3-4 times better RSA/DSA performance on WIN64A target. Well, on AMD64 CPU,

EMT64T will hardly exhibit better performance...

1 files changed, 28 insertions, 0 deletions

diff --git a/crypto/bn/bn_asm.c b/crypto/bn/bn_asm.c
index 19978085b2..99bc2de491 100644
--- a/crypto/bn/bn_asm.c
+++ b/crypto/bn/bn_asm.c
@@ -459,6 +459,34 @@ BN_ULONG bn_sub_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b, int n)
 #define sqr_add_c2(a,i,j,c0,c1,c2) \
 	mul_add_c2((a)[i],(a)[j],c0,c1,c2)
 
+#elif defined(BN_UMULT_LOHI)
+
+#define mul_add_c(a,b,c0,c1,c2)	{	\
+	BN_ULONG ta=(a),tb=(b);		\
+	BN_UMULT_LOHI(t1,t2,ta,tb);	\
+	c0 += t1; t2 += (c0<t1)?1:0;	\
+	c1 += t2; c2 += (c1<t2)?1:0;	\
+	}
+
+#define mul_add_c2(a,b,c0,c1,c2) {	\
+	BN_ULONG ta=(a),tb=(b),t0;	\
+	BN_UMULT_LOHI(t0,t1,ta,tb);	\
+	t2 = t1+t1; c2 += (t2<t1)?1:0;	\
+	t1 = t0+t0; t2 += (t1<t0)?1:0;	\
+	c0 += t1; t2 += (c0<t1)?1:0;	\
+	c1 += t2; c2 += (c1<t2)?1:0;	\
+	}
+
+#define sqr_add_c(a,i,c0,c1,c2)	{	\
+	BN_ULONG ta=(a)[i];		\
+	BN_UMULT_LOHI(t1,t2,ta,ta);	\
+	c0 += t1; t2 += (c0<t1)?1:0;	\
+	c1 += t2; c2 += (c1<t2)?1:0;	\
+	}
+
+#define sqr_add_c2(a,i,j,c0,c1,c2)	\
+	mul_add_c2((a)[i],(a)[j],c0,c1,c2)
+
 #elif defined(BN_UMULT_HIGH)
 
 #define mul_add_c(a,b,c0,c1,c2)	{	\