From f9215b53fba24648f5647eabd3cbfb40d93e6cde Mon Sep 17 00:00:00 2001
From: Moinak Ghosh <moinakg@gmail.com>
Date: Fri, 3 Aug 2012 23:19:38 +0530
Subject: [PATCH] Fix buffer size calculation when decompressing LZ4, Zlib and
 Bzip2 compressed chunks. Slight SSE optimization in LZ4HC.

---
 bzip2_compress.c |  2 +-
 lz4/lz4hc.c      | 25 +++++++++++++++++++++++++
 lz4_compress.c   |  6 ++----
 main.c           |  2 +-
 zlib_compress.c  |  2 +-
 5 files changed, 30 insertions(+), 7 deletions(-)

diff --git a/bzip2_compress.c b/bzip2_compress.c
index b5bf39a..fbb8f54 100644
--- a/bzip2_compress.c
+++ b/bzip2_compress.c
@@ -206,7 +206,7 @@ bzip2_decompress(void *src, size_t srclen, void *dst, size_t *dstlen,
 	}
 
 	/* normal termination */
-	*dstlen = _dstlen;
+	*dstlen = *dstlen - _dstlen;
 	BZ2_bzDecompressEnd(&bzs);
 	return (0);
 }
diff --git a/lz4/lz4hc.c b/lz4/lz4hc.c
index 74499ac..fbb046c 100644
--- a/lz4/lz4hc.c
+++ b/lz4/lz4hc.c
@@ -102,6 +102,9 @@
 //**************************************
 #include <stdlib.h>   // calloc, free
 #include <string.h>   // memset, memcpy
+#ifdef __USE_SSE_INTRIN__
+#include <emmintrin.h>
+#endif
 #include "lz4hc.h"
 
 #define ALLOCATOR(s) calloc(1,s)
@@ -358,6 +361,17 @@ inline static int LZ4HC_InsertAndFindBestMatch (LZ4HC_Data_Structure* hc4, const
 			const BYTE* reft = ref+MINMATCH;
 			const BYTE* ipt = ip+MINMATCH;
 
+#ifdef __USE_SSE_INTRIN__
+			while (ipt<matchlimit-15) {
+				int mask;
+				__m128i span1 = _mm_loadu_si128((__m128i *)(reft));
+				__m128i span2 = _mm_loadu_si128((__m128i *)(ipt));
+				mask = _mm_movemask_epi8(_mm_cmpeq_epi8(span1, span2)) ^ 0xffff;
+				if (!mask) { ipt+=16; reft+=16; continue; }
+				ipt += __builtin_ctz(mask);
+				goto _endCount;
+			}
+#endif
 			while (ipt<matchlimit-(STEPSIZE-1))
 			{
 				UARCH diff = AARCH(reft) ^ AARCH(ipt);
@@ -402,6 +416,17 @@ inline static int LZ4HC_InsertAndGetWiderMatch (LZ4HC_Data_Structure* hc4, const
 			const BYTE* ipt = ip+MINMATCH;
 			const BYTE* startt = ip;
 
+#ifdef __USE_SSE_INTRIN__
+			while (ipt<matchlimit-15) {
+				int mask;
+				__m128i span1 = _mm_loadu_si128((__m128i *)(reft));
+				__m128i span2 = _mm_loadu_si128((__m128i *)(ipt));
+				mask = _mm_movemask_epi8(_mm_cmpeq_epi8(span1, span2)) ^ 0xffff;
+				if (!mask) { ipt+=16; reft+=16; continue; }
+				ipt += __builtin_ctz(mask);
+				goto _endCount;
+			}
+#endif
 			while (ipt<matchlimit-(STEPSIZE-1))
 			{
 				UARCH diff = AARCH(reft) ^ AARCH(ipt);
diff --git a/lz4_compress.c b/lz4_compress.c
index 6405f81..60d5cd2 100644
--- a/lz4_compress.c
+++ b/lz4_compress.c
@@ -85,7 +85,6 @@ lz4_compress(void *src, size_t srclen, void *dst, size_t *dstlen,
 
 	if (lzdat->level == 1) {
 		rv = LZ4_compress(src, dst, _srclen);
-
 	} else if (lzdat->level == 2) {
 		rv = LZ4_compress(src, dst, _srclen);
 		if (rv == 0) {
@@ -127,15 +126,14 @@ lz4_decompress(void *src, size_t srclen, void *dst, size_t *dstlen,
 
 		sz1 = ntohl(*((int *)src));
 		rv = LZ4_uncompress(src + sizeof (int), dst, sz1);
-		if (rv == 0) {
+		if (rv != sz1) {
 			return (-1);
 		}
 		memcpy(src, dst, sz1);
 		rv = LZ4_uncompress(src, dst, _dstlen);
 	}
-	if (rv == 0) {
+	if (rv != srclen) {
 		return (-1);
 	}
-	*dstlen = rv;
 	return (0);
 }
diff --git a/main.c b/main.c
index 6974740..259ab01 100644
--- a/main.c
+++ b/main.c
@@ -149,7 +149,7 @@ show_compression_stats(uint64_t chunksize)
 /*
  * This routine is called in multiple threads. Calls the decompression handler
  * as encoded in the file header. For adaptive mode the handler adapt_decompress()
- * in turns looks at the chunk header and call the actualy decompression
+ * in turns looks at the chunk header and calls the actual decompression
  * routine.
  */
 static void *
diff --git a/zlib_compress.c b/zlib_compress.c
index 382b612..147edeb 100644
--- a/zlib_compress.c
+++ b/zlib_compress.c
@@ -225,7 +225,7 @@ zlib_decompress(void *src, size_t srclen, void *dst, size_t *dstlen,
 		}
 	}
 
-	*dstlen = _dstlen;
+	*dstlen = *dstlen - _dstlen;
 	inflateEnd(&zs);
 	return (0);
 }