]>
Commit | Line | Data |
---|---|---|
41b76195 | 1 | #include "../bn_lcl.h" |
93c4ba07 | 2 | #if !(defined(__GNUC__) && __GNUC__>=2) |
c8a0d0aa AP |
3 | # include "../bn_asm.c" /* kind of dirty hack for Sun Studio */ |
4 | #else | |
6977c7e2 | 5 | /*- |
2f98abbc AP |
6 | * x86_64 BIGNUM accelerator version 0.1, December 2002. |
7 | * | |
8 | * Implemented by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL | |
9 | * project. | |
10 | * | |
11 | * Rights for redistribution and usage in source and binary forms are | |
12 | * granted according to the OpenSSL license. Warranty of any kind is | |
13 | * disclaimed. | |
14 | * | |
15 | * Q. Version 0.1? It doesn't sound like Andy, he used to assign real | |
16 | * versions, like 1.0... | |
17 | * A. Well, that's because this code is basically a quick-n-dirty | |
18 | * proof-of-concept hack. As you can see it's implemented with | |
19 | * inline assembler, which means that you're bound to GCC and that | |
1809e858 | 20 | * there might be enough room for further improvement. |
2f98abbc AP |
21 | * |
22 | * Q. Why inline assembler? | |
1809e858 AP |
23 | * A. x86_64 features own ABI which I'm not familiar with. This is |
24 | * why I decided to let the compiler take care of subroutine | |
25 | * prologue/epilogue as well as register allocation. For reference. | |
26 | * Win64 implements different ABI for AMD64, different from Linux. | |
2f98abbc AP |
27 | * |
28 | * Q. How much faster does it get? | |
1809e858 AP |
29 | * A. 'apps/openssl speed rsa dsa' output with no-asm: |
30 | * | |
31 | * sign verify sign/s verify/s | |
32 | * rsa 512 bits 0.0006s 0.0001s 1683.8 18456.2 | |
33 | * rsa 1024 bits 0.0028s 0.0002s 356.0 6407.0 | |
34 | * rsa 2048 bits 0.0172s 0.0005s 58.0 1957.8 | |
35 | * rsa 4096 bits 0.1155s 0.0018s 8.7 555.6 | |
36 | * sign verify sign/s verify/s | |
37 | * dsa 512 bits 0.0005s 0.0006s 2100.8 1768.3 | |
38 | * dsa 1024 bits 0.0014s 0.0018s 692.3 559.2 | |
39 | * dsa 2048 bits 0.0049s 0.0061s 204.7 165.0 | |
40 | * | |
41 | * 'apps/openssl speed rsa dsa' output with this module: | |
42 | * | |
43 | * sign verify sign/s verify/s | |
44 | * rsa 512 bits 0.0004s 0.0000s 2767.1 33297.9 | |
45 | * rsa 1024 bits 0.0012s 0.0001s 867.4 14674.7 | |
46 | * rsa 2048 bits 0.0061s 0.0002s 164.0 5270.0 | |
47 | * rsa 4096 bits 0.0384s 0.0006s 26.1 1650.8 | |
48 | * sign verify sign/s verify/s | |
49 | * dsa 512 bits 0.0002s 0.0003s 4442.2 3786.3 | |
50 | * dsa 1024 bits 0.0005s 0.0007s 1835.1 1497.4 | |
51 | * dsa 2048 bits 0.0016s 0.0020s 620.4 504.6 | |
52 | * | |
53 | * For the reference. IA-32 assembler implementation performs | |
54 | * very much like 64-bit code compiled with no-asm on the same | |
55 | * machine. | |
2f98abbc AP |
56 | */ |
57 | ||
0e05b51f | 58 | #if defined(_WIN64) || !defined(__LP64__) |
93c4ba07 AP |
59 | #define BN_ULONG unsigned long long |
60 | #else | |
2f98abbc | 61 | #define BN_ULONG unsigned long |
93c4ba07 | 62 | #endif |
2f98abbc | 63 | |
41b76195 DSH |
64 | #undef mul |
65 | #undef mul_add | |
66 | ||
6977c7e2 | 67 | /*- |
2f98abbc AP |
68 |