diff --git a/ed25519.dll b/ed25519.dll
index 2dd29d5..33f9938 100644
Binary files a/ed25519.dll and b/ed25519.dll differ
diff --git a/readme.md b/readme.md
index 3ba1314..7a7a602 100644
--- a/readme.md
+++ b/readme.md
@@ -20,14 +20,15 @@ Performance
 On a Windows machine with an Intel Pentium B970 @ 2.3GHz I got the following
 speeds (running on only one a single core):
 
-    Seed + key generation: 489us
-    Message signing (short message): 251us
-    Message verifying (short message): 772us
-    Scalar addition: 358us
-    Key exchange: 724us 
+    Seed + key generation: 144us
+    Message signing (short message): 87us
+    Message verifying (short message): 228us
+    Scalar addition: 100us
+    Key exchange: 220us 
 
 The speeds on other machines may vary. Sign/verify times will be higher with
-longer messages.
+longer messages. The implementation significantly benefits from 64 bit
+architectures, if possible compile as 64 bit.
 
 
 Usage
diff --git a/src/add_scalar.c b/src/add_scalar.c
index 74dc801..91b64d3 100644
--- a/src/add_scalar.c
+++ b/src/add_scalar.c
@@ -10,7 +10,7 @@ void ed25519_add_scalar(unsigned char *public_key, unsigned char *private_key, c
     unsigned char n[32]; 
     ge_p3 nB;
     ge_p1p1 A_p1p1;
-    ge_p3 A;
+    ge_p2 A;
     ge_p3 public_key_unpacked;
     ge_cached T;
 
@@ -39,9 +39,9 @@ void ed25519_add_scalar(unsigned char *public_key, unsigned char *private_key, c
 
         /* A = n*B + T */
         ge_add(&A_p1p1, &nB, &T);
-        ge_p1p1_to_p3(&A, &A_p1p1);
+        ge_p1p1_to_p2(&A, &A_p1p1);
         
         /* pack public key */
-        ge_p3_tobytes(public_key, &A);
+        ge_tobytes(public_key, &A);
     }
 }
diff --git a/test.c b/test.c
index ec1eeef..264832f 100644
--- a/test.c
+++ b/test.c
@@ -116,7 +116,7 @@ int main(int argc, char *argv[]) {
     }
     end = clock();
 
-    printf("%fus per signature\n", ((double) ((end - start) * 1000)) / CLOCKS_PER_SEC / i * 1000);
+    printf("%fus per keypair\n", ((double) ((end - start) * 1000)) / CLOCKS_PER_SEC / i * 1000);
 
 	printf("testing key exchange performance: ");
     start = clock();
@@ -125,7 +125,7 @@ int main(int argc, char *argv[]) {
     }
     end = clock();
 
-    printf("%fus per signature\n", ((double) ((end - start) * 1000)) / CLOCKS_PER_SEC / i * 1000);
+    printf("%fus per shared secret\n", ((double) ((end - start) * 1000)) / CLOCKS_PER_SEC / i * 1000);
 
     return 0;
 }