#else
-#if !defined(NO_ASM) && !defined(NO_INLINE_ASM) && !defined(PEDANTIC) && !defined(BN_DIV3W)
+#if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) \
+ && !defined(PEDANTIC) && !defined(BN_DIV3W)
# if defined(__GNUC__) && __GNUC__>=2
-# if defined(__i386)
+# if defined(__i386) || defined (__i386__)
/*
* There were two reasons for implementing this template:
* - GNU C generates a call to a function (__udivdi3 to be exact)
q; \
})
# define REMAINDER_IS_ALREADY_CALCULATED
+# elif defined(__x86_64) && defined(SIXTY_FOUR_BIT_LONG)
+ /*
+ * Same story here, but it's 128-bit by 64-bit division. Wow!
+ * <appro@fy.chalmers.se>
+ */
+# define bn_div_words(n0,n1,d0) \
+ ({ asm volatile ( \
+ "divq %4" \
+ : "=a"(q), "=d"(rem) \
+ : "a"(n1), "d"(n0), "g"(d0) \
+ : "cc"); \
+ q; \
+ })
+# define REMAINDER_IS_ALREADY_CALCULATED
# endif /* __<cpu> */
# endif /* __GNUC__ */
-#endif /* NO_ASM */
+#endif /* OPENSSL_NO_ASM */
/* BN_div computes dv := num / divisor, rounding towards zero, and sets up
int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
BN_CTX *ctx)
{
- int norm_shift,i,j,loop;
+ int norm_shift,i,loop;
BIGNUM *tmp,wnum,*snum,*sdiv,*res;
BN_ULONG *resp,*wnump;
BN_ULONG d0,d1;
int num_n,div_n;
+ if (dv)
+ bn_check_top(dv);
+ if (rm)
+ bn_check_top(rm);
bn_check_top(num);
bn_check_top(divisor);
res=BN_CTX_get(ctx);
else res=dv;
if (sdiv == NULL || res == NULL) goto err;
- tmp->neg=0;
/* First we normalise the numbers */
norm_shift=BN_BITS2-((BN_num_bits(divisor))%BN_BITS2);
- BN_lshift(sdiv,divisor,norm_shift);
+ if (!(BN_lshift(sdiv,divisor,norm_shift))) goto err;
sdiv->neg=0;
norm_shift+=BN_BITS2;
- BN_lshift(snum,num,norm_shift);
+ if (!(BN_lshift(snum,num,norm_shift))) goto err;
snum->neg=0;
div_n=sdiv->top;
num_n=snum->top;
loop=num_n-div_n;
-
/* Lets setup a 'window' into snum
* This is the part that corresponds to the current
* 'area' being divided */
- BN_init(&wnum);
- wnum.d= &(snum->d[loop]);
- wnum.top= div_n;
- wnum.dmax= snum->dmax+1; /* a bit of a lie */
+ wnum.neg = 0;
+ wnum.d = &(snum->d[loop]);
+ wnum.top = div_n;
+ /* only needed when BN_ucmp messes up the values between top and max */
+ wnum.dmax = snum->dmax - loop; /* so we don't step out of bounds */
/* Get the top 2 words of sdiv */
- /* i=sdiv->top; */
+ /* div_n=sdiv->top; */
d0=sdiv->d[div_n-1];
d1=(div_n == 1)?0:sdiv->d[div_n-2];
if (BN_ucmp(&wnum,sdiv) >= 0)
{
- if (!BN_usub(&wnum,&wnum,sdiv)) goto err;
+ /* If BN_DEBUG_RAND is defined BN_ucmp changes (via
+ * bn_pollute) the const bignum arguments =>
+ * clean the values between top and max again */
+ bn_clear_top2max(&wnum);
+ bn_sub_words(wnum.d, wnum.d, sdiv->d, div_n);
*resp=1;
- res->d[res->top-1]=1;
}
else
res->top--;
- resp--;
+ /* if res->top == 0 then clear the neg value otherwise decrease
+ * the resp pointer */
+ if (res->top == 0)
+ res->neg = 0;
+ else
+ resp--;
- for (i=0; i<loop-1; i++)
+ for (i=0; i<loop-1; i++, wnump--, resp--)
{
BN_ULONG q,l0;
-#ifdef BN_DIV3W
+ /* the first part of the loop uses the top two words of
+ * snum and sdiv to calculate a BN_ULONG q such that
+ * | wnum - sdiv * q | < sdiv */
+#if defined(BN_DIV3W) && !defined(OPENSSL_NO_ASM)
+ BN_ULONG bn_div_3_words(BN_ULONG*,BN_ULONG,BN_ULONG);
q=bn_div_3_words(wnump,d1,d0);
#else
BN_ULONG n0,n1,rem=0;
q=(BN_ULONG)(((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0);
#else
q=bn_div_words(n0,n1,d0);
+#ifdef BN_DEBUG_LEVITTE
+ fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\
+X) -> 0x%08X\n",
+ n0, n1, d0, q);
+#endif
#endif
#ifndef REMAINDER_IS_ALREADY_CALCULATED
BN_ULONG t2l,t2h,ql,qh;
q=bn_div_words(n0,n1,d0);
+#ifdef BN_DEBUG_LEVITTE
+ fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\
+X) -> 0x%08X\n",
+ n0, n1, d0, q);
+#endif
#ifndef REMAINDER_IS_ALREADY_CALCULATED
rem=(n1-q*d0)&BN_MASK2;
#endif
-#ifdef BN_UMULT_HIGH
+#if defined(BN_UMULT_LOHI)
+ BN_UMULT_LOHI(t2l,t2h,d1,q);
+#elif defined(BN_UMULT_HIGH)
t2l = d1 * q;
t2h = BN_UMULT_HIGH(d1,q);
#else
#endif /* !BN_DIV3W */
l0=bn_mul_words(tmp->d,sdiv->d,div_n,q);
- wnum.d--; wnum.top++;
tmp->d[div_n]=l0;
- for (j=div_n+1; j>0; j--)
- if (tmp->d[j-1]) break;
- tmp->top=j;
-
- j=wnum.top;
- BN_sub(&wnum,&wnum,tmp);
-
- snum->top=snum->top+wnum.top-j;
-
- if (wnum.neg)
+ wnum.d--;
+ /* ingore top values of the bignums just sub the two
+ * BN_ULONG arrays with bn_sub_words */
+ if (bn_sub_words(wnum.d, wnum.d, tmp->d, div_n+1))
{
+ /* Note: As we have considered only the leading
+ * two BN_ULONGs in the calculation of q, sdiv * q
+ * might be greater than wnum (but then (q-1) * sdiv
+ * is less or equal than wnum)
+ */
q--;
- j=wnum.top;
- BN_add(&wnum,&wnum,sdiv);
- snum->top+=wnum.top-j;
+ if (bn_add_words(wnum.d, wnum.d, sdiv->d, div_n))
+ /* we can't have an overflow here (assuming
+ * that q != 0, but if q == 0 then tmp is
+ * zero anyway) */
+ (*wnump)++;
}
- *(resp--)=q;
- wnump--;
+ /* store part of the result */
+ *resp = q;
}
+ bn_correct_top(snum);
if (rm != NULL)
{
+ /* Keep a copy of the neg flag in num because if rm==num
+ * BN_rshift() will overwrite it.
+ */
+ int neg = num->neg;
BN_rshift(rm,snum,norm_shift);
if (!BN_is_zero(rm))
- rm->neg = num->neg;
+ rm->neg = neg;
+ bn_check_top(rm);
}
BN_CTX_end(ctx);
return(1);
err:
+ if (rm)
+ bn_check_top(rm);
BN_CTX_end(ctx);
return(0);
}