X-Git-Url: https://wimlib.net/git/?a=blobdiff_plain;f=include%2Fwimlib%2Fdecompress_common.h;h=ef7ebe953960cd33c2e91efd341c37f4ba717441;hb=76b57cc8cf113194136a7e264644a37515650ccc;hp=40fa49ab4438796e2df043f95f1fecc9d491cc8f;hpb=67c5f2c5954d07c25f57b6d9aedb73fb3720ec0b;p=wimlib

diff --git a/include/wimlib/decompress_common.h b/include/wimlib/decompress_common.h
index 40fa49ab..ef7ebe95 100644
--- a/include/wimlib/decompress_common.h
+++ b/include/wimlib/decompress_common.h
@@ -9,7 +9,6 @@
 
 #include "wimlib/assert.h"
 #include "wimlib/compiler.h"
-#include "wimlib/error.h"
 #include "wimlib/endianness.h"
 #include "wimlib/types.h"
 
@@ -165,7 +164,7 @@ bitstream_read_byte(struct input_bitstream *istream)
  * lzms-decompress.c.  */
 static inline u16
 read_huffsym(struct input_bitstream *istream, const u16 decode_table[],
-	     unsigned num_syms, unsigned table_bits, unsigned max_codeword_len)
+	     unsigned table_bits, unsigned max_codeword_len)
 {
 	u16 entry;
 	u16 key_bits;
@@ -200,4 +199,56 @@ make_huffman_decode_table(u16 decode_table[], unsigned num_syms,
 			  unsigned num_bits, const u8 lens[],
 			  unsigned max_codeword_len);
 
-#endif /* _WIMLIB_DECOMPRESS_H */
+
+/*
+ * Copy a LZ77 match at (dst - offset) to dst.
+ *
+ * The length and offset must be already validated --- that is, (dst - offset)
+ * can't underrun the output buffer, and (dst + length) can't overrun the output
+ * buffer.  Also, the length cannot be 0.
+ *
+ * @winend points to the byte past the end of the output buffer.
+ * This function won't write any data beyond this position.
+ */
+static inline void
+lz_copy(u8 *dst, unsigned length, unsigned offset, const u8 *winend)
+{
+	const u8 *src = dst - offset;
+#if defined(__x86_64__) || defined(__i386__)
+	/* Copy one 'unsigned long' at a time.  On i386 and x86_64 this is
+	 * faster than copying one byte at a time, unless the data is
+	 * near-random and all the matches have very short lengths.  Note that
+	 * since this requires unaligned memory accesses, it won't necessarily
+	 * be faster on every architecture.
+	 *
+	 * Also note that we might copy more than the length of the match.  For
+	 * example, if an 'unsigned long' is 8 bytes and the match is of length
+	 * 5, then we'll simply copy 8 bytes.  This is okay as long as we don't
+	 * write beyond the end of the output buffer, hence the check for
+	 * (winend - (dst + length) >= sizeof(unsigned long) - 1).  */
+	if (offset >= sizeof(unsigned long) &&
+			winend - (dst + length) >= sizeof(unsigned long) - 1)
+	{
+		/* Access memory through a packed struct.  This tricks the
+		 * compiler into allowing unaligned memory accesses.  */
+		struct ulong_wrapper {
+			unsigned long v;
+		} _packed_attribute;
+
+		const u8 *end = dst + length;
+		do {
+			unsigned long v = ((struct ulong_wrapper *)src)->v;
+			((struct ulong_wrapper *)dst)->v = v;
+			dst += sizeof(unsigned long);
+			src += sizeof(unsigned long);
+		} while (dst < end);
+
+		return;
+	}
+#endif
+	do {
+		*dst++ = *src++;
+	} while (--length);
+}
+
+#endif /* _WIMLIB_DECOMPRESS_COMMON_H */