#define MONT_WORD /* use the faster word-based algorithm */
+#if defined(MONT_WORD) && defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32)
+static int BN_from_montgomery_word(BIGNUM *ret, BIGNUM *r, BN_MONT_CTX *mont);
+#endif
+
+
+
int BN_mod_mul_montgomery(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
BN_MONT_CTX *mont, BN_CTX *ctx)
{
if (num>1 && a->top==num && b->top==num)
{
if (bn_wexpand(r,num) == NULL) return(0);
+#if defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32) /* non-default compile */
+ if (bn_mul_mont(r->d,a->d,b->d,mont->N.d,mont->n0,num))
+#else
if (bn_mul_mont(r->d,a->d,b->d,mont->N.d,&mont->n0,num))
+#endif
{
r->neg = a->neg^b->neg;
r->top = num;
if (!BN_mul(tmp,a,b,ctx)) goto err;
}
/* reduce from aRR to aR */
+#if defined(MONT_WORD) && defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32)
+ if (!BN_from_montgomery_word(r,tmp,mont)) goto err;
+#else
if (!BN_from_montgomery(r,tmp,mont,ctx)) goto err;
+#endif
bn_check_top(r);
ret=1;
err:
return(ret);
}
+#if defined(MONT_WORD) && defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32)
+static int BN_from_montgomery_word(BIGNUM *ret, BIGNUM *r, BN_MONT_CTX *mont)
+ {
+ BIGNUM *n;
+ BN_ULONG *ap,*np,*rp,n0,v,*nrp;
+ int al,nl,max,i,x,ri;
+
+ n= &(mont->N);
+ /* mont->ri is the size of mont->N in bits (rounded up
+ to the word size) */
+ al=ri=mont->ri/BN_BITS2;
+
+ nl=n->top;
+ if ((al == 0) || (nl == 0)) { ret->top=0; return(1); }
+
+ max=(nl+al+1); /* allow for overflow (no?) XXX */
+ if (bn_wexpand(r,max) == NULL) return(0);
+
+ r->neg^=n->neg;
+ np=n->d;
+ rp=r->d;
+ nrp= &(r->d[nl]);
+
+ /* clear the top words of T */
+ for (i=r->top; i<max; i++) /* memset? XXX */
+ r->d[i]=0;
+
+ r->top=max;
+ n0=mont->n0[0];
+
+#ifdef BN_COUNT
+ fprintf(stderr,"word BN_from_montgomery_word %d * %d\n",nl,nl);
+#endif
+ for (i=0; i<nl; i++)
+ {
+#ifdef __TANDEM
+ {
+ long long t1;
+ long long t2;
+ long long t3;
+ t1 = rp[0] * (n0 & 0177777);
+ t2 = 037777600000l;
+ t2 = n0 & t2;
+ t3 = rp[0] & 0177777;
+ t2 = (t3 * t2) & BN_MASK2;
+ t1 = t1 + t2;
+ v=bn_mul_add_words(rp,np,nl,(BN_ULONG) t1);
+ }
+#else
+ v=bn_mul_add_words(rp,np,nl,(rp[0]*n0)&BN_MASK2);
+#endif
+ nrp++;
+ rp++;
+ if (((nrp[-1]+=v)&BN_MASK2) >= v)
+ continue;
+ else
+ {
+ if (((++nrp[0])&BN_MASK2) != 0) continue;
+ if (((++nrp[1])&BN_MASK2) != 0) continue;
+ for (x=2; (((++nrp[x])&BN_MASK2) == 0); x++) ;
+ }
+ }
+ bn_correct_top(r);
+
+ /* mont->ri will be a multiple of the word size and below code
+ * is kind of BN_rshift(ret,r,mont->ri) equivalent */
+ if (r->top <= ri)
+ {
+ ret->top=0;
+ return(1);
+ }
+ al=r->top-ri;
+
+ if (bn_wexpand(ret,ri) == NULL) return(0);
+ x=0-(((al-ri)>>(sizeof(al)*8-1))&1);
+ ret->top=x=(ri&~x)|(al&x); /* min(ri,al) */
+ ret->neg=r->neg;
+
+ rp=ret->d;
+ ap=&(r->d[ri]);
+
+ {
+ size_t m1,m2;
+
+ v=bn_sub_words(rp,ap,np,ri);
+ /* this ----------------^^ works even in al<ri case
+ * thanks to zealous zeroing of top of the vector in the
+ * beginning. */
+
+ /* if (al==ri && !v) || al>ri) nrp=rp; else nrp=ap; */
+ /* in other words if subtraction result is real, then
+ * trick unconditional memcpy below to perform in-place
+ * "refresh" instead of actual copy. */
+ m1=0-(size_t)(((al-ri)>>(sizeof(al)*8-1))&1); /* al<ri */
+ m2=0-(size_t)(((ri-al)>>(sizeof(al)*8-1))&1); /* al>ri */
+ m1|=m2; /* (al!=ri) */
+ m1|=(0-(size_t)v); /* (al!=ri || v) */
+ m1&=~m2; /* (al!=ri || v) && !al>ri */
+ nrp=(BN_ULONG *)(((size_t)rp&~m1)|((size_t)ap&m1));
+ }
+
+ /* 'i<ri' is chosen to eliminate dependency on input data, even
+ * though it results in redundant copy in al<ri case. */
+ for (i=0,ri-=4; i<ri; i+=4)
+ {
+ BN_ULONG t1,t2,t3,t4;
+
+ t1=nrp[i+0];
+ t2=nrp[i+1];
+ t3=nrp[i+2]; ap[i+0]=0;
+ t4=nrp[i+3]; ap[i+1]=0;
+ rp[i+0]=t1; ap[i+2]=0;
+ rp[i+1]=t2; ap[i+3]=0;
+ rp[i+2]=t3;
+ rp[i+3]=t4;
+ }
+ for (ri+=4; i<ri; i++)
+ rp[i]=nrp[i], ap[i]=0;
+ bn_correct_top(r);
+ bn_correct_top(ret);
+ bn_check_top(ret);
+
+ return(1);
+ }
+
+int BN_from_montgomery(BIGNUM *ret, const BIGNUM *a, BN_MONT_CTX *mont,
+ BN_CTX *ctx)
+ {
+ int retn=0;
+ BIGNUM *t;
+
+ BN_CTX_start(ctx);
+ if ((t = BN_CTX_get(ctx)) && BN_copy(t,a))
+ retn = BN_from_montgomery_word(ret,t,mont);
+ BN_CTX_end(ctx);
+ return retn;
+ }
+#else
+
int BN_from_montgomery(BIGNUM *ret, const BIGNUM *a, BN_MONT_CTX *mont,
BN_CTX *ctx)
{
BN_CTX_end(ctx);
return(retn);
}
+#endif /* defined(MONT_WORD) && defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32) */
BN_MONT_CTX *BN_MONT_CTX_new(void)
{
BN_init(&(ctx->RR));
BN_init(&(ctx->N));
BN_init(&(ctx->Ni));
+#if defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32) /* non-default compile */
+ ctx->n0[0] = ctx->n0[1] = 0;
+#else
+ ctx->n0 = 0;
+#endif
ctx->flags=0;
}
mont->ri=(BN_num_bits(mod)+(BN_BITS2-1))/BN_BITS2*BN_BITS2;
BN_zero(R);
+#if defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32) /* non-default compile */
+ if (!(BN_set_bit(R,2*BN_BITS2))) goto err; /* R */
+#else
if (!(BN_set_bit(R,BN_BITS2))) goto err; /* R */
+#endif
buf[0]=mod->d[0]; /* tmod = N mod word size */
buf[1]=0;
tmod.top = buf[0] != 0 ? 1 : 0;
tmod.dmax=2;
tmod.neg=0;
+
+#if defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32)
+ tmod.top=0;
+ if ((buf[0] = mod->d[0])) tmod.top=1;
+ if ((buf[1] = mod->top>1 ? mod->d[1] : 0)) tmod.top=2;
+
+ if ((BN_mod_inverse(Ri,R,&tmod,ctx)) == NULL)
+ goto err;
+ if (!BN_lshift(Ri,Ri,2*BN_BITS2)) goto err; /* R*Ri */
+ if (!BN_is_zero(Ri))
+ {
+ if (!BN_sub_word(Ri,1)) goto err;
+ }
+ else /* if N mod word size == 1 */
+ {
+ if (bn_expand(Ri,(int)sizeof(BN_ULONG)*2) == NULL)
+ goto err;
+ /* Ri-- (mod double word size) */
+ Ri->neg=0;
+ Ri->d[0]=BN_MASK2;
+ Ri->d[1]=BN_MASK2;
+ Ri->top=2;
+ }
+ if (!BN_div(Ri,NULL,Ri,&tmod,ctx)) goto err;
+ /* Ni = (R*Ri-1)/N,
+ * keep only couple of least significant words: */
+ mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0;
+ mont->n0[1] = (Ri->top > 1) ? Ri->d[1] : 0;
+#else
/* Ri = R^-1 mod N*/
if ((BN_mod_inverse(Ri,R,&tmod,ctx)) == NULL)
goto err;
/* Ni = (R*Ri-1)/N,
* keep only least significant word: */
mont->n0 = (Ri->top > 0) ? Ri->d[0] : 0;
+#endif
}
#else /* !MONT_WORD */
{ /* bignum version */
if (!BN_copy(&(to->N),&(from->N))) return NULL;
if (!BN_copy(&(to->Ni),&(from->Ni))) return NULL;
to->ri=from->ri;
+#if defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32) /* non-default compile */
+ to->n0[0]=from->n0[0];
+ to->n0[1]=from->n0[1];
+#else
to->n0=from->n0;
+#endif
return(to);
}