| 1 | #! /usr/bin/env perl
|
|---|
| 2 | # Copyright 2011-2022 The OpenSSL Project Authors. All Rights Reserved.
|
|---|
| 3 | #
|
|---|
| 4 | # Licensed under the Apache License 2.0 (the "License"). You may not use
|
|---|
| 5 | # this file except in compliance with the License. You can obtain a copy
|
|---|
| 6 | # in the file LICENSE in the source distribution or at
|
|---|
| 7 | # https://www.openssl.org/source/license.html
|
|---|
| 8 |
|
|---|
| 9 |
|
|---|
| 10 | # ====================================================================
|
|---|
| 11 | # Written by Andy Polyakov <[email protected]> for the OpenSSL
|
|---|
| 12 | # project. The module is, however, dual licensed under OpenSSL and
|
|---|
| 13 | # CRYPTOGAMS licenses depending on where you obtain it. For further
|
|---|
| 14 | # details see http://www.openssl.org/~appro/cryptogams/.
|
|---|
| 15 | # ====================================================================
|
|---|
| 16 |
|
|---|
| 17 | # 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 | # $output is the last argument if it looks like a file (it has an extension)
|
|---|
| 35 | # $flavour is the first argument if it doesn't look like a file
|
|---|
| 36 | $output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
|
|---|
| 37 | $flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
|
|---|
| 38 |
|
|---|
| 39 | $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
|
|---|
| 40 |
|
|---|
| 41 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
|
|---|
| 42 | ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
|
|---|
| 43 | ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
|
|---|
| 44 | die "can't locate x86_64-xlate.pl";
|
|---|
| 45 |
|
|---|
| 46 | open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""
|
|---|
| 47 | or die "can't call $xlate: $!";
|
|---|
| 48 | *STDOUT=*OUT;
|
|---|
| 49 |
|
|---|
| 50 | if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
|
|---|
| 51 | =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
|
|---|
| 52 | $addx = ($1>=2.23);
|
|---|
| 53 | }
|
|---|
| 54 |
|
|---|
| 55 | if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
|
|---|
| 56 | `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
|
|---|
| 57 | $addx = ($1>=2.10);
|
|---|
| 58 | }
|
|---|
| 59 |
|
|---|
| 60 | if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
|
|---|
| 61 | `ml64 2>&1` =~ /Version ([0-9]+)\./) {
|
|---|
| 62 | $addx = ($1>=12);
|
|---|
| 63 | }
|
|---|
| 64 |
|
|---|
| 65 | if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:clang|LLVM) version|.*based on LLVM) ([0-9]+)\.([0-9]+)/) {
|
|---|
| 66 | my $ver = $2 + $3/100.0; # 3.1->3.01, 3.10->3.10
|
|---|
| 67 | $addx = ($ver>=3.03);
|
|---|
| 68 | }
|
|---|
| 69 |
|
|---|
| 70 | # int bn_mul_mont_gather5(
|
|---|
| 71 | $rp="%rdi"; # BN_ULONG *rp,
|
|---|
| 72 | $ap="%rsi"; # const BN_ULONG *ap,
|
|---|
| 73 | $bp="%rdx"; # const BN_ULONG *bp,
|
|---|
| 74 | $np="%rcx"; # const BN_ULONG *np,
|
|---|
| 75 | $n0="%r8"; # const BN_ULONG *n0,
|
|---|
| 76 | $num="%r9"; # int num,
|
|---|
| 77 | # int idx); # 0 to 2^5-1, "index" in $bp holding
|
|---|
| 78 | # pre-computed powers of a', interlaced
|
|---|
| 79 | # in such manner that b[0] is $bp[idx],
|
|---|
| 80 | # b[1] is [2^5+idx], etc.
|
|---|
| 81 | $lo0="%r10";
|
|---|
| 82 | $hi0="%r11";
|
|---|
| 83 | $hi1="%r13";
|
|---|
| 84 | $i="%r14";
|
|---|
| 85 | $j="%r15";
|
|---|
| 86 | $m0="%rbx";
|
|---|
| 87 | $m1="%rbp";
|
|---|
| 88 |
|
|---|
| 89 | $code=<<___;
|
|---|
| 90 | .text
|
|---|
| 91 |
|
|---|
| 92 | .extern OPENSSL_ia32cap_P
|
|---|
| 93 |
|
|---|
| 94 | .globl bn_mul_mont_gather5
|
|---|
| 95 | .type bn_mul_mont_gather5,\@function,6
|
|---|
| 96 | .align 64
|
|---|
| 97 | bn_mul_mont_gather5:
|
|---|
| 98 | .cfi_startproc
|
|---|
| 99 | mov ${num}d,${num}d
|
|---|
| 100 | mov %rsp,%rax
|
|---|
| 101 | .cfi_def_cfa_register %rax
|
|---|
| 102 | test \$7,${num}d
|
|---|
| 103 | jnz .Lmul_enter
|
|---|
| 104 | ___
|
|---|
| 105 | $code.=<<___ if ($addx);
|
|---|
| 106 | mov OPENSSL_ia32cap_P+8(%rip),%r11d
|
|---|
| 107 | ___
|
|---|
| 108 | $code.=<<___;
|
|---|
| 109 | jmp .Lmul4x_enter
|
|---|
| 110 |
|
|---|
| 111 | .align 16
|
|---|
| 112 | .Lmul_enter:
|
|---|
| 113 | movd `($win64?56:8)`(%rsp),%xmm5 # load 7th argument
|
|---|
| 114 | push %rbx
|
|---|
| 115 | .cfi_push %rbx
|
|---|
| 116 | push %rbp
|
|---|
| 117 | .cfi_push %rbp
|
|---|
| 118 | push %r12
|
|---|
| 119 | .cfi_push %r12
|
|---|
| 120 | push %r13
|
|---|
| 121 | .cfi_push %r13
|
|---|
| 122 | push %r14
|
|---|
| 123 | .cfi_push %r14
|
|---|
| 124 | push %r15
|
|---|
| 125 | .cfi_push %r15
|
|---|
| 126 |
|
|---|
| 127 | neg $num
|
|---|
| 128 | mov %rsp,%r11
|
|---|
| 129 | lea -280(%rsp,$num,8),%r10 # future alloca(8*(num+2)+256+8)
|
|---|
| 130 | neg $num # restore $num
|
|---|
| 131 | and \$-1024,%r10 # minimize TLB usage
|
|---|
| 132 |
|
|---|
| 133 | # An OS-agnostic version of __chkstk.
|
|---|
| 134 | #
|
|---|
| 135 | # Some OSes (Windows) insist on stack being "wired" to
|
|---|
| 136 | # physical memory in strictly sequential manner, i.e. if stack
|
|---|
| 137 | # allocation spans two pages, then reference to farmost one can
|
|---|
| 138 | # be punishable by SEGV. But page walking can do good even on
|
|---|
| 139 | # other OSes, because it guarantees that villain thread hits
|
|---|
| 140 | # the guard page before it can make damage to innocent one...
|
|---|
| 141 | sub %r10,%r11
|
|---|
| 142 | and \$-4096,%r11
|
|---|
| 143 | lea (%r10,%r11),%rsp
|
|---|
| 144 | mov (%rsp),%r11
|
|---|
| 145 | cmp %r10,%rsp
|
|---|
| 146 | ja .Lmul_page_walk
|
|---|
| 147 | jmp .Lmul_page_walk_done
|
|---|
| 148 |
|
|---|
| 149 | .Lmul_page_walk:
|
|---|
| 150 | lea -4096(%rsp),%rsp
|
|---|
| 151 | mov (%rsp),%r11
|
|---|
| 152 | cmp %r10,%rsp
|
|---|
| 153 | ja .Lmul_page_walk
|
|---|
| 154 | .Lmul_page_walk_done:
|
|---|
| 155 |
|
|---|
| 156 | lea .Linc(%rip),%r10
|
|---|
| 157 | mov %rax,8(%rsp,$num,8) # tp[num+1]=%rsp
|
|---|
| 158 | .cfi_cfa_expression %rsp+8,$num,8,mul,plus,deref,+8
|
|---|
| 159 | .Lmul_body:
|
|---|
| 160 |
|
|---|
| 161 | lea 128($bp),%r12 # reassign $bp (+size optimization)
|
|---|
| 162 | ___
|
|---|
| 163 | $bp="%r12";
|
|---|
| 164 | $STRIDE=2**5*8; # 5 is "window size"
|
|---|
| 165 | $N=$STRIDE/4; # should match cache line size
|
|---|
| 166 | $code.=<<___;
|
|---|
| 167 | movdqa 0(%r10),%xmm0 # 00000001000000010000000000000000
|
|---|
| 168 | movdqa 16(%r10),%xmm1 # 00000002000000020000000200000002
|
|---|
| 169 | lea 24-112(%rsp,$num,8),%r10# place the mask after tp[num+3] (+ICache optimization)
|
|---|
| 170 | and \$-16,%r10
|
|---|
| 171 |
|
|---|
| 172 | pshufd \$0,%xmm5,%xmm5 # broadcast index
|
|---|
| 173 | movdqa %xmm1,%xmm4
|
|---|
| 174 | movdqa %xmm1,%xmm2
|
|---|
| 175 | ___
|
|---|
| 176 | ########################################################################
|
|---|
| 177 | # calculate mask by comparing 0..31 to index and save result to stack
|
|---|
| 178 | #
|
|---|
| 179 | $code.=<<___;
|
|---|
| 180 | paddd %xmm0,%xmm1
|
|---|
| 181 | pcmpeqd %xmm5,%xmm0 # compare to 1,0
|
|---|
| 182 | .byte 0x67
|
|---|
| 183 | movdqa %xmm4,%xmm3
|
|---|
| 184 | ___
|
|---|
| 185 | for($k=0;$k<$STRIDE/16-4;$k+=4) {
|
|---|
| 186 | $code.=<<___;
|
|---|
| 187 | paddd %xmm1,%xmm2
|
|---|
| 188 | pcmpeqd %xmm5,%xmm1 # compare to 3,2
|
|---|
| 189 | movdqa %xmm0,`16*($k+0)+112`(%r10)
|
|---|
| 190 | movdqa %xmm4,%xmm0
|
|---|
| 191 |
|
|---|
| 192 | paddd %xmm2,%xmm3
|
|---|
| 193 | pcmpeqd %xmm5,%xmm2 # compare to 5,4
|
|---|
| 194 | movdqa %xmm1,`16*($k+1)+112`(%r10)
|
|---|
| 195 | movdqa %xmm4,%xmm1
|
|---|
| 196 |
|
|---|
| 197 | paddd %xmm3,%xmm0
|
|---|
| 198 | pcmpeqd %xmm5,%xmm3 # compare to 7,6
|
|---|
| 199 | movdqa %xmm2,`16*($k+2)+112`(%r10)
|
|---|
| 200 | movdqa %xmm4,%xmm2
|
|---|
| 201 |
|
|---|
| 202 | paddd %xmm0,%xmm1
|
|---|
| 203 | pcmpeqd %xmm5,%xmm0
|
|---|
| 204 | movdqa %xmm3,`16*($k+3)+112`(%r10)
|
|---|
| 205 | movdqa %xmm4,%xmm3
|
|---|
| 206 | ___
|
|---|
| 207 | }
|
|---|
| 208 | $code.=<<___; # last iteration can be optimized
|
|---|
| 209 | paddd %xmm1,%xmm2
|
|---|
| 210 | pcmpeqd %xmm5,%xmm1
|
|---|
| 211 | movdqa %xmm0,`16*($k+0)+112`(%r10)
|
|---|
| 212 |
|
|---|
| 213 | paddd %xmm2,%xmm3
|
|---|
| 214 | .byte 0x67
|
|---|
| 215 | pcmpeqd %xmm5,%xmm2
|
|---|
| 216 | movdqa %xmm1,`16*($k+1)+112`(%r10)
|
|---|
| 217 |
|
|---|
| 218 | pcmpeqd %xmm5,%xmm3
|
|---|
| 219 | movdqa %xmm2,`16*($k+2)+112`(%r10)
|
|---|
| 220 | pand `16*($k+0)-128`($bp),%xmm0 # while it's still in register
|
|---|
| 221 |
|
|---|
| 222 | pand `16*($k+1)-128`($bp),%xmm1
|
|---|
| 223 | pand `16*($k+2)-128`($bp),%xmm2
|
|---|
| 224 | movdqa %xmm3,`16*($k+3)+112`(%r10)
|
|---|
| 225 | pand `16*($k+3)-128`($bp),%xmm3
|
|---|
| 226 | por %xmm2,%xmm0
|
|---|
| 227 | por %xmm3,%xmm1
|
|---|
| 228 | ___
|
|---|
| 229 | for($k=0;$k<$STRIDE/16-4;$k+=4) {
|
|---|
| 230 | $code.=<<___;
|
|---|
| 231 | movdqa `16*($k+0)-128`($bp),%xmm4
|
|---|
| 232 | movdqa `16*($k+1)-128`($bp),%xmm5
|
|---|
| 233 | movdqa `16*($k+2)-128`($bp),%xmm2
|
|---|
| 234 | pand `16*($k+0)+112`(%r10),%xmm4
|
|---|
| 235 | movdqa `16*($k+3)-128`($bp),%xmm3
|
|---|
| 236 | pand `16*($k+1)+112`(%r10),%xmm5
|
|---|
| 237 | por %xmm4,%xmm0
|
|---|
| 238 | pand `16*($k+2)+112`(%r10),%xmm2
|
|---|
| 239 | por %xmm5,%xmm1
|
|---|
| 240 | pand `16*($k+3)+112`(%r10),%xmm3
|
|---|
| 241 | por %xmm2,%xmm0
|
|---|
| 242 | por %xmm3,%xmm1
|
|---|
| 243 | ___
|
|---|
| 244 | }
|
|---|
| 245 | $code.=<<___;
|
|---|
| 246 | por %xmm1,%xmm0
|
|---|
| 247 | pshufd \$0x4e,%xmm0,%xmm1
|
|---|
| 248 | por %xmm1,%xmm0
|
|---|
| 249 | lea $STRIDE($bp),$bp
|
|---|
| 250 | movq %xmm0,$m0 # m0=bp[0]
|
|---|
| 251 |
|
|---|
| 252 | mov ($n0),$n0 # pull n0[0] value
|
|---|
| 253 | mov ($ap),%rax
|
|---|
| 254 |
|
|---|
| 255 | xor $i,$i # i=0
|
|---|
| 256 | xor $j,$j # j=0
|
|---|
| 257 |
|
|---|
| 258 | mov $n0,$m1
|
|---|
| 259 | mulq $m0 # ap[0]*bp[0]
|
|---|
| 260 | mov %rax,$lo0
|
|---|
| 261 | mov ($np),%rax
|
|---|
| 262 |
|
|---|
| 263 | imulq $lo0,$m1 # "tp[0]"*n0
|
|---|
| 264 | mov %rdx,$hi0
|
|---|
| 265 |
|
|---|
| 266 | mulq $m1 # np[0]*m1
|
|---|
| 267 | add %rax,$lo0 # discarded
|
|---|
| 268 | mov 8($ap),%rax
|
|---|
| 269 | adc \$0,%rdx
|
|---|
| 270 | mov %rdx,$hi1
|
|---|
| 271 |
|
|---|
| 272 | lea 1($j),$j # j++
|
|---|
| 273 | jmp .L1st_enter
|
|---|
| 274 |
|
|---|
| 275 | .align 16
|
|---|
| 276 | .L1st:
|
|---|
| 277 | add %rax,$hi1
|
|---|
| 278 | mov ($ap,$j,8),%rax
|
|---|
| 279 | adc \$0,%rdx
|
|---|
| 280 | add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
|
|---|
| 281 | mov $lo0,$hi0
|
|---|
| 282 | adc \$0,%rdx
|
|---|
| 283 | mov $hi1,-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 284 | mov %rdx,$hi1
|
|---|
| 285 |
|
|---|
| 286 | .L1st_enter:
|
|---|
| 287 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 288 | add %rax,$hi0
|
|---|
| 289 | mov ($np,$j,8),%rax
|
|---|
| 290 | adc \$0,%rdx
|
|---|
| 291 | lea 1($j),$j # j++
|
|---|
| 292 | mov %rdx,$lo0
|
|---|
| 293 |
|
|---|
| 294 | mulq $m1 # np[j]*m1
|
|---|
| 295 | cmp $num,$j
|
|---|
| 296 | jne .L1st # note that upon exit $j==$num, so
|
|---|
| 297 | # they can be used interchangeably
|
|---|
| 298 |
|
|---|
| 299 | add %rax,$hi1
|
|---|
| 300 | adc \$0,%rdx
|
|---|
| 301 | add $hi0,$hi1 # np[j]*m1+ap[j]*bp[0]
|
|---|
| 302 | adc \$0,%rdx
|
|---|
| 303 | mov $hi1,-16(%rsp,$num,8) # tp[num-1]
|
|---|
| 304 | mov %rdx,$hi1
|
|---|
| 305 | mov $lo0,$hi0
|
|---|
| 306 |
|
|---|
| 307 | xor %rdx,%rdx
|
|---|
| 308 | add $hi0,$hi1
|
|---|
| 309 | adc \$0,%rdx
|
|---|
| 310 | mov $hi1,-8(%rsp,$num,8)
|
|---|
| 311 | mov %rdx,(%rsp,$num,8) # store upmost overflow bit
|
|---|
| 312 |
|
|---|
| 313 | lea 1($i),$i # i++
|
|---|
| 314 | jmp .Louter
|
|---|
| 315 | .align 16
|
|---|
| 316 | .Louter:
|
|---|
| 317 | lea 24+128(%rsp,$num,8),%rdx # where 256-byte mask is (+size optimization)
|
|---|
| 318 | and \$-16,%rdx
|
|---|
| 319 | pxor %xmm4,%xmm4
|
|---|
| 320 | pxor %xmm5,%xmm5
|
|---|
| 321 | ___
|
|---|
| 322 | for($k=0;$k<$STRIDE/16;$k+=4) {
|
|---|
| 323 | $code.=<<___;
|
|---|
| 324 | movdqa `16*($k+0)-128`($bp),%xmm0
|
|---|
| 325 | movdqa `16*($k+1)-128`($bp),%xmm1
|
|---|
| 326 | movdqa `16*($k+2)-128`($bp),%xmm2
|
|---|
| 327 | movdqa `16*($k+3)-128`($bp),%xmm3
|
|---|
| 328 | pand `16*($k+0)-128`(%rdx),%xmm0
|
|---|
| 329 | pand `16*($k+1)-128`(%rdx),%xmm1
|
|---|
| 330 | por %xmm0,%xmm4
|
|---|
| 331 | pand `16*($k+2)-128`(%rdx),%xmm2
|
|---|
| 332 | por %xmm1,%xmm5
|
|---|
| 333 | pand `16*($k+3)-128`(%rdx),%xmm3
|
|---|
| 334 | por %xmm2,%xmm4
|
|---|
| 335 | por %xmm3,%xmm5
|
|---|
| 336 | ___
|
|---|
| 337 | }
|
|---|
| 338 | $code.=<<___;
|
|---|
| 339 | por %xmm5,%xmm4
|
|---|
| 340 | pshufd \$0x4e,%xmm4,%xmm0
|
|---|
| 341 | por %xmm4,%xmm0
|
|---|
| 342 | lea $STRIDE($bp),$bp
|
|---|
| 343 |
|
|---|
| 344 | mov ($ap),%rax # ap[0]
|
|---|
| 345 | movq %xmm0,$m0 # m0=bp[i]
|
|---|
| 346 |
|
|---|
| 347 | xor $j,$j # j=0
|
|---|
| 348 | mov $n0,$m1
|
|---|
| 349 | mov (%rsp),$lo0
|
|---|
| 350 |
|
|---|
| 351 | mulq $m0 # ap[0]*bp[i]
|
|---|
| 352 | add %rax,$lo0 # ap[0]*bp[i]+tp[0]
|
|---|
| 353 | mov ($np),%rax
|
|---|
| 354 | adc \$0,%rdx
|
|---|
| 355 |
|
|---|
| 356 | imulq $lo0,$m1 # tp[0]*n0
|
|---|
| 357 | mov %rdx,$hi0
|
|---|
| 358 |
|
|---|
| 359 | mulq $m1 # np[0]*m1
|
|---|
| 360 | add %rax,$lo0 # discarded
|
|---|
| 361 | mov 8($ap),%rax
|
|---|
| 362 | adc \$0,%rdx
|
|---|
| 363 | mov 8(%rsp),$lo0 # tp[1]
|
|---|
| 364 | mov %rdx,$hi1
|
|---|
| 365 |
|
|---|
| 366 | lea 1($j),$j # j++
|
|---|
| 367 | jmp .Linner_enter
|
|---|
| 368 |
|
|---|
| 369 | .align 16
|
|---|
| 370 | .Linner:
|
|---|
| 371 | add %rax,$hi1
|
|---|
| 372 | mov ($ap,$j,8),%rax
|
|---|
| 373 | adc \$0,%rdx
|
|---|
| 374 | add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
|
|---|
| 375 | mov (%rsp,$j,8),$lo0
|
|---|
| 376 | adc \$0,%rdx
|
|---|
| 377 | mov $hi1,-16(%rsp,$j,8) # tp[j-1]
|
|---|
| 378 | mov %rdx,$hi1
|
|---|
| 379 |
|
|---|
| 380 | .Linner_enter:
|
|---|
| 381 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 382 | add %rax,$hi0
|
|---|
| 383 | mov ($np,$j,8),%rax
|
|---|
| 384 | adc \$0,%rdx
|
|---|
| 385 | add $hi0,$lo0 # ap[j]*bp[i]+tp[j]
|
|---|
| 386 | mov %rdx,$hi0
|
|---|
| 387 | adc \$0,$hi0
|
|---|
| 388 | lea 1($j),$j # j++
|
|---|
| 389 |
|
|---|
| 390 | mulq $m1 # np[j]*m1
|
|---|
| 391 | cmp $num,$j
|
|---|
| 392 | jne .Linner # note that upon exit $j==$num, so
|
|---|
| 393 | # they can be used interchangeably
|
|---|
| 394 | add %rax,$hi1
|
|---|
| 395 | adc \$0,%rdx
|
|---|
| 396 | add $lo0,$hi1 # np[j]*m1+ap[j]*bp[i]+tp[j]
|
|---|
| 397 | mov (%rsp,$num,8),$lo0
|
|---|
| 398 | adc \$0,%rdx
|
|---|
| 399 | mov $hi1,-16(%rsp,$num,8) # tp[num-1]
|
|---|
| 400 | mov %rdx,$hi1
|
|---|
| 401 |
|
|---|
| 402 | xor %rdx,%rdx
|
|---|
| 403 | add $hi0,$hi1
|
|---|
| 404 | adc \$0,%rdx
|
|---|
| 405 | add $lo0,$hi1 # pull upmost overflow bit
|
|---|
| 406 | adc \$0,%rdx
|
|---|
| 407 | mov $hi1,-8(%rsp,$num,8)
|
|---|
| 408 | mov %rdx,(%rsp,$num,8) # store upmost overflow bit
|
|---|
| 409 |
|
|---|
| 410 | lea 1($i),$i # i++
|
|---|
| 411 | cmp $num,$i
|
|---|
| 412 | jb .Louter
|
|---|
| 413 |
|
|---|
| 414 | xor $i,$i # i=0 and clear CF!
|
|---|
| 415 | mov (%rsp),%rax # tp[0]
|
|---|
| 416 | lea (%rsp),$ap # borrow ap for tp
|
|---|
| 417 | mov $num,$j # j=num
|
|---|
| 418 | jmp .Lsub
|
|---|
| 419 | .align 16
|
|---|
| 420 | .Lsub: sbb ($np,$i,8),%rax
|
|---|
| 421 | mov %rax,($rp,$i,8) # rp[i]=tp[i]-np[i]
|
|---|
| 422 | mov 8($ap,$i,8),%rax # tp[i+1]
|
|---|
| 423 | lea 1($i),$i # i++
|
|---|
| 424 | dec $j # doesn't affect CF!
|
|---|
| 425 | jnz .Lsub
|
|---|
| 426 |
|
|---|
| 427 | sbb \$0,%rax # handle upmost overflow bit
|
|---|
| 428 | mov \$-1,%rbx
|
|---|
| 429 | xor %rax,%rbx
|
|---|
| 430 | xor $i,$i
|
|---|
| 431 | mov $num,$j # j=num
|
|---|
| 432 |
|
|---|
| 433 | .Lcopy: # conditional copy
|
|---|
| 434 | mov ($rp,$i,8),%rcx
|
|---|
| 435 | mov (%rsp,$i,8),%rdx
|
|---|
| 436 | and %rbx,%rcx
|
|---|
| 437 | and %rax,%rdx
|
|---|
| 438 | mov $i,(%rsp,$i,8) # zap temporary vector
|
|---|
| 439 | or %rcx,%rdx
|
|---|
| 440 | mov %rdx,($rp,$i,8) # rp[i]=tp[i]
|
|---|
| 441 | lea 1($i),$i
|
|---|
| 442 | sub \$1,$j
|
|---|
| 443 | jnz .Lcopy
|
|---|
| 444 |
|
|---|
| 445 | mov 8(%rsp,$num,8),%rsi # restore %rsp
|
|---|
| 446 | .cfi_def_cfa %rsi,8
|
|---|
| 447 | mov \$1,%rax
|
|---|
| 448 |
|
|---|
| 449 | mov -48(%rsi),%r15
|
|---|
| 450 | .cfi_restore %r15
|
|---|
| 451 | mov -40(%rsi),%r14
|
|---|
| 452 | .cfi_restore %r14
|
|---|
| 453 | mov -32(%rsi),%r13
|
|---|
| 454 | .cfi_restore %r13
|
|---|
| 455 | mov -24(%rsi),%r12
|
|---|
| 456 | .cfi_restore %r12
|
|---|
| 457 | mov -16(%rsi),%rbp
|
|---|
| 458 | .cfi_restore %rbp
|
|---|
| 459 | mov -8(%rsi),%rbx
|
|---|
| 460 | .cfi_restore %rbx
|
|---|
| 461 | lea (%rsi),%rsp
|
|---|
| 462 | .cfi_def_cfa_register %rsp
|
|---|
| 463 | .Lmul_epilogue:
|
|---|
| 464 | ret
|
|---|
| 465 | .cfi_endproc
|
|---|
| 466 | .size bn_mul_mont_gather5,.-bn_mul_mont_gather5
|
|---|
| 467 | ___
|
|---|
| 468 | {{{
|
|---|
| 469 | my @A=("%r10","%r11");
|
|---|
| 470 | my @N=("%r13","%rdi");
|
|---|
| 471 | $code.=<<___;
|
|---|
| 472 | .type bn_mul4x_mont_gather5,\@function,6
|
|---|
| 473 | .align 32
|
|---|
| 474 | bn_mul4x_mont_gather5:
|
|---|
| 475 | .cfi_startproc
|
|---|
| 476 | .byte 0x67
|
|---|
| 477 | mov %rsp,%rax
|
|---|
| 478 | .cfi_def_cfa_register %rax
|
|---|
| 479 | .Lmul4x_enter:
|
|---|
| 480 | ___
|
|---|
| 481 | $code.=<<___ if ($addx);
|
|---|
| 482 | and \$0x80108,%r11d
|
|---|
| 483 | cmp \$0x80108,%r11d # check for AD*X+BMI2+BMI1
|
|---|
| 484 | je .Lmulx4x_enter
|
|---|
| 485 | ___
|
|---|
| 486 | $code.=<<___;
|
|---|
| 487 | push %rbx
|
|---|
| 488 | .cfi_push %rbx
|
|---|
| 489 | push %rbp
|
|---|
| 490 | .cfi_push %rbp
|
|---|
| 491 | push %r12
|
|---|
| 492 | .cfi_push %r12
|
|---|
| 493 | push %r13
|
|---|
| 494 | .cfi_push %r13
|
|---|
| 495 | push %r14
|
|---|
| 496 | .cfi_push %r14
|
|---|
| 497 | push %r15
|
|---|
| 498 | .cfi_push %r15
|
|---|
| 499 | .Lmul4x_prologue:
|
|---|
| 500 |
|
|---|
| 501 | .byte 0x67
|
|---|
| 502 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 503 | lea ($num,$num,2),%r10 # 3*$num in bytes
|
|---|
| 504 | neg $num # -$num
|
|---|
| 505 |
|
|---|
| 506 | ##############################################################
|
|---|
| 507 | # Ensure that stack frame doesn't alias with $rptr+3*$num
|
|---|
| 508 | # modulo 4096, which covers ret[num], am[num] and n[num]
|
|---|
| 509 | # (see bn_exp.c). This is done to allow memory disambiguation
|
|---|
| 510 | # logic do its magic. [Extra [num] is allocated in order
|
|---|
| 511 | # to align with bn_power5's frame, which is cleansed after
|
|---|
| 512 | # completing exponentiation. Extra 256 bytes is for power mask
|
|---|
| 513 | # calculated from 7th argument, the index.]
|
|---|
| 514 | #
|
|---|
| 515 | lea -320(%rsp,$num,2),%r11
|
|---|
| 516 | mov %rsp,%rbp
|
|---|
| 517 | sub $rp,%r11
|
|---|
| 518 | and \$4095,%r11
|
|---|
| 519 | cmp %r11,%r10
|
|---|
| 520 | jb .Lmul4xsp_alt
|
|---|
| 521 | sub %r11,%rbp # align with $rp
|
|---|
| 522 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*num*8+256)
|
|---|
| 523 | jmp .Lmul4xsp_done
|
|---|
| 524 |
|
|---|
| 525 | .align 32
|
|---|
| 526 | .Lmul4xsp_alt:
|
|---|
| 527 | lea 4096-320(,$num,2),%r10
|
|---|
| 528 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*num*8+256)
|
|---|
| 529 | sub %r10,%r11
|
|---|
| 530 | mov \$0,%r10
|
|---|
| 531 | cmovc %r10,%r11
|
|---|
| 532 | sub %r11,%rbp
|
|---|
| 533 | .Lmul4xsp_done:
|
|---|
| 534 | and \$-64,%rbp
|
|---|
| 535 | mov %rsp,%r11
|
|---|
| 536 | sub %rbp,%r11
|
|---|
| 537 | and \$-4096,%r11
|
|---|
| 538 | lea (%rbp,%r11),%rsp
|
|---|
| 539 | mov (%rsp),%r10
|
|---|
| 540 | cmp %rbp,%rsp
|
|---|
| 541 | ja .Lmul4x_page_walk
|
|---|
| 542 | jmp .Lmul4x_page_walk_done
|
|---|
| 543 |
|
|---|
| 544 | .Lmul4x_page_walk:
|
|---|
| 545 | lea -4096(%rsp),%rsp
|
|---|
| 546 | mov (%rsp),%r10
|
|---|
| 547 | cmp %rbp,%rsp
|
|---|
| 548 | ja .Lmul4x_page_walk
|
|---|
| 549 | .Lmul4x_page_walk_done:
|
|---|
| 550 |
|
|---|
| 551 | neg $num
|
|---|
| 552 |
|
|---|
| 553 | mov %rax,40(%rsp)
|
|---|
| 554 | .cfi_cfa_expression %rsp+40,deref,+8
|
|---|
| 555 | .Lmul4x_body:
|
|---|
| 556 |
|
|---|
| 557 | call mul4x_internal
|
|---|
| 558 |
|
|---|
| 559 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 560 | .cfi_def_cfa %rsi,8
|
|---|
| 561 | mov \$1,%rax
|
|---|
| 562 |
|
|---|
| 563 | mov -48(%rsi),%r15
|
|---|
| 564 | .cfi_restore %r15
|
|---|
| 565 | mov -40(%rsi),%r14
|
|---|
| 566 | .cfi_restore %r14
|
|---|
| 567 | mov -32(%rsi),%r13
|
|---|
| 568 | .cfi_restore %r13
|
|---|
| 569 | mov -24(%rsi),%r12
|
|---|
| 570 | .cfi_restore %r12
|
|---|
| 571 | mov -16(%rsi),%rbp
|
|---|
| 572 | .cfi_restore %rbp
|
|---|
| 573 | mov -8(%rsi),%rbx
|
|---|
| 574 | .cfi_restore %rbx
|
|---|
| 575 | lea (%rsi),%rsp
|
|---|
| 576 | .cfi_def_cfa_register %rsp
|
|---|
| 577 | .Lmul4x_epilogue:
|
|---|
| 578 | ret
|
|---|
| 579 | .cfi_endproc
|
|---|
| 580 | .size bn_mul4x_mont_gather5,.-bn_mul4x_mont_gather5
|
|---|
| 581 |
|
|---|
| 582 | .type mul4x_internal,\@abi-omnipotent
|
|---|
| 583 | .align 32
|
|---|
| 584 | mul4x_internal:
|
|---|
| 585 | .cfi_startproc
|
|---|
| 586 | shl \$5,$num # $num was in bytes
|
|---|
| 587 | movd `($win64?56:8)`(%rax),%xmm5 # load 7th argument, index
|
|---|
| 588 | lea .Linc(%rip),%rax
|
|---|
| 589 | lea 128(%rdx,$num),%r13 # end of powers table (+size optimization)
|
|---|
| 590 | shr \$5,$num # restore $num
|
|---|
| 591 | ___
|
|---|
| 592 | $bp="%r12";
|
|---|
| 593 | $STRIDE=2**5*8; # 5 is "window size"
|
|---|
| 594 | $N=$STRIDE/4; # should match cache line size
|
|---|
| 595 | $tp=$i;
|
|---|
| 596 | $code.=<<___;
|
|---|
| 597 | movdqa 0(%rax),%xmm0 # 00000001000000010000000000000000
|
|---|
| 598 | movdqa 16(%rax),%xmm1 # 00000002000000020000000200000002
|
|---|
| 599 | lea 88-112(%rsp,$num),%r10 # place the mask after tp[num+1] (+ICache optimization)
|
|---|
| 600 | lea 128(%rdx),$bp # size optimization
|
|---|
| 601 |
|
|---|
| 602 | pshufd \$0,%xmm5,%xmm5 # broadcast index
|
|---|
| 603 | movdqa %xmm1,%xmm4
|
|---|
| 604 | .byte 0x67,0x67
|
|---|
| 605 | movdqa %xmm1,%xmm2
|
|---|
| 606 | ___
|
|---|
| 607 | ########################################################################
|
|---|
| 608 | # calculate mask by comparing 0..31 to index and save result to stack
|
|---|
| 609 | #
|
|---|
| 610 | $code.=<<___;
|
|---|
| 611 | paddd %xmm0,%xmm1
|
|---|
| 612 | pcmpeqd %xmm5,%xmm0 # compare to 1,0
|
|---|
| 613 | .byte 0x67
|
|---|
| 614 | movdqa %xmm4,%xmm3
|
|---|
| 615 | ___
|
|---|
| 616 | for($i=0;$i<$STRIDE/16-4;$i+=4) {
|
|---|
| 617 | $code.=<<___;
|
|---|
| 618 | paddd %xmm1,%xmm2
|
|---|
| 619 | pcmpeqd %xmm5,%xmm1 # compare to 3,2
|
|---|
| 620 | movdqa %xmm0,`16*($i+0)+112`(%r10)
|
|---|
| 621 | movdqa %xmm4,%xmm0
|
|---|
| 622 |
|
|---|
| 623 | paddd %xmm2,%xmm3
|
|---|
| 624 | pcmpeqd %xmm5,%xmm2 # compare to 5,4
|
|---|
| 625 | movdqa %xmm1,`16*($i+1)+112`(%r10)
|
|---|
| 626 | movdqa %xmm4,%xmm1
|
|---|
| 627 |
|
|---|
| 628 | paddd %xmm3,%xmm0
|
|---|
| 629 | pcmpeqd %xmm5,%xmm3 # compare to 7,6
|
|---|
| 630 | movdqa %xmm2,`16*($i+2)+112`(%r10)
|
|---|
| 631 | movdqa %xmm4,%xmm2
|
|---|
| 632 |
|
|---|
| 633 | paddd %xmm0,%xmm1
|
|---|
| 634 | pcmpeqd %xmm5,%xmm0
|
|---|
| 635 | movdqa %xmm3,`16*($i+3)+112`(%r10)
|
|---|
| 636 | movdqa %xmm4,%xmm3
|
|---|
| 637 | ___
|
|---|
| 638 | }
|
|---|
| 639 | $code.=<<___; # last iteration can be optimized
|
|---|
| 640 | paddd %xmm1,%xmm2
|
|---|
| 641 | pcmpeqd %xmm5,%xmm1
|
|---|
| 642 | movdqa %xmm0,`16*($i+0)+112`(%r10)
|
|---|
| 643 |
|
|---|
| 644 | paddd %xmm2,%xmm3
|
|---|
| 645 | .byte 0x67
|
|---|
| 646 | pcmpeqd %xmm5,%xmm2
|
|---|
| 647 | movdqa %xmm1,`16*($i+1)+112`(%r10)
|
|---|
| 648 |
|
|---|
| 649 | pcmpeqd %xmm5,%xmm3
|
|---|
| 650 | movdqa %xmm2,`16*($i+2)+112`(%r10)
|
|---|
| 651 | pand `16*($i+0)-128`($bp),%xmm0 # while it's still in register
|
|---|
| 652 |
|
|---|
| 653 | pand `16*($i+1)-128`($bp),%xmm1
|
|---|
| 654 | pand `16*($i+2)-128`($bp),%xmm2
|
|---|
| 655 | movdqa %xmm3,`16*($i+3)+112`(%r10)
|
|---|
| 656 | pand `16*($i+3)-128`($bp),%xmm3
|
|---|
| 657 | por %xmm2,%xmm0
|
|---|
| 658 | por %xmm3,%xmm1
|
|---|
| 659 | ___
|
|---|
| 660 | for($i=0;$i<$STRIDE/16-4;$i+=4) {
|
|---|
| 661 | $code.=<<___;
|
|---|
| 662 | movdqa `16*($i+0)-128`($bp),%xmm4
|
|---|
| 663 | movdqa `16*($i+1)-128`($bp),%xmm5
|
|---|
| 664 | movdqa `16*($i+2)-128`($bp),%xmm2
|
|---|
| 665 | pand `16*($i+0)+112`(%r10),%xmm4
|
|---|
| 666 | movdqa `16*($i+3)-128`($bp),%xmm3
|
|---|
| 667 | pand `16*($i+1)+112`(%r10),%xmm5
|
|---|
| 668 | por %xmm4,%xmm0
|
|---|
| 669 | pand `16*($i+2)+112`(%r10),%xmm2
|
|---|
| 670 | por %xmm5,%xmm1
|
|---|
| 671 | pand `16*($i+3)+112`(%r10),%xmm3
|
|---|
| 672 | por %xmm2,%xmm0
|
|---|
| 673 | por %xmm3,%xmm1
|
|---|
| 674 | ___
|
|---|
| 675 | }
|
|---|
| 676 | $code.=<<___;
|
|---|
| 677 | por %xmm1,%xmm0
|
|---|
| 678 | pshufd \$0x4e,%xmm0,%xmm1
|
|---|
| 679 | por %xmm1,%xmm0
|
|---|
| 680 | lea $STRIDE($bp),$bp
|
|---|
| 681 | movq %xmm0,$m0 # m0=bp[0]
|
|---|
| 682 |
|
|---|
| 683 | mov %r13,16+8(%rsp) # save end of b[num]
|
|---|
| 684 | mov $rp, 56+8(%rsp) # save $rp
|
|---|
| 685 |
|
|---|
| 686 | mov ($n0),$n0 # pull n0[0] value
|
|---|
| 687 | mov ($ap),%rax
|
|---|
| 688 | lea ($ap,$num),$ap # end of a[num]
|
|---|
| 689 | neg $num
|
|---|
| 690 |
|
|---|
| 691 | mov $n0,$m1
|
|---|
| 692 | mulq $m0 # ap[0]*bp[0]
|
|---|
| 693 | mov %rax,$A[0]
|
|---|
| 694 | mov ($np),%rax
|
|---|
| 695 |
|
|---|
| 696 | imulq $A[0],$m1 # "tp[0]"*n0
|
|---|
| 697 | lea 64+8(%rsp),$tp
|
|---|
| 698 | mov %rdx,$A[1]
|
|---|
| 699 |
|
|---|
| 700 | mulq $m1 # np[0]*m1
|
|---|
| 701 | add %rax,$A[0] # discarded
|
|---|
| 702 | mov 8($ap,$num),%rax
|
|---|
| 703 | adc \$0,%rdx
|
|---|
| 704 | mov %rdx,$N[1]
|
|---|
| 705 |
|
|---|
| 706 | mulq $m0
|
|---|
| 707 | add %rax,$A[1]
|
|---|
| 708 | mov 8*1($np),%rax
|
|---|
| 709 | adc \$0,%rdx
|
|---|
| 710 | mov %rdx,$A[0]
|
|---|
| 711 |
|
|---|
| 712 | mulq $m1
|
|---|
| 713 | add %rax,$N[1]
|
|---|
| 714 | mov 16($ap,$num),%rax
|
|---|
| 715 | adc \$0,%rdx
|
|---|
| 716 | add $A[1],$N[1]
|
|---|
| 717 | lea 4*8($num),$j # j=4
|
|---|
| 718 | lea 8*4($np),$np
|
|---|
| 719 | adc \$0,%rdx
|
|---|
| 720 | mov $N[1],($tp)
|
|---|
| 721 | mov %rdx,$N[0]
|
|---|
| 722 | jmp .L1st4x
|
|---|
| 723 |
|
|---|
| 724 | .align 32
|
|---|
| 725 | .L1st4x:
|
|---|
| 726 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 727 | add %rax,$A[0]
|
|---|
| 728 | mov -8*2($np),%rax
|
|---|
| 729 | lea 32($tp),$tp
|
|---|
| 730 | adc \$0,%rdx
|
|---|
| 731 | mov %rdx,$A[1]
|
|---|
| 732 |
|
|---|
| 733 | mulq $m1 # np[j]*m1
|
|---|
| 734 | add %rax,$N[0]
|
|---|
| 735 | mov -8($ap,$j),%rax
|
|---|
| 736 | adc \$0,%rdx
|
|---|
| 737 | add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 738 | adc \$0,%rdx
|
|---|
| 739 | mov $N[0],-24($tp) # tp[j-1]
|
|---|
| 740 | mov %rdx,$N[1]
|
|---|
| 741 |
|
|---|
| 742 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 743 | add %rax,$A[1]
|
|---|
| 744 | mov -8*1($np),%rax
|
|---|
| 745 | adc \$0,%rdx
|
|---|
| 746 | mov %rdx,$A[0]
|
|---|
| 747 |
|
|---|
| 748 | mulq $m1 # np[j]*m1
|
|---|
| 749 | add %rax,$N[1]
|
|---|
| 750 | mov ($ap,$j),%rax
|
|---|
| 751 | adc \$0,%rdx
|
|---|
| 752 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 753 | adc \$0,%rdx
|
|---|
| 754 | mov $N[1],-16($tp) # tp[j-1]
|
|---|
| 755 | mov %rdx,$N[0]
|
|---|
| 756 |
|
|---|
| 757 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 758 | add %rax,$A[0]
|
|---|
| 759 | mov 8*0($np),%rax
|
|---|
| 760 | adc \$0,%rdx
|
|---|
| 761 | mov %rdx,$A[1]
|
|---|
| 762 |
|
|---|
| 763 | mulq $m1 # np[j]*m1
|
|---|
| 764 | add %rax,$N[0]
|
|---|
| 765 | mov 8($ap,$j),%rax
|
|---|
| 766 | adc \$0,%rdx
|
|---|
| 767 | add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 768 | adc \$0,%rdx
|
|---|
| 769 | mov $N[0],-8($tp) # tp[j-1]
|
|---|
| 770 | mov %rdx,$N[1]
|
|---|
| 771 |
|
|---|
| 772 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 773 | add %rax,$A[1]
|
|---|
| 774 | mov 8*1($np),%rax
|
|---|
| 775 | adc \$0,%rdx
|
|---|
| 776 | mov %rdx,$A[0]
|
|---|
| 777 |
|
|---|
| 778 | mulq $m1 # np[j]*m1
|
|---|
| 779 | add %rax,$N[1]
|
|---|
| 780 | mov 16($ap,$j),%rax
|
|---|
| 781 | adc \$0,%rdx
|
|---|
| 782 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 783 | lea 8*4($np),$np
|
|---|
| 784 | adc \$0,%rdx
|
|---|
| 785 | mov $N[1],($tp) # tp[j-1]
|
|---|
| 786 | mov %rdx,$N[0]
|
|---|
| 787 |
|
|---|
| 788 | add \$32,$j # j+=4
|
|---|
| 789 | jnz .L1st4x
|
|---|
| 790 |
|
|---|
| 791 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 792 | add %rax,$A[0]
|
|---|
| 793 | mov -8*2($np),%rax
|
|---|
| 794 | lea 32($tp),$tp
|
|---|
| 795 | adc \$0,%rdx
|
|---|
| 796 | mov %rdx,$A[1]
|
|---|
| 797 |
|
|---|
| 798 | mulq $m1 # np[j]*m1
|
|---|
| 799 | add %rax,$N[0]
|
|---|
| 800 | mov -8($ap),%rax
|
|---|
| 801 | adc \$0,%rdx
|
|---|
| 802 | add $A[0],$N[0] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 803 | adc \$0,%rdx
|
|---|
| 804 | mov $N[0],-24($tp) # tp[j-1]
|
|---|
| 805 | mov %rdx,$N[1]
|
|---|
| 806 |
|
|---|
| 807 | mulq $m0 # ap[j]*bp[0]
|
|---|
| 808 | add %rax,$A[1]
|
|---|
| 809 | mov -8*1($np),%rax
|
|---|
| 810 | adc \$0,%rdx
|
|---|
| 811 | mov %rdx,$A[0]
|
|---|
| 812 |
|
|---|
| 813 | mulq $m1 # np[j]*m1
|
|---|
| 814 | add %rax,$N[1]
|
|---|
| 815 | mov ($ap,$num),%rax # ap[0]
|
|---|
| 816 | adc \$0,%rdx
|
|---|
| 817 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[0]
|
|---|
| 818 | adc \$0,%rdx
|
|---|
| 819 | mov $N[1],-16($tp) # tp[j-1]
|
|---|
| 820 | mov %rdx,$N[0]
|
|---|
| 821 |
|
|---|
| 822 | lea ($np,$num),$np # rewind $np
|
|---|
| 823 |
|
|---|
| 824 | xor $N[1],$N[1]
|
|---|
| 825 | add $A[0],$N[0]
|
|---|
| 826 | adc \$0,$N[1]
|
|---|
| 827 | mov $N[0],-8($tp)
|
|---|
| 828 |
|
|---|
| 829 | jmp .Louter4x
|
|---|
| 830 |
|
|---|
| 831 | .align 32
|
|---|
| 832 | .Louter4x:
|
|---|
| 833 | lea 16+128($tp),%rdx # where 256-byte mask is (+size optimization)
|
|---|
| 834 | pxor %xmm4,%xmm4
|
|---|
| 835 | pxor %xmm5,%xmm5
|
|---|
| 836 | ___
|
|---|
| 837 | for($i=0;$i<$STRIDE/16;$i+=4) {
|
|---|
| 838 | $code.=<<___;
|
|---|
| 839 | movdqa `16*($i+0)-128`($bp),%xmm0
|
|---|
| 840 | movdqa `16*($i+1)-128`($bp),%xmm1
|
|---|
| 841 | movdqa `16*($i+2)-128`($bp),%xmm2
|
|---|
| 842 | movdqa `16*($i+3)-128`($bp),%xmm3
|
|---|
| 843 | pand `16*($i+0)-128`(%rdx),%xmm0
|
|---|
| 844 | pand `16*($i+1)-128`(%rdx),%xmm1
|
|---|
| 845 | por %xmm0,%xmm4
|
|---|
| 846 | pand `16*($i+2)-128`(%rdx),%xmm2
|
|---|
| 847 | por %xmm1,%xmm5
|
|---|
| 848 | pand `16*($i+3)-128`(%rdx),%xmm3
|
|---|
| 849 | por %xmm2,%xmm4
|
|---|
| 850 | por %xmm3,%xmm5
|
|---|
| 851 | ___
|
|---|
| 852 | }
|
|---|
| 853 | $code.=<<___;
|
|---|
| 854 | por %xmm5,%xmm4
|
|---|
| 855 | pshufd \$0x4e,%xmm4,%xmm0
|
|---|
| 856 | por %xmm4,%xmm0
|
|---|
| 857 | lea $STRIDE($bp),$bp
|
|---|
| 858 | movq %xmm0,$m0 # m0=bp[i]
|
|---|
| 859 |
|
|---|
| 860 | mov ($tp,$num),$A[0]
|
|---|
| 861 | mov $n0,$m1
|
|---|
| 862 | mulq $m0 # ap[0]*bp[i]
|
|---|
| 863 | add %rax,$A[0] # ap[0]*bp[i]+tp[0]
|
|---|
| 864 | mov ($np),%rax
|
|---|
| 865 | adc \$0,%rdx
|
|---|
| 866 |
|
|---|
| 867 | imulq $A[0],$m1 # tp[0]*n0
|
|---|
| 868 | mov %rdx,$A[1]
|
|---|
| 869 | mov $N[1],($tp) # store upmost overflow bit
|
|---|
| 870 |
|
|---|
| 871 | lea ($tp,$num),$tp # rewind $tp
|
|---|
| 872 |
|
|---|
| 873 | mulq $m1 # np[0]*m1
|
|---|
| 874 | add %rax,$A[0] # "$N[0]", discarded
|
|---|
| 875 | mov 8($ap,$num),%rax
|
|---|
| 876 | adc \$0,%rdx
|
|---|
| 877 | mov %rdx,$N[1]
|
|---|
| 878 |
|
|---|
| 879 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 880 | add %rax,$A[1]
|
|---|
| 881 | mov 8*1($np),%rax
|
|---|
| 882 | adc \$0,%rdx
|
|---|
| 883 | add 8($tp),$A[1] # +tp[1]
|
|---|
| 884 | adc \$0,%rdx
|
|---|
| 885 | mov %rdx,$A[0]
|
|---|
| 886 |
|
|---|
| 887 | mulq $m1 # np[j]*m1
|
|---|
| 888 | add %rax,$N[1]
|
|---|
| 889 | mov 16($ap,$num),%rax
|
|---|
| 890 | adc \$0,%rdx
|
|---|
| 891 | add $A[1],$N[1] # np[j]*m1+ap[j]*bp[i]+tp[j]
|
|---|
| 892 | lea 4*8($num),$j # j=4
|
|---|
| 893 | lea 8*4($np),$np
|
|---|
| 894 | adc \$0,%rdx
|
|---|
| 895 | mov %rdx,$N[0]
|
|---|
| 896 | jmp .Linner4x
|
|---|
| 897 |
|
|---|
| 898 | .align 32
|
|---|
| 899 | .Linner4x:
|
|---|
| 900 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 901 | add %rax,$A[0]
|
|---|
| 902 | mov -8*2($np),%rax
|
|---|
| 903 | adc \$0,%rdx
|
|---|
| 904 | add 16($tp),$A[0] # ap[j]*bp[i]+tp[j]
|
|---|
| 905 | lea 32($tp),$tp
|
|---|
| 906 | adc \$0,%rdx
|
|---|
| 907 | mov %rdx,$A[1]
|
|---|
| 908 |
|
|---|
| 909 | mulq $m1 # np[j]*m1
|
|---|
| 910 | add %rax,$N[0]
|
|---|
| 911 | mov -8($ap,$j),%rax
|
|---|
| 912 | adc \$0,%rdx
|
|---|
| 913 | add $A[0],$N[0]
|
|---|
| 914 | adc \$0,%rdx
|
|---|
| 915 | mov $N[1],-32($tp) # tp[j-1]
|
|---|
| 916 | mov %rdx,$N[1]
|
|---|
| 917 |
|
|---|
| 918 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 919 | add %rax,$A[1]
|
|---|
| 920 | mov -8*1($np),%rax
|
|---|
| 921 | adc \$0,%rdx
|
|---|
| 922 | add -8($tp),$A[1]
|
|---|
| 923 | adc \$0,%rdx
|
|---|
| 924 | mov %rdx,$A[0]
|
|---|
| 925 |
|
|---|
| 926 | mulq $m1 # np[j]*m1
|
|---|
| 927 | add %rax,$N[1]
|
|---|
| 928 | mov ($ap,$j),%rax
|
|---|
| 929 | adc \$0,%rdx
|
|---|
| 930 | add $A[1],$N[1]
|
|---|
| 931 | adc \$0,%rdx
|
|---|
| 932 | mov $N[0],-24($tp) # tp[j-1]
|
|---|
| 933 | mov %rdx,$N[0]
|
|---|
| 934 |
|
|---|
| 935 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 936 | add %rax,$A[0]
|
|---|
| 937 | mov 8*0($np),%rax
|
|---|
| 938 | adc \$0,%rdx
|
|---|
| 939 | add ($tp),$A[0] # ap[j]*bp[i]+tp[j]
|
|---|
| 940 | adc \$0,%rdx
|
|---|
| 941 | mov %rdx,$A[1]
|
|---|
| 942 |
|
|---|
| 943 | mulq $m1 # np[j]*m1
|
|---|
| 944 | add %rax,$N[0]
|
|---|
| 945 | mov 8($ap,$j),%rax
|
|---|
| 946 | adc \$0,%rdx
|
|---|
| 947 | add $A[0],$N[0]
|
|---|
| 948 | adc \$0,%rdx
|
|---|
| 949 | mov $N[1],-16($tp) # tp[j-1]
|
|---|
| 950 | mov %rdx,$N[1]
|
|---|
| 951 |
|
|---|
| 952 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 953 | add %rax,$A[1]
|
|---|
| 954 | mov 8*1($np),%rax
|
|---|
| 955 | adc \$0,%rdx
|
|---|
| 956 | add 8($tp),$A[1]
|
|---|
| 957 | adc \$0,%rdx
|
|---|
| 958 | mov %rdx,$A[0]
|
|---|
| 959 |
|
|---|
| 960 | mulq $m1 # np[j]*m1
|
|---|
| 961 | add %rax,$N[1]
|
|---|
| 962 | mov 16($ap,$j),%rax
|
|---|
| 963 | adc \$0,%rdx
|
|---|
| 964 | add $A[1],$N[1]
|
|---|
| 965 | lea 8*4($np),$np
|
|---|
| 966 | adc \$0,%rdx
|
|---|
| 967 | mov $N[0],-8($tp) # tp[j-1]
|
|---|
| 968 | mov %rdx,$N[0]
|
|---|
| 969 |
|
|---|
| 970 | add \$32,$j # j+=4
|
|---|
| 971 | jnz .Linner4x
|
|---|
| 972 |
|
|---|
| 973 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 974 | add %rax,$A[0]
|
|---|
| 975 | mov -8*2($np),%rax
|
|---|
| 976 | adc \$0,%rdx
|
|---|
| 977 | add 16($tp),$A[0] # ap[j]*bp[i]+tp[j]
|
|---|
| 978 | lea 32($tp),$tp
|
|---|
| 979 | adc \$0,%rdx
|
|---|
| 980 | mov %rdx,$A[1]
|
|---|
| 981 |
|
|---|
| 982 | mulq $m1 # np[j]*m1
|
|---|
| 983 | add %rax,$N[0]
|
|---|
| 984 | mov -8($ap),%rax
|
|---|
| 985 | adc \$0,%rdx
|
|---|
| 986 | add $A[0],$N[0]
|
|---|
| 987 | adc \$0,%rdx
|
|---|
| 988 | mov $N[1],-32($tp) # tp[j-1]
|
|---|
| 989 | mov %rdx,$N[1]
|
|---|
| 990 |
|
|---|
| 991 | mulq $m0 # ap[j]*bp[i]
|
|---|
| 992 | add %rax,$A[1]
|
|---|
| 993 | mov $m1,%rax
|
|---|
| 994 | mov -8*1($np),$m1
|
|---|
| 995 | adc \$0,%rdx
|
|---|
| 996 | add -8($tp),$A[1]
|
|---|
| 997 | adc \$0,%rdx
|
|---|
| 998 | mov %rdx,$A[0]
|
|---|
| 999 |
|
|---|
| 1000 | mulq $m1 # np[j]*m1
|
|---|
| 1001 | add %rax,$N[1]
|
|---|
| 1002 | mov ($ap,$num),%rax # ap[0]
|
|---|
| 1003 | adc \$0,%rdx
|
|---|
| 1004 | add $A[1],$N[1]
|
|---|
| 1005 | adc \$0,%rdx
|
|---|
| 1006 | mov $N[0],-24($tp) # tp[j-1]
|
|---|
| 1007 | mov %rdx,$N[0]
|
|---|
| 1008 |
|
|---|
| 1009 | mov $N[1],-16($tp) # tp[j-1]
|
|---|
| 1010 | lea ($np,$num),$np # rewind $np
|
|---|
| 1011 |
|
|---|
| 1012 | xor $N[1],$N[1]
|
|---|
| 1013 | add $A[0],$N[0]
|
|---|
| 1014 | adc \$0,$N[1]
|
|---|
| 1015 | add ($tp),$N[0] # pull upmost overflow bit
|
|---|
| 1016 | adc \$0,$N[1] # upmost overflow bit
|
|---|
| 1017 | mov $N[0],-8($tp)
|
|---|
| 1018 |
|
|---|
| 1019 | cmp 16+8(%rsp),$bp
|
|---|
| 1020 | jb .Louter4x
|
|---|
| 1021 | ___
|
|---|
| 1022 | if (1) {
|
|---|
| 1023 | $code.=<<___;
|
|---|
| 1024 | xor %rax,%rax
|
|---|
| 1025 | sub $N[0],$m1 # compare top-most words
|
|---|
| 1026 | adc $j,$j # $j is zero
|
|---|
| 1027 | or $j,$N[1]
|
|---|
| 1028 | sub $N[1],%rax # %rax=-$N[1]
|
|---|
| 1029 | lea ($tp,$num),%rbx # tptr in .sqr4x_sub
|
|---|
| 1030 | mov ($np),%r12
|
|---|
| 1031 | lea ($np),%rbp # nptr in .sqr4x_sub
|
|---|
| 1032 | mov %r9,%rcx
|
|---|
| 1033 | sar \$3+2,%rcx
|
|---|
| 1034 | mov 56+8(%rsp),%rdi # rptr in .sqr4x_sub
|
|---|
| 1035 | dec %r12 # so that after 'not' we get -n[0]
|
|---|
| 1036 | xor %r10,%r10
|
|---|
| 1037 | mov 8*1(%rbp),%r13
|
|---|
| 1038 | mov 8*2(%rbp),%r14
|
|---|
| 1039 | mov 8*3(%rbp),%r15
|
|---|
| 1040 | jmp .Lsqr4x_sub_entry
|
|---|
| 1041 | ___
|
|---|
| 1042 | } else {
|
|---|
| 1043 | my @ri=("%rax",$bp,$m0,$m1);
|
|---|
| 1044 | my $rp="%rdx";
|
|---|
| 1045 | $code.=<<___
|
|---|
| 1046 | xor \$1,$N[1]
|
|---|
| 1047 | lea ($tp,$num),$tp # rewind $tp
|
|---|
| 1048 | sar \$5,$num # cf=0
|
|---|
| 1049 | lea ($np,$N[1],8),$np
|
|---|
| 1050 | mov 56+8(%rsp),$rp # restore $rp
|
|---|
| 1051 | jmp .Lsub4x
|
|---|
| 1052 |
|
|---|
| 1053 | .align 32
|
|---|
| 1054 | .Lsub4x:
|
|---|
| 1055 | .byte 0x66
|
|---|
| 1056 | mov 8*0($tp),@ri[0]
|
|---|
| 1057 | mov 8*1($tp),@ri[1]
|
|---|
| 1058 | .byte 0x66
|
|---|
| 1059 | sbb 16*0($np),@ri[0]
|
|---|
| 1060 | mov 8*2($tp),@ri[2]
|
|---|
| 1061 | sbb 16*1($np),@ri[1]
|
|---|
| 1062 | mov 3*8($tp),@ri[3]
|
|---|
| 1063 | lea 4*8($tp),$tp
|
|---|
| 1064 | sbb 16*2($np),@ri[2]
|
|---|
| 1065 | mov @ri[0],8*0($rp)
|
|---|
| 1066 | sbb 16*3($np),@ri[3]
|
|---|
| 1067 | lea 16*4($np),$np
|
|---|
| 1068 | mov @ri[1],8*1($rp)
|
|---|
| 1069 | mov @ri[2],8*2($rp)
|
|---|
| 1070 | mov @ri[3],8*3($rp)
|
|---|
| 1071 | lea 8*4($rp),$rp
|
|---|
| 1072 |
|
|---|
| 1073 | inc $num
|
|---|
| 1074 | jnz .Lsub4x
|
|---|
| 1075 |
|
|---|
| 1076 | ret
|
|---|
| 1077 | ___
|
|---|
| 1078 | }
|
|---|
| 1079 | $code.=<<___;
|
|---|
| 1080 | .cfi_endproc
|
|---|
| 1081 | .size mul4x_internal,.-mul4x_internal
|
|---|
| 1082 | ___
|
|---|
| 1083 | }}}
|
|---|
| 1084 | |
|---|
| 1085 | {{{
|
|---|
| 1086 | ######################################################################
|
|---|
| 1087 | # void bn_power5(
|
|---|
| 1088 | my $rptr="%rdi"; # BN_ULONG *rptr,
|
|---|
| 1089 | my $aptr="%rsi"; # const BN_ULONG *aptr,
|
|---|
| 1090 | my $bptr="%rdx"; # const void *table,
|
|---|
| 1091 | my $nptr="%rcx"; # const BN_ULONG *nptr,
|
|---|
| 1092 | my $n0 ="%r8"; # const BN_ULONG *n0);
|
|---|
| 1093 | my $num ="%r9"; # int num, has to be divisible by 8
|
|---|
| 1094 | # int pwr
|
|---|
| 1095 |
|
|---|
| 1096 | my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
|
|---|
| 1097 | my @A0=("%r10","%r11");
|
|---|
| 1098 | my @A1=("%r12","%r13");
|
|---|
| 1099 | my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
|
|---|
| 1100 |
|
|---|
| 1101 | $code.=<<___;
|
|---|
| 1102 | .globl bn_power5
|
|---|
| 1103 | .type bn_power5,\@function,6
|
|---|
| 1104 | .align 32
|
|---|
| 1105 | bn_power5:
|
|---|
| 1106 | .cfi_startproc
|
|---|
| 1107 | mov %rsp,%rax
|
|---|
| 1108 | .cfi_def_cfa_register %rax
|
|---|
| 1109 | ___
|
|---|
| 1110 | $code.=<<___ if ($addx);
|
|---|
| 1111 | mov OPENSSL_ia32cap_P+8(%rip),%r11d
|
|---|
| 1112 | and \$0x80108,%r11d
|
|---|
| 1113 | cmp \$0x80108,%r11d # check for AD*X+BMI2+BMI1
|
|---|
| 1114 | je .Lpowerx5_enter
|
|---|
| 1115 | ___
|
|---|
| 1116 | $code.=<<___;
|
|---|
| 1117 | push %rbx
|
|---|
| 1118 | .cfi_push %rbx
|
|---|
| 1119 | push %rbp
|
|---|
| 1120 | .cfi_push %rbp
|
|---|
| 1121 | push %r12
|
|---|
| 1122 | .cfi_push %r12
|
|---|
| 1123 | push %r13
|
|---|
| 1124 | .cfi_push %r13
|
|---|
| 1125 | push %r14
|
|---|
| 1126 | .cfi_push %r14
|
|---|
| 1127 | push %r15
|
|---|
| 1128 | .cfi_push %r15
|
|---|
| 1129 | .Lpower5_prologue:
|
|---|
| 1130 |
|
|---|
| 1131 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 1132 | lea ($num,$num,2),%r10d # 3*$num
|
|---|
| 1133 | neg $num
|
|---|
| 1134 | mov ($n0),$n0 # *n0
|
|---|
| 1135 |
|
|---|
| 1136 | ##############################################################
|
|---|
| 1137 | # Ensure that stack frame doesn't alias with $rptr+3*$num
|
|---|
| 1138 | # modulo 4096, which covers ret[num], am[num] and n[num]
|
|---|
| 1139 | # (see bn_exp.c). This is done to allow memory disambiguation
|
|---|
| 1140 | # logic do its magic. [Extra 256 bytes is for power mask
|
|---|
| 1141 | # calculated from 7th argument, the index.]
|
|---|
| 1142 | #
|
|---|
| 1143 | lea -320(%rsp,$num,2),%r11
|
|---|
| 1144 | mov %rsp,%rbp
|
|---|
| 1145 | sub $rptr,%r11
|
|---|
| 1146 | and \$4095,%r11
|
|---|
| 1147 | cmp %r11,%r10
|
|---|
| 1148 | jb .Lpwr_sp_alt
|
|---|
| 1149 | sub %r11,%rbp # align with $aptr
|
|---|
| 1150 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*num*8+256)
|
|---|
| 1151 | jmp .Lpwr_sp_done
|
|---|
| 1152 |
|
|---|
| 1153 | .align 32
|
|---|
| 1154 | .Lpwr_sp_alt:
|
|---|
| 1155 | lea 4096-320(,$num,2),%r10
|
|---|
| 1156 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*num*8+256)
|
|---|
| 1157 | sub %r10,%r11
|
|---|
| 1158 | mov \$0,%r10
|
|---|
| 1159 | cmovc %r10,%r11
|
|---|
| 1160 | sub %r11,%rbp
|
|---|
| 1161 | .Lpwr_sp_done:
|
|---|
| 1162 | and \$-64,%rbp
|
|---|
| 1163 | mov %rsp,%r11
|
|---|
| 1164 | sub %rbp,%r11
|
|---|
| 1165 | and \$-4096,%r11
|
|---|
| 1166 | lea (%rbp,%r11),%rsp
|
|---|
| 1167 | mov (%rsp),%r10
|
|---|
| 1168 | cmp %rbp,%rsp
|
|---|
| 1169 | ja .Lpwr_page_walk
|
|---|
| 1170 | jmp .Lpwr_page_walk_done
|
|---|
| 1171 |
|
|---|
| 1172 | .Lpwr_page_walk:
|
|---|
| 1173 | lea -4096(%rsp),%rsp
|
|---|
| 1174 | mov (%rsp),%r10
|
|---|
| 1175 | cmp %rbp,%rsp
|
|---|
| 1176 | ja .Lpwr_page_walk
|
|---|
| 1177 | .Lpwr_page_walk_done:
|
|---|
| 1178 |
|
|---|
| 1179 | mov $num,%r10
|
|---|
| 1180 | neg $num
|
|---|
| 1181 |
|
|---|
| 1182 | ##############################################################
|
|---|
| 1183 | # Stack layout
|
|---|
| 1184 | #
|
|---|
| 1185 | # +0 saved $num, used in reduction section
|
|---|
| 1186 | # +8 &t[2*$num], used in reduction section
|
|---|
| 1187 | # +32 saved *n0
|
|---|
| 1188 | # +40 saved %rsp
|
|---|
| 1189 | # +48 t[2*$num]
|
|---|
| 1190 | #
|
|---|
| 1191 | mov $n0, 32(%rsp)
|
|---|
| 1192 | mov %rax, 40(%rsp) # save original %rsp
|
|---|
| 1193 | .cfi_cfa_expression %rsp+40,deref,+8
|
|---|
| 1194 | .Lpower5_body:
|
|---|
| 1195 | movq $rptr,%xmm1 # save $rptr, used in sqr8x
|
|---|
| 1196 | movq $nptr,%xmm2 # save $nptr
|
|---|
| 1197 | movq %r10, %xmm3 # -$num, used in sqr8x
|
|---|
| 1198 | movq $bptr,%xmm4
|
|---|
| 1199 |
|
|---|
| 1200 | call __bn_sqr8x_internal
|
|---|
| 1201 | call __bn_post4x_internal
|
|---|
| 1202 | call __bn_sqr8x_internal
|
|---|
| 1203 | call __bn_post4x_internal
|
|---|
| 1204 | call __bn_sqr8x_internal
|
|---|
| 1205 | call __bn_post4x_internal
|
|---|
| 1206 | call __bn_sqr8x_internal
|
|---|
| 1207 | call __bn_post4x_internal
|
|---|
| 1208 | call __bn_sqr8x_internal
|
|---|
| 1209 | call __bn_post4x_internal
|
|---|
| 1210 |
|
|---|
| 1211 | movq %xmm2,$nptr
|
|---|
| 1212 | movq %xmm4,$bptr
|
|---|
| 1213 | mov $aptr,$rptr
|
|---|
| 1214 | mov 40(%rsp),%rax
|
|---|
| 1215 | lea 32(%rsp),$n0
|
|---|
| 1216 |
|
|---|
| 1217 | call mul4x_internal
|
|---|
| 1218 |
|
|---|
| 1219 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 1220 | .cfi_def_cfa %rsi,8
|
|---|
| 1221 | mov \$1,%rax
|
|---|
| 1222 | mov -48(%rsi),%r15
|
|---|
| 1223 | .cfi_restore %r15
|
|---|
| 1224 | mov -40(%rsi),%r14
|
|---|
| 1225 | .cfi_restore %r14
|
|---|
| 1226 | mov -32(%rsi),%r13
|
|---|
| 1227 | .cfi_restore %r13
|
|---|
| 1228 | mov -24(%rsi),%r12
|
|---|
| 1229 | .cfi_restore %r12
|
|---|
| 1230 | mov -16(%rsi),%rbp
|
|---|
| 1231 | .cfi_restore %rbp
|
|---|
| 1232 | mov -8(%rsi),%rbx
|
|---|
| 1233 | .cfi_restore %rbx
|
|---|
| 1234 | lea (%rsi),%rsp
|
|---|
| 1235 | .cfi_def_cfa_register %rsp
|
|---|
| 1236 | .Lpower5_epilogue:
|
|---|
| 1237 | ret
|
|---|
| 1238 | .cfi_endproc
|
|---|
| 1239 | .size bn_power5,.-bn_power5
|
|---|
| 1240 |
|
|---|
| 1241 | .globl bn_sqr8x_internal
|
|---|
| 1242 | .hidden bn_sqr8x_internal
|
|---|
| 1243 | .type bn_sqr8x_internal,\@abi-omnipotent
|
|---|
| 1244 | .align 32
|
|---|
| 1245 | bn_sqr8x_internal:
|
|---|
| 1246 | __bn_sqr8x_internal:
|
|---|
| 1247 | .cfi_startproc
|
|---|
| 1248 | ##############################################################
|
|---|
| 1249 | # Squaring part:
|
|---|
| 1250 | #
|
|---|
| 1251 | # a) multiply-n-add everything but a[i]*a[i];
|
|---|
| 1252 | # b) shift result of a) by 1 to the left and accumulate
|
|---|
| 1253 | # a[i]*a[i] products;
|
|---|
| 1254 | #
|
|---|
| 1255 | ##############################################################
|
|---|
| 1256 | # a[1]a[0]
|
|---|
| 1257 | # a[2]a[0]
|
|---|
| 1258 | # a[3]a[0]
|
|---|
| 1259 | # a[2]a[1]
|
|---|
| 1260 | # a[4]a[0]
|
|---|
| 1261 | # a[3]a[1]
|
|---|
| 1262 | # a[5]a[0]
|
|---|
| 1263 | # a[4]a[1]
|
|---|
| 1264 | # a[3]a[2]
|
|---|
| 1265 | # a[6]a[0]
|
|---|
| 1266 | # a[5]a[1]
|
|---|
| 1267 | # a[4]a[2]
|
|---|
| 1268 | # a[7]a[0]
|
|---|
| 1269 | # a[6]a[1]
|
|---|
| 1270 | # a[5]a[2]
|
|---|
| 1271 | # a[4]a[3]
|
|---|
| 1272 | # a[7]a[1]
|
|---|
| 1273 | # a[6]a[2]
|
|---|
| 1274 | # a[5]a[3]
|
|---|
| 1275 | # a[7]a[2]
|
|---|
| 1276 | # a[6]a[3]
|
|---|
| 1277 | # a[5]a[4]
|
|---|
| 1278 | # a[7]a[3]
|
|---|
| 1279 | # a[6]a[4]
|
|---|
| 1280 | # a[7]a[4]
|
|---|
| 1281 | # a[6]a[5]
|
|---|
| 1282 | # a[7]a[5]
|
|---|
| 1283 | # a[7]a[6]
|
|---|
| 1284 | # a[1]a[0]
|
|---|
| 1285 | # a[2]a[0]
|
|---|
| 1286 | # a[3]a[0]
|
|---|
| 1287 | # a[4]a[0]
|
|---|
| 1288 | # a[5]a[0]
|
|---|
| 1289 | # a[6]a[0]
|
|---|
| 1290 | # a[7]a[0]
|
|---|
| 1291 | # a[2]a[1]
|
|---|
| 1292 | # a[3]a[1]
|
|---|
| 1293 | # a[4]a[1]
|
|---|
| 1294 | # a[5]a[1]
|
|---|
| 1295 | # a[6]a[1]
|
|---|
| 1296 | # a[7]a[1]
|
|---|
| 1297 | # a[3]a[2]
|
|---|
| 1298 | # a[4]a[2]
|
|---|
| 1299 | # a[5]a[2]
|
|---|
| 1300 | # a[6]a[2]
|
|---|
| 1301 | # a[7]a[2]
|
|---|
| 1302 | # a[4]a[3]
|
|---|
| 1303 | # a[5]a[3]
|
|---|
| 1304 | # a[6]a[3]
|
|---|
| 1305 | # a[7]a[3]
|
|---|
| 1306 | # a[5]a[4]
|
|---|
| 1307 | # a[6]a[4]
|
|---|
| 1308 | # a[7]a[4]
|
|---|
| 1309 | # a[6]a[5]
|
|---|
| 1310 | # a[7]a[5]
|
|---|
| 1311 | # a[7]a[6]
|
|---|
| 1312 | # a[0]a[0]
|
|---|
| 1313 | # a[1]a[1]
|
|---|
| 1314 | # a[2]a[2]
|
|---|
| 1315 | # a[3]a[3]
|
|---|
| 1316 | # a[4]a[4]
|
|---|
| 1317 | # a[5]a[5]
|
|---|
| 1318 | # a[6]a[6]
|
|---|
| 1319 | # a[7]a[7]
|
|---|
| 1320 |
|
|---|
| 1321 | lea 32(%r10),$i # $i=-($num-32)
|
|---|
| 1322 | lea ($aptr,$num),$aptr # end of a[] buffer, ($aptr,$i)=&ap[2]
|
|---|
| 1323 |
|
|---|
| 1324 | mov $num,$j # $j=$num
|
|---|
| 1325 |
|
|---|
| 1326 | # comments apply to $num==8 case
|
|---|
| 1327 | mov -32($aptr,$i),$a0 # a[0]
|
|---|
| 1328 | lea 48+8(%rsp,$num,2),$tptr # end of tp[] buffer, &tp[2*$num]
|
|---|
| 1329 | mov -24($aptr,$i),%rax # a[1]
|
|---|
| 1330 | lea -32($tptr,$i),$tptr # end of tp[] window, &tp[2*$num-"$i"]
|
|---|
| 1331 | mov -16($aptr,$i),$ai # a[2]
|
|---|
| 1332 | mov %rax,$a1
|
|---|
| 1333 |
|
|---|
| 1334 | mul $a0 # a[1]*a[0]
|
|---|
| 1335 | mov %rax,$A0[0] # a[1]*a[0]
|
|---|
| 1336 | mov $ai,%rax # a[2]
|
|---|
| 1337 | mov %rdx,$A0[1]
|
|---|
| 1338 | mov $A0[0],-24($tptr,$i) # t[1]
|
|---|
| 1339 |
|
|---|
| 1340 | mul $a0 # a[2]*a[0]
|
|---|
| 1341 | add %rax,$A0[1]
|
|---|
| 1342 | mov $ai,%rax
|
|---|
| 1343 | adc \$0,%rdx
|
|---|
| 1344 | mov $A0[1],-16($tptr,$i) # t[2]
|
|---|
| 1345 | mov %rdx,$A0[0]
|
|---|
| 1346 |
|
|---|
| 1347 |
|
|---|
| 1348 | mov -8($aptr,$i),$ai # a[3]
|
|---|
| 1349 | mul $a1 # a[2]*a[1]
|
|---|
| 1350 | mov %rax,$A1[0] # a[2]*a[1]+t[3]
|
|---|
| 1351 | mov $ai,%rax
|
|---|
| 1352 | mov %rdx,$A1[1]
|
|---|
| 1353 |
|
|---|
| 1354 | lea ($i),$j
|
|---|
| 1355 | mul $a0 # a[3]*a[0]
|
|---|
| 1356 | add %rax,$A0[0] # a[3]*a[0]+a[2]*a[1]+t[3]
|
|---|
| 1357 | mov $ai,%rax
|
|---|
| 1358 | mov %rdx,$A0[1]
|
|---|
| 1359 | adc \$0,$A0[1]
|
|---|
| 1360 | add $A1[0],$A0[0]
|
|---|
| 1361 | adc \$0,$A0[1]
|
|---|
| 1362 | mov $A0[0],-8($tptr,$j) # t[3]
|
|---|
| 1363 | jmp .Lsqr4x_1st
|
|---|
| 1364 |
|
|---|
| 1365 | .align 32
|
|---|
| 1366 | .Lsqr4x_1st:
|
|---|
| 1367 | mov ($aptr,$j),$ai # a[4]
|
|---|
| 1368 | mul $a1 # a[3]*a[1]
|
|---|
| 1369 | add %rax,$A1[1] # a[3]*a[1]+t[4]
|
|---|
| 1370 | mov $ai,%rax
|
|---|
| 1371 | mov %rdx,$A1[0]
|
|---|
| 1372 | adc \$0,$A1[0]
|
|---|
| 1373 |
|
|---|
| 1374 | mul $a0 # a[4]*a[0]
|
|---|
| 1375 | add %rax,$A0[1] # a[4]*a[0]+a[3]*a[1]+t[4]
|
|---|
| 1376 | mov $ai,%rax # a[3]
|
|---|
| 1377 | mov 8($aptr,$j),$ai # a[5]
|
|---|
| 1378 | mov %rdx,$A0[0]
|
|---|
| 1379 | adc \$0,$A0[0]
|
|---|
| 1380 | add $A1[1],$A0[1]
|
|---|
| 1381 | adc \$0,$A0[0]
|
|---|
| 1382 |
|
|---|
| 1383 |
|
|---|
| 1384 | mul $a1 # a[4]*a[3]
|
|---|
| 1385 | add %rax,$A1[0] # a[4]*a[3]+t[5]
|
|---|
| 1386 | mov $ai,%rax
|
|---|
| 1387 | mov $A0[1],($tptr,$j) # t[4]
|
|---|
| 1388 | mov %rdx,$A1[1]
|
|---|
| 1389 | adc \$0,$A1[1]
|
|---|
| 1390 |
|
|---|
| 1391 | mul $a0 # a[5]*a[2]
|
|---|
| 1392 | add %rax,$A0[0] # a[5]*a[2]+a[4]*a[3]+t[5]
|
|---|
| 1393 | mov $ai,%rax
|
|---|
| 1394 | mov 16($aptr,$j),$ai # a[6]
|
|---|
| 1395 | mov %rdx,$A0[1]
|
|---|
| 1396 | adc \$0,$A0[1]
|
|---|
| 1397 | add $A1[0],$A0[0]
|
|---|
| 1398 | adc \$0,$A0[1]
|
|---|
| 1399 |
|
|---|
| 1400 | mul $a1 # a[5]*a[3]
|
|---|
| 1401 | add %rax,$A1[1] # a[5]*a[3]+t[6]
|
|---|
| 1402 | mov $ai,%rax
|
|---|
| 1403 | mov $A0[0],8($tptr,$j) # t[5]
|
|---|
| 1404 | mov %rdx,$A1[0]
|
|---|
| 1405 | adc \$0,$A1[0]
|
|---|
| 1406 |
|
|---|
| 1407 | mul $a0 # a[6]*a[2]
|
|---|
| 1408 | add %rax,$A0[1] # a[6]*a[2]+a[5]*a[3]+t[6]
|
|---|
| 1409 | mov $ai,%rax # a[3]
|
|---|
| 1410 | mov 24($aptr,$j),$ai # a[7]
|
|---|
| 1411 | mov %rdx,$A0[0]
|
|---|
| 1412 | adc \$0,$A0[0]
|
|---|
| 1413 | add $A1[1],$A0[1]
|
|---|
| 1414 | adc \$0,$A0[0]
|
|---|
| 1415 |
|
|---|
| 1416 |
|
|---|
| 1417 | mul $a1 # a[6]*a[5]
|
|---|
| 1418 | add %rax,$A1[0] # a[6]*a[5]+t[7]
|
|---|
| 1419 | mov $ai,%rax
|
|---|
| 1420 | mov $A0[1],16($tptr,$j) # t[6]
|
|---|
| 1421 | mov %rdx,$A1[1]
|
|---|
| 1422 | adc \$0,$A1[1]
|
|---|
| 1423 | lea 32($j),$j
|
|---|
| 1424 |
|
|---|
| 1425 | mul $a0 # a[7]*a[4]
|
|---|
| 1426 | add %rax,$A0[0] # a[7]*a[4]+a[6]*a[5]+t[6]
|
|---|
| 1427 | mov $ai,%rax
|
|---|
| 1428 | mov %rdx,$A0[1]
|
|---|
| 1429 | adc \$0,$A0[1]
|
|---|
| 1430 | add $A1[0],$A0[0]
|
|---|
| 1431 | adc \$0,$A0[1]
|
|---|
| 1432 | mov $A0[0],-8($tptr,$j) # t[7]
|
|---|
| 1433 |
|
|---|
| 1434 | cmp \$0,$j
|
|---|
| 1435 | jne .Lsqr4x_1st
|
|---|
| 1436 |
|
|---|
| 1437 | mul $a1 # a[7]*a[5]
|
|---|
| 1438 | add %rax,$A1[1]
|
|---|
| 1439 | lea 16($i),$i
|
|---|
| 1440 | adc \$0,%rdx
|
|---|
| 1441 | add $A0[1],$A1[1]
|
|---|
| 1442 | adc \$0,%rdx
|
|---|
| 1443 |
|
|---|
| 1444 | mov $A1[1],($tptr) # t[8]
|
|---|
| 1445 | mov %rdx,$A1[0]
|
|---|
| 1446 | mov %rdx,8($tptr) # t[9]
|
|---|
| 1447 | jmp .Lsqr4x_outer
|
|---|
| 1448 |
|
|---|
| 1449 | .align 32
|
|---|
| 1450 | .Lsqr4x_outer: # comments apply to $num==6 case
|
|---|
| 1451 | mov -32($aptr,$i),$a0 # a[0]
|
|---|
| 1452 | lea 48+8(%rsp,$num,2),$tptr # end of tp[] buffer, &tp[2*$num]
|
|---|
| 1453 | mov -24($aptr,$i),%rax # a[1]
|
|---|
| 1454 | lea -32($tptr,$i),$tptr # end of tp[] window, &tp[2*$num-"$i"]
|
|---|
| 1455 | mov -16($aptr,$i),$ai # a[2]
|
|---|
| 1456 | mov %rax,$a1
|
|---|
| 1457 |
|
|---|
| 1458 | mul $a0 # a[1]*a[0]
|
|---|
| 1459 | mov -24($tptr,$i),$A0[0] # t[1]
|
|---|
| 1460 | add %rax,$A0[0] # a[1]*a[0]+t[1]
|
|---|
| 1461 | mov $ai,%rax # a[2]
|
|---|
| 1462 | adc \$0,%rdx
|
|---|
| 1463 | mov $A0[0],-24($tptr,$i) # t[1]
|
|---|
| 1464 | mov %rdx,$A0[1]
|
|---|
| 1465 |
|
|---|
| 1466 | mul $a0 # a[2]*a[0]
|
|---|
| 1467 | add %rax,$A0[1]
|
|---|
| 1468 | mov $ai,%rax
|
|---|
| 1469 | adc \$0,%rdx
|
|---|
| 1470 | add -16($tptr,$i),$A0[1] # a[2]*a[0]+t[2]
|
|---|
| 1471 | mov %rdx,$A0[0]
|
|---|
| 1472 | adc \$0,$A0[0]
|
|---|
| 1473 | mov $A0[1],-16($tptr,$i) # t[2]
|
|---|
| 1474 |
|
|---|
| 1475 | xor $A1[0],$A1[0]
|
|---|
| 1476 |
|
|---|
| 1477 | mov -8($aptr,$i),$ai # a[3]
|
|---|
| 1478 | mul $a1 # a[2]*a[1]
|
|---|
| 1479 | add %rax,$A1[0] # a[2]*a[1]+t[3]
|
|---|
| 1480 | mov $ai,%rax
|
|---|
| 1481 | adc \$0,%rdx
|
|---|
| 1482 | add -8($tptr,$i),$A1[0]
|
|---|
| 1483 | mov %rdx,$A1[1]
|
|---|
| 1484 | adc \$0,$A1[1]
|
|---|
| 1485 |
|
|---|
| 1486 | mul $a0 # a[3]*a[0]
|
|---|
| 1487 | add %rax,$A0[0] # a[3]*a[0]+a[2]*a[1]+t[3]
|
|---|
| 1488 | mov $ai,%rax
|
|---|
| 1489 | adc \$0,%rdx
|
|---|
| 1490 | add $A1[0],$A0[0]
|
|---|
| 1491 | mov %rdx,$A0[1]
|
|---|
| 1492 | adc \$0,$A0[1]
|
|---|
| 1493 | mov $A0[0],-8($tptr,$i) # t[3]
|
|---|
| 1494 |
|
|---|
| 1495 | lea ($i),$j
|
|---|
| 1496 | jmp .Lsqr4x_inner
|
|---|
| 1497 |
|
|---|
| 1498 | .align 32
|
|---|
| 1499 | .Lsqr4x_inner:
|
|---|
| 1500 | mov ($aptr,$j),$ai # a[4]
|
|---|
| 1501 | mul $a1 # a[3]*a[1]
|
|---|
| 1502 | add %rax,$A1[1] # a[3]*a[1]+t[4]
|
|---|
| 1503 | mov $ai,%rax
|
|---|
| 1504 | mov %rdx,$A1[0]
|
|---|
| 1505 | adc \$0,$A1[0]
|
|---|
| 1506 | add ($tptr,$j),$A1[1]
|
|---|
| 1507 | adc \$0,$A1[0]
|
|---|
| 1508 |
|
|---|
| 1509 | .byte 0x67
|
|---|
| 1510 | mul $a0 # a[4]*a[0]
|
|---|
| 1511 | add %rax,$A0[1] # a[4]*a[0]+a[3]*a[1]+t[4]
|
|---|
| 1512 | mov $ai,%rax # a[3]
|
|---|
| 1513 | mov 8($aptr,$j),$ai # a[5]
|
|---|
| 1514 | mov %rdx,$A0[0]
|
|---|
| 1515 | adc \$0,$A0[0]
|
|---|
| 1516 | add $A1[1],$A0[1]
|
|---|
| 1517 | adc \$0,$A0[0]
|
|---|
| 1518 |
|
|---|
| 1519 | mul $a1 # a[4]*a[3]
|
|---|
| 1520 | add %rax,$A1[0] # a[4]*a[3]+t[5]
|
|---|
| 1521 | mov $A0[1],($tptr,$j) # t[4]
|
|---|
| 1522 | mov $ai,%rax
|
|---|
| 1523 | mov %rdx,$A1[1]
|
|---|
| 1524 | adc \$0,$A1[1]
|
|---|
| 1525 | add 8($tptr,$j),$A1[0]
|
|---|
| 1526 | lea 16($j),$j # j++
|
|---|
| 1527 | adc \$0,$A1[1]
|
|---|
| 1528 |
|
|---|
| 1529 | mul $a0 # a[5]*a[2]
|
|---|
| 1530 | add %rax,$A0[0] # a[5]*a[2]+a[4]*a[3]+t[5]
|
|---|
| 1531 | mov $ai,%rax
|
|---|
| 1532 | adc \$0,%rdx
|
|---|
| 1533 | add $A1[0],$A0[0]
|
|---|
| 1534 | mov %rdx,$A0[1]
|
|---|
| 1535 | adc \$0,$A0[1]
|
|---|
| 1536 | mov $A0[0],-8($tptr,$j) # t[5], "preloaded t[1]" below
|
|---|
| 1537 |
|
|---|
| 1538 | cmp \$0,$j
|
|---|
| 1539 | jne .Lsqr4x_inner
|
|---|
| 1540 |
|
|---|
| 1541 | .byte 0x67
|
|---|
| 1542 | mul $a1 # a[5]*a[3]
|
|---|
| 1543 | add %rax,$A1[1]
|
|---|
| 1544 | adc \$0,%rdx
|
|---|
| 1545 | add $A0[1],$A1[1]
|
|---|
| 1546 | adc \$0,%rdx
|
|---|
| 1547 |
|
|---|
| 1548 | mov $A1[1],($tptr) # t[6], "preloaded t[2]" below
|
|---|
| 1549 | mov %rdx,$A1[0]
|
|---|
| 1550 | mov %rdx,8($tptr) # t[7], "preloaded t[3]" below
|
|---|
| 1551 |
|
|---|
| 1552 | add \$16,$i
|
|---|
| 1553 | jnz .Lsqr4x_outer
|
|---|
| 1554 |
|
|---|
| 1555 | # comments apply to $num==4 case
|
|---|
| 1556 | mov -32($aptr),$a0 # a[0]
|
|---|
| 1557 | lea 48+8(%rsp,$num,2),$tptr # end of tp[] buffer, &tp[2*$num]
|
|---|
| 1558 | mov -24($aptr),%rax # a[1]
|
|---|
| 1559 | lea -32($tptr,$i),$tptr # end of tp[] window, &tp[2*$num-"$i"]
|
|---|
| 1560 | mov -16($aptr),$ai # a[2]
|
|---|
| 1561 | mov %rax,$a1
|
|---|
| 1562 |
|
|---|
| 1563 | mul $a0 # a[1]*a[0]
|
|---|
| 1564 | add %rax,$A0[0] # a[1]*a[0]+t[1], preloaded t[1]
|
|---|
| 1565 | mov $ai,%rax # a[2]
|
|---|
| 1566 | mov %rdx,$A0[1]
|
|---|
| 1567 | adc \$0,$A0[1]
|
|---|
| 1568 |
|
|---|
| 1569 | mul $a0 # a[2]*a[0]
|
|---|
| 1570 | add %rax,$A0[1]
|
|---|
| 1571 | mov $ai,%rax
|
|---|
| 1572 | mov $A0[0],-24($tptr) # t[1]
|
|---|
| 1573 | mov %rdx,$A0[0]
|
|---|
| 1574 | adc \$0,$A0[0]
|
|---|
| 1575 | add $A1[1],$A0[1] # a[2]*a[0]+t[2], preloaded t[2]
|
|---|
| 1576 | mov -8($aptr),$ai # a[3]
|
|---|
| 1577 | adc \$0,$A0[0]
|
|---|
| 1578 |
|
|---|
| 1579 | mul $a1 # a[2]*a[1]
|
|---|
| 1580 | add %rax,$A1[0] # a[2]*a[1]+t[3], preloaded t[3]
|
|---|
| 1581 | mov $ai,%rax
|
|---|
| 1582 | mov $A0[1],-16($tptr) # t[2]
|
|---|
| 1583 | mov %rdx,$A1[1]
|
|---|
| 1584 | adc \$0,$A1[1]
|
|---|
| 1585 |
|
|---|
| 1586 | mul $a0 # a[3]*a[0]
|
|---|
| 1587 | add %rax,$A0[0] # a[3]*a[0]+a[2]*a[1]+t[3]
|
|---|
| 1588 | mov $ai,%rax
|
|---|
| 1589 | mov %rdx,$A0[1]
|
|---|
| 1590 | adc \$0,$A0[1]
|
|---|
| 1591 | add $A1[0],$A0[0]
|
|---|
| 1592 | adc \$0,$A0[1]
|
|---|
| 1593 | mov $A0[0],-8($tptr) # t[3]
|
|---|
| 1594 |
|
|---|
| 1595 | mul $a1 # a[3]*a[1]
|
|---|
| 1596 | add %rax,$A1[1]
|
|---|
| 1597 | mov -16($aptr),%rax # a[2]
|
|---|
| 1598 | adc \$0,%rdx
|
|---|
| 1599 | add $A0[1],$A1[1]
|
|---|
| 1600 | adc \$0,%rdx
|
|---|
| 1601 |
|
|---|
| 1602 | mov $A1[1],($tptr) # t[4]
|
|---|
| 1603 | mov %rdx,$A1[0]
|
|---|
| 1604 | mov %rdx,8($tptr) # t[5]
|
|---|
| 1605 |
|
|---|
| 1606 | mul $ai # a[2]*a[3]
|
|---|
| 1607 | ___
|
|---|
| 1608 | {
|
|---|
| 1609 | my ($shift,$carry)=($a0,$a1);
|
|---|
| 1610 | my @S=(@A1,$ai,$n0);
|
|---|
| 1611 | $code.=<<___;
|
|---|
| 1612 | add \$16,$i
|
|---|
| 1613 | xor $shift,$shift
|
|---|
| 1614 | sub $num,$i # $i=16-$num
|
|---|
| 1615 | xor $carry,$carry
|
|---|
| 1616 |
|
|---|
| 1617 | add $A1[0],%rax # t[5]
|
|---|
| 1618 | adc \$0,%rdx
|
|---|
| 1619 | mov %rax,8($tptr) # t[5]
|
|---|
| 1620 | mov %rdx,16($tptr) # t[6]
|
|---|
| 1621 | mov $carry,24($tptr) # t[7]
|
|---|
| 1622 |
|
|---|
| 1623 | mov -16($aptr,$i),%rax # a[0]
|
|---|
| 1624 | lea 48+8(%rsp),$tptr
|
|---|
| 1625 | xor $A0[0],$A0[0] # t[0]
|
|---|
| 1626 | mov 8($tptr),$A0[1] # t[1]
|
|---|
| 1627 |
|
|---|
| 1628 | lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
|
|---|
| 1629 | shr \$63,$A0[0]
|
|---|
| 1630 | lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
|
|---|
| 1631 | shr \$63,$A0[1]
|
|---|
| 1632 | or $A0[0],$S[1] # | t[2*i]>>63
|
|---|
| 1633 | mov 16($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1634 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1635 | mul %rax # a[i]*a[i]
|
|---|
| 1636 | neg $carry # mov $carry,cf
|
|---|
| 1637 | mov 24($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1638 | adc %rax,$S[0]
|
|---|
| 1639 | mov -8($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1640 | mov $S[0],($tptr)
|
|---|
| 1641 | adc %rdx,$S[1]
|
|---|
| 1642 |
|
|---|
| 1643 | lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1 | shift
|
|---|
| 1644 | mov $S[1],8($tptr)
|
|---|
| 1645 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1646 | shr \$63,$A0[0]
|
|---|
| 1647 | lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
|
|---|
| 1648 | shr \$63,$A0[1]
|
|---|
| 1649 | or $A0[0],$S[3] # | t[2*i]>>63
|
|---|
| 1650 | mov 32($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1651 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1652 | mul %rax # a[i]*a[i]
|
|---|
| 1653 | neg $carry # mov $carry,cf
|
|---|
| 1654 | mov 40($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1655 | adc %rax,$S[2]
|
|---|
| 1656 | mov 0($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1657 | mov $S[2],16($tptr)
|
|---|
| 1658 | adc %rdx,$S[3]
|
|---|
| 1659 | lea 16($i),$i
|
|---|
| 1660 | mov $S[3],24($tptr)
|
|---|
| 1661 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1662 | lea 64($tptr),$tptr
|
|---|
| 1663 | jmp .Lsqr4x_shift_n_add
|
|---|
| 1664 |
|
|---|
| 1665 | .align 32
|
|---|
| 1666 | .Lsqr4x_shift_n_add:
|
|---|
| 1667 | lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
|
|---|
| 1668 | shr \$63,$A0[0]
|
|---|
| 1669 | lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
|
|---|
| 1670 | shr \$63,$A0[1]
|
|---|
| 1671 | or $A0[0],$S[1] # | t[2*i]>>63
|
|---|
| 1672 | mov -16($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1673 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1674 | mul %rax # a[i]*a[i]
|
|---|
| 1675 | neg $carry # mov $carry,cf
|
|---|
| 1676 | mov -8($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1677 | adc %rax,$S[0]
|
|---|
| 1678 | mov -8($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1679 | mov $S[0],-32($tptr)
|
|---|
| 1680 | adc %rdx,$S[1]
|
|---|
| 1681 |
|
|---|
| 1682 | lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1 | shift
|
|---|
| 1683 | mov $S[1],-24($tptr)
|
|---|
| 1684 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1685 | shr \$63,$A0[0]
|
|---|
| 1686 | lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
|
|---|
| 1687 | shr \$63,$A0[1]
|
|---|
| 1688 | or $A0[0],$S[3] # | t[2*i]>>63
|
|---|
| 1689 | mov 0($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1690 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1691 | mul %rax # a[i]*a[i]
|
|---|
| 1692 | neg $carry # mov $carry,cf
|
|---|
| 1693 | mov 8($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1694 | adc %rax,$S[2]
|
|---|
| 1695 | mov 0($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1696 | mov $S[2],-16($tptr)
|
|---|
| 1697 | adc %rdx,$S[3]
|
|---|
| 1698 |
|
|---|
| 1699 | lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
|
|---|
| 1700 | mov $S[3],-8($tptr)
|
|---|
| 1701 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1702 | shr \$63,$A0[0]
|
|---|
| 1703 | lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
|
|---|
| 1704 | shr \$63,$A0[1]
|
|---|
| 1705 | or $A0[0],$S[1] # | t[2*i]>>63
|
|---|
| 1706 | mov 16($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1707 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1708 | mul %rax # a[i]*a[i]
|
|---|
| 1709 | neg $carry # mov $carry,cf
|
|---|
| 1710 | mov 24($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1711 | adc %rax,$S[0]
|
|---|
| 1712 | mov 8($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1713 | mov $S[0],0($tptr)
|
|---|
| 1714 | adc %rdx,$S[1]
|
|---|
| 1715 |
|
|---|
| 1716 | lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1 | shift
|
|---|
| 1717 | mov $S[1],8($tptr)
|
|---|
| 1718 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1719 | shr \$63,$A0[0]
|
|---|
| 1720 | lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
|
|---|
| 1721 | shr \$63,$A0[1]
|
|---|
| 1722 | or $A0[0],$S[3] # | t[2*i]>>63
|
|---|
| 1723 | mov 32($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1724 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1725 | mul %rax # a[i]*a[i]
|
|---|
| 1726 | neg $carry # mov $carry,cf
|
|---|
| 1727 | mov 40($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1728 | adc %rax,$S[2]
|
|---|
| 1729 | mov 16($aptr,$i),%rax # a[i+1] # prefetch
|
|---|
| 1730 | mov $S[2],16($tptr)
|
|---|
| 1731 | adc %rdx,$S[3]
|
|---|
| 1732 | mov $S[3],24($tptr)
|
|---|
| 1733 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1734 | lea 64($tptr),$tptr
|
|---|
| 1735 | add \$32,$i
|
|---|
| 1736 | jnz .Lsqr4x_shift_n_add
|
|---|
| 1737 |
|
|---|
| 1738 | lea ($shift,$A0[0],2),$S[0] # t[2*i]<<1 | shift
|
|---|
| 1739 | .byte 0x67
|
|---|
| 1740 | shr \$63,$A0[0]
|
|---|
| 1741 | lea ($j,$A0[1],2),$S[1] # t[2*i+1]<<1 |
|
|---|
| 1742 | shr \$63,$A0[1]
|
|---|
| 1743 | or $A0[0],$S[1] # | t[2*i]>>63
|
|---|
| 1744 | mov -16($tptr),$A0[0] # t[2*i+2] # prefetch
|
|---|
| 1745 | mov $A0[1],$shift # shift=t[2*i+1]>>63
|
|---|
| 1746 | mul %rax # a[i]*a[i]
|
|---|
| 1747 | neg $carry # mov $carry,cf
|
|---|
| 1748 | mov -8($tptr),$A0[1] # t[2*i+2+1] # prefetch
|
|---|
| 1749 | adc %rax,$S[0]
|
|---|
| 1750 | mov -8($aptr),%rax # a[i+1] # prefetch
|
|---|
| 1751 | mov $S[0],-32($tptr)
|
|---|
| 1752 | adc %rdx,$S[1]
|
|---|
| 1753 |
|
|---|
| 1754 | lea ($shift,$A0[0],2),$S[2] # t[2*i]<<1|shift
|
|---|
| 1755 | mov $S[1],-24($tptr)
|
|---|
| 1756 | sbb $carry,$carry # mov cf,$carry
|
|---|
| 1757 | shr \$63,$A0[0]
|
|---|
| 1758 | lea ($j,$A0[1],2),$S[3] # t[2*i+1]<<1 |
|
|---|
| 1759 | shr \$63,$A0[1]
|
|---|
| 1760 | or $A0[0],$S[3] # | t[2*i]>>63
|
|---|
| 1761 | mul %rax # a[i]*a[i]
|
|---|
| 1762 | neg $carry # mov $carry,cf
|
|---|
| 1763 | adc %rax,$S[2]
|
|---|
| 1764 | adc %rdx,$S[3]
|
|---|
| 1765 | mov $S[2],-16($tptr)
|
|---|
| 1766 | mov $S[3],-8($tptr)
|
|---|
| 1767 | ___
|
|---|
| 1768 | } |
|---|
| 1769 |
|
|---|
| 1770 | ######################################################################
|
|---|
| 1771 | # Montgomery reduction part, "word-by-word" algorithm.
|
|---|
| 1772 | #
|
|---|
| 1773 | # This new path is inspired by multiple submissions from Intel, by
|
|---|
| 1774 | # Shay Gueron, Vlad Krasnov, Erdinc Ozturk, James Guilford,
|
|---|
| 1775 | # Vinodh Gopal...
|
|---|
| 1776 | {
|
|---|
| 1777 | my ($nptr,$tptr,$carry,$m0)=("%rbp","%rdi","%rsi","%rbx");
|
|---|
| 1778 |
|
|---|
| 1779 | $code.=<<___;
|
|---|
| 1780 | movq %xmm2,$nptr
|
|---|
| 1781 | __bn_sqr8x_reduction:
|
|---|
| 1782 | xor %rax,%rax
|
|---|
| 1783 | lea ($nptr,$num),%rcx # end of n[]
|
|---|
| 1784 | lea 48+8(%rsp,$num,2),%rdx # end of t[] buffer
|
|---|
| 1785 | mov %rcx,0+8(%rsp)
|
|---|
| 1786 | lea 48+8(%rsp,$num),$tptr # end of initial t[] window
|
|---|
| 1787 | mov %rdx,8+8(%rsp)
|
|---|
| 1788 | neg $num
|
|---|
| 1789 | jmp .L8x_reduction_loop
|
|---|
| 1790 |
|
|---|
| 1791 | .align 32
|
|---|
| 1792 | .L8x_reduction_loop:
|
|---|
| 1793 | lea ($tptr,$num),$tptr # start of current t[] window
|
|---|
| 1794 | .byte 0x66
|
|---|
| 1795 | mov 8*0($tptr),$m0
|
|---|
| 1796 | mov 8*1($tptr),%r9
|
|---|
| 1797 | mov 8*2($tptr),%r10
|
|---|
| 1798 | mov 8*3($tptr),%r11
|
|---|
| 1799 | mov 8*4($tptr),%r12
|
|---|
| 1800 | mov 8*5($tptr),%r13
|
|---|
| 1801 | mov 8*6($tptr),%r14
|
|---|
| 1802 | mov 8*7($tptr),%r15
|
|---|
| 1803 | mov %rax,(%rdx) # store top-most carry bit
|
|---|
| 1804 | lea 8*8($tptr),$tptr
|
|---|
| 1805 |
|
|---|
| 1806 | .byte 0x67
|
|---|
| 1807 | mov $m0,%r8
|
|---|
| 1808 | imulq 32+8(%rsp),$m0 # n0*a[0]
|
|---|
| 1809 | mov 8*0($nptr),%rax # n[0]
|
|---|
| 1810 | mov \$8,%ecx
|
|---|
| 1811 | jmp .L8x_reduce
|
|---|
| 1812 |
|
|---|
| 1813 | .align 32
|
|---|
| 1814 | .L8x_reduce:
|
|---|
| 1815 | mulq $m0
|
|---|
| 1816 | mov 8*1($nptr),%rax # n[1]
|
|---|
| 1817 | neg %r8
|
|---|
| 1818 | mov %rdx,%r8
|
|---|
| 1819 | adc \$0,%r8
|
|---|
| 1820 |
|
|---|
| 1821 | mulq $m0
|
|---|
| 1822 | add %rax,%r9
|
|---|
| 1823 | mov 8*2($nptr),%rax
|
|---|
| 1824 | adc \$0,%rdx
|
|---|
| 1825 | add %r9,%r8
|
|---|
| 1826 | mov $m0,48-8+8(%rsp,%rcx,8) # put aside n0*a[i]
|
|---|
| 1827 | mov %rdx,%r9
|
|---|
| 1828 | adc \$0,%r9
|
|---|
| 1829 |
|
|---|
| 1830 | mulq $m0
|
|---|
| 1831 | add %rax,%r10
|
|---|
| 1832 | mov 8*3($nptr),%rax
|
|---|
| 1833 | adc \$0,%rdx
|
|---|
| 1834 | add %r10,%r9
|
|---|
| 1835 | mov 32+8(%rsp),$carry # pull n0, borrow $carry
|
|---|
| 1836 | mov %rdx,%r10
|
|---|
| 1837 | adc \$0,%r10
|
|---|
| 1838 |
|
|---|
| 1839 | mulq $m0
|
|---|
| 1840 | add %rax,%r11
|
|---|
| 1841 | mov 8*4($nptr),%rax
|
|---|
| 1842 | adc \$0,%rdx
|
|---|
| 1843 | imulq %r8,$carry # modulo-scheduled
|
|---|
| 1844 | add %r11,%r10
|
|---|
| 1845 | mov %rdx,%r11
|
|---|
| 1846 | adc \$0,%r11
|
|---|
| 1847 |
|
|---|
| 1848 | mulq $m0
|
|---|
| 1849 | add %rax,%r12
|
|---|
| 1850 | mov 8*5($nptr),%rax
|
|---|
| 1851 | adc \$0,%rdx
|
|---|
| 1852 | add %r12,%r11
|
|---|
| 1853 | mov %rdx,%r12
|
|---|
| 1854 | adc \$0,%r12
|
|---|
| 1855 |
|
|---|
| 1856 | mulq $m0
|
|---|
| 1857 | add %rax,%r13
|
|---|
| 1858 | mov 8*6($nptr),%rax
|
|---|
| 1859 | adc \$0,%rdx
|
|---|
| 1860 | add %r13,%r12
|
|---|
| 1861 | mov %rdx,%r13
|
|---|
| 1862 | adc \$0,%r13
|
|---|
| 1863 |
|
|---|
| 1864 | mulq $m0
|
|---|
| 1865 | add %rax,%r14
|
|---|
| 1866 | mov 8*7($nptr),%rax
|
|---|
| 1867 | adc \$0,%rdx
|
|---|
| 1868 | add %r14,%r13
|
|---|
| 1869 | mov %rdx,%r14
|
|---|
| 1870 | adc \$0,%r14
|
|---|
| 1871 |
|
|---|
| 1872 | mulq $m0
|
|---|
| 1873 | mov $carry,$m0 # n0*a[i]
|
|---|
| 1874 | add %rax,%r15
|
|---|
| 1875 | mov 8*0($nptr),%rax # n[0]
|
|---|
| 1876 | adc \$0,%rdx
|
|---|
| 1877 | add %r15,%r14
|
|---|
| 1878 | mov %rdx,%r15
|
|---|
| 1879 | adc \$0,%r15
|
|---|
| 1880 |
|
|---|
| 1881 | dec %ecx
|
|---|
| 1882 | jnz .L8x_reduce
|
|---|
| 1883 |
|
|---|
| 1884 | lea 8*8($nptr),$nptr
|
|---|
| 1885 | xor %rax,%rax
|
|---|
| 1886 | mov 8+8(%rsp),%rdx # pull end of t[]
|
|---|
| 1887 | cmp 0+8(%rsp),$nptr # end of n[]?
|
|---|
| 1888 | jae .L8x_no_tail
|
|---|
| 1889 |
|
|---|
| 1890 | .byte 0x66
|
|---|
| 1891 | add 8*0($tptr),%r8
|
|---|
| 1892 | adc 8*1($tptr),%r9
|
|---|
| 1893 | adc 8*2($tptr),%r10
|
|---|
| 1894 | adc 8*3($tptr),%r11
|
|---|
| 1895 | adc 8*4($tptr),%r12
|
|---|
| 1896 | adc 8*5($tptr),%r13
|
|---|
| 1897 | adc 8*6($tptr),%r14
|
|---|
| 1898 | adc 8*7($tptr),%r15
|
|---|
| 1899 | sbb $carry,$carry # top carry
|
|---|
| 1900 |
|
|---|
| 1901 | mov 48+56+8(%rsp),$m0 # pull n0*a[0]
|
|---|
| 1902 | mov \$8,%ecx
|
|---|
| 1903 | mov 8*0($nptr),%rax
|
|---|
| 1904 | jmp .L8x_tail
|
|---|
| 1905 |
|
|---|
| 1906 | .align 32
|
|---|
| 1907 | .L8x_tail:
|
|---|
| 1908 | mulq $m0
|
|---|
| 1909 | add %rax,%r8
|
|---|
| 1910 | mov 8*1($nptr),%rax
|
|---|
| 1911 | mov %r8,($tptr) # save result
|
|---|
| 1912 | mov %rdx,%r8
|
|---|
| 1913 | adc \$0,%r8
|
|---|
| 1914 |
|
|---|
| 1915 | mulq $m0
|
|---|
| 1916 | add %rax,%r9
|
|---|
| 1917 | mov 8*2($nptr),%rax
|
|---|
| 1918 | adc \$0,%rdx
|
|---|
| 1919 | add %r9,%r8
|
|---|
| 1920 | lea 8($tptr),$tptr # $tptr++
|
|---|
| 1921 | mov %rdx,%r9
|
|---|
| 1922 | adc \$0,%r9
|
|---|
| 1923 |
|
|---|
| 1924 | mulq $m0
|
|---|
| 1925 | add %rax,%r10
|
|---|
| 1926 | mov 8*3($nptr),%rax
|
|---|
| 1927 | adc \$0,%rdx
|
|---|
| 1928 | add %r10,%r9
|
|---|
| 1929 | mov %rdx,%r10
|
|---|
| 1930 | adc \$0,%r10
|
|---|
| 1931 |
|
|---|
| 1932 | mulq $m0
|
|---|
| 1933 | add %rax,%r11
|
|---|
| 1934 | mov 8*4($nptr),%rax
|
|---|
| 1935 | adc \$0,%rdx
|
|---|
| 1936 | add %r11,%r10
|
|---|
| 1937 | mov %rdx,%r11
|
|---|
| 1938 | adc \$0,%r11
|
|---|
| 1939 |
|
|---|
| 1940 | mulq $m0
|
|---|
| 1941 | add %rax,%r12
|
|---|
| 1942 | mov 8*5($nptr),%rax
|
|---|
| 1943 | adc \$0,%rdx
|
|---|
| 1944 | add %r12,%r11
|
|---|
| 1945 | mov %rdx,%r12
|
|---|
| 1946 | adc \$0,%r12
|
|---|
| 1947 |
|
|---|
| 1948 | mulq $m0
|
|---|
| 1949 | add %rax,%r13
|
|---|
| 1950 | mov 8*6($nptr),%rax
|
|---|
| 1951 | adc \$0,%rdx
|
|---|
| 1952 | add %r13,%r12
|
|---|
| 1953 | mov %rdx,%r13
|
|---|
| 1954 | adc \$0,%r13
|
|---|
| 1955 |
|
|---|
| 1956 | mulq $m0
|
|---|
| 1957 | add %rax,%r14
|
|---|
| 1958 | mov 8*7($nptr),%rax
|
|---|
| 1959 | adc \$0,%rdx
|
|---|
| 1960 | add %r14,%r13
|
|---|
| 1961 | mov %rdx,%r14
|
|---|
| 1962 | adc \$0,%r14
|
|---|
| 1963 |
|
|---|
| 1964 | mulq $m0
|
|---|
| 1965 | mov 48-16+8(%rsp,%rcx,8),$m0# pull n0*a[i]
|
|---|
| 1966 | add %rax,%r15
|
|---|
| 1967 | adc \$0,%rdx
|
|---|
| 1968 | add %r15,%r14
|
|---|
| 1969 | mov 8*0($nptr),%rax # pull n[0]
|
|---|
| 1970 | mov %rdx,%r15
|
|---|
| 1971 | adc \$0,%r15
|
|---|
| 1972 |
|
|---|
| 1973 | dec %ecx
|
|---|
| 1974 | jnz .L8x_tail
|
|---|
| 1975 |
|
|---|
| 1976 | lea 8*8($nptr),$nptr
|
|---|
| 1977 | mov 8+8(%rsp),%rdx # pull end of t[]
|
|---|
| 1978 | cmp 0+8(%rsp),$nptr # end of n[]?
|
|---|
| 1979 | jae .L8x_tail_done # break out of loop
|
|---|
| 1980 |
|
|---|
| 1981 | mov 48+56+8(%rsp),$m0 # pull n0*a[0]
|
|---|
| 1982 | neg $carry
|
|---|
| 1983 | mov 8*0($nptr),%rax # pull n[0]
|
|---|
| 1984 | adc 8*0($tptr),%r8
|
|---|
| 1985 | adc 8*1($tptr),%r9
|
|---|
| 1986 | adc 8*2($tptr),%r10
|
|---|
| 1987 | adc 8*3($tptr),%r11
|
|---|
| 1988 | adc 8*4($tptr),%r12
|
|---|
| 1989 | adc 8*5($tptr),%r13
|
|---|
| 1990 | adc 8*6($tptr),%r14
|
|---|
| 1991 | adc 8*7($tptr),%r15
|
|---|
| 1992 | sbb $carry,$carry # top carry
|
|---|
| 1993 |
|
|---|
| 1994 | mov \$8,%ecx
|
|---|
| 1995 | jmp .L8x_tail
|
|---|
| 1996 |
|
|---|
| 1997 | .align 32
|
|---|
| 1998 | .L8x_tail_done:
|
|---|
| 1999 | xor %rax,%rax
|
|---|
| 2000 | add (%rdx),%r8 # can this overflow?
|
|---|
| 2001 | adc \$0,%r9
|
|---|
| 2002 | adc \$0,%r10
|
|---|
| 2003 | adc \$0,%r11
|
|---|
| 2004 | adc \$0,%r12
|
|---|
| 2005 | adc \$0,%r13
|
|---|
| 2006 | adc \$0,%r14
|
|---|
| 2007 | adc \$0,%r15
|
|---|
| 2008 | adc \$0,%rax
|
|---|
| 2009 |
|
|---|
| 2010 | neg $carry
|
|---|
| 2011 | .L8x_no_tail:
|
|---|
| 2012 | adc 8*0($tptr),%r8
|
|---|
| 2013 | adc 8*1($tptr),%r9
|
|---|
| 2014 | adc 8*2($tptr),%r10
|
|---|
| 2015 | adc 8*3($tptr),%r11
|
|---|
| 2016 | adc 8*4($tptr),%r12
|
|---|
| 2017 | adc 8*5($tptr),%r13
|
|---|
| 2018 | adc 8*6($tptr),%r14
|
|---|
| 2019 | adc 8*7($tptr),%r15
|
|---|
| 2020 | adc \$0,%rax # top-most carry
|
|---|
| 2021 | mov -8($nptr),%rcx # np[num-1]
|
|---|
| 2022 | xor $carry,$carry
|
|---|
| 2023 |
|
|---|
| 2024 | movq %xmm2,$nptr # restore $nptr
|
|---|
| 2025 |
|
|---|
| 2026 | mov %r8,8*0($tptr) # store top 512 bits
|
|---|
| 2027 | mov %r9,8*1($tptr)
|
|---|
| 2028 | movq %xmm3,$num # $num is %r9, can't be moved upwards
|
|---|
| 2029 | mov %r10,8*2($tptr)
|
|---|
| 2030 | mov %r11,8*3($tptr)
|
|---|
| 2031 | mov %r12,8*4($tptr)
|
|---|
| 2032 | mov %r13,8*5($tptr)
|
|---|
| 2033 | mov %r14,8*6($tptr)
|
|---|
| 2034 | mov %r15,8*7($tptr)
|
|---|
| 2035 | lea 8*8($tptr),$tptr
|
|---|
| 2036 |
|
|---|
| 2037 | cmp %rdx,$tptr # end of t[]?
|
|---|
| 2038 | jb .L8x_reduction_loop
|
|---|
| 2039 | ret
|
|---|
| 2040 | .cfi_endproc
|
|---|
| 2041 | .size bn_sqr8x_internal,.-bn_sqr8x_internal
|
|---|
| 2042 | ___
|
|---|
| 2043 | } |
|---|
| 2044 |
|
|---|
| 2045 | ##############################################################
|
|---|
| 2046 | # Post-condition, 4x unrolled
|
|---|
| 2047 | #
|
|---|
| 2048 | {
|
|---|
| 2049 | my ($tptr,$nptr)=("%rbx","%rbp");
|
|---|
| 2050 | $code.=<<___;
|
|---|
| 2051 | .type __bn_post4x_internal,\@abi-omnipotent
|
|---|
| 2052 | .align 32
|
|---|
| 2053 | __bn_post4x_internal:
|
|---|
| 2054 | .cfi_startproc
|
|---|
| 2055 | mov 8*0($nptr),%r12
|
|---|
| 2056 | lea (%rdi,$num),$tptr # %rdi was $tptr above
|
|---|
| 2057 | mov $num,%rcx
|
|---|
| 2058 | movq %xmm1,$rptr # restore $rptr
|
|---|
| 2059 | neg %rax
|
|---|
| 2060 | movq %xmm1,$aptr # prepare for back-to-back call
|
|---|
| 2061 | sar \$3+2,%rcx
|
|---|
| 2062 | dec %r12 # so that after 'not' we get -n[0]
|
|---|
| 2063 | xor %r10,%r10
|
|---|
| 2064 | mov 8*1($nptr),%r13
|
|---|
| 2065 | mov 8*2($nptr),%r14
|
|---|
| 2066 | mov 8*3($nptr),%r15
|
|---|
| 2067 | jmp .Lsqr4x_sub_entry
|
|---|
| 2068 |
|
|---|
| 2069 | .align 16
|
|---|
| 2070 | .Lsqr4x_sub:
|
|---|
| 2071 | mov 8*0($nptr),%r12
|
|---|
| 2072 | mov 8*1($nptr),%r13
|
|---|
| 2073 | mov 8*2($nptr),%r14
|
|---|
| 2074 | mov 8*3($nptr),%r15
|
|---|
| 2075 | .Lsqr4x_sub_entry:
|
|---|
| 2076 | lea 8*4($nptr),$nptr
|
|---|
| 2077 | not %r12
|
|---|
| 2078 | not %r13
|
|---|
| 2079 | not %r14
|
|---|
| 2080 | not %r15
|
|---|
| 2081 | and %rax,%r12
|
|---|
| 2082 | and %rax,%r13
|
|---|
| 2083 | and %rax,%r14
|
|---|
| 2084 | and %rax,%r15
|
|---|
| 2085 |
|
|---|
| 2086 | neg %r10 # mov %r10,%cf
|
|---|
| 2087 | adc 8*0($tptr),%r12
|
|---|
| 2088 | adc 8*1($tptr),%r13
|
|---|
| 2089 | adc 8*2($tptr),%r14
|
|---|
| 2090 | adc 8*3($tptr),%r15
|
|---|
| 2091 | mov %r12,8*0($rptr)
|
|---|
| 2092 | lea 8*4($tptr),$tptr
|
|---|
| 2093 | mov %r13,8*1($rptr)
|
|---|
| 2094 | sbb %r10,%r10 # mov %cf,%r10
|
|---|
| 2095 | mov %r14,8*2($rptr)
|
|---|
| 2096 | mov %r15,8*3($rptr)
|
|---|
| 2097 | lea 8*4($rptr),$rptr
|
|---|
| 2098 |
|
|---|
| 2099 | inc %rcx # pass %cf
|
|---|
| 2100 | jnz .Lsqr4x_sub
|
|---|
| 2101 |
|
|---|
| 2102 | mov $num,%r10 # prepare for back-to-back call
|
|---|
| 2103 | neg $num # restore $num
|
|---|
| 2104 | ret
|
|---|
| 2105 | .cfi_endproc
|
|---|
| 2106 | .size __bn_post4x_internal,.-__bn_post4x_internal
|
|---|
| 2107 | ___
|
|---|
| 2108 | }
|
|---|
| 2109 | }}}
|
|---|
| 2110 | |
|---|
| 2111 |
|
|---|
| 2112 | if ($addx) {{{
|
|---|
| 2113 | my $bp="%rdx"; # restore original value
|
|---|
| 2114 |
|
|---|
| 2115 | $code.=<<___;
|
|---|
| 2116 | .type bn_mulx4x_mont_gather5,\@function,6
|
|---|
| 2117 | .align 32
|
|---|
| 2118 | bn_mulx4x_mont_gather5:
|
|---|
| 2119 | .cfi_startproc
|
|---|
| 2120 | mov %rsp,%rax
|
|---|
| 2121 | .cfi_def_cfa_register %rax
|
|---|
| 2122 | .Lmulx4x_enter:
|
|---|
| 2123 | push %rbx
|
|---|
| 2124 | .cfi_push %rbx
|
|---|
| 2125 | push %rbp
|
|---|
| 2126 | .cfi_push %rbp
|
|---|
| 2127 | push %r12
|
|---|
| 2128 | .cfi_push %r12
|
|---|
| 2129 | push %r13
|
|---|
| 2130 | .cfi_push %r13
|
|---|
| 2131 | push %r14
|
|---|
| 2132 | .cfi_push %r14
|
|---|
| 2133 | push %r15
|
|---|
| 2134 | .cfi_push %r15
|
|---|
| 2135 | .Lmulx4x_prologue:
|
|---|
| 2136 |
|
|---|
| 2137 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 2138 | lea ($num,$num,2),%r10 # 3*$num in bytes
|
|---|
| 2139 | neg $num # -$num
|
|---|
| 2140 | mov ($n0),$n0 # *n0
|
|---|
| 2141 |
|
|---|
| 2142 | ##############################################################
|
|---|
| 2143 | # Ensure that stack frame doesn't alias with $rptr+3*$num
|
|---|
| 2144 | # modulo 4096, which covers ret[num], am[num] and n[num]
|
|---|
| 2145 | # (see bn_exp.c). This is done to allow memory disambiguation
|
|---|
| 2146 | # logic do its magic. [Extra [num] is allocated in order
|
|---|
| 2147 | # to align with bn_power5's frame, which is cleansed after
|
|---|
| 2148 | # completing exponentiation. Extra 256 bytes is for power mask
|
|---|
| 2149 | # calculated from 7th argument, the index.]
|
|---|
| 2150 | #
|
|---|
| 2151 | lea -320(%rsp,$num,2),%r11
|
|---|
| 2152 | mov %rsp,%rbp
|
|---|
| 2153 | sub $rp,%r11
|
|---|
| 2154 | and \$4095,%r11
|
|---|
| 2155 | cmp %r11,%r10
|
|---|
| 2156 | jb .Lmulx4xsp_alt
|
|---|
| 2157 | sub %r11,%rbp # align with $aptr
|
|---|
| 2158 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*$num*8+256)
|
|---|
| 2159 | jmp .Lmulx4xsp_done
|
|---|
| 2160 |
|
|---|
| 2161 | .Lmulx4xsp_alt:
|
|---|
| 2162 | lea 4096-320(,$num,2),%r10
|
|---|
| 2163 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*$num*8+256)
|
|---|
| 2164 | sub %r10,%r11
|
|---|
| 2165 | mov \$0,%r10
|
|---|
| 2166 | cmovc %r10,%r11
|
|---|
| 2167 | sub %r11,%rbp
|
|---|
| 2168 | .Lmulx4xsp_done:
|
|---|
| 2169 | and \$-64,%rbp # ensure alignment
|
|---|
| 2170 | mov %rsp,%r11
|
|---|
| 2171 | sub %rbp,%r11
|
|---|
| 2172 | and \$-4096,%r11
|
|---|
| 2173 | lea (%rbp,%r11),%rsp
|
|---|
| 2174 | mov (%rsp),%r10
|
|---|
| 2175 | cmp %rbp,%rsp
|
|---|
| 2176 | ja .Lmulx4x_page_walk
|
|---|
| 2177 | jmp .Lmulx4x_page_walk_done
|
|---|
| 2178 |
|
|---|
| 2179 | .Lmulx4x_page_walk:
|
|---|
| 2180 | lea -4096(%rsp),%rsp
|
|---|
| 2181 | mov (%rsp),%r10
|
|---|
| 2182 | cmp %rbp,%rsp
|
|---|
| 2183 | ja .Lmulx4x_page_walk
|
|---|
| 2184 | .Lmulx4x_page_walk_done:
|
|---|
| 2185 |
|
|---|
| 2186 | ##############################################################
|
|---|
| 2187 | # Stack layout
|
|---|
| 2188 | # +0 -num
|
|---|
| 2189 | # +8 off-loaded &b[i]
|
|---|
| 2190 | # +16 end of b[num]
|
|---|
| 2191 | # +24 inner counter
|
|---|
| 2192 | # +32 saved n0
|
|---|
| 2193 | # +40 saved %rsp
|
|---|
| 2194 | # +48
|
|---|
| 2195 | # +56 saved rp
|
|---|
| 2196 | # +64 tmp[num+1]
|
|---|
| 2197 | #
|
|---|
| 2198 | mov $n0, 32(%rsp) # save *n0
|
|---|
| 2199 | mov %rax,40(%rsp) # save original %rsp
|
|---|
| 2200 | .cfi_cfa_expression %rsp+40,deref,+8
|
|---|
| 2201 | .Lmulx4x_body:
|
|---|
| 2202 | call mulx4x_internal
|
|---|
| 2203 |
|
|---|
| 2204 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 2205 | .cfi_def_cfa %rsi,8
|
|---|
| 2206 | mov \$1,%rax
|
|---|
| 2207 |
|
|---|
| 2208 | mov -48(%rsi),%r15
|
|---|
| 2209 | .cfi_restore %r15
|
|---|
| 2210 | mov -40(%rsi),%r14
|
|---|
| 2211 | .cfi_restore %r14
|
|---|
| 2212 | mov -32(%rsi),%r13
|
|---|
| 2213 | .cfi_restore %r13
|
|---|
| 2214 | mov -24(%rsi),%r12
|
|---|
| 2215 | .cfi_restore %r12
|
|---|
| 2216 | mov -16(%rsi),%rbp
|
|---|
| 2217 | .cfi_restore %rbp
|
|---|
| 2218 | mov -8(%rsi),%rbx
|
|---|
| 2219 | .cfi_restore %rbx
|
|---|
| 2220 | lea (%rsi),%rsp
|
|---|
| 2221 | .cfi_def_cfa_register %rsp
|
|---|
| 2222 | .Lmulx4x_epilogue:
|
|---|
| 2223 | ret
|
|---|
| 2224 | .cfi_endproc
|
|---|
| 2225 | .size bn_mulx4x_mont_gather5,.-bn_mulx4x_mont_gather5
|
|---|
| 2226 |
|
|---|
| 2227 | .type mulx4x_internal,\@abi-omnipotent
|
|---|
| 2228 | .align 32
|
|---|
| 2229 | mulx4x_internal:
|
|---|
| 2230 | .cfi_startproc
|
|---|
| 2231 | mov $num,8(%rsp) # save -$num (it was in bytes)
|
|---|
| 2232 | mov $num,%r10
|
|---|
| 2233 | neg $num # restore $num
|
|---|
| 2234 | shl \$5,$num
|
|---|
| 2235 | neg %r10 # restore $num
|
|---|
| 2236 | lea 128($bp,$num),%r13 # end of powers table (+size optimization)
|
|---|
| 2237 | shr \$5+5,$num
|
|---|
| 2238 | movd `($win64?56:8)`(%rax),%xmm5 # load 7th argument
|
|---|
| 2239 | sub \$1,$num
|
|---|
| 2240 | lea .Linc(%rip),%rax
|
|---|
| 2241 | mov %r13,16+8(%rsp) # end of b[num]
|
|---|
| 2242 | mov $num,24+8(%rsp) # inner counter
|
|---|
| 2243 | mov $rp, 56+8(%rsp) # save $rp
|
|---|
| 2244 | ___
|
|---|
| 2245 | my ($aptr, $bptr, $nptr, $tptr, $mi, $bi, $zero, $num)=
|
|---|
| 2246 | ("%rsi","%rdi","%rcx","%rbx","%r8","%r9","%rbp","%rax");
|
|---|
| 2247 | my $rptr=$bptr;
|
|---|
| 2248 | my $STRIDE=2**5*8; # 5 is "window size"
|
|---|
| 2249 | my $N=$STRIDE/4; # should match cache line size
|
|---|
| 2250 | $code.=<<___;
|
|---|
| 2251 | movdqa 0(%rax),%xmm0 # 00000001000000010000000000000000
|
|---|
| 2252 | movdqa 16(%rax),%xmm1 # 00000002000000020000000200000002
|
|---|
| 2253 | lea 88-112(%rsp,%r10),%r10 # place the mask after tp[num+1] (+ICache optimization)
|
|---|
| 2254 | lea 128($bp),$bptr # size optimization
|
|---|
| 2255 |
|
|---|
| 2256 | pshufd \$0,%xmm5,%xmm5 # broadcast index
|
|---|
| 2257 | movdqa %xmm1,%xmm4
|
|---|
| 2258 | .byte 0x67
|
|---|
| 2259 | movdqa %xmm1,%xmm2
|
|---|
| 2260 | ___
|
|---|
| 2261 | ########################################################################
|
|---|
| 2262 | # calculate mask by comparing 0..31 to index and save result to stack
|
|---|
| 2263 | #
|
|---|
| 2264 | $code.=<<___;
|
|---|
| 2265 | .byte 0x67
|
|---|
| 2266 | paddd %xmm0,%xmm1
|
|---|
| 2267 | pcmpeqd %xmm5,%xmm0 # compare to 1,0
|
|---|
| 2268 | movdqa %xmm4,%xmm3
|
|---|
| 2269 | ___
|
|---|
| 2270 | for($i=0;$i<$STRIDE/16-4;$i+=4) {
|
|---|
| 2271 | $code.=<<___;
|
|---|
| 2272 | paddd %xmm1,%xmm2
|
|---|
| 2273 | pcmpeqd %xmm5,%xmm1 # compare to 3,2
|
|---|
| 2274 | movdqa %xmm0,`16*($i+0)+112`(%r10)
|
|---|
| 2275 | movdqa %xmm4,%xmm0
|
|---|
| 2276 |
|
|---|
| 2277 | paddd %xmm2,%xmm3
|
|---|
| 2278 | pcmpeqd %xmm5,%xmm2 # compare to 5,4
|
|---|
| 2279 | movdqa %xmm1,`16*($i+1)+112`(%r10)
|
|---|
| 2280 | movdqa %xmm4,%xmm1
|
|---|
| 2281 |
|
|---|
| 2282 | paddd %xmm3,%xmm0
|
|---|
| 2283 | pcmpeqd %xmm5,%xmm3 # compare to 7,6
|
|---|
| 2284 | movdqa %xmm2,`16*($i+2)+112`(%r10)
|
|---|
| 2285 | movdqa %xmm4,%xmm2
|
|---|
| 2286 |
|
|---|
| 2287 | paddd %xmm0,%xmm1
|
|---|
| 2288 | pcmpeqd %xmm5,%xmm0
|
|---|
| 2289 | movdqa %xmm3,`16*($i+3)+112`(%r10)
|
|---|
| 2290 | movdqa %xmm4,%xmm3
|
|---|
| 2291 | ___
|
|---|
| 2292 | }
|
|---|
| 2293 | $code.=<<___; # last iteration can be optimized
|
|---|
| 2294 | .byte 0x67
|
|---|
| 2295 | paddd %xmm1,%xmm2
|
|---|
| 2296 | pcmpeqd %xmm5,%xmm1
|
|---|
| 2297 | movdqa %xmm0,`16*($i+0)+112`(%r10)
|
|---|
| 2298 |
|
|---|
| 2299 | paddd %xmm2,%xmm3
|
|---|
| 2300 | pcmpeqd %xmm5,%xmm2
|
|---|
| 2301 | movdqa %xmm1,`16*($i+1)+112`(%r10)
|
|---|
| 2302 |
|
|---|
| 2303 | pcmpeqd %xmm5,%xmm3
|
|---|
| 2304 | movdqa %xmm2,`16*($i+2)+112`(%r10)
|
|---|
| 2305 |
|
|---|
| 2306 | pand `16*($i+0)-128`($bptr),%xmm0 # while it's still in register
|
|---|
| 2307 | pand `16*($i+1)-128`($bptr),%xmm1
|
|---|
| 2308 | pand `16*($i+2)-128`($bptr),%xmm2
|
|---|
| 2309 | movdqa %xmm3,`16*($i+3)+112`(%r10)
|
|---|
| 2310 | pand `16*($i+3)-128`($bptr),%xmm3
|
|---|
| 2311 | por %xmm2,%xmm0
|
|---|
| 2312 | por %xmm3,%xmm1
|
|---|
| 2313 | ___
|
|---|
| 2314 | for($i=0;$i<$STRIDE/16-4;$i+=4) {
|
|---|
| 2315 | $code.=<<___;
|
|---|
| 2316 | movdqa `16*($i+0)-128`($bptr),%xmm4
|
|---|
| 2317 | movdqa `16*($i+1)-128`($bptr),%xmm5
|
|---|
| 2318 | movdqa `16*($i+2)-128`($bptr),%xmm2
|
|---|
| 2319 | pand `16*($i+0)+112`(%r10),%xmm4
|
|---|
| 2320 | movdqa `16*($i+3)-128`($bptr),%xmm3
|
|---|
| 2321 | pand `16*($i+1)+112`(%r10),%xmm5
|
|---|
| 2322 | por %xmm4,%xmm0
|
|---|
| 2323 | pand `16*($i+2)+112`(%r10),%xmm2
|
|---|
| 2324 | por %xmm5,%xmm1
|
|---|
| 2325 | pand `16*($i+3)+112`(%r10),%xmm3
|
|---|
| 2326 | por %xmm2,%xmm0
|
|---|
| 2327 | por %xmm3,%xmm1
|
|---|
| 2328 | ___
|
|---|
| 2329 | }
|
|---|
| 2330 | $code.=<<___;
|
|---|
| 2331 | pxor %xmm1,%xmm0
|
|---|
| 2332 | pshufd \$0x4e,%xmm0,%xmm1
|
|---|
| 2333 | por %xmm1,%xmm0
|
|---|
| 2334 | lea $STRIDE($bptr),$bptr
|
|---|
| 2335 | movq %xmm0,%rdx # bp[0]
|
|---|
| 2336 | lea 64+8*4+8(%rsp),$tptr
|
|---|
| 2337 |
|
|---|
| 2338 | mov %rdx,$bi
|
|---|
| 2339 | mulx 0*8($aptr),$mi,%rax # a[0]*b[0]
|
|---|
| 2340 | mulx 1*8($aptr),%r11,%r12 # a[1]*b[0]
|
|---|
| 2341 | add %rax,%r11
|
|---|
| 2342 | mulx 2*8($aptr),%rax,%r13 # ...
|
|---|
| 2343 | adc %rax,%r12
|
|---|
| 2344 | adc \$0,%r13
|
|---|
| 2345 | mulx 3*8($aptr),%rax,%r14
|
|---|
| 2346 |
|
|---|
| 2347 | mov $mi,%r15
|
|---|
| 2348 | imulq 32+8(%rsp),$mi # "t[0]"*n0
|
|---|
| 2349 | xor $zero,$zero # cf=0, of=0
|
|---|
| 2350 | mov $mi,%rdx
|
|---|
| 2351 |
|
|---|
| 2352 | mov $bptr,8+8(%rsp) # off-load &b[i]
|
|---|
| 2353 |
|
|---|
| 2354 | lea 4*8($aptr),$aptr
|
|---|
| 2355 | adcx %rax,%r13
|
|---|
| 2356 | adcx $zero,%r14 # cf=0
|
|---|
| 2357 |
|
|---|
| 2358 | mulx 0*8($nptr),%rax,%r10
|
|---|
| 2359 | adcx %rax,%r15 # discarded
|
|---|
| 2360 | adox %r11,%r10
|
|---|
| 2361 | mulx 1*8($nptr),%rax,%r11
|
|---|
| 2362 | adcx %rax,%r10
|
|---|
| 2363 | adox %r12,%r11
|
|---|
| 2364 | mulx 2*8($nptr),%rax,%r12
|
|---|
| 2365 | mov 24+8(%rsp),$bptr # counter value
|
|---|
| 2366 | mov %r10,-8*4($tptr)
|
|---|
| 2367 | adcx %rax,%r11
|
|---|
| 2368 | adox %r13,%r12
|
|---|
| 2369 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 2370 | mov $bi,%rdx
|
|---|
| 2371 | mov %r11,-8*3($tptr)
|
|---|
| 2372 | adcx %rax,%r12
|
|---|
| 2373 | adox $zero,%r15 # of=0
|
|---|
| 2374 | lea 4*8($nptr),$nptr
|
|---|
| 2375 | mov %r12,-8*2($tptr)
|
|---|
| 2376 | jmp .Lmulx4x_1st
|
|---|
| 2377 |
|
|---|
| 2378 | .align 32
|
|---|
| 2379 | .Lmulx4x_1st:
|
|---|
| 2380 | adcx $zero,%r15 # cf=0, modulo-scheduled
|
|---|
| 2381 | mulx 0*8($aptr),%r10,%rax # a[4]*b[0]
|
|---|
| 2382 | adcx %r14,%r10
|
|---|
| 2383 | mulx 1*8($aptr),%r11,%r14 # a[5]*b[0]
|
|---|
| 2384 | adcx %rax,%r11
|
|---|
| 2385 | mulx 2*8($aptr),%r12,%rax # ...
|
|---|
| 2386 | adcx %r14,%r12
|
|---|
| 2387 | mulx 3*8($aptr),%r13,%r14
|
|---|
| 2388 | .byte 0x67,0x67
|
|---|
| 2389 | mov $mi,%rdx
|
|---|
| 2390 | adcx %rax,%r13
|
|---|
| 2391 | adcx $zero,%r14 # cf=0
|
|---|
| 2392 | lea 4*8($aptr),$aptr
|
|---|
| 2393 | lea 4*8($tptr),$tptr
|
|---|
| 2394 |
|
|---|
| 2395 | adox %r15,%r10
|
|---|
| 2396 | mulx 0*8($nptr),%rax,%r15
|
|---|
| 2397 | adcx %rax,%r10
|
|---|
| 2398 | adox %r15,%r11
|
|---|
| 2399 | mulx 1*8($nptr),%rax,%r15
|
|---|
| 2400 | adcx %rax,%r11
|
|---|
| 2401 | adox %r15,%r12
|
|---|
| 2402 | mulx 2*8($nptr),%rax,%r15
|
|---|
| 2403 | mov %r10,-5*8($tptr)
|
|---|
| 2404 | adcx %rax,%r12
|
|---|
| 2405 | mov %r11,-4*8($tptr)
|
|---|
| 2406 | adox %r15,%r13
|
|---|
| 2407 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 2408 | mov $bi,%rdx
|
|---|
| 2409 | mov %r12,-3*8($tptr)
|
|---|
| 2410 | adcx %rax,%r13
|
|---|
| 2411 | adox $zero,%r15
|
|---|
| 2412 | lea 4*8($nptr),$nptr
|
|---|
| 2413 | mov %r13,-2*8($tptr)
|
|---|
| 2414 |
|
|---|
| 2415 | dec $bptr # of=0, pass cf
|
|---|
| 2416 | jnz .Lmulx4x_1st
|
|---|
| 2417 |
|
|---|
| 2418 | mov 8(%rsp),$num # load -num
|
|---|
| 2419 | adc $zero,%r15 # modulo-scheduled
|
|---|
| 2420 | lea ($aptr,$num),$aptr # rewind $aptr
|
|---|
| 2421 | add %r15,%r14
|
|---|
| 2422 | mov 8+8(%rsp),$bptr # re-load &b[i]
|
|---|
| 2423 | adc $zero,$zero # top-most carry
|
|---|
| 2424 | mov %r14,-1*8($tptr)
|
|---|
| 2425 | jmp .Lmulx4x_outer
|
|---|
| 2426 |
|
|---|
| 2427 | .align 32
|
|---|
| 2428 | .Lmulx4x_outer:
|
|---|
| 2429 | lea 16-256($tptr),%r10 # where 256-byte mask is (+density control)
|
|---|
| 2430 | pxor %xmm4,%xmm4
|
|---|
| 2431 | .byte 0x67,0x67
|
|---|
| 2432 | pxor %xmm5,%xmm5
|
|---|
| 2433 | ___
|
|---|
| 2434 | for($i=0;$i<$STRIDE/16;$i+=4) {
|
|---|
| 2435 | $code.=<<___;
|
|---|
| 2436 | movdqa `16*($i+0)-128`($bptr),%xmm0
|
|---|
| 2437 | movdqa `16*($i+1)-128`($bptr),%xmm1
|
|---|
| 2438 | movdqa `16*($i+2)-128`($bptr),%xmm2
|
|---|
| 2439 | pand `16*($i+0)+256`(%r10),%xmm0
|
|---|
| 2440 | movdqa `16*($i+3)-128`($bptr),%xmm3
|
|---|
| 2441 | pand `16*($i+1)+256`(%r10),%xmm1
|
|---|
| 2442 | por %xmm0,%xmm4
|
|---|
| 2443 | pand `16*($i+2)+256`(%r10),%xmm2
|
|---|
| 2444 | por %xmm1,%xmm5
|
|---|
| 2445 | pand `16*($i+3)+256`(%r10),%xmm3
|
|---|
| 2446 | por %xmm2,%xmm4
|
|---|
| 2447 | por %xmm3,%xmm5
|
|---|
| 2448 | ___
|
|---|
| 2449 | }
|
|---|
| 2450 | $code.=<<___;
|
|---|
| 2451 | por %xmm5,%xmm4
|
|---|
| 2452 | pshufd \$0x4e,%xmm4,%xmm0
|
|---|
| 2453 | por %xmm4,%xmm0
|
|---|
| 2454 | lea $STRIDE($bptr),$bptr
|
|---|
| 2455 | movq %xmm0,%rdx # m0=bp[i]
|
|---|
| 2456 |
|
|---|
| 2457 | mov $zero,($tptr) # save top-most carry
|
|---|
| 2458 | lea 4*8($tptr,$num),$tptr # rewind $tptr
|
|---|
| 2459 | mulx 0*8($aptr),$mi,%r11 # a[0]*b[i]
|
|---|
| 2460 | xor $zero,$zero # cf=0, of=0
|
|---|
| 2461 | mov %rdx,$bi
|
|---|
| 2462 | mulx 1*8($aptr),%r14,%r12 # a[1]*b[i]
|
|---|
| 2463 | adox -4*8($tptr),$mi # +t[0]
|
|---|
| 2464 | adcx %r14,%r11
|
|---|
| 2465 | mulx 2*8($aptr),%r15,%r13 # ...
|
|---|
| 2466 | adox -3*8($tptr),%r11
|
|---|
| 2467 | adcx %r15,%r12
|
|---|
| 2468 | mulx 3*8($aptr),%rdx,%r14
|
|---|
| 2469 | adox -2*8($tptr),%r12
|
|---|
| 2470 | adcx %rdx,%r13
|
|---|
| 2471 | lea ($nptr,$num),$nptr # rewind $nptr
|
|---|
| 2472 | lea 4*8($aptr),$aptr
|
|---|
| 2473 | adox -1*8($tptr),%r13
|
|---|
| 2474 | adcx $zero,%r14
|
|---|
| 2475 | adox $zero,%r14
|
|---|
| 2476 |
|
|---|
| 2477 | mov $mi,%r15
|
|---|
| 2478 | imulq 32+8(%rsp),$mi # "t[0]"*n0
|
|---|
| 2479 |
|
|---|
| 2480 | mov $mi,%rdx
|
|---|
| 2481 | xor $zero,$zero # cf=0, of=0
|
|---|
| 2482 | mov $bptr,8+8(%rsp) # off-load &b[i]
|
|---|
| 2483 |
|
|---|
| 2484 | mulx 0*8($nptr),%rax,%r10
|
|---|
| 2485 | adcx %rax,%r15 # discarded
|
|---|
| 2486 | adox %r11,%r10
|
|---|
| 2487 | mulx 1*8($nptr),%rax,%r11
|
|---|
| 2488 | adcx %rax,%r10
|
|---|
| 2489 | adox %r12,%r11
|
|---|
| 2490 | mulx 2*8($nptr),%rax,%r12
|
|---|
| 2491 | adcx %rax,%r11
|
|---|
| 2492 | adox %r13,%r12
|
|---|
| 2493 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 2494 | mov $bi,%rdx
|
|---|
| 2495 | mov 24+8(%rsp),$bptr # counter value
|
|---|
| 2496 | mov %r10,-8*4($tptr)
|
|---|
| 2497 | adcx %rax,%r12
|
|---|
| 2498 | mov %r11,-8*3($tptr)
|
|---|
| 2499 | adox $zero,%r15 # of=0
|
|---|
| 2500 | mov %r12,-8*2($tptr)
|
|---|
| 2501 | lea 4*8($nptr),$nptr
|
|---|
| 2502 | jmp .Lmulx4x_inner
|
|---|
| 2503 |
|
|---|
| 2504 | .align 32
|
|---|
| 2505 | .Lmulx4x_inner:
|
|---|
| 2506 | mulx 0*8($aptr),%r10,%rax # a[4]*b[i]
|
|---|
| 2507 | adcx $zero,%r15 # cf=0, modulo-scheduled
|
|---|
| 2508 | adox %r14,%r10
|
|---|
| 2509 | mulx 1*8($aptr),%r11,%r14 # a[5]*b[i]
|
|---|
| 2510 | adcx 0*8($tptr),%r10
|
|---|
| 2511 | adox %rax,%r11
|
|---|
| 2512 | mulx 2*8($aptr),%r12,%rax # ...
|
|---|
| 2513 | adcx 1*8($tptr),%r11
|
|---|
| 2514 | adox %r14,%r12
|
|---|
| 2515 | mulx 3*8($aptr),%r13,%r14
|
|---|
| 2516 | mov $mi,%rdx
|
|---|
| 2517 | adcx 2*8($tptr),%r12
|
|---|
| 2518 | adox %rax,%r13
|
|---|
| 2519 | adcx 3*8($tptr),%r13
|
|---|
| 2520 | adox $zero,%r14 # of=0
|
|---|
| 2521 | lea 4*8($aptr),$aptr
|
|---|
| 2522 | lea 4*8($tptr),$tptr
|
|---|
| 2523 | adcx $zero,%r14 # cf=0
|
|---|
| 2524 |
|
|---|
| 2525 | adox %r15,%r10
|
|---|
| 2526 | mulx 0*8($nptr),%rax,%r15
|
|---|
| 2527 | adcx %rax,%r10
|
|---|
| 2528 | adox %r15,%r11
|
|---|
| 2529 | mulx 1*8($nptr),%rax,%r15
|
|---|
| 2530 | adcx %rax,%r11
|
|---|
| 2531 | adox %r15,%r12
|
|---|
| 2532 | mulx 2*8($nptr),%rax,%r15
|
|---|
| 2533 | mov %r10,-5*8($tptr)
|
|---|
| 2534 | adcx %rax,%r12
|
|---|
| 2535 | adox %r15,%r13
|
|---|
| 2536 | mov %r11,-4*8($tptr)
|
|---|
| 2537 | mulx 3*8($nptr),%rax,%r15
|
|---|
| 2538 | mov $bi,%rdx
|
|---|
| 2539 | lea 4*8($nptr),$nptr
|
|---|
| 2540 | mov %r12,-3*8($tptr)
|
|---|
| 2541 | adcx %rax,%r13
|
|---|
| 2542 | adox $zero,%r15
|
|---|
| 2543 | mov %r13,-2*8($tptr)
|
|---|
| 2544 |
|
|---|
| 2545 | dec $bptr # of=0, pass cf
|
|---|
| 2546 | jnz .Lmulx4x_inner
|
|---|
| 2547 |
|
|---|
| 2548 | mov 0+8(%rsp),$num # load -num
|
|---|
| 2549 | adc $zero,%r15 # modulo-scheduled
|
|---|
| 2550 | sub 0*8($tptr),$bptr # pull top-most carry to %cf
|
|---|
| 2551 | mov 8+8(%rsp),$bptr # re-load &b[i]
|
|---|
| 2552 | mov 16+8(%rsp),%r10
|
|---|
| 2553 | adc %r15,%r14
|
|---|
| 2554 | lea ($aptr,$num),$aptr # rewind $aptr
|
|---|
| 2555 | adc $zero,$zero # top-most carry
|
|---|
| 2556 | mov %r14,-1*8($tptr)
|
|---|
| 2557 |
|
|---|
| 2558 | cmp %r10,$bptr
|
|---|
| 2559 | jb .Lmulx4x_outer
|
|---|
| 2560 |
|
|---|
| 2561 | mov -8($nptr),%r10
|
|---|
| 2562 | mov $zero,%r8
|
|---|
| 2563 | mov ($nptr,$num),%r12
|
|---|
| 2564 | lea ($nptr,$num),%rbp # rewind $nptr
|
|---|
| 2565 | mov $num,%rcx
|
|---|
| 2566 | lea ($tptr,$num),%rdi # rewind $tptr
|
|---|
| 2567 | xor %eax,%eax
|
|---|
| 2568 | xor %r15,%r15
|
|---|
| 2569 | sub %r14,%r10 # compare top-most words
|
|---|
| 2570 | adc %r15,%r15
|
|---|
| 2571 | or %r15,%r8
|
|---|
| 2572 | sar \$3+2,%rcx
|
|---|
| 2573 | sub %r8,%rax # %rax=-%r8
|
|---|
| 2574 | mov 56+8(%rsp),%rdx # restore rp
|
|---|
| 2575 | dec %r12 # so that after 'not' we get -n[0]
|
|---|
| 2576 | mov 8*1(%rbp),%r13
|
|---|
| 2577 | xor %r8,%r8
|
|---|
| 2578 | mov 8*2(%rbp),%r14
|
|---|
| 2579 | mov 8*3(%rbp),%r15
|
|---|
| 2580 | jmp .Lsqrx4x_sub_entry # common post-condition
|
|---|
| 2581 | .cfi_endproc
|
|---|
| 2582 | .size mulx4x_internal,.-mulx4x_internal
|
|---|
| 2583 | ___
|
|---|
| 2584 | } |
|---|
| 2585 | {
|
|---|
| 2586 | ######################################################################
|
|---|
| 2587 | # void bn_power5(
|
|---|
| 2588 | my $rptr="%rdi"; # BN_ULONG *rptr,
|
|---|
| 2589 | my $aptr="%rsi"; # const BN_ULONG *aptr,
|
|---|
| 2590 | my $bptr="%rdx"; # const void *table,
|
|---|
| 2591 | my $nptr="%rcx"; # const BN_ULONG *nptr,
|
|---|
| 2592 | my $n0 ="%r8"; # const BN_ULONG *n0);
|
|---|
| 2593 | my $num ="%r9"; # int num, has to be divisible by 8
|
|---|
| 2594 | # int pwr);
|
|---|
| 2595 |
|
|---|
| 2596 | my ($i,$j,$tptr)=("%rbp","%rcx",$rptr);
|
|---|
| 2597 | my @A0=("%r10","%r11");
|
|---|
| 2598 | my @A1=("%r12","%r13");
|
|---|
| 2599 | my ($a0,$a1,$ai)=("%r14","%r15","%rbx");
|
|---|
| 2600 |
|
|---|
| 2601 | $code.=<<___;
|
|---|
| 2602 | .type bn_powerx5,\@function,6
|
|---|
| 2603 | .align 32
|
|---|
| 2604 | bn_powerx5:
|
|---|
| 2605 | .cfi_startproc
|
|---|
| 2606 | mov %rsp,%rax
|
|---|
| 2607 | .cfi_def_cfa_register %rax
|
|---|
| 2608 | .Lpowerx5_enter:
|
|---|
| 2609 | push %rbx
|
|---|
| 2610 | .cfi_push %rbx
|
|---|
| 2611 | push %rbp
|
|---|
| 2612 | .cfi_push %rbp
|
|---|
| 2613 | push %r12
|
|---|
| 2614 | .cfi_push %r12
|
|---|
| 2615 | push %r13
|
|---|
| 2616 | .cfi_push %r13
|
|---|
| 2617 | push %r14
|
|---|
| 2618 | .cfi_push %r14
|
|---|
| 2619 | push %r15
|
|---|
| 2620 | .cfi_push %r15
|
|---|
| 2621 | .Lpowerx5_prologue:
|
|---|
| 2622 |
|
|---|
| 2623 | shl \$3,${num}d # convert $num to bytes
|
|---|
| 2624 | lea ($num,$num,2),%r10 # 3*$num in bytes
|
|---|
| 2625 | neg $num
|
|---|
| 2626 | mov ($n0),$n0 # *n0
|
|---|
| 2627 |
|
|---|
| 2628 | ##############################################################
|
|---|
| 2629 | # Ensure that stack frame doesn't alias with $rptr+3*$num
|
|---|
| 2630 | # modulo 4096, which covers ret[num], am[num] and n[num]
|
|---|
| 2631 | # (see bn_exp.c). This is done to allow memory disambiguation
|
|---|
| 2632 | # logic do its magic. [Extra 256 bytes is for power mask
|
|---|
| 2633 | # calculated from 7th argument, the index.]
|
|---|
| 2634 | #
|
|---|
| 2635 | lea -320(%rsp,$num,2),%r11
|
|---|
| 2636 | mov %rsp,%rbp
|
|---|
| 2637 | sub $rptr,%r11
|
|---|
| 2638 | and \$4095,%r11
|
|---|
| 2639 | cmp %r11,%r10
|
|---|
| 2640 | jb .Lpwrx_sp_alt
|
|---|
| 2641 | sub %r11,%rbp # align with $aptr
|
|---|
| 2642 | lea -320(%rbp,$num,2),%rbp # future alloca(frame+2*$num*8+256)
|
|---|
| 2643 | jmp .Lpwrx_sp_done
|
|---|
| 2644 |
|
|---|
| 2645 | .align 32
|
|---|
| 2646 | .Lpwrx_sp_alt:
|
|---|
| 2647 | lea 4096-320(,$num,2),%r10
|
|---|
| 2648 | lea -320(%rbp,$num,2),%rbp # alloca(frame+2*$num*8+256)
|
|---|
| 2649 | sub %r10,%r11
|
|---|
| 2650 | mov \$0,%r10
|
|---|
| 2651 | cmovc %r10,%r11
|
|---|
| 2652 | sub %r11,%rbp
|
|---|
| 2653 | .Lpwrx_sp_done:
|
|---|
| 2654 | and \$-64,%rbp
|
|---|
| 2655 | mov %rsp,%r11
|
|---|
| 2656 | sub %rbp,%r11
|
|---|
| 2657 | and \$-4096,%r11
|
|---|
| 2658 | lea (%rbp,%r11),%rsp
|
|---|
| 2659 | mov (%rsp),%r10
|
|---|
| 2660 | cmp %rbp,%rsp
|
|---|
| 2661 | ja .Lpwrx_page_walk
|
|---|
| 2662 | jmp .Lpwrx_page_walk_done
|
|---|
| 2663 |
|
|---|
| 2664 | .Lpwrx_page_walk:
|
|---|
| 2665 | lea -4096(%rsp),%rsp
|
|---|
| 2666 | mov (%rsp),%r10
|
|---|
| 2667 | cmp %rbp,%rsp
|
|---|
| 2668 | ja .Lpwrx_page_walk
|
|---|
| 2669 | .Lpwrx_page_walk_done:
|
|---|
| 2670 |
|
|---|
| 2671 | mov $num,%r10
|
|---|
| 2672 | neg $num
|
|---|
| 2673 |
|
|---|
| 2674 | ##############################################################
|
|---|
| 2675 | # Stack layout
|
|---|
| 2676 | #
|
|---|
| 2677 | # +0 saved $num, used in reduction section
|
|---|
| 2678 | # +8 &t[2*$num], used in reduction section
|
|---|
| 2679 | # +16 intermediate carry bit
|
|---|
| 2680 | # +24 top-most carry bit, used in reduction section
|
|---|
| 2681 | # +32 saved *n0
|
|---|
| 2682 | # +40 saved %rsp
|
|---|
| 2683 | # +48 t[2*$num]
|
|---|
| 2684 | #
|
|---|
| 2685 | pxor %xmm0,%xmm0
|
|---|
| 2686 | movq $rptr,%xmm1 # save $rptr
|
|---|
| 2687 | movq $nptr,%xmm2 # save $nptr
|
|---|
| 2688 | movq %r10, %xmm3 # -$num
|
|---|
| 2689 | movq $bptr,%xmm4
|
|---|
| 2690 | mov $n0, 32(%rsp)
|
|---|
| 2691 | mov %rax, 40(%rsp) # save original %rsp
|
|---|
| 2692 | .cfi_cfa_expression %rsp+40,deref,+8
|
|---|
| 2693 | .Lpowerx5_body:
|
|---|
| 2694 |
|
|---|
| 2695 | call __bn_sqrx8x_internal
|
|---|
| 2696 | call __bn_postx4x_internal
|
|---|
| 2697 | call __bn_sqrx8x_internal
|
|---|
| 2698 | call __bn_postx4x_internal
|
|---|
| 2699 | call __bn_sqrx8x_internal
|
|---|
| 2700 | call __bn_postx4x_internal
|
|---|
| 2701 | call __bn_sqrx8x_internal
|
|---|
| 2702 | call __bn_postx4x_internal
|
|---|
| 2703 | call __bn_sqrx8x_internal
|
|---|
| 2704 | call __bn_postx4x_internal
|
|---|
| 2705 |
|
|---|
| 2706 | mov %r10,$num # -num
|
|---|
| 2707 | mov $aptr,$rptr
|
|---|
| 2708 | movq %xmm2,$nptr
|
|---|
| 2709 | movq %xmm4,$bptr
|
|---|
| 2710 | mov 40(%rsp),%rax
|
|---|
| 2711 |
|
|---|
| 2712 | call mulx4x_internal
|
|---|
| 2713 |
|
|---|
| 2714 | mov 40(%rsp),%rsi # restore %rsp
|
|---|
| 2715 | .cfi_def_cfa %rsi,8
|
|---|
| 2716 | mov \$1,%rax
|
|---|
| 2717 |
|
|---|
| 2718 | mov -48(%rsi),%r15
|
|---|
| 2719 | .cfi_restore %r15
|
|---|
| 2720 | mov -40(%rsi),%r14
|
|---|
| 2721 | .cfi_restore %r14
|
|---|
| 2722 | mov -32(%rsi),%r13
|
|---|
| 2723 | .cfi_restore %r13
|
|---|
| 2724 | mov -24(%rsi),%r12
|
|---|
| 2725 | .cfi_restore %r12
|
|---|
| 2726 | mov -16(%rsi),%rbp
|
|---|
| 2727 | .cfi_restore %rbp
|
|---|
| 2728 | mov -8(%rsi),%rbx
|
|---|
| 2729 | .cfi_restore %rbx
|
|---|
| 2730 | lea (%rsi),%rsp
|
|---|
| 2731 | .cfi_def_cfa_register %rsp
|
|---|
| 2732 | .Lpowerx5_epilogue:
|
|---|
| 2733 | ret
|
|---|
| 2734 | .cfi_endproc
|
|---|
| 2735 | .size bn_powerx5,.-bn_powerx5
|
|---|
| 2736 |
|
|---|
| 2737 | .globl bn_sqrx8x_internal
|
|---|
| 2738 | .hidden bn_sqrx8x_internal
|
|---|
| 2739 | .type bn_sqrx8x_internal,\@abi-omnipotent
|
|---|
| 2740 | .align 32
|
|---|
| 2741 | bn_sqrx8x_internal:
|
|---|
| 2742 | __bn_sqrx8x_internal:
|
|---|
| 2743 | .cfi_startproc
|
|---|
| 2744 | ##################################################################
|
|---|
| 2745 | # Squaring part:
|
|---|
| 2746 | #
|
|---|
| 2747 | # a) multiply-n-add everything but a[i]*a[i];
|
|---|
| 2748 | # b) shift result of a) by 1 to the left and accumulate
|
|---|
| 2749 | # a[i]*a[i] products;
|
|---|
| 2750 | #
|
|---|
| 2751 | ##################################################################
|
|---|
| 2752 | # 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]
|
|---|
| 2753 | # a[1]a[0]
|
|---|
| 2754 | # a[2]a[0]
|
|---|
| 2755 | # a[3]a[0]
|
|---|
| 2756 | # a[2]a[1]
|
|---|
| 2757 | # a[3]a[1]
|
|---|
| 2758 | # a[3]a[2]
|
|---|
| 2759 | #
|
|---|
| 2760 | # a[4]a[0]
|
|---|
| 2761 | # a[5]a[0]
|
|---|
| 2762 | # a[6]a[0]
|
|---|
| 2763 | # a[7]a[0]
|
|---|
| 2764 | # a[4]a[1]
|
|---|
| 2765 | # a[5]a[1]
|
|---|
| 2766 | # a[6]a[1]
|
|---|
| 2767 | # a[7]a[1]
|
|---|
| 2768 | # a[4]a[2]
|
|---|
| 2769 | # a[5]a[2]
|
|---|
| 2770 | # a[6]a[2]
|
|---|
| 2771 | # a[7]a[2]
|
|---|
| 2772 | # a[4]a[3]
|
|---|
| 2773 | # a[5]a[3]
|
|---|
| 2774 | # a[6]a[3]
|
|---|
| 2775 | # a[7]a[3]
|
|---|
| 2776 | #
|
|---|
| 2777 | # a[5]a[4]
|
|---|
| 2778 | # a[6]a[4]
|
|---|
| 2779 | # a[7]a[4]
|
|---|
| 2780 | # a[6]a[5]
|
|---|
| 2781 | # a[7]a[5]
|
|---|
| 2782 | # a[7]a[6]
|
|---|
| 2783 | # 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]
|
|---|
| 2784 | ___
|
|---|
| 2785 | {
|
|---|
| 2786 | my ($zero,$carry)=("%rbp","%rcx");
|
|---|
| 2787 | my $aaptr=$zero;
|
|---|
| 2788 | $code.=<<___;
|
|---|
| 2789 | lea 48+8(%rsp),$tptr
|
|---|
| 2790 | lea ($aptr,$num),$aaptr
|
|---|
| 2791 | mov $num,0+8(%rsp) # save $num
|
|---|
| 2792 | mov $aaptr,8+8(%rsp) # save end of $aptr
|
|---|
| 2793 | jmp .Lsqr8x_zero_start
|
|---|
| 2794 |
|
|---|
| 2795 | .align 32
|
|---|
| 2796 | .byte 0x66,0x66,0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00
|
|---|
| 2797 | .Lsqrx8x_zero:
|
|---|
| 2798 | .byte 0x3e
|
|---|
| 2799 | movdqa %xmm0,0*8($tptr)
|
|---|
| 2800 | movdqa %xmm0,2*8($tptr)
|
|---|
| 2801 | movdqa %xmm0,4*8($tptr)
|
|---|
| 2802 | movdqa %xmm0,6*8($tptr)
|
|---|
| 2803 | .Lsqr8x_zero_start: # aligned at 32
|
|---|
| 2804 | movdqa %xmm0,8*8($tptr)
|
|---|
| 2805 | movdqa %xmm0,10*8($tptr)
|
|---|
| 2806 | movdqa %xmm0,12*8($tptr)
|
|---|
| 2807 | movdqa %xmm0,14*8($tptr)
|
|---|
| 2808 | lea 16*8($tptr),$tptr
|
|---|
| 2809 | sub \$64,$num
|
|---|
| 2810 | jnz .Lsqrx8x_zero
|
|---|
| 2811 |
|
|---|
| 2812 | mov 0*8($aptr),%rdx # a[0], modulo-scheduled
|
|---|
| 2813 | #xor %r9,%r9 # t[1], ex-$num, zero already
|
|---|
| 2814 | xor %r10,%r10
|
|---|
| 2815 | xor %r11,%r11
|
|---|
| 2816 | xor %r12,%r12
|
|---|
| 2817 | xor %r13,%r13
|
|---|
| 2818 | xor %r14,%r14
|
|---|
| 2819 | xor %r15,%r15
|
|---|
| 2820 | lea 48+8(%rsp),$tptr
|
|---|
| 2821 | xor $zero,$zero # cf=0, cf=0
|
|---|
| 2822 | jmp .Lsqrx8x_outer_loop
|
|---|
| 2823 |
|
|---|
| 2824 | .align 32
|
|---|
| 2825 | .Lsqrx8x_outer_loop:
|
|---|
| 2826 | mulx 1*8($aptr),%r8,%rax # a[1]*a[0]
|
|---|
| 2827 | adcx %r9,%r8 # a[1]*a[0]+=t[1]
|
|---|
| 2828 | adox %rax,%r10
|
|---|
| 2829 | mulx 2*8($aptr),%r9,%rax # a[2]*a[0]
|
|---|
| 2830 | adcx %r10,%r9
|
|---|
| 2831 | adox %rax,%r11
|
|---|
| 2832 | .byte 0xc4,0xe2,0xab,0xf6,0x86,0x18,0x00,0x00,0x00 # mulx 3*8($aptr),%r10,%rax # ...
|
|---|
| 2833 | adcx %r11,%r10
|
|---|
| 2834 | adox %rax,%r12
|
|---|
| 2835 | .byte 0xc4,0xe2,0xa3,0xf6,0x86,0x20,0x00,0x00,0x00 # mulx 4*8($aptr),%r11,%rax
|
|---|
| 2836 | adcx %r12,%r11
|
|---|
| 2837 | adox %rax,%r13
|
|---|
| 2838 | mulx 5*8($aptr),%r12,%rax
|
|---|
| 2839 | adcx %r13,%r12
|
|---|
| 2840 | adox %rax,%r14
|
|---|
| 2841 | mulx 6*8($aptr),%r13,%rax
|
|---|
| 2842 | adcx %r14,%r13
|
|---|
| 2843 | adox %r15,%rax
|
|---|
| 2844 | mulx 7*8($aptr),%r14,%r15
|
|---|
| 2845 | mov 1*8($aptr),%rdx # a[1]
|
|---|
| 2846 | adcx %rax,%r14
|
|---|
| 2847 | adox $zero,%r15
|
|---|
| 2848 | adc 8*8($tptr),%r15
|
|---|
| 2849 | mov %r8,1*8($tptr) # t[1]
|
|---|
| 2850 | mov %r9,2*8($tptr) # t[2]
|
|---|
| 2851 | sbb $carry,$carry # mov %cf,$carry
|
|---|
| 2852 | xor $zero,$zero # cf=0, of=0
|
|---|
| 2853 |
|
|---|
| 2854 |
|
|---|
| 2855 | mulx 2*8($aptr),%r8,%rbx # a[2]*a[1]
|
|---|
| 2856 | mulx 3*8($aptr),%r9,%rax # a[3]*a[1]
|
|---|
| 2857 | adcx %r10,%r8
|
|---|
| 2858 | adox %rbx,%r9
|
|---|
| 2859 | mulx 4*8($aptr),%r10,%rbx # ...
|
|---|
| 2860 | adcx %r11,%r9
|
|---|
| 2861 | adox %rax,%r10
|
|---|
| 2862 | .byte 0xc4,0xe2,0xa3,0xf6,0x86,0x28,0x00,0x00,0x00 # mulx 5*8($aptr),%r11,%rax
|
|---|
| 2863 | adcx %r12,%r10
|
|---|
| 2864 | adox %rbx,%r11
|
|---|
| 2865 | .byte 0xc4,0xe2,0x9b,0xf6,0x9e,0x30,0x00,0x00,0x00 # mulx 6*8($aptr),%r12,%rbx
|
|---|
| 2866 | adcx %r13,%r11
|
|---|
| 2867 | adox %r14,%r12
|
|---|
| 2868 | .byte 0xc4,0x62,0x93,0xf6,0xb6,0x38,0x00,0x00,0x00 # mulx 7*8($aptr),%r13,%r14
|
|---|
| 2869 | mov 2*8($aptr),%rdx # a[2]
|
|---|
| 2870 | adcx %rax,%r12
|
|---|
| 2871 | adox %rbx,%r13
|
|---|
| 2872 | adcx %r15,%r13
|
|---|
| 2873 | adox $zero,%r14 # of=0
|
|---|
| 2874 | adcx $zero,%r14 # cf=0
|
|---|
| 2875 |
|
|---|
| 2876 | mov %r8,3*8($tptr) # t[3]
|
|---|
| 2877 | mov %r9,4*8($tptr) # t[4]
|
|---|
| 2878 |
|
|---|
| 2879 | mulx 3*8($aptr),%r8,%rbx # a[3]*a[2]
|
|---|
| 2880 | mulx 4*8($aptr),%r9,%rax # a[4]*a[2]
|
|---|
| 2881 | adcx %r10,%r8
|
|---|
| 2882 | adox %rbx,%r9
|
|---|
| 2883 | mulx 5*8($aptr),%r10,%rbx # ...
|
|---|
| 2884 | adcx %r11,%r9
|
|---|
| 2885 | adox %rax,%r10
|
|---|
| 2886 | .byte 0xc4,0xe2,0xa3,0xf6,0x86,0x30,0x00,0x00,0x00 # mulx 6*8($aptr),%r11,%rax
|
|---|
| 2887 | adcx %r12,%r10
|
|---|
| 2888 | adox %r13,%r11
|
|---|
| 2889 | .byte 0xc4,0x62,0x9b,0xf6,0xae,0x38,0x00,0x00,0x00 # mulx 7*8($aptr),%r12,%r13
|
|---|
| 2890 | .byte 0x3e
|
|---|
| 2891 | mov 3*8($aptr),%rdx # a[3]
|
|---|
| 2892 | adcx %rbx,%r11
|
|---|
| 2893 | adox %rax,%r12
|
|---|
| 2894 | adcx %r14,%r12
|
|---|
| 2895 | mov %r8,5*8($tptr) # t[5]
|
|---|
| 2896 | mov %r9,6*8($tptr) # t[6]
|
|---|
| 2897 | mulx 4*8($aptr),%r8,%rax # a[4]*a[3]
|
|---|
| 2898 | adox $zero,%r13 # of=0
|
|---|
| 2899 | adcx $zero,%r13 # cf=0
|
|---|
| 2900 |
|
|---|
| 2901 | mulx 5*8($aptr),%r9,%rbx # a[5]*a[3]
|
|---|
| 2902 | adcx %r10,%r8
|
|---|
| 2903 | adox %rax,%r9
|
|---|
| 2904 | mulx 6*8($aptr),%r10,%rax # ...
|
|---|
| 2905 | adcx %r11,%r9
|
|---|
| 2906 | adox %r12,%r10
|
|---|
| 2907 | mulx 7*8($aptr),%r11,%r12
|
|---|
| 2908 | mov 4*8($aptr),%rdx # a[4]
|
|---|
| 2909 | mov 5*8($aptr),%r14 # a[5]
|
|---|
| 2910 | adcx %rbx,%r10
|
|---|
| 2911 | adox %rax,%r11
|
|---|
| 2912 | mov 6*8($aptr),%r15 # a[6]
|
|---|
| 2913 | adcx %r13,%r11
|
|---|
| 2914 | adox $zero,%r12 # of=0
|
|---|
| 2915 | adcx $zero,%r12 # cf=0
|
|---|
| 2916 |
|
|---|
| 2917 | mov %r8,7*8($tptr) # t[7]
|
|---|
| 2918 | mov %r9,8*8($tptr) # t[8]
|
|---|
| 2919 |
|
|---|
| 2920 | mulx %r14,%r9,%rax # a[5]*a[4]
|
|---|
| 2921 | mov 7*8($aptr),%r8 # a[7]
|
|---|
| 2922 | adcx %r10,%r9
|
|---|
| 2923 | mulx %r15,%r10,%rbx # a[6]*a[4]
|
|---|
| 2924 | adox %rax,%r10
|
|---|
| 2925 | adcx %r11,%r10
|
|---|
| 2926 | mulx %r8,%r11,%rax # a[7]*a[4]
|
|---|
| 2927 | mov %r14,%rdx # a[5]
|
|---|
| 2928 | adox %rbx,%r11
|
|---|
| 2929 | adcx %r12,%r11
|
|---|
| 2930 | #adox $zero,%rax # of=0
|
|---|
| 2931 | adcx $zero,%rax # cf=0
|
|---|
| 2932 |
|
|---|
| 2933 | mulx %r15,%r14,%rbx # a[6]*a[5]
|
|---|
| 2934 | mulx %r8,%r12,%r13 # a[7]*a[5]
|
|---|
| 2935 | mov %r15,%rdx # a[6]
|
|---|
| 2936 | lea 8*8($aptr),$aptr
|
|---|
| 2937 | adcx %r14,%r11
|
|---|
| 2938 | adox %rbx,%r12
|
|---|
| 2939 | adcx %rax,%r12
|
|---|
| 2940 | adox $zero,%r13
|
|---|
| 2941 |
|
|---|
| 2942 | .byte 0x67,0x67
|
|---|
| 2943 | mulx %r8,%r8,%r14 # a[7]*a[6]
|
|---|
| 2944 | adcx %r8,%r13
|
|---|
| 2945 | adcx $zero,%r14
|
|---|
| 2946 |
|
|---|
| 2947 | cmp 8+8(%rsp),$aptr
|
|---|
| 2948 | je .Lsqrx8x_outer_break
|
|---|
| 2949 |
|
|---|
| 2950 | neg $carry # mov $carry,%cf
|
|---|
| 2951 | mov \$-8,%rcx
|
|---|
| 2952 | mov $zero,%r15
|
|---|
| 2953 | mov 8*8($tptr),%r8
|
|---|
| 2954 | adcx 9*8($tptr),%r9 # +=t[9]
|
|---|
| 2955 | adcx 10*8($tptr),%r10 # ...
|
|---|
| 2956 | adcx 11*8($tptr),%r11
|
|---|
| 2957 | adc 12*8($tptr),%r12
|
|---|
| 2958 | adc 13*8($tptr),%r13
|
|---|
| 2959 | adc 14*8($tptr),%r14
|
|---|
| 2960 | adc 15*8($tptr),%r15
|
|---|
| 2961 | lea ($aptr),$aaptr
|
|---|
| 2962 | lea 2*64($tptr),$tptr
|
|---|
| 2963 | sbb %rax,%rax # mov %cf,$carry
|
|---|
| 2964 |
|
|---|
| 2965 | mov -64($aptr),%rdx # a[0]
|
|---|
| 2966 | mov %rax,16+8(%rsp) # offload $carry
|
|---|
| 2967 | mov $tptr,24+8(%rsp)
|
|---|
| 2968 |
|
|---|
| 2969 | #lea 8*8($tptr),$tptr # see 2*8*8($tptr) above
|
|---|
| 2970 | xor %eax,%eax # cf=0, of=0
|
|---|
| 2971 | jmp .Lsqrx8x_loop
|
|---|
| 2972 |
|
|---|
| 2973 | .align 32
|
|---|
| 2974 | .Lsqrx8x_loop:
|
|---|
| 2975 | mov %r8,%rbx
|
|---|
| 2976 | mulx 0*8($aaptr),%rax,%r8 # a[8]*a[i]
|
|---|
| 2977 | adcx %rax,%rbx # +=t[8]
|
|---|
| 2978 | adox %r9,%r8
|
|---|
| 2979 |
|
|---|
| 2980 | mulx 1*8($aaptr),%rax,%r9 # ...
|
|---|
| 2981 | adcx %rax,%r8
|
|---|
| 2982 | adox %r10,%r9
|
|---|
| 2983 |
|
|---|
| 2984 | mulx 2*8($aaptr),%rax,%r10
|
|---|
| 2985 | adcx %rax,%r9
|
|---|
| 2986 | adox %r11,%r10
|
|---|
| 2987 |
|
|---|
| 2988 | mulx 3*8($aaptr),%rax,%r11
|
|---|
| 2989 | adcx %rax,%r10
|
|---|
| 2990 | adox %r12,%r11
|
|---|
| 2991 |
|
|---|
| 2992 | .byte 0xc4,0x62,0xfb,0xf6,0xa5,0x20,0x00,0x00,0x00 # mulx 4*8($aaptr),%rax,%r12
|
|---|
| 2993 | adcx %rax,%r11
|
|---|
| 2994 | adox %r13,%r12
|
|---|
| 2995 |
|
|---|
| 2996 | mulx 5*8($aaptr),%rax,%r13
|
|---|
| 2997 | adcx %rax,%r12
|
|---|
| 2998 | adox %r14,%r13
|
|---|
| 2999 |
|
|---|
| 3000 | mulx 6*8($aaptr),%rax,%r14
|
|---|
| 3001 | mov %rbx,($tptr,%rcx,8) # store t[8+i]
|
|---|
| 3002 | mov \$0,%ebx
|
|---|
| 3003 | adcx %rax,%r13
|
|---|
| 3004 | adox %r15,%r14
|
|---|
| 3005 |
|
|---|
| 3006 | .byte 0xc4,0x62,0xfb,0xf6,0xbd,0x38,0x00,0x00,0x00 # mulx 7*8($aaptr),%rax,%r15
|
|---|
| 3007 | mov 8($aptr,%rcx,8),%rdx # a[i]
|
|---|
| 3008 | adcx %rax,%r14
|
|---|
| 3009 | adox %rbx,%r15 # %rbx is 0, of=0
|
|---|
| 3010 | adcx %rbx,%r15 # cf=0
|
|---|
| 3011 |
|
|---|
| 3012 | .byte 0x67
|
|---|
| 3013 | inc %rcx # of=0
|
|---|
| 3014 | jnz .Lsqrx8x_loop
|
|---|
| 3015 |
|
|---|
| 3016 | lea 8*8($aaptr),$aaptr
|
|---|
| 3017 | mov \$-8,%rcx
|
|---|
| 3018 | cmp 8+8(%rsp),$aaptr # done?
|
|---|
| 3019 | je .Lsqrx8x_break
|
|---|
| 3020 |
|
|---|
| 3021 | sub 16+8(%rsp),%rbx # mov 16(%rsp),%cf
|
|---|
| 3022 | .byte 0x66
|
|---|
| 3023 | mov -64($aptr),%rdx
|
|---|
| 3024 | adcx 0*8($tptr),%r8
|
|---|
| 3025 | adcx 1*8($tptr),%r9
|
|---|
| 3026 | adc 2*8($tptr),%r10
|
|---|
| 3027 | adc 3*8($tptr),%r11
|
|---|
| 3028 | adc 4*8($tptr),%r12
|
|---|
| 3029 | adc 5*8($tptr),%r13
|
|---|
| 3030 | adc 6*8($tptr),%r14
|
|---|
| 3031 | adc 7*8($tptr),%r15
|
|---|
| 3032 | lea 8*8($tptr),$tptr
|
|---|
| 3033 | .byte 0x67
|
|---|
| 3034 | sbb %rax,%rax # mov %cf,%rax
|
|---|
| 3035 | xor %ebx,%ebx # cf=0, of=0
|
|---|
| 3036 | mov %rax,16+8(%rsp) # offload carry
|
|---|
| 3037 | jmp .Lsqrx8x_loop
|
|---|
| 3038 |
|
|---|
| 3039 | .align 32
|
|---|
| 3040 | .Lsqrx8x_break:
|
|---|
| 3041 | xor $zero,$zero
|
|---|
| 3042 | sub 16+8(%rsp),%rbx # mov 16(%rsp),%cf
|
|---|
| 3043 | adcx $zero,%r8
|
|---|
| 3044 | mov 24+8(%rsp),$carry # initial $tptr, borrow $carry
|
|---|
| 3045 | adcx $zero,%r9
|
|---|
| 3046 | mov 0*8($aptr),%rdx # a[8], modulo-scheduled
|
|---|
| 3047 | adc \$0,%r10
|
|---|
| 3048 | mov %r8,0*8($tptr)
|
|---|
| 3049 | adc \$0,%r11
|
|---|
| 3050 | adc \$0,%r12
|
|---|
| 3051 | adc \$0,%r13
|
|---|
| 3052 | adc \$0,%r14
|
|---|
| 3053 | adc \$0,%r15
|
|---|
| 3054 | cmp $carry,$tptr # cf=0, of=0
|
|---|
| 3055 | je .Lsqrx8x_outer_loop
|
|---|
| 3056 |
|
|---|
| 3057 | mov %r9,1*8($tptr)
|
|---|
| 3058 | mov 1*8($carry),%r9
|
|---|
| 3059 | mov %r10,2*8($tptr)
|
|---|
| 3060 | mov 2*8($carry),%r10
|
|---|
| 3061 | mov %r11,3*8($tptr)
|
|---|
| 3062 | mov 3*8($carry),%r11
|
|---|
| 3063 | mov %r12,4*8($tptr)
|
|---|
| 3064 | mov 4*8($carry),%r12
|
|---|
| 3065 | mov %r13,5*8($tptr)
|
|---|
| 3066 | mov 5*8($carry),%r13
|
|---|
| 3067 | mov %r14,6*8($tptr)
|
|---|
| 3068 | mov 6*8($carry),%r14
|
|---|
| 3069 | mov %r15,7*8($tptr)
|
|---|
| 3070 | mov 7*8($carry),%r15
|
|---|
| 3071 | mov $carry,$tptr
|
|---|
| 3072 | jmp .Lsqrx8x_outer_loop
|
|---|
| 3073 |
|
|---|
| 3074 | .align 32
|
|---|
| 3075 | .Lsqrx8x_outer_break:
|
|---|
| 3076 | mov %r9,9*8($tptr) # t[9]
|
|---|
| 3077 | movq %xmm3,%rcx # -$num
|
|---|
| 3078 | mov %r10,10*8($tptr) # ...
|
|---|
| 3079 | mov %r11,11*8($tptr)
|
|---|
| 3080 | mov %r12,12*8($tptr)
|
|---|
| 3081 | mov %r13,13*8($tptr)
|
|---|
| 3082 | mov %r14,14*8($tptr)
|
|---|
| 3083 | ___
|
|---|
| 3084 | } |
|---|
| 3085 | {
|
|---|
| 3086 | my $i="%rcx";
|
|---|
| 3087 | $code.=<<___;
|
|---|
| 3088 | lea 48+8(%rsp),$tptr
|
|---|
| 3089 | mov ($aptr,$i),%rdx # a[0]
|
|---|
| 3090 |
|
|---|
| 3091 | mov 8($tptr),$A0[1] # t[1]
|
|---|
| 3092 | xor $A0[0],$A0[0] # t[0], of=0, cf=0
|
|---|
| 3093 | mov 0+8(%rsp),$num # restore $num
|
|---|
| 3094 | adox $A0[1],$A0[1]
|
|---|
| 3095 | mov 16($tptr),$A1[0] # t[2] # prefetch
|
|---|
| 3096 | mov 24($tptr),$A1[1] # t[3] # prefetch
|
|---|
| 3097 | #jmp .Lsqrx4x_shift_n_add # happens to be aligned
|
|---|
| 3098 |
|
|---|
| 3099 | .align 32
|
|---|
| 3100 | .Lsqrx4x_shift_n_add:
|
|---|
| 3101 | mulx %rdx,%rax,%rbx
|
|---|
| 3102 | adox $A1[0],$A1[0]
|
|---|
| 3103 | adcx $A0[0],%rax
|
|---|
| 3104 | .byte 0x48,0x8b,0x94,0x0e,0x08,0x00,0x00,0x00 # mov 8($aptr,$i),%rdx # a[i+1] # prefetch
|
|---|
| 3105 | .byte 0x4c,0x8b,0x97,0x20,0x00,0x00,0x00 # mov 32($tptr),$A0[0] # t[2*i+4] # prefetch
|
|---|
| 3106 | adox $A1[1],$A1[1]
|
|---|
| 3107 | adcx $A0[1],%rbx
|
|---|
| 3108 | mov 40($tptr),$A0[1] # t[2*i+4+1] # prefetch
|
|---|
| 3109 | mov %rax,0($tptr)
|
|---|
| 3110 | mov %rbx,8($tptr)
|
|---|
| 3111 |
|
|---|
| 3112 | mulx %rdx,%rax,%rbx
|
|---|
| 3113 | adox $A0[0],$A0[0]
|
|---|
| 3114 | adcx $A1[0],%rax
|
|---|
| 3115 | mov 16($aptr,$i),%rdx # a[i+2] # prefetch
|
|---|
| 3116 | mov 48($tptr),$A1[0] # t[2*i+6] # prefetch
|
|---|
| 3117 | adox $A0[1],$A0[1]
|
|---|
| 3118 | adcx $A1[1],%rbx
|
|---|
| 3119 | mov 56($tptr),$A1[1] # t[2*i+6+1] # prefetch
|
|---|
| 3120 | mov %rax,16($tptr)
|
|---|
| 3121 | mov %rbx,24($tptr)
|
|---|
| 3122 |
|
|---|
| 3123 | mulx %rdx,%rax,%rbx
|
|---|
| 3124 | adox $A1[0],$A1[0]
|
|---|
| 3125 | adcx $A0[0],%rax
|
|---|
| 3126 | mov 24($aptr,$i),%rdx # a[i+3] # prefetch
|
|---|
| 3127 | lea 32($i),$i
|
|---|
| 3128 | mov 64($tptr),$A0[0] # t[2*i+8] # prefetch
|
|---|
| 3129 | adox $A1[1],$A1[1]
|
|---|
| 3130 | adcx $A0[1],%rbx
|
|---|
| 3131 | mov 72($tptr),$A0[1] # t[2*i+8+1] # prefetch
|
|---|
| 3132 | mov %rax,32($tptr)
|
|---|
| 3133 | mov %rbx,40($tptr)
|
|---|
| 3134 |
|
|---|
| 3135 | mulx %rdx,%rax,%rbx
|
|---|
| 3136 | adox $A0[0],$A0[0]
|
|---|
| 3137 | adcx $A1[0],%rax
|
|---|
| 3138 | jrcxz .Lsqrx4x_shift_n_add_break
|
|---|
| 3139 | .byte 0x48,0x8b,0x94,0x0e,0x00,0x00,0x00,0x00 # mov 0($aptr,$i),%rdx # a[i+4] # prefetch
|
|---|
| 3140 | adox $A0[1],$A0[1]
|
|---|
| 3141 | adcx $A1[1],%rbx
|
|---|
| 3142 | mov 80($tptr),$A1[0] # t[2*i+10] # prefetch
|
|---|
| 3143 | mov 88($tptr),$A1[1] # t[2*i+10+1] # prefetch
|
|---|
| 3144 | mov %rax,48($tptr)
|
|---|
| 3145 | mov %rbx,56($tptr)
|
|---|
| 3146 | lea 64($tptr),$tptr
|
|---|
| 3147 | nop
|
|---|
| 3148 | jmp .Lsqrx4x_shift_n_add
|
|---|
| 3149 |
|
|---|
| 3150 | .align 32
|
|---|
| 3151 | .Lsqrx4x_shift_n_add_break:
|
|---|
| 3152 | adcx $A1[1],%rbx
|
|---|
| 3153 | mov %rax,48($tptr)
|
|---|
| 3154 | mov %rbx,56($tptr)
|
|---|
| 3155 | lea 64($tptr),$tptr # end of t[] buffer
|
|---|
| 3156 | ___
|
|---|
| 3157 | } |
|---|
| 3158 |
|
|---|
| 3159 | ######################################################################
|
|---|
| 3160 | # Montgomery reduction part, "word-by-word" algorithm.
|
|---|
| 3161 | #
|
|---|
| 3162 | # This new path is inspired by multiple submissions from Intel, by
|
|---|
| 3163 | # Shay Gueron, Vlad Krasnov, Erdinc Ozturk, James Guilford,
|
|---|
| 3164 | # Vinodh Gopal...
|
|---|
| 3165 | {
|
|---|
| 3166 | my ($nptr,$carry,$m0)=("%rbp","%rsi","%rdx");
|
|---|
| 3167 |
|
|---|
| 3168 | $code.=<<___;
|
|---|
| 3169 | movq %xmm2,$nptr
|
|---|
| 3170 | __bn_sqrx8x_reduction:
|
|---|
| 3171 | xor %eax,%eax # initial top-most carry bit
|
|---|
| 3172 | mov 32+8(%rsp),%rbx # n0
|
|---|
| 3173 | mov 48+8(%rsp),%rdx # "%r8", 8*0($tptr)
|
|---|
| 3174 | lea -8*8($nptr,$num),%rcx # end of n[]
|
|---|
| 3175 | #lea 48+8(%rsp,$num,2),$tptr # end of t[] buffer
|
|---|
| 3176 | mov %rcx, 0+8(%rsp) # save end of n[]
|
|---|
| 3177 | mov $tptr,8+8(%rsp) # save end of t[]
|
|---|
| 3178 |
|
|---|
| 3179 | lea 48+8(%rsp),$tptr # initial t[] window
|
|---|
| 3180 | jmp .Lsqrx8x_reduction_loop
|
|---|
| 3181 |
|
|---|
| 3182 | .align 32
|
|---|
| 3183 | .Lsqrx8x_reduction_loop:
|
|---|
| 3184 | mov 8*1($tptr),%r9
|
|---|
| 3185 | mov 8*2($tptr),%r10
|
|---|
| 3186 | mov 8*3($tptr),%r11
|
|---|
| 3187 | mov 8*4($tptr),%r12
|
|---|
| 3188 | mov %rdx,%r8
|
|---|
| 3189 | imulq %rbx,%rdx # n0*a[i]
|
|---|
| 3190 | mov 8*5($tptr),%r13
|
|---|
| 3191 | mov 8*6($tptr),%r14
|
|---|
| 3192 | mov 8*7($tptr),%r15
|
|---|
| 3193 | mov %rax,24+8(%rsp) # store top-most carry bit
|
|---|
| 3194 |
|
|---|
| 3195 | lea 8*8($tptr),$tptr
|
|---|
| 3196 | xor $carry,$carry # cf=0,of=0
|
|---|
| 3197 | mov \$-8,%rcx
|
|---|
| 3198 | jmp .Lsqrx8x_reduce
|
|---|
| 3199 |
|
|---|
| 3200 | .align 32
|
|---|
| 3201 | .Lsqrx8x_reduce:
|
|---|
| 3202 | mov %r8, %rbx
|
|---|
| 3203 | mulx 8*0($nptr),%rax,%r8 # n[0]
|
|---|
| 3204 | adcx %rbx,%rax # discarded
|
|---|
| 3205 | adox %r9,%r8
|
|---|
| 3206 |
|
|---|
| 3207 | mulx 8*1($nptr),%rbx,%r9 # n[1]
|
|---|
| 3208 | adcx %rbx,%r8
|
|---|
| 3209 | adox %r10,%r9
|
|---|
| 3210 |
|
|---|
| 3211 | mulx 8*2($nptr),%rbx,%r10
|
|---|
| 3212 | adcx %rbx,%r9
|
|---|
| 3213 | adox %r11,%r10
|
|---|
| 3214 |
|
|---|
| 3215 | mulx 8*3($nptr),%rbx,%r11
|
|---|
| 3216 | adcx %rbx,%r10
|
|---|
| 3217 | adox %r12,%r11
|
|---|
| 3218 |
|
|---|
| 3219 | .byte 0xc4,0x62,0xe3,0xf6,0xa5,0x20,0x00,0x00,0x00 # mulx 8*4($nptr),%rbx,%r12
|
|---|
| 3220 | mov %rdx,%rax
|
|---|
| 3221 | mov %r8,%rdx
|
|---|
| 3222 | adcx %rbx,%r11
|
|---|
| 3223 | adox %r13,%r12
|
|---|
| 3224 |
|
|---|
| 3225 | mulx 32+8(%rsp),%rbx,%rdx # %rdx discarded
|
|---|
| 3226 | mov %rax,%rdx
|
|---|
| 3227 | mov %rax,64+48+8(%rsp,%rcx,8) # put aside n0*a[i]
|
|---|
| 3228 |
|
|---|
| 3229 | mulx 8*5($nptr),%rax,%r13
|
|---|
| 3230 | adcx %rax,%r12
|
|---|
| 3231 | adox %r14,%r13
|
|---|
| 3232 |
|
|---|
| 3233 | mulx 8*6($nptr),%rax,%r14
|
|---|
| 3234 | adcx %rax,%r13
|
|---|
| 3235 | adox %r15,%r14
|
|---|
| 3236 |
|
|---|
| 3237 | mulx 8*7($nptr),%rax,%r15
|
|---|
| 3238 | mov %rbx,%rdx
|
|---|
| 3239 | adcx %rax,%r14
|
|---|
| 3240 | adox $carry,%r15 # $carry is 0
|
|---|
| 3241 | adcx $carry,%r15 # cf=0
|
|---|
| 3242 |
|
|---|
| 3243 | .byte 0x67,0x67,0x67
|
|---|
| 3244 | inc %rcx # of=0
|
|---|
| 3245 | jnz .Lsqrx8x_reduce
|
|---|
| 3246 |
|
|---|
| 3247 | mov $carry,%rax # xor %rax,%rax
|
|---|
| 3248 | cmp 0+8(%rsp),$nptr # end of n[]?
|
|---|
| 3249 | jae .Lsqrx8x_no_tail
|
|---|
| 3250 |
|
|---|
| 3251 | mov 48+8(%rsp),%rdx # pull n0*a[0]
|
|---|
| 3252 | add 8*0($tptr),%r8
|
|---|
| 3253 | lea 8*8($nptr),$nptr
|
|---|
| 3254 | mov \$-8,%rcx
|
|---|
| 3255 | adcx 8*1($tptr),%r9
|
|---|
| 3256 | adcx 8*2($tptr),%r10
|
|---|
| 3257 | adc 8*3($tptr),%r11
|
|---|
| 3258 | adc 8*4($tptr),%r12
|
|---|
| 3259 | adc 8*5($tptr),%r13
|
|---|
| 3260 | adc 8*6($tptr),%r14
|
|---|
| 3261 | adc 8*7($tptr),%r15
|
|---|
| 3262 | lea 8*8($tptr),$tptr
|
|---|
| 3263 | sbb %rax,%rax # top carry
|
|---|
| 3264 |
|
|---|
| 3265 | xor $carry,$carry # of=0, cf=0
|
|---|
| 3266 | mov %rax,16+8(%rsp)
|
|---|
| 3267 | jmp .Lsqrx8x_tail
|
|---|
| 3268 |
|
|---|
| 3269 | .align 32
|
|---|
| 3270 | .Lsqrx8x_tail:
|
|---|
| 3271 | mov %r8,%rbx
|
|---|
| 3272 | mulx 8*0($nptr),%rax,%r8
|
|---|
| 3273 | adcx %rax,%rbx
|
|---|
| 3274 | adox %r9,%r8
|
|---|
| 3275 |
|
|---|
| 3276 | mulx 8*1($nptr),%rax,%r9
|
|---|
| 3277 | adcx %rax,%r8
|
|---|
| 3278 | adox %r10,%r9
|
|---|
| 3279 |
|
|---|
| 3280 | mulx 8*2($nptr),%rax,%r10
|
|---|
| 3281 | adcx %rax,%r9
|
|---|
| 3282 | adox %r11,%r10
|
|---|
| 3283 |
|
|---|
| 3284 | mulx 8*3($nptr),%rax,%r11
|
|---|
| 3285 | adcx %rax,%r10
|
|---|
| 3286 | adox %r12,%r11
|
|---|
| 3287 |
|
|---|
| 3288 | .byte 0xc4,0x62,0xfb,0xf6,0xa5,0x20,0x00,0x00,0x00 # mulx 8*4($nptr),%rax,%r12
|
|---|
| 3289 | adcx %rax,%r11
|
|---|
| 3290 | adox %r13,%r12
|
|---|
| 3291 |
|
|---|
| 3292 | mulx 8*5($nptr),%rax,%r13
|
|---|
| 3293 | adcx %rax,%r12
|
|---|
| 3294 | adox %r14,%r13
|
|---|
| 3295 |
|
|---|
| 3296 | mulx 8*6($nptr),%rax,%r14
|
|---|
| 3297 | adcx %rax,%r13
|
|---|
| 3298 | adox %r15,%r14
|
|---|
| 3299 |
|
|---|
| 3300 | mulx 8*7($nptr),%rax,%r15
|
|---|
| 3301 | mov 72+48+8(%rsp,%rcx,8),%rdx # pull n0*a[i]
|
|---|
| 3302 | adcx %rax,%r14
|
|---|
| 3303 | adox $carry,%r15
|
|---|
| 3304 | mov %rbx,($tptr,%rcx,8) # save result
|
|---|
| 3305 | mov %r8,%rbx
|
|---|
| 3306 | adcx $carry,%r15 # cf=0
|
|---|
| 3307 |
|
|---|
| 3308 | inc %rcx # of=0
|
|---|
| 3309 | jnz .Lsqrx8x_tail
|
|---|
| 3310 |
|
|---|
| 3311 | cmp 0+8(%rsp),$nptr # end of n[]?
|
|---|
| 3312 | jae .Lsqrx8x_tail_done # break out of loop
|
|---|
| 3313 |
|
|---|
| 3314 | sub 16+8(%rsp),$carry # mov 16(%rsp),%cf
|
|---|
| 3315 | mov 48+8(%rsp),%rdx # pull n0*a[0]
|
|---|
| 3316 | lea 8*8($nptr),$nptr
|
|---|
| 3317 | adc 8*0($tptr),%r8
|
|---|
| 3318 | adc 8*1($tptr),%r9
|
|---|
| 3319 | adc 8*2($tptr),%r10
|
|---|
| 3320 | adc 8*3($tptr),%r11
|
|---|
| 3321 | adc 8*4($tptr),%r12
|
|---|
| 3322 | adc 8*5($tptr),%r13
|
|---|
| 3323 | adc 8*6($tptr),%r14
|
|---|
| 3324 | adc 8*7($tptr),%r15
|
|---|
| 3325 | lea 8*8($tptr),$tptr
|
|---|
| 3326 | sbb %rax,%rax
|
|---|
| 3327 | sub \$8,%rcx # mov \$-8,%rcx
|
|---|
| 3328 |
|
|---|
| 3329 | xor $carry,$carry # of=0, cf=0
|
|---|
| 3330 | mov %rax,16+8(%rsp)
|
|---|
| 3331 | jmp .Lsqrx8x_tail
|
|---|
| 3332 |
|
|---|
| 3333 | .align 32
|
|---|
| 3334 | .Lsqrx8x_tail_done:
|
|---|
| 3335 | xor %rax,%rax
|
|---|
| 3336 | add 24+8(%rsp),%r8 # can this overflow?
|
|---|
| 3337 | adc \$0,%r9
|
|---|
| 3338 | adc \$0,%r10
|
|---|
| 3339 | adc \$0,%r11
|
|---|
| 3340 | adc \$0,%r12
|
|---|
| 3341 | adc \$0,%r13
|
|---|
| 3342 | adc \$0,%r14
|
|---|
| 3343 | adc \$0,%r15
|
|---|
| 3344 | adc \$0,%rax
|
|---|
| 3345 |
|
|---|
| 3346 | sub 16+8(%rsp),$carry # mov 16(%rsp),%cf
|
|---|
| 3347 | .Lsqrx8x_no_tail: # %cf is 0 if jumped here
|
|---|
| 3348 | adc 8*0($tptr),%r8
|
|---|
| 3349 | movq %xmm3,%rcx
|
|---|
| 3350 | adc 8*1($tptr),%r9
|
|---|
| 3351 | mov 8*7($nptr),$carry
|
|---|
| 3352 | movq %xmm2,$nptr # restore $nptr
|
|---|
| 3353 | adc 8*2($tptr),%r10
|
|---|
| 3354 | adc 8*3($tptr),%r11
|
|---|
| 3355 | adc 8*4($tptr),%r12
|
|---|
| 3356 | adc 8*5($tptr),%r13
|
|---|
| 3357 | adc 8*6($tptr),%r14
|
|---|
| 3358 | adc 8*7($tptr),%r15
|
|---|
| 3359 | adc \$0,%rax # top-most carry
|
|---|
| 3360 |
|
|---|
| 3361 | mov 32+8(%rsp),%rbx # n0
|
|---|
| 3362 | mov 8*8($tptr,%rcx),%rdx # modulo-scheduled "%r8"
|
|---|
| 3363 |
|
|---|
| 3364 | mov %r8,8*0($tptr) # store top 512 bits
|
|---|
| 3365 | lea 8*8($tptr),%r8 # borrow %r8
|
|---|
| 3366 | mov %r9,8*1($tptr)
|
|---|
| 3367 | mov %r10,8*2($tptr)
|
|---|
| 3368 | mov %r11,8*3($tptr)
|
|---|
| 3369 | mov %r12,8*4($tptr)
|
|---|
| 3370 | mov %r13,8*5($tptr)
|
|---|
| 3371 | mov %r14,8*6($tptr)
|
|---|
| 3372 | mov %r15,8*7($tptr)
|
|---|
| 3373 |
|
|---|
| 3374 | lea 8*8($tptr,%rcx),$tptr # start of current t[] window
|
|---|
| 3375 | cmp 8+8(%rsp),%r8 # end of t[]?
|
|---|
| 3376 | jb .Lsqrx8x_reduction_loop
|
|---|
| 3377 | ret
|
|---|
| 3378 | .cfi_endproc
|
|---|
| 3379 | .size bn_sqrx8x_internal,.-bn_sqrx8x_internal
|
|---|
| 3380 | ___
|
|---|
| 3381 | } |
|---|
| 3382 |
|
|---|
| 3383 | ##############################################################
|
|---|
| 3384 | # Post-condition, 4x unrolled
|
|---|
| 3385 | #
|
|---|
| 3386 | {
|
|---|
| 3387 | my ($rptr,$nptr)=("%rdx","%rbp");
|
|---|
| 3388 | $code.=<<___;
|
|---|
| 3389 | .align 32
|
|---|
| 3390 | __bn_postx4x_internal:
|
|---|
| 3391 | .cfi_startproc
|
|---|
| 3392 | mov 8*0($nptr),%r12
|
|---|
| 3393 | mov %rcx,%r10 # -$num
|
|---|
| 3394 | mov %rcx,%r9 # -$num
|
|---|
| 3395 | neg %rax
|
|---|
| 3396 | sar \$3+2,%rcx
|
|---|
| 3397 | #lea 48+8(%rsp,%r9),$tptr
|
|---|
| 3398 | movq %xmm1,$rptr # restore $rptr
|
|---|
| 3399 | movq %xmm1,$aptr # prepare for back-to-back call
|
|---|
| 3400 | dec %r12 # so that after 'not' we get -n[0]
|
|---|
| 3401 | mov 8*1($nptr),%r13
|
|---|
| 3402 | xor %r8,%r8
|
|---|
| 3403 | mov 8*2($nptr),%r14
|
|---|
| 3404 | mov 8*3($nptr),%r15
|
|---|
| 3405 | jmp .Lsqrx4x_sub_entry
|
|---|
| 3406 |
|
|---|
| 3407 | .align 16
|
|---|
| 3408 | .Lsqrx4x_sub:
|
|---|
| 3409 | mov 8*0($nptr),%r12
|
|---|
| 3410 | mov 8*1($nptr),%r13
|
|---|
| 3411 | mov 8*2($nptr),%r14
|
|---|
| 3412 | mov 8*3($nptr),%r15
|
|---|
| 3413 | .Lsqrx4x_sub_entry:
|
|---|
| 3414 | andn %rax,%r12,%r12
|
|---|
| 3415 | lea 8*4($nptr),$nptr
|
|---|
| 3416 | andn %rax,%r13,%r13
|
|---|
| 3417 | andn %rax,%r14,%r14
|
|---|
| 3418 | andn %rax,%r15,%r15
|
|---|
| 3419 |
|
|---|
| 3420 | neg %r8 # mov %r8,%cf
|
|---|
| 3421 | adc 8*0($tptr),%r12
|
|---|
| 3422 | adc 8*1($tptr),%r13
|
|---|
| 3423 | adc 8*2($tptr),%r14
|
|---|
| 3424 | adc 8*3($tptr),%r15
|
|---|
| 3425 | mov %r12,8*0($rptr)
|
|---|
| 3426 | lea 8*4($tptr),$tptr
|
|---|
| 3427 | mov %r13,8*1($rptr)
|
|---|
| 3428 | sbb %r8,%r8 # mov %cf,%r8
|
|---|
| 3429 | mov %r14,8*2($rptr)
|
|---|
| 3430 | mov %r15,8*3($rptr)
|
|---|
| 3431 | lea 8*4($rptr),$rptr
|
|---|
| 3432 |
|
|---|
| 3433 | inc %rcx
|
|---|
| 3434 | jnz .Lsqrx4x_sub
|
|---|
| 3435 |
|
|---|
| 3436 | neg %r9 # restore $num
|
|---|
| 3437 |
|
|---|
| 3438 | ret
|
|---|
| 3439 | .cfi_endproc
|
|---|
| 3440 | .size __bn_postx4x_internal,.-__bn_postx4x_internal
|
|---|
| 3441 | ___
|
|---|
| 3442 | }
|
|---|
| 3443 | }}}
|
|---|
| 3444 | {
|
|---|
| 3445 | my ($inp,$num,$tbl,$idx)=$win64?("%rcx","%edx","%r8", "%r9d") : # Win64 order
|
|---|
| 3446 | ("%rdi","%esi","%rdx","%ecx"); # Unix order
|
|---|
| 3447 | my $out=$inp;
|
|---|
| 3448 | my $STRIDE=2**5*8;
|
|---|
| 3449 | my $N=$STRIDE/4;
|
|---|
| 3450 |
|
|---|
| 3451 | $code.=<<___;
|
|---|
| 3452 | .globl bn_get_bits5
|
|---|
| 3453 | .type bn_get_bits5,\@abi-omnipotent
|
|---|
| 3454 | .align 16
|
|---|
| 3455 | bn_get_bits5:
|
|---|
| 3456 | .cfi_startproc
|
|---|
| 3457 | lea 0($inp),%r10
|
|---|
| 3458 | lea 1($inp),%r11
|
|---|
| 3459 | mov $num,%ecx
|
|---|
| 3460 | shr \$4,$num
|
|---|
| 3461 | and \$15,%ecx
|
|---|
| 3462 | lea -8(%ecx),%eax
|
|---|
| 3463 | cmp \$11,%ecx
|
|---|
| 3464 | cmova %r11,%r10
|
|---|
| 3465 | cmova %eax,%ecx
|
|---|
| 3466 | movzw (%r10,$num,2),%eax
|
|---|
| 3467 | shrl %cl,%eax
|
|---|
| 3468 | and \$31,%eax
|
|---|
| 3469 | ret
|
|---|
| 3470 | .cfi_endproc
|
|---|
| 3471 | .size bn_get_bits5,.-bn_get_bits5
|
|---|
| 3472 |
|
|---|
| 3473 | .globl bn_scatter5
|
|---|
| 3474 | .type bn_scatter5,\@abi-omnipotent
|
|---|
| 3475 | .align 16
|
|---|
| 3476 | bn_scatter5:
|
|---|
| 3477 | .cfi_startproc
|
|---|
| 3478 | cmp \$0, $num
|
|---|
| 3479 | jz .Lscatter_epilogue
|
|---|
| 3480 | lea ($tbl,$idx,8),$tbl
|
|---|
| 3481 | .Lscatter:
|
|---|
| 3482 | mov ($inp),%rax
|
|---|
| 3483 | lea 8($inp),$inp
|
|---|
| 3484 | mov %rax,($tbl)
|
|---|
| 3485 | lea 32*8($tbl),$tbl
|
|---|
| 3486 | sub \$1,$num
|
|---|
| 3487 | jnz .Lscatter
|
|---|
| 3488 | .Lscatter_epilogue:
|
|---|
| 3489 | ret
|
|---|
| 3490 | .cfi_endproc
|
|---|
| 3491 | .size bn_scatter5,.-bn_scatter5
|
|---|
| 3492 |
|
|---|
| 3493 | .globl bn_gather5
|
|---|
| 3494 | .type bn_gather5,\@abi-omnipotent
|
|---|
| 3495 | .align 32
|
|---|
| 3496 | bn_gather5:
|
|---|
| 3497 | .LSEH_begin_bn_gather5: # Win64 thing, but harmless in other cases
|
|---|
| 3498 | .cfi_startproc
|
|---|
| 3499 | # I can't trust assembler to use specific encoding:-(
|
|---|
| 3500 | .byte 0x4c,0x8d,0x14,0x24 #lea (%rsp),%r10
|
|---|
| 3501 | .byte 0x48,0x81,0xec,0x08,0x01,0x00,0x00 #sub $0x108,%rsp
|
|---|
| 3502 | lea .Linc(%rip),%rax
|
|---|
| 3503 | and \$-16,%rsp # shouldn't be formally required
|
|---|
| 3504 |
|
|---|
| 3505 | movd $idx,%xmm5
|
|---|
| 3506 | movdqa 0(%rax),%xmm0 # 00000001000000010000000000000000
|
|---|
| 3507 | movdqa 16(%rax),%xmm1 # 00000002000000020000000200000002
|
|---|
| 3508 | lea 128($tbl),%r11 # size optimization
|
|---|
| 3509 | lea 128(%rsp),%rax # size optimization
|
|---|
| 3510 |
|
|---|
| 3511 | pshufd \$0,%xmm5,%xmm5 # broadcast $idx
|
|---|
| 3512 | movdqa %xmm1,%xmm4
|
|---|
| 3513 | movdqa %xmm1,%xmm2
|
|---|
| 3514 | ___
|
|---|
| 3515 | ########################################################################
|
|---|
| 3516 | # calculate mask by comparing 0..31 to $idx and save result to stack
|
|---|
| 3517 | #
|
|---|
| 3518 | for($i=0;$i<$STRIDE/16;$i+=4) {
|
|---|
| 3519 | $code.=<<___;
|
|---|
| 3520 | paddd %xmm0,%xmm1
|
|---|
| 3521 | pcmpeqd %xmm5,%xmm0 # compare to 1,0
|
|---|
| 3522 | ___
|
|---|
| 3523 | $code.=<<___ if ($i);
|
|---|
| 3524 | movdqa %xmm3,`16*($i-1)-128`(%rax)
|
|---|
| 3525 | ___
|
|---|
| 3526 | $code.=<<___;
|
|---|
| 3527 | movdqa %xmm4,%xmm3
|
|---|
| 3528 |
|
|---|
| 3529 | paddd %xmm1,%xmm2
|
|---|
| 3530 | pcmpeqd %xmm5,%xmm1 # compare to 3,2
|
|---|
| 3531 | movdqa %xmm0,`16*($i+0)-128`(%rax)
|
|---|
| 3532 | movdqa %xmm4,%xmm0
|
|---|
| 3533 |
|
|---|
| 3534 | paddd %xmm2,%xmm3
|
|---|
| 3535 | pcmpeqd %xmm5,%xmm2 # compare to 5,4
|
|---|
| 3536 | movdqa %xmm1,`16*($i+1)-128`(%rax)
|
|---|
| 3537 | movdqa %xmm4,%xmm1
|
|---|
| 3538 |
|
|---|
| 3539 | paddd %xmm3,%xmm0
|
|---|
| 3540 | pcmpeqd %xmm5,%xmm3 # compare to 7,6
|
|---|
| 3541 | movdqa %xmm2,`16*($i+2)-128`(%rax)
|
|---|
| 3542 | movdqa %xmm4,%xmm2
|
|---|
| 3543 | ___
|
|---|
| 3544 | }
|
|---|
| 3545 | $code.=<<___;
|
|---|
| 3546 | movdqa %xmm3,`16*($i-1)-128`(%rax)
|
|---|
| 3547 | jmp .Lgather
|
|---|
| 3548 |
|
|---|
| 3549 | .align 32
|
|---|
| 3550 | .Lgather:
|
|---|
| 3551 | pxor %xmm4,%xmm4
|
|---|
| 3552 | pxor %xmm5,%xmm5
|
|---|
| 3553 | ___
|
|---|
| 3554 | for($i=0;$i<$STRIDE/16;$i+=4) {
|
|---|
| 3555 | $code.=<<___;
|
|---|
| 3556 | movdqa `16*($i+0)-128`(%r11),%xmm0
|
|---|
| 3557 | movdqa `16*($i+1)-128`(%r11),%xmm1
|
|---|
| 3558 | movdqa `16*($i+2)-128`(%r11),%xmm2
|
|---|
| 3559 | pand `16*($i+0)-128`(%rax),%xmm0
|
|---|
| 3560 | movdqa `16*($i+3)-128`(%r11),%xmm3
|
|---|
| 3561 | pand `16*($i+1)-128`(%rax),%xmm1
|
|---|
| 3562 | por %xmm0,%xmm4
|
|---|
| 3563 | pand `16*($i+2)-128`(%rax),%xmm2
|
|---|
| 3564 | por %xmm1,%xmm5
|
|---|
| 3565 | pand `16*($i+3)-128`(%rax),%xmm3
|
|---|
| 3566 | por %xmm2,%xmm4
|
|---|
| 3567 | por %xmm3,%xmm5
|
|---|
| 3568 | ___
|
|---|
| 3569 | }
|
|---|
| 3570 | $code.=<<___;
|
|---|
| 3571 | por %xmm5,%xmm4
|
|---|
| 3572 | lea $STRIDE(%r11),%r11
|
|---|
| 3573 | pshufd \$0x4e,%xmm4,%xmm0
|
|---|
| 3574 | por %xmm4,%xmm0
|
|---|
| 3575 | movq %xmm0,($out) # m0=bp[0]
|
|---|
| 3576 | lea 8($out),$out
|
|---|
| 3577 | sub \$1,$num
|
|---|
| 3578 | jnz .Lgather
|
|---|
| 3579 |
|
|---|
| 3580 | lea (%r10),%rsp
|
|---|
| 3581 | ret
|
|---|
| 3582 | .LSEH_end_bn_gather5:
|
|---|
| 3583 | .cfi_endproc
|
|---|
| 3584 | .size bn_gather5,.-bn_gather5
|
|---|
| 3585 | ___
|
|---|
| 3586 | }
|
|---|
| 3587 | $code.=<<___;
|
|---|
| 3588 | .align 64
|
|---|
| 3589 | .Linc:
|
|---|
| 3590 | .long 0,0, 1,1
|
|---|
| 3591 | .long 2,2, 2,2
|
|---|
| 3592 | .asciz "Montgomery Multiplication with scatter/gather for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
|
|---|
| 3593 | ___
|
|---|
| 3594 |
|
|---|
| 3595 | # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
|
|---|
| 3596 | # CONTEXT *context,DISPATCHER_CONTEXT *disp)
|
|---|
| 3597 | if ($win64) {
|
|---|
| 3598 | $rec="%rcx";
|
|---|
| 3599 | $frame="%rdx";
|
|---|
| 3600 | $context="%r8";
|
|---|
| 3601 | $disp="%r9";
|
|---|
| 3602 |
|
|---|
| 3603 | $code.=<<___;
|
|---|
| 3604 | .extern __imp_RtlVirtualUnwind
|
|---|
| 3605 | .type mul_handler,\@abi-omnipotent
|
|---|
| 3606 | .align 16
|
|---|
| 3607 | mul_handler:
|
|---|
| 3608 | push %rsi
|
|---|
| 3609 | push %rdi
|
|---|
| 3610 | push %rbx
|
|---|
| 3611 | push %rbp
|
|---|
| 3612 | push %r12
|
|---|
| 3613 | push %r13
|
|---|
| 3614 | push %r14
|
|---|
| 3615 | push %r15
|
|---|
| 3616 | pushfq
|
|---|
| 3617 | sub \$64,%rsp
|
|---|
| 3618 |
|
|---|
| 3619 | mov 120($context),%rax # pull context->Rax
|
|---|
| 3620 | mov 248($context),%rbx # pull context->Rip
|
|---|
| 3621 |
|
|---|
| 3622 | mov 8($disp),%rsi # disp->ImageBase
|
|---|
| 3623 | mov 56($disp),%r11 # disp->HandlerData
|
|---|
| 3624 |
|
|---|
| 3625 | mov 0(%r11),%r10d # HandlerData[0]
|
|---|
| 3626 | lea (%rsi,%r10),%r10 # end of prologue label
|
|---|
| 3627 | cmp %r10,%rbx # context->Rip<end of prologue label
|
|---|
| 3628 | jb .Lcommon_seh_tail
|
|---|
| 3629 |
|
|---|
| 3630 | mov 4(%r11),%r10d # HandlerData[1]
|
|---|
| 3631 | lea (%rsi,%r10),%r10 # beginning of body label
|
|---|
| 3632 | cmp %r10,%rbx # context->Rip<body label
|
|---|
| 3633 | jb .Lcommon_pop_regs
|
|---|
| 3634 |
|
|---|
| 3635 | mov 152($context),%rax # pull context->Rsp
|
|---|
| 3636 |
|
|---|
| 3637 | mov 8(%r11),%r10d # HandlerData[2]
|
|---|
| 3638 | lea (%rsi,%r10),%r10 # epilogue label
|
|---|
| 3639 | cmp %r10,%rbx # context->Rip>=epilogue label
|
|---|
| 3640 | jae .Lcommon_seh_tail
|
|---|
| 3641 |
|
|---|
| 3642 | lea .Lmul_epilogue(%rip),%r10
|
|---|
| 3643 | cmp %r10,%rbx
|
|---|
| 3644 | ja .Lbody_40
|
|---|
| 3645 |
|
|---|
| 3646 | mov 192($context),%r10 # pull $num
|
|---|
| 3647 | mov 8(%rax,%r10,8),%rax # pull saved stack pointer
|
|---|
| 3648 |
|
|---|
| 3649 | jmp .Lcommon_pop_regs
|
|---|
| 3650 |
|
|---|
| 3651 | .Lbody_40:
|
|---|
| 3652 | mov 40(%rax),%rax # pull saved stack pointer
|
|---|
| 3653 | .Lcommon_pop_regs:
|
|---|
| 3654 | mov -8(%rax),%rbx
|
|---|
| 3655 | mov -16(%rax),%rbp
|
|---|
| 3656 | mov -24(%rax),%r12
|
|---|
| 3657 | mov -32(%rax),%r13
|
|---|
| 3658 | mov -40(%rax),%r14
|
|---|
| 3659 | mov -48(%rax),%r15
|
|---|
| 3660 | mov %rbx,144($context) # restore context->Rbx
|
|---|
| 3661 | mov %rbp,160($context) # restore context->Rbp
|
|---|
| 3662 | mov %r12,216($context) # restore context->R12
|
|---|
| 3663 | mov %r13,224($context) # restore context->R13
|
|---|
| 3664 | mov %r14,232($context) # restore context->R14
|
|---|
| 3665 | mov %r15,240($context) # restore context->R15
|
|---|
| 3666 |
|
|---|
| 3667 | .Lcommon_seh_tail:
|
|---|
| 3668 | mov 8(%rax),%rdi
|
|---|
| 3669 | mov 16(%rax),%rsi
|
|---|
| 3670 | mov %rax,152($context) # restore context->Rsp
|
|---|
| 3671 | mov %rsi,168($context) # restore context->Rsi
|
|---|
| 3672 | mov %rdi,176($context) # restore context->Rdi
|
|---|
| 3673 |
|
|---|
| 3674 | mov 40($disp),%rdi # disp->ContextRecord
|
|---|
| 3675 | mov $context,%rsi # context
|
|---|
| 3676 | mov \$154,%ecx # sizeof(CONTEXT)
|
|---|
| 3677 | .long 0xa548f3fc # cld; rep movsq
|
|---|
| 3678 |
|
|---|
| 3679 | mov $disp,%rsi
|
|---|
| 3680 | xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
|
|---|
| 3681 | mov 8(%rsi),%rdx # arg2, disp->ImageBase
|
|---|
| 3682 | mov 0(%rsi),%r8 # arg3, disp->ControlPc
|
|---|
| 3683 | mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
|
|---|
| 3684 | mov 40(%rsi),%r10 # disp->ContextRecord
|
|---|
| 3685 | lea 56(%rsi),%r11 # &disp->HandlerData
|
|---|
| 3686 | lea 24(%rsi),%r12 # &disp->EstablisherFrame
|
|---|
| 3687 | mov %r10,32(%rsp) # arg5
|
|---|
| 3688 | mov %r11,40(%rsp) # arg6
|
|---|
| 3689 | mov %r12,48(%rsp) # arg7
|
|---|
| 3690 | mov %rcx,56(%rsp) # arg8, (NULL)
|
|---|
| 3691 | call *__imp_RtlVirtualUnwind(%rip)
|
|---|
| 3692 |
|
|---|
| 3693 | mov \$1,%eax # ExceptionContinueSearch
|
|---|
| 3694 | add \$64,%rsp
|
|---|
| 3695 | popfq
|
|---|
| 3696 | pop %r15
|
|---|
| 3697 | pop %r14
|
|---|
| 3698 | pop %r13
|
|---|
| 3699 | pop %r12
|
|---|
| 3700 | pop %rbp
|
|---|
| 3701 | pop %rbx
|
|---|
| 3702 | pop %rdi
|
|---|
| 3703 | pop %rsi
|
|---|
| 3704 | ret
|
|---|
| 3705 | .size mul_handler,.-mul_handler
|
|---|
| 3706 |
|
|---|
| 3707 | .section .pdata
|
|---|
| 3708 | .align 4
|
|---|
| 3709 | .rva .LSEH_begin_bn_mul_mont_gather5
|
|---|
| 3710 | .rva .LSEH_end_bn_mul_mont_gather5
|
|---|
| 3711 | .rva .LSEH_info_bn_mul_mont_gather5
|
|---|
| 3712 |
|
|---|
| 3713 | .rva .LSEH_begin_bn_mul4x_mont_gather5
|
|---|
| 3714 | .rva .LSEH_end_bn_mul4x_mont_gather5
|
|---|
| 3715 | .rva .LSEH_info_bn_mul4x_mont_gather5
|
|---|
| 3716 |
|
|---|
| 3717 | .rva .LSEH_begin_bn_power5
|
|---|
| 3718 | .rva .LSEH_end_bn_power5
|
|---|
| 3719 | .rva .LSEH_info_bn_power5
|
|---|
| 3720 | ___
|
|---|
| 3721 | $code.=<<___ if ($addx);
|
|---|
| 3722 | .rva .LSEH_begin_bn_mulx4x_mont_gather5
|
|---|
| 3723 | .rva .LSEH_end_bn_mulx4x_mont_gather5
|
|---|
| 3724 | .rva .LSEH_info_bn_mulx4x_mont_gather5
|
|---|
| 3725 |
|
|---|
| 3726 | .rva .LSEH_begin_bn_powerx5
|
|---|
| 3727 | .rva .LSEH_end_bn_powerx5
|
|---|
| 3728 | .rva .LSEH_info_bn_powerx5
|
|---|
| 3729 | ___
|
|---|
| 3730 | $code.=<<___;
|
|---|
| 3731 | .rva .LSEH_begin_bn_gather5
|
|---|
| 3732 | .rva .LSEH_end_bn_gather5
|
|---|
| 3733 | .rva .LSEH_info_bn_gather5
|
|---|
| 3734 |
|
|---|
| 3735 | .section .xdata
|
|---|
| 3736 | .align 8
|
|---|
| 3737 | .LSEH_info_bn_mul_mont_gather5:
|
|---|
| 3738 | .byte 9,0,0,0
|
|---|
| 3739 | .rva mul_handler
|
|---|
| 3740 | .rva .Lmul_body,.Lmul_body,.Lmul_epilogue # HandlerData[]
|
|---|
| 3741 | .align 8
|
|---|
| 3742 | .LSEH_info_bn_mul4x_mont_gather5:
|
|---|
| 3743 | .byte 9,0,0,0
|
|---|
| 3744 | .rva mul_handler
|
|---|
| 3745 | .rva .Lmul4x_prologue,.Lmul4x_body,.Lmul4x_epilogue # HandlerData[]
|
|---|
| 3746 | .align 8
|
|---|
| 3747 | .LSEH_info_bn_power5:
|
|---|
| 3748 | .byte 9,0,0,0
|
|---|
| 3749 | .rva mul_handler
|
|---|
| 3750 | .rva .Lpower5_prologue,.Lpower5_body,.Lpower5_epilogue # HandlerData[]
|
|---|
| 3751 | ___
|
|---|
| 3752 | $code.=<<___ if ($addx);
|
|---|
| 3753 | .align 8
|
|---|
| 3754 | .LSEH_info_bn_mulx4x_mont_gather5:
|
|---|
| 3755 | .byte 9,0,0,0
|
|---|
| 3756 | .rva mul_handler
|
|---|
| 3757 | .rva .Lmulx4x_prologue,.Lmulx4x_body,.Lmulx4x_epilogue # HandlerData[]
|
|---|
| 3758 | .align 8
|
|---|
| 3759 | .LSEH_info_bn_powerx5:
|
|---|
| 3760 | .byte 9,0,0,0
|
|---|
| 3761 | .rva mul_handler
|
|---|
| 3762 | .rva .Lpowerx5_prologue,.Lpowerx5_body,.Lpowerx5_epilogue # HandlerData[]
|
|---|
| 3763 | ___
|
|---|
| 3764 | $code.=<<___;
|
|---|
| 3765 | .align 8
|
|---|
| 3766 | .LSEH_info_bn_gather5:
|
|---|
| 3767 | .byte 0x01,0x0b,0x03,0x0a
|
|---|
| 3768 | .byte 0x0b,0x01,0x21,0x00 # sub rsp,0x108
|
|---|
| 3769 | .byte 0x04,0xa3,0x00,0x00 # lea r10,(rsp)
|
|---|
| 3770 | .align 8
|
|---|
| 3771 | ___
|
|---|
| 3772 | }
|
|---|
| 3773 |
|
|---|
| 3774 | $code =~ s/\`([^\`]*)\`/eval($1)/gem;
|
|---|
| 3775 |
|
|---|
| 3776 | print $code;
|
|---|
| 3777 | close STDOUT or die "error closing STDOUT: $!";
|
|---|
