dedup

commit eaed3251aefdb0cac2fb61970ccf2f8f52a097a2
parent 99e39cb4a6a9a5b21bae79d2e8c304019565ad93
Author: sin <sin@2f30.org>
Date:   Sat,  6 Apr 2019 09:33:34 +0100

Remove match_pattern and inline the code in the caller

Diffstat:
M
chunker.c
 | 
20
+++++++-------------

1 file changed, 7 insertions(+), 13 deletions(-)
diff --git a/chunker.c b/chunker.c
@@ -83,16 +83,6 @@ buzh_update(uint32_t sum, uint8_t out, uint8_t in, size_t size)
 	return ROTL(sum, 1) ^ ROTL(buz[out], size % 32) ^ buz[in];
 }
 
-static inline int
-match_pattern(struct chunker *chunker, size_t chunk_size, uint32_t sum)
-{
-	if (chunk_size >= chunker->max_size)
-		return 1;
-	if (chunk_size < chunker->min_size)
-		return 0;
-	return (sum & chunker->mask) == 0;
-}
-
 static size_t
 get_chunk_size(struct chunker *chunker)
 {
@@ -107,8 +97,8 @@ get_chunk_size(struct chunker *chunker)
 
 	/*
 	 * To achieve better deduplication, we chunk blocks based on a
-	 * recurring pattern occuring on the data stream. A fixed window
-	 * of WINSIZE bytes is slid over the data, and a rolling hash is
+	 * recurring pattern occuring on the data stream. We slide a fixed
+	 * window of WINSIZE bytes over the data, and a rolling hash is
 	 * computed for this window.
 	 * When the rolling hash matches a given pattern the block is chunked
 	 * at the end of that window.
@@ -125,7 +115,11 @@ get_chunk_size(struct chunker *chunker)
 			sum = buzh_update(sum, out, in, win_size);
 		}
 
-		if (match_pattern(chunker, chunk_size, sum))
+		if (chunk_size < chunker->min_size)
+			continue;
+
+		if (chunk_size >= chunker->max_size ||
+		    (sum & chunker->mask) == 0)
 			return chunk_size;
 	}
 	return max_chunk_size;