# include <alloca.h>
#endif
+#undef RSAZ_ENABLED
+#if defined(OPENSSL_BN_ASM_MONT) && \
+ (defined(__x86_64) || defined(__x86_64__) || \
+ defined(_M_AMD64) || defined(_M_X64))
+# include "rsaz_exp.h"
+# define RSAZ_ENABLED
+#endif
+
+#undef SPARC_T4_MONT
#if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc))
# include "sparc_arch.h"
extern unsigned int OPENSSL_sparcv9cap_P[];
+# define SPARC_T4_MONT
#endif
/* maximum precomputation table size for *variable* sliding windows */
wend=0; /* The bottom bit of the window */
#if 1 /* by Shay Gueron's suggestion */
- j = mont->N.top; /* borrow j */
- if (bn_wexpand(r,j) == NULL) goto err;
- r->d[0] = (0-m->d[0])&BN_MASK2; /* 2^(top*BN_BITS2) - m */
- for(i=1;i<j;i++) r->d[i] = (~m->d[i])&BN_MASK2;
- r->top = j;
-#else
- if (!BN_to_montgomery(r,BN_value_one(),mont,ctx)) goto err;
+ j = m->top; /* borrow j */
+ if (m->d[j-1] & (((BN_ULONG)1)<<(BN_BITS2-1)))
+ {
+ if (bn_wexpand(r,j) == NULL) goto err;
+ /* 2^(top*BN_BITS2) - m */
+ r->d[0] = (0-m->d[0])&BN_MASK2;
+ for(i=1;i<j;i++) r->d[i] = (~m->d[i])&BN_MASK2;
+ r->top = j;
+ /* Upper words will be zero if the corresponding words of 'm'
+ * were 0xfff[...], so decrement r->top accordingly. */
+ bn_correct_top(r);
+ }
+ else
#endif
+ if (!BN_to_montgomery(r,BN_value_one(),mont,ctx)) goto err;
for (;;)
{
if (BN_is_bit_set(p,wstart) == 0)
start=0;
if (wstart < 0) break;
}
-#if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc))
+#if defined(SPARC_T4_MONT)
if (OPENSSL_sparcv9cap_P[0]&(SPARCV9_VIS3|SPARCV9_PREFER_FPU))
{
j = mont->N.top; /* borrow j */
return(ret);
}
-#if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc))
+#if defined(SPARC_T4_MONT)
static BN_ULONG bn_get_bits(const BIGNUM *a, int bitpos)
{
BN_ULONG ret=0;
int powerbufLen = 0;
unsigned char *powerbuf=NULL;
BIGNUM tmp, am;
-#if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc))
+#if defined(SPARC_T4_MONT)
unsigned int t4=0;
#endif
if (!BN_MONT_CTX_set(mont,m,ctx)) goto err;
}
+#ifdef RSAZ_ENABLED
+ /*
+ * If the size of the operands allow it, perform the optimized
+ * RSAZ exponentiation. For further information see
+ * crypto/bn/rsaz_exp.c and accompanying assembly modules.
+ */
+ if ((16 == a->top) && (16 == p->top) && (BN_num_bits(m) == 1024)
+ && rsaz_avx2_eligible())
+ {
+ if (NULL == bn_wexpand(rr, 16)) goto err;
+ RSAZ_1024_mod_exp_avx2(rr->d, a->d, p->d, m->d, mont->RR.d, mont->n0[0]);
+ rr->top = 16;
+ rr->neg = 0;
+ bn_correct_top(rr);
+ ret = 1;
+ goto err;
+ }
+ else if ((8 == a->top) && (8 == p->top) && (BN_num_bits(m) == 512))
+ {
+ if (NULL == bn_wexpand(rr,8)) goto err;
+ RSAZ_512_mod_exp(rr->d, a->d, p->d, m->d, mont->n0[0], mont->RR.d);
+ rr->top = 8;
+ rr->neg = 0;
+ bn_correct_top(rr);
+ ret = 1;
+ goto err;
+ }
+#endif
+
/* Get the window size to use with size of p. */
window = BN_window_bits_for_ctime_exponent_size(bits);
-#if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc))
+#if defined(SPARC_T4_MONT)
if (window>=5 && (top&15)==0 && top<=64 &&
(OPENSSL_sparcv9cap_P[1]&(CFR_MONTMUL|CFR_MONTSQR))==
(CFR_MONTMUL|CFR_MONTSQR) &&
else
#endif
#if defined(OPENSSL_BN_ASM_MONT5)
- if (window==6 && bits<=1024) window=5; /* ~5% improvement of 2048-bit RSA sign */
+ if (window>=5)
+ {
+ window=5; /* ~5% improvement for RSA2048 sign, and even for RSA4096 */
+ if ((top&7)==0) powerbufLen += 2*top*sizeof(m->d[0]);
+ }
#endif
(void)0;
* powers of am, am itself and tmp.
*/
numPowers = 1 << window;
- powerbufLen = sizeof(m->d[0])*(top*numPowers +
+ powerbufLen += sizeof(m->d[0])*(top*numPowers +
((2*top)>numPowers?(2*top):numPowers));
#ifdef alloca
if (powerbufLen < 3072)
/* prepare a^0 in Montgomery domain */
#if 1 /* by Shay Gueron's suggestion */
- tmp.d[0] = (0-m->d[0])&BN_MASK2; /* 2^(top*BN_BITS2) - m */
- for (i=1;i<top;i++)
- tmp.d[i] = (~m->d[i])&BN_MASK2;
- tmp.top = top;
-#else
- if (!BN_to_montgomery(&tmp,BN_value_one(),mont,ctx)) goto err;
+ if (m->d[top-1] & (((BN_ULONG)1)<<(BN_BITS2-1)))
+ {
+ /* 2^(top*BN_BITS2) - m */
+ tmp.d[0] = (0-m->d[0])&BN_MASK2;
+ for (i=1;i<top;i++) tmp.d[i] = (~m->d[i])&BN_MASK2;
+ tmp.top = top;
+ }
+ else
#endif
+ if (!BN_to_montgomery(&tmp,BN_value_one(),mont,ctx)) goto err;
/* prepare a^1 in Montgomery domain */
if (a->neg || BN_ucmp(a,m) >= 0)
}
else if (!BN_to_montgomery(&am,a,mont,ctx)) goto err;
-#if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc))
+#if defined(SPARC_T4_MONT)
if (t4)
{
typedef int (*bn_pwr5_mont_f)(BN_ULONG *tp,const BN_ULONG *np,
/* Dedicated window==4 case improves 512-bit RSA sign by ~15%, but as
* 512-bit RSA is hardly relevant, we omit it to spare size... */
- if (window==5)
+ if (window==5 && top>1)
{
void bn_mul_mont_gather5(BN_ULONG *rp,const BN_ULONG *ap,
const void *table,const BN_ULONG *np,
void *table,size_t power);
void bn_gather5(BN_ULONG *out,size_t num,
void *table,size_t power);
+ void bn_power5(BN_ULONG *rp,const BN_ULONG *ap,
+ const void *table,const BN_ULONG *np,
+ const BN_ULONG *n0,int num,int power);
+ int bn_get_bits5(const BN_ULONG *ap,int off);
+ int bn_from_montgomery(BN_ULONG *rp,const BN_ULONG *ap,
+ const BN_ULONG *not_used,const BN_ULONG *np,
+ const BN_ULONG *n0,int num);
- BN_ULONG *np=mont->N.d, *n0=mont->n0;
+ BN_ULONG *np=mont->N.d, *n0=mont->n0, *np2;
/* BN_to_montgomery can contaminate words above .top
* [in BN_DEBUG[_DEBUG] build]... */
for (i=am.top; i<top; i++) am.d[i]=0;
for (i=tmp.top; i<top; i++) tmp.d[i]=0;
+ if (top&7)
+ np2 = np;
+ else
+ for (np2=am.d+top,i=0; i<top; i++) np2[2*i]=np[i];
+
bn_scatter5(tmp.d,top,powerbuf,0);
bn_scatter5(am.d,am.top,powerbuf,1);
bn_mul_mont(tmp.d,am.d,am.d,np,n0,top);
for (i=3; i<32; i++)
{
/* Calculate a^i = a^(i-1) * a */
- bn_mul_mont_gather5(tmp.d,am.d,powerbuf,np,n0,top,i-1);
+ bn_mul_mont_gather5(tmp.d,am.d,powerbuf,np2,n0,top,i-1);
bn_scatter5(tmp.d,top,powerbuf,i);
}
#else
for (i=3; i<8; i+=2)
{
int j;
- bn_mul_mont_gather5(tmp.d,am.d,powerbuf,np,n0,top,i-1);
+ bn_mul_mont_gather5(tmp.d,am.d,powerbuf,np2,n0,top,i-1);
bn_scatter5(tmp.d,top,powerbuf,i);
for (j=2*i; j<32; j*=2)
{
}
for (; i<16; i+=2)
{
- bn_mul_mont_gather5(tmp.d,am.d,powerbuf,np,n0,top,i-1);
+ bn_mul_mont_gather5(tmp.d,am.d,powerbuf,np2,n0,top,i-1);
bn_scatter5(tmp.d,top,powerbuf,i);
bn_mul_mont(tmp.d,tmp.d,tmp.d,np,n0,top);
bn_scatter5(tmp.d,top,powerbuf,2*i);
}
for (; i<32; i+=2)
{
- bn_mul_mont_gather5(tmp.d,am.d,powerbuf,np,n0,top,i-1);
+ bn_mul_mont_gather5(tmp.d,am.d,powerbuf,np2,n0,top,i-1);
bn_scatter5(tmp.d,top,powerbuf,i);
}
#endif
/* Scan the exponent one window at a time starting from the most
* significant bits.
*/
- while (bits >= 0)
+ if (top&7)
+ while (bits >= 0)
{
for (wvalue=0, i=0; i<5; i++,bits--)
wvalue = (wvalue<<1)+BN_is_bit_set(p,bits);
bn_mul_mont(tmp.d,tmp.d,tmp.d,np,n0,top);
bn_mul_mont_gather5(tmp.d,tmp.d,powerbuf,np,n0,top,wvalue);
}
+ else
+ {
+ while (bits >= 0)
+ {
+ wvalue = bn_get_bits5(p->d,bits-4);
+ bits-=5;
+ bn_power5(tmp.d,tmp.d,powerbuf,np2,n0,top,wvalue);
+ }
+ }
+ ret=bn_from_montgomery(tmp.d,tmp.d,NULL,np2,n0,top);
tmp.top=top;
bn_correct_top(&tmp);
+ if (ret)
+ {
+ if (!BN_copy(rr,&tmp)) ret=0;
+ goto err; /* non-zero ret means it's not error */
+ }
}
else
#endif
}
/* Convert the final result from montgomery to standard format */
-#if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc))
+#if defined(SPARC_T4_MONT)
if (OPENSSL_sparcv9cap_P[0]&(SPARCV9_VIS3|SPARCV9_PREFER_FPU))
{
am.d[0] = 1; /* borrow am */
projects / openssl.git / blobdiff