3 files changed, 13 insertions, 4 deletions
diff --git a/Makefile b/Makefile
index 76f3163..62fe2a6 100644
--- a/Makefile
+++ b/Makefile
@@ -14,7 +14,7 @@ lz4demo64: lz4.c lz4.h bench.c lz4demo.c
 	gcc      -O3 -I. -std=c99 -Wall -W -Wundef -Wno-implicit-function-declaration lz4.c bench.c lz4demo.c -o $(OUTPUT64)
 
 lz4demo32: lz4.c lz4.h bench.c lz4demo.c
-	gcc -m32 -O3 -I. -std=c99 -Wall -W -Wundef -Wno-implicit-function-declaration lz4.c bench.c lz4demo.c -o $(OUTPUT32)
+	gcc -m32 -Os -march=native -I. -std=c99 -Wall -W -Wundef -Wno-implicit-function-declaration lz4.c bench.c lz4demo.c -o $(OUTPUT32)
 
 clean:
 	rm -f core *.o $(OUTPUT32) $(OUTPUT64)
diff --git a/lz4.c b/lz4.c
index df62be6..952d58d 100644
--- a/lz4.c
+++ b/lz4.c
@@ -215,10 +215,10 @@ typedef struct _U64_S { U64 v; } U64_S;
 
 #if (defined(LZ4_BIG_ENDIAN) && !defined(BIG_ENDIAN_NATIVE_BUT_INCOMPATIBLE))
 #define LZ4_READ_LITTLEENDIAN_16(d,s,p) { U16 v = A16(p); v = bswap16(v); d = (s) - v; }
-#define LZ4_WRITE_LITTLEENDIAN_16(p,i) { U16 v = (U16)(i); v = bswap16(v); A16(p) = v; p+=2; }
+#define LZ4_WRITE_LITTLEENDIAN_16(p,i)  { U16 v = (U16)(i); v = bswap16(v); A16(p) = v; p+=2; }
 #else		// Little Endian
 #define LZ4_READ_LITTLEENDIAN_16(d,s,p) { d = (s) - A16(p); }
-#define LZ4_WRITE_LITTLEENDIAN_16(p,v) { A16(p) = v; p+=2; }
+#define LZ4_WRITE_LITTLEENDIAN_16(p,v)  { A16(p) = v; p+=2; }
 #endif
 
 
diff --git a/lz4demo.c b/lz4demo.c
index 0d7b020..61bf75c 100644
--- a/lz4demo.c
+++ b/lz4demo.c
@@ -45,10 +45,19 @@
 //**************************************
 // Compiler functions
 //**************************************
+#define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+
 #if defined(_MSC_VER)    // Visual Studio 
 #define swap32 _byteswap_ulong
-#else    // GCC assumed
+#elif GCC_VERSION >= 402
 #define swap32 __builtin_bswap32
+#else
+static inline unsigned int swap32(unsigned int x) {
+			return	((x << 24) & 0xff000000 ) |
+				((x <<  8) & 0x00ff0000 ) |
+				((x >>  8) & 0x0000ff00 ) |
+				((x >> 24) & 0x000000ff );
+		 }
 #endif