x86_64-gcc.c

开源项目openssl的源码

/*
 * x86_64 BIGNUM accelerator version 0.1, December 2002.
 *
 * Implemented by Andy Polyakov <[e-mail]> for the OpenSSL
 * project.
 *
 * Rights for redistribution and usage in source and binary forms are
 * granted according to the OpenSSL license. Warranty of any kind is
 * disclaimed.
 *
 * Q. Version 0.1? It doesn't sound like Andy, he used to assign real
 *    versions, like 1.0...
 * A. Well, that's because this code is basically a quick-n-dirty
 *    proof-of-concept hack. As you can see it's implemented with
 *    inline assembler, which means that you're bound to GCC and that
 *    there must be a room for fine-tuning.
 *
 * Q. Why inline assembler?
 * A. x86_64 features own ABI I'm not familiar with. Which is why
 *    I decided to let the compiler take care of subroutine
 *    prologue/epilogue as well as register allocation.
 *
 * Q. How much faster does it get?
 * A. Unfortunately people sitting on x86_64 hardware are prohibited
 *    to disclose the performance numbers, so they (SuSE labs to be
 *    specific) wouldn't tell me. However! Very similar coding technique
 *    (reaching out for 128-bit result from 64x64-bit multiplication)
 *    results in >3 times performance improvement on MIPS and I see no
 *    reason why gain on x86_64 would be so much different:-)
 */

#define BN_ULONG unsigned long

/*
 * "m"(a), "+m"(r)	is the way to favor DirectPath