| 1 | #! /usr/bin/env perl
|
|---|
| 2 | # Copyright 2011-2016 The OpenSSL Project Authors. All Rights Reserved.
|
|---|
| 3 | #
|
|---|
| 4 | # Licensed under the OpenSSL license (the "License"). You may not use
|
|---|
| 5 | # this file except in compliance with the License. You can obtain a copy
|
|---|
| 6 | # in the file LICENSE in the source distribution or at
|
|---|
| 7 | # https://www.openssl.org/source/license.html
|
|---|
| 8 |
|
|---|
| 9 | #
|
|---|
| 10 | # ====================================================================
|
|---|
| 11 | # Written by Andy Polyakov <[email protected]> for the OpenSSL
|
|---|
| 12 | # project. The module is, however, dual licensed under OpenSSL and
|
|---|
| 13 | # CRYPTOGAMS licenses depending on where you obtain it. For further
|
|---|
| 14 | # details see http://www.openssl.org/~appro/cryptogams/.
|
|---|
| 15 | # ====================================================================
|
|---|
| 16 | #
|
|---|
| 17 | # May 2011
|
|---|
| 18 | #
|
|---|
| 19 | # The module implements bn_GF2m_mul_2x2 polynomial multiplication used
|
|---|
| 20 | # in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
|
|---|
| 21 | # the time being... Except that it has two code paths: code suitable
|
|---|
| 22 | # for any x86_64 CPU and PCLMULQDQ one suitable for Westmere and
|
|---|
| 23 | # later. Improvement varies from one benchmark and µ-arch to another.
|
|---|
| 24 | # Vanilla code path is at most 20% faster than compiler-generated code
|
|---|
| 25 | # [not very impressive], while PCLMULQDQ - whole 85%-160% better on
|
|---|
| 26 | # 163- and 571-bit ECDH benchmarks on Intel CPUs. Keep in mind that
|
|---|
| 27 | # these coefficients are not ones for bn_GF2m_mul_2x2 itself, as not
|
|---|
| 28 | # all CPU time is burnt in it...
|
|---|
| 29 |
|
|---|
| 30 | $flavour = shift;
|
|---|
| 31 | $output = shift;
|
|---|
| 32 | if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
|
|---|
| 33 |
|
|---|
| 34 | $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
|
|---|
| 35 |
|
|---|
| 36 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
|
|---|
| 37 | ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
|
|---|
| 38 | ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
|
|---|
| 39 | die "can't locate x86_64-xlate.pl";
|
|---|
| 40 |
|
|---|
| 41 | open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
|
|---|
| 42 | *STDOUT=*OUT;
|
|---|
| 43 |
|
|---|
| 44 | ($lo,$hi)=("%rax","%rdx"); $a=$lo;
|
|---|
| 45 | ($i0,$i1)=("%rsi","%rdi");
|
|---|
| 46 | ($t0,$t1)=("%rbx","%rcx");
|
|---|
| 47 | ($b,$mask)=("%rbp","%r8");
|
|---|
| 48 | ($a1,$a2,$a4,$a8,$a12,$a48)=map("%r$_",(9..15));
|
|---|
| 49 | ($R,$Tx)=("%xmm0","%xmm1");
|
|---|
| 50 |
|
|---|
| 51 | $code.=<<___;
|
|---|
| 52 | .text
|
|---|
| 53 |
|
|---|
| 54 | .type _mul_1x1,\@abi-omnipotent
|
|---|
| 55 | .align 16
|
|---|
| 56 | _mul_1x1:
|
|---|
| 57 | sub \$128+8,%rsp
|
|---|
| 58 | mov \$-1,$a1
|
|---|
| 59 | lea ($a,$a),$i0
|
|---|
| 60 | shr \$3,$a1
|
|---|
| 61 | lea (,$a,4),$i1
|
|---|
| 62 | and $a,$a1 # a1=a&0x1fffffffffffffff
|
|---|
| 63 | lea (,$a,8),$a8
|
|---|
| 64 | sar \$63,$a # broadcast 63rd bit
|
|---|
| 65 | lea ($a1,$a1),$a2
|
|---|
| 66 | sar \$63,$i0 # broadcast 62nd bit
|
|---|
| 67 | lea (,$a1,4),$a4
|
|---|
| 68 | and $b,$a
|
|---|
| 69 | sar \$63,$i1 # boardcast 61st bit
|
|---|
| 70 | mov $a,$hi # $a is $lo
|
|---|
| 71 | shl \$63,$lo
|
|---|
| 72 | and $b,$i0
|
|---|
| 73 | shr \$1,$hi
|
|---|
| 74 | mov $i0,$t1
|
|---|
| 75 | shl \$62,$i0
|
|---|
| 76 | and $b,$i1
|
|---|
| 77 | shr \$2,$t1
|
|---|
| 78 | xor $i0,$lo
|
|---|
| 79 | mov $i1,$t0
|
|---|
| 80 | shl \$61,$i1
|
|---|
| 81 | xor $t1,$hi
|
|---|
| 82 | shr \$3,$t0
|
|---|
| 83 | xor $i1,$lo
|
|---|
| 84 | xor $t0,$hi
|
|---|
| 85 |
|
|---|
| 86 | mov $a1,$a12
|
|---|
| 87 | movq \$0,0(%rsp) # tab[0]=0
|
|---|
| 88 | xor $a2,$a12 # a1^a2
|
|---|
| 89 | mov $a1,8(%rsp) # tab[1]=a1
|
|---|
| 90 | mov $a4,$a48
|
|---|
| 91 | mov $a2,16(%rsp) # tab[2]=a2
|
|---|
| 92 | xor $a8,$a48 # a4^a8
|
|---|
| 93 | mov $a12,24(%rsp) # tab[3]=a1^a2
|
|---|
| 94 |
|
|---|
| 95 | xor $a4,$a1
|
|---|
| 96 | mov $a4,32(%rsp) # tab[4]=a4
|
|---|
| 97 | xor $a4,$a2
|
|---|
| 98 | mov $a1,40(%rsp) # tab[5]=a1^a4
|
|---|
| 99 | xor $a4,$a12
|
|---|
| 100 | mov $a2,48(%rsp) # tab[6]=a2^a4
|
|---|
| 101 | xor $a48,$a1 # a1^a4^a4^a8=a1^a8
|
|---|
| 102 | mov $a12,56(%rsp) # tab[7]=a1^a2^a4
|
|---|
| 103 | xor $a48,$a2 # a2^a4^a4^a8=a1^a8
|
|---|
| 104 |
|
|---|
| 105 | mov $a8,64(%rsp) # tab[8]=a8
|
|---|
| 106 | xor $a48,$a12 # a1^a2^a4^a4^a8=a1^a2^a8
|
|---|
| 107 | mov $a1,72(%rsp) # tab[9]=a1^a8
|
|---|
| 108 | xor $a4,$a1 # a1^a8^a4
|
|---|
| 109 | mov $a2,80(%rsp) # tab[10]=a2^a8
|
|---|
| 110 | xor $a4,$a2 # a2^a8^a4
|
|---|
| 111 | mov $a12,88(%rsp) # tab[11]=a1^a2^a8
|
|---|
| 112 |
|
|---|
| 113 | xor $a4,$a12 # a1^a2^a8^a4
|
|---|
| 114 | mov $a48,96(%rsp) # tab[12]=a4^a8
|
|---|
| 115 | mov $mask,$i0
|
|---|
| 116 | mov $a1,104(%rsp) # tab[13]=a1^a4^a8
|
|---|
| 117 | and $b,$i0
|
|---|
| 118 | mov $a2,112(%rsp) # tab[14]=a2^a4^a8
|
|---|
| 119 | shr \$4,$b
|
|---|
| 120 | mov $a12,120(%rsp) # tab[15]=a1^a2^a4^a8
|
|---|
| 121 | mov $mask,$i1
|
|---|
| 122 | and $b,$i1
|
|---|
| 123 | shr \$4,$b
|
|---|
| 124 |
|
|---|
| 125 | movq (%rsp,$i0,8),$R # half of calculations is done in SSE2
|
|---|
| 126 | mov $mask,$i0
|
|---|
| 127 | and $b,$i0
|
|---|
| 128 | shr \$4,$b
|
|---|
| 129 | ___
|
|---|
| 130 | for ($n=1;$n<8;$n++) {
|
|---|
| 131 | $code.=<<___;
|
|---|
| 132 | mov (%rsp,$i1,8),$t1
|
|---|
| 133 | mov $mask,$i1
|
|---|
| 134 | mov $t1,$t0
|
|---|
| 135 | shl \$`8*$n-4`,$t1
|
|---|
| 136 | and $b,$i1
|
|---|
| 137 | movq (%rsp,$i0,8),$Tx
|
|---|
| 138 | shr \$`64-(8*$n-4)`,$t0
|
|---|
| 139 | xor $t1,$lo
|
|---|
| 140 | pslldq \$$n,$Tx
|
|---|
| 141 | mov $mask,$i0
|
|---|
| 142 | shr \$4,$b
|
|---|
| 143 | xor $t0,$hi
|
|---|
| 144 | and $b,$i0
|
|---|
| 145 | shr \$4,$b
|
|---|
| 146 | pxor $Tx,$R
|
|---|
| 147 | ___
|
|---|
| 148 | }
|
|---|
| 149 | $code.=<<___;
|
|---|
| 150 | mov (%rsp,$i1,8),$t1
|
|---|
| 151 | mov $t1,$t0
|
|---|
| 152 | shl \$`8*$n-4`,$t1
|
|---|
| 153 | movq $R,$i0
|
|---|
| 154 | shr \$`64-(8*$n-4)`,$t0
|
|---|
| 155 | xor $t1,$lo
|
|---|
| 156 | psrldq \$8,$R
|
|---|
| 157 | xor $t0,$hi
|
|---|
| 158 | movq $R,$i1
|
|---|
| 159 | xor $i0,$lo
|
|---|
| 160 | xor $i1,$hi
|
|---|
| 161 |
|
|---|
| 162 | add \$128+8,%rsp
|
|---|
| 163 | ret
|
|---|
| 164 | .Lend_mul_1x1:
|
|---|
| 165 | .size _mul_1x1,.-_mul_1x1
|
|---|
| 166 | ___
|
|---|
| 167 |
|
|---|
| 168 | ($rp,$a1,$a0,$b1,$b0) = $win64? ("%rcx","%rdx","%r8", "%r9","%r10") : # Win64 order
|
|---|
| 169 | ("%rdi","%rsi","%rdx","%rcx","%r8"); # Unix order
|
|---|
| 170 |
|
|---|
| 171 | $code.=<<___;
|
|---|
| 172 | .extern OPENSSL_ia32cap_P
|
|---|
| 173 | .globl bn_GF2m_mul_2x2
|
|---|
| 174 | .type bn_GF2m_mul_2x2,\@abi-omnipotent
|
|---|
| 175 | .align 16
|
|---|
| 176 | bn_GF2m_mul_2x2:
|
|---|
| 177 | mov OPENSSL_ia32cap_P(%rip),%rax
|
|---|
| 178 | bt \$33,%rax
|
|---|
| 179 | jnc .Lvanilla_mul_2x2
|
|---|
| 180 |
|
|---|
| 181 | movq $a1,%xmm0
|
|---|
| 182 | movq $b1,%xmm1
|
|---|
| 183 | movq $a0,%xmm2
|
|---|
| 184 | ___
|
|---|
| 185 | $code.=<<___ if ($win64);
|
|---|
| 186 | movq 40(%rsp),%xmm3
|
|---|
| 187 | ___
|
|---|
| 188 | $code.=<<___ if (!$win64);
|
|---|
| 189 | movq $b0,%xmm3
|
|---|
| 190 | ___
|
|---|
| 191 | $code.=<<___;
|
|---|
| 192 | movdqa %xmm0,%xmm4
|
|---|
| 193 | movdqa %xmm1,%xmm5
|
|---|
| 194 | pclmulqdq \$0,%xmm1,%xmm0 # a1·b1
|
|---|
| 195 | pxor %xmm2,%xmm4
|
|---|
| 196 | pxor %xmm3,%xmm5
|
|---|
| 197 | pclmulqdq \$0,%xmm3,%xmm2 # a0·b0
|
|---|
| 198 | pclmulqdq \$0,%xmm5,%xmm4 # (a0+a1)·(b0+b1)
|
|---|
| 199 | xorps %xmm0,%xmm4
|
|---|
| 200 | xorps %xmm2,%xmm4 # (a0+a1)·(b0+b1)-a0·b0-a1·b1
|
|---|
| 201 | movdqa %xmm4,%xmm5
|
|---|
| 202 | pslldq \$8,%xmm4
|
|---|
| 203 | psrldq \$8,%xmm5
|
|---|
| 204 | pxor %xmm4,%xmm2
|
|---|
| 205 | pxor %xmm5,%xmm0
|
|---|
| 206 | movdqu %xmm2,0($rp)
|
|---|
| 207 | movdqu %xmm0,16($rp)
|
|---|
| 208 | ret
|
|---|
| 209 |
|
|---|
| 210 | .align 16
|
|---|
| 211 | .Lvanilla_mul_2x2:
|
|---|
| 212 | lea -8*17(%rsp),%rsp
|
|---|
| 213 | ___
|
|---|
| 214 | $code.=<<___ if ($win64);
|
|---|
| 215 | mov `8*17+40`(%rsp),$b0
|
|---|
| 216 | mov %rdi,8*15(%rsp)
|
|---|
| 217 | mov %rsi,8*16(%rsp)
|
|---|
| 218 | ___
|
|---|
| 219 | $code.=<<___;
|
|---|
| 220 | mov %r14,8*10(%rsp)
|
|---|
| 221 | mov %r13,8*11(%rsp)
|
|---|
| 222 | mov %r12,8*12(%rsp)
|
|---|
| 223 | mov %rbp,8*13(%rsp)
|
|---|
| 224 | mov %rbx,8*14(%rsp)
|
|---|
| 225 | .Lbody_mul_2x2:
|
|---|
| 226 | mov $rp,32(%rsp) # save the arguments
|
|---|
| 227 | mov $a1,40(%rsp)
|
|---|
| 228 | mov $a0,48(%rsp)
|
|---|
| 229 | mov $b1,56(%rsp)
|
|---|
| 230 | mov $b0,64(%rsp)
|
|---|
| 231 |
|
|---|
| 232 | mov \$0xf,$mask
|
|---|
| 233 | mov $a1,$a
|
|---|
| 234 | mov $b1,$b
|
|---|
| 235 | call _mul_1x1 # a1·b1
|
|---|
| 236 | mov $lo,16(%rsp)
|
|---|
| 237 | mov $hi,24(%rsp)
|
|---|
| 238 |
|
|---|
| 239 | mov 48(%rsp),$a
|
|---|
| 240 | mov 64(%rsp),$b
|
|---|
| 241 | call _mul_1x1 # a0·b0
|
|---|
| 242 | mov $lo,0(%rsp)
|
|---|
| 243 | mov $hi,8(%rsp)
|
|---|
| 244 |
|
|---|
| 245 | mov 40(%rsp),$a
|
|---|
| 246 | mov 56(%rsp),$b
|
|---|
| 247 | xor 48(%rsp),$a
|
|---|
| 248 | xor 64(%rsp),$b
|
|---|
| 249 | call _mul_1x1 # (a0+a1)·(b0+b1)
|
|---|
| 250 | ___
|
|---|
| 251 | @r=("%rbx","%rcx","%rdi","%rsi");
|
|---|
| 252 | $code.=<<___;
|
|---|
| 253 | mov 0(%rsp),@r[0]
|
|---|
| 254 | mov 8(%rsp),@r[1]
|
|---|
| 255 | mov 16(%rsp),@r[2]
|
|---|
| 256 | mov 24(%rsp),@r[3]
|
|---|
| 257 | mov 32(%rsp),%rbp
|
|---|
| 258 |
|
|---|
| 259 | xor $hi,$lo
|
|---|
| 260 | xor @r[1],$hi
|
|---|
| 261 | xor @r[0],$lo
|
|---|
| 262 | mov @r[0],0(%rbp)
|
|---|
| 263 | xor @r[2],$hi
|
|---|
| 264 | mov @r[3],24(%rbp)
|
|---|
| 265 | xor @r[3],$lo
|
|---|
| 266 | xor @r[3],$hi
|
|---|
| 267 | xor $hi,$lo
|
|---|
| 268 | mov $hi,16(%rbp)
|
|---|
| 269 | mov $lo,8(%rbp)
|
|---|
| 270 |
|
|---|
| 271 | mov 8*10(%rsp),%r14
|
|---|
| 272 | mov 8*11(%rsp),%r13
|
|---|
| 273 | mov 8*12(%rsp),%r12
|
|---|
| 274 | mov 8*13(%rsp),%rbp
|
|---|
| 275 | mov 8*14(%rsp),%rbx
|
|---|
| 276 | ___
|
|---|
| 277 | $code.=<<___ if ($win64);
|
|---|
| 278 | mov 8*15(%rsp),%rdi
|
|---|
| 279 | mov 8*16(%rsp),%rsi
|
|---|
| 280 | ___
|
|---|
| 281 | $code.=<<___;
|
|---|
| 282 | lea 8*17(%rsp),%rsp
|
|---|
| 283 | ret
|
|---|
| 284 | .Lend_mul_2x2:
|
|---|
| 285 | .size bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
|
|---|
| 286 | .asciz "GF(2^m) Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
|
|---|
| 287 | .align 16
|
|---|
| 288 | ___
|
|---|
| 289 |
|
|---|
| 290 | # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
|
|---|
| 291 | # CONTEXT *context,DISPATCHER_CONTEXT *disp)
|
|---|
| 292 | if ($win64) {
|
|---|
| 293 | $rec="%rcx";
|
|---|
| 294 | $frame="%rdx";
|
|---|
| 295 | $context="%r8";
|
|---|
| 296 | $disp="%r9";
|
|---|
| 297 |
|
|---|
| 298 | $code.=<<___;
|
|---|
| 299 | .extern __imp_RtlVirtualUnwind
|
|---|
| 300 |
|
|---|
| 301 | .type se_handler,\@abi-omnipotent
|
|---|
| 302 | .align 16
|
|---|
| 303 | se_handler:
|
|---|
| 304 | push %rsi
|
|---|
| 305 | push %rdi
|
|---|
| 306 | push %rbx
|
|---|
| 307 | push %rbp
|
|---|
| 308 | push %r12
|
|---|
| 309 | push %r13
|
|---|
| 310 | push %r14
|
|---|
| 311 | push %r15
|
|---|
| 312 | pushfq
|
|---|
| 313 | sub \$64,%rsp
|
|---|
| 314 |
|
|---|
| 315 | mov 152($context),%rax # pull context->Rsp
|
|---|
| 316 | mov 248($context),%rbx # pull context->Rip
|
|---|
| 317 |
|
|---|
| 318 | lea .Lbody_mul_2x2(%rip),%r10
|
|---|
| 319 | cmp %r10,%rbx # context->Rip<"prologue" label
|
|---|
| 320 | jb .Lin_prologue
|
|---|
| 321 |
|
|---|
| 322 | mov 8*10(%rax),%r14 # mimic epilogue
|
|---|
| 323 | mov 8*11(%rax),%r13
|
|---|
| 324 | mov 8*12(%rax),%r12
|
|---|
| 325 | mov 8*13(%rax),%rbp
|
|---|
| 326 | mov 8*14(%rax),%rbx
|
|---|
| 327 | mov 8*15(%rax),%rdi
|
|---|
| 328 | mov 8*16(%rax),%rsi
|
|---|
| 329 |
|
|---|
| 330 | mov %rbx,144($context) # restore context->Rbx
|
|---|
| 331 | mov %rbp,160($context) # restore context->Rbp
|
|---|
| 332 | mov %rsi,168($context) # restore context->Rsi
|
|---|
| 333 | mov %rdi,176($context) # restore context->Rdi
|
|---|
| 334 | mov %r12,216($context) # restore context->R12
|
|---|
| 335 | mov %r13,224($context) # restore context->R13
|
|---|
| 336 | mov %r14,232($context) # restore context->R14
|
|---|
| 337 |
|
|---|
| 338 | .Lin_prologue:
|
|---|
| 339 | lea 8*17(%rax),%rax
|
|---|
| 340 | mov %rax,152($context) # restore context->Rsp
|
|---|
| 341 |
|
|---|
| 342 | mov 40($disp),%rdi # disp->ContextRecord
|
|---|
| 343 | mov $context,%rsi # context
|
|---|
| 344 | mov \$154,%ecx # sizeof(CONTEXT)
|
|---|
| 345 | .long 0xa548f3fc # cld; rep movsq
|
|---|
| 346 |
|
|---|
| 347 | mov $disp,%rsi
|
|---|
| 348 | xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
|
|---|
| 349 | mov 8(%rsi),%rdx # arg2, disp->ImageBase
|
|---|
| 350 | mov 0(%rsi),%r8 # arg3, disp->ControlPc
|
|---|
| 351 | mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
|
|---|
| 352 | mov 40(%rsi),%r10 # disp->ContextRecord
|
|---|
| 353 | lea 56(%rsi),%r11 # &disp->HandlerData
|
|---|
| 354 | lea 24(%rsi),%r12 # &disp->EstablisherFrame
|
|---|
| 355 | mov %r10,32(%rsp) # arg5
|
|---|
| 356 | mov %r11,40(%rsp) # arg6
|
|---|
| 357 | mov %r12,48(%rsp) # arg7
|
|---|
| 358 | mov %rcx,56(%rsp) # arg8, (NULL)
|
|---|
| 359 | call *__imp_RtlVirtualUnwind(%rip)
|
|---|
| 360 |
|
|---|
| 361 | mov \$1,%eax # ExceptionContinueSearch
|
|---|
| 362 | add \$64,%rsp
|
|---|
| 363 | popfq
|
|---|
| 364 | pop %r15
|
|---|
| 365 | pop %r14
|
|---|
| 366 | pop %r13
|
|---|
| 367 | pop %r12
|
|---|
| 368 | pop %rbp
|
|---|
| 369 | pop %rbx
|
|---|
| 370 | pop %rdi
|
|---|
| 371 | pop %rsi
|
|---|
| 372 | ret
|
|---|
| 373 | .size se_handler,.-se_handler
|
|---|
| 374 |
|
|---|
| 375 | .section .pdata
|
|---|
| 376 | .align 4
|
|---|
| 377 | .rva _mul_1x1
|
|---|
| 378 | .rva .Lend_mul_1x1
|
|---|
| 379 | .rva .LSEH_info_1x1
|
|---|
| 380 |
|
|---|
| 381 | .rva .Lvanilla_mul_2x2
|
|---|
| 382 | .rva .Lend_mul_2x2
|
|---|
| 383 | .rva .LSEH_info_2x2
|
|---|
| 384 | .section .xdata
|
|---|
| 385 | .align 8
|
|---|
| 386 | .LSEH_info_1x1:
|
|---|
| 387 | .byte 0x01,0x07,0x02,0x00
|
|---|
| 388 | .byte 0x07,0x01,0x11,0x00 # sub rsp,128+8
|
|---|
| 389 | .LSEH_info_2x2:
|
|---|
| 390 | .byte 9,0,0,0
|
|---|
| 391 | .rva se_handler
|
|---|
| 392 | ___
|
|---|
| 393 | }
|
|---|
| 394 |
|
|---|
| 395 | $code =~ s/\`([^\`]*)\`/eval($1)/gem;
|
|---|
| 396 | print $code;
|
|---|
| 397 | close STDOUT;
|
|---|
