liblzma: Rename crc32_aarch64.h to crc32_arm64.h.

Even though the proper name for the architecture is aarch64, this
project uses ARM64 throughout. So the rename is for consistency.

Additionally, crc32_arm64.h was slightly refactored for the following
changes:

   * Added MSVC, FreeBSD, and macOS support in
     is_arch_extension_supported().

   * crc32_arch_optimized() now checks the size when aligning the
     buffer.

   * crc32_arch_optimized() loop conditions were slightly modified to
     avoid both decrementing the size and incrementing the buffer
     pointer.

   * Use the intrinsic wrappers defined in <arm_acle.h> because GCC and
     Clang name them differently.

   * Minor spacing and comment changes.

6 files changed, 122 insertions, 116 deletions

diff --git a/CMakeLists.txt b/CMakeLists.txt
index b21090f7..408b1605 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -230,7 +230,7 @@ add_library(liblzma
     src/liblzma/check/check.h
     src/liblzma/check/crc_common.h
     src/liblzma/check/crc_x86_clmul.h
-    src/liblzma/check/crc32_aarch64.h
+    src/liblzma/check/crc32_arm64.h
     src/liblzma/common/block_util.c
     src/liblzma/common/common.c
     src/liblzma/common/common.h
diff --git a/src/liblzma/check/Makefile.inc b/src/liblzma/check/Makefile.inc
index e7f87c85..da7ee0db 100644
--- a/src/liblzma/check/Makefile.inc
+++ b/src/liblzma/check/Makefile.inc
@@ -16,7 +16,7 @@ liblzma_la_SOURCES += \
 	check/check.h \
 	check/crc_common.h \
 	check/crc_x86_clmul.h \
-	check/crc32_aarch64.h
+	check/crc32_arm64.h
 
 if COND_SMALL
 liblzma_la_SOURCES += check/crc32_small.c
diff --git a/src/liblzma/check/crc32_aarch64.h b/src/liblzma/check/crc32_aarch64.h
deleted file mode 100644
index 77b14af4..00000000
--- a/src/liblzma/check/crc32_aarch64.h
+++ /dev/null
@@ -1,109 +0,0 @@
-///////////////////////////////////////////////////////////////////////////////
-//
-/// \file       crc32_aarch64.c
-/// \brief      CRC32 calculation with aarch64 optimization
-//
-//  Authors:    Chenxi Mao
-//
-//  This file has been put into the public domain.
-//  You can do whatever you want with this file.
-//
-///////////////////////////////////////////////////////////////////////////////
-#ifdef LZMA_CRC_CRC32_AARCH64_H
-#	error crc_arm64_clmul.h was included twice.
-#endif
-#define LZMA_CRC_CRC32_AARCH64_H
-#include <sys/auxv.h>
-// EDG-based compilers (Intel's classic compiler and compiler for E2K) can
-// define __GNUC__ but the attribute must not be used with them.
-// The new Clang-based ICX needs the attribute.
-//
-// NOTE: Build systems check for this too, keep them in sync with this.
-#if (defined(__GNUC__) || defined(__clang__)) && !defined(__EDG__)
-#	define crc_attr_target \
-        __attribute__((__target__("+crc")))
-#else
-#	define crc_attr_target
-#endif
-#ifdef BUILDING_CRC32_AARCH64
-crc_attr_target
-crc_attr_no_sanitize_address
-static uint32_t
-crc32_arch_optimized(const uint8_t *buf, size_t size, uint32_t crc)
-{
-	crc = ~crc;
-	while ((uintptr_t)(buf) & 7) {
-		crc = __builtin_aarch64_crc32b(crc, *buf);
-		buf++;
-		size--;
-	}
-	for (;size>=8;size-=8,buf+=8) {
-		crc = __builtin_aarch64_crc32x(crc, aligned_read64le(buf));
-	}
-	for (;size>0;size--,buf++)
-		crc = __builtin_aarch64_crc32b(crc, *buf);
-	return ~crc;
-}
-#endif
-#ifdef BUILDING_CRC64_AARCH64
-//FIXME: there is no crc64_arch_optimized implementation,
-// to make compiler happy, add crc64_generic here.
-#ifdef WORDS_BIGENDIAN
-#	define A1(x) ((x) >> 56)
-#else
-#	define A1 A
-#endif
-crc_attr_target
-crc_attr_no_sanitize_address
-static uint64_t
-crc64_arch_optimized(const uint8_t *buf, size_t size, uint64_t crc)
-{
-	crc = ~crc;
-
-#ifdef WORDS_BIGENDIAN
-	crc = bswap64(crc);
-#endif
-
-	if (size > 4) {
-		while ((uintptr_t)(buf) & 3) {
-			crc = lzma_crc64_table[0][*buf++ ^ A1(crc)] ^ S8(crc);
-			--size;
-		}
-
-		const uint8_t *const limit = buf + (size & ~(size_t)(3));
-		size &= (size_t)(3);
-
-		while (buf < limit) {
-#ifdef WORDS_BIGENDIAN
-			const uint32_t tmp = (uint32_t)(crc >> 32)
-					^ aligned_read32ne(buf);
-#else
-			const uint32_t tmp = (uint32_t)crc
-					^ aligned_read32ne(buf);
-#endif
-			buf += 4;
-
-			crc = lzma_crc64_table[3][A(tmp)]
-			    ^ lzma_crc64_table[2][B(tmp)]
-			    ^ S32(crc)
-			    ^ lzma_crc64_table[1][C(tmp)]
-			    ^ lzma_crc64_table[0][D(tmp)];
-		}
-	}
-
-	while (size-- != 0)
-		crc = lzma_crc64_table[0][*buf++ ^ A1(crc)] ^ S8(crc);
-
-#ifdef WORDS_BIGENDIAN
-	crc = bswap64(crc);
-#endif
-
-	return ~crc;
-}
-#endif
-static inline bool
-is_arch_extension_supported(void)
-{
-	return (getauxval(AT_HWCAP) & HWCAP_CRC32)!=0;
-}
-
diff --git a/src/liblzma/check/crc32_arm64.h b/src/liblzma/check/crc32_arm64.h
new file mode 100644
index 00000000..a1888ea7
--- /dev/null
+++ b/src/liblzma/check/crc32_arm64.h
@@ -0,0 +1,119 @@
+///////////////////////////////////////////////////////////////////////////////
+//
+/// \file       crc32_arm64.h
+/// \brief      CRC32 calculation with ARM64 optimization
+//
+//  Authors:    Chenxi Mao
+//              Jia Tan
+//
+//  This file has been put into the public domain.
+//  You can do whatever you want with this file.
+//
+///////////////////////////////////////////////////////////////////////////////
+
+
+#ifndef LZMA_CRC32_ARM64_H
+#define LZMA_CRC32_ARM64_H
+
+// MSVC always has the CRC intrinsics available when building for ARM64
+// there is no need to include any header files.
+#ifndef _MSC_VER
+#	include <arm_acle.h>
+#endif
+
+#if defined(CRC32_GENERIC) && defined(CRC32_ARCH_OPTIMIZED)
+#	if defined(HAVE_GETAUXVAL) || defined(HAVE_ELF_AUX_INFO)
+#		include <sys/auxv.h>
+#	elif defined(_WIN32)
+#		include <processthreadsapi.h>
+#	elif defined(__APPLE__) && defined(HAVE_SYSCTLBYNAME)
+#		include <sys/sysctl.h>
+#	endif
+#endif
+
+// Some EDG-based compilers support ARM64 and define __GNUC__
+// (such as Nvidia's nvcc), but do not support function attributes.
+//
+// NOTE: Build systems check for this too, keep them in sync with this.
+#if (defined(__GNUC__) || defined(__clang__)) && !defined(__EDG__)
+#	define crc_attr_target \
+        __attribute__((__target__("+crc")))
+#else
+#	define crc_attr_target
+#endif
+
+
+crc_attr_target
+static uint32_t
+crc32_arch_optimized(const uint8_t *buf, size_t size, uint32_t crc)
+{
+	crc = ~crc;
+
+	// Align the input buffer because this was shown to be
+	// significantly faster than unaligned accesses.
+	const size_t align_amount = my_min(size, (8 - (uintptr_t)buf) & 7);
+
+	for (const uint8_t *limit = buf + align_amount; buf < limit; ++buf)
+		crc = __crc32b(crc, *buf);
+
+	size -= align_amount;
+
+	// Process 8 bytes at a time. The end point is determined by
+	// ignoring the least significant three bits of size to ensure
+	// we do not process past the bounds of the buffer. This guarentees
+	// that limit is a multiple of 8 and is strictly less than size.
+	for (const uint8_t *limit = buf + (size & ~((size_t)7));
+			buf < limit; buf += 8)
+		crc = __crc32d(crc, aligned_read64le(buf));
+
+	// Process the remaining bytes that are not 8 byte aligned.
+	for (const uint8_t *limit = buf + (size & 7); buf < limit; ++buf)
+		crc = __crc32b(crc, *buf);
+
+	return ~crc;
+}
+
+
+#if defined(CRC32_GENERIC) && defined(CRC32_ARCH_OPTIMIZED)
+static inline bool
+is_arch_extension_supported(void)
+{
+#if defined(HAVE_GETAUXVAL)
+	return (getauxval(AT_HWCAP) & HWCAP_CRC32) != 0;
+
+#elif defined(HAVE_ELF_AUX_INFO)
+	unsigned long feature_flags;
+
+	elf_aux_info(AT_HWCAP, &feature_flags, sizeof(feature_flags));
+	return feature_flags & HWCAP_CRC32 != 0;
+
+#elif defined(_WIN32)
+	return IsProcessorFeaturePresent(
+			PF_ARM_V8_CRC32_INSTRUCTIONS_AVAILABLE);
+
+#elif defined(__APPLE__) && defined(HAVE_SYSCTLBYNAME)
+	int has_crc32 = 0;
+	size_t size = sizeof(has_crc32);
+
+	// The sysctlbyname() function requires a string identifier for the
+	// CPU feature it tests. The Apple documentation lists the string
+	// "hw.optional.armv8_crc32", which can be found here:
+	// (https://developer.apple.com/documentation/kernel/1387446-sysctlbyname/determining_instruction_set_characteristics#3915619)
+	int err = sysctlbyname("hw.optional.armv8_crc32", &has_crc32,
+			&size, NULL, 0);
+
+	return !err && has_crc32;
+
+#else
+	// If a runtime detection method cannot be found, then this must
+	// be a compile time error. The checks in crc_common.h should ensure
+	// a runtime detection method is always found if this function is
+	// built. It would be possible to just return false here, but this
+	// is inefficient for binary size and runtime since only the generic
+	// method could ever be used.
+#	error Runtime detection method unavailable.
+#endif
+}
+#endif
+
+#endif // LZMA_CRC32_ARM64_H
diff --git a/src/liblzma/check/crc32_fast.c b/src/liblzma/check/crc32_fast.c
index be034bdc..0b667d8b 100644
--- a/src/liblzma/check/crc32_fast.c
+++ b/src/liblzma/check/crc32_fast.c
@@ -19,8 +19,7 @@
 #	define BUILDING_CRC32_CLMUL
 #	include "crc_x86_clmul.h"
 #elif defined(CRC32_ARM64)
-#	define BUILDING_CRC32_AARCH64
-#	include "crc32_aarch64.h"
+#	include "crc32_arm64.h"
 #endif
 
 
diff --git a/src/liblzma/check/crc64_fast.c b/src/liblzma/check/crc64_fast.c
index 3d94ed3f..d1ab6862 100644
--- a/src/liblzma/check/crc64_fast.c
+++ b/src/liblzma/check/crc64_fast.c
@@ -17,9 +17,6 @@
 #if defined(CRC_X86_CLMUL)
 #	define BUILDING_CRC64_CLMUL
 #	include "crc_x86_clmul.h"
-#elif defined(CRC32_ARM64)
-#	define BUILDING_CRC64_AARCH64
-#	include "crc32_aarch64.h"
 #endif