UltraSPARC performance "tune-up."
authorAndy Polyakov <appro@openssl.org>
Fri, 27 Dec 2002 14:51:49 +0000 (14:51 +0000)
committerAndy Polyakov <appro@openssl.org>
Fri, 27 Dec 2002 14:51:49 +0000 (14:51 +0000)
PROBLEMS
crypto/md5/md5_locl.h

index 56bc73816b352a74849a9d1e985cae8968a2f55f..4bf31303be998a94a9b4044801ed443cdc725208 100644 (file)
--- a/PROBLEMS
+++ b/PROBLEMS
@@ -62,3 +62,11 @@ What happens is that gcc might optimize a little too agressively, and
 you end up with an extra incrementation when *header != '4'.
 
 We recommend that you upgrade gcc to as high a 3.x version as you can.
+
+* solaris64-sparcv9-cc SHA-1 performance with WorkShop 6 compiler.
+
+As subject suggests SHA-1 might perform poorly (4 times slower)
+if compiled with WorkShop 6 compiler and -xarch=v9. The cause for
+this seems to be the fact that compiler emits multiplication to
+perform shift operations:-( To work the problem around configure
+with './Configure solaris64-sparcv9-cc -DMD32_REG_T=int'.
index 34c5257306dbc35edfa30fac2790c94ad8fc1c04..85ae99210ceb6cf05797b97f93c89b12af1238f0 100644 (file)
@@ -68,7 +68,7 @@
 #ifdef MD5_ASM
 # if defined(__i386) || defined(__i386__) || defined(_M_IX86) || defined(__INTEL__)
 #  define md5_block_host_order md5_block_asm_host_order
-# elif defined(__sparc) && defined(OPENSSL_SYS_ULTRASPARC)
+# elif defined(__sparc) && defined(OPENSSL_SYSNAME_ULTRASPARC)
    void md5_block_asm_data_order_aligned (MD5_CTX *c, const MD5_LONG *p,int num);
 #  define HASH_BLOCK_DATA_ORDER_ALIGNED md5_block_asm_data_order_aligned
 # endif