Commit dc434bbc authored by Bodo Möller's avatar Bodo Möller
Browse files

Slightly faster DSA verification (BN_mod_exp2_mont), 

marginally faster BN_mod_exp for 1024 bit exponents.
parent 947b3b8b

CHANGES

+12 −1
Original line number Diff line number Diff line
@@ -4,6 +4,17 @@ 


 Changes between 0.9.5a and 0.9.6  [xx XXX 2000]



  *) Re-implement BN_mod_exp2_mont using independent (and larger) windows.

     This makes DSA verification about 2 % faster.

     [Bodo Moeller]



  *) Increase maximum window size in BN_mod_exp_... to 6 bits instead of 5

     (meaning that now 2^5 values will be precomputed, which is only 4 KB

     plus overhead for 1024 bit moduli).

     This makes exponentiations about 0.5 % faster for 1024 bit

     exponents (as measured by "openssl speed rsa2048").

     [Bodo Moeller]



  *) Rename memory handling macros to avoid conflicts with other

     software:

          Malloc         =>  OPENSSL_malloc
@@ -13,7 +24,7 @@ 
     [Richard Levitte]



  *) New function BN_mod_exp_mont_word for small bases (roughly 15-20%

     faster than BN_mod_exp_mont).

     faster than BN_mod_exp_mont, i.e. 7.5-10% for a full DH exchange).

     [Bodo Moeller]



  *) CygWin32 support.

TABLE

+11 −11
Original line number Diff line number Diff line
@@ -632,7 +632,7 @@ $dso_scheme = 


*** debug-ben

$cc           = gcc

$cflags       = -DBN_DEBUG -DREF_CHECK -DBN_CTX_DEBUG -DCRYPTO_MDEBUG -DPEDANTIC -O2 -pedantic -Wall -Wshadow -Werror -pipe

$cflags       = -DBN_DEBUG -DREF_CHECK -DBN_CTX_DEBUG -DCRYPTO_MDEBUG -DPEDANTIC -DDEBUG_SAFESTACK -O2 -pedantic -Wall -Wshadow -Werror -pipe

$unistd       = 

$thread_cflag = (unknown)

$lflags       =
@@ -650,7 +650,7 @@ $dso_scheme = 


*** debug-ben-debug

$cc           = gcc

$cflags       = -DBN_DEBUG -DREF_CHECK -DBN_CTX_DEBUG -DCRYPTO_MDEBUG -g3 -O2 -pedantic -Wall -Wshadow -Werror -pipe

$cflags       = -DBN_DEBUG -DREF_CHECK -DBN_CTX_DEBUG -DCRYPTO_MDEBUG -DPEDANTIC -DDEBUG_SAFESTACK -g3 -O2 -pedantic -Wall -Wshadow -Werror -pipe

$unistd       = 

$thread_cflag = (unknown)

$lflags       =
@@ -1228,7 +1228,7 @@ $dso_scheme = 
$cc           = cc

$cflags       = -n32 -O2 -use_readonly_const -DTERMIOS -DB_ENDIAN -DBN_DIV3W

$unistd       = 

$thread_cflag = (unknown)

$thread_cflag = -D_SGI_MP_SOURCE

$lflags       = 

$bn_ops       = DES_PTR RC4_CHAR RC4_CHUNK_LL DES_RISC2 DES_UNROLL BF_PTR SIXTY_FOUR_BIT

$bn_obj       = asm/mips3.o
@@ -1246,7 +1246,7 @@ $dso_scheme = 
$cc           = gcc

$cflags       = -mabi=n32 -mmips-as -O3 -DTERMIOS -DB_ENDIAN -DBN_DIV3W

$unistd       = 

$thread_cflag = (unknown)

$thread_cflag = -D_SGI_MP_SOURCE

$lflags       = 

$bn_ops       = MD2_CHAR RC4_INDEX RC4_CHAR RC4_CHUNK_LL DES_UNROLL DES_RISC2 DES_PTR BF_PTR SIXTY_FOUR_BIT

$bn_obj       = asm/mips3.o
@@ -1264,7 +1264,7 @@ $dso_scheme = 
$cc           = cc

$cflags       = -64 -mips4 -O2 -use_readonly_const -DTERMIOS -DB_ENDIAN -DBN_DIV3W

$unistd       = 

$thread_cflag = (unknown)

$thread_cflag = -D_SGI_MP_SOURCE

$lflags       = 

$bn_ops       = RC4_CHAR RC4_CHUNK DES_RISC2 DES_UNROLL SIXTY_FOUR_BIT_LONG

$bn_obj       = asm/mips3.o
@@ -1282,7 +1282,7 @@ $dso_scheme = 
$cc           = gcc

$cflags       = -mabi=64 -mips4 -mmips-as -O3 -DTERMIOS -DB_ENDIAN -DBN_DIV3W

$unistd       = 

$thread_cflag = (unknown)

$thread_cflag = -D_SGI_MP_SOURCE

$lflags       = 

$bn_ops       = RC4_CHAR RC4_CHUNK DES_RISC2 DES_UNROLL SIXTY_FOUR_BIT_LONG

$bn_obj       = asm/mips3.o
@@ -1300,7 +1300,7 @@ $dso_scheme = 
$cc           = ccc

$cflags       = -fast -readonly_strings -DL_ENDIAN -DTERMIO

$unistd       = 

$thread_cflag = (unknown)

$thread_cflag = -D_REENTRANT

$lflags       = 

$bn_ops       = SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_INT DES_PTR DES_RISC1 DES_UNROLL

$bn_obj       = asm/alpha.o
@@ -1318,7 +1318,7 @@ $dso_scheme = 
$cc           = gcc

$cflags       = -O3 -DL_ENDIAN -DTERMIO

$unistd       = 

$thread_cflag = (unknown)

$thread_cflag = -D_REENTRANT

$lflags       = 

$bn_ops       = SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL

$bn_obj       = asm/alpha.o
@@ -1336,7 +1336,7 @@ $dso_scheme = 
$cc           = ccc

$cflags       = -fast -readonly_strings -DL_ENDIAN -DTERMIO

$unistd       = 

$thread_cflag = (unknown)

$thread_cflag = -D_REENTRANT

$lflags       = 

$bn_ops       = SIXTY_FOUR_BIT_LONG RC4_CHUNK DES_INT DES_PTR DES_RISC1 DES_UNROLL

$bn_obj       = asm/alpha.o
@@ -1354,7 +1354,7 @@ $dso_scheme = 
$cc           = gcc

$cflags       = -O3 -DL_ENDIAN -DTERMIO

$unistd       = 

$thread_cflag = (unknown)

$thread_cflag = -D_REENTRANT

$lflags       = 

$bn_ops       = SIXTY_FOUR_BIT_LONG RC4_CHUNK DES_RISC1 DES_UNROLL

$bn_obj       = asm/alpha.o
@@ -1859,7 +1859,7 @@ $cc = gcc 
$cflags       = -O3 -mv8 -Dssize_t=int

$unistd       = 

$thread_cflag = (unknown)

$lflags       = 

$lflags       = -liberty

$bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL DES_PTR DES_RISC1

$bn_obj       = 

$des_obj      =

crypto/bn/bn.h

+1 −0
Original line number Diff line number Diff line
@@ -485,6 +485,7 @@ BN_ULONG bn_sub_words(BN_ULONG *rp, BN_ULONG *ap, BN_ULONG *bp,int num); 
#define BN_F_BN_CTX_NEW					 106

#define BN_F_BN_DIV					 107

#define BN_F_BN_EXPAND2					 108

#define BN_F_BN_MOD_EXP2_MONT				 118

#define BN_F_BN_MOD_EXP_MONT				 109

#define BN_F_BN_MOD_EXP_MONT_WORD			 117

#define BN_F_BN_MOD_INVERSE				 110

crypto/bn/bn_err.c

+2 −1
Original line number Diff line number Diff line
@@ -76,8 +76,9 @@ static ERR_STRING_DATA BN_str_functs[]= 
{ERR_PACK(0,BN_F_BN_CTX_NEW,0),	"BN_CTX_new"},

{ERR_PACK(0,BN_F_BN_DIV,0),	"BN_div"},

{ERR_PACK(0,BN_F_BN_EXPAND2,0),	"bn_expand2"},

{ERR_PACK(0,BN_F_BN_MOD_EXP2_MONT,0),	"BN_mod_exp2_mont"},

{ERR_PACK(0,BN_F_BN_MOD_EXP_MONT,0),	"BN_mod_exp_mont"},

{ERR_PACK(0,BN_F_BN_MOD_EXP_MONT_WORD,0),	"BN_MOD_EXP_MONT_WORD"},

{ERR_PACK(0,BN_F_BN_MOD_EXP_MONT_WORD,0),	"BN_mod_exp_mont_word"},

{ERR_PACK(0,BN_F_BN_MOD_INVERSE,0),	"BN_mod_inverse"},

{ERR_PACK(0,BN_F_BN_MOD_MUL_RECIPROCAL,0),	"BN_mod_mul_reciprocal"},

{ERR_PACK(0,BN_F_BN_MPI2BN,0),	"BN_mpi2bn"},

crypto/bn/bn_exp.c

+37 −52
Original line number Diff line number Diff line
@@ -121,7 +121,7 @@ 
#endif





#define TABLE_SIZE	16

#define TABLE_SIZE	32



/* slow but works */

int BN_mod_mul(BIGNUM *ret, BIGNUM *a, BIGNUM *b, const BIGNUM *m, BN_CTX *ctx)
@@ -427,18 +427,12 @@ int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, 
	ts=1;



	if (!BN_mod(&(val[0]),a,m,ctx)) goto err;		/* 1 */



	window = BN_window_bits_for_exponent_size(bits);

	if (window > 1)

		{

		if (!BN_mod_mul_reciprocal(aa,&(val[0]),&(val[0]),&recp,ctx))

			goto err;				/* 2 */



	if (bits <= 17) /* This is probably 3 or 0x10001, so just do singles */

		window=1;

	else if (bits >= 256)

		window=5;	/* max size of window */

	else if (bits >= 128)

		window=4;

	else

		window=3;



		j=1<<(window-1);

		for (i=1; i<j; i++)

			{
@@ -447,6 +441,7 @@ int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, 
				goto err;

			}

		ts=i;

		}

		

	start=1;	/* This is used to avoid multiplication etc

			 * when there is only the value '1' in the
@@ -574,17 +569,11 @@ int BN_mod_exp_mont(BIGNUM *rr, BIGNUM *a, const BIGNUM *p, 
	else

		aa=a;

	if (!BN_to_montgomery(&(val[0]),aa,mont,ctx)) goto err; /* 1 */

	if (!BN_mod_mul_montgomery(d,&(val[0]),&(val[0]),mont,ctx)) goto err; /* 2 */



	if (bits <= 20) /* This is probably 3 or 0x10001, so just do singles */

		window=1;

	else if (bits >= 256)

		window=5;	/* max size of window */

	else if (bits >= 128)

		window=4;

	else

		window=3;



	window = BN_window_bits_for_exponent_size(bits);

	if (window > 1)

		{

		if (!BN_mod_mul_montgomery(d,&(val[0]),&(val[0]),mont,ctx)) goto err; /* 2 */

		j=1<<(window-1);

		for (i=1; i<j; i++)

			{
@@ -593,6 +582,7 @@ int BN_mod_exp_mont(BIGNUM *rr, BIGNUM *a, const BIGNUM *p, 
				goto err;

			}

		ts=i;

		}



	start=1;	/* This is used to avoid multiplication etc

			 * when there is only the value '1' in the
@@ -787,18 +777,12 @@ int BN_mod_exp_simple(BIGNUM *r, BIGNUM *a, BIGNUM *p, BIGNUM *m, 
	BN_init(&(val[0]));

	ts=1;

	if (!BN_mod(&(val[0]),a,m,ctx)) goto err;		/* 1 */



	window = BN_window_bits_for_exponent_size(bits);

	if (window > 1)

		{

		if (!BN_mod_mul(d,&(val[0]),&(val[0]),m,ctx))

			goto err;				/* 2 */



	if (bits <= 17) /* This is probably 3 or 0x10001, so just do singles */

		window=1;

	else if (bits >= 256)

		window=5;	/* max size of window */

	else if (bits >= 128)

		window=4;

	else

		window=3;



		j=1<<(window-1);

		for (i=1; i<j; i++)

			{
@@ -807,6 +791,7 @@ int BN_mod_exp_simple(BIGNUM *r, BIGNUM *a, BIGNUM *p, BIGNUM *m, 
				goto err;

			}

		ts=i;

		}



	start=1;	/* This is used to avoid multiplication etc

			 * when there is only the value '1' in the