int BN_num_bits_word(BN_ULONG l)
{
- static char bits[256]={
+ static const char bits[256]={
0,1,2,2,3,3,3,3,4,4,4,4,4,4,4,4,
5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
}
}
-int BN_num_bits(BIGNUM *a)
+int BN_num_bits(const BIGNUM *a)
{
BN_ULONG l;
int i;
BIGNUM *bn_expand2(BIGNUM *b, int words)
{
- BN_ULONG *A,*B,*a;
- int i,j;
+ BN_ULONG *A,*a;
+ const BN_ULONG *B;
+ int i;
bn_check_top(b);
BNerr(BN_F_BN_EXPAND2,ERR_R_MALLOC_FAILURE);
return(NULL);
}
-memset(A,0x5c,sizeof(BN_ULONG)*(words+1));
#if 1
B=b->d;
/* Check if the previous number needs to be copied */
if (B != NULL)
{
+#if 0
/* This lot is an unrolled loop to copy b->top
* BN_ULONGs from B to A
*/
+/*
+ * I have nothing against unrolling but it's usually done for
+ * several reasons, namely:
+ * - minimize percentage of decision making code, i.e. branches;
+ * - avoid cache trashing;
+ * - make it possible to schedule loads earlier;
+ * Now let's examine the code below. The cornerstone of C is
+ * "programmer is always right" and that's what we love it for:-)
+ * For this very reason C compilers have to be paranoid when it
+ * comes to data aliasing and assume the worst. Yeah, but what
+ * does it mean in real life? This means that loop body below will
+ * be compiled to sequence of loads immediately followed by stores
+ * as compiler assumes the worst, something in A==B+1 style. As a
+ * result CPU pipeline is going to starve for incoming data. Secondly
+ * if A and B happen to share same cache line such code is going to
+ * cause severe cache trashing. Both factors have severe impact on
+ * performance of modern CPUs and this is the reason why this
+ * particulare piece of code is #ifdefed away and replaced by more
+ * "friendly" version found in #else section below. This comment
+ * also applies to BN_copy function.
+ *
+ * <appro@fy.chalmers.se>
+ */
for (i=b->top&(~7); i>0; i-=8)
{
A[0]=B[0]; A[1]=B[1]; A[2]=B[2]; A[3]=B[3];
*/
;
}
+#else
+ for (i=b->top>>2; i>0; i--,A+=4,B+=4)
+ {
+ /*
+ * The fact that the loop is unrolled
+ * 4-wise is a tribute to Intel. It's
+ * the one that doesn't have enough
+ * registers to accomodate more data.
+ * I'd unroll it 8-wise otherwise:-)
+ *
+ * <appro@fy.chalmers.se>
+ */
+ BN_ULONG a0,a1,a2,a3;
+ a0=B[0]; a1=B[1]; a2=B[2]; a3=B[3];
+ A[0]=a0; A[1]=a1; A[2]=a2; A[3]=a3;
+ }
+ switch (b->top&3)
+ {
+ case 3: A[2]=B[2];
+ case 2: A[1]=B[1];
+ case 1: A[0]=B[0];
+ case 0: ; /* ultrix cc workaround, see above */
+ }
+#endif
Free(b->d);
}
/* Now need to zero any data between b->top and b->max */
- B= &(b->d[b->top]);
- j=(b->max - b->top) & ~7;
- for (i=0; i<j; i+=8)
+ A= &(b->d[b->top]);
+ for (i=(b->max - b->top)>>3; i>0; i--,A+=8)
{
- B[0]=0; B[1]=0; B[2]=0; B[3]=0;
- B[4]=0; B[5]=0; B[6]=0; B[7]=0;
- B+=8;
- }
- j=(b->max - b->top) & 7;
- for (i=0; i<j; i++)
- {
- B[0]=0;
- B++;
+ A[0]=0; A[1]=0; A[2]=0; A[3]=0;
+ A[4]=0; A[5]=0; A[6]=0; A[7]=0;
}
+ for (i=(b->max - b->top)&7; i>0; i--,A++)
+ A[0]=0;
#else
- memcpy(a->d,b->d,sizeof(b->d[0])*b->top);
+ memset(A,0,sizeof(BN_ULONG)*(words+1));
+ memcpy(A,b->d,sizeof(b->d[0])*b->top);
+ b->d=a;
+ b->max=words;
#endif
/* memset(&(p[b->max]),0,((words+1)-b->max)*sizeof(BN_ULONG)); */
return(b);
}
-BIGNUM *BN_dup(BIGNUM *a)
+BIGNUM *BN_dup(const BIGNUM *a)
{
BIGNUM *r;
return((BIGNUM *)BN_copy(r,a));
}
-BIGNUM *BN_copy(BIGNUM *a, BIGNUM *b)
+BIGNUM *BN_copy(BIGNUM *a, const BIGNUM *b)
{
int i;
- BN_ULONG *A,*B;
+ BN_ULONG *A;
+ const BN_ULONG *B;
bn_check_top(b);
#if 1
A=a->d;
B=b->d;
- for (i=b->top&(~7); i>0; i-=8)
+ for (i=b->top>>2; i>0; i--,A+=4,B+=4)
{
- A[0]=B[0];
- A[1]=B[1];
- A[2]=B[2];
- A[3]=B[3];
- A[4]=B[4];
- A[5]=B[5];
- A[6]=B[6];
- A[7]=B[7];
- A+=8;
- B+=8;
+ BN_ULONG a0,a1,a2,a3;
+ a0=B[0]; a1=B[1]; a2=B[2]; a3=B[3];
+ A[0]=a0; A[1]=a1; A[2]=a2; A[3]=a3;
}
- switch (b->top&7)
+ switch (b->top&3)
{
- case 7:
- A[6]=B[6];
- case 6:
- A[5]=B[5];
- case 5:
- A[4]=B[4];
- case 4:
- A[3]=B[3];
- case 3:
- A[2]=B[2];
- case 2:
- A[1]=B[1];
- case 1:
- A[0]=B[0];
- case 0:
- /* I need the 'case 0' entry for utrix cc.
- * If the optimiser is turned on, it does the
- * switch table by doing
- * a=top&7
- * a--;
- * goto jump_table[a];
- * If top is 0, this makes us jump to 0xffffffc which is
- * rather bad :-(.
- * eric 23-Apr-1998
- */
- ;
+ case 3: A[2]=B[2];
+ case 2: A[1]=B[1];
+ case 1: A[0]=B[0];
+ case 0: ; /* ultrix cc workaround, see comments in bn_expand2 */
}
#else
memcpy(a->d,b->d,sizeof(b->d[0])*b->top);
#ifndef SIXTY_FOUR_BIT /* the data item > unsigned long */
ret<<=BN_BITS4; /* stops the compiler complaining */
ret<<=BN_BITS4;
+#else
+ ret=0;
#endif
ret|=a->d[i];
}
#ifndef SIXTY_FOUR_BIT /* the data item > unsigned long */
w>>=BN_BITS4;
w>>=BN_BITS4;
+#else
+ w=0;
#endif
a->d[i]=(BN_ULONG)w&BN_MASK2;
if (a->d[i] != 0) a->top=i+1;
}
/* ignore negative */
-int BN_bn2bin(BIGNUM *a, unsigned char *to)
+int BN_bn2bin(const BIGNUM *a, unsigned char *to)
{
int n,i;
BN_ULONG l;
return(n);
}
-int BN_ucmp(BIGNUM *a, BIGNUM *b)
+int BN_ucmp(const BIGNUM *a, const BIGNUM *b)
{
int i;
BN_ULONG t1,t2,*ap,*bp;
return(0);
}
-int BN_cmp(BIGNUM *a, BIGNUM *b)
+int BN_cmp(const BIGNUM *a, const BIGNUM *b)
{
int i;
int gt,lt;
a->top=i+1;
}
- a->d[i]|=(1L<<j);
+ a->d[i]|=(((BN_ULONG)1)<<j);
return(1);
}
j=n%BN_BITS2;
if (a->top <= i) return(0);
- a->d[i]&=(~(1L<<j));
+ a->d[i]&=(~(((BN_ULONG)1)<<j));
bn_fix_top(a);
return(1);
}
-int BN_is_bit_set(BIGNUM *a, int n)
+int BN_is_bit_set(const BIGNUM *a, int n)
{
int i,j;
projects / openssl.git / blobdiff