Optimizations for digest final functions

* cipher/md4.c (md4_final): Avoid byte-by-byte buffer setting when
padding; Merge extra and last block processing.
* cipher/md5.c (md5_final): Ditto.
* cipher/rmd160.c (rmd160_final): Ditto.
* cipher/sha1.c (sha1_final): Ditto.
* cipher/sha256.c (sha256_final): Ditto.
* cipher/sm3.c (sm3_final): Ditto.
* cipher/tiger.c (tiger_final): Ditto.
* cipher/sha512.c (sha512_final): Avoid byte-by-byte buffer setting
when padding.
* cipher/stribog.c (stribog_final): Ditto.
* cipher/whirlpool.c (whirlpool_final): Ditto.
--

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>

1 files changed, 17 insertions, 11 deletions

diff --git a/cipher/tiger.c b/cipher/tiger.c
index d24d1603..0319b711 100644
--- a/cipher/tiger.c
+++ b/cipher/tiger.c
@@ -760,22 +760,26 @@ tiger_final( void *context )
   if( hd->bctx.count < 56 )  /* enough room */
     {
       hd->bctx.buf[hd->bctx.count++] = pad;
-      while( hd->bctx.count < 56 )
-        hd->bctx.buf[hd->bctx.count++] = 0;  /* pad */
+      if (hd->bctx.count < 56)
+	memset (&hd->bctx.buf[hd->bctx.count], 0, 56 - hd->bctx.count);
+      hd->bctx.count = 56;
+      /* append the 64 bit count */
+      buf_put_le32(hd->bctx.buf + 56, lsb);
+      buf_put_le32(hd->bctx.buf + 60, msb);
+      burn = transform( hd, hd->bctx.buf, 1 );
     }
   else  /* need one extra block */
     {
       hd->bctx.buf[hd->bctx.count++] = pad; /* pad character */
-      while( hd->bctx.count < 64 )
-        hd->bctx.buf[hd->bctx.count++] = 0;
-      _gcry_md_block_write(hd, NULL, 0);  /* flush */;
-      memset(hd->bctx.buf, 0, 56 ); /* fill next block with zeroes */
+      /* fill pad and next block with zeroes */
+      memset (&hd->bctx.buf[hd->bctx.count], 0, 64 - hd->bctx.count + 56);
+      hd->bctx.count = 64 + 56;
+
+      /* append the 64 bit count */
+      buf_put_le32(hd->bctx.buf + 64 + 56, lsb);
+      buf_put_le32(hd->bctx.buf + 64 + 60, msb);
+      burn = transform( hd, hd->bctx.buf, 2 );
     }
-  /* append the 64 bit count */
-  buf_put_le32(hd->bctx.buf + 56, lsb);
-  buf_put_le32(hd->bctx.buf + 60, msb);
-  burn = transform( hd, hd->bctx.buf, 1 );
-  _gcry_burn_stack (burn);
 
   p = hd->bctx.buf;
 #define X(a) do { buf_put_be64(p, hd->a); p += 8; } while(0)
@@ -794,6 +798,8 @@ tiger_final( void *context )
     }
 #undef X
 #undef Y
+
+  _gcry_burn_stack (burn);
 }
 
 static byte *