diff options
author | Andy Polyakov <appro@openssl.org> | 2006-10-11 11:55:11 +0000 |
---|---|---|
committer | Andy Polyakov <appro@openssl.org> | 2006-10-11 11:55:11 +0000 |
commit | c69ed6ea396b184fd56327b475a441081fcade8e (patch) | |
tree | 934b6a730dd5eebcf82ec3b1015f6c2ee96ada1f | |
parent | 55a08fac680d5f35096c54ef2fdc6185c314fc89 (diff) |
Re-implement md32_common.h [make it simpler!] and eliminate code rendered
redundant as result.
-rw-r--r-- | crypto/md32_common.h | 343 | ||||
-rw-r--r-- | crypto/md4/md4_dgst.c | 93 | ||||
-rw-r--r-- | crypto/md4/md4_locl.h | 45 | ||||
-rw-r--r-- | crypto/md5/asm/md5-586.pl | 2 | ||||
-rwxr-xr-x | crypto/md5/asm/md5-x86_64.pl | 8 | ||||
-rw-r--r-- | crypto/md5/md5_dgst.c | 110 | ||||
-rw-r--r-- | crypto/md5/md5_locl.h | 51 | ||||
-rw-r--r-- | crypto/ripemd/asm/rmd-586.pl | 4 | ||||
-rw-r--r-- | crypto/ripemd/rmd_dgst.c | 205 | ||||
-rw-r--r-- | crypto/ripemd/rmd_locl.h | 15 | ||||
-rw-r--r-- | crypto/sha/sha256.c | 61 | ||||
-rw-r--r-- | crypto/sha/sha_locl.h | 309 |
12 files changed, 156 insertions, 1090 deletions
diff --git a/crypto/md32_common.h b/crypto/md32_common.h index 3ed16f380f..ce956df382 100644 --- a/crypto/md32_common.h +++ b/crypto/md32_common.h @@ -1,6 +1,6 @@ /* crypto/md32_common.h */ /* ==================================================================== - * Copyright (c) 1999-2002 The OpenSSL Project. All rights reserved. + * Copyright (c) 1999-2006 The OpenSSL Project. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions @@ -47,10 +47,6 @@ * OF THE POSSIBILITY OF SUCH DAMAGE. * ==================================================================== * - * This product includes cryptographic software written by Eric Young - * (eay@cryptsoft.com). This product includes software written by Tim - * Hudson (tjh@cryptsoft.com). - * */ /* @@ -76,40 +72,27 @@ * typedef struct { * ... * HASH_LONG Nl,Nh; + * either { * HASH_LONG data[HASH_LBLOCK]; + * unsigned char data[HASH_CBLOCK]; + * }; * unsigned int num; * ... * } HASH_CTX; + * data[] vector is expected to be zeroed upon first call to + * HASH_UPDATE. * HASH_UPDATE * name of "Update" function, implemented here. * HASH_TRANSFORM * name of "Transform" function, implemented here. * HASH_FINAL * name of "Final" function, implemented here. - * HASH_BLOCK_HOST_ORDER - * name of "block" function treating *aligned* input message - * in host byte order, implemented externally. * HASH_BLOCK_DATA_ORDER - * name of "block" function treating *unaligned* input message - * in original (data) byte order, implemented externally (it - * actually is optional if data and host are of the same - * "endianess"). + * name of "block" function capable of treating *unaligned* input + * message in original (data) byte order, implemented externally. * HASH_MAKE_STRING * macro convering context variables to an ASCII hash string. * - * Optional macros: - * - * B_ENDIAN or L_ENDIAN - * defines host byte-order. - * HASH_LONG_LOG2 - * defaults to 2 if not states otherwise. - * HASH_LBLOCK - * assumed to be HASH_CBLOCK/4 if not stated otherwise. - * HASH_BLOCK_DATA_ORDER_ALIGNED - * alternative "block" function capable of treating - * aligned input message in original (data) order, - * implemented externally. - * * MD5 example: * * #define DATA_ORDER_IS_LITTLE_ENDIAN @@ -118,11 +101,9 @@ * #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 * #define HASH_BLOCK_DATA_ORDER md5_block_data_order * * <appro@fy.chalmers.se> @@ -152,27 +133,9 @@ #error "HASH_FINAL must be defined!" #endif -#ifndef HASH_BLOCK_HOST_ORDER -#error "HASH_BLOCK_HOST_ORDER must be defined!" -#endif - -#if 0 -/* - * Moved below as it's required only if HASH_BLOCK_DATA_ORDER_ALIGNED - * isn't defined. - */ #ifndef HASH_BLOCK_DATA_ORDER #error "HASH_BLOCK_DATA_ORDER must be defined!" #endif -#endif - -#ifndef HASH_LBLOCK -#define HASH_LBLOCK (HASH_CBLOCK/4) -#endif - -#ifndef HASH_LONG_LOG2 -#define HASH_LONG_LOG2 2 -#endif /* * Engage compiler specific rotate intrinsic function if available. @@ -219,70 +182,10 @@ # endif #endif /* PEDANTIC */ -#if HASH_LONG_LOG2==2 /* Engage only if sizeof(HASH_LONG)== 4 */ -/* A nice byte order reversal from Wei Dai <weidai@eskimo.com> */ -#ifdef ROTATE -/* 5 instructions with rotate instruction, else 9 */ -#define REVERSE_FETCH32(a,l) ( \ - l=*(const HASH_LONG *)(a), \ - ((ROTATE(l,8)&0x00FF00FF)|(ROTATE((l&0x00FF00FF),24))) \ - ) -#else -/* 6 instructions with rotate instruction, else 8 */ -#define REVERSE_FETCH32(a,l) ( \ - l=*(const HASH_LONG *)(a), \ - l=(((l>>8)&0x00FF00FF)|((l&0x00FF00FF)<<8)), \ - ROTATE(l,16) \ - ) -/* - * Originally the middle line started with l=(((l&0xFF00FF00)>>8)|... - * It's rewritten as above for two reasons: - * - RISCs aren't good at long constants and have to explicitely - * compose 'em with several (well, usually 2) instructions in a - * register before performing the actual operation and (as you - * already realized:-) having same constant should inspire the - * compiler to permanently allocate the only register for it; - * - most modern CPUs have two ALUs, but usually only one has - * circuitry for shifts:-( this minor tweak inspires compiler - * to schedule shift instructions in a better way... - * - * <appro@fy.chalmers.se> - */ -#endif -#endif - #ifndef ROTATE #define ROTATE(a,n) (((a)<<(n))|(((a)&0xffffffff)>>(32-(n)))) #endif -/* - * Make some obvious choices. E.g., HASH_BLOCK_DATA_ORDER_ALIGNED - * and HASH_BLOCK_HOST_ORDER ought to be the same if input data - * and host are of the same "endianess". It's possible to mask - * this with blank #define HASH_BLOCK_DATA_ORDER though... - * - * <appro@fy.chalmers.se> - */ -#if defined(B_ENDIAN) -# if defined(DATA_ORDER_IS_BIG_ENDIAN) -# if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2 -# define HASH_BLOCK_DATA_ORDER_ALIGNED HASH_BLOCK_HOST_ORDER -# endif -# endif -#elif defined(L_ENDIAN) -# if defined(DATA_ORDER_IS_LITTLE_ENDIAN) -# if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2 -# define HASH_BLOCK_DATA_ORDER_ALIGNED HASH_BLOCK_HOST_ORDER -# endif -# endif -#endif - -#if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) -#ifndef HASH_BLOCK_DATA_ORDER -#error "HASH_BLOCK_DATA_ORDER must be defined!" -#endif -#endif - #if defined(DATA_ORDER_IS_BIG_ENDIAN) #ifndef PEDANTIC @@ -314,29 +217,6 @@ l|=(((unsigned long)(*((c)++))) ), \ l) #endif -#define HOST_p_c2l(c,l,n) { \ - switch (n) { \ - case 0: l =((unsigned long)(*((c)++)))<<24; \ - case 1: l|=((unsigned long)(*((c)++)))<<16; \ - case 2: l|=((unsigned long)(*((c)++)))<< 8; \ - case 3: l|=((unsigned long)(*((c)++))); \ - } } -#define HOST_p_c2l_p(c,l,sc,len) { \ - switch (sc) { \ - case 0: l =((unsigned long)(*((c)++)))<<24; \ - if (--len == 0) break; \ - case 1: l|=((unsigned long)(*((c)++)))<<16; \ - if (--len == 0) break; \ - case 2: l|=((unsigned long)(*((c)++)))<< 8; \ - } } -/* NOTE the pointer is not incremented at the end of this */ -#define HOST_c2l_p(c,l,n) { \ - l=0; (c)+=n; \ - switch (n) { \ - case 3: l =((unsigned long)(*(--(c))))<< 8; \ - case 2: l|=((unsigned long)(*(--(c))))<<16; \ - case 1: l|=((unsigned long)(*(--(c))))<<24; \ - } } #ifndef HOST_l2c #define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l)>>24)&0xff), \ *((c)++)=(unsigned char)(((l)>>16)&0xff), \ @@ -362,29 +242,6 @@ l|=(((unsigned long)(*((c)++)))<<24), \ l) #endif -#define HOST_p_c2l(c,l,n) { \ - switch (n) { \ - case 0: l =((unsigned long)(*((c)++))); \ - case 1: l|=((unsigned long)(*((c)++)))<< 8; \ - case 2: l|=((unsigned long)(*((c)++)))<<16; \ - case 3: l|=((unsigned long)(*((c)++)))<<24; \ - } } -#define HOST_p_c2l_p(c,l,sc,len) { \ - switch (sc) { \ - case 0: l =((unsigned long)(*((c)++))); \ - if (--len == 0) break; \ - case 1: l|=((unsigned long)(*((c)++)))<< 8; \ - if (--len == 0) break; \ - case 2: l|=((unsigned long)(*((c)++)))<<16; \ - } } -/* NOTE the pointer is not incremented at the end of this */ -#define HOST_c2l_p(c,l,n) { \ - l=0; (c)+=n; \ - switch (n) { \ - case 3: l =((unsigned long)(*(--(c))))<<16; \ - case 2: l|=((unsigned long)(*(--(c))))<< 8; \ - case 1: l|=((unsigned long)(*(--(c)))); \ - } } #ifndef HOST_l2c #define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l) )&0xff), \ *((c)++)=(unsigned char)(((l)>> 8)&0xff), \ @@ -402,9 +259,9 @@ int HASH_UPDATE (HASH_CTX *c, const void *data_, size_t len) { const unsigned char *data=data_; - register HASH_LONG * p; - register HASH_LONG l; - size_t sw,sc,ew,ec; + unsigned char *p; + HASH_LONG l; + size_t n; if (len==0) return 1; @@ -416,101 +273,43 @@ int HASH_UPDATE (HASH_CTX *c, const void *data_, size_t len) c->Nh+=(len>>29); /* might cause compiler warning on 16-bit */ c->Nl=l; - if (c->num != 0) + n = c->num; + if (n != 0) { - p=c->data; - sw=c->num>>2; - sc=c->num&0x03; + p=(unsigned char *)c->data; - if ((c->num+len) >= HASH_CBLOCK) + if ((n+len) >= HASH_CBLOCK) { - l=p[sw]; HOST_p_c2l(data,l,sc); p[sw++]=l; - for (; sw<HASH_LBLOCK; sw++) - { - HOST_c2l(data,l); p[sw]=l; - } - HASH_BLOCK_HOST_ORDER (c,p,1); - len-=(HASH_CBLOCK-c->num); - c->num=0; - /* drop through and do the rest */ + memcpy (p+n,data,HASH_CBLOCK-n); + HASH_BLOCK_DATA_ORDER (c,p,1); + n = HASH_CBLOCK-n; + data += n; + len -= n; + c->num = 0; + memset (p,0,HASH_CBLOCK); /* keep it zeroed */ } else { - c->num+=(unsigned int)len; - if ((sc+len) < 4) /* ugly, add char's to a word */ - { - l=p[sw]; HOST_p_c2l_p(data,l,sc,len); p[sw]=l; - } - else - { - ew=(c->num>>2); - ec=(c->num&0x03); - if (sc) - l=p[sw]; - HOST_p_c2l(data,l,sc); - p[sw++]=l; - for (; sw < ew; sw++) - { - HOST_c2l(data,l); p[sw]=l; - } - if (ec) - { - HOST_c2l_p(data,l,ec); p[sw]=l; - } - } + memcpy (p+n,data,len); + c->num += (unsigned int)len; return 1; } } - sw=len/HASH_CBLOCK; - if (sw > 0) + n = len/HASH_CBLOCK; + if (n > 0) { -#if defined(HASH_BLOCK_DATA_ORDER_ALIGNED) - /* - * Note that HASH_BLOCK_DATA_ORDER_ALIGNED gets defined - * only if sizeof(HASH_LONG)==4. - */ - if ((((size_t)data)%4) == 0) - { - /* data is properly aligned so that we can cast it: */ - HASH_BLOCK_DATA_ORDER_ALIGNED (c,(const HASH_LONG *)data,sw); - sw*=HASH_CBLOCK; - data+=sw; - len-=sw; - } - else -#if !defined(HASH_BLOCK_DATA_ORDER) - while (sw--) - { - memcpy (p=c->data,data,HASH_CBLOCK); - HASH_BLOCK_DATA_ORDER_ALIGNED(c,p,1); - data+=HASH_CBLOCK; - len-=HASH_CBLOCK; - } -#endif -#endif -#if defined(HASH_BLOCK_DATA_ORDER) - { - HASH_BLOCK_DATA_ORDER(c,data,sw); - sw*=HASH_CBLOCK; - data+=sw; - len-=sw; - } -#endif + HASH_BLOCK_DATA_ORDER (c,data,n); + n *= HASH_CBLOCK; + data += n; + len -= n; } - if (len!=0) + if (len != 0) { - p = c->data; + p = (unsigned char *)c->data; c->num = len; - ew=len>>2; /* words to copy */ - ec=len&0x03; - for (; ew; ew--,p++) - { - HOST_c2l(data,l); *p=l; - } - HOST_c2l_p(data,l,ec); - *p=l; + memcpy (p,data,len); } return 1; } @@ -518,73 +317,36 @@ int HASH_UPDATE (HASH_CTX *c, const void *data_, size_t len) void HASH_TRANSFORM (HASH_CTX *c, const unsigned char *data) { -#if defined(HASH_BLOCK_DATA_ORDER_ALIGNED) - if ((((size_t)data)%4) == 0) - /* data is properly aligned so that we can cast it: */ - HASH_BLOCK_DATA_ORDER_ALIGNED (c,(const HASH_LONG *)data,1); - else -#if !defined(HASH_BLOCK_DATA_ORDER) - { - memcpy (c->data,data,HASH_CBLOCK); - HASH_BLOCK_DATA_ORDER_ALIGNED (c,c->data,1); - } -#endif -#endif -#if defined(HASH_BLOCK_DATA_ORDER) HASH_BLOCK_DATA_ORDER (c,data,1); -#endif } int HASH_FINAL (unsigned char *md, HASH_CTX *c) { - register HASH_LONG *p; - register unsigned long l; - register int i,j; - static const unsigned char end[4]={0x80,0x00,0x00,0x00}; - const unsigned char *cp=end; - - /* c->num should definitly have room for at least one more byte. */ - p=c->data; - i=c->num>>2; - j=c->num&0x03; - -#if 0 - /* purify often complains about the following line as an - * Uninitialized Memory Read. While this can be true, the - * following p_c2l macro will reset l when that case is true. - * This is because j&0x03 contains the number of 'valid' bytes - * already in p[i]. If and only if j&0x03 == 0, the UMR will - * occur but this is also the only time p_c2l will do - * l= *(cp++) instead of l|= *(cp++) - * Many thanks to Alex Tang <altitude@cic.net> for pickup this - * 'potential bug' */ -#ifdef PURIFY - if (j==0) p[i]=0; /* Yeah, but that's not the way to fix it:-) */ -#endif - l=p[i]; -#else - l = (j==0) ? 0 : p[i]; -#endif - HOST_p_c2l(cp,l,j); p[i++]=l; /* i is the next 'undefined word' */ + unsigned char *p = (unsigned char *)c->data; + size_t n = c->num; - if (i>(HASH_LBLOCK-2)) /* save room for Nl and Nh */ + p[n] = 0x80; /* there is always room for one */ + n++; + + if (n > (HASH_CBLOCK-8)) { - if (i<HASH_LBLOCK) p[i]=0; - HASH_BLOCK_HOST_ORDER (c,p,1); - i=0; + HASH_BLOCK_DATA_ORDER (c,p,1); + memset (p,0,HASH_CBLOCK); } - for (; i<(HASH_LBLOCK-2); i++) - p[i]=0; + p += HASH_CBLOCK-8; #if defined(DATA_ORDER_IS_BIG_ENDIAN) - p[HASH_LBLOCK-2]=c->Nh; - p[HASH_LBLOCK-1]=c->Nl; + (void)HOST_l2c(c->Nh,p); + (void)HOST_l2c(c->Nl,p); #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN) - p[HASH_LBLOCK-2]=c->Nl; - p[HASH_LBLOCK-1]=c->Nh; + (void)HOST_l2c(c->Nl,p); + (void)HOST_l2c(c->Nh,p); #endif - HASH_BLOCK_HOST_ORDER (c,p,1); + p -= HASH_CBLOCK; + HASH_BLOCK_DATA_ORDER (c,p,1); + c->num=0; + memset (p,0,HASH_CBLOCK); #ifndef HASH_MAKE_STRING #error "HASH_MAKE_STRING must be defined!" @@ -592,11 +354,6 @@ int HASH_FINAL (unsigned char *md, HASH_CTX *c) HASH_MAKE_STRING(c,md); #endif - c->num=0; - /* clear stuff, HASH_BLOCK may be leaving some stuff on the stack - * but I'm not worried :-) - OPENSSL_cleanse((void *)c,sizeof(HASH_CTX)); - */ return 1; } diff --git a/crypto/md4/md4_dgst.c b/crypto/md4/md4_dgst.c index d4c7057f13..e5b08f3fce 100644 --- a/crypto/md4/md4_dgst.c +++ b/crypto/md4/md4_dgst.c @@ -72,89 +72,14 @@ const char *MD4_version="MD4" OPENSSL_VERSION_PTEXT; int MD4_Init(MD4_CTX *c) { + memset (c,0,sizeof(*c)); c->A=INIT_DATA_A; c->B=INIT_DATA_B; c->C=INIT_DATA_C; c->D=INIT_DATA_D; - c->Nl=0; - c->Nh=0; - c->num=0; return 1; } -#ifndef md4_block_host_order -void md4_block_host_order (MD4_CTX *c, const void *data, size_t num) - { - const MD4_LONG *X=data; - register unsigned MD32_REG_T A,B,C,D; - - A=c->A; - B=c->B; - C=c->C; - D=c->D; - - for (;num--;X+=HASH_LBLOCK) - { - /* Round 0 */ - R0(A,B,C,D,X[ 0], 3,0); - R0(D,A,B,C,X[ 1], 7,0); - R0(C,D,A,B,X[ 2],11,0); - R0(B,C,D,A,X[ 3],19,0); - R0(A,B,C,D,X[ 4], 3,0); - R0(D,A,B,C,X[ 5], 7,0); - R0(C,D,A,B,X[ 6],11,0); - R0(B,C,D,A,X[ 7],19,0); - R0(A,B,C,D,X[ 8], 3,0); - R0(D,A,B,C,X[ 9], 7,0); - R0(C,D,A,B,X[10],11,0); - R0(B,C,D,A,X[11],19,0); - R0(A,B,C,D,X[12], 3,0); - R0(D,A,B,C,X[13], 7,0); - R0(C,D,A,B,X[14],11,0); - R0(B,C,D,A,X[15],19,0); - /* Round 1 */ - R1(A,B,C,D,X[ 0], 3,0x5A827999L); - R1(D,A,B,C,X[ 4], 5,0x5A827999L); - R1(C,D,A,B,X[ 8], 9,0x5A827999L); - R1(B,C,D,A,X[12],13,0x5A827999L); - R1(A,B,C,D,X[ 1], 3,0x5A827999L); - R1(D,A,B,C,X[ 5], 5,0x5A827999L); - R1(C,D,A,B,X[ 9], 9,0x5A827999L); - R1(B,C,D,A,X[13],13,0x5A827999L); - R1(A,B,C,D,X[ 2], 3,0x5A827999L); - R1(D,A,B,C,X[ 6], 5,0x5A827999L); - R1(C,D,A,B,X[10], 9,0x5A827999L); - R1(B,C,D,A,X[14],13,0x5A827999L); - R1(A,B,C,D,X[ 3], 3,0x5A827999L); - R1(D,A,B,C,X[ 7], 5,0x5A827999L); - R1(C,D,A,B,X[11], 9,0x5A827999L); - R1(B,C,D,A,X[15],13,0x5A827999L); - /* Round 2 */ - R2(A,B,C,D,X[ 0], 3,0x6ED9EBA1); - R2(D,A,B,C,X[ 8], 9,0x6ED9EBA1); - R2(C,D,A,B,X[ 4],11,0x6ED9EBA1); - R2(B,C,D,A,X[12],15,0x6ED9EBA1); - R2(A,B,C,D,X[ 2], 3,0x6ED9EBA1); - R2(D,A,B,C,X[10], 9,0x6ED9EBA1); - R2(C,D,A,B,X[ 6],11,0x6ED9EBA1); - R2(B,C,D,A,X[14],15,0x6ED9EBA1); - R2(A,B,C,D,X[ 1], 3,0x6ED9EBA1); - R2(D,A,B,C,X[ 9], 9,0x6ED9EBA1); - R2(C,D,A,B,X[ 5],11,0x6ED9EBA1); - R2(B,C,D,A,X[13],15,0x6ED9EBA1); - R2(A,B,C,D,X[ 3], 3,0x6ED9EBA1); - R2(D,A,B,C,X[11], 9,0x6ED9EBA1); - R2(C,D,A,B,X[ 7],11,0x6ED9EBA1); - R2(B,C,D,A,X[15],15,0x6ED9EBA1); - - A = c->A += A; - B = c->B += B; - C = c->C += C; - D = c->D += D; - } - } -#endif - #ifndef md4_block_data_order #ifdef X #undef X @@ -240,19 +165,3 @@ void md4_block_data_order (MD4_CTX *c, const void *data_, size_t num) } } #endif - -#ifdef undef -int printit(unsigned long *l) - { - int i,ii; - - for (i=0; i<2; i++) - { - for (ii=0; ii<8; ii++) - { - fprintf(stderr,"%08lx ",l[i*8+ii]); - } - fprintf(stderr,"\n"); - } - } -#endif diff --git a/crypto/md4/md4_locl.h b/crypto/md4/md4_locl.h index 625b587fde..c8085b0ead 100644 --- a/crypto/md4/md4_locl.h +++ b/crypto/md4/md4_locl.h @@ -65,44 +65,13 @@ #define MD4_LONG_LOG2 2 /* default to 32 bits */ #endif -void md4_block_host_order (MD4_CTX *c, const void *p,size_t num); void md4_block_data_order (MD4_CTX *c, const void *p,size_t num); -#if defined(__i386) || defined(__i386__) || defined(_M_IX86) || defined(__INTEL__) || \ - defined(__x86_64) || defined(__x86_64__) || defined(_M_AMD64) || defined(_M_X64) -# if !defined(B_ENDIAN) -/* - * *_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 "endianness" 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:-) - * - * <appro@fy.chalmers.se> - */ -# define md4_block_data_order md4_block_host_order -# endif -#endif - #define DATA_ORDER_IS_LITTLE_ENDIAN #define HASH_LONG MD4_LONG -#define HASH_LONG_LOG2 MD4_LONG_LOG2 #define HASH_CTX MD4_CTX #define HASH_CBLOCK MD4_CBLOCK -#define HASH_LBLOCK MD4_LBLOCK #define HASH_UPDATE MD4_Update #define HASH_TRANSFORM MD4_Transform #define HASH_FINAL MD4_Final @@ -113,21 +82,7 @@ void md4_block_data_order (MD4_CTX *c, const void *p,size_t num); ll=(c)->C; HOST_l2c(ll,(s)); \ ll=(c)->D; HOST_l2c(ll,(s)); \ } while (0) -#define HASH_BLOCK_HOST_ORDER md4_block_host_order -#if !defined(L_ENDIAN) || defined(md4_block_data_order) #define HASH_BLOCK_DATA_ORDER md4_block_data_order -/* - * Little-endians (Intel and Alpha) feel better without this. - * It looks like memcpy does better job than generic - * md4_block_data_order on copying-n-aligning input data. - * But frankly 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. - * - * <appro@fy.chalmers.se> - */ -#endif #include "md32_common.h" diff --git a/crypto/md5/asm/md5-586.pl b/crypto/md5/asm/md5-586.pl index fa3fa3bed5..76ac235f7d 100644 --- a/crypto/md5/asm/md5-586.pl +++ b/crypto/md5/asm/md5-586.pl @@ -29,7 +29,7 @@ $X="esi"; 0, 7, 14, 5, 12, 3, 10, 1, 8, 15, 6, 13, 4, 11, 2, 9, # R3 ); -&md5_block("md5_block_asm_host_order"); +&md5_block("md5_block_asm_data_order"); &asm_finish(); sub Np diff --git a/crypto/md5/asm/md5-x86_64.pl b/crypto/md5/asm/md5-x86_64.pl index 75b4085a7b..4330eddb17 100755 --- a/crypto/md5/asm/md5-x86_64.pl +++ b/crypto/md5/asm/md5-x86_64.pl @@ -115,9 +115,9 @@ $code .= <<EOF; .text .align 16 -.globl md5_block_asm_host_order -.type md5_block_asm_host_order,\@function,3 -md5_block_asm_host_order: +.globl md5_block_asm_data_order +.type md5_block_asm_data_order,\@function,3 +md5_block_asm_data_order: push %rbp push %rbx push %r12 @@ -243,7 +243,7 @@ $code .= <<EOF; pop %rbx pop %rbp ret -.size md5_block_asm_host_order,.-md5_block_asm_host_order +.size md5_block_asm_data_order,.-md5_block_asm_data_order EOF print $code; diff --git a/crypto/md5/md5_dgst.c b/crypto/md5/md5_dgst.c index f97f48e55b..2708e167f7 100644 --- a/crypto/md5/md5_dgst.c +++ b/crypto/md5/md5_dgst.c @@ -72,106 +72,14 @@ const char *MD5_version="MD5" OPENSSL_VERSION_PTEXT; int MD5_Init(MD5_CTX *c) { + memset (c,0,sizeof(*c)); c->A=INIT_DATA_A; c->B=INIT_DATA_B; c->C=INIT_DATA_C; c->D=INIT_DATA_D; - c->Nl=0; - c->Nh=0; - c->num=0; return 1; } -#ifndef md5_block_host_order -void md5_block_host_order (MD5_CTX *c, const void *data, size_t num) - { - const MD5_LONG *X=data; - register unsigned MD32_REG_T A,B,C,D; - - A=c->A; - B=c->B; - C=c->C; - D=c->D; - - for (;num--;X+=HASH_LBLOCK) - { - /* Round 0 */ - R0(A,B,C,D,X[ 0], 7,0xd76aa478L); - R0(D,A,B,C,X[ 1],12,0xe8c7b756L); - R0(C,D,A,B,X[ 2],17,0x242070dbL); - R0(B,C,D,A,X[ 3],22,0xc1bdceeeL); - R0(A,B,C,D,X[ 4], 7,0xf57c0fafL); - R0(D,A,B,C,X[ 5],12,0x4787c62aL); - R0(C,D,A,B,X[ 6],17,0xa8304613L); - R0(B,C,D,A,X[ 7],22,0xfd469501L); - R0(A,B,C,D,X[ 8], 7,0x698098d8L); - R0(D,A,B,C,X[ 9],12,0x8b44f7afL); - R0(C,D,A,B,X[10],17,0xffff5bb1L); - R0(B,C,D,A,X[11],22,0x895cd7beL); - R0(A,B,C,D,X[12], 7,0x6b901122L); - R0(D,A,B,C,X[13],12,0xfd987193L); - R0(C,D,A,B,X[14],17,0xa679438eL); - R0(B,C,D,A,X[15],22,0x49b40821L); - /* Round 1 */ - R1(A,B,C,D,X[ 1], 5,0xf61e2562L); - R1(D,A,B,C,X[ 6], 9,0xc040b340L); - R1(C,D,A,B,X[11],14,0x265e5a51L); - R1(B,C,D,A,X[ 0],20,0xe9b6c7aaL); - R1(A,B,C,D,X[ 5], 5,0xd62f105dL); - R1(D,A,B,C,X[10], 9,0x02441453L); - R1(C,D,A,B,X[15],14,0xd8a1e681L); - R1(B,C,D,A,X[ 4],20,0xe7d3fbc8L); - R1(A,B,C,D,X[ 9], 5,0x21e1cde6L); - R1(D,A,B,C,X[14], 9,0xc33707d6L); - R1(C,D,A,B,X[ 3],14,0xf4d50d87L); - R1(B,C,D,A,X[ 8],20,0x455a14edL); - R1(A,B,C,D,X[13], 5,0xa9e3e905L); - R1(D,A,B,C,X[ 2], 9,0xfcefa3f8L); - R1(C,D,A,B,X[ 7],14,0x676f02d9L); - R1(B,C,D,A,X[12],20,0x8d2a4c8aL); - /* Round 2 */ - R2(A,B,C,D,X[ 5], 4,0xfffa3942L); - R2(D,A,B,C,X[ 8],11,0x8771f681L); - R2(C,D,A,B,X[11],16,0x6d9d6122L); - R2(B,C,D,A,X[14],23,0xfde5380cL); - R2(A,B,C,D,X[ 1], 4,0xa4beea44L); - R2(D,A,B,C,X[ 4],11,0x4bdecfa9L); - R2(C,D,A,B,X[ 7],16,0xf6bb4b60L); - R2(B,C,D,A,X[10],23,0xbebfbc70L); - R2(A,B,C,D,X[13], 4,0x289b7ec6L); - R2(D,A,B,C,X[ 0],11,0xeaa127faL); - R2(C,D,A,B,X[ 3],16,0xd4ef3085L); - R2(B,C,D,A,X[ 6],23,0x04881d05L); - R2(A,B,C,D,X[ 9], 4,0xd9d4d039L); - R2(D,A,B,C,X[12],11,0xe6db99e5L); - R2(C,D,A,B,X[15],16,0x1fa27cf8L); - R2(B,C,D,A,X[ 2],23,0xc4ac5665L); - /* Round 3 */ - R3(A,B,C,D,X[ 0], 6,0xf4292244L); - R3(D,A,B,C,X[ 7],10,0x432aff97L); - R3(C,D,A,B,X[14],15,0xab9423a7L); - R3(B,C,D,A,X[ 5],21,0xfc93a039L); - R3(A,B,C,D,X[12], 6,0x655b59c3L); - R3(D,A,B,C,X[ 3],10,0x8f0ccc92L); - R3(C,D,A,B,X[10],15,0xffeff47dL); - R3(B,C,D,A,X[ 1],21,0x85845dd1L); - R3(A,B,C,D,X[ 8], 6,0x6fa87e4fL); - R3(D,A,B,C,X[15],10,0xfe2ce6e0L); - R3(C,D,A,B,X[ 6],15,0xa3014314L); - R3(B,C,D,A,X[13],21,0x4e0811a1L); - R3(A,B,C,D,X[ 4], 6,0xf7537e82L); - R3(D,A,B,C,X[11],10,0xbd3af235L); - R3(C,D,A,B,X[ 2],15,0x2ad7d2bbL); - R3(B,C,D,A,X[ 9],21,0xeb86d391L); - - A = c->A += A; - B = c->B += B; - C = c->C += C; - D = c->D += D; - } - } -#endif - #ifndef md5_block_data_order #ifdef X #undef X @@ -274,19 +182,3 @@ void md5_block_data_order (MD5_CTX *c, const void *data_, size_t num) } } #endif - -#ifdef undef -int printit(unsigned long *l) - { - int i,ii; - - for (i=0; i<2; i++) - { - for (ii=0; ii<8; ii++) - { - fprintf(stderr,"%08lx ",l[i*8+ii]); - } - fprintf(stderr,"\n"); - } - } -#endif diff --git a/crypto/md5/md5_locl.h b/crypto/md5/md5_locl.h index d1375dd9f4..968d577995 100644 --- a/crypto/md5/md5_locl.h +++ b/crypto/md5/md5_locl.h @@ -68,54 +68,19 @@ #ifdef MD5_ASM # 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_host_order md5_block_asm_host_order -# elif defined(__sparc) && defined(OPENSSL_SYS_ULTRASPARC) - void md5_block_asm_data_order_aligned (MD5_CTX *c, const MD5_LONG *p,size_t num); -# define HASH_BLOCK_DATA_ORDER_ALIGNED md5_block_asm_data_order_aligned +# define md5_block_data_order md5_block_asm_data_order # elif defined(__ia64) || defined(__ia64__) || defined(_M_IA64) -# define md5_block_host_order md5_block_asm_host_order # define md5_block_data_order md5_block_asm_data_order # endif #endif -void md5_block_host_order (MD5_CTX *c, const void *p,size_t num); void md5_block_data_order (MD5_CTX *c, const void *p,size_t num); -#if defined(__i386) || defined(__i386__) || defined(_M_IX86) || defined(__INTEL__) || \ - defined(__x86_64) || defined(__x86_64__) || defined(_M_AMD64) || defined(_M_X64) -# if !defined(B_ENDIAN) -/* - * *_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 "endianness" 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:-) - * - * <appro@fy.chalmers.se> - */ -# define md5_block_data_order md5_block_host_order -# endif -#endif - #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 @@ -126,21 +91,7 @@ void md5_block_data_order (MD5_CTX *c, const void *p,size_t num); ll=(c)->C; HOST_l2c(ll,(s)); \ ll=(c)->D; HOST_l2c(ll,(s)); \ } while (0) -#define HASH_BLOCK_HOST_ORDER md5_block_host_order -#if !defined(L_ENDIAN) || defined(md5_block_data_order) #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 frankly 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. - * - * <appro@fy.chalmers.se> - */ -#endif #include "md32_common.h" diff --git a/crypto/ripemd/asm/rmd-586.pl b/crypto/ripemd/asm/rmd-586.pl index 0ab6f76bff..4f3c4c967f 100644 --- a/crypto/ripemd/asm/rmd-586.pl +++ b/crypto/ripemd/asm/rmd-586.pl @@ -1,7 +1,7 @@ #!/usr/local/bin/perl # Normal is the -# ripemd160_block_asm_host_order(RIPEMD160_CTX *c, ULONG *X,int blocks); +# ripemd160_block_asm_data_order(RIPEMD160_CTX *c, ULONG *X,int blocks); $normal=0; @@ -56,7 +56,7 @@ $KR3=0x7A6D76E9; 8, 5,12, 9,12, 5,14, 6, 8,13, 6, 5,15,13,11,11, ); -&ripemd160_block("ripemd160_block_asm_host_order"); +&ripemd160_block("ripemd160_block_asm_data_order"); &asm_finish(); sub Xv diff --git a/crypto/ripemd/rmd_dgst.c b/crypto/ripemd/rmd_dgst.c index 03a286dfcc..b04d7bf548 100644 --- a/crypto/ripemd/rmd_dgst.c +++ b/crypto/ripemd/rmd_dgst.c @@ -71,218 +71,15 @@ const char *RMD160_version="RIPE-MD160" OPENSSL_VERSION_PTEXT; int RIPEMD160_Init(RIPEMD160_CTX *c) { + memset (c,0,sizeof(*c)); c->A=RIPEMD160_A; c->B=RIPEMD160_B; c->C=RIPEMD160_C; c->D=RIPEMD160_D; c->E=RIPEMD160_E; - c->Nl=0; - c->Nh=0; - c->num=0; return 1; } -#ifndef ripemd160_block_host_order -#ifdef X -#undef X -#endif -#define X(i) XX[i] -void ripemd160_block_host_order (RIPEMD160_CTX *ctx, const void *p, size_t num) - { - const RIPEMD160_LONG *XX=p; - register unsigned MD32_REG_T A,B,C,D,E; - register unsigned MD32_REG_T a,b,c,d,e; - - for (;num--;XX+=HASH_LBLOCK) - { - - A=ctx->A; B=ctx->B; C=ctx->C; D=ctx->D; E=ctx->E; - - RIP1(A,B,C,D,E,WL00,SL00); - RIP1(E,A,B,C,D,WL01,SL01); - RIP1(D,E,A,B,C,WL02,SL02); - RIP1(C,D,E,A,B,WL03,SL03); - RIP1(B,C,D,E,A,WL04,SL04); - RIP1(A,B,C,D,E,WL05,SL05); - RIP1(E,A,B,C,D,WL06,SL06); - RIP1(D,E,A,B,C,WL07,SL07); |