SHA-1/SSSE3: Improve performance on large buffers

* cipher/sha1-ssse3-amd64.S (RNBLKS): New.
(_gcry_sha1_transform_amd64_ssse3): Handle multiple input blocks, with
software pipelining of next data block processing.
* cipher/sha1.c [USE_SSSE3] (_gcry_sha1_transform_amd64_ssse3): Add
'nblks'.
(transform) [USE_SSSE3]: Pass nblks to assembly function.
--

Patch gives small improvement for large buffer processing, on Intel i5-4570
speed goes from 4.80 c/B to 4.61 c/B.

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

1 files changed, 4 insertions, 11 deletions

diff --git a/cipher/sha1.c b/cipher/sha1.c
index 53f75381..8040e766 100644
--- a/cipher/sha1.c
+++ b/cipher/sha1.c
@@ -234,7 +234,8 @@ transform_blk (void *ctx, const unsigned char *data)
 
 #ifdef USE_SSSE3
 unsigned int
-_gcry_sha1_transform_amd64_ssse3 (void *state, const unsigned char *data);
+_gcry_sha1_transform_amd64_ssse3 (void *state, const unsigned char *data,
+                                  size_t nblks);
 #endif
 
 
@@ -246,16 +247,8 @@ transform (void *ctx, const unsigned char *data, size_t nblks)
 
 #ifdef USE_SSSE3
   if (hd->use_ssse3)
-    {
-      do
-        {
-          burn = _gcry_sha1_transform_amd64_ssse3 (&hd->h0, data);
-          data += 64;
-        }
-      while (--nblks);
-
-      return burn + 4 * sizeof(void*);
-    }
+    return _gcry_sha1_transform_amd64_ssse3 (&hd->h0, data, nblks)
+           + 4 * sizeof(void*);
 #endif
 
   do