mov ($bp),$m0 # m0=bp[0]
|
|---|
| 400 | mov ($ap),%rax
|
|---|
| 401 |
|
|---|
| 402 | xor $i,$i # i=0
|
|---|
| 403 | xor $j,$j # j=0
|
|---|
| 404 |
|
|---|
| 405 | mov $n0,$m1
|
|---|
| 406 | mulq $m0 # ap[0]*bp[0]
|
|---|
| 407 | mov %rax,$A[0]
|
|---|
| 408 | mov ($np),%rax
|
|---|
| 409 |
|
|---|
| 410 | imulq $A[0],$m1 # "tp[0]"*n0
|
|---|
| 411 | mov %rdx,$A[1]
|
|---|
| 412 |
|
|---|
| 413 | mulq $m1 # np[0]*m1
|
|---|
| 414 | add %rax,$A[0] # discarded
|
|---|
| 415 | mov 8($ap),%rax
|
|---|
| 416 | adc \$0,%rdx
|
|---|
| 417 | mov %rdx,$N[1]
|
|---|
| 418 |
|
|---|
| 419 | mulq $m0
|
|---|
| 420 | add %rax,$A[1]
|
|---|
| 421 | mov 8($np),%rax
|
|---|
| 422 | adc \$0,%rdx
|
|---|
| 423 | mov %rdx,$A[0]
|
|---|
| 424 |
|
|---|
| 425 | mulq $m1
|
|---|
| 426 | add %rax,$N[1]
|
|---|
| 427 | mov 16($ap),%rax
|
|---|
| 428 | adc \$0,%rdx
|
|---|
| 429 | add $A[1],$N[1]
|
|---|
| 430 | lea 4($j),$j # j++
|
|---|
| 431 | adc \$0,%rdx
|
|---|
| 432 | mov $N[1],(%rsp)
|
|---|
| 433 | mov %rdx,$N[0]
|
|---|
| 434 | jmp .L1st4x
|
|---|
| 435 | .align 16
|
|---|
| 436 | .L1st4x:
|
|---|
| 437 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 438 | add %rax,$A[0]
|
|---|
| 439 | mov -16($np,$j,8),%rax
|
|---|
| 440 | adc \$0,%rdx
|
|---|
| 441 | mov %rdx,$A[1]
|
|---|
| 442 |
|
|---|
| 443 | mulq $m1 # np[j]*m1
|
|---|
| 444 | add %rax,$N[0]
|
|---|
| 445 | mov -8($ap,$j,8),%rax
|
|---|
| 446 | adc \$0,%rdx
|
|---|
| 447 | add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 448 | adc \$0,%rdx
|
|---|
| 449 | mov $N[0],-24(%rsp,$j,8) # tp[j-1]
|
|---|
| 450 | mov %rdx,$N[1]
|
|---|
| 451 |
|
|---|
| 452 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 453 | add %rax,$A[1]
|
|---|
| 454 | mov -8($np,$j,8),%rax
|
|---|
| 455 | adc \$0,%rdx
|
|---|
| 456 | mov %rdx,$A[0]
|
|---|
| 457 |
|
|---|
| 458 | mulq $m1 # np[j]*m1
|
|---|
| 459 | add %rax,$N[1]
|
|---|
| 460 | mov ($ap,$j,8),%rax
|
|---|
| 461 | adc \$0,%rdx
|
|---|
| 462 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 463 | adc \$0,%rdx
|
|---|
| 464 | mov $N[1],-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 465 | mov %rdx,$N[0]
|
|---|
| 466 |
|
|---|
| 467 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 468 | add %rax,$A[0]
|
|---|
| 469 | mov ($np,$j,8),%rax
|
|---|
| 470 | adc \$0,%rdx
|
|---|
| 471 | mov %rdx,$A[1]
|
|---|
| 472 |
|
|---|
| 473 | mulq $m1 # np[j]*m1
|
|---|
| 474 | add %rax,$N[0]
|
|---|
| 475 | mov 8($ap,$j,8),%rax
|
|---|
| 476 | adc \$0,%rdx
|
|---|
| 477 | add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 478 | adc \$0,%rdx
|
|---|
| 479 | mov $N[0],-8(%rsp,$j,8) # tp[j-1]
|
|---|
| 480 | mov %rdx,$N[1]
|
|---|
| 481 |
|
|---|
| 482 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 483 | add %rax,$A[1]
|
|---|
| 484 | mov 8($np,$j,8),%rax
|
|---|
| 485 | adc \$0,%rdx
|
|---|
| 486 | lea 4($j),$j # j++
|
|---|
| 487 | mov %rdx,$A[0]
|
|---|
| 488 |
|
|---|
| 489 | mulq $m1 # np[j]*m1
|
|---|
| 490 | add %rax,$N[1]
|
|---|
| 491 | mov -16($ap,$j,8),%rax
|
|---|
| 492 | adc \$0,%rdx
|
|---|
| 493 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 494 | adc \$0,%rdx
|
|---|
| 495 | mov $N[1],-32(%rsp,$j,8) # tp[j-1]
|
|---|
| 496 | mov %rdx,$N[0]
|
|---|
| 497 | cmp $num,$j
|
|---|
| 498 | jb .L1st4x
|
|---|
| 499 |
|
|---|
| 500 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 501 | add %rax,$A[0]
|
|---|
| 502 | mov -16($np,$j,8),%rax
|
|---|
| 503 | adc \$0,%rdx
|
|---|
| 504 | mov %rdx,$A[1]
|
|---|
| 505 |
|
|---|
| 506 | mulq $m1 # np[j]*m1
|
|---|
| 507 | add %rax,$N[0]
|
|---|
| 508 | mov -8($ap,$j,8),%rax
|
|---|
| 509 | adc \$0,%rdx
|
|---|
| 510 | add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 511 | adc \$0,%rdx
|
|---|
| 512 | mov $N[0],-24(%rsp,$j,8) # tp[j-1]
|
|---|
| 513 | mov %rdx,$N[1]
|
|---|
| 514 |
|
|---|
| 515 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 516 | add %rax,$A[1]
|
|---|
| 517 | mov -8($np,$j,8),%rax
|
|---|
| 518 | adc \$0,%rdx
|
|---|
| 519 | mov %rdx,$A[0]
|
|---|
| 520 |
|
|---|
| 521 | mulq $m1 # np[j]*m1
|
|---|
| 522 | add %rax,$N[1]
|
|---|
| 523 | mov ($ap),%rax # ap[0]
|
|---|
| 524 | adc \$0,%rdx
|
|---|
| 525 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 526 | adc \$0,%rdx
|
|---|
| 527 | mov $N[1],-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 528 | mov %rdx,$N[0]
|
|---|
| 529 |
|
|---|
| 530 | xor $N[1],$N[1]
|
|---|
| 531 | add $A[0],$N[0]
|
|---|
| 532 | adc \$0,$N[1]
|
|---|
| 533 | mov $N[0],-8(%rsp,$j,8)
|
|---|
| 534 | mov $N[1],(%rsp,$j,8) # store upmost overflow bit
|
|---|
| 535 |
|
|---|
| 536 | lea 1($i),$i # i++
|
|---|
| 537 | .align 4
|
|---|
| 538 | .Louter4x:
|
|---|
| 539 | mov ($bp,$i,8),$m0 # m0=bp[i]
|
|---|
| 540 | xor $j,$j # j=0
|
|---|
| 541 | mov (%rsp),$A[0]
|
|---|
| 542 | mov $n0,$m1
|
|---|
| 543 | mulq $m0 # ap[0]*bp[i]
|
|---|
| 544 | add %rax,$A[0] # ap[0]*bp[i]+tp[0]
|
|---|
| 545 | mov ($np),%rax
|
|---|
| 546 | adc \$0,%rdx
|
|---|
| 547 |
|
|---|
| 548 | imulq $A[0],$m1 # tp[0]*n0
|
|---|
| 549 | mov %rdx,$A[1]
|
|---|
| 550 |
|
|---|
| 551 | mulq $m1 # np[0]*m1
|
|---|
| 552 | add %rax,$A[0] # "$N[0]", discarded
|
|---|
| 553 | mov 8($ap),%rax
|
|---|
| 554 | adc \$0,%rdx
|
|---|
| 555 | mov %rdx,$N[1]
|
|---|
| 556 |
|
|---|
| 557 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 558 | add %rax,$A[1]
|
|---|
| 559 | mov 8($np),%rax
|
|---|
| 560 | adc \$0,%rdx
|
|---|
| 561 | add 8(%rsp),$A[1] # +tp[1]
|
|---|
| 562 | adc \$0,%rdx
|
|---|
| 563 | mov %rdx,$A[0]
|
|---|
| 564 |
|
|---|
| 565 | mulq $m1 # np[j]*m1
|
|---|
| 566 | add %rax,$N[1]
|
|---|
| 567 | mov 16($ap),%rax
|
|---|
| 568 | adc \$0,%rdx
|
|---|
| 569 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
|
|---|
| 570 | lea 4($j),$j # j+=2
|
|---|
| 571 | adc \$0,%rdx
|
|---|
| 572 | mov $N[1],(%rsp) # tp[j-1]
|
|---|
| 573 | mov %rdx,$N[0]
|
|---|
| 574 | jmp .Linner4x
|
|---|
| 575 | .align 16
|
|---|
| 576 | .Linner4x:
|
|---|
| 577 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 578 | add %rax,$A[0]
|
|---|
| 579 | mov -16($np,$j,8),%rax
|
|---|
| 580 | adc \$0,%rdx
|
|---|
| 581 | add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
|
|---|
| 582 | adc \$0,%rdx
|
|---|
| 583 | mov %rdx,$A[1]
|
|---|
| 584 |
|
|---|
| 585 | mulq $m1 # np[j]*m1
|
|---|
| 586 | add %rax,$N[0]
|
|---|
| 587 | mov -8($ap,$j,8),%rax
|
|---|
| 588 | adc \$0,%rdx
|
|---|
| 589 | add $A[0],$N[0]
|
|---|
| 590 | adc \$0,%rdx
|
|---|
| 591 | mov $N[0],-24(%rsp,$j,8) # tp[j-1]
|
|---|
| 592 | mov %rdx,$N[1]
|
|---|
| 593 |
|
|---|
| 594 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 595 | add %rax,$A[1]
|
|---|
| 596 | mov -8($np,$j,8),%rax
|
|---|
| 597 | adc \$0,%rdx
|
|---|
| 598 | add -8(%rsp,$j,8),$A[1]
|
|---|
| 599 | adc \$0,%rdx
|
|---|
| 600 | mov %rdx,$A[0]
|
|---|
| 601 |
|
|---|
| 602 | mulq $m1 # np[j]*m1
|
|---|
| 603 | add %rax,$N[1]
|
|---|
| 604 | mov ($ap,$j,8),%rax
|
|---|
| 605 | adc \$0,%rdx
|
|---|
| 606 | add $A[1],$N[1]
|
|---|
| 607 | adc \$0,%rdx
|
|---|
| 608 | mov $N[1],-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 609 | mov %rdx,$N[0]
|
|---|
| 610 |
|
|---|
| 611 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 612 | add %rax,$A[0]
|
|---|
| 613 | mov ($np,$j,8),%rax
|
|---|
| 614 | adc \$0,%rdx
|
|---|
| 615 | add (%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
|
|---|
| 616 | adc \$0,%rdx
|
|---|
| 617 | mov %rdx,$A[1]
|
|---|
| 618 |
|
|---|
| 619 | mulq $m1 # np[j]*m1
|
|---|
| 620 | add %rax,$N[0]
|
|---|
| 621 | mov 8($ap,$j,8),%rax
|
|---|
| 622 | adc \$0,%rdx
|
|---|
| 623 | add $A[0],$N[0]
|
|---|
| 624 | adc \$0,%rdx
|
|---|
| 625 | mov $N[0],-8(%rsp,$j,8) # tp[j-1]
|
|---|
| 626 | mov %rdx,$N[1]
|
|---|
| 627 |
|
|---|
| 628 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 629 | add %rax,$A[1]
|
|---|
| 630 | mov 8($np,$j,8),%rax
|
|---|
| 631 | adc \$0,%rdx
|
|---|
| 632 | add 8(%rsp,$j,8),$A[1]
|
|---|
| 633 | adc \$0,%rdx
|
|---|
| 634 | lea 4($j),$j # j++
|
|---|
| 635 | mov %rdx,$A[0]
|
|---|
| 636 |
|
|---|
| 637 | mulq $m1 # np[j]*m1
|
|---|
| 638 | add %rax,$N[1]
|
|---|
| 639 | mov -16($ap,$j,8),%rax
|
|---|
| 640 | adc \$0,%rdx
|
|---|
| 641 | add $A[1],$N[1]
|
|---|
| 642 | adc \$0,%rdx
|
|---|
| 643 | mov $N[1],-32(%rsp,$j,8) # tp[j-1]
|
|---|
| 644 | mov %rdx,$N[0]
|
|---|
| 645 | cmp $num,$j
|
|---|
| 646 | jb .Linner4x
|
|---|
| 647 |
|
|---|
| 648 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 649 | add %rax,$A[0]
|
|---|
| 650 | mov -16($np,$j,8),%rax
|
|---|
| 651 | adc \$0,%rdx
|
|---|
| 652 | add -16(%rsp,$j,8),$A[0] # ap[j]*bp[i]+tp[j]
|
|---|
| 653 | adc \$0,%rdx
|
|---|
| 654 | mov %rdx,$A[1]
|
|---|
| 655 |
|
|---|
| 656 | mulq $m1 # np[j]*m1
|
|---|
| 657 | add %rax,$N[0]
|
|---|
| 658 | mov -8($ap,$j,8),%rax
|
|---|
| 659 | adc \$0,%rdx
|
|---|
| 660 | add $A[0],$N[0]
|
|---|
| 661 | adc \$0,%rdx
|
|---|
| 662 | mov $N[0],-24(%rsp,$j,8) # tp[j-1]
|
|---|
| 663 | mov %rdx,$N[1]
|
|---|
| 664 |
|
|---|
| 665 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 666 | add %rax,$A[1]
|
|---|
| 667 | mov -8($np,$j,8),%rax
|
|---|
| 668 | adc \$0,%rdx
|
|---|
| 669 | add -8(%rsp,$j,8),$A[1]
|
|---|
| 670 | adc \$0,%rdx
|
|---|
| 671 | lea 1($i),$i # i++
|
|---|
| 672 | mov %rdx,$A[0]
|
|---|
| 673 |
|
|---|
| 674 | mulq $m1 # np[j]*m1
|
|---|
| 675 | add %rax,$N[1]
|
|---|
| 676 | mov ($ap),%rax # ap[0]
|
|---|
| 677 | adc \$0,%rdx
|
|---|
| 678 | add $A[1],$N[1]
|
|---|
| 679 | adc \$0,%rdx
|
|---|
| 680 | mov $N[1],-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 681 | mov %rdx,$N[0]
|
|---|
| 682 |
|
|---|
| 683 | xor $N[1],$N[1]
|
|---|
| 684 | add $A[0],$N[0]
|
|---|
| 685 | adc \$0,$N[1]
|
|---|
| 686 | add (%rsp,$num,8),$N[0] # pull upmost overflow bit
|
|---|
| 687 | adc \$0,$N[1]
|
|---|
| 688 | mov $N[0],-8(%rsp,$j,8)
|
|---|
| 689 | mov $N[1],(%rsp,$j,8) # store upmost overflow bit
|
|---|
| 690 |
|
|---|
| 691 | cmp $num,$i
|
|---|
| 692 | jb .Louter4x
|
|---|
| 693 | ___
|
|---|
| 694 | {
|
|---|
| 695 | my @ri=("%rax","%rdx",$m0,$m1);
|
|---|
| 696 | $code.=<<___;
|
|---|
| 697 | mov 16(%rsp,$num,8),$rp # restore $rp
|
|---|
| 698 | mov 0(%rsp),@ri[0] # tp[0]
|
|---|
| 699 | pxor %xmm0,%xmm0
|
|---|
| 700 | mov 8(%rsp),@ri[1] # tp[1]
|
|---|
| 701 | shr \$2,$num # num/=4
|
|---|
| 702 | lea (%rsp),$ap # borrow ap for tp
|
|---|
| 703 | xor $i,$i # i=0 and clear CF!
|
|---|
| 704 |
|
|---|
| 705 | sub 0($np),@ri[0]
|
|---|
| 706 | mov 16($ap),@ri[2] # tp[2]
|
|---|
| 707 | mov 24($ap),@ri[3] # tp[3]
|
|---|
| 708 | sbb 8($np),@ri[1]
|
|---|
| 709 | lea -1($num),$j # j=num/4-1
|
|---|
| 710 | jmp .Lsub4x
|
|---|
| 711 | .align 16
|
|---|
| 712 | .Lsub4x:
|
|---|
| 713 | mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 714 | mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 715 | sbb 16($np,$i,8),@ri[2]
|
|---|
| 716 | mov 32($ap,$i,8),@ri[0] # tp[i+1]
|
|---|
| 717 | mov 40($ap,$i,8),@ri[1]
|
|---|
| 718 | sbb 24($np,$i,8),@ri[3]
|
|---|
| 719 | mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 720 | mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 721 | sbb 32($np,$i,8),@ri[0]
|
|---|
| 722 | mov 48($ap,$i,8),@ri[2]
|
|---|
| 723 | mov 56($ap,$i,8),@ri[3]
|
|---|
| 724 | sbb 40($np,$i,8),@ri[1]
|
|---|
| 725 | lea 4($i),$i # i++
|
|---|
| 726 | dec $j # doesnn't affect CF!
|
|---|
| 727 | jnz .Lsub4x
|
|---|
| 728 |
|
|---|
| 729 | mov @ri[0],0($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 730 | mov 32($ap,$i,8),@ri[0] # load overflow bit
|
|---|
| 731 | sbb 16($np,$i,8),@ri[2]
|
|---|
| 732 | mov @ri[1],8($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 733 | sbb 24($np,$i,8),@ri[3]
|
|---|
| 734 | mov @ri[2],16($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 735 |
|
|---|
| 736 | sbb \$0,@ri[0] # handle upmost overflow bit
|
|---|
| 737 | mov @ri[3],24($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 738 | xor $i,$i # i=0
|
|---|
| 739 | and @ri[0],$ap
|
|---|
| 740 | not @ri[0]
|
|---|
| 741 | mov $rp,$np
|
|---|
| 742 | and @ri[0],$np
|
|---|
| 743 | lea -1($num),$j
|
|---|
| 744 | or $np,$ap # ap=borrow?tp:rp
|
|---|
| 745 |
|
|---|
| 746 | movdqu ($ap),%xmm1
|
|---|
| 747 | movdqa %xmm0,(%rsp)
|
|---|
| 748 | movdqu %xmm1,($rp)
|
|---|
| 749 | jmp .Lcopy4x
|
|---|
| 750 | .align 16
|
|---|
| 751 | .Lcopy4x: # copy or in-place refresh
|
|---|
| 752 | movdqu 16($ap,$i),%xmm2
|
|---|
| 753 | movdqu 32($ap,$i),%xmm1
|
|---|
| 754 | movdqa %xmm0,16(%rsp,$i)
|
|---|
| 755 | movdqu %xmm2,16($rp,$i)
|
|---|
| 756 | movdqa %xmm0,32(%rsp,$i)
|
|---|
| 757 | movdqu %xmm1,32($rp,$i)
|
|---|
| 758 | lea 32($i),$i
|
|---|
| 759 | dec $j
|
|---|
| 760 | jnz .Lcopy4x
|
|---|
| 761 |
|
|---|
| 762 | shl \$2,$num
|
|---|
| 763 | movdqu 16($ap,$i),%xmm2
|
|---|
| 764 | movdqa %xmm0,16(%rsp,$i)
|
|---|
| 765 | movdqu %xmm2,16($rp,$i)
|
|---|
| 766 | ___
|
|---|
| 767 | }
|
|---|
| 768 | $code.=<<___;
|
|---|
| 769 | mov 8(%rsp,$num,8),%rsi # restore %rsp
|
|---|
| 770 | mov \$1,%rax
|
|---|
| 771 | mov -48(%rsi),%r15
|
|---|
| 772 | mov -40(%rsi),%r14
|
|---|
| 773 | mov -32(%rsi),%r13
|
|---|
| 774 | mov -24(%rsi),%r12
|
|---|
| 775 | mov -16(%rsi),%rbp
|
|---|
| 776 | mov -8(%rsi),%rbx
|
|---|
| 777 | lea (%rsi),%rsp
|
|---|
| 778 | .Lmul4x_epilogue:
|
|---|
| 779 | ret
|
|---|
| 780 | .size bn_mul4x_mont,.-bn_mul4x_mont
|
|---|
| 781 | ___
|
|---|
| 782 | }}}
|
|---|
| 783 | |
|---|
| 784 | {{{
|
|---|
| 785 | ######################################################################
|
|---|
| 786 | # void bn_sqr8x_mont(
|
|---|
| 787 | my $rptr="%rdi"; # const BN_ULONG *rptr,
|
|---|
| 788 | my $aptr="%rsi"; # const BN_ULONG *aptr,
|
|---|
| 789 | my $bptr="%rdx"; # not used
|
|---|
| 790 | my $nptr="%rcx"; # const BN_ULONG *nptr,
|
|---|
| 791 | my $n0 ="%r8"; # const BN_ULONG *n0);
|
|---|
| 792 | my $num ="%r9"; # int num, has to be divisible by 8
|
|---|
| 793 |
|
|---|
| 794 | my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
|
|---|
| 795 | my @A0=("%r10","%r11");
|
|---|
| 796 | my @A1=("%r12","%r13");
|
|---|
| 797 | my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
|
|---|
| 798 |
|
|---|
| 799 | $code.=<<___ if ($addx);
|
|---|
| 800 | .extern bn_sqrx8x_internal # see x86_64-mont5 module
|
|---|
| 801 | ___
|
|---|
| 802 | $code.=<<___;
|
|---|
| 803 | .extern bn_sqr8x_internal # see x86_64-mont5 module
|
|---|
| 804 |
|
|---|
| 805 | .type bn_sqr8x_mont,\@function,6
|
|---|
| 806 | .align 32
|
|---|
| 807 | bn_sqr8x_mont:
|
|---|
| 808 | mov %rsp,%rax
|
|---|
| 809 | .Lsqr8x_enter:
|
|---|
| 810 | push %rbx
|
|---|
| 811 | push %rbp
|
|---|
| 812 | push %r12
|
|---|
| 813 | push %r13
|
|---|
| 814 | push %r14
|
|---|
| 815 | push %r15
|
|---|
| 816 | .Lsqr8x_prologue:
|
|---|
| 817 |
|
|---|
| 818 | mov ${num}d,%r10d
|
|---|
| 819 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 820 | shl \$3+2,%r10 # 4*$num
|
|---|
| 821 | neg $num
|
|---|
| 822 |
|
|---|
| 823 | ##############################################################
|
|---|
| 824 | # ensure that stack frame doesn't alias with $aptr modulo
|
|---|
| 825 | # 4096. this is done to allow memory disambiguation logic
|
|---|
| 826 | # do its job.
|
|---|
| 827 | #
|
|---|
| 828 | lea -64(%rsp,$num,2),%r11
|
|---|
| 829 | mov %rsp,%rbp
|
|---|
| 830 | mov ($n0),$n0 # *n0
|
|---|
| 831 | sub $aptr,%r11
|
|---|
| 832 | and \$4095,%r11
|
|---|
| 833 | cmp %r11,%r10
|
|---|
| 834 | jb .Lsqr8x_sp_alt
|
|---|
| 835 | sub %r11,%rbp # align with $aptr
|
|---|
| 836 | lea -64(%rbp,$num,2),%rbp # future alloca(frame+2*$num)
|
|---|
| 837 | jmp .Lsqr8x_sp_done
|
|---|
| 838 |
|
|---|
| 839 | .align 32
|
|---|
| 840 | .Lsqr8x_sp_alt:
|
|---|
| 841 | lea 4096-64(,$num,2),%r10 # 4096-frame-2*$num
|
|---|
| 842 | lea -64(%rbp,$num,2),%rbp # future alloca(frame+2*$num)
|
|---|
| 843 | sub %r10,%r11
|
|---|
| 844 | mov \$0,%r10
|
|---|
| 845 | cmovc %r10,%r11
|
|---|
| 846 | sub %r11,%rbp
|
|---|
| 847 | .Lsqr8x_sp_done:
|
|---|
| 848 | and \$-64,%rbp
|
|---|
| 849 | mov %rsp,%r11
|
|---|
| 850 | sub %rbp,%r11
|
|---|
| 851 | and \$-4096,%r11
|
|---|
| 852 | lea (%rbp,%r11),%rsp
|
|---|
| 853 | mov (%rsp),%r10
|
|---|
| 854 | cmp %rbp,%rsp
|
|---|
| 855 | ja .Lsqr8x_page_walk
|
|---|
| 856 | jmp .Lsqr8x_page_walk_done
|
|---|
| 857 |
|
|---|
| 858 | .align 16
|
|---|
| 859 | .Lsqr8x_page_walk:
|
|---|
| 860 | lea -4096(%rsp),%rsp
|
|---|
| 861 | mov (%rsp),%r10
|
|---|
| 862 | cmp %rbp,%rsp
|
|---|
| 863 | ja .Lsqr8x_page_walk
|
|---|
| 864 | .Lsqr8x_page_walk_done:
|
|---|
| 865 |
|
|---|
| 866 | mov $num,%r10
|
|---|
| 867 | neg $num
|
|---|
| 868 |
|
|---|
| 869 | mov $n0, 32(%rsp)
|
|---|
| 870 | mov %rax, 40(%rsp) # save original %rsp
|
|---|
| 871 | .Lsqr8x_body:
|
|---|
| 872 |
|
|---|
| 873 | movq $nptr, %xmm2 # save pointer to modulus
|
|---|
| 874 | pxor %xmm0,%xmm0
|
|---|
| 875 | movq $rptr,%xmm1 # save $rptr
|
|---|
| 876 | movq %r10, %xmm3 # -$num
|
|---|
| 877 | ___
|
|---|
| 878 | $code.=<<___ if ($addx);
|
|---|
| 879 | mov OPENSSL_ia32cap_P+8(%rip),%eax
|
|---|
| 880 | and \$0x80100,%eax
|
|---|
| 881 | cmp \$0x80100,%eax
|
|---|
| 882 | jne .Lsqr8x_nox
|
|---|
| 883 |
|
|---|
| 884 | call bn_sqrx8x_internal # see x86_64-mont5 module
|
|---|
| 885 | # %rax top-most carry
|
|---|
| 886 | # %rbp nptr
|
|---|
| 887 | # %rcx -8*num
|
|---|
| 888 | # %r8 end of tp[2*num]
|
|---|
| 889 | lea (%r8,%rcx),%rbx
|
|---|
| 890 | mov %rcx,$num
|
|---|
| 891 | mov %rcx,%rdx
|
|---|
| 892 | movq %xmm1,$rptr
|
|---|
| 893 | sar \$3+2,%rcx # %cf=0
|
|---|
| 894 | jmp .Lsqr8x_sub
|
|---|
| 895 |
|
|---|
| 896 | .align 32
|
|---|
| 897 | .Lsqr8x_nox:
|
|---|
| 898 | ___
|
|---|
| 899 | $code.=<<___;
|
|---|
| 900 | call bn_sqr8x_internal # see x86_64-mont5 module
|
|---|
| 901 | # %rax top-most carry
|
|---|
| 902 | # %rbp nptr
|
|---|
| 903 | # %r8 -8*num
|
|---|
| 904 | # %rdi end of tp[2*num]
|
|---|
| 905 | lea (%rdi,$num),%rbx
|
|---|
| 906 | mov $num,%rcx
|
|---|
| 907 | mov $num,%rdx
|
|---|
| 908 | movq %xmm1,$rptr
|
|---|
| 909 | sar \$3+2,%rcx # %cf=0
|
|---|
| 910 | jmp .Lsqr8x_sub
|
|---|
| 911 |
|
|---|
| 912 | .align 32
|
|---|
| 913 | .Lsqr8x_sub:
|
|---|
| 914 | mov 8*0(%rbx),%r12
|
|---|
| 915 | mov 8*1(%rbx),%r13
|
|---|
| 916 | mov 8*2(%rbx),%r14
|
|---|
| 917 | mov 8*3(%rbx),%r15
|
|---|
| 918 | lea 8*4(%rbx),%rbx
|
|---|
| 919 | sbb 8*0(%rbp),%r12
|
|---|
| 920 | sbb 8*1(%rbp),%r13
|
|---|
| 921 | sbb 8*2(%rbp),%r14
|
|---|
| 922 | sbb 8*3(%rbp),%r15
|
|---|
| 923 | lea 8*4(%rbp),%rbp
|
|---|
| 924 | mov %r12,8*0($rptr)
|
|---|
| 925 | mov %r13,8*1($rptr)
|
|---|
| 926 | mov %r14,8*2($rptr)
|
|---|
| 927 | mov %r15,8*3($rptr)
|
|---|
| 928 | lea 8*4($rptr),$rptr
|
|---|
| 929 | inc %rcx # preserves %cf
|
|---|
| 930 | jnz .Lsqr8x_sub
|
|---|
| 931 |
|
|---|
| 932 | sbb \$0,%rax # top-most carry
|
|---|
| 933 | lea (%rbx,$num),%rbx # rewind
|
|---|
| 934 | lea ($rptr,$num),$rptr # rewind
|
|---|
| 935 |
|
|---|
| 936 | movq %rax,%xmm1
|
|---|
| 937 | pxor %xmm0,%xmm0
|
|---|
| 938 | pshufd \$0,%xmm1,%xmm1
|
|---|
| 939 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 940 | jmp .Lsqr8x_cond_copy
|
|---|
| 941 |
|
|---|
| 942 | .align 32
|
|---|
| 943 | .Lsqr8x_cond_copy:
|
|---|
| 944 | movdqa 16*0(%rbx),%xmm2
|
|---|
| 945 | movdqa 16*1(%rbx),%xmm3
|
|---|
| 946 | lea 16*2(%rbx),%rbx
|
|---|
| 947 | movdqu 16*0($rptr),%xmm4
|
|---|
| 948 | movdqu 16*1($rptr),%xmm5
|
|---|
| 949 | lea 16*2($rptr),$rptr
|
|---|
| 950 | movdqa %xmm0,-16*2(%rbx) # zero tp
|
|---|
| 951 | movdqa %xmm0,-16*1(%rbx)
|
|---|
| 952 | movdqa %xmm0,-16*2(%rbx,%rdx)
|
|---|
| 953 | movdqa %xmm0,-16*1(%rbx,%rdx)
|
|---|
| 954 | pcmpeqd %xmm1,%xmm0
|
|---|
| 955 | pand %xmm1,%xmm2
|
|---|
| 956 | pand %xmm1,%xmm3
|
|---|
| 957 | pand %xmm0,%xmm4
|
|---|
| 958 | pand %xmm0,%xmm5
|
|---|
| 959 | pxor %xmm0,%xmm0
|
|---|
| 960 | por %xmm2,%xmm4
|
|---|
| 961 | por %xmm3,%xmm5
|
|---|
| 962 | movdqu %xmm4,-16*2($rptr)
|
|---|
| 963 | movdqu %xmm5,-16*1($rptr)
|
|---|
| 964 | add \$32,$num
|
|---|
| 965 | jnz .Lsqr8x_cond_copy
|
|---|
| 966 |
|
|---|
| 967 | mov \$1,%rax
|
|---|
| 968 | mov -48(%rsi),%r15
|
|---|
| 969 | mov -40(%rsi),%r14
|
|---|
| 970 | mov -32(%rsi),%r13
|
|---|
| 971 | mov -24(%rsi),%r12
|
|---|
| 972 | mov -16(%rsi),%rbp
|
|---|
| 973 | mov -8(%rsi),%rbx
|
|---|
| 974 | lea (%rsi),%rsp
|
|---|
| 975 | .Lsqr8x_epilogue:
|
|---|
| 976 | ret
|
|---|
| 977 | .size bn_sqr8x_mont,.-bn_sqr8x_mont
|
|---|
| 978 | ___
|
|---|
| 979 | }}}
|
|---|
| 980 | |
|---|
| 981 |
|
|---|
| 982 | if ($addx) {{{
|
|---|
| 983 | my $bp="%rdx"; # original value
|
|---|
| 984 |
|
|---|
| 985 | $code.=<<___;
|
|---|
| 986 | .type bn_mulx4x_mont,\@function,6
|
|---|
| 987 | .align 32
|
|---|
| 988 | bn_mulx4x_mont:
|
|---|
| 989 | mov %rsp,%rax
|
|---|
| 990 | .Lmulx4x_enter:
|
|---|
| 991 | push %rbx
|
|---|
| 992 | push %rbp
|
|---|
| 993 | push %r12
|
|---|
| 994 | push %r13
|
|---|
| 995 | push %r14
|
|---|
| 996 | push %r15
|
|---|
| 997 | .Lmulx4x_prologue:
|
|---|
| 998 |
|
|---|
| 999 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 1000 | xor %r10,%r10
|
|---|
| 1001 | sub $num,%r10 # -$num
|
|---|
| 1002 | mov ($n0),$n0 # *n0
|
|---|
| 1003 | lea -72(%rsp,%r10),%rbp # future alloca(frame+$num+8)
|
|---|
| 1004 | and \$-128,%rbp
|
|---|
| 1005 | mov %rsp,%r11
|
|---|
| 1006 | sub %rbp,%r11
|
|---|
| 1007 | and \$-4096,%r11
|
|---|
| 1008 | lea (%rbp,%r11),%rsp
|
|---|
| 1009 | mov (%rsp),%r10
|
|---|
| 1010 | cmp %rbp,%rsp
|
|---|
| 1011 | ja .Lmulx4x_page_walk
|
|---|
| 1012 | jmp .Lmulx4x_page_walk_done
|
|---|
| 1013 |
|
|---|
| 1014 | .align 16
|
|---|
| 1015 | .Lmulx4x_page_walk:
|
|---|
| 1016 | lea -4096(%rsp),%rsp
|
|---|
| 1017 | mov (%rsp),%r10
|
|---|
| 1018 | cmp %rbp,%rsp
|
|---|
| 1019 | ja .Lmulx4x_page_walk
|
|---|
| 1020 | .Lmulx4x_page_walk_done:
|
|---|
| 1021 |
|
|---|
| 1022 | lea ($bp,$num),%r10
|
|---|
| 1023 | ##############################################################
|
|---|
| 1024 | # Stack layout
|
|---|
| 1025 | # +0 num
|
|---|
| 1026 | # +8 off-loaded &b[i]
|
|---|
| 1027 | # +16 end of b[num]
|
|---|
| 1028 | # +24 saved n0
|
|---|
| 1029 | # +32 saved rp
|
|---|
| 1030 | # +40 saved %rsp
|
|---|
| 1031 | # +48 inner counter
|
|---|
| 1032 | # +56
|
|---|
| 1033 | # +64 tmp[num+1]
|
|---|
| 1034 | #
|
|---|
| 1035 | mov $num,0(%rsp) # save $num
|
|---|
| 1036 | shr \$5,$num
|
|---|
| 1037 | mov %r10,16(%rsp) # end of b[num]
|
|---|
| 1038 | sub \$1,$num
|
|---|
| 1039 | mov $n0, 24(%rsp) # save *n0
|
|---|
| 1040 | mov $rp, 32(%rsp) # save $rp
|
|---|
| 1041 | mov %rax,40(%rsp) # save original %rsp
|
|---|
| 1042 | mov $num,48(%rsp) # inner counter
|
|---|
| 1043 | jmp .Lmulx4x_body
|
|---|
| 1044 |
|
|---|
| 1045 | .align 32
|
|---|
| 1046 | .Lmulx4x_body:
|
|---|
| 1047 | ___
|
|---|
| 1048 | my ($aptr, $bptr, $nptr, $tptr, $mi, $bi, $zero, $num)=
|
|---|
| 1049 | ("%rsi","%rdi","%rcx","%rbx","%r8","%r9","%rbp","%rax");
|
|---|
| 1050 | my $rptr=$bptr;
|
|---|
| 1051 | $code.=<<___;
|
|---|
| 1052 | lea 8($bp),$bptr
|
|---|
| 1053 | mov ($bp),%rdx # b[0], $bp==%rdx actually
|
|---|
| 1054 | lea 64+32(%rsp),$tptr
|
|---|
| 1055 | mov %rdx,$bi
|
|---|
| 1056 |
|
|---|
| 1057 | mulx 0*8($aptr),$mi,%rax # a[0]*b[0]
|
|---|
| 1058 | mulx 1*8($aptr),%r11,%r14 # a[1]*b[0]
|
|---|
| 1059 | add %rax,%r11
|
|---|
| 1060 | mov $bptr,8(%rsp) # off-load &b[i]
|
|---|
| 1061 | mulx 2*8($aptr),%r12,%r13 # ...
|
|---|
| 1062 | adc %r14,%r12
|
|---|
| 1063 | adc \$0,%r13
|
|---|
| 1064 |
|
|---|
| 1065 | mov $mi,$bptr # borrow $bptr
|
|---|
| 1066 | imulq 24(%rsp),$mi # "t[0]"*n0
|
|---|
| 1067 | xor $zero,$zero # cf=0, of=0
|
|---|
| 1068 |
|
|---|
| 1069 | mulx 3*8($aptr),%rax,%r14
|
|---|
| 1070 | mov $mi,%rdx
|
|---|
| 1071 | lea 4*8($aptr),$aptr
|
|---|
| 1072 | adcx %rax,%r13
|
|---|
| 1073 | adcx $zero,%r14 # cf=0
|
|---|
| 1074 |
|
|---|
| 1075 | mulx 0*8($nptr),%rax,%r10
|
|---|
| 1076 | adcx %rax,$bptr # discarded
|
|---|
| 1077 | adox %r11,%r10
|
|---|
| 1078 | mulx 1*8($nptr),%rax,%r11
|
|---|
| 1079 | adcx %rax,%r10
|
|---|
| 1080 | adox %r12,%r11
|
|---|
| 1081 | .byte 0xc4,0x62,0xfb,0xf6,0xa1,0x10,0x00,0x00,0x00 # mulx 2*8($nptr),%rax,%r12
|
|---|
| 1082 | mov 48(%rsp),$bptr # counter value
|
|---|
| 1083 | mov %r10,-4*8($tptr)
|
|---|
| 1084 | adcx %rax,%r11
|
|---|
| 1085 | adox %r13,%r12
|
|---|
| 1086 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 1087 | mov $bi,%rdx
|
|---|
| 1088 | mov %r11,-3*8($tptr)
|
|---|
| 1089 | adcx %rax,%r12
|
|---|
| 1090 | adox $zero,%r15 # of=0
|
|---|
| 1091 | lea 4*8($nptr),$nptr
|
|---|
| 1092 | mov %r12,-2*8($tptr)
|
|---|
| 1093 |
|
|---|
| 1094 | jmp .Lmulx4x_1st
|
|---|
| 1095 |
|
|---|
| 1096 | .align 32
|
|---|
| 1097 | .Lmulx4x_1st:
|
|---|
| 1098 | adcx $zero,%r15 # cf=0, modulo-scheduled
|
|---|
| 1099 | mulx 0*8($aptr),%r10,%rax # a[4]*b[0]
|
|---|
| 1100 | adcx %r14,%r10
|
|---|
| 1101 | mulx 1*8($aptr),%r11,%r14 # a[5]*b[0]
|
|---|
| 1102 | adcx %rax,%r11
|
|---|
| 1103 | mulx 2*8($aptr),%r12,%rax # ...
|
|---|
| 1104 | adcx %r14,%r12
|
|---|
| 1105 | mulx 3*8($aptr),%r13,%r14
|
|---|
| 1106 | .byte 0x67,0x67
|
|---|
| 1107 | mov $mi,%rdx
|
|---|
| 1108 | adcx %rax,%r13
|
|---|
| 1109 | adcx $zero,%r14 # cf=0
|
|---|
| 1110 | lea 4*8($aptr),$aptr
|
|---|
| 1111 | lea 4*8($tptr),$tptr
|
|---|
| 1112 |
|
|---|
| 1113 | adox %r15,%r10
|
|---|
| 1114 | mulx 0*8($nptr),%rax,%r15
|
|---|
| 1115 | adcx %rax,%r10
|
|---|
| 1116 | adox %r15,%r11
|
|---|
| 1117 | mulx 1*8($nptr),%rax,%r15
|
|---|
| 1118 | adcx %rax,%r11
|
|---|
| 1119 | adox %r15,%r12
|
|---|
| 1120 | mulx 2*8($nptr),%rax,%r15
|
|---|
| 1121 | mov %r10,-5*8($tptr)
|
|---|
| 1122 | adcx %rax,%r12
|
|---|
| 1123 | mov %r11,-4*8($tptr)
|
|---|
| 1124 | adox %r15,%r13
|
|---|
| 1125 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 1126 | mov $bi,%rdx
|
|---|
| 1127 | mov %r12,-3*8($tptr)
|
|---|
| 1128 | adcx %rax,%r13
|
|---|
| 1129 | adox $zero,%r15
|
|---|
| 1130 | lea 4*8($nptr),$nptr
|
|---|
| 1131 | mov %r13,-2*8($tptr)
|
|---|
| 1132 |
|
|---|
| 1133 | dec $bptr # of=0, pass cf
|
|---|
| 1134 | jnz .Lmulx4x_1st
|
|---|
| 1135 |
|
|---|
| 1136 | mov 0(%rsp),$num # load num
|
|---|
| 1137 | mov 8(%rsp),$bptr # re-load &b[i]
|
|---|
| 1138 | adc $zero,%r15 # modulo-scheduled
|
|---|
| 1139 | add %r15,%r14
|
|---|
| 1140 | sbb %r15,%r15 # top-most carry
|
|---|
| 1141 | mov %r14,-1*8($tptr)
|
|---|
| 1142 | jmp .Lmulx4x_outer
|
|---|
| 1143 |
|
|---|
| 1144 | .align 32
|
|---|
| 1145 | .Lmulx4x_outer:
|
|---|
| 1146 | mov ($bptr),%rdx # b[i]
|
|---|
| 1147 | lea 8($bptr),$bptr # b++
|
|---|
| 1148 | sub $num,$aptr # rewind $aptr
|
|---|
| 1149 | mov %r15,($tptr) # save top-most carry
|
|---|
| 1150 | lea 64+4*8(%rsp),$tptr
|
|---|
| 1151 | sub $num,$nptr # rewind $nptr
|
|---|
| 1152 |
|
|---|
| 1153 | mulx 0*8($aptr),$mi,%r11 # a[0]*b[i]
|
|---|
| 1154 | xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
|
|---|
| 1155 | mov %rdx,$bi
|
|---|
| 1156 | mulx 1*8($aptr),%r14,%r12 # a[1]*b[i]
|
|---|
| 1157 | adox -4*8($tptr),$mi
|
|---|
| 1158 | adcx %r14,%r11
|
|---|
| 1159 | mulx 2*8($aptr),%r15,%r13 # ...
|
|---|
| 1160 | adox -3*8($tptr),%r11
|
|---|
| 1161 | adcx %r15,%r12
|
|---|
| 1162 | adox -2*8($tptr),%r12
|
|---|
| 1163 | adcx $zero,%r13
|
|---|
| 1164 | adox $zero,%r13
|
|---|
| 1165 |
|
|---|
| 1166 | mov $bptr,8(%rsp) # off-load &b[i]
|
|---|
| 1167 | mov $mi,%r15
|
|---|
| 1168 | imulq 24(%rsp),$mi # "t[0]"*n0
|
|---|
| 1169 | xor %ebp,%ebp # xor $zero,$zero # cf=0, of=0
|
|---|
| 1170 |
|
|---|
| 1171 | mulx 3*8($aptr),%rax,%r14
|
|---|
| 1172 | mov $mi,%rdx
|
|---|
| 1173 | adcx %rax,%r13
|
|---|
| 1174 | adox -1*8($tptr),%r13
|
|---|
| 1175 | adcx $zero,%r14
|
|---|
| 1176 | lea 4*8($aptr),$aptr
|
|---|
| 1177 | adox $zero,%r14
|
|---|
| 1178 |
|
|---|
| 1179 | mulx 0*8($nptr),%rax,%r10
|
|---|
| 1180 | adcx %rax,%r15 # discarded
|
|---|
| 1181 | adox %r11,%r10
|
|---|
| 1182 | mulx 1*8($nptr),%rax,%r11
|
|---|
| 1183 | adcx %rax,%r10
|
|---|
| 1184 | adox %r12,%r11
|
|---|
| 1185 | mulx 2*8($nptr),%rax,%r12
|
|---|
| 1186 | mov %r10,-4*8($tptr)
|
|---|
| 1187 | adcx %rax,%r11
|
|---|
| 1188 | adox %r13,%r12
|
|---|
| 1189 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 1190 | mov $bi,%rdx
|
|---|
| 1191 | mov %r11,-3*8($tptr)
|
|---|
| 1192 | lea 4*8($nptr),$nptr
|
|---|
| 1193 | adcx %rax,%r12
|
|---|
| 1194 | adox $zero,%r15 # of=0
|
|---|
| 1195 | mov 48(%rsp),$bptr # counter value
|
|---|
| 1196 | mov %r12,-2*8($tptr)
|
|---|
| 1197 |
|
|---|
| 1198 | jmp .Lmulx4x_inner
|
|---|
| 1199 |
|
|---|
| 1200 | .align 32
|
|---|
| 1201 | .Lmulx4x_inner:
|
|---|
| 1202 | mulx 0*8($aptr),%r10,%rax # a[4]*b[i]
|
|---|
| 1203 | adcx $zero,%r15 # cf=0, modulo-scheduled
|
|---|
| 1204 | adox %r14,%r10
|
|---|
| 1205 | mulx 1*8($aptr),%r11,%r14 # a[5]*b[i]
|
|---|
| 1206 | adcx 0*8($tptr),%r10
|
|---|
| 1207 | adox %rax,%r11
|
|---|
| 1208 | mulx 2*8($aptr),%r12,%rax # ...
|
|---|
| 1209 | adcx 1*8($tptr),%r11
|
|---|
| 1210 | adox %r14,%r12
|
|---|
| 1211 | mulx 3*8($aptr),%r13,%r14
|
|---|
| 1212 | mov $mi,%rdx
|
|---|
| 1213 | adcx 2*8($tptr),%r12
|
|---|
| 1214 | adox %rax,%r13
|
|---|
| 1215 | adcx 3*8($tptr),%r13
|
|---|
| 1216 | adox $zero,%r14 # of=0
|
|---|
| 1217 | lea 4*8($aptr),$aptr
|
|---|
| 1218 | lea 4*8($tptr),$tptr
|
|---|
| 1219 | adcx $zero,%r14 # cf=0
|
|---|
| 1220 |
|
|---|
| 1221 | adox %r15,%r10
|
|---|
| 1222 | mulx 0*8($nptr),%rax,%r15
|
|---|
| 1223 | adcx %rax,%r10
|
|---|
| 1224 | adox %r15,%r11
|
|---|
| 1225 | mulx 1*8($nptr),%rax,%r15
|
|---|
| 1226 | adcx %rax,%r11
|
|---|
| 1227 | adox %r15,%r12
|
|---|
| 1228 | mulx 2*8($nptr),%rax,%r15
|
|---|
| 1229 | mov %r10,-5*8($tptr)
|
|---|
| 1230 | adcx %rax,%r12
|
|---|
| 1231 | adox %r15,%r13
|
|---|
| 1232 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 1233 | mov $bi,%rdx
|
|---|
| 1234 | mov %r11,-4*8($tptr)
|
|---|
| 1235 | mov %r12,-3*8($tptr)
|
|---|
| 1236 | adcx %rax,%r13
|
|---|
| 1237 | adox $zero,%r15
|
|---|
| 1238 | lea 4*8($nptr),$nptr
|
|---|
| 1239 | mov %r13,-2*8($tptr)
|
|---|
| 1240 |
|
|---|
| 1241 | dec $bptr # of=0, pass cf
|
|---|
| 1242 | jnz .Lmulx4x_inner
|
|---|
| 1243 |
|
|---|
| 1244 | mov 0(%rsp),$num # load num
|
|---|
| 1245 | mov 8(%rsp),$bptr # re-load &b[i]
|
|---|
| 1246 | adc $zero,%r15 # modulo-scheduled
|
|---|
| 1247 | sub 0*8($tptr),$zero # pull top-most carry
|
|---|
| 1248 | adc %r15,%r14
|
|---|
| 1249 | sbb %r15,%r15 # top-most carry
|
|---|
| 1250 | mov %r14,-1*8($tptr)
|
|---|
| 1251 |
|
|---|
| 1252 | cmp 16(%rsp),$bptr
|
|---|
| 1253 | jne .Lmulx4x_outer
|
|---|
| 1254 |
|
|---|
| 1255 | lea 64(%rsp),$tptr
|
|---|
| 1256 | sub $num,$nptr # rewind $nptr
|
|---|
| 1257 | neg %r15
|
|---|
| 1258 | mov $num,%rdx
|
|---|
| 1259 | shr \$3+2,$num # %cf=0
|
|---|
| 1260 | mov 32(%rsp),$rptr # restore rp
|
|---|
| 1261 | jmp .Lmulx4x_sub
|
|---|
| 1262 |
|
|---|
| 1263 | .align 32
|
|---|
| 1264 | .Lmulx4x_sub:
|
|---|
| 1265 | mov 8*0($tptr),%r11
|
|---|
| 1266 | mov 8*1($tptr),%r12
|
|---|
| 1267 | mov 8*2($tptr),%r13
|
|---|
| 1268 | mov 8*3($tptr),%r14
|
|---|
| 1269 | lea 8*4($tptr),$tptr
|
|---|
| 1270 | sbb 8*0($nptr),%r11
|
|---|
| 1271 | sbb 8*1($nptr),%r12
|
|---|
| 1272 | sbb 8*2($nptr),%r13
|
|---|
| 1273 | sbb 8*3($nptr),%r14
|
|---|
| 1274 | lea 8*4($nptr),$nptr
|
|---|
| 1275 | mov %r11,8*0($rptr)
|
|---|
| 1276 | mov %r12,8*1($rptr)
|
|---|
| 1277 | mov %r13,8*2($rptr)
|
|---|
| 1278 | mov %r14,8*3($rptr)
|
|---|
| 1279 | lea 8*4($rptr),$rptr
|
|---|
| 1280 | dec $num # preserves %cf
|
|---|
| 1281 | jnz .Lmulx4x_sub
|
|---|
| 1282 |
|
|---|
| 1283 | sbb \$0,%r15 # top-most carry
|
|---|
| 1284 | lea 64(%rsp),$tptr
|
|---|
| 1285 | sub %rdx,$rptr # rewind
|
|---|
| 1286 |
|
|---|
| 1287 | movq %r15,%xmm1
|
|---|
| 1288 | pxor %xmm0,%xmm0
|
|---|
| 1289 | pshufd \$0,%xmm1,%xmm1
|
|---|
| 1290 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 1291 | jmp .Lmulx4x_cond_copy
|
|---|
| 1292 |
|
|---|
| 1293 | .align 32
|
|---|
| 1294 | .Lmulx4x_cond_copy:
|
|---|
| 1295 | movdqa 16*0($tptr),%xmm2
|
|---|
| 1296 | movdqa 16*1($tptr),%xmm3
|
|---|
| 1297 | lea 16*2($tptr),$tptr
|
|---|
| 1298 | movdqu 16*0($rptr),%xmm4
|
|---|
| 1299 | movdqu 16*1($rptr),%xmm5
|
|---|
| 1300 | lea 16*2($rptr),$rptr
|
|---|
| 1301 | movdqa %xmm0,-16*2($tptr) # zero tp
|
|---|
| 1302 | movdqa %xmm0,-16*1($tptr)
|
|---|
| 1303 | pcmpeqd %xmm1,%xmm0
|
|---|
| 1304 | pand %xmm1,%xmm2
|
|---|
| 1305 | pand %xmm1,%xmm3
|
|---|
| 1306 | pand %xmm0,%xmm4
|
|---|
| 1307 | pand %xmm0,%xmm5
|
|---|
| 1308 | pxor %xmm0,%xmm0
|
|---|
| 1309 | por %xmm2,%xmm4
|
|---|
| 1310 | por %xmm3,%xmm5
|
|---|
| 1311 | movdqu %xmm4,-16*2($rptr)
|
|---|
| 1312 | movdqu %xmm5,-16*1($rptr)
|
|---|
| 1313 | sub \$32,%rdx
|
|---|
| 1314 | jnz .Lmulx4x_cond_copy
|
|---|
| 1315 |
|
|---|
| 1316 | mov %rdx,($tptr)
|
|---|
| 1317 |
|
|---|
| 1318 | mov \$1,%rax
|
|---|
| 1319 | mov -48(%rsi),%r15
|
|---|
| 1320 | mov -40(%rsi),%r14
|
|---|
| 1321 | mov -32(%rsi),%r13
|
|---|
| 1322 | mov -24(%rsi),%r12
|
|---|
| 1323 | mov -16(%rsi),%rbp
|
|---|
| 1324 | mov -8(%rsi),%rbx
|
|---|
| 1325 | lea (%rsi),%rsp
|
|---|
| 1326 | .Lmulx4x_epilogue:
|
|---|
| 1327 | ret
|
|---|
| 1328 | .size bn_mulx4x_mont,.-bn_mulx4x_mont
|
|---|
| 1329 | ___
|
|---|
| 1330 | }}}
|
|---|
| 1331 | $code.=<<___;
|
|---|
| 1332 | .asciz "Montgomery Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
|
|---|
| 1333 | .align 16
|
|---|
| 1334 | ___
|
|---|
| 1335 |
|
|---|
| 1336 | # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
|
|---|
| 1337 | # CONTEXT *context,DISPATCHER_CONTEXT *disp)
|
|---|
| 1338 | if ($win64) {
|
|---|
| 1339 | $rec="%rcx";
|
|---|
| 1340 | $frame="%rdx";
|
|---|
| 1341 | $context="%r8";
|
|---|
| 1342 | $disp="%r9";
|
|---|
| 1343 |
|
|---|
| 1344 | $code.=<<___;
|
|---|
| 1345 | .extern __imp_RtlVirtualUnwind
|
|---|
| 1346 | .type mul_handler,\@abi-omnipotent
|
|---|
| 1347 | .align 16
|
|---|
| 1348 | mul_handler:
|
|---|
| 1349 | push %rsi
|
|---|
| 1350 | push %rdi
|
|---|
| 1351 | push %rbx
|
|---|
| 1352 | push %rbp
|
|---|
| 1353 | push %r12
|
|---|
| 1354 | push %r13
|
|---|
| 1355 | push %r14
|
|---|
| 1356 | push %r15
|
|---|
| 1357 | pushfq
|
|---|
| 1358 | sub \$64,%rsp
|
|---|
| 1359 |
|
|---|
| 1360 | mov 120($context),%rax # pull context->Rax
|
|---|
| 1361 | mov 248($context),%rbx # pull context->Rip
|
|---|
| 1362 |
|
|---|
| 1363 | mov 8($disp),%rsi # disp->ImageBase
|
|---|
| 1364 | mov 56($disp),%r11 # disp->HandlerData
|
|---|
| 1365 |
|
|---|
| 1366 | mov 0(%r11),%r10d # HandlerData[0]
|
|---|
| 1367 | lea (%rsi,%r10),%r10 # end of prologue label
|
|---|
| 1368 | cmp %r10,%rbx # context->Rip<end of prologue label
|
|---|
| 1369 | jb .Lcommon_seh_tail
|
|---|
| 1370 |
|
|---|
| 1371 | mov 152($context),%rax # pull context->Rsp
|
|---|
| 1372 |
|
|---|
| 1373 | mov 4(%r11),%r10d # HandlerData[1]
|
|---|
| 1374 | lea (%rsi,%r10),%r10 # epilogue label
|
|---|
| 1375 | cmp %r10,%rbx # context->Rip>=epilogue label
|
|---|
| 1376 | jae .Lcommon_seh_tail
|
|---|
| 1377 |
|
|---|
| 1378 | mov 192($context),%r10 # pull $num
|
|---|
| 1379 | mov 8(%rax,%r10,8),%rax # pull saved stack pointer
|
|---|
| 1380 |
|
|---|
| 1381 | jmp .Lcommon_pop_regs
|
|---|
| 1382 | .size mul_handler,.-mul_handler
|
|---|
| 1383 |
|
|---|
| 1384 | .type sqr_handler,\@abi-omnipotent
|
|---|
| 1385 | .align 16
|
|---|
| 1386 | sqr_handler:
|
|---|
| 1387 | push %rsi
|
|---|
| 1388 | push %rdi
|
|---|
| 1389 | push %rbx
|
|---|
| 1390 | push %rbp
|
|---|
| 1391 | push %r12
|
|---|
| 1392 | push %r13
|
|---|
| 1393 | push %r14
|
|---|
| 1394 | push %r15
|
|---|
| 1395 | pushfq
|
|---|
| 1396 | sub \$64,%rsp
|
|---|
| 1397 |
|
|---|
| 1398 | mov 120($context),%rax # pull context->Rax
|
|---|
| 1399 | mov 248($context),%rbx # pull context->Rip
|
|---|
| 1400 |
|
|---|
| 1401 | mov 8($disp),%rsi # disp->ImageBase
|
|---|
| 1402 | mov 56($disp),%r11 # disp->HandlerData
|
|---|
| 1403 |
|
|---|
| 1404 | mov 0(%r11),%r10d # HandlerData[0]
|
|---|
| 1405 | lea (%rsi,%r10),%r10 # end of prologue label
|
|---|
| 1406 | cmp %r10,%rbx # context->Rip<.Lsqr_body
|
|---|
| 1407 | jb .Lcommon_seh_tail
|
|---|
| 1408 |
|
|---|
| 1409 | mov 4(%r11),%r10d # HandlerData[1]
|
|---|
| 1410 | lea (%rsi,%r10),%r10 # body label
|
|---|
| 1411 | cmp %r10,%rbx # context->Rip>=.Lsqr_epilogue
|
|---|
| 1412 | jb .Lcommon_pop_regs
|
|---|
| 1413 |
|
|---|
| 1414 | mov 152($context),%rax # pull context->Rsp
|
|---|
| 1415 |
|
|---|
| 1416 | mov 8(%r11),%r10d # HandlerData[2]
|
|---|
| 1417 | lea (%rsi,%r10),%r10 # epilogue label
|
|---|
| 1418 | cmp %r10,%rbx # context->Rip>=.Lsqr_epilogue
|
|---|
| 1419 | jae .Lcommon_seh_tail
|
|---|
| 1420 |
|
|---|
| 1421 | mov 40(%rax),%rax # pull saved stack pointer
|
|---|
| 1422 |
|
|---|
| 1423 | .Lcommon_pop_regs:
|
|---|
| 1424 | mov -8(%rax),%rbx
|
|---|
| 1425 | mov -16(%rax),%rbp
|
|---|
| 1426 | mov -24(%rax),%r12
|
|---|
| 1427 | mov -32(%rax),%r13
|
|---|
| 1428 | mov -40(%rax),%r14
|
|---|
| 1429 | mov -48(%rax),%r15
|
|---|
| 1430 | mov %rbx,144($context) # restore context->Rbx
|
|---|
| 1431 | mov %rbp,160($context) # restore context->Rbp
|
|---|
| 1432 | mov %r12,216($context) # restore context->R12
|
|---|
| 1433 | mov %r13,224($context) # restore context->R13
|
|---|
| 1434 | mov %r14,232($context) # restore context->R14
|
|---|
| 1435 | mov %r15,240($context) # restore context->R15
|
|---|
| 1436 |
|
|---|
| 1437 | .Lcommon_seh_tail:
|
|---|
| 1438 | mov 8(%rax),%rdi
|
|---|
| 1439 | mov 16(%rax),%rsi
|
|---|
| 1440 | mov %rax,152($context) # restore context->Rsp
|
|---|
| 1441 | mov %rsi,168($context) # restore context->Rsi
|
|---|
| 1442 | mov %rdi,176($context) # restore context->Rdi
|
|---|
| 1443 |
|
|---|
| 1444 | mov 40($disp),%rdi # disp->ContextRecord
|
|---|
| 1445 | mov $context,%rsi # context
|
|---|
| 1446 | mov \$154,%ecx # sizeof(CONTEXT)
|
|---|
| 1447 | .long 0xa548f3fc # cld; rep movsq
|
|---|
| 1448 |
|
|---|
| 1449 | mov $disp,%rsi
|
|---|
| 1450 | xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
|
|---|
| 1451 | mov 8(%rsi),%rdx # arg2, disp->ImageBase
|
|---|
| 1452 | mov 0(%rsi),%r8 # arg3, disp->ControlPc
|
|---|
| 1453 | mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
|
|---|
| 1454 | mov 40(%rsi),%r10 # disp->ContextRecord
|
|---|
| 1455 | lea 56(%rsi),%r11 # &disp->HandlerData
|
|---|
| 1456 | lea 24(%rsi),%r12 # &disp->EstablisherFrame
|
|---|
| 1457 | mov %r10,32(%rsp) # arg5
|
|---|
| 1458 | mov %r11,40(%rsp) # arg6
|
|---|
| 1459 | mov %r12,48(%rsp) # arg7
|
|---|
| 1460 | mov %rcx,56(%rsp) # arg8, (NULL)
|
|---|
| 1461 | call *__imp_RtlVirtualUnwind(%rip)
|
|---|
| 1462 |
|
|---|
| 1463 | mov \$1,%eax # ExceptionContinueSearch
|
|---|
| 1464 | add \$64,%rsp
|
|---|
| 1465 | popfq
|
|---|
| 1466 | pop %r15
|
|---|
| 1467 | pop %r14
|
|---|
| 1468 | pop %r13
|
|---|
| 1469 | pop %r12
|
|---|
| 1470 | pop %rbp
|
|---|
| 1471 | pop %rbx
|
|---|
| 1472 | pop %rdi
|
|---|
| 1473 | pop %rsi
|
|---|
| 1474 | ret
|
|---|
| 1475 | .size sqr_handler,.-sqr_handler
|
|---|
| 1476 |
|
|---|
| 1477 | .section .pdata
|
|---|
| 1478 | .align 4
|
|---|
| 1479 | .rva .LSEH_begin_bn_mul_mont
|
|---|
| 1480 | .rva .LSEH_end_bn_mul_mont
|
|---|
| 1481 | .rva .LSEH_info_bn_mul_mont
|
|---|
| 1482 |
|
|---|
| 1483 | .rva .LSEH_begin_bn_mul4x_mont
|
|---|
| 1484 | .rva .LSEH_end_bn_mul4x_mont
|
|---|
| 1485 | .rva .LSEH_info_bn_mul4x_mont
|
|---|
| 1486 |
|
|---|
| 1487 | .rva .LSEH_begin_bn_sqr8x_mont
|
|---|
| 1488 | .rva .LSEH_end_bn_sqr8x_mont
|
|---|
| 1489 | .rva .LSEH_info_bn_sqr8x_mont
|
|---|
| 1490 | ___
|
|---|
| 1491 | $code.=<<___ if ($addx);
|
|---|
| 1492 | .rva .LSEH_begin_bn_mulx4x_mont
|
|---|
| 1493 | .rva .LSEH_end_bn_mulx4x_mont
|
|---|
| 1494 | .rva .LSEH_info_bn_mulx4x_mont
|
|---|
| 1495 | ___
|
|---|
| 1496 | $code.=<<___;
|
|---|
| 1497 | .section .xdata
|
|---|
| 1498 | .align 8
|
|---|
| 1499 | .LSEH_info_bn_mul_mont:
|
|---|
| 1500 | .byte 9,0,0,0
|
|---|
| 1501 | .rva mul_handler
|
|---|
| 1502 | .rva .Lmul_body,.Lmul_epilogue # HandlerData[]
|
|---|
| 1503 | .LSEH_info_bn_mul4x_mont:
|
|---|
| 1504 | .byte 9,0,0,0
|
|---|
| 1505 | .rva mul_handler
|
|---|
| 1506 | .rva .Lmul4x_body,.Lmul4x_epilogue # HandlerData[]
|
|---|
| 1507 | .LSEH_info_bn_sqr8x_mont:
|
|---|
| 1508 | .byte 9,0,0,0
|
|---|
| 1509 | .rva sqr_handler
|
|---|
| 1510 | .rva .Lsqr8x_prologue,.Lsqr8x_body,.Lsqr8x_epilogue # HandlerData[]
|
|---|
| 1511 | .align 8
|
|---|
| 1512 | ___
|
|---|
| 1513 | $code.=<<___ if ($addx);
|
|---|
| 1514 | .LSEH_info_bn_mulx4x_mont:
|
|---|
| 1515 | .byte 9,0,0,0
|
|---|
| 1516 | .rva sqr_handler
|
|---|
| 1517 | .rva .Lmulx4x_prologue,.Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[]
|
|---|
| 1518 | .align 8
|
|---|
| 1519 | ___
|
|---|
| 1520 | }
|
|---|
| 1521 |
|
|---|
| 1522 | print $code;
|
|---|
| 1523 | close STDOUT;
|
|---|
