| 1 | #! /usr/bin/env perl
|
|---|
| 2 | # Copyright 2011-2020 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 | # August 2011.
|
|---|
| 18 | #
|
|---|
| 19 | # Companion to x86_64-mont.pl that optimizes cache-timing attack
|
|---|
| 20 | # countermeasures. The subroutines are produced by replacing bp[i]
|
|---|
| 21 | # references in their x86_64-mont.pl counterparts with cache-neutral
|
|---|
| 22 | # references to powers table computed in BN_mod_exp_mont_consttime.
|
|---|
| 23 | # In addition subroutine that scatters elements of the powers table
|
|---|
| 24 | # is implemented, so that scatter-/gathering can be tuned without
|
|---|
| 25 | # bn_exp.c modifications.
|
|---|
| 26 |
|
|---|
| 27 | # August 2013.
|
|---|
| 28 | #
|
|---|
| 29 | # Add MULX/AD*X code paths and additional interfaces to optimize for
|
|---|
| 30 | # branch prediction unit. For input lengths that are multiples of 8
|
|---|
| 31 | # the np argument is not just modulus value, but one interleaved
|
|---|
| 32 | # with 0. This is to optimize post-condition...
|
|---|
| 33 |
|
|---|
| 34 | $flavour = shift;
|
|---|
| 35 | $output = shift;
|
|---|
| 36 | if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
|
|---|
| 37 |
|
|---|
| 38 | $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
|
|---|
| 39 |
|
|---|
| 40 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
|
|---|
| 41 | ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
|
|---|
| 42 | ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
|
|---|
| 43 | die "can't locate x86_64-xlate.pl";
|
|---|
| 44 |
|
|---|
| 45 | open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
|
|---|
| 46 | *STDOUT=*OUT;
|
|---|
| 47 |
|
|---|
| 48 | if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
|
|---|
| 49 | =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
|
|---|
| 50 | $addx = ($1>=2.23);
|
|---|
| 51 | }
|
|---|
| 52 |
|
|---|
| 53 | if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
|
|---|
| 54 | `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
|
|---|
| 55 | $addx = ($1>=2.10);
|
|---|
| 56 | }
|
|---|
| 57 |
|
|---|
| 58 | if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
|
|---|
| 59 | `ml64 2>&1` =~ /Version ([0-9]+)\./) {
|
|---|
| 60 | $addx = ($1>=12);
|
|---|
| 61 | }
|
|---|
| 62 |
|
|---|
| 63 | if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:clang|LLVM) version|.*based on LLVM) ([0-9]+)\.([0-9]+)/) {
|
|---|
| 64 | my $ver = $2 + $3/100.0; # 3.1->3.01, 3.10->3.10
|
|---|
| 65 | $addx = ($ver>=3.03);
|
|---|
| 66 | }
|
|---|
| 67 |
|
|---|
| 68 | # int bn_mul_mont_gather5(
|
|---|
| 69 | $rp="%rdi"; # BN_ULONG *rp,
|
|---|
| 70 | $ap="%rsi"; # const BN_ULONG *ap,
|
|---|
| 71 | $bp="%rdx"; # const BN_ULONG *bp,
|
|---|
| 72 | $np="%rcx"; # const BN_ULONG *np,
|
|---|
| 73 | $n0="%r8"; # const BN_ULONG *n0,
|
|---|
| 74 | $num="%r9"; # int num,
|
|---|
| 75 | # int idx); # 0 to 2^5-1, "index" in $bp holding
|
|---|
| 76 | # pre-computed powers of a', interlaced
|
|---|
| 77 | # in such manner that b[0] is $bp[idx],
|
|---|
| 78 | # b[1] is [2^5+idx], etc.
|
|---|
| 79 | $lo0="%r10";
|
|---|
| 80 | $hi0="%r11";
|
|---|
| 81 | $hi1="%r13";
|
|---|
| 82 | $i="%r14";
|
|---|
| 83 | $j="%r15";
|
|---|
| 84 | $m0="%rbx";
|
|---|
| 85 | $m1="%rbp";
|
|---|
| 86 |
|
|---|
| 87 | $code=<<___;
|
|---|
| 88 | .text
|
|---|
| 89 |
|
|---|
| 90 | .extern OPENSSL_ia32cap_P
|
|---|
| 91 |
|
|---|
| 92 | .globl bn_mul_mont_gather5
|
|---|
| 93 | .type bn_mul_mont_gather5,\@function,6
|
|---|
| 94 | .align 64
|
|---|
| 95 | bn_mul_mont_gather5:
|
|---|
| 96 | .cfi_startproc
|
|---|
| 97 | mov ${num}d,${num}d
|
|---|
| 98 | mov %rsp,%rax
|
|---|
| 99 | .cfi_def_cfa_register %rax
|
|---|
| 100 | test \$7,${num}d
|
|---|
| 101 | jnz .Lmul_enter
|
|---|
| 102 | ___
|
|---|
| 103 | $code.=<<___ if ($addx);
|
|---|
| 104 | mov OPENSSL_ia32cap_P+8(%rip),%r11d
|
|---|
| 105 | ___
|
|---|
| 106 | $code.=<<___;
|
|---|
| 107 | jmp .Lmul4x_enter
|
|---|
| 108 |
|
|---|
| 109 | .align 16
|
|---|
| 110 | .Lmul_enter:
|
|---|
| 111 | movd `($win64?56:8)`(%rsp),%xmm5 # load 7th argument
|
|---|
| 112 | push %rbx
|
|---|
| 113 | .cfi_push %rbx
|
|---|
| 114 | push %rbp
|
|---|
| 115 | .cfi_push %rbp
|
|---|
| 116 | push %r12
|
|---|
| 117 | .cfi_push %r12
|
|---|
| 118 | push %r13
|
|---|
| 119 | .cfi_push %r13
|
|---|
| 120 | push %r14
|
|---|
| 121 | .cfi_push %r14
|
|---|
| 122 | push %r15
|
|---|
| 123 | .cfi_push %r15
|
|---|
| 124 |
|
|---|
| 125 | neg $num
|
|---|
| 126 | mov %rsp,%r11
|
|---|
| 127 | lea -280(%rsp,$num,8),%r10 # future alloca(8*(num+2)+256+8)
|
|---|
| 128 | neg $num # restore $num
|
|---|
| 129 | and \$-1024,%r10 # minimize TLB usage
|
|---|
| 130 |
|
|---|
| 131 | # An OS-agnostic version of __chkstk.
|
|---|
| 132 | #
|
|---|
| 133 | # Some OSes (Windows) insist on stack being "wired" to
|
|---|
| 134 | # physical memory in strictly sequential manner, i.e. if stack
|
|---|
| 135 | # allocation spans two pages, then reference to farmost one can
|
|---|
| 136 | # be punishable by SEGV. But page walking can do good even on
|
|---|
| 137 | # other OSes, because it guarantees that villain thread hits
|
|---|
| 138 | # the guard page before it can make damage to innocent one...
|
|---|
| 139 | sub %r10,%r11
|
|---|
| 140 | and \$-4096,%r11
|
|---|
| 141 | lea (%r10,%r11),%rsp
|
|---|
| 142 | mov (%rsp),%r11
|
|---|
| 143 | cmp %r10,%rsp
|
|---|
| 144 | ja .Lmul_page_walk
|
|---|
| 145 | jmp .Lmul_page_walk_done
|
|---|
| 146 |
|
|---|
| 147 | .Lmul_page_walk:
|
|---|
| 148 | lea -4096(%rsp),%rsp
|
|---|
| 149 | mov (%rsp),%r11
|
|---|
| 150 | cmp %r10,%rsp
|
|---|
| 151 | ja .Lmul_page_walk
|
|---|
| 152 | .Lmul_page_walk_done:
|
|---|
| 153 |
|
|---|
| 154 | lea .Linc(%rip),%r10
|
|---|
| 155 | mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
|
|---|
| 156 | .cfi_cfa_expression %rsp+8,$num,8,mul,plus,deref,+8
|
|---|
| 157 | .Lmul_body:
|
|---|
| 158 |
|
|---|
| 159 | lea 128($bp),%r12 # reassign $bp (+size optimization)
|
|---|
| 160 | ___
|
|---|
| 161 | $bp="%r12";
|
|---|
| 162 | $STRIDE=2**5*8; # 5 is "window size"
|
|---|
| 163 | $N=$STRIDE/4; # should match cache line size
|
|---|
| 164 | $code.=<<___;
|
|---|
| 165 | movdqa 0(%r10),%xmm0 # 00000001000000010000000000000000
|
|---|
| 166 | movdqa 16(%r10),%xmm1 # 00000002000000020000000200000002
|
|---|
| 167 | lea 24-112(%rsp,$num,8),%r10# place the mask after tp[num+3] (+ICache optimization)
|
|---|
| 168 | and \$-16,%r10
|
|---|
| 169 |
|
|---|
| 170 | pshufd \$0,%xmm5,%xmm5 # broadcast index
|
|---|
| 171 | movdqa %xmm1,%xmm4
|
|---|
| 172 | movdqa %xmm1,%xmm2
|
|---|
| 173 | ___
|
|---|
| 174 | ########################################################################
|
|---|
| 175 | # calculate mask by comparing 0..31 to index and save result to stack
|
|---|
| 176 | #
|
|---|
| 177 | $code.=<<___;
|
|---|
| 178 | paddd %xmm0,%xmm1
|
|---|
| 179 | pcmpeqd %xmm5,%xmm0 # compare to 1,0
|
|---|
| 180 | .byte 0x67
|
|---|
| 181 | movdqa %xmm4,%xmm3
|
|---|
| 182 | ___
|
|---|
| 183 | for($k=0;$k<$STRIDE/16-4;$k+=4) {
|
|---|
| 184 | $code.=<<___;
|
|---|
| 185 | paddd %xmm1,%xmm2
|
|---|
| 186 | pcmpeqd %xmm5,%xmm1 # compare to 3,2
|
|---|
| 187 | movdqa %xmm0,`16*($k+0)+112`(%r10)
|
|---|
| 188 | movdqa %xmm4,%xmm0
|
|---|
| 189 |
|
|---|
| 190 | paddd %xmm2,%xmm3
|
|---|
| 191 | pcmpeqd %xmm5,%xmm2 # compare to 5,4
|
|---|
| 192 | movdqa %xmm1,`16*($k+1)+112`(%r10)
|
|---|
| 193 | movdqa %xmm4,%xmm1
|
|---|
| 194 |
|
|---|
| 195 | paddd %xmm3,%xmm0
|
|---|
| 196 | pcmpeqd %xmm5,%xmm3 # compare to 7,6
|
|---|
| 197 | movdqa %xmm2,`16*($k+2)+112`(%r10)
|
|---|
| 198 | movdqa %xmm4,%xmm2
|
|---|
| 199 |
|
|---|
| 200 | paddd %xmm0,%xmm1
|
|---|
| 201 | pcmpeqd %xmm5,%xmm0
|
|---|
| 202 | movdqa %xmm3,`16*($k+3)+112`(%r10)
|
|---|
| 203 | movdqa %xmm4,%xmm3
|
|---|
| 204 | ___
|
|---|
| 205 | }
|
|---|
| 206 | $code.=<<___; # last iteration can be optimized
|
|---|
| 207 | paddd %xmm1,%xmm2
|
|---|
| 208 | pcmpeqd %xmm5,%xmm1
|
|---|
| 209 | movdqa %xmm0,`16*($k+0)+112`(%r10)
|
|---|
| 210 |
|
|---|
| 211 | paddd %xmm2,%xmm3
|
|---|
| 212 | .byte 0x67
|
|---|
| 213 | pcmpeqd %xmm5,%xmm2
|
|---|
| 214 | movdqa %xmm1,`16*($k+1)+112`(%r10)
|
|---|
| 215 |
|
|---|
| 216 | pcmpeqd %xmm5,%xmm3
|
|---|
| 217 | movdqa %xmm2,`16*($k+2)+112`(%r10)
|
|---|
| 218 | pand `16*($k+0)-128`($bp),%xmm0 # while it's still in register
|
|---|
| 219 |
|
|---|
| 220 | pand `16*($k+1)-128`($bp),%xmm1
|
|---|
| 221 | pand `16*($k+2)-128`($bp),%xmm2
|
|---|
| 222 | movdqa %xmm3,`16*($k+3)+112`(%r10)
|
|---|
| 223 | pand `16*($k+3)-128`($bp),%xmm3
|
|---|
| 224 | por %xmm2,%xmm0
|
|---|
| 225 | por %xmm3,%xmm1
|
|---|
| 226 | ___
|
|---|
| 227 | for($k=0;$k<$STRIDE/16-4;$k+=4) {
|
|---|
| 228 | $code.=<<___;
|
|---|
| 229 | movdqa `16*($k+0)-128`($bp),%xmm4
|
|---|
| 230 | movdqa `16*($k+1)-128`($bp),%xmm5
|
|---|
| 231 | movdqa `16*($k+2)-128`($bp),%xmm2
|
|---|
| 232 | pand `16*($k+0)+112`(%r10),%xmm4
|
|---|
| 233 | movdqa `16*($k+3)-128`($bp),%xmm3
|
|---|
| 234 | pand `16*($k+1)+112`(%r10),%xmm5
|
|---|
| 235 | por %xmm4,%xmm0
|
|---|
| 236 | pand `16*($k+2)+112`(%r10),%xmm2
|
|---|
| 237 | por %xmm5,%xmm1
|
|---|
| 238 | pand `16*($k+3)+112`(%r10),%xmm3
|
|---|
| 239 | por %xmm2,%xmm0
|
|---|
| 240 | por %xmm3,%xmm1
|
|---|
| 241 | ___
|
|---|
| 242 | }
|
|---|
| 243 | $code.=<<___;
|
|---|
| 244 | por %xmm1,%xmm0
|
|---|
| 245 | pshufd \$0x4e,%xmm0,%xmm1
|
|---|
| 246 | por %xmm1,%xmm0
|
|---|
| 247 | lea $STRIDE($bp),$bp
|
|---|
| 248 | movq %xmm0,$m0 # m0=bp[0]
|
|---|
| 249 |
|
|---|
| 250 | mov ($n0),$n0 # pull n0[0] value
|
|---|
| 251 | mov ($ap),%rax
|
|---|
| 252 |
|
|---|
| 253 | xor $i,$i # i=0
|
|---|
| 254 | xor $j,$j # j=0
|
|---|
| 255 |
|
|---|
| 256 | mov $n0,$m1
|
|---|
| 257 | mulq $m0 # ap[0]*bp[0]
|
|---|
| 258 | mov %rax,$lo0
|
|---|
| 259 | mov ($np),%rax
|
|---|
| 260 |
|
|---|
| 261 | imulq $lo0,$m1 # "tp[0]"*n0
|
|---|
| 262 | mov %rdx,$hi0
|
|---|
| 263 |
|
|---|
| 264 | mulq $m1 # np[0]*m1
|
|---|
| 265 | add %rax,$lo0 # discarded
|
|---|
| 266 | mov 8($ap),%rax
|
|---|
| 267 | adc \$0,%rdx
|
|---|
| 268 | mov %rdx,$hi1
|
|---|
| 269 |
|
|---|
| 270 | lea 1($j),$j # j++
|
|---|
| 271 | jmp .L1st_enter
|
|---|
| 272 |
|
|---|
| 273 | .align 16
|
|---|
| 274 | .L1st:
|
|---|
| 275 | add %rax,$hi1
|
|---|
| 276 | mov ($ap,$j,8),%rax
|
|---|
| 277 | adc \$0,%rdx
|
|---|
| 278 | add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
|
|---|
| 279 | mov $lo0,$hi0
|
|---|
| 280 | adc \$0,%rdx
|
|---|
| 281 | mov $hi1,-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 282 | mov %rdx,$hi1
|
|---|
| 283 |
|
|---|
| 284 | .L1st_enter:
|
|---|
| 285 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 286 | add %rax,$hi0
|
|---|
| 287 | mov ($np,$j,8),%rax
|
|---|
| 288 | adc \$0,%rdx
|
|---|
| 289 | lea 1($j),$j # j++
|
|---|
| 290 | mov %rdx,$lo0
|
|---|
| 291 |
|
|---|
| 292 | mulq $m1 # np[j]*m1
|
|---|
| 293 | cmp $num,$j
|
|---|
| 294 | jne .L1st # note that upon exit $j==$num, so
|
|---|
| 295 | # they can be used interchangeably
|
|---|
| 296 |
|
|---|
| 297 | add %rax,$hi1
|
|---|
| 298 | adc \$0,%rdx
|
|---|
| 299 | add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
|
|---|
| 300 | adc \$0,%rdx
|
|---|
| 301 | mov $hi1,-16(%rsp,$num,8) # tp[num-1]
|
|---|
| 302 | mov %rdx,$hi1
|
|---|
| 303 | mov $lo0,$hi0
|
|---|
| 304 |
|
|---|
| 305 | xor %rdx,%rdx
|
|---|
| 306 | add $hi0,$hi1
|
|---|
| 307 | adc \$0,%rdx
|
|---|
| 308 | mov $hi1,-8(%rsp,$num,8)
|
|---|
| 309 | mov %rdx,(%rsp,$num,8) # store upmost overflow bit
|
|---|
| 310 |
|
|---|
| 311 | lea 1($i),$i # i++
|
|---|
| 312 | jmp .Louter
|
|---|
| 313 | .align 16
|
|---|
| 314 | .Louter:
|
|---|
| 315 | lea 24+128(%rsp,$num,8),%rdx # where 256-byte mask is (+size optimization)
|
|---|
| 316 | and \$-16,%rdx
|
|---|
| 317 | pxor %xmm4,%xmm4
|
|---|
| 318 | pxor %xmm5,%xmm5
|
|---|
| 319 | ___
|
|---|
| 320 | for($k=0;$k<$STRIDE/16;$k+=4) {
|
|---|
| 321 | $code.=<<___;
|
|---|
| 322 | movdqa `16*($k+0)-128`($bp),%xmm0
|
|---|
| 323 | movdqa `16*($k+1)-128`($bp),%xmm1
|
|---|
| 324 | movdqa `16*($k+2)-128`($bp),%xmm2
|
|---|
| 325 | movdqa `16*($k+3)-128`($bp),%xmm3
|
|---|
| 326 | pand `16*($k+0)-128`(%rdx),%xmm0
|
|---|
| 327 | pand `16*($k+1)-128`(%rdx),%xmm1
|
|---|
| 328 | por %xmm0,%xmm4
|
|---|
| 329 | pand `16*($k+2)-128`(%rdx),%xmm2
|
|---|
| 330 | por %xmm1,%xmm5
|
|---|
| 331 | pand `16*($k+3)-128`(%rdx),%xmm3
|
|---|
| 332 | por %xmm2,%xmm4
|
|---|
| 333 | por %xmm3,%xmm5
|
|---|
| 334 | ___
|
|---|
| 335 | }
|
|---|
| 336 | $code.=<<___;
|
|---|
| 337 | por %xmm5,%xmm4
|
|---|
| 338 | pshufd \$0x4e,%xmm4,%xmm0
|
|---|
| 339 | por %xmm4,%xmm0
|
|---|
| 340 | lea $STRIDE($bp),$bp
|
|---|
| 341 |
|
|---|
| 342 | mov ($ap),%rax # ap[0]
|
|---|
| 343 | movq %xmm0,$m0 # m0=bp[i]
|
|---|
| 344 |
|
|---|
| 345 | xor $j,$j # j=0
|
|---|
| 346 | mov $n0,$m1
|
|---|
| 347 | mov (%rsp),$lo0
|
|---|
| 348 |
|
|---|
| 349 | mulq $m0 # ap[0]*bp[i]
|
|---|
| 350 | add %rax,$lo0 # ap[0]*bp[i]+tp[0]
|
|---|
| 351 | mov ($np),%rax
|
|---|
| 352 | adc \$0,%rdx
|
|---|
| 353 |
|
|---|
| 354 | imulq $lo0,$m1 # tp[0]*n0
|
|---|
| 355 | mov %rdx,$hi0
|
|---|
| 356 |
|
|---|
| 357 | mulq $m1 # np[0]*m1
|
|---|
| 358 | add %rax,$lo0 # discarded
|
|---|
| 359 | mov 8($ap),%rax
|
|---|
| 360 | adc \$0,%rdx
|
|---|
| 361 | mov 8(%rsp),$lo0 # tp[1]
|
|---|
| 362 | mov %rdx,$hi1
|
|---|
| 363 |
|
|---|
| 364 | lea 1($j),$j # j++
|
|---|
| 365 | jmp .Linner_enter
|
|---|
| 366 |
|
|---|
| 367 | .align 16
|
|---|
| 368 | .Linner:
|
|---|
| 369 | add %rax,$hi1
|
|---|
| 370 | mov ($ap,$j,8),%rax
|
|---|
| 371 | adc \$0,%rdx
|
|---|
| 372 | add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
|
|---|
| 373 | mov (%rsp,$j,8),$lo0
|
|---|
| 374 | adc \$0,%rdx
|
|---|
| 375 | mov $hi1,-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 376 | mov %rdx,$hi1
|
|---|
| 377 |
|
|---|
| 378 | .Linner_enter:
|
|---|
| 379 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 380 | add %rax,$hi0
|
|---|
| 381 | mov ($np,$j,8),%rax
|
|---|
| 382 | adc \$0,%rdx
|
|---|
| 383 | add $hi0,$lo0 # ap[j]*bp[i]+tp[j]
|
|---|
| 384 | mov %rdx,$hi0
|
|---|
| 385 | adc \$0,$hi0
|
|---|
| 386 | lea 1($j),$j # j++
|
|---|
| 387 |
|
|---|
| 388 | mulq $m1 # np[j]*m1
|
|---|
| 389 | cmp $num,$j
|
|---|
| 390 | jne .Linner # note that upon exit $j==$num, so
|
|---|
| 391 | # they can be used interchangeably
|
|---|
| 392 | add %rax,$hi1
|
|---|
| 393 | adc \$0,%rdx
|
|---|
| 394 | add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
|
|---|
| 395 | mov (%rsp,$num,8),$lo0
|
|---|
| 396 | adc \$0,%rdx
|
|---|
| 397 | mov $hi1,-16(%rsp,$num,8) # tp[num-1]
|
|---|
| 398 | mov %rdx,$hi1
|
|---|
| 399 |
|
|---|
| 400 | xor %rdx,%rdx
|
|---|
| 401 | add $hi0,$hi1
|
|---|
| 402 | adc \$0,%rdx
|
|---|
| 403 | add $lo0,$hi1 # pull upmost overflow bit
|
|---|
| 404 | adc \$0,%rdx
|
|---|
| 405 | mov $hi1,-8(%rsp,$num,8)
|
|---|
| 406 | mov %rdx,(%rsp,$num,8) # store upmost overflow bit
|
|---|
| 407 |
|
|---|
| 408 | lea 1($i),$i # i++
|
|---|
| 409 | cmp $num,$i
|
|---|
| 410 | jb .Louter
|
|---|
| 411 |
|
|---|
| 412 | xor $i,$i # i=0 and clear CF!
|
|---|
| 413 | mov (%rsp),%rax # tp[0]
|
|---|
| 414 | lea (%rsp),$ap # borrow ap for tp
|
|---|
| 415 | mov $num,$j # j=num
|
|---|
| 416 | jmp .Lsub
|
|---|
| 417 | .align 16
|
|---|
| 418 | .Lsub: sbb ($np,$i,8),%rax
|
|---|
| 419 | mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 420 | mov 8($ap,$i,8),%rax # tp[i+1]
|
|---|
| 421 | lea 1($i),$i # i++
|
|---|
| 422 | dec $j # doesn't affect CF!
|
|---|
| 423 | jnz .Lsub
|
|---|
| 424 |
|
|---|
| 425 | sbb \$0,%rax # handle upmost overflow bit
|
|---|
| 426 | mov \$-1,%rbx
|
|---|
| 427 | xor %rax,%rbx
|
|---|
| 428 | xor $i,$i
|
|---|
| 429 | mov $num,$j # j=num
|
|---|
| 430 |
|
|---|
| 431 | .Lcopy: # conditional copy
|
|---|
| 432 | mov ($rp,$i,8),%rcx
|
|---|
| 433 | mov (%rsp,$i,8),%rdx
|
|---|
| 434 | and %rbx,%rcx
|
|---|
| 435 | and %rax,%rdx
|
|---|
| 436 | mov $i,(%rsp,$i,8) # zap temporary vector
|
|---|
| 437 | or %rcx,%rdx
|
|---|
| 438 | mov %rdx,($rp,$i,8) # rp[i]=tp[i]
|
|---|
| 439 | lea 1($i),$i
|
|---|
| 440 | sub \$1,$j
|
|---|
| 441 | jnz .Lcopy
|
|---|
| 442 |
|
|---|
| 443 | mov 8(%rsp,$num,8),%rsi # restore %rsp
|
|---|
| 444 | .cfi_def_cfa %rsi,8
|
|---|
| 445 | mov \$1,%rax
|
|---|
| 446 |
|
|---|
| 447 | mov -48(%rsi),%r15
|
|---|
| 448 | .cfi_restore %r15
|
|---|
| 449 | mov -40(%rsi),%r14
|
|---|
| 450 | .cfi_restore %r14
|
|---|
| 451 | mov -32(%rsi),%r13
|
|---|
| 452 | .cfi_restore %r13
|
|---|
| 453 | mov -24(%rsi),%r12
|
|---|
| 454 | .cfi_restore %r12
|
|---|
| 455 | mov -16(%rsi),%rbp
|
|---|
| 456 | .cfi_restore %rbp
|
|---|
| 457 | mov -8(%rsi),%rbx
|
|---|
| 458 | .cfi_restore %rbx
|
|---|
| 459 | lea (%rsi),%rsp
|
|---|
| 460 | .cfi_def_cfa_register %rsp
|
|---|
| 461 | .Lmul_epilogue:
|
|---|
| 462 | ret
|
|---|
| 463 | .cfi_endproc
|
|---|
| 464 | .size bn_mul_mont_gather5,.-bn_mul_mont_gather5
|
|---|
| 465 | ___
|
|---|
| 466 | {{{
|
|---|
| 467 | my @A=("%r10","%r11");
|
|---|
| 468 | my @N=("%r13","%rdi");
|
|---|
| 469 | $code.=<<___;
|
|---|
| 470 | .type bn_mul4x_mont_gather5,\@function,6
|
|---|
| 471 | .align 32
|
|---|
| 472 | bn_mul4x_mont_gather5:
|
|---|
| 473 | .cfi_startproc
|
|---|
| 474 | .byte 0x67
|
|---|
| 475 | mov %rsp,%rax
|
|---|
| 476 | .cfi_def_cfa_register %rax
|
|---|
| 477 | .Lmul4x_enter:
|
|---|
| 478 | ___
|
|---|
| 479 | $code.=<<___ if ($addx);
|
|---|
| 480 | and \$0x80108,%r11d
|
|---|
| 481 | cmp \$0x80108,%r11d # check for AD*X+BMI2+BMI1
|
|---|
| 482 | je .Lmulx4x_enter
|
|---|
| 483 | ___
|
|---|
| 484 | $code.=<<___;
|
|---|
| 485 | push %rbx
|
|---|
| 486 | .cfi_push %rbx
|
|---|
| 487 | push %rbp
|
|---|
| 488 | .cfi_push %rbp
|
|---|
| 489 | push %r12
|
|---|
| 490 | .cfi_push %r12
|
|---|
| 491 | push %r13
|
|---|
| 492 | .cfi_push %r13
|
|---|
| 493 | push %r14
|
|---|
| 494 | .cfi_push %r14
|
|---|
| 495 | push %r15
|
|---|
| 496 | .cfi_push %r15
|
|---|
| 497 | .Lmul4x_prologue:
|
|---|
| 498 |
|
|---|
| 499 | .byte 0x67
|
|---|
| 500 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 501 | lea ($num,$num,2),%r10 # 3*$num in bytes
|
|---|
| 502 | neg $num # -$num
|
|---|
| 503 |
|
|---|
| 504 | ##############################################################
|
|---|
| 505 | # Ensure that stack frame doesn't alias with $rptr+3*$num
|
|---|
| 506 | # modulo 4096, which covers ret[num], am[num] and n[num]
|
|---|
| 507 | # (see bn_exp.c). This is done to allow memory disambiguation
|
|---|
| 508 | # logic do its magic. [Extra [num] is allocated in order
|
|---|
| 509 | # to align with bn_power5's frame, which is cleansed after
|
|---|
| 510 | # completing exponentiation. Extra 256 bytes is for power mask
|
|---|
| 511 | # calculated from 7th argument, the index.]
|
|---|
| 512 | #
|
|---|
| 513 | lea -320(%rsp,$num,2),%r11
|
|---|
| 514 | mov %rsp,%rbp
|
|---|
| 515 | sub $rp,%r11
|
|---|
| 516 | and \$4095,%r11
|
|---|
| 517 | cmp %r11,%r10
|
|---|
| 518 | jb .Lmul4xsp_alt
|
|---|
| 519 | sub %r11,%rbp # align with $rp
|
|---|
| 520 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*num*8+256)
|
|---|
| 521 | jmp .Lmul4xsp_done
|
|---|
| 522 |
|
|---|
| 523 | .align 32
|
|---|
| 524 | .Lmul4xsp_alt:
|
|---|
| 525 | lea 4096-320(,$num,2),%r10
|
|---|
| 526 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*num*8+256)
|
|---|
| 527 | sub %r10,%r11
|
|---|
| 528 | mov \$0,%r10
|
|---|
| 529 | cmovc %r10,%r11
|
|---|
| 530 | sub %r11,%rbp
|
|---|
| 531 | .Lmul4xsp_done:
|
|---|
| 532 | and \$-64,%rbp
|
|---|
| 533 | mov %rsp,%r11
|
|---|
| 534 | sub %rbp,%r11
|
|---|
| 535 | and \$-4096,%r11
|
|---|
| 536 | lea (%rbp,%r11),%rsp
|
|---|
| 537 | mov (%rsp),%r10
|
|---|
| 538 | cmp %rbp,%rsp
|
|---|
| 539 | ja .Lmul4x_page_walk
|
|---|
| 540 | jmp .Lmul4x_page_walk_done
|
|---|
| 541 |
|
|---|
| 542 | .Lmul4x_page_walk:
|
|---|
| 543 | lea -4096(%rsp),%rsp
|
|---|
| 544 | mov (%rsp),%r10
|
|---|
| 545 | cmp %rbp,%rsp
|
|---|
| 546 | ja .Lmul4x_page_walk
|
|---|
| 547 | .Lmul4x_page_walk_done:
|
|---|
| 548 |
|
|---|
| 549 | neg $num
|
|---|
| 550 |
|
|---|
| 551 | mov %rax,40(%rsp)
|
|---|
| 552 | .cfi_cfa_expression %rsp+40,deref,+8
|
|---|
| 553 | .Lmul4x_body:
|
|---|
| 554 |
|
|---|
| 555 | call mul4x_internal
|
|---|
| 556 |
|
|---|
| 557 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 558 | .cfi_def_cfa %rsi,8
|
|---|
| 559 | mov \$1,%rax
|
|---|
| 560 |
|
|---|
| 561 | mov -48(%rsi),%r15
|
|---|
| 562 | .cfi_restore %r15
|
|---|
| 563 | mov -40(%rsi),%r14
|
|---|
| 564 | .cfi_restore %r14
|
|---|
| 565 | mov -32(%rsi),%r13
|
|---|
| 566 | .cfi_restore %r13
|
|---|
| 567 | mov -24(%rsi),%r12
|
|---|
| 568 | .cfi_restore %r12
|
|---|
| 569 | mov -16(%rsi),%rbp
|
|---|
| 570 | .cfi_restore %rbp
|
|---|
| 571 | mov -8(%rsi),%rbx
|
|---|
| 572 | .cfi_restore %rbx
|
|---|
| 573 | lea (%rsi),%rsp
|
|---|
| 574 | .cfi_def_cfa_register %rsp
|
|---|
| 575 | .Lmul4x_epilogue:
|
|---|
| 576 | ret
|
|---|
| 577 | .cfi_endproc
|
|---|
| 578 | .size bn_mul4x_mont_gather5,.-bn_mul4x_mont_gather5
|
|---|
| 579 |
|
|---|
| 580 | .type mul4x_internal,\@abi-omnipotent
|
|---|
| 581 | .align 32
|
|---|
| 582 | mul4x_internal:
|
|---|
| 583 | .cfi_startproc
|
|---|
| 584 | shl \$5,$num # $num was in bytes
|
|---|
| 585 | movd `($win64?56:8)`(%rax),%xmm5 # load 7th argument, index
|
|---|
| 586 | lea .Linc(%rip),%rax
|
|---|
| 587 | lea 128(%rdx,$num),%r13 # end of powers table (+size optimization)
|
|---|
| 588 | shr \$5,$num # restore $num
|
|---|
| 589 | ___
|
|---|
| 590 | $bp="%r12";
|
|---|
| 591 | $STRIDE=2**5*8; # 5 is "window size"
|
|---|
| 592 | $N=$STRIDE/4; # should match cache line size
|
|---|
| 593 | $tp=$i;
|
|---|
| 594 | $code.=<<___;
|
|---|
| 595 | movdqa 0(%rax),%xmm0 # 00000001000000010000000000000000
|
|---|
| 596 | movdqa 16(%rax),%xmm1 # 00000002000000020000000200000002
|
|---|
| 597 | lea 88-112(%rsp,$num),%r10 # place the mask after tp[num+1] (+ICache optimization)
|
|---|
| 598 | lea 128(%rdx),$bp # size optimization
|
|---|
| 599 |
|
|---|
| 600 | pshufd \$0,%xmm5,%xmm5 # broadcast index
|
|---|
| 601 | movdqa %xmm1,%xmm4
|
|---|
| 602 | .byte 0x67,0x67
|
|---|
| 603 | movdqa %xmm1,%xmm2
|
|---|
| 604 | ___
|
|---|
| 605 | ########################################################################
|
|---|
| 606 | # calculate mask by comparing 0..31 to index and save result to stack
|
|---|
| 607 | #
|
|---|
| 608 | $code.=<<___;
|
|---|
| 609 | paddd %xmm0,%xmm1
|
|---|
| 610 | pcmpeqd %xmm5,%xmm0 # compare to 1,0
|
|---|
| 611 | .byte 0x67
|
|---|
| 612 | movdqa %xmm4,%xmm3
|
|---|
| 613 | ___
|
|---|
| 614 | for($i=0;$i<$STRIDE/16-4;$i+=4) {
|
|---|
| 615 | $code.=<<___;
|
|---|
| 616 | paddd %xmm1,%xmm2
|
|---|
| 617 | pcmpeqd %xmm5,%xmm1 # compare to 3,2
|
|---|
| 618 | movdqa %xmm0,`16*($i+0)+112`(%r10)
|
|---|
| 619 | movdqa %xmm4,%xmm0
|
|---|
| 620 |
|
|---|
| 621 | paddd %xmm2,%xmm3
|
|---|
| 622 | pcmpeqd %xmm5,%xmm2 # compare to 5,4
|
|---|
| 623 | movdqa %xmm1,`16*($i+1)+112`(%r10)
|
|---|
| 624 | movdqa %xmm4,%xmm1
|
|---|
| 625 |
|
|---|
| 626 | paddd %xmm3,%xmm0
|
|---|
| 627 | pcmpeqd %xmm5,%xmm3 # compare to 7,6
|
|---|
| 628 | movdqa %xmm2,`16*($i+2)+112`(%r10)
|
|---|
| 629 | movdqa %xmm4,%xmm2
|
|---|
| 630 |
|
|---|
| 631 | paddd %xmm0,%xmm1
|
|---|
| 632 | pcmpeqd %xmm5,%xmm0
|
|---|
| 633 | movdqa %xmm3,`16*($i+3)+112`(%r10)
|
|---|
| 634 | movdqa %xmm4,%xmm3
|
|---|
| 635 | ___
|
|---|
| 636 | }
|
|---|
| 637 | $code.=<<___; # last iteration can be optimized
|
|---|
| 638 | paddd %xmm1,%xmm2
|
|---|
| 639 | pcmpeqd %xmm5,%xmm1
|
|---|
| 640 | movdqa %xmm0,`16*($i+0)+112`(%r10)
|
|---|
| 641 |
|
|---|
| 642 | paddd %xmm2,%xmm3
|
|---|
| 643 | .byte 0x67
|
|---|
| 644 | pcmpeqd %xmm5,%xmm2
|
|---|
| 645 | movdqa %xmm1,`16*($i+1)+112`(%r10)
|
|---|
| 646 |
|
|---|
| 647 | pcmpeqd %xmm5,%xmm3
|
|---|
| 648 | movdqa %xmm2,`16*($i+2)+112`(%r10)
|
|---|
| 649 | pand `16*($i+0)-128`($bp),%xmm0 # while it's still in register
|
|---|
| 650 |
|
|---|
| 651 | pand `16*($i+1)-128`($bp),%xmm1
|
|---|
| 652 | pand `16*($i+2)-128`($bp),%xmm2
|
|---|
| 653 | movdqa %xmm3,`16*($i+3)+112`(%r10)
|
|---|
| 654 | pand `16*($i+3)-128`($bp),%xmm3
|
|---|
| 655 | por %xmm2,%xmm0
|
|---|
| 656 | por %xmm3,%xmm1
|
|---|
| 657 | ___
|
|---|
| 658 | for($i=0;$i<$STRIDE/16-4;$i+=4) {
|
|---|
| 659 | $code.=<<___;
|
|---|
| 660 | movdqa `16*($i+0)-128`($bp),%xmm4
|
|---|
| 661 | movdqa `16*($i+1)-128`($bp),%xmm5
|
|---|
| 662 | movdqa `16*($i+2)-128`($bp),%xmm2
|
|---|
| 663 | pand `16*($i+0)+112`(%r10),%xmm4
|
|---|
| 664 | movdqa `16*($i+3)-128`($bp),%xmm3
|
|---|
| 665 | pand `16*($i+1)+112`(%r10),%xmm5
|
|---|
| 666 | por %xmm4,%xmm0
|
|---|
| 667 | pand `16*($i+2)+112`(%r10),%xmm2
|
|---|
| 668 | por %xmm5,%xmm1
|
|---|
| 669 | pand `16*($i+3)+112`(%r10),%xmm3
|
|---|
| 670 | por %xmm2,%xmm0
|
|---|
| 671 | por %xmm3,%xmm1
|
|---|
| 672 | ___
|
|---|
| 673 | }
|
|---|
| 674 | $code.=<<___;
|
|---|
| 675 | por %xmm1,%xmm0
|
|---|
| 676 | pshufd \$0x4e,%xmm0,%xmm1
|
|---|
| 677 | por %xmm1,%xmm0
|
|---|
| 678 | lea $STRIDE($bp),$bp
|
|---|
| 679 | movq %xmm0,$m0 # m0=bp[0]
|
|---|
| 680 |
|
|---|
| 681 | mov %r13,16+8(%rsp) # save end of b[num]
|
|---|
| 682 | mov $rp, 56+8(%rsp) # save $rp
|
|---|
| 683 |
|
|---|
| 684 | mov ($n0),$n0 # pull n0[0] value
|
|---|
| 685 | mov ($ap),%rax
|
|---|
| 686 | lea ($ap,$num),$ap # end of a[num]
|
|---|
| 687 | neg $num
|
|---|
| 688 |
|
|---|
| 689 | mov $n0,$m1
|
|---|
| 690 | mulq $m0 # ap[0]*bp[0]
|
|---|
| 691 | mov %rax,$A[0]
|
|---|
| 692 | mov ($np),%rax
|
|---|
| 693 |
|
|---|
| 694 | imulq $A[0],$m1 # "tp[0]"*n0
|
|---|
| 695 | lea 64+8(%rsp),$tp
|
|---|
| 696 | mov %rdx,$A[1]
|
|---|
| 697 |
|
|---|
| 698 | mulq $m1 # np[0]*m1
|
|---|
| 699 | add %rax,$A[0] # discarded
|
|---|
| 700 | mov 8($ap,$num),%rax
|
|---|
| 701 | adc \$0,%rdx
|
|---|
| 702 | mov %rdx,$N[1]
|
|---|
| 703 |
|
|---|
| 704 | mulq $m0
|
|---|
| 705 | add %rax,$A[1]
|
|---|
| 706 | mov 8*1($np),%rax
|
|---|
| 707 | adc \$0,%rdx
|
|---|
| 708 | mov %rdx,$A[0]
|
|---|
| 709 |
|
|---|
| 710 | mulq $m1
|
|---|
| 711 | add %rax,$N[1]
|
|---|
| 712 | mov 16($ap,$num),%rax
|
|---|
| 713 | adc \$0,%rdx
|
|---|
| 714 | add $A[1],$N[1]
|
|---|
| 715 | lea 4*8($num),$j # j=4
|
|---|
| 716 | lea 8*4($np),$np
|
|---|
| 717 | adc \$0,%rdx
|
|---|
| 718 | mov $N[1],($tp)
|
|---|
| 719 | mov %rdx,$N[0]
|
|---|
| 720 | jmp .L1st4x
|
|---|
| 721 |
|
|---|
| 722 | .align 32
|
|---|
| 723 | .L1st4x:
|
|---|
| 724 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 725 | add %rax,$A[0]
|
|---|
| 726 | mov -8*2($np),%rax
|
|---|
| 727 | lea 32($tp),$tp
|
|---|
| 728 | adc \$0,%rdx
|
|---|
| 729 | mov %rdx,$A[1]
|
|---|
| 730 |
|
|---|
| 731 | mulq $m1 # np[j]*m1
|
|---|
| 732 | add %rax,$N[0]
|
|---|
| 733 | mov -8($ap,$j),%rax
|
|---|
| 734 | adc \$0,%rdx
|
|---|
| 735 | add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 736 | adc \$0,%rdx
|
|---|
| 737 | mov $N[0],-24($tp) # tp[j-1]
|
|---|
| 738 | mov %rdx,$N[1]
|
|---|
| 739 |
|
|---|
| 740 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 741 | add %rax,$A[1]
|
|---|
| 742 | mov -8*1($np),%rax
|
|---|
| 743 | adc \$0,%rdx
|
|---|
| 744 | mov %rdx,$A[0]
|
|---|
| 745 |
|
|---|
| 746 | mulq $m1 # np[j]*m1
|
|---|
| 747 | add %rax,$N[1]
|
|---|
| 748 | mov ($ap,$j),%rax
|
|---|
| 749 | adc \$0,%rdx
|
|---|
| 750 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 751 | adc \$0,%rdx
|
|---|
| 752 | mov $N[1],-16($tp) # tp[j-1]
|
|---|
| 753 | mov %rdx,$N[0]
|
|---|
| 754 |
|
|---|
| 755 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 756 | add %rax,$A[0]
|
|---|
| 757 | mov 8*0($np),%rax
|
|---|
| 758 | adc \$0,%rdx
|
|---|
| 759 | mov %rdx,$A[1]
|
|---|
| 760 |
|
|---|
| 761 | mulq $m1 # np[j]*m1
|
|---|
| 762 | add %rax,$N[0]
|
|---|
| 763 | mov 8($ap,$j),%rax
|
|---|
| 764 | adc \$0,%rdx
|
|---|
| 765 | add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 766 | adc \$0,%rdx
|
|---|
| 767 | mov $N[0],-8($tp) # tp[j-1]
|
|---|
| 768 | mov %rdx,$N[1]
|
|---|
| 769 |
|
|---|
| 770 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 771 | add %rax,$A[1]
|
|---|
| 772 | mov 8*1($np),%rax
|
|---|
| 773 | adc \$0,%rdx
|
|---|
| 774 | mov %rdx,$A[0]
|
|---|
| 775 |
|
|---|
| 776 | mulq $m1 # np[j]*m1
|
|---|
| 777 | add %rax,$N[1]
|
|---|
| 778 | mov 16($ap,$j),%rax
|
|---|
| 779 | adc \$0,%rdx
|
|---|
| 780 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 781 | lea 8*4($np),$np
|
|---|
| 782 | adc \$0,%rdx
|
|---|
| 783 | mov $N[1],($tp) # tp[j-1]
|
|---|
| 784 | mov %rdx,$N[0]
|
|---|
| 785 |
|
|---|
| 786 | add \$32,$j # j+=4
|
|---|
| 787 | jnz .L1st4x
|
|---|
| 788 |
|
|---|
| 789 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 790 | add %rax,$A[0]
|
|---|
| 791 | mov -8*2($np),%rax
|
|---|
| 792 | lea 32($tp),$tp
|
|---|
| 793 | adc \$0,%rdx
|
|---|
| 794 | mov %rdx,$A[1]
|
|---|
| 795 |
|
|---|
| 796 | mulq $m1 # np[j]*m1
|
|---|
| 797 | add %rax,$N[0]
|
|---|
| 798 | mov -8($ap),%rax
|
|---|
| 799 | adc \$0,%rdx
|
|---|
| 800 | add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 801 | adc \$0,%rdx
|
|---|
| 802 | mov $N[0],-24($tp) # tp[j-1]
|
|---|
| 803 | mov %rdx,$N[1]
|
|---|
| 804 |
|
|---|
| 805 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 806 | add %rax,$A[1]
|
|---|
| 807 | mov -8*1($np),%rax
|
|---|
| 808 | adc \$0,%rdx
|
|---|
| 809 | mov %rdx,$A[0]
|
|---|
| 810 |
|
|---|
| 811 | mulq $m1 # np[j]*m1
|
|---|
| 812 | add %rax,$N[1]
|
|---|
| 813 | mov ($ap,$num),%rax # ap[0]
|
|---|
| 814 | adc \$0,%rdx
|
|---|
| 815 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 816 | adc \$0,%rdx
|
|---|
| 817 | mov $N[1],-16($tp) # tp[j-1]
|
|---|
| 818 | mov %rdx,$N[0]
|
|---|
| 819 |
|
|---|
| 820 | lea ($np,$num),$np # rewind $np
|
|---|
| 821 |
|
|---|
| 822 | xor $N[1],$N[1]
|
|---|
| 823 | add $A[0],$N[0]
|
|---|
| 824 | adc \$0,$N[1]
|
|---|
| 825 | mov $N[0],-8($tp)
|
|---|
| 826 |
|
|---|
| 827 | jmp .Louter4x
|
|---|
| 828 |
|
|---|
| 829 | .align 32
|
|---|
| 830 | .Louter4x:
|
|---|
| 831 | lea 16+128($tp),%rdx # where 256-byte mask is (+size optimization)
|
|---|
| 832 | pxor %xmm4,%xmm4
|
|---|
| 833 | pxor %xmm5,%xmm5
|
|---|
| 834 | ___
|
|---|
| 835 | for($i=0;$i<$STRIDE/16;$i+=4) {
|
|---|
| 836 | $code.=<<___;
|
|---|
| 837 | movdqa `16*($i+0)-128`($bp),%xmm0
|
|---|
| 838 | movdqa `16*($i+1)-128`($bp),%xmm1
|
|---|
| 839 | movdqa `16*($i+2)-128`($bp),%xmm2
|
|---|
| 840 | movdqa `16*($i+3)-128`($bp),%xmm3
|
|---|
| 841 | pand `16*($i+0)-128`(%rdx),%xmm0
|
|---|
| 842 | pand `16*($i+1)-128`(%rdx),%xmm1
|
|---|
| 843 | por %xmm0,%xmm4
|
|---|
| 844 | pand `16*($i+2)-128`(%rdx),%xmm2
|
|---|
| 845 | por %xmm1,%xmm5
|
|---|
| 846 | pand `16*($i+3)-128`(%rdx),%xmm3
|
|---|
| 847 | por %xmm2,%xmm4
|
|---|
| 848 | por %xmm3,%xmm5
|
|---|
| 849 | ___
|
|---|
| 850 | }
|
|---|
| 851 | $code.=<<___;
|
|---|
| 852 | por %xmm5,%xmm4
|
|---|
| 853 | pshufd \$0x4e,%xmm4,%xmm0
|
|---|
| 854 | por %xmm4,%xmm0
|
|---|
| 855 | lea $STRIDE($bp),$bp
|
|---|
| 856 | movq %xmm0,$m0 # m0=bp[i]
|
|---|
| 857 |
|
|---|
| 858 | mov ($tp,$num),$A[0]
|
|---|
| 859 | mov $n0,$m1
|
|---|
| 860 | mulq $m0 # ap[0]*bp[i]
|
|---|
| 861 | add %rax,$A[0] # ap[0]*bp[i]+tp[0]
|
|---|
| 862 | mov ($np),%rax
|
|---|
| 863 | adc \$0,%rdx
|
|---|
| 864 |
|
|---|
| 865 | imulq $A[0],$m1 # tp[0]*n0
|
|---|
| 866 | mov %rdx,$A[1]
|
|---|
| 867 | mov $N[1],($tp) # store upmost overflow bit
|
|---|
| 868 |
|
|---|
| 869 | lea ($tp,$num),$tp # rewind $tp
|
|---|
| 870 |
|
|---|
| 871 | mulq $m1 # np[0]*m1
|
|---|
| 872 | add %rax,$A[0] # "$N[0]", discarded
|
|---|
| 873 | mov 8($ap,$num),%rax
|
|---|
| 874 | adc \$0,%rdx
|
|---|
| 875 | mov %rdx,$N[1]
|
|---|
| 876 |
|
|---|
| 877 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 878 | add %rax,$A[1]
|
|---|
| 879 | mov 8*1($np),%rax
|
|---|
| 880 | adc \$0,%rdx
|
|---|
| 881 | add 8($tp),$A[1] # +tp[1]
|
|---|
| 882 | adc \$0,%rdx
|
|---|
| 883 | mov %rdx,$A[0]
|
|---|
| 884 |
|
|---|
| 885 | mulq $m1 # np[j]*m1
|
|---|
| 886 | add %rax,$N[1]
|
|---|
| 887 | mov 16($ap,$num),%rax
|
|---|
| 888 | adc \$0,%rdx
|
|---|
| 889 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
|
|---|
| 890 | lea 4*8($num),$j # j=4
|
|---|
| 891 | lea 8*4($np),$np
|
|---|
| 892 | adc \$0,%rdx
|
|---|
| 893 | mov %rdx,$N[0]
|
|---|
| 894 | jmp .Linner4x
|
|---|
| 895 |
|
|---|
| 896 | .align 32
|
|---|
| 897 | .Linner4x:
|
|---|
| 898 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 899 | add %rax,$A[0]
|
|---|
| 900 | mov -8*2($np),%rax
|
|---|
| 901 | adc \$0,%rdx
|
|---|
| 902 | add 16($tp),$A[0] # ap[j]*bp[i]+tp[j]
|
|---|
| 903 | lea 32($tp),$tp
|
|---|
| 904 | adc \$0,%rdx
|
|---|
| 905 | mov %rdx,$A[1]
|
|---|
| 906 |
|
|---|
| 907 | mulq $m1 # np[j]*m1
|
|---|
| 908 | add %rax,$N[0]
|
|---|
| 909 | mov -8($ap,$j),%rax
|
|---|
| 910 | adc \$0,%rdx
|
|---|
| 911 | add $A[0],$N[0]
|
|---|
| 912 | adc \$0,%rdx
|
|---|
| 913 | mov $N[1],-32($tp) # tp[j-1]
|
|---|
| 914 | mov %rdx,$N[1]
|
|---|
| 915 |
|
|---|
| 916 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 917 | add %rax,$A[1]
|
|---|
| 918 | mov -8*1($np),%rax
|
|---|
| 919 | adc \$0,%rdx
|
|---|
| 920 | add -8($tp),$A[1]
|
|---|
| 921 | adc \$0,%rdx
|
|---|
| 922 | mov %rdx,$A[0]
|
|---|
| 923 |
|
|---|
| 924 | mulq $m1 # np[j]*m1
|
|---|
| 925 | add %rax,$N[1]
|
|---|
| 926 | mov ($ap,$j),%rax
|
|---|
| 927 | adc \$0,%rdx
|
|---|
| 928 | add $A[1],$N[1]
|
|---|
| 929 | adc \$0,%rdx
|
|---|
| 930 | mov $N[0],-24($tp) # tp[j-1]
|
|---|
| 931 | mov %rdx,$N[0]
|
|---|
| 932 |
|
|---|
| 933 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 934 | add %rax,$A[0]
|
|---|
| 935 | mov 8*0($np),%rax
|
|---|
| 936 | adc \$0,%rdx
|
|---|
| 937 | add ($tp),$A[0] # ap[j]*bp[i]+tp[j]
|
|---|
| 938 | adc \$0,%rdx
|
|---|
| 939 | mov %rdx,$A[1]
|
|---|
| 940 |
|
|---|
| 941 | mulq $m1 # np[j]*m1
|
|---|
| 942 | add %rax,$N[0]
|
|---|
| 943 | mov 8($ap,$j),%rax
|
|---|
| 944 | adc \$0,%rdx
|
|---|
| 945 | add $A[0],$N[0]
|
|---|
| 946 | adc \$0,%rdx
|
|---|
| 947 | mov $N[1],-16($tp) # tp[j-1]
|
|---|
| 948 | mov %rdx,$N[1]
|
|---|
| 949 |
|
|---|
| 950 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 951 | add %rax,$A[1]
|
|---|
| 952 | mov 8*1($np),%rax
|
|---|
| 953 | adc \$0,%rdx
|
|---|
| 954 | add 8($tp),$A[1]
|
|---|
| 955 | adc \$0,%rdx
|
|---|
| 956 | mov %rdx,$A[0]
|
|---|
| 957 |
|
|---|
| 958 | mulq $m1 # np[j]*m1
|
|---|
| 959 | add %rax,$N[1]
|
|---|
| 960 | mov 16($ap,$j),%rax
|
|---|
| 961 | adc \$0,%rdx
|
|---|
| 962 | add $A[1],$N[1]
|
|---|
| 963 | lea 8*4($np),$np
|
|---|
| 964 | adc \$0,%rdx
|
|---|
| 965 | mov $N[0],-8($tp) # tp[j-1]
|
|---|
| 966 | mov %rdx,$N[0]
|
|---|
| 967 |
|
|---|
| 968 | add \$32,$j # j+=4
|
|---|
| 969 | jnz .Linner4x
|
|---|
| 970 |
|
|---|
| 971 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 972 | add %rax,$A[0]
|
|---|
| 973 | mov -8*2($np),%rax
|
|---|
| 974 | adc \$0,%rdx
|
|---|
| 975 | add 16($tp),$A[0] # ap[j]*bp[i]+tp[j]
|
|---|
| 976 | lea 32($tp),$tp
|
|---|
| 977 | adc \$0,%rdx
|
|---|
| 978 | mov %rdx,$A[1]
|
|---|
| 979 |
|
|---|
| 980 | mulq $m1 # np[j]*m1
|
|---|
| 981 | add %rax,$N[0]
|
|---|
| 982 | mov -8($ap),%rax
|
|---|
| 983 | adc \$0,%rdx
|
|---|
| 984 | add $A[0],$N[0]
|
|---|
| 985 | adc \$0,%rdx
|
|---|
| 986 | mov $N[1],-32($tp) # tp[j-1]
|
|---|
| 987 | mov %rdx,$N[1]
|
|---|
| 988 |
|
|---|
| 989 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 990 | add %rax,$A[1]
|
|---|
| 991 | mov $m1,%rax
|
|---|
| 992 | mov -8*1($np),$m1
|
|---|
| 993 | adc \$0,%rdx
|
|---|
| 994 | add -8($tp),$A[1]
|
|---|
| 995 | adc \$0,%rdx
|
|---|
| 996 | mov %rdx,$A[0]
|
|---|
| 997 |
|
|---|
| 998 | mulq $m1 # np[j]*m1
|
|---|
| 999 | add %rax,$N[1]
|
|---|
| 1000 | mov ($ap,$num),%rax # ap[0]
|
|---|
| 1001 | adc \$0,%rdx
|
|---|
| 1002 | add $A[1],$N[1]
|
|---|
| 1003 | adc \$0,%rdx
|
|---|
| 1004 | mov $N[0],-24($tp) # tp[j-1]
|
|---|
| 1005 | mov %rdx,$N[0]
|
|---|
| 1006 |
|
|---|
| 1007 | mov $N[1],-16($tp) # tp[j-1]
|
|---|
| 1008 | lea ($np,$num),$np # rewind $np
|
|---|
| 1009 |
|
|---|
| 1010 | xor $N[1],$N[1]
|
|---|
| 1011 | add $A[0],$N[0]
|
|---|
| 1012 | adc \$0,$N[1]
|
|---|
| 1013 | add ($tp),$N[0] # pull upmost overflow bit
|
|---|
| 1014 | adc \$0,$N[1] # upmost overflow bit
|
|---|
| 1015 | mov $N[0],-8($tp)
|
|---|
| 1016 |
|
|---|
| 1017 | cmp 16+8(%rsp),$bp
|
|---|
| 1018 | jb .Louter4x
|
|---|
| 1019 | ___
|
|---|
| 1020 | if (1) {
|
|---|
| 1021 | $code.=<<___;
|
|---|
| 1022 | xor %rax,%rax
|
|---|
| 1023 | sub $N[0],$m1 # compare top-most words
|
|---|
| 1024 | adc $j,$j # $j is zero
|
|---|
| 1025 | or $j,$N[1]
|
|---|
| 1026 | sub $N[1],%rax # %rax=-$N[1]
|
|---|
| 1027 | lea ($tp,$num),%rbx # tptr in .sqr4x_sub
|
|---|
| 1028 | mov ($np),%r12
|
|---|
| 1029 | lea ($np),%rbp # nptr in .sqr4x_sub
|
|---|
| 1030 | mov %r9,%rcx
|
|---|
| 1031 | sar \$3+2,%rcx
|
|---|
| 1032 | mov 56+8(%rsp),%rdi # rptr in .sqr4x_sub
|
|---|
| 1033 | dec %r12 # so that after 'not' we get -n[0]
|
|---|
| 1034 | xor %r10,%r10
|
|---|
| 1035 | mov 8*1(%rbp),%r13
|
|---|
| 1036 | mov 8*2(%rbp),%r14
|
|---|
| 1037 | mov 8*3(%rbp),%r15
|
|---|
| 1038 | jmp .Lsqr4x_sub_entry
|
|---|
| 1039 | ___
|
|---|
| 1040 | } else {
|
|---|
| 1041 | my @ri=("%rax",$bp,$m0,$m1);
|
|---|
| 1042 | my $rp="%rdx";
|
|---|
| 1043 | $code.=<<___
|
|---|
| 1044 | xor \$1,$N[1]
|
|---|
| 1045 | lea ($tp,$num),$tp # rewind $tp
|
|---|
| 1046 | sar \$5,$num # cf=0
|
|---|
| 1047 | lea ($np,$N[1],8),$np
|
|---|
| 1048 | mov 56+8(%rsp),$rp # restore $rp
|
|---|
| 1049 | jmp .Lsub4x
|
|---|
| 1050 |
|
|---|
| 1051 | .align 32
|
|---|
| 1052 | .Lsub4x:
|
|---|
| 1053 | .byte 0x66
|
|---|
| 1054 | mov 8*0($tp),@ri[0]
|
|---|
| 1055 | mov 8*1($tp),@ri[1]
|
|---|
| 1056 | .byte 0x66
|
|---|
| 1057 | sbb 16*0($np),@ri[0]
|
|---|
| 1058 | mov 8*2($tp),@ri[2]
|
|---|
| 1059 | sbb 16*1($np),@ri[1]
|
|---|
| 1060 | mov 3*8($tp),@ri[3]
|
|---|
| 1061 | lea 4*8($tp),$tp
|
|---|
| 1062 | sbb 16*2($np),@ri[2]
|
|---|
| 1063 | mov @ri[0],8*0($rp)
|
|---|
| 1064 | sbb 16*3($np),@ri[3]
|
|---|
| 1065 | lea 16*4($np),$np
|
|---|
| 1066 | mov @ri[1],8*1($rp)
|
|---|
| 1067 | mov @ri[2],8*2($rp)
|
|---|
| 1068 | mov @ri[3],8*3($rp)
|
|---|
| 1069 | lea 8*4($rp),$rp
|
|---|
| 1070 |
|
|---|
| 1071 | inc $num
|
|---|
| 1072 | jnz .Lsub4x
|
|---|
| 1073 |
|
|---|
| 1074 | ret
|
|---|
| 1075 | ___
|
|---|
| 1076 | }
|
|---|
| 1077 | $code.=<<___;
|
|---|
| 1078 | .cfi_endproc
|
|---|
| 1079 | .size mul4x_internal,.-mul4x_internal
|
|---|
| 1080 | ___
|
|---|
| 1081 | }}}
|
|---|
| 1082 | |
|---|
| 1083 | {{{
|
|---|
| 1084 | ######################################################################
|
|---|
| 1085 | # void bn_power5(
|
|---|
| 1086 | my $rptr="%rdi"; # BN_ULONG *rptr,
|
|---|
| 1087 | my $aptr="%rsi"; # const BN_ULONG *aptr,
|
|---|
| 1088 | my $bptr="%rdx"; # const void *table,
|
|---|
| 1089 | my $nptr="%rcx"; # const BN_ULONG *nptr,
|
|---|
| 1090 | my $n0 ="%r8"; # const BN_ULONG *n0);
|
|---|
| 1091 | my $num ="%r9"; # int num, has to be divisible by 8
|
|---|
| 1092 | # int pwr
|
|---|
| 1093 |
|
|---|
| 1094 | my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
|
|---|
| 1095 | my @A0=("%r10","%r11");
|
|---|
| 1096 | my @A1=("%r12","%r13");
|
|---|
| 1097 | my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
|
|---|
| 1098 |
|
|---|
| 1099 | $code.=<<___;
|
|---|
| 1100 | .globl bn_power5
|
|---|
| 1101 | .type bn_power5,\@function,6
|
|---|
| 1102 | .align 32
|
|---|
| 1103 | bn_power5:
|
|---|
| 1104 | .cfi_startproc
|
|---|
| 1105 | mov %rsp,%rax
|
|---|
| 1106 | .cfi_def_cfa_register %rax
|
|---|
| 1107 | ___
|
|---|
| 1108 | $code.=<<___ if ($addx);
|
|---|
| 1109 | mov OPENSSL_ia32cap_P+8(%rip),%r11d
|
|---|
| 1110 | and \$0x80108,%r11d
|
|---|
| 1111 | cmp \$0x80108,%r11d # check for AD*X+BMI2+BMI1
|
|---|
| 1112 | je .Lpowerx5_enter
|
|---|
| 1113 | ___
|
|---|
| 1114 | $code.=<<___;
|
|---|
| 1115 | push %rbx
|
|---|
| 1116 | .cfi_push %rbx
|
|---|
| 1117 | push %rbp
|
|---|
| 1118 | .cfi_push %rbp
|
|---|
| 1119 | push %r12
|
|---|
| 1120 | .cfi_push %r12
|
|---|
| 1121 | push %r13
|
|---|
| 1122 | .cfi_push %r13
|
|---|
| 1123 | push %r14
|
|---|
| 1124 | .cfi_push %r14
|
|---|
| 1125 | push %r15
|
|---|
| 1126 | .cfi_push %r15
|
|---|
| 1127 | .Lpower5_prologue:
|
|---|
| 1128 |
|
|---|
| 1129 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 1130 | lea ($num,$num,2),%r10d # 3*$num
|
|---|
| 1131 | neg $num
|
|---|
| 1132 | mov ($n0),$n0 # *n0
|
|---|
| 1133 |
|
|---|
| 1134 | ##############################################################
|
|---|
| 1135 | # Ensure that stack frame doesn't alias with $rptr+3*$num
|
|---|
| 1136 | # modulo 4096, which covers ret[num], am[num] and n[num]
|
|---|
| 1137 | # (see bn_exp.c). This is done to allow memory disambiguation
|
|---|
| 1138 | # logic do its magic. [Extra 256 bytes is for power mask
|
|---|
| 1139 | # calculated from 7th argument, the index.]
|
|---|
| 1140 | #
|
|---|
| 1141 | lea -320(%rsp,$num,2),%r11
|
|---|
| 1142 | mov %rsp,%rbp
|
|---|
| 1143 | sub $rptr,%r11
|
|---|
| 1144 | and \$4095,%r11
|
|---|
| 1145 | cmp %r11,%r10
|
|---|
| 1146 | jb .Lpwr_sp_alt
|
|---|
| 1147 | sub %r11,%rbp # align with $aptr
|
|---|
| 1148 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*num*8+256)
|
|---|
| 1149 | jmp .Lpwr_sp_done
|
|---|
| 1150 |
|
|---|
| 1151 | .align 32
|
|---|
| 1152 | .Lpwr_sp_alt:
|
|---|
| 1153 | lea 4096-320(,$num,2),%r10
|
|---|
| 1154 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*num*8+256)
|
|---|
| 1155 | sub %r10,%r11
|
|---|
| 1156 | mov \$0,%r10
|
|---|
| 1157 | cmovc %r10,%r11
|
|---|
| 1158 | sub %r11,%rbp
|
|---|
| 1159 | .Lpwr_sp_done:
|
|---|
| 1160 | and \$-64,%rbp
|
|---|
| 1161 | mov %rsp,%r11
|
|---|
| 1162 | sub %rbp,%r11
|
|---|
| 1163 | and \$-4096,%r11
|
|---|
| 1164 | lea (%rbp,%r11),%rsp
|
|---|
| 1165 | mov (%rsp),%r10
|
|---|
| 1166 | cmp %rbp,%rsp
|
|---|
| 1167 | ja .Lpwr_page_walk
|
|---|
| 1168 | jmp .Lpwr_page_walk_done
|
|---|
| 1169 |
|
|---|
| 1170 | .Lpwr_page_walk:
|
|---|
| 1171 | lea -4096(%rsp),%rsp
|
|---|
| 1172 | mov (%rsp),%r10
|
|---|
| 1173 | cmp %rbp,%rsp
|
|---|
| 1174 | ja .Lpwr_page_walk
|
|---|
| 1175 | .Lpwr_page_walk_done:
|
|---|
| 1176 |
|
|---|
| 1177 | mov $num,%r10
|
|---|
| 1178 | neg $num
|
|---|
| 1179 |
|
|---|
| 1180 | ##############################################################
|
|---|
| 1181 | # Stack layout
|
|---|
| 1182 | #
|
|---|
| 1183 | # +0 saved $num, used in reduction section
|
|---|
| 1184 | # +8 &t[2*$num], used in reduction section
|
|---|
| 1185 | # +32 saved *n0
|
|---|
| 1186 | # +40 saved %rsp
|
|---|
| 1187 | # +48 t[2*$num]
|
|---|
| 1188 | #
|
|---|
| 1189 | mov $n0, 32(%rsp)
|
|---|
| 1190 | mov %rax, 40(%rsp) # save original %rsp
|
|---|
| 1191 | .cfi_cfa_expression %rsp+40,deref,+8
|
|---|
| 1192 | .Lpower5_body:
|
|---|
| 1193 | movq $rptr,%xmm1 # save $rptr, used in sqr8x
|
|---|
| 1194 | movq $nptr,%xmm2 # save $nptr
|
|---|
| 1195 | movq %r10, %xmm3 # -$num, used in sqr8x
|
|---|
| 1196 | movq $bptr,%xmm4
|
|---|
| 1197 |
|
|---|
| 1198 | call __bn_sqr8x_internal
|
|---|
| 1199 | call __bn_post4x_internal
|
|---|
| 1200 | call __bn_sqr8x_internal
|
|---|
| 1201 | call __bn_post4x_internal
|
|---|
| 1202 | call __bn_sqr8x_internal
|
|---|
| 1203 | call __bn_post4x_internal
|
|---|
| 1204 | call __bn_sqr8x_internal
|
|---|
| 1205 | call __bn_post4x_internal
|
|---|
| 1206 | call __bn_sqr8x_internal
|
|---|
| 1207 | call __bn_post4x_internal
|
|---|
| 1208 |
|
|---|
| 1209 | movq %xmm2,$nptr
|
|---|
| 1210 | movq %xmm4,$bptr
|
|---|
| 1211 | mov $aptr,$rptr
|
|---|
| 1212 | mov 40(%rsp),%rax
|
|---|
| 1213 | lea 32(%rsp),$n0
|
|---|
| 1214 |
|
|---|
| 1215 | call mul4x_internal
|
|---|
| 1216 |
|
|---|
| 1217 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 1218 | .cfi_def_cfa %rsi,8
|
|---|
| 1219 | mov \$1,%rax
|
|---|
| 1220 | mov -48(%rsi),%r15
|
|---|
| 1221 | .cfi_restore %r15
|
|---|
| 1222 | mov -40(%rsi),%r14
|
|---|
| 1223 | .cfi_restore %r14
|
|---|
| 1224 | mov -32(%rsi),%r13
|
|---|
| 1225 | .cfi_restore %r13
|
|---|
| 1226 | mov -24(%rsi),%r12
|
|---|
| 1227 | .cfi_restore %r12
|
|---|
| 1228 | mov -16(%rsi),%rbp
|
|---|
| 1229 | .cfi_restore %rbp
|
|---|
| 1230 | mov -8(%rsi),%rbx
|
|---|
| 1231 | .cfi_restore %rbx
|
|---|
| 1232 | lea (%rsi),%rsp
|
|---|
| 1233 | .cfi_def_cfa_register %rsp
|
|---|
| 1234 | .Lpower5_epilogue:
|
|---|
| 1235 | ret
|
|---|
| 1236 | .cfi_endproc
|
|---|
| 1237 | .size bn_power5,.-bn_power5
|
|---|
| 1238 |
|
|---|
| 1239 | .globl bn_sqr8x_internal
|
|---|
| 1240 | .hidden bn_sqr8x_internal
|
|---|
| 1241 | .type bn_sqr8x_internal,\@abi-omnipotent
|
|---|
| 1242 | .align 32
|
|---|
| 1243 | bn_sqr8x_internal:
|
|---|
| 1244 | __bn_sqr8x_internal:
|
|---|
| 1245 | .cfi_startproc
|
|---|
| 1246 | ##############################################################
|
|---|
| 1247 | # Squaring part:
|
|---|
| 1248 | #
|
|---|
| 1249 | # a) multiply-n-add everything but a[i]*a[i];
|
|---|
| 1250 | # b) shift result of a) by 1 to the left and accumulate
|
|---|
| 1251 | # a[i]*a[i] products;
|
|---|
| 1252 | #
|
|---|
| 1253 | ##############################################################
|
|---|
| 1254 | # a[1]a[0]
|
|---|
| 1255 | # a[2]a[0]
|
|---|
| 1256 | # a[3]a[0]
|
|---|
| 1257 | # a[2]a[1]
|
|---|
| 1258 | # a[4]a[0]
|
|---|
| 1259 | # a[3]a[1]
|
|---|
| 1260 | # a[5]a[0]
|
|---|
| 1261 | # a[4]a[1]
|
|---|
| 1262 | # a[3]a[2]
|
|---|
| 1263 | # a[6]a[0]
|
|---|
| 1264 | # a[5]a[1]
|
|---|
| 1265 | # a[4]a[2]
|
|---|
| 1266 | # a[7]a[0]
|
|---|
| 1267 | # a[6]a[1]
|
|---|
| 1268 | # a[5]a[2]
|
|---|
| 1269 | # a[4]a[3]
|
|---|
| 1270 | # a[7]a[1]
|
|---|
| 1271 | # a[6]a[2]
|
|---|
| 1272 | # a[5]a[3]
|
|---|
| 1273 | # a[7]a[2]
|
|---|
| 1274 | # a[6]a[3]
|
|---|
| 1275 | # a[5]a[4]
|
|---|
| 1276 | # a[7]a[3]
|
|---|
| 1277 | # a[6]a[4]
|
|---|
| 1278 | # a[7]a[4]
|
|---|
| 1279 | # a[6]a[5]
|
|---|
| 1280 | # a[7]a[5]
|
|---|
| 1281 | # a[7]a[6]
|
|---|
| 1282 | # a[1]a[0]
|
|---|
| 1283 | # a[2]a[0]
|
|---|
| 1284 | # a[3]a[0]
|
|---|
| 1285 | # a[4]a[0]
|
|---|
| 1286 | # a[5]a[0]
|
|---|
| 1287 | # a[6]a[0]
|
|---|
| 1288 | # a[7]a[0]
|
|---|
| 1289 | # a[2]a[1]
|
|---|
| 1290 | # a[3]a[1]
|
|---|
| 1291 | # a[4]a[1]
|
|---|
| 1292 | # a[5]a[1]
|
|---|
| 1293 | # a[6]a[1]
|
|---|
| 1294 | # a[7]a[1]
|
|---|
| 1295 | # a[3]a[2]
|
|---|
| 1296 | # a[4]a[2]
|
|---|
| 1297 | # a[5]a[2]
|
|---|
| 1298 | # a[6]a[2]
|
|---|
| 1299 | # a[7]a[2]
|
|---|
| 1300 | # a[4]a[3]
|
|---|
| 1301 | # a[5]a[3]
|
|---|
| 1302 | # a[6]a[3]
|
|---|
| 1303 | # a[7]a[3]
|
|---|
| 1304 | # a[5]a[4]
|
|---|
| 1305 | # a[6]a[4]
|
|---|
| 1306 | # a[7]a[4]
|
|---|
| 1307 | # a[6]a[5]
|
|---|
| 1308 | # a[7]a[5]
|
|---|
| 1309 | # a[7]a[6]
|
|---|
| 1310 | # a[0]a[0]
|
|---|
| 1311 | # a[1]a[1]
|
|---|
| 1312 | # a[2]a[2]
|
|---|
| 1313 | # a[3]a[3]
|
|---|
| 1314 | # a[4]a[4]
|
|---|
| 1315 | # a[5]a[5]
|
|---|
| 1316 | # a[6]a[6]
|
|---|
| 1317 | # a[7]a[7]
|
|---|
| 1318 |
|
|---|
| 1319 | lea 32(%r10),$i # $i=-($num-32)
|
|---|
| 1320 | lea ($aptr,$num),$aptr # end of a[] buffer, ($aptr,$i)=&ap[2]
|
|---|
| 1321 |
|
|---|
| 1322 | mov $num,$j # $j=$num
|
|---|
| 1323 |
|
|---|
| 1324 | # comments apply to $num==8 case
|
|---|
| 1325 | mov -32($aptr,$i),$a0 # a[0]
|
|---|
| 1326 | lea 48+8(%rsp,$num,2),$tptr # end of tp[] buffer, &tp[2*$num]
|
|---|
| 1327 | mov -24($aptr,$i),%rax # a[1]
|
|---|
| 1328 | lea -32($tptr,$i),$tptr # end of tp[] window, &tp[2*$num-"$i"]
|
|---|
| 1329 | mov -16($aptr,$i),$ai # a[2]
|
|---|
| 1330 | mov %rax,$a1
|
|---|
| 1331 |
|
|---|
| 1332 | mul $a0 # a[1]*a[0]
|
|---|
| 1333 | mov %rax,$A0[0] # a[1]*a[0]
|
|---|
| 1334 | mov $ai,%rax # a[2]
|
|---|
| 1335 | mov %rdx,$A0[1]
|
|---|
| 1336 | mov $A0[0],-24($tptr,$i) # t[1]
|
|---|
| 1337 |
|
|---|
| 1338 | mul $a0 # a[2]*a[0]
|
|---|
| 1339 | add %rax,$A0[1]
|
|---|
| 1340 | mov $ai,%rax
|
|---|
| 1341 | adc \$0,%rdx
|
|---|
| 1342 | mov $A0[1],-16($tptr,$i) # t[2]
|
|---|
| 1343 | mov %rdx,$A0[0]
|
|---|
| 1344 |
|
|---|
| 1345 |
|
|---|
| 1346 | mov -8($aptr,$i),$ai # a[3]
|
|---|
| 1347 | mul $a1 # a[2]*a[1]
|
|---|
| 1348 | mov %rax,$A1[0] # a[2]*a[1]+t[3]
|
|---|
| 1349 | mov $ai,%rax
|
|---|
| 1350 | mov %rdx,$A1[1]
|
|---|
| 1351 |
|
|---|
| 1352 | lea ($i),$j
|
|---|
| 1353 | mul $a0 # a[3]*a[0]
|
|---|
| 1354 | add %rax,$A0[0] # a[3]*a[0]+a[2]*a[1]+t[3]
|
|---|
| 1355 | mov $ai,%rax
|
|---|
| 1356 | mov %rdx,$A0[1]
|
|---|
| 1357 | adc \$0,$A0[1]
|
|---|
| 1358 | add $A1[0],$A0[0]
|
|---|
| 1359 | adc \$0,$A0[1]
|
|---|
| 1360 | mov $A0[0],-8($tptr,$j) # t[3]
|
|---|
| 1361 | jmp .Lsqr4x_1st
|
|---|
| 1362 |
|
|---|
| 1363 | .align 32
|
|---|
| 1364 | .Lsqr4x_1st:
|
|---|
| 1365 | mov ($aptr,$j),$ai # a[4]
|
|---|
| 1366 | mul $a1 # a[3]*a[1]
|
|---|
| 1367 | add %rax,$A1[1] # a[3]*a[1]+t[4]
|
|---|
| 1368 | mov $ai,%rax
|
|---|
| 1369 | mov %rdx,$A1[0]
|
|---|
| 1370 | adc \$0,$A1[0]
|
|---|
| 1371 |
|
|---|
| 1372 | mul $a0 # a[4]*a[0]
|
|---|
| 1373 | add %rax,$A0[1] # a[4]*a[0]+a[3]*a[1]+t[4]
|
|---|
| 1374 | mov $ai,%rax # a[3]
|
|---|
| 1375 | mov 8($aptr,$j),$ai # a[5]
|
|---|
| 1376 | mov %rdx,$A0[0]
|
|---|
| 1377 | adc \$0,$A0[0]
|
|---|
| 1378 | add $A1[1],$A0[1]
|
|---|
| 1379 | adc \$0,$A0[0]
|
|---|
| 1380 |
|
|---|
| 1381 |
|
|---|
| 1382 | mul $a1 # a[4]*a[3]
|
|---|
| 1383 | add %rax,$A1[0] # a[4]*a[3]+t[5]
|
|---|
| 1384 | mov $ai,%rax
|
|---|
| 1385 | mov $A0[1],($tptr,$j) # t[4]
|
|---|
| 1386 | mov %rdx,$A1[1]
|
|---|
| 1387 | adc \$0,$A1[1]
|
|---|
| 1388 |
|
|---|
| 1389 | mul $a0 # a[5]*a[2]
|
|---|
| 1390 | add %rax,$A0[0] # a[5]*a[2]+a[4]*a[3]+t[5]
|
|---|
| 1391 | mov $ai,%rax
|
|---|
| 1392 | mov 16($aptr,$j),$ai # a[6]
|
|---|
| 1393 | mov %rdx,$A0[1]
|
|---|
| 1394 | adc \$0,$A0[1]
|
|---|
| 1395 | add $A1[0],$A0[0]
|
|---|
| 1396 | adc \$0,$A0[1]
|
|---|
| 1397 |
|
|---|
| 1398 | mul $a1 # a[5]*a[3]
|
|---|
| 1399 | add %rax,$A1[1] # a[5]*a[3]+t[6]
|
|---|
| 1400 | mov $ai,%rax
|
|---|
| 1401 | mov $A0[0],8($tptr,$j) # t[5]
|
|---|
| 1402 | mov %rdx,$A1[0]
|
|---|
| 1403 | adc \$0,$A1[0]
|
|---|
| 1404 |
|
|---|
| 1405 | mul $a0 # a[6]*a[2]
|
|---|
| 1406 | add %rax,$A0[1] # a[6]*a[2]+a[5]*a[3]+t[6]
|
|---|
| 1407 | mov $ai,%rax # a[3]
|
|---|
| 1408 | mov 24($aptr,$j),$ai # a[7]
|
|---|
| 1409 | mov %rdx,$A0[0]
|
|---|
| 1410 | adc \$0,$A0[0]
|
|---|
| 1411 | add $A1[1],$A0[1]
|
|---|
| 1412 | adc \$0,$A0[0]
|
|---|
| 1413 |
|
|---|
| 1414 |
|
|---|
| 1415 | mul $a1 # a[6]*a[5]
|
|---|
| 1416 | add %rax,$A1[0] # a[6]*a[5]+t[7]
|
|---|
| 1417 | mov $ai,%rax
|
|---|
| 1418 | mov $A0[1],16($tptr,$j) # t[6]
|
|---|
| 1419 | mov %rdx,$A1[1]
|
|---|
| 1420 | adc \$0,$A1[1]
|
|---|
| 1421 | lea 32($j),$j
|
|---|
| 1422 |
|
|---|
| 1423 | mul $a0 # a[7]*a[4]
|
|---|
| 1424 | add %rax,$A0[0] # a[7]*a[4]+a[6]*a[5]+t[6]
|
|---|
| 1425 | mov $ai,%rax
|
|---|
| 1426 | mov %rdx,$A0[1]
|
|---|
| 1427 | adc \$0,$A0[1]
|
|---|
| 1428 | add $A1[0],$A0[0]
|
|---|
| 1429 | adc \$0,$A0[1]
|
|---|
| 1430 | mov $A0[0],-8($tptr,$j) # t[7]
|
|---|
| 1431 |
|
|---|
| 1432 | cmp \$0,$j
|
|---|
| 1433 | jne .Lsqr4x_1st
|
|---|
| 1434 |
|
|---|
| 1435 | mul $a1 # a[7]*a[5]
|
|---|
| 1436 | add %rax,$A1[1]
|
|---|
| 1437 | lea 16($i),$i
|
|---|
| 1438 | adc \$0,%rdx
|
|---|
| 1439 | add $A0[1],$A1[1]
|
|---|
| 1440 | adc \$0,%rdx
|
|---|
| 1441 |
|
|---|
| 1442 | mov $A1[1],($tptr) # t[8]
|
|---|
| 1443 | mov %rdx,$A1[0]
|
|---|
| 1444 | mov %rdx,8($tptr) # t[9]
|
|---|
| 1445 | jmp .Lsqr4x_outer
|
|---|
| 1446 |
|
|---|
| 1447 | .align 32
|
|---|
| 1448 | .Lsqr4x_outer: # comments apply to $num==6 case
|
|---|
| 1449 | mov -32($aptr,$i),$a0 # a[0]
|
|---|
| 1450 | lea 48+8(%rsp,$num,2),$tptr # end of tp[] buffer, &tp[2*$num]
|
|---|
| 1451 | mov -24($aptr,$i),%rax # a[1]
|
|---|
| 1452 | lea -32($tptr,$i),$tptr # end of tp[] window, &tp[2*$num-"$i"]
|
|---|
| 1453 | mov -16($aptr,$i),$ai # a[2]
|
|---|
| 1454 | mov %rax,$a1
|
|---|
| 1455 |
|
|---|
| 1456 | mul $a0 # a[1]*a[0]
|
|---|
| 1457 | mov -24($tptr,$i),$A0[0] # t[1]
|
|---|
| 1458 | add %rax,$A0[0] # a[1]*a[0]+t[1]
|
|---|
| 1459 | mov $ai,%rax # a[2]
|
|---|
| 1460 | adc \$0,%rdx
|
|---|
| 1461 | mov $A0[0],-24($tptr,$i) # t[1]
|
|---|
| 1462 | mov %rdx,$A0[1]
|
|---|
| 1463 |
|
|---|
| 1464 | mul $a0 # a[2]*a[0]
|
|---|
| 1465 | add %rax,$A0[1]
|
|---|
| 1466 | mov $ai,%rax
|
|---|
| 1467 | adc \$0,%rdx
|
|---|
| 1468 | add -16($tptr,$i),$A0[1] # a[2]*a[0]+t[2]
|
|---|
| 1469 | mov %rdx,$A0[0]
|
|---|
| 1470 | adc \$0,$A0[0]
|
|---|
| 1471 | mov $A0[1],-16($tptr,$i) # t[2]
|
|---|
| 1472 |
|
|---|
| 1473 | xor $A1[0],$A1[0]
|
|---|
| 1474 |
|
|---|
| 1475 | mov -8($aptr,$i),$ai # a[3]
|
|---|
| 1476 | mul $a1 # a[2]*a[1]
|
|---|
| 1477 | add %rax,$A1[0] # a[2]*a[1]+t[3]
|
|---|
| 1478 | mov $ai,%rax
|
|---|
| 1479 | adc \$0,%rdx
|
|---|
| 1480 | add -8($tptr,$i),$A1[0]
|
|---|
| 1481 | mov %rdx,$A1[1]
|
|---|
| 1482 | adc \$0,$A1[1]
|
|---|
| 1483 |
|
|---|
| 1484 | mul $a0 # a[3]*a[0]
|
|---|
| 1485 | add %rax,$A0[0] # a[3]*a[0]+a[2]*a[1]+t[3]
|
|---|
| 1486 | mov $ai,%rax
|
|---|
| 1487 | adc \$0,%rdx
|
|---|
| 1488 | add $A1[0],$A0[0]
|
|---|
| 1489 | mov %rdx,$A0[1]
|
|---|
| 1490 | adc \$0,$A0[1]
|
|---|
| 1491 | mov $A0[0],-8($tptr,$i) # t[3]
|
|---|
| 1492 |
|
|---|
| 1493 | lea ($i),$j
|
|---|
| 1494 | jmp .Lsqr4x_inner
|
|---|
| 1495 |
|
|---|
| 1496 | .align 32
|
|---|
| 1497 | .Lsqr4x_inner:
|
|---|
| 1498 | mov ($aptr,$j),$ai # a[4]
|
|---|
| 1499 | mul $a1 # a[3]*a[1]
|
|---|
| 1500 | add %rax,$A1[1] # a[3]*a[1]+t[4]
|
|---|
| 1501 | mov $ai,%rax
|
|---|
| 1502 | mov %rdx,$A1[0]
|
|---|
| 1503 | adc \$0,$A1[0]
|
|---|
| 1504 | add ($tptr,$j),$A1[1]
|
|---|
| 1505 | adc \$0,$A1[0]
|
|---|
| 1506 |
|
|---|
| 1507 | .byte 0x67
|
|---|
| 1508 | mul $a0 # a[4]*a[0]
|
|---|
| 1509 | add %rax,$A0[1] # a[4]*a[0]+a[3]*a[1]+t[4]
|
|---|
| 1510 | mov $ai,%rax # a[3]
|
|---|
| 1511 | mov 8($aptr,$j),$ai # a[5]
|
|---|
| 1512 | mov %rdx,$A0[0]
|
|---|
| 1513 | adc \$0,$A0[0]
|
|---|
| 1514 | add $A1[1],$A0[1]
|
|---|
| 1515 | adc \$0,$A0[0]
|
|---|
| 1516 |
|
|---|
| 1517 | mul $a1 # a[4]*a[3]
|
|---|
| 1518 | add %rax,$A1[0] # a[4]*a[3]+t[5]
|
|---|
| 1519 | mov $A0[1],($tptr,$j) # t[4]
|
|---|
| 1520 | mov $ai,%rax
|
|---|
| 1521 | mov %rdx,$A1[1]
|
|---|
| 1522 | adc \$0,$A1[1]
|
|---|
| 1523 | add 8($tptr,$j),$A1[0]
|
|---|
| 1524 | lea 16($j),$j # j++
|
|---|
| 1525 | adc \$0,$A1[1]
|
|---|
| 1526 |
|
|---|
| 1527 | mul $a0 # a[5]*a[2]
|
|---|
| 1528 | add %rax,$A0[0] # a[5]*a[2]+a[4]*a[3]+t[5]
|
|---|
| 1529 | mov $ai,%rax
|
|---|
| 1530 | adc \$0,%rdx
|
|---|
| 1531 | add $A1[0],$A0[0]
|
|---|
| 1532 | mov %rdx,$A0[1]
|
|---|
| 1533 | adc \$0,$A0[1]
|
|---|
| 1534 | mov $A0[0],-8($tptr,$j) # t[5], "preloaded t[1]" below
|
|---|
| 1535 |
|
|---|
| 1536 | cmp \$0,$j
|
|---|
| 1537 | jne .Lsqr4x_inner
|
|---|
| 1538 |
|
|---|
| 1539 | .byte 0x67
|
|---|
| 1540 | mul $a1 # a[5]*a[3]
|
|---|
| 1541 | add %rax,$A1[1]
|
|---|
| 1542 | adc \$0,%rdx
|
|---|
| 1543 | add $A0[1],$A1[1]
|
|---|
| 1544 | adc \$0,%rdx
|
|---|
| 1545 |
|
|---|
| 1546 | mov $A1[1],($tptr) # t[6], "preloaded t[2]" below
|
|---|
| 1547 | mov %rdx,$A1[0]
|
|---|
| 1548 | mov %rdx,8($tptr) # t[7], "preloaded t[3]" below
|
|---|
| 1549 |
|
|---|
| 1550 | add \$16,$i
|
|---|
| 1551 | jnz .Lsqr4x_outer
|
|---|
| 1552 |
|
|---|
| 1553 | # comments apply to $num==4 case
|
|---|
| 1554 | mov -32($aptr),$a0 # a[0]
|
|---|
| 1555 | lea 48+8(%rsp,$num,2),$tptr # end of tp[] buffer, &tp[2*$num]
|
|---|
| 1556 | mov -24($aptr),%rax # a[1]
|
|---|
| 1557 | lea -32($tptr,$i),$tptr # end of tp[] window, &tp[2*$num-"$i"]
|
|---|
| 1558 | mov -16($aptr),$ai # a[2]
|
|---|
| 1559 | mov %rax,$a1
|
|---|
| 1560 |
|
|---|
| 1561 | mul $a0 # a[1]*a[0]
|
|---|
| 1562 | add %rax,$A0[0] # a[1]*a[0]+t[1], preloaded t[1]
|
|---|
| 1563 | mov $ai,%rax # a[2]
|
|---|
| 1564 | mov %rdx,$A0[1]
|
|---|
| 1565 | adc \$0,$A0[1]
|
|---|
| 1566 |
|
|---|
| 1567 | mul $a0 # a[2]*a[0]
|
|---|
| 1568 | add %rax,$A0[1]
|
|---|
| 1569 | mov $ai,%rax
|
|---|
| 1570 | mov $A0[0],-24($tptr) # t[1]
|
|---|
| 1571 | mov %rdx,$A0[0]
|
|---|
| 1572 | adc \$0,$A0[0]
|
|---|
| 1573 | add $A1[1],$A0[1] # a[2]*a[0]+t[2], preloaded t[2]
|
|---|
| 1574 | mov -8($aptr),$ai # a[3]
|
|---|
| 1575 | adc \$0,$A0[0]
|
|---|
| 1576 |
|
|---|
| 1577 | mul $a1 # a[2]*a[1]
|
|---|
| 1578 | add %rax,$A1[0] # a[2]*a[1]+t[3], preloaded t[3]
|
|---|
| 1579 | mov $ai,%rax
|
|---|
| 1580 | mov $A0[1],-16($tptr) # t[2]
|
|---|
| 1581 | mov %rdx,$A1[1]
|
|---|
| 1582 | adc \$0,$A1[1]
|
|---|
| 1583 |
|
|---|
| 1584 | mul $a0 # a[3]*a[0]
|
|---|
| 1585 | add %rax,$A0[0] # a[3]*a[0]+a[2]*a[1]+t[3]
|
|---|
| 1586 | mov $ai,%rax
|
|---|
| 1587 | mov %rdx,$A0[1]
|
|---|
| 1588 | adc \$0,$A0[1]
|
|---|
| 1589 | add $A1[0],$A0[0]
|
|---|
| 1590 | adc \$0,$A0[1]
|
|---|
| 1591 | mov $A0[0],-8($tptr) # t[3]
|
|---|
| 1592 |
|
|---|
| 1593 | mul $a1 # a[3]*a[1]
|
|---|
| 1594 | add %rax,$A1[1]
|
|---|
| 1595 | mov -16($aptr),%rax # a[2]
|
|---|
| 1596 | adc \$0,%rdx
|
|---|
| 1597 | add $A0[1],$A1[1]
|
|---|
| 1598 | adc \$0,%rdx
|
|---|
| 1599 |
|
|---|
| 1600 | mov $A1[1],($tptr) # t[4]
|
|---|
| 1601 | mov %rdx,$A1[0]
|
|---|
| 1602 | mov %rdx,8($tptr) # t[5]
|
|---|
| 1603 |
|
|---|
| 1604 | mul $ai # a[2]*a[3]
|
|---|
| 1605 | ___
|
|---|
| 1606 | {
|
|---|
| 1607 | my ($shift,$carry)=($a0,$a1);
|
|---|
| 1608 | my @S=(@A1,$ai,$n0);
|
|---|
| 1609 | $code.=<<___;
|
|---|
| 1610 | add \$16,$i
|
|---|
| 1611 | xor $shift,$shift
|
|---|
| 1612 | sub $num,$i # $i=16-$num
|
|---|
| 1613 | xor $carry,$carry
|
|---|
| 1614 |
|
|---|
| 1615 | add $A1[0],%rax # t[5]
|
|---|
| 1616 | adc \$0,%rdx
|
|---|
| 1617 | mov %rax,8($tptr) # t[5]
|
|---|
| 1618 | mov %rdx,16($tptr) # t[6]
|
|---|
| 1619 | mov $carry,24($tptr) # t[7]
|
|---|
| 1620 |
|
|---|
| 1621 | mov -16($aptr,$i),%rax # a[0]
|
|---|
| 1622 | lea 48+8(%rsp),$tptr
|
|---|
| 1623 | xor $A0[0],$A0[0] # t[0]
|
|---|
| 1624 | mov 8($tptr),$A0[1] # t[1]
|
|---|
| 1625 |
|
|---|
| 1626 | lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
|
|---|
| 1627 | shr \$63,$A0[0]
|
|---|
| 1628 | lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
|
|---|
| 1629 | shr \$63,$A0[1]
|
|---|
| 1630 | or $A0[0],$S[1] # | t[2*i]>>63
|
|---|
| 1631 | mov 16($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1632 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1633 | mul %rax # a[i]*a[i]
|
|---|
| 1634 | neg $carry # mov $carry,cf
|
|---|
| 1635 | mov 24($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1636 | adc %rax,$S[0]
|
|---|
| 1637 | mov -8($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1638 | mov $S[0],($tptr)
|
|---|
| 1639 | adc %rdx,$S[1]
|
|---|
| 1640 |
|
|---|
| 1641 | lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1 | shift
|
|---|
| 1642 | mov $S[1],8($tptr)
|
|---|
| 1643 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1644 | shr \$63,$A0[0]
|
|---|
| 1645 | lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
|
|---|
| 1646 | shr \$63,$A0[1]
|
|---|
| 1647 | or $A0[0],$S[3] # | t[2*i]>>63
|
|---|
| 1648 | mov 32($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1649 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1650 | mul %rax # a[i]*a[i]
|
|---|
| 1651 | neg $carry # mov $carry,cf
|
|---|
| 1652 | mov 40($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1653 | adc %rax,$S[2]
|
|---|
| 1654 | mov 0($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1655 | mov $S[2],16($tptr)
|
|---|
| 1656 | adc %rdx,$S[3]
|
|---|
| 1657 | lea 16($i),$i
|
|---|
| 1658 | mov $S[3],24($tptr)
|
|---|
| 1659 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1660 | lea 64($tptr),$tptr
|
|---|
| 1661 | jmp .Lsqr4x_shift_n_add
|
|---|
| 1662 |
|
|---|
| 1663 | .align 32
|
|---|
| 1664 | .Lsqr4x_shift_n_add:
|
|---|
| 1665 | lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
|
|---|
| 1666 | shr \$63,$A0[0]
|
|---|
| 1667 | lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
|
|---|
| 1668 | shr \$63,$A0[1]
|
|---|
| 1669 | or $A0[0],$S[1] # | t[2*i]>>63
|
|---|
| 1670 | mov -16($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1671 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1672 | mul %rax # a[i]*a[i]
|
|---|
| 1673 | neg $carry # mov $carry,cf
|
|---|
| 1674 | mov -8($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1675 | adc %rax,$S[0]
|
|---|
| 1676 | mov -8($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1677 | mov $S[0],-32($tptr)
|
|---|
| 1678 | adc %rdx,$S[1]
|
|---|
| 1679 |
|
|---|
| 1680 | lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1 | shift
|
|---|
| 1681 | mov $S[1],-24($tptr)
|
|---|
| 1682 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1683 | shr \$63,$A0[0]
|
|---|
| 1684 | lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
|
|---|
| 1685 | shr \$63,$A0[1]
|
|---|
| 1686 | or $A0[0],$S[3] # | t[2*i]>>63
|
|---|
| 1687 | mov 0($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1688 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1689 | mul %rax # a[i]*a[i]
|
|---|
| 1690 | neg $carry # mov $carry,cf
|
|---|
| 1691 | mov 8($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1692 | adc %rax,$S[2]
|
|---|
| 1693 | mov 0($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1694 | mov $S[2],-16($tptr)
|
|---|
| 1695 | adc %rdx,$S[3]
|
|---|
| 1696 |
|
|---|
| 1697 | lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
|
|---|
| 1698 | mov $S[3],-8($tptr)
|
|---|
| 1699 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1700 | shr \$63,$A0[0]
|
|---|
| 1701 | lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
|
|---|
| 1702 | shr \$63,$A0[1]
|
|---|
| 1703 | or $A0[0],$S[1] # | t[2*i]>>63
|
|---|
| 1704 | mov 16($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1705 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1706 | mul %rax # a[i]*a[i]
|
|---|
| 1707 | neg $carry # mov $carry,cf
|
|---|
| 1708 | mov 24($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1709 | adc %rax,$S[0]
|
|---|
| 1710 | mov 8($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1711 | mov $S[0],0($tptr)
|
|---|
| 1712 | adc %rdx,$S[1]
|
|---|
| 1713 |
|
|---|
| 1714 | lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1 | shift
|
|---|
| 1715 | mov $S[1],8($tptr)
|
|---|
| 1716 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1717 | shr \$63,$A0[0]
|
|---|
| 1718 | lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
|
|---|
| 1719 | shr \$63,$A0[1]
|
|---|
| 1720 | or $A0[0],$S[3] # | t[2*i]>>63
|
|---|
| 1721 | mov 32($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1722 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1723 | mul %rax # a[i]*a[i]
|
|---|
| 1724 | neg $carry # mov $carry,cf
|
|---|
| 1725 | mov 40($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1726 | adc %rax,$S[2]
|
|---|
| 1727 | mov 16($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1728 | mov $S[2],16($tptr)
|
|---|
| 1729 | adc %rdx,$S[3]
|
|---|
| 1730 | mov $S[3],24($tptr)
|
|---|
| 1731 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1732 | lea 64($tptr),$tptr
|
|---|
| 1733 | add \$32,$i
|
|---|
| 1734 | jnz .Lsqr4x_shift_n_add
|
|---|
| 1735 |
|
|---|
| 1736 | lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
|
|---|
| 1737 | .byte 0x67
|
|---|
| 1738 | shr \$63,$A0[0]
|
|---|
| 1739 | lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
|
|---|
| 1740 | shr \$63,$A0[1]
|
|---|
| 1741 | or $A0[0],$S[1] # | t[2*i]>>63
|
|---|
| 1742 | mov -16($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1743 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1744 | mul %rax # a[i]*a[i]
|
|---|
| 1745 | neg $carry # mov $carry,cf
|
|---|
| 1746 | mov -8($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1747 | adc %rax,$S[0]
|
|---|
| 1748 | mov -8($aptr),%rax # a[i+1] # prefetch
|
|---|
| 1749 | mov $S[0],-32($tptr)
|
|---|
| 1750 | adc %rdx,$S[1]
|
|---|
| 1751 |
|
|---|
| 1752 | lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1|shift
|
|---|
| 1753 | mov $S[1],-24($tptr)
|
|---|
| 1754 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1755 | shr \$63,$A0[0]
|
|---|
| 1756 | lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
|
|---|
| 1757 | shr \$63,$A0[1]
|
|---|
| 1758 | or $A0[0],$S[3] # | t[2*i]>>63
|
|---|
| 1759 | mul %rax # a[i]*a[i]
|
|---|
| 1760 | neg $carry # mov $carry,cf
|
|---|
| 1761 | adc %rax,$S[2]
|
|---|
| 1762 | adc %rdx,$S[3]
|
|---|
| 1763 | mov $S[2],-16($tptr)
|
|---|
| 1764 | mov $S[3],-8($tptr)
|
|---|
| 1765 | ___
|
|---|
| 1766 | } |
|---|
| 1767 |
|
|---|
| 1768 | ######################################################################
|
|---|
| 1769 | # Montgomery reduction part, "word-by-word" algorithm.
|
|---|
| 1770 | #
|
|---|
| 1771 | # This new path is inspired by multiple submissions from Intel, by
|
|---|
| 1772 | # Shay Gueron, Vlad Krasnov, Erdinc Ozturk, James Guilford,
|
|---|
| 1773 | # Vinodh Gopal...
|
|---|
| 1774 | {
|
|---|
| 1775 | my ($nptr,$tptr,$carry,$m0)=("%rbp","%rdi","%rsi","%rbx");
|
|---|
| 1776 |
|
|---|
| 1777 | $code.=<<___;
|
|---|
| 1778 | movq %xmm2,$nptr
|
|---|
| 1779 | __bn_sqr8x_reduction:
|
|---|
| 1780 | xor %rax,%rax
|
|---|
| 1781 | lea ($nptr,$num),%rcx # end of n[]
|
|---|
| 1782 | lea 48+8(%rsp,$num,2),%rdx # end of t[] buffer
|
|---|
| 1783 | mov %rcx,0+8(%rsp)
|
|---|
| 1784 | lea 48+8(%rsp,$num),$tptr # end of initial t[] window
|
|---|
| 1785 | mov %rdx,8+8(%rsp)
|
|---|
| 1786 | neg $num
|
|---|
| 1787 | jmp .L8x_reduction_loop
|
|---|
| 1788 |
|
|---|
| 1789 | .align 32
|
|---|
| 1790 | .L8x_reduction_loop:
|
|---|
| 1791 | lea ($tptr,$num),$tptr # start of current t[] window
|
|---|
| 1792 | .byte 0x66
|
|---|
| 1793 | mov 8*0($tptr),$m0
|
|---|
| 1794 | mov 8*1($tptr),%r9
|
|---|
| 1795 | mov 8*2($tptr),%r10
|
|---|
| 1796 | mov 8*3($tptr),%r11
|
|---|
| 1797 | mov 8*4($tptr),%r12
|
|---|
| 1798 | mov 8*5($tptr),%r13
|
|---|
| 1799 | mov 8*6($tptr),%r14
|
|---|
| 1800 | mov 8*7($tptr),%r15
|
|---|
| 1801 | mov %rax,(%rdx) # store top-most carry bit
|
|---|
| 1802 | lea 8*8($tptr),$tptr
|
|---|
| 1803 |
|
|---|
| 1804 | .byte 0x67
|
|---|
| 1805 | mov $m0,%r8
|
|---|
| 1806 | imulq 32+8(%rsp),$m0 # n0*a[0]
|
|---|
| 1807 | mov 8*0($nptr),%rax # n[0]
|
|---|
| 1808 | mov \$8,%ecx
|
|---|
| 1809 | jmp .L8x_reduce
|
|---|
| 1810 |
|
|---|
| 1811 | .align 32
|
|---|
| 1812 | .L8x_reduce:
|
|---|
| 1813 | mulq $m0
|
|---|
| 1814 | mov 8*1($nptr),%rax # n[1]
|
|---|
| 1815 | neg %r8
|
|---|
| 1816 | mov %rdx,%r8
|
|---|
| 1817 | adc \$0,%r8
|
|---|
| 1818 |
|
|---|
| 1819 | mulq $m0
|
|---|
| 1820 | add %rax,%r9
|
|---|
| 1821 | mov 8*2($nptr),%rax
|
|---|
| 1822 | adc \$0,%rdx
|
|---|
| 1823 | add %r9,%r8
|
|---|
| 1824 | mov $m0,48-8+8(%rsp,%rcx,8) # put aside n0*a[i]
|
|---|
| 1825 | mov %rdx,%r9
|
|---|
| 1826 | adc \$0,%r9
|
|---|
| 1827 |
|
|---|
| 1828 | mulq $m0
|
|---|
| 1829 | add %rax,%r10
|
|---|
| 1830 | mov 8*3($nptr),%rax
|
|---|
| 1831 | adc \$0,%rdx
|
|---|
| 1832 | add %r10,%r9
|
|---|
| 1833 | mov 32+8(%rsp),$carry # pull n0, borrow $carry
|
|---|
| 1834 | mov %rdx,%r10
|
|---|
| 1835 | adc \$0,%r10
|
|---|
| 1836 |
|
|---|
| 1837 | mulq $m0
|
|---|
| 1838 | add %rax,%r11
|
|---|
| 1839 | mov 8*4($nptr),%rax
|
|---|
| 1840 | adc \$0,%rdx
|
|---|
| 1841 | imulq %r8,$carry # modulo-scheduled
|
|---|
| 1842 | add %r11,%r10
|
|---|
| 1843 | mov %rdx,%r11
|
|---|
| 1844 | adc \$0,%r11
|
|---|
| 1845 |
|
|---|
| 1846 | mulq $m0
|
|---|
| 1847 | add %rax,%r12
|
|---|
| 1848 | mov 8*5($nptr),%rax
|
|---|
| 1849 | adc \$0,%rdx
|
|---|
| 1850 | add %r12,%r11
|
|---|
| 1851 | mov %rdx,%r12
|
|---|
| 1852 | adc \$0,%r12
|
|---|
| 1853 |
|
|---|
| 1854 | mulq $m0
|
|---|
| 1855 | add %rax,%r13
|
|---|
| 1856 | mov 8*6($nptr),%rax
|
|---|
| 1857 | adc \$0,%rdx
|
|---|
| 1858 | add %r13,%r12
|
|---|
| 1859 | mov %rdx,%r13
|
|---|
| 1860 | adc \$0,%r13
|
|---|
| 1861 |
|
|---|
| 1862 | mulq $m0
|
|---|
| 1863 | add %rax,%r14
|
|---|
| 1864 | mov 8*7($nptr),%rax
|
|---|
| 1865 | adc \$0,%rdx
|
|---|
| 1866 | add %r14,%r13
|
|---|
| 1867 | mov %rdx,%r14
|
|---|
| 1868 | adc \$0,%r14
|
|---|
| 1869 |
|
|---|
| 1870 | mulq $m0
|
|---|
| 1871 | mov $carry,$m0 # n0*a[i]
|
|---|
| 1872 | add %rax,%r15
|
|---|
| 1873 | mov 8*0($nptr),%rax # n[0]
|
|---|
| 1874 | adc \$0,%rdx
|
|---|
| 1875 | add %r15,%r14
|
|---|
| 1876 | mov %rdx,%r15
|
|---|
| 1877 | adc \$0,%r15
|
|---|
| 1878 |
|
|---|
| 1879 | dec %ecx
|
|---|
| 1880 | jnz .L8x_reduce
|
|---|
| 1881 |
|
|---|
| 1882 | lea 8*8($nptr),$nptr
|
|---|
| 1883 | xor %rax,%rax
|
|---|
| 1884 | mov 8+8(%rsp),%rdx # pull end of t[]
|
|---|
| 1885 | cmp 0+8(%rsp),$nptr # end of n[]?
|
|---|
| 1886 | jae .L8x_no_tail
|
|---|
| 1887 |
|
|---|
| 1888 | .byte 0x66
|
|---|
| 1889 | add 8*0($tptr),%r8
|
|---|
| 1890 | adc 8*1($tptr),%r9
|
|---|
| 1891 | adc 8*2($tptr),%r10
|
|---|
| 1892 | adc 8*3($tptr),%r11
|
|---|
| 1893 | adc 8*4($tptr),%r12
|
|---|
| 1894 | adc 8*5($tptr),%r13
|
|---|
| 1895 | adc 8*6($tptr),%r14
|
|---|
| 1896 | adc 8*7($tptr),%r15
|
|---|
| 1897 | sbb $carry,$carry # top carry
|
|---|
| 1898 |
|
|---|
| 1899 | mov 48+56+8(%rsp),$m0 # pull n0*a[0]
|
|---|
| 1900 | mov \$8,%ecx
|
|---|
| 1901 | mov 8*0($nptr),%rax
|
|---|
| 1902 | jmp .L8x_tail
|
|---|
| 1903 |
|
|---|
| 1904 | .align 32
|
|---|
| 1905 | .L8x_tail:
|
|---|
| 1906 | mulq $m0
|
|---|
| 1907 | add %rax,%r8
|
|---|
| 1908 | mov 8*1($nptr),%rax
|
|---|
| 1909 | mov %r8,($tptr) # save result
|
|---|
| 1910 | mov %rdx,%r8
|
|---|
| 1911 | adc \$0,%r8
|
|---|
| 1912 |
|
|---|
| 1913 | mulq $m0
|
|---|
| 1914 | add %rax,%r9
|
|---|
| 1915 | mov 8*2($nptr),%rax
|
|---|
| 1916 | adc \$0,%rdx
|
|---|
| 1917 | add %r9,%r8
|
|---|
| 1918 | lea 8($tptr),$tptr # $tptr++
|
|---|
| 1919 | mov %rdx,%r9
|
|---|
| 1920 | adc \$0,%r9
|
|---|
| 1921 |
|
|---|
| 1922 | mulq $m0
|
|---|
| 1923 | add %rax,%r10
|
|---|
| 1924 | mov 8*3($nptr),%rax
|
|---|
| 1925 | adc \$0,%rdx
|
|---|
| 1926 | add %r10,%r9
|
|---|
| 1927 | mov %rdx,%r10
|
|---|
| 1928 | adc \$0,%r10
|
|---|
| 1929 |
|
|---|
| 1930 | mulq $m0
|
|---|
| 1931 | add %rax,%r11
|
|---|
| 1932 | mov 8*4($nptr),%rax
|
|---|
| 1933 | adc \$0,%rdx
|
|---|
| 1934 | add %r11,%r10
|
|---|
| 1935 | mov %rdx,%r11
|
|---|
| 1936 | adc \$0,%r11
|
|---|
| 1937 |
|
|---|
| 1938 | mulq $m0
|
|---|
| 1939 | add %rax,%r12
|
|---|
| 1940 | mov 8*5($nptr),%rax
|
|---|
| 1941 | adc \$0,%rdx
|
|---|
| 1942 | add %r12,%r11
|
|---|
| 1943 | mov %rdx,%r12
|
|---|
| 1944 | adc \$0,%r12
|
|---|
| 1945 |
|
|---|
| 1946 | mulq $m0
|
|---|
| 1947 | add %rax,%r13
|
|---|
| 1948 | mov 8*6($nptr),%rax
|
|---|
| 1949 | adc \$0,%rdx
|
|---|
| 1950 | add %r13,%r12
|
|---|
| 1951 | mov %rdx,%r13
|
|---|
| 1952 | adc \$0,%r13
|
|---|
| 1953 |
|
|---|
| 1954 | mulq $m0
|
|---|
| 1955 | add %rax,%r14
|
|---|
| 1956 | mov 8*7($nptr),%rax
|
|---|
| 1957 | adc \$0,%rdx
|
|---|
| 1958 | add %r14,%r13
|
|---|
| 1959 | mov %rdx,%r14
|
|---|
| 1960 | adc \$0,%r14
|
|---|
| 1961 |
|
|---|
| 1962 | mulq $m0
|
|---|
| 1963 | mov 48-16+8(%rsp,%rcx,8),$m0# pull n0*a[i]
|
|---|
| 1964 | add %rax,%r15
|
|---|
| 1965 | adc \$0,%rdx
|
|---|
| 1966 | add %r15,%r14
|
|---|
| 1967 | mov 8*0($nptr),%rax # pull n[0]
|
|---|
| 1968 | mov %rdx,%r15
|
|---|
| 1969 | adc \$0,%r15
|
|---|
| 1970 |
|
|---|
| 1971 | dec %ecx
|
|---|
| 1972 | jnz .L8x_tail
|
|---|
| 1973 |
|
|---|
| 1974 | lea 8*8($nptr),$nptr
|
|---|
| 1975 | mov 8+8(%rsp),%rdx # pull end of t[]
|
|---|
| 1976 | cmp 0+8(%rsp),$nptr # end of n[]?
|
|---|
| 1977 | jae .L8x_tail_done # break out of loop
|
|---|
| 1978 |
|
|---|
| 1979 | mov 48+56+8(%rsp),$m0 # pull n0*a[0]
|
|---|
| 1980 | neg $carry
|
|---|
| 1981 | mov 8*0($nptr),%rax # pull n[0]
|
|---|
| 1982 | adc 8*0($tptr),%r8
|
|---|
| 1983 | adc 8*1($tptr),%r9
|
|---|
| 1984 | adc 8*2($tptr),%r10
|
|---|
| 1985 | adc 8*3($tptr),%r11
|
|---|
| 1986 | adc 8*4($tptr),%r12
|
|---|
| 1987 | adc 8*5($tptr),%r13
|
|---|
| 1988 | adc 8*6($tptr),%r14
|
|---|
| 1989 | adc 8*7($tptr),%r15
|
|---|
| 1990 | sbb $carry,$carry # top carry
|
|---|
| 1991 |
|
|---|
| 1992 | mov \$8,%ecx
|
|---|
| 1993 | jmp .L8x_tail
|
|---|
| 1994 |
|
|---|
| 1995 | .align 32
|
|---|
| 1996 | .L8x_tail_done:
|
|---|
| 1997 | xor %rax,%rax
|
|---|
| 1998 | add (%rdx),%r8 # can this overflow?
|
|---|
| 1999 | adc \$0,%r9
|
|---|
| 2000 | adc \$0,%r10
|
|---|
| 2001 | adc \$0,%r11
|
|---|
| 2002 | adc \$0,%r12
|
|---|
| 2003 | adc \$0,%r13
|
|---|
| 2004 | adc \$0,%r14
|
|---|
| 2005 | adc \$0,%r15
|
|---|
| 2006 | adc \$0,%rax
|
|---|
| 2007 |
|
|---|
| 2008 | neg $carry
|
|---|
| 2009 | .L8x_no_tail:
|
|---|
| 2010 | adc 8*0($tptr),%r8
|
|---|
| 2011 | adc 8*1($tptr),%r9
|
|---|
| 2012 | adc 8*2($tptr),%r10
|
|---|
| 2013 | adc 8*3($tptr),%r11
|
|---|
| 2014 | adc 8*4($tptr),%r12
|
|---|
| 2015 | adc 8*5($tptr),%r13
|
|---|
| 2016 | adc 8*6($tptr),%r14
|
|---|
| 2017 | adc 8*7($tptr),%r15
|
|---|
| 2018 | adc \$0,%rax # top-most carry
|
|---|
| 2019 | mov -8($nptr),%rcx # np[num-1]
|
|---|
| 2020 | xor $carry,$carry
|
|---|
| 2021 |
|
|---|
| 2022 | movq %xmm2,$nptr # restore $nptr
|
|---|
| 2023 |
|
|---|
| 2024 | mov %r8,8*0($tptr) # store top 512 bits
|
|---|
| 2025 | mov %r9,8*1($tptr)
|
|---|
| 2026 | movq %xmm3,$num # $num is %r9, can't be moved upwards
|
|---|
| 2027 | mov %r10,8*2($tptr)
|
|---|
| 2028 | mov %r11,8*3($tptr)
|
|---|
| 2029 | mov %r12,8*4($tptr)
|
|---|
| 2030 | mov %r13,8*5($tptr)
|
|---|
| 2031 | mov %r14,8*6($tptr)
|
|---|
| 2032 | mov %r15,8*7($tptr)
|
|---|
| 2033 | lea 8*8($tptr),$tptr
|
|---|
| 2034 |
|
|---|
| 2035 | cmp %rdx,$tptr # end of t[]?
|
|---|
| 2036 | jb .L8x_reduction_loop
|
|---|
| 2037 | ret
|
|---|
| 2038 | .cfi_endproc
|
|---|
| 2039 | .size bn_sqr8x_internal,.-bn_sqr8x_internal
|
|---|
| 2040 | ___
|
|---|
| 2041 | } |
|---|
| 2042 |
|
|---|
| 2043 | ##############################################################
|
|---|
| 2044 | # Post-condition, 4x unrolled
|
|---|
| 2045 | #
|
|---|
| 2046 | {
|
|---|
| 2047 | my ($tptr,$nptr)=("%rbx","%rbp");
|
|---|
| 2048 | $code.=<<___;
|
|---|
| 2049 | .type __bn_post4x_internal,\@abi-omnipotent
|
|---|
| 2050 | .align 32
|
|---|
| 2051 | __bn_post4x_internal:
|
|---|
| 2052 | .cfi_startproc
|
|---|
| 2053 | mov 8*0($nptr),%r12
|
|---|
| 2054 | lea (%rdi,$num),$tptr # %rdi was $tptr above
|
|---|
| 2055 | mov $num,%rcx
|
|---|
| 2056 | movq %xmm1,$rptr # restore $rptr
|
|---|
| 2057 | neg %rax
|
|---|
| 2058 | movq %xmm1,$aptr # prepare for back-to-back call
|
|---|
| 2059 | sar \$3+2,%rcx
|
|---|
| 2060 | dec %r12 # so that after 'not' we get -n[0]
|
|---|
| 2061 | xor %r10,%r10
|
|---|
| 2062 | mov 8*1($nptr),%r13
|
|---|
| 2063 | mov 8*2($nptr),%r14
|
|---|
| 2064 | mov 8*3($nptr),%r15
|
|---|
| 2065 | jmp .Lsqr4x_sub_entry
|
|---|
| 2066 |
|
|---|
| 2067 | .align 16
|
|---|
| 2068 | .Lsqr4x_sub:
|
|---|
| 2069 | mov 8*0($nptr),%r12
|
|---|
| 2070 | mov 8*1($nptr),%r13
|
|---|
| 2071 | mov 8*2($nptr),%r14
|
|---|
| 2072 | mov 8*3($nptr),%r15
|
|---|
| 2073 | .Lsqr4x_sub_entry:
|
|---|
| 2074 | lea 8*4($nptr),$nptr
|
|---|
| 2075 | not %r12
|
|---|
| 2076 | not %r13
|
|---|
| 2077 | not %r14
|
|---|
| 2078 | not %r15
|
|---|
| 2079 | and %rax,%r12
|
|---|
| 2080 | and %rax,%r13
|
|---|
| 2081 | and %rax,%r14
|
|---|
| 2082 | and %rax,%r15
|
|---|
| 2083 |
|
|---|
| 2084 | neg %r10 # mov %r10,%cf
|
|---|
| 2085 | adc 8*0($tptr),%r12
|
|---|
| 2086 | adc 8*1($tptr),%r13
|
|---|
| 2087 | adc 8*2($tptr),%r14
|
|---|
| 2088 | adc 8*3($tptr),%r15
|
|---|
| 2089 | mov %r12,8*0($rptr)
|
|---|
| 2090 | lea 8*4($tptr),$tptr
|
|---|
| 2091 | mov %r13,8*1($rptr)
|
|---|
| 2092 | sbb %r10,%r10 # mov %cf,%r10
|
|---|
| 2093 | mov %r14,8*2($rptr)
|
|---|
| 2094 | mov %r15,8*3($rptr)
|
|---|
| 2095 | lea 8*4($rptr),$rptr
|
|---|
| 2096 |
|
|---|
| 2097 | inc %rcx # pass %cf
|
|---|
| 2098 | jnz .Lsqr4x_sub
|
|---|
| 2099 |
|
|---|
| 2100 | mov $num,%r10 # prepare for back-to-back call
|
|---|
| 2101 | neg $num # restore $num
|
|---|
| 2102 | ret
|
|---|
| 2103 | .cfi_endproc
|
|---|
| 2104 | .size __bn_post4x_internal,.-__bn_post4x_internal
|
|---|
| 2105 | ___
|
|---|
| 2106 | }
|
|---|
| 2107 | {
|
|---|
| 2108 | $code.=<<___;
|
|---|
| 2109 | .globl bn_from_montgomery
|
|---|
| 2110 | .type bn_from_montgomery,\@abi-omnipotent
|
|---|
| 2111 | .align 32
|
|---|
| 2112 | bn_from_montgomery:
|
|---|
| 2113 | .cfi_startproc
|
|---|
| 2114 | testl \$7,`($win64?"48(%rsp)":"%r9d")`
|
|---|
| 2115 | jz bn_from_mont8x
|
|---|
| 2116 | xor %eax,%eax
|
|---|
| 2117 | ret
|
|---|
| 2118 | .cfi_endproc
|
|---|
| 2119 | .size bn_from_montgomery,.-bn_from_montgomery
|
|---|
| 2120 |
|
|---|
| 2121 | .type bn_from_mont8x,\@function,6
|
|---|
| 2122 | .align 32
|
|---|
| 2123 | bn_from_mont8x:
|
|---|
| 2124 | .cfi_startproc
|
|---|
| 2125 | .byte 0x67
|
|---|
| 2126 | mov %rsp,%rax
|
|---|
| 2127 | .cfi_def_cfa_register %rax
|
|---|
| 2128 | push %rbx
|
|---|
| 2129 | .cfi_push %rbx
|
|---|
| 2130 | push %rbp
|
|---|
| 2131 | .cfi_push %rbp
|
|---|
| 2132 | push %r12
|
|---|
| 2133 | .cfi_push %r12
|
|---|
| 2134 | push %r13
|
|---|
| 2135 | .cfi_push %r13
|
|---|
| 2136 | push %r14
|
|---|
| 2137 | .cfi_push %r14
|
|---|
| 2138 | push %r15
|
|---|
| 2139 | .cfi_push %r15
|
|---|
| 2140 | .Lfrom_prologue:
|
|---|
| 2141 |
|
|---|
| 2142 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 2143 | lea ($num,$num,2),%r10 # 3*$num in bytes
|
|---|
| 2144 | neg $num
|
|---|
| 2145 | mov ($n0),$n0 # *n0
|
|---|
| 2146 |
|
|---|
| 2147 | ##############################################################
|
|---|
| 2148 | # Ensure that stack frame doesn't alias with $rptr+3*$num
|
|---|
| 2149 | # modulo 4096, which covers ret[num], am[num] and n[num]
|
|---|
| 2150 | # (see bn_exp.c). The stack is allocated to aligned with
|
|---|
| 2151 | # bn_power5's frame, and as bn_from_montgomery happens to be
|
|---|
| 2152 | # last operation, we use the opportunity to cleanse it.
|
|---|
| 2153 | #
|
|---|
| 2154 | lea -320(%rsp,$num,2),%r11
|
|---|
| 2155 | mov %rsp,%rbp
|
|---|
| 2156 | sub $rptr,%r11
|
|---|
| 2157 | and \$4095,%r11
|
|---|
| 2158 | cmp %r11,%r10
|
|---|
| 2159 | jb .Lfrom_sp_alt
|
|---|
| 2160 | sub %r11,%rbp # align with $aptr
|
|---|
| 2161 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*$num*8+256)
|
|---|
| 2162 | jmp .Lfrom_sp_done
|
|---|
| 2163 |
|
|---|
| 2164 | .align 32
|
|---|
| 2165 | .Lfrom_sp_alt:
|
|---|
| 2166 | lea 4096-320(,$num,2),%r10
|
|---|
| 2167 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*$num*8+256)
|
|---|
| 2168 | sub %r10,%r11
|
|---|
| 2169 | mov \$0,%r10
|
|---|
| 2170 | cmovc %r10,%r11
|
|---|
| 2171 | sub %r11,%rbp
|
|---|
| 2172 | .Lfrom_sp_done:
|
|---|
| 2173 | and \$-64,%rbp
|
|---|
| 2174 | mov %rsp,%r11
|
|---|
| 2175 | sub %rbp,%r11
|
|---|
| 2176 | and \$-4096,%r11
|
|---|
| 2177 | lea (%rbp,%r11),%rsp
|
|---|
| 2178 | mov (%rsp),%r10
|
|---|
| 2179 | cmp %rbp,%rsp
|
|---|
| 2180 | ja .Lfrom_page_walk
|
|---|
| 2181 | jmp .Lfrom_page_walk_done
|
|---|
| 2182 |
|
|---|
| 2183 | .Lfrom_page_walk:
|
|---|
| 2184 | lea -4096(%rsp),%rsp
|
|---|
| 2185 | mov (%rsp),%r10
|
|---|
| 2186 | cmp %rbp,%rsp
|
|---|
| 2187 | ja .Lfrom_page_walk
|
|---|
| 2188 | .Lfrom_page_walk_done:
|
|---|
| 2189 |
|
|---|
| 2190 | mov $num,%r10
|
|---|
| 2191 | neg $num
|
|---|
| 2192 |
|
|---|
| 2193 | ##############################################################
|
|---|
| 2194 | # Stack layout
|
|---|
| 2195 | #
|
|---|
| 2196 | # +0 saved $num, used in reduction section
|
|---|
| 2197 | # +8 &t[2*$num], used in reduction section
|
|---|
| 2198 | # +32 saved *n0
|
|---|
| 2199 | # +40 saved %rsp
|
|---|
| 2200 | # +48 t[2*$num]
|
|---|
| 2201 | #
|
|---|
| 2202 | mov $n0, 32(%rsp)
|
|---|
| 2203 | mov %rax, 40(%rsp) # save original %rsp
|
|---|
| 2204 | .cfi_cfa_expression %rsp+40,deref,+8
|
|---|
| 2205 | .Lfrom_body:
|
|---|
| 2206 | mov $num,%r11
|
|---|
| 2207 | lea 48(%rsp),%rax
|
|---|
| 2208 | pxor %xmm0,%xmm0
|
|---|
| 2209 | jmp .Lmul_by_1
|
|---|
| 2210 |
|
|---|
| 2211 | .align 32
|
|---|
| 2212 | .Lmul_by_1:
|
|---|
| 2213 | movdqu ($aptr),%xmm1
|
|---|
| 2214 | movdqu 16($aptr),%xmm2
|
|---|
| 2215 | movdqu 32($aptr),%xmm3
|
|---|
| 2216 | movdqa %xmm0,(%rax,$num)
|
|---|
| 2217 | movdqu 48($aptr),%xmm4
|
|---|
| 2218 | movdqa %xmm0,16(%rax,$num)
|
|---|
| 2219 | .byte 0x48,0x8d,0xb6,0x40,0x00,0x00,0x00 # lea 64($aptr),$aptr
|
|---|
| 2220 | movdqa %xmm1,(%rax)
|
|---|
| 2221 | movdqa %xmm0,32(%rax,$num)
|
|---|
| 2222 | movdqa %xmm2,16(%rax)
|
|---|
| 2223 | movdqa %xmm0,48(%rax,$num)
|
|---|
| 2224 | movdqa %xmm3,32(%rax)
|
|---|
| 2225 | movdqa %xmm4,48(%rax)
|
|---|
| 2226 | lea 64(%rax),%rax
|
|---|
| 2227 | sub \$64,%r11
|
|---|
| 2228 | jnz .Lmul_by_1
|
|---|
| 2229 |
|
|---|
| 2230 | movq $rptr,%xmm1
|
|---|
| 2231 | movq $nptr,%xmm2
|
|---|
| 2232 | .byte 0x67
|
|---|
| 2233 | mov $nptr,%rbp
|
|---|
| 2234 | movq %r10, %xmm3 # -num
|
|---|
| 2235 | ___
|
|---|
| 2236 | $code.=<<___ if ($addx);
|
|---|
| 2237 | mov OPENSSL_ia32cap_P+8(%rip),%r11d
|
|---|
| 2238 | and \$0x80108,%r11d
|
|---|
| 2239 | cmp \$0x80108,%r11d # check for AD*X+BMI2+BMI1
|
|---|
| 2240 | jne .Lfrom_mont_nox
|
|---|
| 2241 |
|
|---|
| 2242 | lea (%rax,$num),$rptr
|
|---|
| 2243 | call __bn_sqrx8x_reduction
|
|---|
| 2244 | call __bn_postx4x_internal
|
|---|
| 2245 |
|
|---|
| 2246 | pxor %xmm0,%xmm0
|
|---|
| 2247 | lea 48(%rsp),%rax
|
|---|
| 2248 | jmp .Lfrom_mont_zero
|
|---|
| 2249 |
|
|---|
| 2250 | .align 32
|
|---|
| 2251 | .Lfrom_mont_nox:
|
|---|
| 2252 | ___
|
|---|
| 2253 | $code.=<<___;
|
|---|
| 2254 | call __bn_sqr8x_reduction
|
|---|
| 2255 | call __bn_post4x_internal
|
|---|
| 2256 |
|
|---|
| 2257 | pxor %xmm0,%xmm0
|
|---|
| 2258 | lea 48(%rsp),%rax
|
|---|
| 2259 | jmp .Lfrom_mont_zero
|
|---|
| 2260 |
|
|---|
| 2261 | .align 32
|
|---|
| 2262 | .Lfrom_mont_zero:
|
|---|
| 2263 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 2264 | .cfi_def_cfa %rsi,8
|
|---|
| 2265 | movdqa %xmm0,16*0(%rax)
|
|---|
| 2266 | movdqa %xmm0,16*1(%rax)
|
|---|
| 2267 | movdqa %xmm0,16*2(%rax)
|
|---|
| 2268 | movdqa %xmm0,16*3(%rax)
|
|---|
| 2269 | lea 16*4(%rax),%rax
|
|---|
| 2270 | sub \$32,$num
|
|---|
| 2271 | jnz .Lfrom_mont_zero
|
|---|
| 2272 |
|
|---|
| 2273 | mov \$1,%rax
|
|---|
| 2274 | mov -48(%rsi),%r15
|
|---|
| 2275 | .cfi_restore %r15
|
|---|
| 2276 | mov -40(%rsi),%r14
|
|---|
| 2277 | .cfi_restore %r14
|
|---|
| 2278 | mov -32(%rsi),%r13
|
|---|
| 2279 | .cfi_restore %r13
|
|---|
| 2280 | mov -24(%rsi),%r12
|
|---|
| 2281 | .cfi_restore %r12
|
|---|
| 2282 | mov -16(%rsi),%rbp
|
|---|
| 2283 | .cfi_restore %rbp
|
|---|
| 2284 | mov -8(%rsi),%rbx
|
|---|
| 2285 | .cfi_restore %rbx
|
|---|
| 2286 | lea (%rsi),%rsp
|
|---|
| 2287 | .cfi_def_cfa_register %rsp
|
|---|
| 2288 | .Lfrom_epilogue:
|
|---|
| 2289 | ret
|
|---|
| 2290 | .cfi_endproc
|
|---|
| 2291 | .size bn_from_mont8x,.-bn_from_mont8x
|
|---|
| 2292 | ___
|
|---|
| 2293 | }
|
|---|
| 2294 | }}}
|
|---|
| 2295 | |
|---|
| 2296 |
|
|---|
| 2297 | if ($addx) {{{
|
|---|
| 2298 | my $bp="%rdx"; # restore original value
|
|---|
| 2299 |
|
|---|
| 2300 | $code.=<<___;
|
|---|
| 2301 | .type bn_mulx4x_mont_gather5,\@function,6
|
|---|
| 2302 | .align 32
|
|---|
| 2303 | bn_mulx4x_mont_gather5:
|
|---|
| 2304 | .cfi_startproc
|
|---|
| 2305 | mov %rsp,%rax
|
|---|
| 2306 | .cfi_def_cfa_register %rax
|
|---|
| 2307 | .Lmulx4x_enter:
|
|---|
| 2308 | push %rbx
|
|---|
| 2309 | .cfi_push %rbx
|
|---|
| 2310 | push %rbp
|
|---|
| 2311 | .cfi_push %rbp
|
|---|
| 2312 | push %r12
|
|---|
| 2313 | .cfi_push %r12
|
|---|
| 2314 | push %r13
|
|---|
| 2315 | .cfi_push %r13
|
|---|
| 2316 | push %r14
|
|---|
| 2317 | .cfi_push %r14
|
|---|
| 2318 | push %r15
|
|---|
| 2319 | .cfi_push %r15
|
|---|
| 2320 | .Lmulx4x_prologue:
|
|---|
| 2321 |
|
|---|
| 2322 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 2323 | lea ($num,$num,2),%r10 # 3*$num in bytes
|
|---|
| 2324 | neg $num # -$num
|
|---|
| 2325 | mov ($n0),$n0 # *n0
|
|---|
| 2326 |
|
|---|
| 2327 | ##############################################################
|
|---|
| 2328 | # Ensure that stack frame doesn't alias with $rptr+3*$num
|
|---|
| 2329 | # modulo 4096, which covers ret[num], am[num] and n[num]
|
|---|
| 2330 | # (see bn_exp.c). This is done to allow memory disambiguation
|
|---|
| 2331 | # logic do its magic. [Extra [num] is allocated in order
|
|---|
| 2332 | # to align with bn_power5's frame, which is cleansed after
|
|---|
| 2333 | # completing exponentiation. Extra 256 bytes is for power mask
|
|---|
| 2334 | # calculated from 7th argument, the index.]
|
|---|
| 2335 | #
|
|---|
| 2336 | lea -320(%rsp,$num,2),%r11
|
|---|
| 2337 | mov %rsp,%rbp
|
|---|
| 2338 | sub $rp,%r11
|
|---|
| 2339 | and \$4095,%r11
|
|---|
| 2340 | cmp %r11,%r10
|
|---|
| 2341 | jb .Lmulx4xsp_alt
|
|---|
| 2342 | sub %r11,%rbp # align with $aptr
|
|---|
| 2343 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*$num*8+256)
|
|---|
| 2344 | jmp .Lmulx4xsp_done
|
|---|
| 2345 |
|
|---|
| 2346 | .Lmulx4xsp_alt:
|
|---|
| 2347 | lea 4096-320(,$num,2),%r10
|
|---|
| 2348 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*$num*8+256)
|
|---|
| 2349 | sub %r10,%r11
|
|---|
| 2350 | mov \$0,%r10
|
|---|
| 2351 | cmovc %r10,%r11
|
|---|
| 2352 | sub %r11,%rbp
|
|---|
| 2353 | .Lmulx4xsp_done:
|
|---|
| 2354 | and \$-64,%rbp # ensure alignment
|
|---|
| 2355 | mov %rsp,%r11
|
|---|
| 2356 | sub %rbp,%r11
|
|---|
| 2357 | and \$-4096,%r11
|
|---|
| 2358 | lea (%rbp,%r11),%rsp
|
|---|
| 2359 | mov (%rsp),%r10
|
|---|
| 2360 | cmp %rbp,%rsp
|
|---|
| 2361 | ja .Lmulx4x_page_walk
|
|---|
| 2362 | jmp .Lmulx4x_page_walk_done
|
|---|
| 2363 |
|
|---|
| 2364 | .Lmulx4x_page_walk:
|
|---|
| 2365 | lea -4096(%rsp),%rsp
|
|---|
| 2366 | mov (%rsp),%r10
|
|---|
| 2367 | cmp %rbp,%rsp
|
|---|
| 2368 | ja .Lmulx4x_page_walk
|
|---|
| 2369 | .Lmulx4x_page_walk_done:
|
|---|
| 2370 |
|
|---|
| 2371 | ##############################################################
|
|---|
| 2372 | # Stack layout
|
|---|
| 2373 | # +0 -num
|
|---|
| 2374 | # +8 off-loaded &b[i]
|
|---|
| 2375 | # +16 end of b[num]
|
|---|
| 2376 | # +24 inner counter
|
|---|
| 2377 | # +32 saved n0
|
|---|
| 2378 | # +40 saved %rsp
|
|---|
| 2379 | # +48
|
|---|
| 2380 | # +56 saved rp
|
|---|
| 2381 | # +64 tmp[num+1]
|
|---|
| 2382 | #
|
|---|
| 2383 | mov $n0, 32(%rsp) # save *n0
|
|---|
| 2384 | mov %rax,40(%rsp) # save original %rsp
|
|---|
| 2385 | .cfi_cfa_expression %rsp+40,deref,+8
|
|---|
| 2386 | .Lmulx4x_body:
|
|---|
| 2387 | call mulx4x_internal
|
|---|
| 2388 |
|
|---|
| 2389 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 2390 | .cfi_def_cfa %rsi,8
|
|---|
| 2391 | mov \$1,%rax
|
|---|
| 2392 |
|
|---|
| 2393 | mov -48(%rsi),%r15
|
|---|
| 2394 | .cfi_restore %r15
|
|---|
| 2395 | mov -40(%rsi),%r14
|
|---|
| 2396 | .cfi_restore %r14
|
|---|
| 2397 | mov -32(%rsi),%r13
|
|---|
| 2398 | .cfi_restore %r13
|
|---|
| 2399 | mov -24(%rsi),%r12
|
|---|
| 2400 | .cfi_restore %r12
|
|---|
| 2401 | mov -16(%rsi),%rbp
|
|---|
| 2402 | .cfi_restore %rbp
|
|---|
| 2403 | mov -8(%rsi),%rbx
|
|---|
| 2404 | .cfi_restore %rbx
|
|---|
| 2405 | lea (%rsi),%rsp
|
|---|
| 2406 | .cfi_def_cfa_register %rsp
|
|---|
| 2407 | .Lmulx4x_epilogue:
|
|---|
| 2408 | ret
|
|---|
| 2409 | .cfi_endproc
|
|---|
| 2410 | .size bn_mulx4x_mont_gather5,.-bn_mulx4x_mont_gather5
|
|---|
| 2411 |
|
|---|
| 2412 | .type mulx4x_internal,\@abi-omnipotent
|
|---|
| 2413 | .align 32
|
|---|
| 2414 | mulx4x_internal:
|
|---|
| 2415 | .cfi_startproc
|
|---|
| 2416 | mov $num,8(%rsp) # save -$num (it was in bytes)
|
|---|
| 2417 | mov $num,%r10
|
|---|
| 2418 | neg $num # restore $num
|
|---|
| 2419 | shl \$5,$num
|
|---|
| 2420 | neg %r10 # restore $num
|
|---|
| 2421 | lea 128($bp,$num),%r13 # end of powers table (+size optimization)
|
|---|
| 2422 | shr \$5+5,$num
|
|---|
| 2423 | movd `($win64?56:8)`(%rax),%xmm5 # load 7th argument
|
|---|
| 2424 | sub \$1,$num
|
|---|
| 2425 | lea .Linc(%rip),%rax
|
|---|
| 2426 | mov %r13,16+8(%rsp) # end of b[num]
|
|---|
| 2427 | mov $num,24+8(%rsp) # inner counter
|
|---|
| 2428 | mov $rp, 56+8(%rsp) # save $rp
|
|---|
| 2429 | ___
|
|---|
| 2430 | my ($aptr, $bptr, $nptr, $tptr, $mi, $bi, $zero, $num)=
|
|---|
| 2431 | ("%rsi","%rdi","%rcx","%rbx","%r8","%r9","%rbp","%rax");
|
|---|
| 2432 | my $rptr=$bptr;
|
|---|
| 2433 | my $STRIDE=2**5*8; # 5 is "window size"
|
|---|
| 2434 | my $N=$STRIDE/4; # should match cache line size
|
|---|
| 2435 | $code.=<<___;
|
|---|
| 2436 | movdqa 0(%rax),%xmm0 # 00000001000000010000000000000000
|
|---|
| 2437 | movdqa 16(%rax),%xmm1 # 00000002000000020000000200000002
|
|---|
| 2438 | lea 88-112(%rsp,%r10),%r10 # place the mask after tp[num+1] (+ICache optimization)
|
|---|
| 2439 | lea 128($bp),$bptr # size optimization
|
|---|
| 2440 |
|
|---|
| 2441 | pshufd \$0,%xmm5,%xmm5 # broadcast index
|
|---|
| 2442 | movdqa %xmm1,%xmm4
|
|---|
| 2443 | .byte 0x67
|
|---|
| 2444 | movdqa %xmm1,%xmm2
|
|---|
| 2445 | ___
|
|---|
| 2446 | ########################################################################
|
|---|
| 2447 | # calculate mask by comparing 0..31 to index and save result to stack
|
|---|
| 2448 | #
|
|---|
| 2449 | $code.=<<___;
|
|---|
| 2450 | .byte 0x67
|
|---|
| 2451 | paddd %xmm0,%xmm1
|
|---|
| 2452 | pcmpeqd %xmm5,%xmm0 # compare to 1,0
|
|---|
| 2453 | movdqa %xmm4,%xmm3
|
|---|
| 2454 | ___
|
|---|
| 2455 | for($i=0;$i<$STRIDE/16-4;$i+=4) {
|
|---|
| 2456 | $code.=<<___;
|
|---|
| 2457 | paddd %xmm1,%xmm2
|
|---|
| 2458 | pcmpeqd %xmm5,%xmm1 # compare to 3,2
|
|---|
| 2459 | movdqa %xmm0,`16*($i+0)+112`(%r10)
|
|---|
| 2460 | movdqa %xmm4,%xmm0
|
|---|
| 2461 |
|
|---|
| 2462 | paddd %xmm2,%xmm3
|
|---|
| 2463 | pcmpeqd %xmm5,%xmm2 # compare to 5,4
|
|---|
| 2464 | movdqa %xmm1,`16*($i+1)+112`(%r10)
|
|---|
| 2465 | movdqa %xmm4,%xmm1
|
|---|
| 2466 |
|
|---|
| 2467 | paddd %xmm3,%xmm0
|
|---|
| 2468 | pcmpeqd %xmm5,%xmm3 # compare to 7,6
|
|---|
| 2469 | movdqa %xmm2,`16*($i+2)+112`(%r10)
|
|---|
| 2470 | movdqa %xmm4,%xmm2
|
|---|
| 2471 |
|
|---|
| 2472 | paddd %xmm0,%xmm1
|
|---|
| 2473 | pcmpeqd %xmm5,%xmm0
|
|---|
| 2474 | movdqa %xmm3,`16*($i+3)+112`(%r10)
|
|---|
| 2475 | movdqa %xmm4,%xmm3
|
|---|
| 2476 | ___
|
|---|
| 2477 | }
|
|---|
| 2478 | $code.=<<___; # last iteration can be optimized
|
|---|
| 2479 | .byte 0x67
|
|---|
| 2480 | paddd %xmm1,%xmm2
|
|---|
| 2481 | pcmpeqd %xmm5,%xmm1
|
|---|
| 2482 | movdqa %xmm0,`16*($i+0)+112`(%r10)
|
|---|
| 2483 |
|
|---|
| 2484 | paddd %xmm2,%xmm3
|
|---|
| 2485 | pcmpeqd %xmm5,%xmm2
|
|---|
| 2486 | movdqa %xmm1,`16*($i+1)+112`(%r10)
|
|---|
| 2487 |
|
|---|
| 2488 | pcmpeqd %xmm5,%xmm3
|
|---|
| 2489 | movdqa %xmm2,`16*($i+2)+112`(%r10)
|
|---|
| 2490 |
|
|---|
| 2491 | pand `16*($i+0)-128`($bptr),%xmm0 # while it's still in register
|
|---|
| 2492 | pand `16*($i+1)-128`($bptr),%xmm1
|
|---|
| 2493 | pand `16*($i+2)-128`($bptr),%xmm2
|
|---|
| 2494 | movdqa %xmm3,`16*($i+3)+112`(%r10)
|
|---|
| 2495 | pand `16*($i+3)-128`($bptr),%xmm3
|
|---|
| 2496 | por %xmm2,%xmm0
|
|---|
| 2497 | por %xmm3,%xmm1
|
|---|
| 2498 | ___
|
|---|
| 2499 | for($i=0;$i<$STRIDE/16-4;$i+=4) {
|
|---|
| 2500 | $code.=<<___;
|
|---|
| 2501 | movdqa `16*($i+0)-128`($bptr),%xmm4
|
|---|
| 2502 | movdqa `16*($i+1)-128`($bptr),%xmm5
|
|---|
| 2503 | movdqa `16*($i+2)-128`($bptr),%xmm2
|
|---|
| 2504 | pand `16*($i+0)+112`(%r10),%xmm4
|
|---|
| 2505 | movdqa `16*($i+3)-128`($bptr),%xmm3
|
|---|
| 2506 | pand `16*($i+1)+112`(%r10),%xmm5
|
|---|
| 2507 | por %xmm4,%xmm0
|
|---|
| 2508 | pand `16*($i+2)+112`(%r10),%xmm2
|
|---|
| 2509 | por %xmm5,%xmm1
|
|---|
| 2510 | pand `16*($i+3)+112`(%r10),%xmm3
|
|---|
| 2511 | por %xmm2,%xmm0
|
|---|
| 2512 | por %xmm3,%xmm1
|
|---|
| 2513 | ___
|
|---|
| 2514 | }
|
|---|
| 2515 | $code.=<<___;
|
|---|
| 2516 | pxor %xmm1,%xmm0
|
|---|
| 2517 | pshufd \$0x4e,%xmm0,%xmm1
|
|---|
| 2518 | por %xmm1,%xmm0
|
|---|
| 2519 | lea $STRIDE($bptr),$bptr
|
|---|
| 2520 | movq %xmm0,%rdx # bp[0]
|
|---|
| 2521 | lea 64+8*4+8(%rsp),$tptr
|
|---|
| 2522 |
|
|---|
| 2523 | mov %rdx,$bi
|
|---|
| 2524 | mulx 0*8($aptr),$mi,%rax # a[0]*b[0]
|
|---|
| 2525 | mulx 1*8($aptr),%r11,%r12 # a[1]*b[0]
|
|---|
| 2526 | add %rax,%r11
|
|---|
| 2527 | mulx 2*8($aptr),%rax,%r13 # ...
|
|---|
| 2528 | adc %rax,%r12
|
|---|
| 2529 | adc \$0,%r13
|
|---|
| 2530 | mulx 3*8($aptr),%rax,%r14
|
|---|
| 2531 |
|
|---|
| 2532 | mov $mi,%r15
|
|---|
| 2533 | imulq 32+8(%rsp),$mi # "t[0]"*n0
|
|---|
| 2534 | xor $zero,$zero # cf=0, of=0
|
|---|
| 2535 | mov $mi,%rdx
|
|---|
| 2536 |
|
|---|
| 2537 | mov $bptr,8+8(%rsp) # off-load &b[i]
|
|---|
| 2538 |
|
|---|
| 2539 | lea 4*8($aptr),$aptr
|
|---|
| 2540 | adcx %rax,%r13
|
|---|
| 2541 | adcx $zero,%r14 # cf=0
|
|---|
| 2542 |
|
|---|
| 2543 | mulx 0*8($nptr),%rax,%r10
|
|---|
| 2544 | adcx %rax,%r15 # discarded
|
|---|
| 2545 | adox %r11,%r10
|
|---|
| 2546 | mulx 1*8($nptr),%rax,%r11
|
|---|
| 2547 | adcx %rax,%r10
|
|---|
| 2548 | adox %r12,%r11
|
|---|
| 2549 | mulx 2*8($nptr),%rax,%r12
|
|---|
| 2550 | mov 24+8(%rsp),$bptr # counter value
|
|---|
| 2551 | mov %r10,-8*4($tptr)
|
|---|
| 2552 | adcx %rax,%r11
|
|---|
| 2553 | adox %r13,%r12
|
|---|
| 2554 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 2555 | mov $bi,%rdx
|
|---|
| 2556 | mov %r11,-8*3($tptr)
|
|---|
| 2557 | adcx %rax,%r12
|
|---|
| 2558 | adox $zero,%r15 # of=0
|
|---|
| 2559 | lea 4*8($nptr),$nptr
|
|---|
| 2560 | mov %r12,-8*2($tptr)
|
|---|
| 2561 | jmp .Lmulx4x_1st
|
|---|
| 2562 |
|
|---|
| 2563 | .align 32
|
|---|
| 2564 | .Lmulx4x_1st:
|
|---|
| 2565 | adcx $zero,%r15 # cf=0, modulo-scheduled
|
|---|
| 2566 | mulx 0*8($aptr),%r10,%rax # a[4]*b[0]
|
|---|
| 2567 | adcx %r14,%r10
|
|---|
| 2568 | mulx 1*8($aptr),%r11,%r14 # a[5]*b[0]
|
|---|
| 2569 | adcx %rax,%r11
|
|---|
| 2570 | mulx 2*8($aptr),%r12,%rax # ...
|
|---|
| 2571 | adcx %r14,%r12
|
|---|
| 2572 | mulx 3*8($aptr),%r13,%r14
|
|---|
| 2573 | .byte 0x67,0x67
|
|---|
| 2574 | mov $mi,%rdx
|
|---|
| 2575 | adcx %rax,%r13
|
|---|
| 2576 | adcx $zero,%r14 # cf=0
|
|---|
| 2577 | lea 4*8($aptr),$aptr
|
|---|
| 2578 | lea 4*8($tptr),$tptr
|
|---|
| 2579 |
|
|---|
| 2580 | adox %r15,%r10
|
|---|
| 2581 | mulx 0*8($nptr),%rax,%r15
|
|---|
| 2582 | adcx %rax,%r10
|
|---|
| 2583 | adox %r15,%r11
|
|---|
| 2584 | mulx 1*8($nptr),%rax,%r15
|
|---|
| 2585 | adcx %rax,%r11
|
|---|
| 2586 | adox %r15,%r12
|
|---|
| 2587 | mulx 2*8($nptr),%rax,%r15
|
|---|
| 2588 | mov %r10,-5*8($tptr)
|
|---|
| 2589 | adcx %rax,%r12
|
|---|
| 2590 | mov %r11,-4*8($tptr)
|
|---|
| 2591 | adox %r15,%r13
|
|---|
| 2592 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 2593 | mov $bi,%rdx
|
|---|
| 2594 | mov %r12,-3*8($tptr)
|
|---|
| 2595 | adcx %rax,%r13
|
|---|
| 2596 | adox $zero,%r15
|
|---|
| 2597 | lea 4*8($nptr),$nptr
|
|---|
| 2598 | mov %r13,-2*8($tptr)
|
|---|
| 2599 |
|
|---|
| 2600 | dec $bptr # of=0, pass cf
|
|---|
| 2601 | jnz .Lmulx4x_1st
|
|---|
| 2602 |
|
|---|
| 2603 | mov 8(%rsp),$num # load -num
|
|---|
| 2604 | adc $zero,%r15 # modulo-scheduled
|
|---|
| 2605 | lea ($aptr,$num),$aptr # rewind $aptr
|
|---|
| 2606 | add %r15,%r14
|
|---|
| 2607 | mov 8+8(%rsp),$bptr # re-load &b[i]
|
|---|
| 2608 | adc $zero,$zero # top-most carry
|
|---|
| 2609 | mov %r14,-1*8($tptr)
|
|---|
| 2610 | jmp .Lmulx4x_outer
|
|---|
| 2611 |
|
|---|
| 2612 | .align 32
|
|---|
| 2613 | .Lmulx4x_outer:
|
|---|
| 2614 | lea 16-256($tptr),%r10 # where 256-byte mask is (+density control)
|
|---|
| 2615 | pxor %xmm4,%xmm4
|
|---|
| 2616 | .byte 0x67,0x67
|
|---|
| 2617 | pxor %xmm5,%xmm5
|
|---|
| 2618 | ___
|
|---|
| 2619 | for($i=0;$i<$STRIDE/16;$i+=4) {
|
|---|
| 2620 | $code.=<<___;
|
|---|
| 2621 | movdqa `16*($i+0)-128`($bptr),%xmm0
|
|---|
| 2622 | movdqa `16*($i+1)-128`($bptr),%xmm1
|
|---|
| 2623 | movdqa `16*($i+2)-128`($bptr),%xmm2
|
|---|
| 2624 | pand `16*($i+0)+256`(%r10),%xmm0
|
|---|
| 2625 | movdqa `16*($i+3)-128`($bptr),%xmm3
|
|---|
| 2626 | pand `16*($i+1)+256`(%r10),%xmm1
|
|---|
| 2627 | por %xmm0,%xmm4
|
|---|
| 2628 | pand `16*($i+2)+256`(%r10),%xmm2
|
|---|
| 2629 | por %xmm1,%xmm5
|
|---|
| 2630 | pand `16*($i+3)+256`(%r10),%xmm3
|
|---|
| 2631 | por %xmm2,%xmm4
|
|---|
| 2632 | por %xmm3,%xmm5
|
|---|
| 2633 | ___
|
|---|
| 2634 | }
|
|---|
| 2635 | $code.=<<___;
|
|---|
| 2636 | por %xmm5,%xmm4
|
|---|
| 2637 | pshufd \$0x4e,%xmm4,%xmm0
|
|---|
| 2638 | por %xmm4,%xmm0
|
|---|
| 2639 | lea $STRIDE($bptr),$bptr
|
|---|
| 2640 | movq %xmm0,%rdx # m0=bp[i]
|
|---|
| 2641 |
|
|---|
| 2642 | mov $zero,($tptr) # save top-most carry
|
|---|
| 2643 | lea 4*8($tptr,$num),$tptr # rewind $tptr
|
|---|
| 2644 | mulx 0*8($aptr),$mi,%r11 # a[0]*b[i]
|
|---|
| 2645 | xor $zero,$zero # cf=0, of=0
|
|---|
| 2646 | mov %rdx,$bi
|
|---|
| 2647 | mulx 1*8($aptr),%r14,%r12 # a[1]*b[i]
|
|---|
| 2648 | adox -4*8($tptr),$mi # +t[0]
|
|---|
| 2649 | adcx %r14,%r11
|
|---|
| 2650 | mulx 2*8($aptr),%r15,%r13 # ...
|
|---|
| 2651 | adox -3*8($tptr),%r11
|
|---|
| 2652 | adcx %r15,%r12
|
|---|
| 2653 | mulx 3*8($aptr),%rdx,%r14
|
|---|
| 2654 | adox -2*8($tptr),%r12
|
|---|
| 2655 | adcx %rdx,%r13
|
|---|
| 2656 | lea ($nptr,$num),$nptr # rewind $nptr
|
|---|
| 2657 | lea 4*8($aptr),$aptr
|
|---|
| 2658 | adox -1*8($tptr),%r13
|
|---|
| 2659 | adcx $zero,%r14
|
|---|
| 2660 | adox $zero,%r14
|
|---|
| 2661 |
|
|---|
| 2662 | mov $mi,%r15
|
|---|
| 2663 | imulq 32+8(%rsp),$mi # "t[0]"*n0
|
|---|
| 2664 |
|
|---|
| 2665 | mov $mi,%rdx
|
|---|
| 2666 | xor $zero,$zero # cf=0, of=0
|
|---|
| 2667 | mov $bptr,8+8(%rsp) # off-load &b[i]
|
|---|
| 2668 |
|
|---|
| 2669 | mulx 0*8($nptr),%rax,%r10
|
|---|
| 2670 | adcx %rax,%r15 # discarded
|
|---|
| 2671 | adox %r11,%r10
|
|---|
| 2672 | mulx 1*8($nptr),%rax,%r11
|
|---|
| 2673 | adcx %rax,%r10
|
|---|
| 2674 | adox %r12,%r11
|
|---|
| 2675 | mulx 2*8($nptr),%rax,%r12
|
|---|
| 2676 | adcx %rax,%r11
|
|---|
| 2677 | adox %r13,%r12
|
|---|
| 2678 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 2679 | mov $bi,%rdx
|
|---|
| 2680 | mov 24+8(%rsp),$bptr # counter value
|
|---|
| 2681 | mov %r10,-8*4($tptr)
|
|---|
| 2682 | adcx %rax,%r12
|
|---|
| 2683 | mov %r11,-8*3($tptr)
|
|---|
| 2684 | adox $zero,%r15 # of=0
|
|---|
| 2685 | mov %r12,-8*2($tptr)
|
|---|
| 2686 | lea 4*8($nptr),$nptr
|
|---|
| 2687 | jmp .Lmulx4x_inner
|
|---|
| 2688 |
|
|---|
| 2689 | .align 32
|
|---|
| 2690 | .Lmulx4x_inner:
|
|---|
| 2691 | mulx 0*8($aptr),%r10,%rax # a[4]*b[i]
|
|---|
| 2692 | adcx $zero,%r15 # cf=0, modulo-scheduled
|
|---|
| 2693 | adox %r14,%r10
|
|---|
| 2694 | mulx 1*8($aptr),%r11,%r14 # a[5]*b[i]
|
|---|
| 2695 | adcx 0*8($tptr),%r10
|
|---|
| 2696 | adox %rax,%r11
|
|---|
| 2697 | mulx 2*8($aptr),%r12,%rax # ...
|
|---|
| 2698 | adcx 1*8($tptr),%r11
|
|---|
| 2699 | adox %r14,%r12
|
|---|
| 2700 | mulx 3*8($aptr),%r13,%r14
|
|---|
| 2701 | mov $mi,%rdx
|
|---|
| 2702 | adcx 2*8($tptr),%r12
|
|---|
| 2703 | adox %rax,%r13
|
|---|
| 2704 | adcx 3*8($tptr),%r13
|
|---|
| 2705 | adox $zero,%r14 # of=0
|
|---|
| 2706 | lea 4*8($aptr),$aptr
|
|---|
| 2707 | lea 4*8($tptr),$tptr
|
|---|
| 2708 | adcx $zero,%r14 # cf=0
|
|---|
| 2709 |
|
|---|
| 2710 | adox %r15,%r10
|
|---|
| 2711 | mulx 0*8($nptr),%rax,%r15
|
|---|
| 2712 | adcx %rax,%r10
|
|---|
| 2713 | adox %r15,%r11
|
|---|
| 2714 | mulx 1*8($nptr),%rax,%r15
|
|---|
| 2715 | adcx %rax,%r11
|
|---|
| 2716 | adox %r15,%r12
|
|---|
| 2717 | mulx 2*8($nptr),%rax,%r15
|
|---|
| 2718 | mov %r10,-5*8($tptr)
|
|---|
| 2719 | adcx %rax,%r12
|
|---|
| 2720 | adox %r15,%r13
|
|---|
| 2721 | mov %r11,-4*8($tptr)
|
|---|
| 2722 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 2723 | mov $bi,%rdx
|
|---|
| 2724 | lea 4*8($nptr),$nptr
|
|---|
| 2725 | mov %r12,-3*8($tptr)
|
|---|
| 2726 | adcx %rax,%r13
|
|---|
| 2727 | adox $zero,%r15
|
|---|
| 2728 | mov %r13,-2*8($tptr)
|
|---|
| 2729 |
|
|---|
| 2730 | dec $bptr # of=0, pass cf
|
|---|
| 2731 | jnz .Lmulx4x_inner
|
|---|
| 2732 |
|
|---|
| 2733 | mov 0+8(%rsp),$num # load -num
|
|---|
| 2734 | adc $zero,%r15 # modulo-scheduled
|
|---|
| 2735 | sub 0*8($tptr),$bptr # pull top-most carry to %cf
|
|---|
| 2736 | mov 8+8(%rsp),$bptr # re-load &b[i]
|
|---|
| 2737 | mov 16+8(%rsp),%r10
|
|---|
| 2738 | adc %r15,%r14
|
|---|
| 2739 | lea ($aptr,$num),$aptr # rewind $aptr
|
|---|
| 2740 | adc $zero,$zero # top-most carry
|
|---|
| 2741 | mov %r14,-1*8($tptr)
|
|---|
| 2742 |
|
|---|
| 2743 | cmp %r10,$bptr
|
|---|
| 2744 | jb .Lmulx4x_outer
|
|---|
| 2745 |
|
|---|
| 2746 | mov -8($nptr),%r10
|
|---|
| 2747 | mov $zero,%r8
|
|---|
| 2748 | mov ($nptr,$num),%r12
|
|---|
| 2749 | lea ($nptr,$num),%rbp # rewind $nptr
|
|---|
| 2750 | mov $num,%rcx
|
|---|
| 2751 | lea ($tptr,$num),%rdi # rewind $tptr
|
|---|
| 2752 | xor %eax,%eax
|
|---|
| 2753 | xor %r15,%r15
|
|---|
| 2754 | sub %r14,%r10 # compare top-most words
|
|---|
| 2755 | adc %r15,%r15
|
|---|
| 2756 | or %r15,%r8
|
|---|
| 2757 | sar \$3+2,%rcx
|
|---|
| 2758 | sub %r8,%rax # %rax=-%r8
|
|---|
| 2759 | mov 56+8(%rsp),%rdx # restore rp
|
|---|
| 2760 | dec %r12 # so that after 'not' we get -n[0]
|
|---|
| 2761 | mov 8*1(%rbp),%r13
|
|---|
| 2762 | xor %r8,%r8
|
|---|
| 2763 | mov 8*2(%rbp),%r14
|
|---|
| 2764 | mov 8*3(%rbp),%r15
|
|---|
| 2765 | jmp .Lsqrx4x_sub_entry # common post-condition
|
|---|
| 2766 | .cfi_endproc
|
|---|
| 2767 | .size mulx4x_internal,.-mulx4x_internal
|
|---|
| 2768 | ___
|
|---|
| 2769 | } |
|---|
| 2770 | {
|
|---|
| 2771 | ######################################################################
|
|---|
| 2772 | # void bn_power5(
|
|---|
| 2773 | my $rptr="%rdi"; # BN_ULONG *rptr,
|
|---|
| 2774 | my $aptr="%rsi"; # const BN_ULONG *aptr,
|
|---|
| 2775 | my $bptr="%rdx"; # const void *table,
|
|---|
| 2776 | my $nptr="%rcx"; # const BN_ULONG *nptr,
|
|---|
| 2777 | my $n0 ="%r8"; # const BN_ULONG *n0);
|
|---|
| 2778 | my $num ="%r9"; # int num, has to be divisible by 8
|
|---|
| 2779 | # int pwr);
|
|---|
| 2780 |
|
|---|
| 2781 | my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
|
|---|
| 2782 | my @A0=("%r10","%r11");
|
|---|
| 2783 | my @A1=("%r12","%r13");
|
|---|
| 2784 | my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
|
|---|
| 2785 |
|
|---|
| 2786 | $code.=<<___;
|
|---|
| 2787 | .type bn_powerx5,\@function,6
|
|---|
| 2788 | .align 32
|
|---|
| 2789 | bn_powerx5:
|
|---|
| 2790 | .cfi_startproc
|
|---|
| 2791 | mov %rsp,%rax
|
|---|
| 2792 | .cfi_def_cfa_register %rax
|
|---|
| 2793 | .Lpowerx5_enter:
|
|---|
| 2794 | push %rbx
|
|---|
| 2795 | .cfi_push %rbx
|
|---|
| 2796 | push %rbp
|
|---|
| 2797 | .cfi_push %rbp
|
|---|
| 2798 | push %r12
|
|---|
| 2799 | .cfi_push %r12
|
|---|
| 2800 | push %r13
|
|---|
| 2801 | .cfi_push %r13
|
|---|
| 2802 | push %r14
|
|---|
| 2803 | .cfi_push %r14
|
|---|
| 2804 | push %r15
|
|---|
| 2805 | .cfi_push %r15
|
|---|
| 2806 | .Lpowerx5_prologue:
|
|---|
| 2807 |
|
|---|
| 2808 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 2809 | lea ($num,$num,2),%r10 # 3*$num in bytes
|
|---|
| 2810 | neg $num
|
|---|
| 2811 | mov ($n0),$n0 # *n0
|
|---|
| 2812 |
|
|---|
| 2813 | ##############################################################
|
|---|
| 2814 | # Ensure that stack frame doesn't alias with $rptr+3*$num
|
|---|
| 2815 | # modulo 4096, which covers ret[num], am[num] and n[num]
|
|---|
| 2816 | # (see bn_exp.c). This is done to allow memory disambiguation
|
|---|
| 2817 | # logic do its magic. [Extra 256 bytes is for power mask
|
|---|
| 2818 | # calculated from 7th argument, the index.]
|
|---|
| 2819 | #
|
|---|
| 2820 | lea -320(%rsp,$num,2),%r11
|
|---|
| 2821 | mov %rsp,%rbp
|
|---|
| 2822 | sub $rptr,%r11
|
|---|
| 2823 | and \$4095,%r11
|
|---|
| 2824 | cmp %r11,%r10
|
|---|
| 2825 | jb .Lpwrx_sp_alt
|
|---|
| 2826 | sub %r11,%rbp # align with $aptr
|
|---|
| 2827 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*$num*8+256)
|
|---|
| 2828 | jmp .Lpwrx_sp_done
|
|---|
| 2829 |
|
|---|
| 2830 | .align 32
|
|---|
| 2831 | .Lpwrx_sp_alt:
|
|---|
| 2832 | lea 4096-320(,$num,2),%r10
|
|---|
| 2833 | lea -320(%rbp,$num,2),%rbp # alloca(frame+2*$num*8+256)
|
|---|
| 2834 | sub %r10,%r11
|
|---|
| 2835 | mov \$0,%r10
|
|---|
| 2836 | cmovc %r10,%r11
|
|---|
| 2837 | sub %r11,%rbp
|
|---|
| 2838 | .Lpwrx_sp_done:
|
|---|
| 2839 | and \$-64,%rbp
|
|---|
| 2840 | mov %rsp,%r11
|
|---|
| 2841 | sub %rbp,%r11
|
|---|
| 2842 | and \$-4096,%r11
|
|---|
| 2843 | lea (%rbp,%r11),%rsp
|
|---|
| 2844 | mov (%rsp),%r10
|
|---|
| 2845 | cmp %rbp,%rsp
|
|---|
| 2846 | ja .Lpwrx_page_walk
|
|---|
| 2847 | jmp .Lpwrx_page_walk_done
|
|---|
| 2848 |
|
|---|
| 2849 | .Lpwrx_page_walk:
|
|---|
| 2850 | lea -4096(%rsp),%rsp
|
|---|
| 2851 | mov (%rsp),%r10
|
|---|
| 2852 | cmp %rbp,%rsp
|
|---|
| 2853 | ja .Lpwrx_page_walk
|
|---|
| 2854 | .Lpwrx_page_walk_done:
|
|---|
| 2855 |
|
|---|
| 2856 | mov $num,%r10
|
|---|
| 2857 | neg $num
|
|---|
| 2858 |
|
|---|
| 2859 | ##############################################################
|
|---|
| 2860 | # Stack layout
|
|---|
| 2861 | #
|
|---|
| 2862 | # +0 saved $num, used in reduction section
|
|---|
| 2863 | # +8 &t[2*$num], used in reduction section
|
|---|
| 2864 | # +16 intermediate carry bit
|
|---|
| 2865 | # +24 top-most carry bit, used in reduction section
|
|---|
| 2866 | # +32 saved *n0
|
|---|
| 2867 | # +40 saved %rsp
|
|---|
| 2868 | # +48 t[2*$num]
|
|---|
| 2869 | #
|
|---|
| 2870 | pxor %xmm0,%xmm0
|
|---|
| 2871 | movq $rptr,%xmm1 # save $rptr
|
|---|
| 2872 | movq $nptr,%xmm2 # save $nptr
|
|---|
| 2873 | movq %r10, %xmm3 # -$num
|
|---|
| 2874 | movq $bptr,%xmm4
|
|---|
| 2875 | mov $n0, 32(%rsp)
|
|---|
| 2876 | mov %rax, 40(%rsp) # save original %rsp
|
|---|
| 2877 | .cfi_cfa_expression %rsp+40,deref,+8
|
|---|
| 2878 | .Lpowerx5_body:
|
|---|
| 2879 |
|
|---|
| 2880 | call __bn_sqrx8x_internal
|
|---|
| 2881 | call __bn_postx4x_internal
|
|---|
| 2882 | call __bn_sqrx8x_internal
|
|---|
| 2883 | call __bn_postx4x_internal
|
|---|
| 2884 | call __bn_sqrx8x_internal
|
|---|
| 2885 | call __bn_postx4x_internal
|
|---|
| 2886 | call __bn_sqrx8x_internal
|
|---|
| 2887 | call __bn_postx4x_internal
|
|---|
| 2888 | call __bn_sqrx8x_internal
|
|---|
| 2889 | call __bn_postx4x_internal
|
|---|
| 2890 |
|
|---|
| 2891 | mov %r10,$num # -num
|
|---|
| 2892 | mov $aptr,$rptr
|
|---|
| 2893 | movq %xmm2,$nptr
|
|---|
| 2894 | movq %xmm4,$bptr
|
|---|
| 2895 | mov 40(%rsp),%rax
|
|---|
| 2896 |
|
|---|
| 2897 | call mulx4x_internal
|
|---|
| 2898 |
|
|---|
| 2899 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 2900 | .cfi_def_cfa %rsi,8
|
|---|
| 2901 | mov \$1,%rax
|
|---|
| 2902 |
|
|---|
| 2903 | mov -48(%rsi),%r15
|
|---|
| 2904 | .cfi_restore %r15
|
|---|
| 2905 | mov -40(%rsi),%r14
|
|---|
| 2906 | .cfi_restore %r14
|
|---|
| 2907 | mov -32(%rsi),%r13
|
|---|
| 2908 | .cfi_restore %r13
|
|---|
| 2909 | mov -24(%rsi),%r12
|
|---|
| 2910 | .cfi_restore %r12
|
|---|
| 2911 | mov -16(%rsi),%rbp
|
|---|
| 2912 | .cfi_restore %rbp
|
|---|
| 2913 | mov -8(%rsi),%rbx
|
|---|
| 2914 | .cfi_restore %rbx
|
|---|
| 2915 | lea (%rsi),%rsp
|
|---|
| 2916 | .cfi_def_cfa_register %rsp
|
|---|
| 2917 | .Lpowerx5_epilogue:
|
|---|
| 2918 | ret
|
|---|
| 2919 | .cfi_endproc
|
|---|
| 2920 | .size bn_powerx5,.-bn_powerx5
|
|---|
| 2921 |
|
|---|
| 2922 | .globl bn_sqrx8x_internal
|
|---|
| 2923 | .hidden bn_sqrx8x_internal
|
|---|
| 2924 | .type bn_sqrx8x_internal,\@abi-omnipotent
|
|---|
| 2925 | .align 32
|
|---|
| 2926 | bn_sqrx8x_internal:
|
|---|
| 2927 | __bn_sqrx8x_internal:
|
|---|
| 2928 | .cfi_startproc
|
|---|
| 2929 | ##################################################################
|
|---|
| 2930 | # Squaring part:
|
|---|
| 2931 | #
|
|---|
| 2932 | # a) multiply-n-add everything but a[i]*a[i];
|
|---|
| 2933 | # b) shift result of a) by 1 to the left and accumulate
|
|---|
| 2934 | # a[i]*a[i] products;
|
|---|
| 2935 | #
|
|---|
| 2936 | ##################################################################
|
|---|
| 2937 | # a[7]a[7]a[6]a[6]a[5]a[5]a[4]a[4]a[3]a[3]a[2]a[2]a[1]a[1]a[0]a[0]
|
|---|
| 2938 | # a[1]a[0]
|
|---|
| 2939 | # a[2]a[0]
|
|---|
| 2940 | # a[3]a[0]
|
|---|
| 2941 | # a[2]a[1]
|
|---|
| 2942 | # a[3]a[1]
|
|---|
| 2943 | # a[3]a[2]
|
|---|
| 2944 | #
|
|---|
| 2945 | # a[4]a[0]
|
|---|
| 2946 | # a[5]a[0]
|
|---|
| 2947 | # a[6]a[0]
|
|---|
| 2948 | # a[7]a[0]
|
|---|
| 2949 | # a[4]a[1]
|
|---|
| 2950 | # a[5]a[1]
|
|---|
| 2951 | # a[6]a[1]
|
|---|
| 2952 | # a[7]a[1]
|
|---|
| 2953 | # a[4]a[2]
|
|---|
| 2954 | # a[5]a[2]
|
|---|
| 2955 | # a[6]a[2]
|
|---|
| 2956 | # a[7]a[2]
|
|---|
| 2957 | # a[4]a[3]
|
|---|
| 2958 | # a[5]a[3]
|
|---|
| 2959 | # a[6]a[3]
|
|---|
| 2960 | # a[7]a[3]
|
|---|
| 2961 | #
|
|---|
| 2962 | # a[5]a[4]
|
|---|
| 2963 | # a[6]a[4]
|
|---|
| 2964 | # a[7]a[4]
|
|---|
| 2965 | # a[6]a[5]
|
|---|
| 2966 | # a[7]a[5]
|
|---|
| 2967 | # a[7]a[6]
|
|---|
| 2968 | # a[7]a[7]a[6]a[6]a[5]a[5]a[4]a[4]a[3]a[3]a[2]a[2]a[1]a[1]a[0]a[0]
|
|---|
| 2969 | ___
|
|---|
| 2970 | {
|
|---|
| 2971 | my ($zero,$carry)=("%rbp","%rcx");
|
|---|
| 2972 | my $aaptr=$zero;
|
|---|
| 2973 | $code.=<<___;
|
|---|
| 2974 | lea 48+8(%rsp),$tptr
|
|---|
| 2975 | lea ($aptr,$num),$aaptr
|
|---|
| 2976 | mov $num,0+8(%rsp) # save $num
|
|---|
| 2977 | mov $aaptr,8+8(%rsp) # save end of $aptr
|
|---|
| 2978 | jmp .Lsqr8x_zero_start
|
|---|
| 2979 |
|
|---|
| 2980 | .align 32
|
|---|
| 2981 | .byte 0x66,0x66,0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00
|
|---|
| 2982 | .Lsqrx8x_zero:
|
|---|
| 2983 | .byte 0x3e
|
|---|
| 2984 | movdqa %xmm0,0*8($tptr)
|
|---|
| 2985 | movdqa %xmm0,2*8($tptr)
|
|---|
| 2986 | movdqa %xmm0,4*8($tptr)
|
|---|
| 2987 | movdqa %xmm0,6*8($tptr)
|
|---|
| 2988 | .Lsqr8x_zero_start: # aligned at 32
|
|---|
| 2989 | movdqa %xmm0,8*8($tptr)
|
|---|
| 2990 | movdqa %xmm0,10*8($tptr)
|
|---|
| 2991 | movdqa %xmm0,12*8($tptr)
|
|---|
| 2992 | movdqa %xmm0,14*8($tptr)
|
|---|
| 2993 | lea 16*8($tptr),$tptr
|
|---|
| 2994 | sub \$64,$num
|
|---|
| 2995 | jnz .Lsqrx8x_zero
|
|---|
| 2996 |
|
|---|
| 2997 | mov 0*8($aptr),%rdx # a[0], modulo-scheduled
|
|---|
| 2998 | #xor %r9,%r9 # t[1], ex-$num, zero already
|
|---|
| 2999 | xor %r10,%r10
|
|---|
| 3000 | xor %r11,%r11
|
|---|
| 3001 | xor %r12,%r12
|
|---|
| 3002 | xor %r13,%r13
|
|---|
| 3003 | xor %r14,%r14
|
|---|
| 3004 | xor %r15,%r15
|
|---|
| 3005 | lea 48+8(%rsp),$tptr
|
|---|
| 3006 | xor $zero,$zero # cf=0, cf=0
|
|---|
| 3007 | jmp .Lsqrx8x_outer_loop
|
|---|
| 3008 |
|
|---|
| 3009 | .align 32
|
|---|
| 3010 | .Lsqrx8x_outer_loop:
|
|---|
| 3011 | mulx 1*8($aptr),%r8,%rax # a[1]*a[0]
|
|---|
| 3012 | adcx %r9,%r8 # a[1]*a[0]+=t[1]
|
|---|
| 3013 | adox %rax,%r10
|
|---|
| 3014 | mulx 2*8($aptr),%r9,%rax # a[2]*a[0]
|
|---|
| 3015 | adcx %r10,%r9
|
|---|
| 3016 | adox %rax,%r11
|
|---|
| 3017 | .byte 0xc4,0xe2,0xab,0xf6,0x86,0x18,0x00,0x00,0x00 # mulx 3*8($aptr),%r10,%rax # ...
|
|---|
| 3018 | adcx %r11,%r10
|
|---|
| 3019 | adox %rax,%r12
|
|---|
| 3020 | .byte 0xc4,0xe2,0xa3,0xf6,0x86,0x20,0x00,0x00,0x00 # mulx 4*8($aptr),%r11,%rax
|
|---|
| 3021 | adcx %r12,%r11
|
|---|
| 3022 | adox %rax,%r13
|
|---|
| 3023 | mulx 5*8($aptr),%r12,%rax
|
|---|
| 3024 | adcx %r13,%r12
|
|---|
| 3025 | adox %rax,%r14
|
|---|
| 3026 | mulx 6*8($aptr),%r13,%rax
|
|---|
| 3027 | adcx %r14,%r13
|
|---|
| 3028 | adox %r15,%rax
|
|---|
| 3029 | mulx 7*8($aptr),%r14,%r15
|
|---|
| 3030 | mov 1*8($aptr),%rdx # a[1]
|
|---|
| 3031 | adcx %rax,%r14
|
|---|
| 3032 | adox $zero,%r15
|
|---|
| 3033 | adc 8*8($tptr),%r15
|
|---|
| 3034 | mov %r8,1*8($tptr) # t[1]
|
|---|
| 3035 | mov %r9,2*8($tptr) # t[2]
|
|---|
| 3036 | sbb $carry,$carry # mov %cf,$carry
|
|---|
| 3037 | xor $zero,$zero # cf=0, of=0
|
|---|
| 3038 |
|
|---|
| 3039 |
|
|---|
| 3040 | mulx 2*8($aptr),%r8,%rbx # a[2]*a[1]
|
|---|
| 3041 | mulx 3*8($aptr),%r9,%rax # a[3]*a[1]
|
|---|
| 3042 | adcx %r10,%r8
|
|---|
| 3043 | adox %rbx,%r9
|
|---|
| 3044 | mulx 4*8($aptr),%r10,%rbx # ...
|
|---|
| 3045 | adcx %r11,%r9
|
|---|
| 3046 | adox %rax,%r10
|
|---|
| 3047 | .byte 0xc4,0xe2,0xa3,0xf6,0x86,0x28,0x00,0x00,0x00 # mulx 5*8($aptr),%r11,%rax
|
|---|
| 3048 | adcx %r12,%r10
|
|---|
| 3049 | adox %rbx,%r11
|
|---|
| 3050 | .byte 0xc4,0xe2,0x9b,0xf6,0x9e,0x30,0x00,0x00,0x00 # mulx 6*8($aptr),%r12,%rbx
|
|---|
| 3051 | adcx %r13,%r11
|
|---|
| 3052 | adox %r14,%r12
|
|---|
| 3053 | .byte 0xc4,0x62,0x93,0xf6,0xb6,0x38,0x00,0x00,0x00 # mulx 7*8($aptr),%r13,%r14
|
|---|
| 3054 | mov 2*8($aptr),%rdx # a[2]
|
|---|
| 3055 | adcx %rax,%r12
|
|---|
| 3056 | adox %rbx,%r13
|
|---|
| 3057 | adcx %r15,%r13
|
|---|
| 3058 | adox $zero,%r14 # of=0
|
|---|
| 3059 | adcx $zero,%r14 # cf=0
|
|---|
| 3060 |
|
|---|
| 3061 | mov %r8,3*8($tptr) # t[3]
|
|---|
| 3062 | mov %r9,4*8($tptr) # t[4]
|
|---|
| 3063 |
|
|---|
| 3064 | mulx 3*8($aptr),%r8,%rbx # a[3]*a[2]
|
|---|
| 3065 | mulx 4*8($aptr),%r9,%rax # a[4]*a[2]
|
|---|
| 3066 | adcx %r10,%r8
|
|---|
| 3067 | adox %rbx,%r9
|
|---|
| 3068 | mulx 5*8($aptr),%r10,%rbx # ...
|
|---|
| 3069 | adcx %r11,%r9
|
|---|
| 3070 | adox %rax,%r10
|
|---|
| 3071 | .byte 0xc4,0xe2,0xa3,0xf6,0x86,0x30,0x00,0x00,0x00 # mulx 6*8($aptr),%r11,%rax
|
|---|
| 3072 | adcx %r12,%r10
|
|---|
| 3073 | adox %r13,%r11
|
|---|
| 3074 | .byte 0xc4,0x62,0x9b,0xf6,0xae,0x38,0x00,0x00,0x00 # mulx 7*8($aptr),%r12,%r13
|
|---|
| 3075 | .byte 0x3e
|
|---|
| 3076 | mov 3*8($aptr),%rdx # a[3]
|
|---|
| 3077 | adcx %rbx,%r11
|
|---|
| 3078 | adox %rax,%r12
|
|---|
| 3079 | adcx %r14,%r12
|
|---|
| 3080 | mov %r8,5*8($tptr) # t[5]
|
|---|
| 3081 | mov %r9,6*8($tptr) # t[6]
|
|---|
| 3082 | mulx 4*8($aptr),%r8,%rax # a[4]*a[3]
|
|---|
| 3083 | adox $zero,%r13 # of=0
|
|---|
| 3084 | adcx $zero,%r13 # cf=0
|
|---|
| 3085 |
|
|---|
| 3086 | mulx 5*8($aptr),%r9,%rbx # a[5]*a[3]
|
|---|
| 3087 | adcx %r10,%r8
|
|---|
| 3088 | adox %rax,%r9
|
|---|
| 3089 | mulx 6*8($aptr),%r10,%rax # ...
|
|---|
| 3090 | adcx %r11,%r9
|
|---|
| 3091 | adox %r12,%r10
|
|---|
| 3092 | mulx 7*8($aptr),%r11,%r12
|
|---|
| 3093 | mov 4*8($aptr),%rdx # a[4]
|
|---|
| 3094 | mov 5*8($aptr),%r14 # a[5]
|
|---|
| 3095 | adcx %rbx,%r10
|
|---|
| 3096 | adox %rax,%r11
|
|---|
| 3097 | mov 6*8($aptr),%r15 # a[6]
|
|---|
| 3098 | adcx %r13,%r11
|
|---|
| 3099 | adox $zero,%r12 # of=0
|
|---|
| 3100 | adcx $zero,%r12 # cf=0
|
|---|
| 3101 |
|
|---|
| 3102 | mov %r8,7*8($tptr) # t[7]
|
|---|
| 3103 | mov %r9,8*8($tptr) # t[8]
|
|---|
| 3104 |
|
|---|
| 3105 | mulx %r14,%r9,%rax # a[5]*a[4]
|
|---|
| 3106 | mov 7*8($aptr),%r8 # a[7]
|
|---|
| 3107 | adcx %r10,%r9
|
|---|
| 3108 | mulx %r15,%r10,%rbx # a[6]*a[4]
|
|---|
| 3109 | adox %rax,%r10
|
|---|
| 3110 | adcx %r11,%r10
|
|---|
| 3111 | mulx %r8,%r11,%rax # a[7]*a[4]
|
|---|
| 3112 | mov %r14,%rdx # a[5]
|
|---|
| 3113 | adox %rbx,%r11
|
|---|
| 3114 | adcx %r12,%r11
|
|---|
| 3115 | #adox $zero,%rax # of=0
|
|---|
| 3116 | adcx $zero,%rax # cf=0
|
|---|
| 3117 |
|
|---|
| 3118 | mulx %r15,%r14,%rbx # a[6]*a[5]
|
|---|
| 3119 | mulx %r8,%r12,%r13 # a[7]*a[5]
|
|---|
| 3120 | mov %r15,%rdx # a[6]
|
|---|
| 3121 | lea 8*8($aptr),$aptr
|
|---|
| 3122 | adcx %r14,%r11
|
|---|
| 3123 | adox %rbx,%r12
|
|---|
| 3124 | adcx %rax,%r12
|
|---|
| 3125 | adox $zero,%r13
|
|---|
| 3126 |
|
|---|
| 3127 | .byte 0x67,0x67
|
|---|
| 3128 | mulx %r8,%r8,%r14 # a[7]*a[6]
|
|---|
| 3129 | adcx %r8,%r13
|
|---|
| 3130 | adcx $zero,%r14
|
|---|
| 3131 |
|
|---|
| 3132 | cmp 8+8(%rsp),$aptr
|
|---|
| 3133 | je .Lsqrx8x_outer_break
|
|---|
| 3134 |
|
|---|
| 3135 | neg $carry # mov $carry,%cf
|
|---|
| 3136 | mov \$-8,%rcx
|
|---|
| 3137 | mov $zero,%r15
|
|---|
| 3138 | mov 8*8($tptr),%r8
|
|---|
| 3139 | adcx 9*8($tptr),%r9 # +=t[9]
|
|---|
| 3140 | adcx 10*8($tptr),%r10 # ...
|
|---|
| 3141 | adcx 11*8($tptr),%r11
|
|---|
| 3142 | adc 12*8($tptr),%r12
|
|---|
| 3143 | adc 13*8($tptr),%r13
|
|---|
| 3144 | adc 14*8($tptr),%r14
|
|---|
| 3145 | adc 15*8($tptr),%r15
|
|---|
| 3146 | lea ($aptr),$aaptr
|
|---|
| 3147 | lea 2*64($tptr),$tptr
|
|---|
| 3148 | sbb %rax,%rax # mov %cf,$carry
|
|---|
| 3149 |
|
|---|
| 3150 | mov -64($aptr),%rdx # a[0]
|
|---|
| 3151 | mov %rax,16+8(%rsp) # offload $carry
|
|---|
| 3152 | mov $tptr,24+8(%rsp)
|
|---|
| 3153 |
|
|---|
| 3154 | #lea 8*8($tptr),$tptr # see 2*8*8($tptr) above
|
|---|
| 3155 | xor %eax,%eax # cf=0, of=0
|
|---|
| 3156 | jmp .Lsqrx8x_loop
|
|---|
| 3157 |
|
|---|
| 3158 | .align 32
|
|---|
| 3159 | .Lsqrx8x_loop:
|
|---|
| 3160 | mov %r8,%rbx
|
|---|
| 3161 | mulx 0*8($aaptr),%rax,%r8 # a[8]*a[i]
|
|---|
| 3162 | adcx %rax,%rbx # +=t[8]
|
|---|
| 3163 | adox %r9,%r8
|
|---|
| 3164 |
|
|---|
| 3165 | mulx 1*8($aaptr),%rax,%r9 # ...
|
|---|
| 3166 | adcx %rax,%r8
|
|---|
| 3167 | adox %r10,%r9
|
|---|
| 3168 |
|
|---|
| 3169 | mulx 2*8($aaptr),%rax,%r10
|
|---|
| 3170 | adcx %rax,%r9
|
|---|
| 3171 | adox %r11,%r10
|
|---|
| 3172 |
|
|---|
| 3173 | mulx 3*8($aaptr),%rax,%r11
|
|---|
| 3174 | adcx %rax,%r10
|
|---|
| 3175 | adox %r12,%r11
|
|---|
| 3176 |
|
|---|
| 3177 | .byte 0xc4,0x62,0xfb,0xf6,0xa5,0x20,0x00,0x00,0x00 # mulx 4*8($aaptr),%rax,%r12
|
|---|
| 3178 | adcx %rax,%r11
|
|---|
| 3179 | adox %r13,%r12
|
|---|
| 3180 |
|
|---|
| 3181 | mulx 5*8($aaptr),%rax,%r13
|
|---|
| 3182 | adcx %rax,%r12
|
|---|
| 3183 | adox %r14,%r13
|
|---|
| 3184 |
|
|---|
| 3185 | mulx 6*8($aaptr),%rax,%r14
|
|---|
| 3186 | mov %rbx,($tptr,%rcx,8) # store t[8+i]
|
|---|
| 3187 | mov \$0,%ebx
|
|---|
| 3188 | adcx %rax,%r13
|
|---|
| 3189 | adox %r15,%r14
|
|---|
| 3190 |
|
|---|
| 3191 | .byte 0xc4,0x62,0xfb,0xf6,0xbd,0x38,0x00,0x00,0x00 # mulx 7*8($aaptr),%rax,%r15
|
|---|
| 3192 | mov 8($aptr,%rcx,8),%rdx # a[i]
|
|---|
| 3193 | adcx %rax,%r14
|
|---|
| 3194 | adox %rbx,%r15 # %rbx is 0, of=0
|
|---|
| 3195 | adcx %rbx,%r15 # cf=0
|
|---|
| 3196 |
|
|---|
| 3197 | .byte 0x67
|
|---|
| 3198 | inc %rcx # of=0
|
|---|
| 3199 | jnz .Lsqrx8x_loop
|
|---|
| 3200 |
|
|---|
| 3201 | lea 8*8($aaptr),$aaptr
|
|---|
| 3202 | mov \$-8,%rcx
|
|---|
| 3203 | cmp 8+8(%rsp),$aaptr # done?
|
|---|
| 3204 | je .Lsqrx8x_break
|
|---|
| 3205 |
|
|---|
| 3206 | sub 16+8(%rsp),%rbx # mov 16(%rsp),%cf
|
|---|
| 3207 | .byte 0x66
|
|---|
| 3208 | mov -64($aptr),%rdx
|
|---|
| 3209 | adcx 0*8($tptr),%r8
|
|---|
| 3210 | adcx 1*8($tptr),%r9
|
|---|
| 3211 | adc 2*8($tptr),%r10
|
|---|
| 3212 | adc 3*8($tptr),%r11
|
|---|
| 3213 | adc 4*8($tptr),%r12
|
|---|
| 3214 | adc 5*8($tptr),%r13
|
|---|
| 3215 | adc 6*8($tptr),%r14
|
|---|
| 3216 | adc 7*8($tptr),%r15
|
|---|
| 3217 | lea 8*8($tptr),$tptr
|
|---|
| 3218 | .byte 0x67
|
|---|
| 3219 | sbb %rax,%rax # mov %cf,%rax
|
|---|
| 3220 | xor %ebx,%ebx # cf=0, of=0
|
|---|
| 3221 | mov %rax,16+8(%rsp) # offload carry
|
|---|
| 3222 | jmp .Lsqrx8x_loop
|
|---|
| 3223 |
|
|---|
| 3224 | .align 32
|
|---|
| 3225 | .Lsqrx8x_break:
|
|---|
| 3226 | xor $zero,$zero
|
|---|
| 3227 | sub 16+8(%rsp),%rbx # mov 16(%rsp),%cf
|
|---|
| 3228 | adcx $zero,%r8
|
|---|
| 3229 | mov 24+8(%rsp),$carry # initial $tptr, borrow $carry
|
|---|
| 3230 | adcx $zero,%r9
|
|---|
| 3231 | mov 0*8($aptr),%rdx # a[8], modulo-scheduled
|
|---|
| 3232 | adc \$0,%r10
|
|---|
| 3233 | mov %r8,0*8($tptr)
|
|---|
| 3234 | adc \$0,%r11
|
|---|
| 3235 | adc \$0,%r12
|
|---|
| 3236 | adc \$0,%r13
|
|---|
| 3237 | adc \$0,%r14
|
|---|
| 3238 | adc \$0,%r15
|
|---|
| 3239 | cmp $carry,$tptr # cf=0, of=0
|
|---|
| 3240 | je .Lsqrx8x_outer_loop
|
|---|
| 3241 |
|
|---|
| 3242 | mov %r9,1*8($tptr)
|
|---|
| 3243 | mov 1*8($carry),%r9
|
|---|
| 3244 | mov %r10,2*8($tptr)
|
|---|
| 3245 | mov 2*8($carry),%r10
|
|---|
| 3246 | mov %r11,3*8($tptr)
|
|---|
| 3247 | mov 3*8($carry),%r11
|
|---|
| 3248 | mov %r12,4*8($tptr)
|
|---|
| 3249 | mov 4*8($carry),%r12
|
|---|
| 3250 | mov %r13,5*8($tptr)
|
|---|
| 3251 | mov 5*8($carry),%r13
|
|---|
| 3252 | mov %r14,6*8($tptr)
|
|---|
| 3253 | mov 6*8($carry),%r14
|
|---|
| 3254 | mov %r15,7*8($tptr)
|
|---|
| 3255 | mov 7*8($carry),%r15
|
|---|
| 3256 | mov $carry,$tptr
|
|---|
| 3257 | jmp .Lsqrx8x_outer_loop
|
|---|
| 3258 |
|
|---|
| 3259 | .align 32
|
|---|
| 3260 | .Lsqrx8x_outer_break:
|
|---|
| 3261 | mov %r9,9*8($tptr) # t[9]
|
|---|
| 3262 | movq %xmm3,%rcx # -$num
|
|---|
| 3263 | mov %r10,10*8($tptr) # ...
|
|---|
| 3264 | mov %r11,11*8($tptr)
|
|---|
| 3265 | mov %r12,12*8($tptr)
|
|---|
| 3266 | mov %r13,13*8($tptr)
|
|---|
| 3267 | mov %r14,14*8($tptr)
|
|---|
| 3268 | ___
|
|---|
| 3269 | } |
|---|
| 3270 | {
|
|---|
| 3271 | my $i="%rcx";
|
|---|
| 3272 | $code.=<<___;
|
|---|
| 3273 | lea 48+8(%rsp),$tptr
|
|---|
| 3274 | mov ($aptr,$i),%rdx # a[0]
|
|---|
| 3275 |
|
|---|
| 3276 | mov 8($tptr),$A0[1] # t[1]
|
|---|
| 3277 | xor $A0[0],$A0[0] # t[0], of=0, cf=0
|
|---|
| 3278 | mov 0+8(%rsp),$num # restore $num
|
|---|
| 3279 | adox $A0[1],$A0[1]
|
|---|
| 3280 | mov 16($tptr),$A1[0] # t[2] # prefetch
|
|---|
| 3281 | mov 24($tptr),$A1[1] # t[3] # prefetch
|
|---|
| 3282 | #jmp .Lsqrx4x_shift_n_add # happens to be aligned
|
|---|
| 3283 |
|
|---|
| 3284 | .align 32
|
|---|
| 3285 | .Lsqrx4x_shift_n_add:
|
|---|
| 3286 | mulx %rdx,%rax,%rbx
|
|---|
| 3287 | adox $A1[0],$A1[0]
|
|---|
| 3288 | adcx $A0[0],%rax
|
|---|
| 3289 | .byte 0x48,0x8b,0x94,0x0e,0x08,0x00,0x00,0x00 # mov 8($aptr,$i),%rdx # a[i+1] # prefetch
|
|---|
| 3290 | .byte 0x4c,0x8b,0x97,0x20,0x00,0x00,0x00 # mov 32($tptr),$A0[0] # t[2*i+4] # prefetch
|
|---|
| 3291 | adox $A1[1],$A1[1]
|
|---|
| 3292 | adcx $A0[1],%rbx
|
|---|
| 3293 | mov 40($tptr),$A0[1] # t[2*i+4+1] # prefetch
|
|---|
| 3294 | mov %rax,0($tptr)
|
|---|
| 3295 | mov %rbx,8($tptr)
|
|---|
| 3296 |
|
|---|
| 3297 | mulx %rdx,%rax,%rbx
|
|---|
| 3298 | adox $A0[0],$A0[0]
|
|---|
| 3299 | adcx $A1[0],%rax
|
|---|
| 3300 | mov 16($aptr,$i),%rdx # a[i+2] # prefetch
|
|---|
| 3301 | mov 48($tptr),$A1[0] # t[2*i+6] # prefetch
|
|---|
| 3302 | adox $A0[1],$A0[1]
|
|---|
| 3303 | adcx $A1[1],%rbx
|
|---|
| 3304 | mov 56($tptr),$A1[1] # t[2*i+6+1] # prefetch
|
|---|
| 3305 | mov %rax,16($tptr)
|
|---|
| 3306 | mov %rbx,24($tptr)
|
|---|
| 3307 |
|
|---|
| 3308 | mulx %rdx,%rax,%rbx
|
|---|
| 3309 | adox $A1[0],$A1[0]
|
|---|
| 3310 | adcx $A0[0],%rax
|
|---|
| 3311 | mov 24($aptr,$i),%rdx # a[i+3] # prefetch
|
|---|
| 3312 | lea 32($i),$i
|
|---|
| 3313 | mov 64($tptr),$A0[0] # t[2*i+8] # prefetch
|
|---|
| 3314 | adox $A1[1],$A1[1]
|
|---|
| 3315 | adcx $A0[1],%rbx
|
|---|
| 3316 | mov 72($tptr),$A0[1] # t[2*i+8+1] # prefetch
|
|---|
| 3317 | mov %rax,32($tptr)
|
|---|
| 3318 | mov %rbx,40($tptr)
|
|---|
| 3319 |
|
|---|
| 3320 | mulx %rdx,%rax,%rbx
|
|---|
| 3321 | adox $A0[0],$A0[0]
|
|---|
| 3322 | adcx $A1[0],%rax
|
|---|
| 3323 | jrcxz .Lsqrx4x_shift_n_add_break
|
|---|
| 3324 | .byte 0x48,0x8b,0x94,0x0e,0x00,0x00,0x00,0x00 # mov 0($aptr,$i),%rdx # a[i+4] # prefetch
|
|---|
| 3325 | adox $A0[1],$A0[1]
|
|---|
| 3326 | adcx $A1[1],%rbx
|
|---|
| 3327 | mov 80($tptr),$A1[0] # t[2*i+10] # prefetch
|
|---|
| 3328 | mov 88($tptr),$A1[1] # t[2*i+10+1] # prefetch
|
|---|
| 3329 | mov %rax,48($tptr)
|
|---|
| 3330 | mov %rbx,56($tptr)
|
|---|
| 3331 | lea 64($tptr),$tptr
|
|---|
| 3332 | nop
|
|---|
| 3333 | jmp .Lsqrx4x_shift_n_add
|
|---|
| 3334 |
|
|---|
| 3335 | .align 32
|
|---|
| 3336 | .Lsqrx4x_shift_n_add_break:
|
|---|
| 3337 | adcx $A1[1],%rbx
|
|---|
| 3338 | mov %rax,48($tptr)
|
|---|
| 3339 | mov %rbx,56($tptr)
|
|---|
| 3340 | lea 64($tptr),$tptr # end of t[] buffer
|
|---|
| 3341 | ___
|
|---|
| 3342 | } |
|---|
| 3343 |
|
|---|
| 3344 | ######################################################################
|
|---|
| 3345 | # Montgomery reduction part, "word-by-word" algorithm.
|
|---|
| 3346 | #
|
|---|
| 3347 | # This new path is inspired by multiple submissions from Intel, by
|
|---|
| 3348 | # Shay Gueron, Vlad Krasnov, Erdinc Ozturk, James Guilford,
|
|---|
| 3349 | # Vinodh Gopal...
|
|---|
| 3350 | {
|
|---|
| 3351 | my ($nptr,$carry,$m0)=("%rbp","%rsi","%rdx");
|
|---|
| 3352 |
|
|---|
| 3353 | $code.=<<___;
|
|---|
| 3354 | movq %xmm2,$nptr
|
|---|
| 3355 | __bn_sqrx8x_reduction:
|
|---|
| 3356 | xor %eax,%eax # initial top-most carry bit
|
|---|
| 3357 | mov 32+8(%rsp),%rbx # n0
|
|---|
| 3358 | mov 48+8(%rsp),%rdx # "%r8", 8*0($tptr)
|
|---|
| 3359 | lea -8*8($nptr,$num),%rcx # end of n[]
|
|---|
| 3360 | #lea 48+8(%rsp,$num,2),$tptr # end of t[] buffer
|
|---|
| 3361 | mov %rcx, 0+8(%rsp) # save end of n[]
|
|---|
| 3362 | mov $tptr,8+8(%rsp) # save end of t[]
|
|---|
| 3363 |
|
|---|
| 3364 | lea 48+8(%rsp),$tptr # initial t[] window
|
|---|
| 3365 | jmp .Lsqrx8x_reduction_loop
|
|---|
| 3366 |
|
|---|
| 3367 | .align 32
|
|---|
| 3368 | .Lsqrx8x_reduction_loop:
|
|---|
| 3369 | mov 8*1($tptr),%r9
|
|---|
| 3370 | mov 8*2($tptr),%r10
|
|---|
| 3371 | mov 8*3($tptr),%r11
|
|---|
| 3372 | mov 8*4($tptr),%r12
|
|---|
| 3373 | mov %rdx,%r8
|
|---|
| 3374 | imulq %rbx,%rdx # n0*a[i]
|
|---|
| 3375 | mov 8*5($tptr),%r13
|
|---|
| 3376 | mov 8*6($tptr),%r14
|
|---|
| 3377 | mov 8*7($tptr),%r15
|
|---|
| 3378 | mov %rax,24+8(%rsp) # store top-most carry bit
|
|---|
| 3379 |
|
|---|
| 3380 | lea 8*8($tptr),$tptr
|
|---|
| 3381 | xor $carry,$carry # cf=0,of=0
|
|---|
| 3382 | mov \$-8,%rcx
|
|---|
| 3383 | jmp .Lsqrx8x_reduce
|
|---|
| 3384 |
|
|---|
| 3385 | .align 32
|
|---|
| 3386 | .Lsqrx8x_reduce:
|
|---|
| 3387 | mov %r8, %rbx
|
|---|
| 3388 | mulx 8*0($nptr),%rax,%r8 # n[0]
|
|---|
| 3389 | adcx %rbx,%rax # discarded
|
|---|
| 3390 | adox %r9,%r8
|
|---|
| 3391 |
|
|---|
| 3392 | mulx 8*1($nptr),%rbx,%r9 # n[1]
|
|---|
| 3393 | adcx %rbx,%r8
|
|---|
| 3394 | adox %r10,%r9
|
|---|
| 3395 |
|
|---|
| 3396 | mulx 8*2($nptr),%rbx,%r10
|
|---|
| 3397 | adcx %rbx,%r9
|
|---|
| 3398 | adox %r11,%r10
|
|---|
| 3399 |
|
|---|
| 3400 | mulx 8*3($nptr),%rbx,%r11
|
|---|
| 3401 | adcx %rbx,%r10
|
|---|
| 3402 | adox %r12,%r11
|
|---|
| 3403 |
|
|---|
| 3404 | .byte 0xc4,0x62,0xe3,0xf6,0xa5,0x20,0x00,0x00,0x00 # mulx 8*4($nptr),%rbx,%r12
|
|---|
| 3405 | mov %rdx,%rax
|
|---|
| 3406 | mov %r8,%rdx
|
|---|
| 3407 | adcx %rbx,%r11
|
|---|
| 3408 | adox %r13,%r12
|
|---|
| 3409 |
|
|---|
| 3410 | mulx 32+8(%rsp),%rbx,%rdx # %rdx discarded
|
|---|
| 3411 | mov %rax,%rdx
|
|---|
| 3412 | mov %rax,64+48+8(%rsp,%rcx,8) # put aside n0*a[i]
|
|---|
| 3413 |
|
|---|
| 3414 | mulx 8*5($nptr),%rax,%r13
|
|---|
| 3415 | adcx %rax,%r12
|
|---|
| 3416 | adox %r14,%r13
|
|---|
| 3417 |
|
|---|
| 3418 | mulx 8*6($nptr),%rax,%r14
|
|---|
| 3419 | adcx %rax,%r13
|
|---|
| 3420 | adox %r15,%r14
|
|---|
| 3421 |
|
|---|
| 3422 | mulx 8*7($nptr),%rax,%r15
|
|---|
| 3423 | mov %rbx,%rdx
|
|---|
| 3424 | adcx %rax,%r14
|
|---|
| 3425 | adox $carry,%r15 # $carry is 0
|
|---|
| 3426 | adcx $carry,%r15 # cf=0
|
|---|
| 3427 |
|
|---|
| 3428 | .byte 0x67,0x67,0x67
|
|---|
| 3429 | inc %rcx # of=0
|
|---|
| 3430 | jnz .Lsqrx8x_reduce
|
|---|
| 3431 |
|
|---|
| 3432 | mov $carry,%rax # xor %rax,%rax
|
|---|
| 3433 | cmp 0+8(%rsp),$nptr # end of n[]?
|
|---|
| 3434 | jae .Lsqrx8x_no_tail
|
|---|
| 3435 |
|
|---|
| 3436 | mov 48+8(%rsp),%rdx # pull n0*a[0]
|
|---|
| 3437 | add 8*0($tptr),%r8
|
|---|
| 3438 | lea 8*8($nptr),$nptr
|
|---|
| 3439 | mov \$-8,%rcx
|
|---|
| 3440 | adcx 8*1($tptr),%r9
|
|---|
| 3441 | adcx 8*2($tptr),%r10
|
|---|
| 3442 | adc 8*3($tptr),%r11
|
|---|
| 3443 | adc 8*4($tptr),%r12
|
|---|
| 3444 | adc 8*5($tptr),%r13
|
|---|
| 3445 | adc 8*6($tptr),%r14
|
|---|
| 3446 | adc 8*7($tptr),%r15
|
|---|
| 3447 | lea 8*8($tptr),$tptr
|
|---|
| 3448 | sbb %rax,%rax # top carry
|
|---|
| 3449 |
|
|---|
| 3450 | xor $carry,$carry # of=0, cf=0
|
|---|
| 3451 | mov %rax,16+8(%rsp)
|
|---|
| 3452 | jmp .Lsqrx8x_tail
|
|---|
| 3453 |
|
|---|
| 3454 | .align 32
|
|---|
| 3455 | .Lsqrx8x_tail:
|
|---|
| 3456 | mov %r8,%rbx
|
|---|
| 3457 | mulx 8*0($nptr),%rax,%r8
|
|---|
| 3458 | adcx %rax,%rbx
|
|---|
| 3459 | adox %r9,%r8
|
|---|
| 3460 |
|
|---|
| 3461 | mulx 8*1($nptr),%rax,%r9
|
|---|
| 3462 | adcx %rax,%r8
|
|---|
| 3463 | adox %r10,%r9
|
|---|
| 3464 |
|
|---|
| 3465 | mulx 8*2($nptr),%rax,%r10
|
|---|
| 3466 | adcx %rax,%r9
|
|---|
| 3467 | adox %r11,%r10
|
|---|
| 3468 |
|
|---|
| 3469 | mulx 8*3($nptr),%rax,%r11
|
|---|
| 3470 | adcx %rax,%r10
|
|---|
| 3471 | adox %r12,%r11
|
|---|
| 3472 |
|
|---|
| 3473 | .byte 0xc4,0x62,0xfb,0xf6,0xa5,0x20,0x00,0x00,0x00 # mulx 8*4($nptr),%rax,%r12
|
|---|
| 3474 | adcx %rax,%r11
|
|---|
| 3475 | adox %r13,%r12
|
|---|
| 3476 |
|
|---|
| 3477 | mulx 8*5($nptr),%rax,%r13
|
|---|
| 3478 | adcx %rax,%r12
|
|---|
| 3479 | adox %r14,%r13
|
|---|
| 3480 |
|
|---|
| 3481 | mulx 8*6($nptr),%rax,%r14
|
|---|
| 3482 | adcx %rax,%r13
|
|---|
| 3483 | adox %r15,%r14
|
|---|
| 3484 |
|
|---|
| 3485 | mulx 8*7($nptr),%rax,%r15
|
|---|
| 3486 | mov 72+48+8(%rsp,%rcx,8),%rdx # pull n0*a[i]
|
|---|
| 3487 | adcx %rax,%r14
|
|---|
| 3488 | adox $carry,%r15
|
|---|
| 3489 | mov %rbx,($tptr,%rcx,8) # save result
|
|---|
| 3490 | mov %r8,%rbx
|
|---|
| 3491 | adcx $carry,%r15 # cf=0
|
|---|
| 3492 |
|
|---|
| 3493 | inc %rcx # of=0
|
|---|
| 3494 | jnz .Lsqrx8x_tail
|
|---|
| 3495 |
|
|---|
| 3496 | cmp 0+8(%rsp),$nptr # end of n[]?
|
|---|
| 3497 | jae .Lsqrx8x_tail_done # break out of loop
|
|---|
| 3498 |
|
|---|
| 3499 | sub 16+8(%rsp),$carry # mov 16(%rsp),%cf
|
|---|
| 3500 | mov 48+8(%rsp),%rdx # pull n0*a[0]
|
|---|
| 3501 | lea 8*8($nptr),$nptr
|
|---|
| 3502 | adc 8*0($tptr),%r8
|
|---|
| 3503 | adc 8*1($tptr),%r9
|
|---|
| 3504 | adc 8*2($tptr),%r10
|
|---|
| 3505 | adc 8*3($tptr),%r11
|
|---|
| 3506 | adc 8*4($tptr),%r12
|
|---|
| 3507 | adc 8*5($tptr),%r13
|
|---|
| 3508 | adc 8*6($tptr),%r14
|
|---|
| 3509 | adc 8*7($tptr),%r15
|
|---|
| 3510 | lea 8*8($tptr),$tptr
|
|---|
| 3511 | sbb %rax,%rax
|
|---|
| 3512 | sub \$8,%rcx # mov \$-8,%rcx
|
|---|
| 3513 |
|
|---|
| 3514 | xor $carry,$carry # of=0, cf=0
|
|---|
| 3515 | mov %rax,16+8(%rsp)
|
|---|
| 3516 | jmp .Lsqrx8x_tail
|
|---|
| 3517 |
|
|---|
| 3518 | .align 32
|
|---|
| 3519 | .Lsqrx8x_tail_done:
|
|---|
| 3520 | xor %rax,%rax
|
|---|
| 3521 | add 24+8(%rsp),%r8 # can this overflow?
|
|---|
| 3522 | adc \$0,%r9
|
|---|
| 3523 | adc \$0,%r10
|
|---|
| 3524 | adc \$0,%r11
|
|---|
| 3525 | adc \$0,%r12
|
|---|
| 3526 | adc \$0,%r13
|
|---|
| 3527 | adc \$0,%r14
|
|---|
| 3528 | adc \$0,%r15
|
|---|
| 3529 | adc \$0,%rax
|
|---|
| 3530 |
|
|---|
| 3531 | sub 16+8(%rsp),$carry # mov 16(%rsp),%cf
|
|---|
| 3532 | .Lsqrx8x_no_tail: # %cf is 0 if jumped here
|
|---|
| 3533 | adc 8*0($tptr),%r8
|
|---|
| 3534 | movq %xmm3,%rcx
|
|---|
| 3535 | adc 8*1($tptr),%r9
|
|---|
| 3536 | mov 8*7($nptr),$carry
|
|---|
| 3537 | movq %xmm2,$nptr # restore $nptr
|
|---|
| 3538 | adc 8*2($tptr),%r10
|
|---|
| 3539 | adc 8*3($tptr),%r11
|
|---|
| 3540 | adc 8*4($tptr),%r12
|
|---|
| 3541 | adc 8*5($tptr),%r13
|
|---|
| 3542 | adc 8*6($tptr),%r14
|
|---|
| 3543 | adc 8*7($tptr),%r15
|
|---|
| 3544 | adc \$0,%rax # top-most carry
|
|---|
| 3545 |
|
|---|
| 3546 | mov 32+8(%rsp),%rbx # n0
|
|---|
| 3547 | mov 8*8($tptr,%rcx),%rdx # modulo-scheduled "%r8"
|
|---|
| 3548 |
|
|---|
| 3549 | mov %r8,8*0($tptr) # store top 512 bits
|
|---|
| 3550 | lea 8*8($tptr),%r8 # borrow %r8
|
|---|
| 3551 | mov %r9,8*1($tptr)
|
|---|
| 3552 | mov %r10,8*2($tptr)
|
|---|
| 3553 | mov %r11,8*3($tptr)
|
|---|
| 3554 | mov %r12,8*4($tptr)
|
|---|
| 3555 | mov %r13,8*5($tptr)
|
|---|
| 3556 | mov %r14,8*6($tptr)
|
|---|
| 3557 | mov %r15,8*7($tptr)
|
|---|
| 3558 |
|
|---|
| 3559 | lea 8*8($tptr,%rcx),$tptr # start of current t[] window
|
|---|
| 3560 | cmp 8+8(%rsp),%r8 # end of t[]?
|
|---|
| 3561 | jb .Lsqrx8x_reduction_loop
|
|---|
| 3562 | ret
|
|---|
| 3563 | .cfi_endproc
|
|---|
| 3564 | .size bn_sqrx8x_internal,.-bn_sqrx8x_internal
|
|---|
| 3565 | ___
|
|---|
| 3566 | } |
|---|
| 3567 |
|
|---|
| 3568 | ##############################################################
|
|---|
| 3569 | # Post-condition, 4x unrolled
|
|---|
| 3570 | #
|
|---|
| 3571 | {
|
|---|
| 3572 | my ($rptr,$nptr)=("%rdx","%rbp");
|
|---|
| 3573 | $code.=<<___;
|
|---|
| 3574 | .align 32
|
|---|
| 3575 | __bn_postx4x_internal:
|
|---|
| 3576 | .cfi_startproc
|
|---|
| 3577 | mov 8*0($nptr),%r12
|
|---|
| 3578 | mov %rcx,%r10 # -$num
|
|---|
| 3579 | mov %rcx,%r9 # -$num
|
|---|
| 3580 | neg %rax
|
|---|
| 3581 | sar \$3+2,%rcx
|
|---|
| 3582 | #lea 48+8(%rsp,%r9),$tptr
|
|---|
| 3583 | movq %xmm1,$rptr # restore $rptr
|
|---|
| 3584 | movq %xmm1,$aptr # prepare for back-to-back call
|
|---|
| 3585 | dec %r12 # so that after 'not' we get -n[0]
|
|---|
| 3586 | mov 8*1($nptr),%r13
|
|---|
| 3587 | xor %r8,%r8
|
|---|
| 3588 | mov 8*2($nptr),%r14
|
|---|
| 3589 | mov 8*3($nptr),%r15
|
|---|
| 3590 | jmp .Lsqrx4x_sub_entry
|
|---|
| 3591 |
|
|---|
| 3592 | .align 16
|
|---|
| 3593 | .Lsqrx4x_sub:
|
|---|
| 3594 | mov 8*0($nptr),%r12
|
|---|
| 3595 | mov 8*1($nptr),%r13
|
|---|
| 3596 | mov 8*2($nptr),%r14
|
|---|
| 3597 | mov 8*3($nptr),%r15
|
|---|
| 3598 | .Lsqrx4x_sub_entry:
|
|---|
| 3599 | andn %rax,%r12,%r12
|
|---|
| 3600 | lea 8*4($nptr),$nptr
|
|---|
| 3601 | andn %rax,%r13,%r13
|
|---|
| 3602 | andn %rax,%r14,%r14
|
|---|
| 3603 | andn %rax,%r15,%r15
|
|---|
| 3604 |
|
|---|
| 3605 | neg %r8 # mov %r8,%cf
|
|---|
| 3606 | adc 8*0($tptr),%r12
|
|---|
| 3607 | adc 8*1($tptr),%r13
|
|---|
| 3608 | adc 8*2($tptr),%r14
|
|---|
| 3609 | adc 8*3($tptr),%r15
|
|---|
| 3610 | mov %r12,8*0($rptr)
|
|---|
| 3611 | lea 8*4($tptr),$tptr
|
|---|
| 3612 | mov %r13,8*1($rptr)
|
|---|
| 3613 | sbb %r8,%r8 # mov %cf,%r8
|
|---|
| 3614 | mov %r14,8*2($rptr)
|
|---|
| 3615 | mov %r15,8*3($rptr)
|
|---|
| 3616 | lea 8*4($rptr),$rptr
|
|---|
| 3617 |
|
|---|
| 3618 | inc %rcx
|
|---|
| 3619 | jnz .Lsqrx4x_sub
|
|---|
| 3620 |
|
|---|
| 3621 | neg %r9 # restore $num
|
|---|
| 3622 |
|
|---|
| 3623 | ret
|
|---|
| 3624 | .cfi_endproc
|
|---|
| 3625 | .size __bn_postx4x_internal,.-__bn_postx4x_internal
|
|---|
| 3626 | ___
|
|---|
| 3627 | }
|
|---|
| 3628 | }}}
|
|---|
| 3629 | {
|
|---|
| 3630 | my ($inp,$num,$tbl,$idx)=$win64?("%rcx","%edx","%r8", "%r9d") : # Win64 order
|
|---|
| 3631 | ("%rdi","%esi","%rdx","%ecx"); # Unix order
|
|---|
| 3632 | my $out=$inp;
|
|---|
| 3633 | my $STRIDE=2**5*8;
|
|---|
| 3634 | my $N=$STRIDE/4;
|
|---|
| 3635 |
|
|---|
| 3636 | $code.=<<___;
|
|---|
| 3637 | .globl bn_get_bits5
|
|---|
| 3638 | .type bn_get_bits5,\@abi-omnipotent
|
|---|
| 3639 | .align 16
|
|---|
| 3640 | bn_get_bits5:
|
|---|
| 3641 | .cfi_startproc
|
|---|
| 3642 | lea 0($inp),%r10
|
|---|
| 3643 | lea 1($inp),%r11
|
|---|
| 3644 | mov $num,%ecx
|
|---|
| 3645 | shr \$4,$num
|
|---|
| 3646 | and \$15,%ecx
|
|---|
| 3647 | lea -8(%ecx),%eax
|
|---|
| 3648 | cmp \$11,%ecx
|
|---|
| 3649 | cmova %r11,%r10
|
|---|
| 3650 | cmova %eax,%ecx
|
|---|
| 3651 | movzw (%r10,$num,2),%eax
|
|---|
| 3652 | shrl %cl,%eax
|
|---|
| 3653 | and \$31,%eax
|
|---|
| 3654 | ret
|
|---|
| 3655 | .cfi_endproc
|
|---|
| 3656 | .size bn_get_bits5,.-bn_get_bits5
|
|---|
| 3657 |
|
|---|
| 3658 | .globl bn_scatter5
|
|---|
| 3659 | .type bn_scatter5,\@abi-omnipotent
|
|---|
| 3660 | .align 16
|
|---|
| 3661 | bn_scatter5:
|
|---|
| 3662 | .cfi_startproc
|
|---|
| 3663 | cmp \$0, $num
|
|---|
| 3664 | jz .Lscatter_epilogue
|
|---|
| 3665 | lea ($tbl,$idx,8),$tbl
|
|---|
| 3666 | .Lscatter:
|
|---|
| 3667 | mov ($inp),%rax
|
|---|
| 3668 | lea 8($inp),$inp
|
|---|
| 3669 | mov %rax,($tbl)
|
|---|
| 3670 | lea 32*8($tbl),$tbl
|
|---|
| 3671 | sub \$1,$num
|
|---|
| 3672 | jnz .Lscatter
|
|---|
| 3673 | .Lscatter_epilogue:
|
|---|
| 3674 | ret
|
|---|
| 3675 | .cfi_endproc
|
|---|
| 3676 | .size bn_scatter5,.-bn_scatter5
|
|---|
| 3677 |
|
|---|
| 3678 | .globl bn_gather5
|
|---|
| 3679 | .type bn_gather5,\@abi-omnipotent
|
|---|
| 3680 | .align 32
|
|---|
| 3681 | bn_gather5:
|
|---|
| 3682 | .LSEH_begin_bn_gather5: # Win64 thing, but harmless in other cases
|
|---|
| 3683 | .cfi_startproc
|
|---|
| 3684 | # I can't trust assembler to use specific encoding:-(
|
|---|
| 3685 | .byte 0x4c,0x8d,0x14,0x24 #lea (%rsp),%r10
|
|---|
| 3686 | .byte 0x48,0x81,0xec,0x08,0x01,0x00,0x00 #sub $0x108,%rsp
|
|---|
| 3687 | lea .Linc(%rip),%rax
|
|---|
| 3688 | and \$-16,%rsp # shouldn't be formally required
|
|---|
| 3689 |
|
|---|
| 3690 | movd $idx,%xmm5
|
|---|
| 3691 | movdqa 0(%rax),%xmm0 # 00000001000000010000000000000000
|
|---|
| 3692 | movdqa 16(%rax),%xmm1 # 00000002000000020000000200000002
|
|---|
| 3693 | lea 128($tbl),%r11 # size optimization
|
|---|
| 3694 | lea 128(%rsp),%rax # size optimization
|
|---|
| 3695 |
|
|---|
| 3696 | pshufd \$0,%xmm5,%xmm5 # broadcast $idx
|
|---|
| 3697 | movdqa %xmm1,%xmm4
|
|---|
| 3698 | movdqa %xmm1,%xmm2
|
|---|
| 3699 | ___
|
|---|
| 3700 | ########################################################################
|
|---|
| 3701 | # calculate mask by comparing 0..31 to $idx and save result to stack
|
|---|
| 3702 | #
|
|---|
| 3703 | for($i=0;$i<$STRIDE/16;$i+=4) {
|
|---|
| 3704 | $code.=<<___;
|
|---|
| 3705 | paddd %xmm0,%xmm1
|
|---|
| 3706 | pcmpeqd %xmm5,%xmm0 # compare to 1,0
|
|---|
| 3707 | ___
|
|---|
| 3708 | $code.=<<___ if ($i);
|
|---|
| 3709 | movdqa %xmm3,`16*($i-1)-128`(%rax)
|
|---|
| 3710 | ___
|
|---|
| 3711 | $code.=<<___;
|
|---|
| 3712 | movdqa %xmm4,%xmm3
|
|---|
| 3713 |
|
|---|
| 3714 | paddd %xmm1,%xmm2
|
|---|
| 3715 | pcmpeqd %xmm5,%xmm1 # compare to 3,2
|
|---|
| 3716 | movdqa %xmm0,`16*($i+0)-128`(%rax)
|
|---|
| 3717 | movdqa %xmm4,%xmm0
|
|---|
| 3718 |
|
|---|
| 3719 | paddd %xmm2,%xmm3
|
|---|
| 3720 | pcmpeqd %xmm5,%xmm2 # compare to 5,4
|
|---|
| 3721 | movdqa %xmm1,`16*($i+1)-128`(%rax)
|
|---|
| 3722 | movdqa %xmm4,%xmm1
|
|---|
| 3723 |
|
|---|
| 3724 | paddd %xmm3,%xmm0
|
|---|
| 3725 | pcmpeqd %xmm5,%xmm3 # compare to 7,6
|
|---|
| 3726 | movdqa %xmm2,`16*($i+2)-128`(%rax)
|
|---|
| 3727 | movdqa %xmm4,%xmm2
|
|---|
| 3728 | ___
|
|---|
| 3729 | }
|
|---|
| 3730 | $code.=<<___;
|
|---|
| 3731 | movdqa %xmm3,`16*($i-1)-128`(%rax)
|
|---|
| 3732 | jmp .Lgather
|
|---|
| 3733 |
|
|---|
| 3734 | .align 32
|
|---|
| 3735 | .Lgather:
|
|---|
| 3736 | pxor %xmm4,%xmm4
|
|---|
| 3737 | pxor %xmm5,%xmm5
|
|---|
| 3738 | ___
|
|---|
| 3739 | for($i=0;$i<$STRIDE/16;$i+=4) {
|
|---|
| 3740 | $code.=<<___;
|
|---|
| 3741 | movdqa `16*($i+0)-128`(%r11),%xmm0
|
|---|
| 3742 | movdqa `16*($i+1)-128`(%r11),%xmm1
|
|---|
| 3743 | movdqa `16*($i+2)-128`(%r11),%xmm2
|
|---|
| 3744 | pand `16*($i+0)-128`(%rax),%xmm0
|
|---|
| 3745 | movdqa `16*($i+3)-128`(%r11),%xmm3
|
|---|
| 3746 | pand `16*($i+1)-128`(%rax),%xmm1
|
|---|
| 3747 | por %xmm0,%xmm4
|
|---|
| 3748 | pand `16*($i+2)-128`(%rax),%xmm2
|
|---|
| 3749 | por %xmm1,%xmm5
|
|---|
| 3750 | pand `16*($i+3)-128`(%rax),%xmm3
|
|---|
| 3751 | por %xmm2,%xmm4
|
|---|
| 3752 | por %xmm3,%xmm5
|
|---|
| 3753 | ___
|
|---|
| 3754 | }
|
|---|
| 3755 | $code.=<<___;
|
|---|
| 3756 | por %xmm5,%xmm4
|
|---|
| 3757 | lea $STRIDE(%r11),%r11
|
|---|
| 3758 | pshufd \$0x4e,%xmm4,%xmm0
|
|---|
| 3759 | por %xmm4,%xmm0
|
|---|
| 3760 | movq %xmm0,($out) # m0=bp[0]
|
|---|
| 3761 | lea 8($out),$out
|
|---|
| 3762 | sub \$1,$num
|
|---|
| 3763 | jnz .Lgather
|
|---|
| 3764 |
|
|---|
| 3765 | lea (%r10),%rsp
|
|---|
| 3766 | ret
|
|---|
| 3767 | .LSEH_end_bn_gather5:
|
|---|
| 3768 | .cfi_endproc
|
|---|
| 3769 | .size bn_gather5,.-bn_gather5
|
|---|
| 3770 | ___
|
|---|
| 3771 | }
|
|---|
| 3772 | $code.=<<___;
|
|---|
| 3773 | .align 64
|
|---|
| 3774 | .Linc:
|
|---|
| 3775 | .long 0,0, 1,1
|
|---|
| 3776 | .long 2,2, 2,2
|
|---|
| 3777 | .asciz "Montgomery Multiplication with scatter/gather for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
|
|---|
| 3778 | ___
|
|---|
| 3779 |
|
|---|
| 3780 | # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
|
|---|
| 3781 | # CONTEXT *context,DISPATCHER_CONTEXT *disp)
|
|---|
| 3782 | if ($win64) {
|
|---|
| 3783 | $rec="%rcx";
|
|---|
| 3784 | $frame="%rdx";
|
|---|
| 3785 | $context="%r8";
|
|---|
| 3786 | $disp="%r9";
|
|---|
| 3787 |
|
|---|
| 3788 | $code.=<<___;
|
|---|
| 3789 | .extern __imp_RtlVirtualUnwind
|
|---|
| 3790 | .type mul_handler,\@abi-omnipotent
|
|---|
| 3791 | .align 16
|
|---|
| 3792 | mul_handler:
|
|---|
| 3793 | push %rsi
|
|---|
| 3794 | push %rdi
|
|---|
| 3795 | push %rbx
|
|---|
| 3796 | push %rbp
|
|---|
| 3797 | push %r12
|
|---|
| 3798 | push %r13
|
|---|
| 3799 | push %r14
|
|---|
| 3800 | push %r15
|
|---|
| 3801 | pushfq
|
|---|
| 3802 | sub \$64,%rsp
|
|---|
| 3803 |
|
|---|
| 3804 | mov 120($context),%rax # pull context->Rax
|
|---|
| 3805 | mov 248($context),%rbx # pull context->Rip
|
|---|
| 3806 |
|
|---|
| 3807 | mov 8($disp),%rsi # disp->ImageBase
|
|---|
| 3808 | mov 56($disp),%r11 # disp->HandlerData
|
|---|
| 3809 |
|
|---|
| 3810 | mov 0(%r11),%r10d # HandlerData[0]
|
|---|
| 3811 | lea (%rsi,%r10),%r10 # end of prologue label
|
|---|
| 3812 | cmp %r10,%rbx # context->Rip<end of prologue label
|
|---|
| 3813 | jb .Lcommon_seh_tail
|
|---|
| 3814 |
|
|---|
| 3815 | mov 4(%r11),%r10d # HandlerData[1]
|
|---|
| 3816 | lea (%rsi,%r10),%r10 # beginning of body label
|
|---|
| 3817 | cmp %r10,%rbx # context->Rip<body label
|
|---|
| 3818 | jb .Lcommon_pop_regs
|
|---|
| 3819 |
|
|---|
| 3820 | mov 152($context),%rax # pull context->Rsp
|
|---|
| 3821 |
|
|---|
| 3822 | mov 8(%r11),%r10d # HandlerData[2]
|
|---|
| 3823 | lea (%rsi,%r10),%r10 # epilogue label
|
|---|
| 3824 | cmp %r10,%rbx # context->Rip>=epilogue label
|
|---|
| 3825 | jae .Lcommon_seh_tail
|
|---|
| 3826 |
|
|---|
| 3827 | lea .Lmul_epilogue(%rip),%r10
|
|---|
| 3828 | cmp %r10,%rbx
|
|---|
| 3829 | ja .Lbody_40
|
|---|
| 3830 |
|
|---|
| 3831 | mov 192($context),%r10 # pull $num
|
|---|
| 3832 | mov 8(%rax,%r10,8),%rax # pull saved stack pointer
|
|---|
| 3833 |
|
|---|
| 3834 | jmp .Lcommon_pop_regs
|
|---|
| 3835 |
|
|---|
| 3836 | .Lbody_40:
|
|---|
| 3837 | mov 40(%rax),%rax # pull saved stack pointer
|
|---|
| 3838 | .Lcommon_pop_regs:
|
|---|
| 3839 | mov -8(%rax),%rbx
|
|---|
| 3840 | mov -16(%rax),%rbp
|
|---|
| 3841 | mov -24(%rax),%r12
|
|---|
| 3842 | mov -32(%rax),%r13
|
|---|
| 3843 | mov -40(%rax),%r14
|
|---|
| 3844 | mov -48(%rax),%r15
|
|---|
| 3845 | mov %rbx,144($context) # restore context->Rbx
|
|---|
| 3846 | mov %rbp,160($context) # restore context->Rbp
|
|---|
| 3847 | mov %r12,216($context) # restore context->R12
|
|---|
| 3848 | mov %r13,224($context) # restore context->R13
|
|---|
| 3849 | mov %r14,232($context) # restore context->R14
|
|---|
| 3850 | mov %r15,240($context) # restore context->R15
|
|---|
| 3851 |
|
|---|
| 3852 | .Lcommon_seh_tail:
|
|---|
| 3853 | mov 8(%rax),%rdi
|
|---|
| 3854 | mov 16(%rax),%rsi
|
|---|
| 3855 | mov %rax,152($context) # restore context->Rsp
|
|---|
| 3856 | mov %rsi,168($context) # restore context->Rsi
|
|---|
| 3857 | mov %rdi,176($context) # restore context->Rdi
|
|---|
| 3858 |
|
|---|
| 3859 | mov 40($disp),%rdi # disp->ContextRecord
|
|---|
| 3860 | mov $context,%rsi # context
|
|---|
| 3861 | mov \$154,%ecx # sizeof(CONTEXT)
|
|---|
| 3862 | .long 0xa548f3fc # cld; rep movsq
|
|---|
| 3863 |
|
|---|
| 3864 | mov $disp,%rsi
|
|---|
| 3865 | xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
|
|---|
| 3866 | mov 8(%rsi),%rdx # arg2, disp->ImageBase
|
|---|
| 3867 | mov 0(%rsi),%r8 # arg3, disp->ControlPc
|
|---|
| 3868 | mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
|
|---|
| 3869 | mov 40(%rsi),%r10 # disp->ContextRecord
|
|---|
| 3870 | lea 56(%rsi),%r11 # &disp->HandlerData
|
|---|
| 3871 | lea 24(%rsi),%r12 # &disp->EstablisherFrame
|
|---|
| 3872 | mov %r10,32(%rsp) # arg5
|
|---|
| 3873 | mov %r11,40(%rsp) # arg6
|
|---|
| 3874 | mov %r12,48(%rsp) # arg7
|
|---|
| 3875 | mov %rcx,56(%rsp) # arg8, (NULL)
|
|---|
| 3876 | call *__imp_RtlVirtualUnwind(%rip)
|
|---|
| 3877 |
|
|---|
| 3878 | mov \$1,%eax # ExceptionContinueSearch
|
|---|
| 3879 | add \$64,%rsp
|
|---|
| 3880 | popfq
|
|---|
| 3881 | pop %r15
|
|---|
| 3882 | pop %r14
|
|---|
| 3883 | pop %r13
|
|---|
| 3884 | pop %r12
|
|---|
| 3885 | pop %rbp
|
|---|
| 3886 | pop %rbx
|
|---|
| 3887 | pop %rdi
|
|---|
| 3888 | pop %rsi
|
|---|
| 3889 | ret
|
|---|
| 3890 | .size mul_handler,.-mul_handler
|
|---|
| 3891 |
|
|---|
| 3892 | .section .pdata
|
|---|
| 3893 | .align 4
|
|---|
| 3894 | .rva .LSEH_begin_bn_mul_mont_gather5
|
|---|
| 3895 | .rva .LSEH_end_bn_mul_mont_gather5
|
|---|
| 3896 | .rva .LSEH_info_bn_mul_mont_gather5
|
|---|
| 3897 |
|
|---|
| 3898 | .rva .LSEH_begin_bn_mul4x_mont_gather5
|
|---|
| 3899 | .rva .LSEH_end_bn_mul4x_mont_gather5
|
|---|
| 3900 | .rva .LSEH_info_bn_mul4x_mont_gather5
|
|---|
| 3901 |
|
|---|
| 3902 | .rva .LSEH_begin_bn_power5
|
|---|
| 3903 | .rva .LSEH_end_bn_power5
|
|---|
| 3904 | .rva .LSEH_info_bn_power5
|
|---|
| 3905 |
|
|---|
| 3906 | .rva .LSEH_begin_bn_from_mont8x
|
|---|
| 3907 | .rva .LSEH_end_bn_from_mont8x
|
|---|
| 3908 | .rva .LSEH_info_bn_from_mont8x
|
|---|
| 3909 | ___
|
|---|
| 3910 | $code.=<<___ if ($addx);
|
|---|
| 3911 | .rva .LSEH_begin_bn_mulx4x_mont_gather5
|
|---|
| 3912 | .rva .LSEH_end_bn_mulx4x_mont_gather5
|
|---|
| 3913 | .rva .LSEH_info_bn_mulx4x_mont_gather5
|
|---|
| 3914 |
|
|---|
| 3915 | .rva .LSEH_begin_bn_powerx5
|
|---|
| 3916 | .rva .LSEH_end_bn_powerx5
|
|---|
| 3917 | .rva .LSEH_info_bn_powerx5
|
|---|
| 3918 | ___
|
|---|
| 3919 | $code.=<<___;
|
|---|
| 3920 | .rva .LSEH_begin_bn_gather5
|
|---|
| 3921 | .rva .LSEH_end_bn_gather5
|
|---|
| 3922 | .rva .LSEH_info_bn_gather5
|
|---|
| 3923 |
|
|---|
| 3924 | .section .xdata
|
|---|
| 3925 | .align 8
|
|---|
| 3926 | .LSEH_info_bn_mul_mont_gather5:
|
|---|
| 3927 | .byte 9,0,0,0
|
|---|
| 3928 | .rva mul_handler
|
|---|
| 3929 | .rva .Lmul_body,.Lmul_body,.Lmul_epilogue # HandlerData[]
|
|---|
| 3930 | .align 8
|
|---|
| 3931 | .LSEH_info_bn_mul4x_mont_gather5:
|
|---|
| 3932 | .byte 9,0,0,0
|
|---|
| 3933 | .rva mul_handler
|
|---|
| 3934 | .rva .Lmul4x_prologue,.Lmul4x_body,.Lmul4x_epilogue # HandlerData[]
|
|---|
| 3935 | .align 8
|
|---|
| 3936 | .LSEH_info_bn_power5:
|
|---|
| 3937 | .byte 9,0,0,0
|
|---|
| 3938 | .rva mul_handler
|
|---|
| 3939 | .rva .Lpower5_prologue,.Lpower5_body,.Lpower5_epilogue # HandlerData[]
|
|---|
| 3940 | .align 8
|
|---|
| 3941 | .LSEH_info_bn_from_mont8x:
|
|---|
| 3942 | .byte 9,0,0,0
|
|---|
| 3943 | .rva mul_handler
|
|---|
| 3944 | .rva .Lfrom_prologue,.Lfrom_body,.Lfrom_epilogue # HandlerData[]
|
|---|
| 3945 | ___
|
|---|
| 3946 | $code.=<<___ if ($addx);
|
|---|
| 3947 | .align 8
|
|---|
| 3948 | .LSEH_info_bn_mulx4x_mont_gather5:
|
|---|
| 3949 | .byte 9,0,0,0
|
|---|
| 3950 | .rva mul_handler
|
|---|
| 3951 | .rva .Lmulx4x_prologue,.Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[]
|
|---|
| 3952 | .align 8
|
|---|
| 3953 | .LSEH_info_bn_powerx5:
|
|---|
| 3954 | .byte 9,0,0,0
|
|---|
| 3955 | .rva mul_handler
|
|---|
| 3956 | .rva .Lpowerx5_prologue,.Lpowerx5_body,.Lpowerx5_epilogue # HandlerData[]
|
|---|
| 3957 | ___
|
|---|
| 3958 | $code.=<<___;
|
|---|
| 3959 | .align 8
|
|---|
| 3960 | .LSEH_info_bn_gather5:
|
|---|
| 3961 | .byte 0x01,0x0b,0x03,0x0a
|
|---|
| 3962 | .byte 0x0b,0x01,0x21,0x00 # sub rsp,0x108
|
|---|
| 3963 | .byte 0x04,0xa3,0x00,0x00 # lea r10,(rsp)
|
|---|
| 3964 | .align 8
|
|---|
| 3965 | ___
|
|---|
| 3966 | }
|
|---|
| 3967 |
|
|---|
| 3968 | $code =~ s/\`([^\`]*)\`/eval($1)/gem;
|
|---|
| 3969 |
|
|---|
| 3970 | print $code;
|
|---|
| 3971 | close STDOUT or die "error closing STDOUT: $!";
|
|---|
