/* crypto/bn/bn_lcl.h */
-/* Copyright (C) 1995-1997 Eric Young (eay@cryptsoft.com)
+/* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
* All rights reserved.
*
* This package is an SSL implementation written
#ifndef HEADER_BN_LCL_H
#define HEADER_BN_LCL_H
-#include "bn.h"
+#include <openssl/bn.h>
#ifdef __cplusplus
extern "C" {
#endif
+/* Pentium pro 16,16,16,32,64 */
+/* Alpha 16,16,16,16.64 */
+#define BN_MULL_SIZE_NORMAL (16) /* 32 */
+#define BN_MUL_RECURSIVE_SIZE_NORMAL (16) /* 32 less than */
+#define BN_SQR_RECURSIVE_SIZE_NORMAL (16) /* 32 */
+#define BN_MUL_LOW_RECURSIVE_SIZE_NORMAL (32) /* 32 */
+#define BN_MONT_CTX_SET_SIZE_WORD (64) /* 32 */
+
+#if 0
+#ifndef BN_MUL_COMBA
+/* #define bn_mul_comba8(r,a,b) bn_mul_normal(r,a,8,b,8) */
+/* #define bn_mul_comba4(r,a,b) bn_mul_normal(r,a,4,b,4) */
+#endif
+
+#ifndef BN_SQR_COMBA
+/* This is probably faster than using the C code - I need to check */
+#define bn_sqr_comba8(r,a) bn_mul_normal(r,a,8,a,8)
+#define bn_sqr_comba4(r,a) bn_mul_normal(r,a,4,a,4)
+#endif
+#endif
+
+#if !defined(NO_ASM) && !defined(PEDANTIC)
+/*
+ * BN_UMULT_HIGH section.
+ *
+ * No, I'm not trying to overwhelm you when stating that the
+ * product of N-bit numbers is 2*N bits wide:-) No, I don't expect
+ * you to be impressed when I say that if the compiler doesn't
+ * support 2*N integer type, then you have to replace every N*N
+ * multiplication with 4 (N/2)*(N/2) accompanied by some shifts
+ * and additions which unavoidably results in severe performance
+ * penalties. Of course provided that the hardware is capable of
+ * producing 2*N result... That's when you normally start
+ * considering assembler implementation. However! It should be
+ * pointed out that some CPUs (most notably Alpha, PowerPC and
+ * upcoming IA-64 family:-) provide *separate* instruction
+ * calculating the upper half of the product placing the result
+ * into a general purpose register. Now *if* the compiler supports
+ * inline assembler, then it's not impossible to implement the
+ * "bignum" routines (and have the compiler optimize 'em)
+ * exhibiting "native" performance in C. That's what BN_UMULT_HIGH
+ * macro is about:-)
+ *
+ * <appro@fy.chalmers.se>
+ */
+# if defined(__alpha) && (defined(SIXTY_FOUR_BIT_LONG) || defined(SIXTY_FOUR_BIT))
+# if defined(__DECC)
+# include <c_asm.h>
+# define BN_UMULT_HIGH(a,b) (BN_ULONG)asm("umulh %a0,%a1,%v0",(a),(b))
+# elif defined(__GNUC__)
+# define BN_UMULT_HIGH(a,b) ({ \
+ register BN_ULONG ret; \
+ asm ("umulh %1,%2,%0" \
+ : "=r"(ret) \
+ : "r"(a), "r"(b)); \
+ ret; })
+# endif /* compiler */
+# elif defined(_ARCH_PPC) && defined(__64BIT__) && defined(SIXTY_FOUR_BIT_LONG)
+# if defined(__GNUC__)
+# define BN_UMULT_HIGH(a,b) ({ \
+ register BN_ULONG ret; \
+ asm ("mulhdu %0,%1,%2" \
+ : "=r"(ret) \
+ : "r"(a), "r"(b)); \
+ ret; })
+# endif /* compiler */
+# endif /* cpu */
+#endif /* NO_ASM */
+
/*************************************************************
* Using the long long type
*/
#define Lw(t) (((BN_ULONG)(t))&BN_MASK2)
#define Hw(t) (((BN_ULONG)((t)>>BN_BITS2))&BN_MASK2)
-#define bn_fix_top(a) \
- { \
- BN_ULONG *fix_top_l; \
- for (fix_top_l= &((a)->d[(a)->top-1]); (a)->top > 0; (a)->top--) \
- if (*(fix_top_l--)) break; \
+/* These are used for internal error checking and are not normally used */
+#ifdef BN_DEBUG
+#define bn_check_top(a) \
+ { if (((a)->top < 0) || ((a)->top > (a)->max)) \
+ { char *nullp=NULL; *nullp='z'; } }
+#define bn_check_num(a) if ((a) < 0) { char *nullp=NULL; *nullp='z'; }
+#else
+#define bn_check_top(a)
+#define bn_check_num(a)
+#endif
+
+/* This macro is to add extra stuff for development checking */
+#ifdef BN_DEBUG
+#define bn_set_max(r) ((r)->max=(r)->top,BN_set_flags((r),BN_FLG_STATIC_DATA))
+#else
+#define bn_set_max(r)
+#endif
+
+/* These macros are used to 'take' a section of a bignum for read only use */
+#define bn_set_low(r,a,n) \
+ { \
+ (r)->top=((a)->top > (n))?(n):(a)->top; \
+ (r)->d=(a)->d; \
+ (r)->neg=(a)->neg; \
+ (r)->flags|=BN_FLG_STATIC_DATA; \
+ bn_set_max(r); \
}
-#define bn_expand(n,b) ((((b)/BN_BITS2) <= (n)->max)?(n):bn_expand2((n),(b)))
+#define bn_set_high(r,a,n) \
+ { \
+ if ((a)->top > (n)) \
+ { \
+ (r)->top=(a)->top-n; \
+ (r)->d= &((a)->d[n]); \
+ } \
+ else \
+ (r)->top=0; \
+ (r)->neg=(a)->neg; \
+ (r)->flags|=BN_FLG_STATIC_DATA; \
+ bn_set_max(r); \
+ }
+
+/* #define bn_expand(n,b) ((((b)/BN_BITS2) <= (n)->max)?(n):bn_expand2((n),(b))) */
#ifdef BN_LLONG
#define mul_add(r,a,w,c) { \
BN_ULLONG t; \
t=(BN_ULLONG)w * (a) + (r) + (c); \
- (r)=Lw(t); \
+ (r)= Lw(t); \
(c)= Hw(t); \
}
#define mul(r,a,w,c) { \
BN_ULLONG t; \
t=(BN_ULLONG)w * (a) + (c); \
- (r)=Lw(t); \
+ (r)= Lw(t); \
(c)= Hw(t); \
}
-#define bn_mul_words(r1,r2,a,b) \
- { \
+#define sqr(r0,r1,a) { \
BN_ULLONG t; \
- t=(BN_ULLONG)(a)*(b); \
- r1=Lw(t); \
- r2=Hw(t); \
+ t=(BN_ULLONG)(a)*(a); \
+ (r0)=Lw(t); \
+ (r1)=Hw(t); ]
+ }
+
+#elif defined(BN_UMULT_HIGH)
+#define mul_add(r,a,w,c) { \
+ BN_ULONG high,low,ret,tmp=(a); \
+ ret = (r); \
+ high= BN_UMULT_HIGH(w,tmp); \
+ ret += (c); \
+ low = (w) * tmp; \
+ (c) = (ret<(c))?1:0; \
+ (c) += high; \
+ ret += low; \
+ (c) += (ret<low)?1:0; \
+ (r) = ret; \
+ }
+
+#define mul(r,a,w,c) { \
+ BN_ULONG high,low,ret,ta=(a); \
+ low = (w) * ta; \
+ high= BN_UMULT_HIGH(w,ta); \
+ ret = low + (c); \
+ (c) = high; \
+ (c) += (ret<low)?1:0; \
+ (r) = ret; \
+ }
+
+#define sqr(r0,r1,a) { \
+ BN_ULONG tmp=(a); \
+ (r0) = tmp * tmp; \
+ (r1) = BN_UMULT_HIGH(tmp,tmp); \
}
#else
lt=(bl)*(lt); \
m1=(bl)*(ht); \
ht =(bh)*(ht); \
- m+=m1; if ((m&BN_MASK2) < m1) ht+=L2HBITS(1L); \
+ m=(m+m1)&BN_MASK2; if (m < m1) ht+=L2HBITS(1L); \
ht+=HBITS(m); \
m1=L2HBITS(m); \
- lt+=m1; if ((lt&BN_MASK2) < m1) ht++; \
+ lt=(lt+m1)&BN_MASK2; if (lt < m1) ht++; \
(l)=lt; \
(h)=ht; \
}
h*=h; \
h+=(m&BN_MASK2h1)>>(BN_BITS4-1); \
m =(m&BN_MASK2l)<<(BN_BITS4+1); \
- l+=m; if ((l&BN_MASK2) < m) h++; \
+ l=(l+m)&BN_MASK2; if (l < m) h++; \
(lo)=l; \
(ho)=h; \
}
mul64(l,h,(bl),(bh)); \
\
/* non-multiply part */ \
- l+=(c); if ((l&BN_MASK2) < (c)) h++; \
+ l=(l+(c))&BN_MASK2; if (l < (c)) h++; \
(c)=(r); \
- l+=(c); if ((l&BN_MASK2) < (c)) h++; \
+ l=(l+(c))&BN_MASK2; if (l < (c)) h++; \
(c)=h&BN_MASK2; \
- (r)=l&BN_MASK2; \
+ (r)=l; \
}
#define mul(r,a,bl,bh,c) { \
(r)=l&BN_MASK2; \
}
-#define bn_mul_words(r1,r2,a,b) \
- { \
- BN_ULONG l,h,bl,bh; \
- \
- h=(a); \
- l=LBITS(h); \
- h=HBITS(h); \
- bh=(b); \
- bl=LBITS(bh); \
- bh=HBITS(bh); \
- \
- mul64(l,h,bl,bh); \
- \
- (r1)=l; \
- (r2)=h; \
- }
#endif
-#ifndef NOPROTO
+OPENSSL_EXTERN int bn_limit_bits;
+OPENSSL_EXTERN int bn_limit_num; /* (1<<bn_limit_bits) */
+/* Recursive 'low' limit */
+OPENSSL_EXTERN int bn_limit_bits_low;
+OPENSSL_EXTERN int bn_limit_num_low; /* (1<<bn_limit_bits_low) */
+/* Do modified 'high' part calculation' */
+OPENSSL_EXTERN int bn_limit_bits_high;
+OPENSSL_EXTERN int bn_limit_num_high; /* (1<<bn_limit_bits_high) */
+OPENSSL_EXTERN int bn_limit_bits_mont;
+OPENSSL_EXTERN int bn_limit_num_mont; /* (1<<bn_limit_bits_mont) */
BIGNUM *bn_expand2(BIGNUM *b, int bits);
-#else
-
-BIGNUM *bn_expand2();
-
-#endif
+void bn_mul_normal(BN_ULONG *r,BN_ULONG *a,int na,BN_ULONG *b,int nb);
+void bn_mul_comba8(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b);
+void bn_mul_comba4(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b);
+void bn_sqr_normal(BN_ULONG *r, BN_ULONG *a, int n, BN_ULONG *tmp);
+void bn_sqr_comba8(BN_ULONG *r,BN_ULONG *a);
+void bn_sqr_comba4(BN_ULONG *r,BN_ULONG *a);
+int bn_cmp_words(BN_ULONG *a,BN_ULONG *b,int n);
+void bn_mul_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2,BN_ULONG *t);
+void bn_mul_part_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,
+ int tn, int n,BN_ULONG *t);
+void bn_sqr_recursive(BN_ULONG *r,BN_ULONG *a, int n2, BN_ULONG *t);
+void bn_mul_low_normal(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b, int n);
+void bn_mul_low_recursive(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,int n2,
+ BN_ULONG *t);
+void bn_mul_high(BN_ULONG *r,BN_ULONG *a,BN_ULONG *b,BN_ULONG *l,int n2,
+ BN_ULONG *t);
#ifdef __cplusplus
}
projects / openssl.git / blobdiff