tuklib_integer: Add 64-bit endianness-converting reads and writes.

Also update the comment in liblzma's memcmplen.h.

Thanks to Michał Górny for the original patch for the reads.

m4/tuklib_integer.m4

@@ -64,8 +64,8 @@ main(void)
 AC_MSG_CHECKING([if unaligned memory access should be used])
 AC_ARG_ENABLE([unaligned-access], AS_HELP_STRING([--enable-unaligned-access],
 		[Enable if the system supports *fast* unaligned memory access
-		with 16-bit and 32-bit integers. By default, this is enabled
+		with 16-bit, 32-bit, and 64-bit integers. By default,
-		only on x86, x86_64, big endian PowerPC,
+		this is enabled only on x86, x86_64, big endian PowerPC,
 		and some ARM systems.]),
 	[], [enable_unaligned_access=auto])
 if test "x$enable_unaligned_access" = xauto ; then
@@ -93,8 +93,8 @@ int main(void) { return 0; }
 fi
 if test "x$enable_unaligned_access" = xyes ; then
 	AC_DEFINE([TUKLIB_FAST_UNALIGNED_ACCESS], [1], [Define to 1 if
-		the system supports fast unaligned access to 16-bit and
+		the system supports fast unaligned access to 16-bit,
-		32-bit integers.])
+		32-bit, and 64-bit integers.])
 	AC_MSG_RESULT([yes])
 else
 	AC_MSG_RESULT([no])

src/common/tuklib_integer.h

@@ -17,8 +17,8 @@
 ///   - Byte swapping: bswapXX(num)
 ///   - Byte order conversions to/from native (byteswaps if Y isn't
 ///     the native endianness): convXXYe(num)
-///   - Unaligned reads (16/32-bit only): readXXYe(ptr)
+///   - Unaligned reads: readXXYe(ptr)
-///   - Unaligned writes (16/32-bit only): writeXXYe(ptr, num)
+///   - Unaligned writes: writeXXYe(ptr, num)
 ///   - Aligned reads: aligned_readXXYe(ptr)
 ///   - Aligned writes: aligned_writeXXYe(ptr, num)
 ///
@@ -343,6 +343,46 @@ read32le(const uint8_t *buf)
 }
 
 
+static inline uint64_t
+read64be(const uint8_t *buf)
+{
+#if defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
+	uint64_t num = read64ne(buf);
+	return conv64be(num);
+#else
+	uint64_t num = (uint64_t)buf[0] << 56;
+	num |= (uint64_t)buf[1] << 48;
+	num |= (uint64_t)buf[2] << 40;
+	num |= (uint64_t)buf[3] << 32;
+	num |= (uint64_t)buf[4] << 24;
+	num |= (uint64_t)buf[5] << 16;
+	num |= (uint64_t)buf[6] << 8;
+	num |= (uint64_t)buf[7];
+	return num;
+#endif
+}
+
+
+static inline uint64_t
+read64le(const uint8_t *buf)
+{
+#if !defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
+	uint64_t num = read64ne(buf);
+	return conv64le(num);
+#else
+	uint64_t num = (uint64_t)buf[0];
+	num |= (uint64_t)buf[1] << 8;
+	num |= (uint64_t)buf[2] << 16;
+	num |= (uint64_t)buf[3] << 24;
+	num |= (uint64_t)buf[4] << 32;
+	num |= (uint64_t)buf[5] << 40;
+	num |= (uint64_t)buf[6] << 48;
+	num |= (uint64_t)buf[7] << 56;
+	return num;
+#endif
+}
+
+
 // NOTE: Possible byte swapping must be done in a macro to allow the compiler
 // to optimize byte swapping of constants when using glibc's or *BSD's
 // byte swapping macros. The actual write is done in an inline function
@@ -350,11 +390,13 @@ read32le(const uint8_t *buf)
 #if defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
 #	define write16be(buf, num) write16ne(buf, conv16be(num))
 #	define write32be(buf, num) write32ne(buf, conv32be(num))
+#	define write64be(buf, num) write64ne(buf, conv64be(num))
 #endif
 
 #if !defined(WORDS_BIGENDIAN) || defined(TUKLIB_FAST_UNALIGNED_ACCESS)
 #	define write16le(buf, num) write16ne(buf, conv16le(num))
 #	define write32le(buf, num) write32ne(buf, conv32le(num))
+#	define write64le(buf, num) write64ne(buf, conv64le(num))
 #endif
 
 

src/liblzma/common/memcmplen.h

@@ -51,10 +51,6 @@ lzma_memcmplen(const uint8_t *buf1, const uint8_t *buf2,
 			|| (defined(__INTEL_COMPILER) && defined(__x86_64__)) \
 			|| (defined(__INTEL_COMPILER) && defined(_M_X64)) \
 			|| (defined(_MSC_VER) && defined(_M_X64)))
-	// NOTE: This will use 64-bit unaligned access which
-	// TUKLIB_FAST_UNALIGNED_ACCESS wasn't meant to permit, but
-	// it's convenient here at least as long as it's x86-64 only.
-	//
 	// I keep this x86-64 only for now since that's where I know this
 	// to be a good method. This may be fine on other 64-bit CPUs too.
 	// On big endian one should use xor instead of subtraction and switch
@@ -84,8 +80,9 @@ lzma_memcmplen(const uint8_t *buf1, const uint8_t *buf2,
 			|| (defined(__INTEL_COMPILER) && defined(__SSE2__)) \
 			|| (defined(_MSC_VER) && defined(_M_IX86_FP) \
 				&& _M_IX86_FP >= 2))
-	// NOTE: Like above, this will use 128-bit unaligned access which
+	// NOTE: This will use 128-bit unaligned access which
-	// TUKLIB_FAST_UNALIGNED_ACCESS wasn't meant to permit.
+	// TUKLIB_FAST_UNALIGNED_ACCESS wasn't meant to permit,
+	// but it's convenient here since this is x86-only.
 	//
 	// SSE2 version for 32-bit and 64-bit x86. On x86-64 the above
 	// version is sometimes significantly faster and sometimes