X-Git-Url: https://git.openssl.org/gitweb/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fmd5%2Fmd5_locl.h;h=432f523c7eff7b467e8b4f3e19779ed3886779d9;hp=cc96bf50d588dfca9a2ee210c8b02d574b4d3cfd;hb=b4f0abd246340b90bb3fa2646814729f0e9d049e;hpb=9dff4cc05100b9989998e4fe061e58d0fbfd6d8e diff --git a/crypto/md5/md5_locl.h b/crypto/md5/md5_locl.h index cc96bf50d5..432f523c7e 100644 --- a/crypto/md5/md5_locl.h +++ b/crypto/md5/md5_locl.h @@ -58,6 +58,7 @@ #include #include +#include #include #ifndef MD5_LONG_LOG2 @@ -65,73 +66,36 @@ #endif #ifdef MD5_ASM -# if defined(__i386) || defined(WIN32) -# define md5_block_host_order md5_block_asm_host_order -# elif defined(__sparc) && defined(ULTRASPARC) - void md5_block_asm_data_order_aligned (MD5_CTX *c, const MD5_LONG *p,int num); -# define HASH_BLOCK_DATA_ORDER_ALIGNED md5_block_asm_data_order_aligned +# if defined(__i386) || defined(__i386__) || defined(_M_IX86) || defined(__INTEL__) || \ + defined(__x86_64) || defined(__x86_64__) || defined(_M_AMD64) || defined(_M_X64) +# define md5_block_data_order md5_block_asm_data_order +# elif defined(__ia64) || defined(__ia64__) || defined(_M_IA64) +# define md5_block_data_order md5_block_asm_data_order +# elif defined(__sparc) || defined(__sparc__) +# define md5_block_data_order md5_block_asm_data_order # endif #endif -void md5_block_host_order (MD5_CTX *c, const void *p,int num); -void md5_block_data_order (MD5_CTX *c, const void *p,int num); - -#if defined(__i386) -/* - * *_block_host_order is expected to handle aligned data while - * *_block_data_order - unaligned. As algorithm and host (x86) - * are in this case of the same "endianess" these two are - * otherwise indistinguishable. But normally you don't want to - * call the same function because unaligned access in places - * where alignment is expected is usually a "Bad Thing". Indeed, - * on RISCs you get punished with BUS ERROR signal or *severe* - * performance degradation. Intel CPUs are in turn perfectly - * capable of loading unaligned data without such drastic side - * effect. Yes, they say it's slower than aligned load, but no - * exception is generated and therefore performance degradation - * is *incomparable* with RISCs. What we should weight here is - * costs of unaligned access against costs of aligning data. - * According to my measurements allowing unaligned access results - * in ~9% performance improvement on Pentium II operating at - * 266MHz. I won't be surprised if the difference will be higher - * on faster systems:-) - * - * - */ -#define md5_block_data_order md5_block_host_order -#endif +void md5_block_data_order (MD5_CTX *c, const void *p,size_t num); #define DATA_ORDER_IS_LITTLE_ENDIAN #define HASH_LONG MD5_LONG -#define HASH_LONG_LOG2 MD5_LONG_LOG2 #define HASH_CTX MD5_CTX #define HASH_CBLOCK MD5_CBLOCK -#define HASH_LBLOCK MD5_LBLOCK #define HASH_UPDATE MD5_Update #define HASH_TRANSFORM MD5_Transform #define HASH_FINAL MD5_Final -#define HASH_BLOCK_HOST_ORDER md5_block_host_order -#if !defined(L_ENDIAN) || defined(md5_block_data_order) +#define HASH_MAKE_STRING(c,s) do { \ + unsigned long ll; \ + ll=(c)->A; (void)HOST_l2c(ll,(s)); \ + ll=(c)->B; (void)HOST_l2c(ll,(s)); \ + ll=(c)->C; (void)HOST_l2c(ll,(s)); \ + ll=(c)->D; (void)HOST_l2c(ll,(s)); \ + } while (0) #define HASH_BLOCK_DATA_ORDER md5_block_data_order -/* - * Little-endians (Intel and Alpha) feel better without this. - * It looks like memcpy does better job than generic - * md5_block_data_order on copying-n-aligning input data. - * But franlky speaking I didn't expect such result on Alpha. - * On the other hand I've got this with egcs-1.0.2 and if - * program is compiled with another (better?) compiler it - * might turn out other way around. - * - * - */ -#endif -#ifndef FLAT_INC -#include "../md32_common.h" -#else #include "md32_common.h" -#endif /* #define F(x,y,z) (((x) & (y)) | ((~(x)) & (z))) @@ -139,7 +103,7 @@ void md5_block_data_order (MD5_CTX *c, const void *p,int num); */ /* As pointed out by Wei Dai , the above can be - * simplified to the code below. Wei attributes these optimisations + * simplified to the code below. Wei attributes these optimizations * to Peter Gutmann's SHS code, and he attributes it to Rich Schroeppel. */ #define F(b,c,d) ((((c) ^ (d)) & (b)) ^ (d))