diff options
author | Andy Polyakov <appro@openssl.org> | 1999-07-31 23:27:41 +0000 |
---|---|---|
committer | Andy Polyakov <appro@openssl.org> | 1999-07-31 23:27:41 +0000 |
commit | 4c22909e31bcb284973243c3b1622503455c667c (patch) | |
tree | 15e9aa575cc3b6d2423121eddcb81b19339b4dd4 /crypto/bn/bn_div.c | |
parent | 8d85b33eb53ca9ae739aaa6a6f30cb65a1ac6135 (diff) | |
download | openssl-4c22909e31bcb284973243c3b1622503455c667c.tar.gz |
Extra i386+gcc bn_div.c tune-up featuring inline division and saving
the remainder left in %edx. Here is the resulting performance improvement
matrix (improvement as a result of this *and* previous tune-up committed
two days ago). The results were obtained by profiling the "div" part of
the crypto/bn/bnspeed.c.
CPU BN_div bn_div_words overall comment
------------------------------------------------------------------------
PII +16% accumulated by +2-3% PII multiplies damn fast! Taking
inlining multiplication out of the loop
didn't make too much difference.
Eliminating of the multiplication
involved in remainder calculation
is the major factor.
Pentium +45% accumulated by +7-9% mull isn't that fast and replacing
inlining multiplications with additions in
the loop has more visible effect:-)
MIPS +75% +12% +20-25% In addition to the taking mults
R10000 out of the loop (giving 12% in the
asm/mips3.s) three mults were
eliminated in BN_div.
Alpha +30% +50% +10-15% Same as above. But remember that
EV4 bn_div_words is a C implementation.
It takes 4 Alpha mults in C to do
the same thing as 1 MIPS mult in
assembler does. So the effect (50%)
is more impressive. But not the
overall one... Well, if Alpha
bn_mul_add would be implemented
in assembler overall improvement
would be closer to MIPS...
Diffstat (limited to 'crypto/bn/bn_div.c')
-rw-r--r-- | crypto/bn/bn_div.c | 41 |
1 files changed, 35 insertions, 6 deletions
diff --git a/crypto/bn/bn_div.c b/crypto/bn/bn_div.c index 03b9152241..e4253f6095 100644 --- a/crypto/bn/bn_div.c +++ b/crypto/bn/bn_div.c @@ -204,15 +204,41 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor, #ifdef BN_DIV3W q=bn_div_3_words(wnump,d0,d1); #else - BN_ULONG n0,n1,rem; + +#if !defined(NO_ASM) +# if defined(__GNUC__) && __GNUC__>=2 +# if defined(__i386) + /* + * There were two reasons for implementing this template: + * - GNU C generates a call to a function (__udivdi3 to be exact) + * in reply to ((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0 (I fail to + * understand why...); + * - divl doesn't only calculate quotient, but also leaves + * remainder in %edx which we can definitely use here:-) + * + * <appro@fy.chalmers.se> + */ +# define bn_div_words(n0,n1,d0) \ + ({ asm volatile ( \ + "divl %4" \ + : "=a"(q), "=d"(rem) \ + : "a"(n1), "d"(n0), "g"(d0) \ + : "cc"); \ + q; \ + }) +# define REMINDER_IS_ALREADY_CALCULATED +# endif /* __<cpu> */ +# endif /* __GNUC__ */ +#endif /* NO_ASM */ + BN_ULONG n0,n1,rem=0; n0=wnump[0]; n1=wnump[-1]; if (n0 == d0) q=BN_MASK2; else -#if defined(BN_LLONG) && defined(BN_DIV2W) - q=((((BN_ULLONG)n0)<<BN_BITS2)|n1)/((BN_ULLONG)d0); +#if defined(BN_LLONG) && defined(BN_DIV2W) && !defined(bn_div_words) + q=((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0; #else q=bn_div_words(n0,n1,d0); #endif @@ -220,13 +246,15 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor, #ifdef BN_LLONG BN_ULLONG t2; +#ifndef REMINDER_IS_ALREADY_CALCULATED /* * rem doesn't have to be BN_ULLONG. The least we * know it's less that d0, isn't it? */ rem=(n1-q*d0)&BN_MASK2; - +#endif t2=(BN_ULLONG)d1*q; + for (;;) { if (t2 <= ((((BN_ULLONG)rem)<<BN_BITS2)|wnump[-2])) @@ -239,12 +267,13 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor, #else BN_ULONG t2l,t2h,ql,qh; +#ifndef REMINDER_IS_ALREADY_CALCULATED /* * It's more than enough with the only multiplication. * See the comment above in BN_LLONG section... */ rem=(n1-q*d0)&BN_MASK2; - +#endif t2l=LBITS(d1); t2h=HBITS(d1); ql =LBITS(q); qh =HBITS(q); mul64(t2l,t2h,ql,qh); /* t2=(BN_ULLONG)d1*q; */ @@ -261,7 +290,7 @@ int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor, } #endif } -#endif /* BN_DIV3W */ +#endif /* !BN_DIV3W */ wnum.d--; wnum.top++; l0=bn_mul_words(tmp->d,sdiv->d,div_n,q); tmp->d[div_n]=l0; |