3 ##############################################################################
5 # Copyright (c) 2012, Intel Corporation #
7 # All rights reserved. #
9 # Redistribution and use in source and binary forms, with or without #
10 # modification, are permitted provided that the following conditions are #
13 # * Redistributions of source code must retain the above copyright #
14 # notice, this list of conditions and the following disclaimer. #
16 # * Redistributions in binary form must reproduce the above copyright #
17 # notice, this list of conditions and the following disclaimer in the #
18 # documentation and/or other materials provided with the #
21 # * Neither the name of the Intel Corporation nor the names of its #
22 # contributors may be used to endorse or promote products derived from #
23 # this software without specific prior written permission. #
26 # THIS SOFTWARE IS PROVIDED BY INTEL CORPORATION ""AS IS"" AND ANY #
27 # EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE #
28 # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR #
29 # PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL CORPORATION OR #
30 # CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, #
31 # EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, #
32 # PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR #
33 # PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF #
34 # LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING #
35 # NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS #
36 # SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #
38 ##############################################################################
39 # Developers and authors: #
40 # Shay Gueron (1, 2), and Vlad Krasnov (1) #
41 # (1) Intel Architecture Group, Microprocessor and Chipset Development, #
42 # Israel Development Center, Haifa, Israel #
43 # (2) University of Haifa #
44 ##############################################################################
46 # [1] S. Gueron, "Efficient Software Implementations of Modular #
47 # Exponentiation", http://eprint.iacr.org/2011/239 #
48 # [2] S. Gueron, V. Krasnov. "Speeding up Big-Numbers Squaring". #
49 # IEEE Proceedings of 9th International Conference on Information #
50 # Technology: New Generations (ITNG 2012), 821-823 (2012). #
51 # [3] S. Gueron, Efficient Software Implementations of Modular Exponentiation#
52 # Journal of Cryptographic Engineering 2:31-43 (2012). #
53 # [4] S. Gueron, V. Krasnov: "[PATCH] Efficient and side channel analysis #
54 # resistant 512-bit and 1024-bit modular exponentiation for optimizing #
55 # RSA1024 and RSA2048 on x86_64 platforms", #
56 # http://rt.openssl.org/Ticket/Display.html?id=2582&user=guest&pass=guest#
57 ##############################################################################
59 # While original submission covers 512- and 1024-bit exponentiation,
60 # this module is limited to 512-bit version only (and as such
61 # accelerates RSA1024 sign). This is because improvement for longer
62 # keys is not high enough to justify the effort, highest measured
63 # was ~5% on Westmere. [This is relative to OpenSSL 1.0.2, upcoming
64 # for the moment of this writing!] Nor does this module implement
65 # "monolithic" complete exponentiation jumbo-subroutine, but adheres
66 # to more modular mixture of C and assembly. And it's optimized even
67 # for processors other than Intel Core family (see table below for
68 # improvement coefficients).
71 # RSA1024 sign/sec this/original |this/rsax(*) this/fips(*)
72 # ----------------+---------------------------
73 # Opteron +13% |+5% +20%
74 # Bulldozer -0% |-1% +10%
76 # Westmere +5% |+14% +17%
77 # Sandy Bridge +2% |+12% +29%
78 # Ivy Bridge +1% |+11% +35%
79 # Haswell(**) -0% |+12% +39%
81 # VIA Nano +70% |+9% +25%
83 # (*) rsax engine and fips numbers are presented for reference
85 # (**) MULX was attempted, but found to give only marginal improvement;
89 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
91 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
93 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
94 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
95 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
96 die "can't locate x86_64-xlate.pl";
98 open OUT,"| \"$^X\" $xlate $flavour $output";
101 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
102 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
106 if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
107 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
111 if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
112 `ml64 2>&1` =~ /Version ([0-9]+)\./) {
116 if (!$addx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9])\.([0-9]+)/) {
117 my $ver = $2 + $3/100.0; # 3.1->3.01, 3.10->3.10
118 $addx = ($ver>=3.03);
121 ($out, $inp, $mod) = ("%rdi", "%rsi", "%rbp"); # common internal API
123 my ($out,$inp,$mod,$n0,$times) = ("%rdi","%rsi","%rdx","%rcx","%r8d");
128 .extern OPENSSL_ia32cap_P
131 .type rsaz_512_sqr,\@function,5
133 rsaz_512_sqr: # 25-29% faster than rsaz_512_mul
143 movq $mod, %rbp # common argument
148 $code.=<<___ if ($addx);
150 andl OPENSSL_ia32cap_P+8(%rip),%r11d
151 cmpl \$0x80100,%r11d # check for MULX and ADO/CX
159 movl $times,128+8(%rsp)
203 addq %r8, %r8 #shlq \$1, %r8
205 adcq %r9, %r9 #shld \$1, %r8, %r9
266 lea (%rcx,%r10,2), %r10 #shld \$1, %rcx, %r10
268 adcq %r11, %r11 #shld \$1, %r10, %r11
306 lea (%rbx,%r12,2), %r12 #shld \$1, %rbx, %r12
324 leaq (%r10,%r13,2), %r13 #shld \$1, %r12, %r13
354 leaq (%rcx,%r14,2), %r14 #shld \$1, %rcx, %r14
372 leaq (%r12,%r15,2),%r15 #shld \$1, %r14, %r15
397 leaq (%rbx,%r8,2), %r8 #shld \$1, %rbx, %r8
412 leaq (%r12,%r9,2), %r9 #shld \$1, %r8, %r9
436 leaq (%rcx,%r10,2), %r10 #shld \$1, %rcx, %r10
444 leaq (%r15,%r11,2), %r11 #shld \$1, %r10, %r11
465 adcq %r12, %r12 #shld \$1, %rbx, %r12
466 adcq %r13, %r13 #shld \$1, %r12, %r13
467 adcq %r14, %r14 #shld \$1, %r13, %r14
497 call __rsaz_512_reduce
509 call __rsaz_512_subtract
513 movl 128+8(%rsp), $times
525 movl $times,128+8(%rsp)
526 movq $out, %xmm0 # off-load
527 movq %rbp, %xmm1 # off-load
531 mulx 16($inp), %rcx, %r10
532 xor %rbp, %rbp # cf=0, of=0
534 mulx 24($inp), %rax, %r11
537 mulx 32($inp), %rcx, %r12
540 mulx 40($inp), %rax, %r13
543 .byte 0xc4,0x62,0xf3,0xf6,0xb6,0x30,0x00,0x00,0x00 # mulx 48($inp), %rcx, %r14
547 .byte 0xc4,0x62,0xfb,0xf6,0xbe,0x38,0x00,0x00,0x00 # mulx 56($inp), %rax, %r15
549 adcx %rbp, %r15 # %rbp is 0
556 mulx %rdx, %rax, %rdx
565 mulx 16($inp), %rax, %rbx
569 .byte 0xc4,0x62,0xc3,0xf6,0x86,0x18,0x00,0x00,0x00 # mulx 24($inp), $out, %r8
573 mulx 32($inp), %rax, %rbx
577 mulx 40($inp), $out, %r8
581 .byte 0xc4,0xe2,0xfb,0xf6,0x9e,0x30,0x00,0x00,0x00 # mulx 48($inp), %rax, %rbx
585 .byte 0xc4,0x62,0xc3,0xf6,0x86,0x38,0x00,0x00,0x00 # mulx 56($inp), $out, %r8
595 mulx %rdx, %rax, %rcx
602 .byte 0x4c,0x89,0x94,0x24,0x18,0x00,0x00,0x00 # mov %r10, 24(%rsp)
605 .byte 0xc4,0x62,0xc3,0xf6,0x8e,0x18,0x00,0x00,0x00 # mulx 24($inp), $out, %r9
609 mulx 32($inp), %rax, %rcx
613 mulx 40($inp), $out, %r9
617 .byte 0xc4,0xe2,0xfb,0xf6,0x8e,0x30,0x00,0x00,0x00 # mulx 48($inp), %rax, %rcx
621 .byte 0xc4,0x62,0xc3,0xf6,0x8e,0x38,0x00,0x00,0x00 # mulx 56($inp), $out, %r9
631 mulx %rdx, %rax, %rdx
638 .byte 0x4c,0x89,0xa4,0x24,0x28,0x00,0x00,0x00 # mov %r12, 40(%rsp)
641 .byte 0xc4,0xe2,0xfb,0xf6,0x9e,0x20,0x00,0x00,0x00 # mulx 32($inp), %rax, %rbx
645 mulx 40($inp), $out, %r10
649 mulx 48($inp), %rax, %rbx
653 mulx 56($inp), $out, %r10
664 mulx %rdx, %rax, %rdx
674 .byte 0xc4,0x62,0xc3,0xf6,0x9e,0x28,0x00,0x00,0x00 # mulx 40($inp), $out, %r11
678 mulx 48($inp), %rax, %rcx
682 mulx 56($inp), $out, %r11
692 mulx %rdx, %rax, %rdx
702 .byte 0xc4,0xe2,0xfb,0xf6,0x9e,0x30,0x00,0x00,0x00 # mulx 48($inp), %rax, %rbx
706 .byte 0xc4,0x62,0xc3,0xf6,0xa6,0x38,0x00,0x00,0x00 # mulx 56($inp), $out, %r12
716 mulx %rdx, %rax, %rdx
726 .byte 0xc4,0x62,0xfb,0xf6,0xae,0x38,0x00,0x00,0x00 # mulx 56($inp), %rax, %r13
736 mulx %rdx, %rax, %rdx
742 .byte 0x4c,0x89,0x9c,0x24,0x60,0x00,0x00,0x00 # mov %r11, 96(%rsp)
743 .byte 0x4c,0x89,0xa4,0x24,0x68,0x00,0x00,0x00 # mov %r12, 104(%rsp)
746 mulx %rdx, %rax, %rdx
758 movq 128(%rsp), %rdx # pull $n0
768 call __rsaz_512_reducex
780 call __rsaz_512_subtract
784 movl 128+8(%rsp), $times
795 leaq 128+24+48(%rsp), %rax
805 .size rsaz_512_sqr,.-rsaz_512_sqr
809 my ($out,$ap,$bp,$mod,$n0) = ("%rdi","%rsi","%rdx","%rcx","%r8");
812 .type rsaz_512_mul,\@function,5
824 movq $out, %xmm0 # off-load arguments
828 $code.=<<___ if ($addx);
830 andl OPENSSL_ia32cap_P+8(%rip),%r11d
831 cmpl \$0x80100,%r11d # check for MULX and ADO/CX
835 movq ($bp), %rbx # pass b[0]
836 movq $bp, %rbp # pass argument
851 call __rsaz_512_reduce
853 $code.=<<___ if ($addx);
858 movq $bp, %rbp # pass argument
859 movq ($bp), %rdx # pass b[0]
865 movq 128(%rsp), %rdx # pull $n0
875 call __rsaz_512_reducex
889 call __rsaz_512_subtract
891 leaq 128+24+48(%rsp), %rax
901 .size rsaz_512_mul,.-rsaz_512_mul
905 my ($out,$ap,$bp,$mod,$n0,$pwr) = ("%rdi","%rsi","%rdx","%rcx","%r8","%r9d");
907 .globl rsaz_512_mul_gather4
908 .type rsaz_512_mul_gather4,\@function,6
910 rsaz_512_mul_gather4:
918 subq \$`128+24+($win64?0xb0:0)`, %rsp
920 $code.=<<___ if ($win64);
921 movaps %xmm6,0xa0(%rsp)
922 movaps %xmm7,0xb0(%rsp)
923 movaps %xmm8,0xc0(%rsp)
924 movaps %xmm9,0xd0(%rsp)
925 movaps %xmm10,0xe0(%rsp)
926 movaps %xmm11,0xf0(%rsp)
927 movaps %xmm12,0x100(%rsp)
928 movaps %xmm13,0x110(%rsp)
929 movaps %xmm14,0x120(%rsp)
930 movaps %xmm15,0x130(%rsp)
935 movdqa .Linc+16(%rip),%xmm1 # 00000002000000020000000200000002
936 movdqa .Linc(%rip),%xmm0 # 00000001000000010000000000000000
938 pshufd \$0,%xmm8,%xmm8 # broadcast $power
942 ########################################################################
943 # calculate mask by comparing 0..15 to $power
945 for($i=0;$i<4;$i++) {
947 paddd %xmm`$i`,%xmm`$i+1`
948 pcmpeqd %xmm8,%xmm`$i`
949 movdqa %xmm7,%xmm`$i+3`
954 paddd %xmm`$i`,%xmm`$i+1`
955 pcmpeqd %xmm8,%xmm`$i`
961 movdqa 16*0($bp),%xmm8
962 movdqa 16*1($bp),%xmm9
963 movdqa 16*2($bp),%xmm10
964 movdqa 16*3($bp),%xmm11
966 movdqa 16*4($bp),%xmm12
968 movdqa 16*5($bp),%xmm13
970 movdqa 16*6($bp),%xmm14
972 movdqa 16*7($bp),%xmm15
986 pshufd \$0x4e,%xmm8,%xmm9
989 $code.=<<___ if ($addx);
991 andl OPENSSL_ia32cap_P+8(%rip),%r11d
992 cmpl \$0x80100,%r11d # check for MULX and ADO/CX
998 movq $n0, 128(%rsp) # off-load arguments
999 movq $out, 128+8(%rsp)
1000 movq $mod, 128+16(%rsp)
1004 mulq %rbx # 0 iteration
1053 jmp .Loop_mul_gather
1057 movdqa 16*0(%rbp),%xmm8
1058 movdqa 16*1(%rbp),%xmm9
1059 movdqa 16*2(%rbp),%xmm10
1060 movdqa 16*3(%rbp),%xmm11
1062 movdqa 16*4(%rbp),%xmm12
1064 movdqa 16*5(%rbp),%xmm13
1066 movdqa 16*6(%rbp),%xmm14
1068 movdqa 16*7(%rbp),%xmm15
1069 leaq 128(%rbp), %rbp
1082 pshufd \$0x4e,%xmm8,%xmm9
1152 jnz .Loop_mul_gather
1163 movq 128+8(%rsp), $out
1164 movq 128+16(%rsp), %rbp
1175 call __rsaz_512_reduce
1177 $code.=<<___ if ($addx);
1178 jmp .Lmul_gather_tail
1184 mov $n0, 128(%rsp) # off-load arguments
1185 mov $out, 128+8(%rsp)
1186 mov $mod, 128+16(%rsp)
1188 mulx ($ap), %rbx, %r8 # 0 iteration
1190 xor %edi, %edi # cf=0, of=0
1192 mulx 8($ap), %rax, %r9
1194 mulx 16($ap), %rbx, %r10
1197 mulx 24($ap), %rax, %r11
1200 mulx 32($ap), %rbx, %r12
1203 mulx 40($ap), %rax, %r13
1206 mulx 48($ap), %rbx, %r14
1209 mulx 56($ap), %rax, %r15
1214 adcx %rdi, %r15 # %rdi is 0
1217 jmp .Loop_mulx_gather
1221 movdqa 16*0(%rbp),%xmm8
1222 movdqa 16*1(%rbp),%xmm9
1223 movdqa 16*2(%rbp),%xmm10
1224 movdqa 16*3(%rbp),%xmm11
1226 movdqa 16*4(%rbp),%xmm12
1228 movdqa 16*5(%rbp),%xmm13
1230 movdqa 16*6(%rbp),%xmm14
1232 movdqa 16*7(%rbp),%xmm15
1233 leaq 128(%rbp), %rbp
1246 pshufd \$0x4e,%xmm8,%xmm9
1250 .byte 0xc4,0x62,0xfb,0xf6,0x86,0x00,0x00,0x00,0x00 # mulx ($ap), %rax, %r8
1254 mulx 8($ap), %rax, %r9
1258 mulx 16($ap), %rax, %r10
1262 .byte 0xc4,0x62,0xfb,0xf6,0x9e,0x18,0x00,0x00,0x00 # mulx 24($ap), %rax, %r11
1266 mulx 32($ap), %rax, %r12
1270 mulx 40($ap), %rax, %r13
1274 .byte 0xc4,0x62,0xfb,0xf6,0xb6,0x30,0x00,0x00,0x00 # mulx 48($ap), %rax, %r14
1279 mulx 56($ap), %rax, %r15
1280 mov %rbx, 64(%rsp,%rcx,8)
1284 adcx %rdi, %r15 # cf=0
1287 jnz .Loop_mulx_gather
1291 mov %r10, 64+16(%rsp)
1292 mov %r11, 64+24(%rsp)
1293 mov %r12, 64+32(%rsp)
1294 mov %r13, 64+40(%rsp)
1295 mov %r14, 64+48(%rsp)
1296 mov %r15, 64+56(%rsp)
1298 mov 128(%rsp), %rdx # pull arguments
1299 mov 128+8(%rsp), $out
1300 mov 128+16(%rsp), %rbp
1311 call __rsaz_512_reducex
1321 adcq 104(%rsp), %r13
1322 adcq 112(%rsp), %r14
1323 adcq 120(%rsp), %r15
1326 call __rsaz_512_subtract
1328 leaq 128+24+48(%rsp), %rax
1330 $code.=<<___ if ($win64);
1331 movaps 0xa0-0xc8(%rax),%xmm6
1332 movaps 0xb0-0xc8(%rax),%xmm7
1333 movaps 0xc0-0xc8(%rax),%xmm8
1334 movaps 0xd0-0xc8(%rax),%xmm9
1335 movaps 0xe0-0xc8(%rax),%xmm10
1336 movaps 0xf0-0xc8(%rax),%xmm11
1337 movaps 0x100-0xc8(%rax),%xmm12
1338 movaps 0x110-0xc8(%rax),%xmm13
1339 movaps 0x120-0xc8(%rax),%xmm14
1340 movaps 0x130-0xc8(%rax),%xmm15
1344 movq -48(%rax), %r15
1345 movq -40(%rax), %r14
1346 movq -32(%rax), %r13
1347 movq -24(%rax), %r12
1348 movq -16(%rax), %rbp
1351 .Lmul_gather4_epilogue:
1353 .size rsaz_512_mul_gather4,.-rsaz_512_mul_gather4
1357 my ($out,$ap,$mod,$n0,$tbl,$pwr) = ("%rdi","%rsi","%rdx","%rcx","%r8","%r9d");
1359 .globl rsaz_512_mul_scatter4
1360 .type rsaz_512_mul_scatter4,\@function,6
1362 rsaz_512_mul_scatter4:
1372 .Lmul_scatter4_body:
1373 leaq ($tbl,$pwr,8), $tbl
1374 movq $out, %xmm0 # off-load arguments
1381 $code.=<<___ if ($addx);
1382 movl \$0x80100,%r11d
1383 andl OPENSSL_ia32cap_P+8(%rip),%r11d
1384 cmpl \$0x80100,%r11d # check for MULX and ADO/CX
1388 movq ($out),%rbx # pass b[0]
1403 call __rsaz_512_reduce
1405 $code.=<<___ if ($addx);
1406 jmp .Lmul_scatter_tail
1410 movq ($out), %rdx # pass b[0]
1411 call __rsaz_512_mulx
1416 movq 128(%rsp), %rdx # pull $n0
1426 call __rsaz_512_reducex
1436 adcq 104(%rsp), %r13
1437 adcq 112(%rsp), %r14
1438 adcq 120(%rsp), %r15
1442 call __rsaz_512_subtract
1444 movq %r8, 128*0($inp) # scatter
1445 movq %r9, 128*1($inp)
1446 movq %r10, 128*2($inp)
1447 movq %r11, 128*3($inp)
1448 movq %r12, 128*4($inp)
1449 movq %r13, 128*5($inp)
1450 movq %r14, 128*6($inp)
1451 movq %r15, 128*7($inp)
1453 leaq 128+24+48(%rsp), %rax
1454 movq -48(%rax), %r15
1455 movq -40(%rax), %r14
1456 movq -32(%rax), %r13
1457 movq -24(%rax), %r12
1458 movq -16(%rax), %rbp
1461 .Lmul_scatter4_epilogue:
1463 .size rsaz_512_mul_scatter4,.-rsaz_512_mul_scatter4
1467 my ($out,$inp,$mod,$n0) = ("%rdi","%rsi","%rdx","%rcx");
1469 .globl rsaz_512_mul_by_one
1470 .type rsaz_512_mul_by_one,\@function,4
1472 rsaz_512_mul_by_one:
1483 $code.=<<___ if ($addx);
1484 movl OPENSSL_ia32cap_P+8(%rip),%eax
1487 movq $mod, %rbp # reassign argument
1500 movdqa %xmm0, (%rsp)
1501 movdqa %xmm0, 16(%rsp)
1502 movdqa %xmm0, 32(%rsp)
1503 movdqa %xmm0, 48(%rsp)
1504 movdqa %xmm0, 64(%rsp)
1505 movdqa %xmm0, 80(%rsp)
1506 movdqa %xmm0, 96(%rsp)
1508 $code.=<<___ if ($addx);
1510 cmpl \$0x80100,%eax # check for MULX and ADO/CX
1514 call __rsaz_512_reduce
1516 $code.=<<___ if ($addx);
1520 movq 128(%rsp), %rdx # pull $n0
1521 call __rsaz_512_reducex
1534 leaq 128+24+48(%rsp), %rax
1535 movq -48(%rax), %r15
1536 movq -40(%rax), %r14
1537 movq -32(%rax), %r13
1538 movq -24(%rax), %r12
1539 movq -16(%rax), %rbp
1542 .Lmul_by_one_epilogue:
1544 .size rsaz_512_mul_by_one,.-rsaz_512_mul_by_one
1547 { # __rsaz_512_reduce
1549 # input: %r8-%r15, %rbp - mod, 128(%rsp) - n0
1551 # clobbers: everything except %rbp and %rdi
1553 .type __rsaz_512_reduce,\@abi-omnipotent
1557 imulq 128+8(%rsp), %rbx
1560 jmp .Lreduction_loop
1591 movq 128+8(%rsp), %rsi
1632 jne .Lreduction_loop
1635 .size __rsaz_512_reduce,.-__rsaz_512_reduce
1639 # __rsaz_512_reducex
1641 # input: %r8-%r15, %rbp - mod, 128(%rsp) - n0
1643 # clobbers: everything except %rbp and %rdi
1645 .type __rsaz_512_reducex,\@abi-omnipotent
1648 #movq 128+8(%rsp), %rdx # pull $n0
1650 xorq %rsi, %rsi # cf=0,of=0
1652 jmp .Lreduction_loopx
1657 mulx 0(%rbp), %rax, %r8
1661 mulx 8(%rbp), %rax, %r9
1665 mulx 16(%rbp), %rbx, %r10
1669 mulx 24(%rbp), %rbx, %r11
1673 .byte 0xc4,0x62,0xe3,0xf6,0xa5,0x20,0x00,0x00,0x00 # mulx 32(%rbp), %rbx, %r12
1679 mulx 128+8(%rsp), %rbx, %rdx
1682 mulx 40(%rbp), %rax, %r13
1686 .byte 0xc4,0x62,0xfb,0xf6,0xb5,0x30,0x00,0x00,0x00 # mulx 48(%rbp), %rax, %r14
1690 mulx 56(%rbp), %rax, %r15
1693 adox %rsi, %r15 # %rsi is 0
1694 adcx %rsi, %r15 # cf=0
1697 jne .Lreduction_loopx
1700 .size __rsaz_512_reducex,.-__rsaz_512_reducex
1703 { # __rsaz_512_subtract
1704 # input: %r8-%r15, %rdi - $out, %rbp - $mod, %rcx - mask
1706 # clobbers: everything but %rdi, %rsi and %rbp
1708 .type __rsaz_512_subtract,\@abi-omnipotent
1710 __rsaz_512_subtract:
1764 .size __rsaz_512_subtract,.-__rsaz_512_subtract
1769 # input: %rsi - ap, %rbp - bp
1771 # clobbers: everything
1772 my ($ap,$bp) = ("%rsi","%rbp");
1774 .type __rsaz_512_mul,\@abi-omnipotent
1915 .size __rsaz_512_mul,.-__rsaz_512_mul
1921 # input: %rsi - ap, %rbp - bp
1923 # clobbers: everything
1924 my ($ap,$bp,$zero) = ("%rsi","%rbp","%rdi");
1926 .type __rsaz_512_mulx,\@abi-omnipotent
1929 mulx ($ap), %rbx, %r8 # initial %rdx preloaded by caller
1932 mulx 8($ap), %rax, %r9
1935 mulx 16($ap), %rbx, %r10
1938 mulx 24($ap), %rax, %r11
1941 mulx 32($ap), %rbx, %r12
1944 mulx 40($ap), %rax, %r13
1947 mulx 48($ap), %rbx, %r14
1950 mulx 56($ap), %rax, %r15
1956 xor $zero, $zero # cf=0,of=0
1962 mulx ($ap), %rax, %r8
1966 mulx 8($ap), %rax, %r9
1970 mulx 16($ap), %rax, %r10
1974 mulx 24($ap), %rax, %r11
1978 .byte 0x3e,0xc4,0x62,0xfb,0xf6,0xa6,0x20,0x00,0x00,0x00 # mulx 32($ap), %rax, %r12
1982 mulx 40($ap), %rax, %r13
1986 mulx 48($ap), %rax, %r14
1990 mulx 56($ap), %rax, %r15
1991 movq 64($bp,%rcx,8), %rdx
1992 movq %rbx, 8+64-8(%rsp,%rcx,8)
1995 adcx $zero, %r15 # cf=0
2001 mulx ($ap), %rax, %r8
2005 .byte 0xc4,0x62,0xfb,0xf6,0x8e,0x08,0x00,0x00,0x00 # mulx 8($ap), %rax, %r9
2009 .byte 0xc4,0x62,0xfb,0xf6,0x96,0x10,0x00,0x00,0x00 # mulx 16($ap), %rax, %r10
2013 mulx 24($ap), %rax, %r11
2017 mulx 32($ap), %rax, %r12
2021 mulx 40($ap), %rax, %r13
2025 .byte 0xc4,0x62,0xfb,0xf6,0xb6,0x30,0x00,0x00,0x00 # mulx 48($ap), %rax, %r14
2029 .byte 0xc4,0x62,0xfb,0xf6,0xbe,0x38,0x00,0x00,0x00 # mulx 56($ap), %rax, %r15
2034 mov %rbx, 8+64-8(%rsp)
2036 mov %r9, 8+64+8(%rsp)
2037 mov %r10, 8+64+16(%rsp)
2038 mov %r11, 8+64+24(%rsp)
2039 mov %r12, 8+64+32(%rsp)
2040 mov %r13, 8+64+40(%rsp)
2041 mov %r14, 8+64+48(%rsp)
2042 mov %r15, 8+64+56(%rsp)
2045 .size __rsaz_512_mulx,.-__rsaz_512_mulx
2049 my ($out,$inp,$power)= $win64 ? ("%rcx","%rdx","%r8d") : ("%rdi","%rsi","%edx");
2051 .globl rsaz_512_scatter4
2052 .type rsaz_512_scatter4,\@abi-omnipotent
2055 leaq ($out,$power,8), $out
2063 leaq 128($out), $out
2067 .size rsaz_512_scatter4,.-rsaz_512_scatter4
2069 .globl rsaz_512_gather4
2070 .type rsaz_512_gather4,\@abi-omnipotent
2074 $code.=<<___ if ($win64);
2075 .LSEH_begin_rsaz_512_gather4:
2076 .byte 0x48,0x81,0xec,0xa8,0x00,0x00,0x00 # sub $0xa8,%rsp
2077 .byte 0x0f,0x29,0x34,0x24 # movaps %xmm6,(%rsp)
2078 .byte 0x0f,0x29,0x7c,0x24,0x10 # movaps %xmm7,0x10(%rsp)
2079 .byte 0x44,0x0f,0x29,0x44,0x24,0x20 # movaps %xmm8,0x20(%rsp)
2080 .byte 0x44,0x0f,0x29,0x4c,0x24,0x30 # movaps %xmm9,0x30(%rsp)
2081 .byte 0x44,0x0f,0x29,0x54,0x24,0x40 # movaps %xmm10,0x40(%rsp)
2082 .byte 0x44,0x0f,0x29,0x5c,0x24,0x50 # movaps %xmm11,0x50(%rsp)
2083 .byte 0x44,0x0f,0x29,0x64,0x24,0x60 # movaps %xmm12,0x60(%rsp)
2084 .byte 0x44,0x0f,0x29,0x6c,0x24,0x70 # movaps %xmm13,0x70(%rsp)
2085 .byte 0x44,0x0f,0x29,0xb4,0x24,0x80,0,0,0 # movaps %xmm14,0x80(%rsp)
2086 .byte 0x44,0x0f,0x29,0xbc,0x24,0x90,0,0,0 # movaps %xmm15,0x90(%rsp)
2090 movdqa .Linc+16(%rip),%xmm1 # 00000002000000020000000200000002
2091 movdqa .Linc(%rip),%xmm0 # 00000001000000010000000000000000
2093 pshufd \$0,%xmm8,%xmm8 # broadcast $power
2097 ########################################################################
2098 # calculate mask by comparing 0..15 to $power
2100 for($i=0;$i<4;$i++) {
2102 paddd %xmm`$i`,%xmm`$i+1`
2103 pcmpeqd %xmm8,%xmm`$i`
2104 movdqa %xmm7,%xmm`$i+3`
2109 paddd %xmm`$i`,%xmm`$i+1`
2110 pcmpeqd %xmm8,%xmm`$i`
2119 movdqa 16*0($inp),%xmm8
2120 movdqa 16*1($inp),%xmm9
2121 movdqa 16*2($inp),%xmm10
2122 movdqa 16*3($inp),%xmm11
2124 movdqa 16*4($inp),%xmm12
2126 movdqa 16*5($inp),%xmm13
2128 movdqa 16*6($inp),%xmm14
2130 movdqa 16*7($inp),%xmm15
2131 leaq 128($inp), $inp
2144 pshufd \$0x4e,%xmm8,%xmm9
2151 $code.=<<___ if ($win64);
2152 movaps 0x00(%rsp),%xmm6
2153 movaps 0x10(%rsp),%xmm7
2154 movaps 0x20(%rsp),%xmm8
2155 movaps 0x30(%rsp),%xmm9
2156 movaps 0x40(%rsp),%xmm10
2157 movaps 0x50(%rsp),%xmm11
2158 movaps 0x60(%rsp),%xmm12
2159 movaps 0x70(%rsp),%xmm13
2160 movaps 0x80(%rsp),%xmm14
2161 movaps 0x90(%rsp),%xmm15
2166 .LSEH_end_rsaz_512_gather4:
2167 .size rsaz_512_gather4,.-rsaz_512_gather4
2176 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
2177 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
2185 .extern __imp_RtlVirtualUnwind
2186 .type se_handler,\@abi-omnipotent
2200 mov 120($context),%rax # pull context->Rax
2201 mov 248($context),%rbx # pull context->Rip
2203 mov 8($disp),%rsi # disp->ImageBase
2204 mov 56($disp),%r11 # disp->HandlerData
2206 mov 0(%r11),%r10d # HandlerData[0]
2207 lea (%rsi,%r10),%r10 # end of prologue label
2208 cmp %r10,%rbx # context->Rip<end of prologue label
2209 jb .Lcommon_seh_tail
2211 mov 152($context),%rax # pull context->Rsp
2213 mov 4(%r11),%r10d # HandlerData[1]
2214 lea (%rsi,%r10),%r10 # epilogue label
2215 cmp %r10,%rbx # context->Rip>=epilogue label
2216 jae .Lcommon_seh_tail
2218 lea 128+24+48(%rax),%rax
2220 lea .Lmul_gather4_epilogue(%rip),%rbx
2222 jne .Lse_not_in_mul_gather4
2226 lea -48-0xa8(%rax),%rsi
2227 lea 512($context),%rdi
2229 .long 0xa548f3fc # cld; rep movsq
2231 .Lse_not_in_mul_gather4:
2238 mov %rbx,144($context) # restore context->Rbx
2239 mov %rbp,160($context) # restore context->Rbp
2240 mov %r12,216($context) # restore context->R12
2241 mov %r13,224($context) # restore context->R13
2242 mov %r14,232($context) # restore context->R14
2243 mov %r15,240($context) # restore context->R15
2248 mov %rax,152($context) # restore context->Rsp
2249 mov %rsi,168($context) # restore context->Rsi
2250 mov %rdi,176($context) # restore context->Rdi
2252 mov 40($disp),%rdi # disp->ContextRecord
2253 mov $context,%rsi # context
2254 mov \$154,%ecx # sizeof(CONTEXT)
2255 .long 0xa548f3fc # cld; rep movsq
2258 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
2259 mov 8(%rsi),%rdx # arg2, disp->ImageBase
2260 mov 0(%rsi),%r8 # arg3, disp->ControlPc
2261 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
2262 mov 40(%rsi),%r10 # disp->ContextRecord
2263 lea 56(%rsi),%r11 # &disp->HandlerData
2264 lea 24(%rsi),%r12 # &disp->EstablisherFrame
2265 mov %r10,32(%rsp) # arg5
2266 mov %r11,40(%rsp) # arg6
2267 mov %r12,48(%rsp) # arg7
2268 mov %rcx,56(%rsp) # arg8, (NULL)
2269 call *__imp_RtlVirtualUnwind(%rip)
2271 mov \$1,%eax # ExceptionContinueSearch
2283 .size se_handler,.-se_handler
2287 .rva .LSEH_begin_rsaz_512_sqr
2288 .rva .LSEH_end_rsaz_512_sqr
2289 .rva .LSEH_info_rsaz_512_sqr
2291 .rva .LSEH_begin_rsaz_512_mul
2292 .rva .LSEH_end_rsaz_512_mul
2293 .rva .LSEH_info_rsaz_512_mul
2295 .rva .LSEH_begin_rsaz_512_mul_gather4
2296 .rva .LSEH_end_rsaz_512_mul_gather4
2297 .rva .LSEH_info_rsaz_512_mul_gather4
2299 .rva .LSEH_begin_rsaz_512_mul_scatter4
2300 .rva .LSEH_end_rsaz_512_mul_scatter4
2301 .rva .LSEH_info_rsaz_512_mul_scatter4
2303 .rva .LSEH_begin_rsaz_512_mul_by_one
2304 .rva .LSEH_end_rsaz_512_mul_by_one
2305 .rva .LSEH_info_rsaz_512_mul_by_one
2307 .rva .LSEH_begin_rsaz_512_gather4
2308 .rva .LSEH_end_rsaz_512_gather4
2309 .rva .LSEH_info_rsaz_512_gather4
2313 .LSEH_info_rsaz_512_sqr:
2316 .rva .Lsqr_body,.Lsqr_epilogue # HandlerData[]
2317 .LSEH_info_rsaz_512_mul:
2320 .rva .Lmul_body,.Lmul_epilogue # HandlerData[]
2321 .LSEH_info_rsaz_512_mul_gather4:
2324 .rva .Lmul_gather4_body,.Lmul_gather4_epilogue # HandlerData[]
2325 .LSEH_info_rsaz_512_mul_scatter4:
2328 .rva .Lmul_scatter4_body,.Lmul_scatter4_epilogue # HandlerData[]
2329 .LSEH_info_rsaz_512_mul_by_one:
2332 .rva .Lmul_by_one_body,.Lmul_by_one_epilogue # HandlerData[]
2333 .LSEH_info_rsaz_512_gather4:
2334 .byte 0x01,0x46,0x16,0x00
2335 .byte 0x46,0xf8,0x09,0x00 # vmovaps 0x90(rsp),xmm15
2336 .byte 0x3d,0xe8,0x08,0x00 # vmovaps 0x80(rsp),xmm14
2337 .byte 0x34,0xd8,0x07,0x00 # vmovaps 0x70(rsp),xmm13
2338 .byte 0x2e,0xc8,0x06,0x00 # vmovaps 0x60(rsp),xmm12
2339 .byte 0x28,0xb8,0x05,0x00 # vmovaps 0x50(rsp),xmm11
2340 .byte 0x22,0xa8,0x04,0x00 # vmovaps 0x40(rsp),xmm10
2341 .byte 0x1c,0x98,0x03,0x00 # vmovaps 0x30(rsp),xmm9
2342 .byte 0x16,0x88,0x02,0x00 # vmovaps 0x20(rsp),xmm8
2343 .byte 0x10,0x78,0x01,0x00 # vmovaps 0x10(rsp),xmm7
2344 .byte 0x0b,0x68,0x00,0x00 # vmovaps 0x00(rsp),xmm6
2345 .byte 0x07,0x01,0x15,0x00 # sub rsp,0xa8
2349 $code =~ s/\`([^\`]*)\`/eval $1/gem;