Add runtime ARM64 AES check

Adds runtime CPU detection for Windows and macOS, and switches feature
detection of AES to runtime like for x86,

So far only on ARM64, since gcc doesn't do function versioning on ARM32,
but clang can, so it could be added later.

Change-Id: Ibe5d60f48cdae3e366a8ecd6263534ba2b09b131
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>

3 files changed, 66 insertions, 32 deletions

diff --git a/src/corelib/global/qsimd.cpp b/src/corelib/global/qsimd.cpp
index 37d6dd4e14..025c50b6e8 100644
--- a/src/corelib/global/qsimd.cpp
+++ b/src/corelib/global/qsimd.cpp
@@ -54,22 +54,19 @@
 #  if !defined(Q_CC_GNU)
 #    include <intrin.h>
 #  endif
-#elif defined(Q_OS_LINUX) && (defined(Q_PROCESSOR_ARM) || defined(Q_PROCESSOR_MIPS_32))
-#include "private/qcore_unix_p.h"
-
-#if QT_CONFIG(getauxval)
+#  if defined(Q_PROCESSOR_ARM64)
+#    include <processthreadsapi.h>
+#  endif
+#elif defined(Q_OS_LINUX) && defined(Q_PROCESSOR_MIPS_32)
+#  include "private/qcore_unix_p.h"
+#elif QT_CONFIG(getauxval) && defined(Q_PROCESSOR_ARM)
 #  include <sys/auxv.h>
-#endif
 
 // the kernel header definitions for HWCAP_*
 // (the ones we need/may need anyway)
 
 // copied from <asm/hwcap.h> (ARM)
-#define HWCAP_CRUNCH    1024
-#define HWCAP_THUMBEE   2048
 #define HWCAP_NEON      4096
-#define HWCAP_VFPv3     8192
-#define HWCAP_VFPv3D16  16384
 
 // copied from <asm/hwcap.h> (ARM):
 #define HWCAP2_AES   (1 << 0)
@@ -84,7 +81,9 @@
 #define AT_HWCAP2 26    /* extension of AT_HWCAP */
 
 #elif defined(Q_CC_GHS)
-#include <INTEGRITY_types.h>
+#  include <INTEGRITY_types.h>
+#elif defined(Q_OS_DARWIN) && defined(Q_PROCESSOR_ARM)
+#  include <sys/sysctl.h>
 #endif
 
 QT_BEGIN_NAMESPACE
@@ -155,8 +154,26 @@ static inline quint64 detectProcessorFeatures()
         return features;
     }
     // fall back to compile-time flags if getauxval failed
-#endif // QT_CONFIG(getauxval)
-
+#elif defined(Q_OS_DARWIN) && defined(Q_PROCESSOR_ARM)
+    unsigned feature;
+    size_t len = sizeof(feature);
+    if (sysctlbyname("hw.optional.neon", &feature, &len, nullptr, 0) == 0)
+        features |= feature ? CpuFeatureNEON : 0;
+    if (sysctlbyname("hw.optional.armv8_crc32", &feature, &len, nullptr, 0) == 0)
+        features |= feature ? CpuFeatureCRC32 : 0;
+    // There is currently no optional value for crypto/AES.
+#if defined(__ARM_FEATURE_CRYPTO)
+    features |= CpuFeatureAES;
+#endif
+    return features;
+#elif defined(Q_OS_WIN) && defined(Q_PROCESSOR_ARM64)
+    features |= CpuFeatureNEON;
+    if (IsProcessorFeaturePresent(PF_ARM_V8_CRC32_INSTRUCTIONS_AVAILABLE) != 0)
+        features |= CpuFeatureCRC32;
+    if (IsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE) != 0)
+        features |= CpuFeatureAES;
+    return features;
+#endif
 #if defined(__ARM_NEON__) || defined(__ARM_NEON)
     features |= CpuFeatureNEON;
 #endif
diff --git a/src/corelib/global/qsimd_p.h b/src/corelib/global/qsimd_p.h
index 2806a29cf1..c8134964af 100644
--- a/src/corelib/global/qsimd_p.h
+++ b/src/corelib/global/qsimd_p.h
@@ -145,17 +145,14 @@
 
 #define QT_COMPILER_SUPPORTS(x)     (QT_COMPILER_SUPPORTS_ ## x - 0)
 
-#if defined(Q_PROCESSOR_ARM)
-#  define QT_COMPILER_SUPPORTS_HERE(x)    (__ARM_FEATURE_ ## x)
-#  if defined(Q_CC_GNU) && !defined(Q_CC_INTEL) && Q_CC_GNU >= 600
+#if defined(Q_PROCESSOR_ARM) && defined(QT_COMPILER_SUPPORTS_SIMD_ALWAYS)
+#  define QT_COMPILER_SUPPORTS_HERE(x)    ((__ARM_FEATURE_ ## x) || (__ ## x ## __) || QT_COMPILER_SUPPORTS(x))
+#  if defined(Q_CC_GNU)
      /* GCC requires attributes for a function */
 #    define QT_FUNCTION_TARGET(x)  __attribute__((__target__(QT_FUNCTION_TARGET_STRING_ ## x)))
 #  else
 #    define QT_FUNCTION_TARGET(x)
 #  endif
-#  if !defined(__ARM_FEATURE_NEON) && defined(__ARM_NEON__)
-#    define __ARM_FEATURE_NEON           // also support QT_COMPILER_SUPPORTS_HERE(NEON)
-#  endif
 #elif defined(Q_PROCESSOR_MIPS)
 #  define QT_COMPILER_SUPPORTS_HERE(x)    (__ ## x ## __)
 #  define QT_FUNCTION_TARGET(x)
@@ -173,6 +170,9 @@
 #  else
 #    define QT_FUNCTION_TARGET(x)
 #  endif
+#elif defined(Q_PROCESSOR_ARM)
+#  define QT_COMPILER_SUPPORTS_HERE(x)    ((__ARM_FEATURE_ ## x) || (__ ## x ## __))
+#  define QT_FUNCTION_TARGET(x)
 #else
 #  define QT_COMPILER_SUPPORTS_HERE(x)    (__ ## x ## __)
 #  define QT_FUNCTION_TARGET(x)
@@ -259,17 +259,14 @@ QT_END_NAMESPACE
 
 #endif  /* Q_PROCESSOR_X86 */
 
-// Clang compiler fix, see http://lists.llvm.org/pipermail/cfe-commits/Week-of-Mon-20160222/151168.html
-// This should be tweaked with an "upper version" of clang once we know which release fixes the
-// issue. At that point we can rely on __ARM_FEATURE_CRC32 again.
-#if defined(Q_CC_CLANG) && defined(Q_OS_DARWIN) && defined (__ARM_FEATURE_CRC32)
-#  undef __ARM_FEATURE_CRC32
-#endif
-
 // NEON intrinsics
 // note: as of GCC 4.9, does not support function targets for ARM
 #if defined(__ARM_NEON) || defined(__ARM_NEON__)
+#if defined(Q_CC_CLANG)
+#define QT_FUNCTION_TARGET_STRING_NEON      "neon"
+#else
 #define QT_FUNCTION_TARGET_STRING_NEON      "+neon" // unused: gcc doesn't support function targets on non-aarch64, and on Aarch64 NEON is always available.
+#endif
 #ifndef __ARM_NEON__
 // __ARM_NEON__ is not defined on AArch64, but we need it in our NEON detection.
 #define __ARM_NEON__
@@ -291,15 +288,30 @@ inline uint8_t vaddv_u8(uint8x8_t v8)
 #endif
 
 #endif
-// AArch64/ARM64
-#if defined(Q_PROCESSOR_ARM_V8) && defined(__ARM_FEATURE_CRC32)
+
+#if defined(Q_PROCESSOR_ARM) && defined(__ARM_FEATURE_CRC32)
+#  include <arm_acle.h>
+#endif
+
 #if defined(Q_PROCESSOR_ARM_64)
-// only available on aarch64
+#if defined(Q_CC_CLANG)
+#define QT_FUNCTION_TARGET_STRING_AES        "crypto"
+#define QT_FUNCTION_TARGET_STRING_CRC32      "crc"
+#elif defined(Q_CC_GNU)
+#define QT_FUNCTION_TARGET_STRING_AES        "+crypto"
 #define QT_FUNCTION_TARGET_STRING_CRC32      "+crc"
 #endif
-#  include <arm_acle.h>
+#elif defined(Q_PROCESSOR_ARM_32)
+#if defined(Q_CC_CLANG)
+#define QT_FUNCTION_TARGET_STRING_AES        "armv8-a,crypto"
+#define QT_FUNCTION_TARGET_STRING_CRC32      "armv8-a,crc"
+#elif defined(Q_CC_GNU)
+#define QT_FUNCTION_TARGET_STRING_AES        "arch=armv8-a+crypto"
+#define QT_FUNCTION_TARGET_STRING_CRC32      "arch=armv8-a+crc"
+#endif
 #endif
 
+
 #ifdef __cplusplus
 #include <qatomic.h>
 
diff --git a/src/corelib/tools/qhash.cpp b/src/corelib/tools/qhash.cpp
index 2221de1461..46917cfbe9 100644
--- a/src/corelib/tools/qhash.cpp
+++ b/src/corelib/tools/qhash.cpp
@@ -527,7 +527,8 @@ lt16:
 }
 #endif
 
-#if defined(__ARM_FEATURE_CRYPTO) && !defined(QT_BOOTSTRAPPED)
+#if defined(Q_PROCESSOR_ARM) && QT_COMPILER_SUPPORTS_HERE(AES) && !defined(QHASH_AES_SANITIZER_BUILD) && !defined(QT_BOOTSTRAPPED)
+QT_FUNCTION_TARGET(AES)
 static size_t aeshash(const uchar *p, size_t len, size_t seed) noexcept
 {
     uint8x16_t key;
@@ -669,10 +670,14 @@ size_t qHashBits(const void *p, size_t size, size_t seed) noexcept
 #ifdef AESHASH
     if (seed && qCpuHasFeature(AES) && qCpuHasFeature(SSE4_2))
         return aeshash(reinterpret_cast<const uchar *>(p), size, seed);
-#elif defined(__ARM_FEATURE_CRYPTO) && !defined(QT_BOOTSTRAPPED)
-    // Do additional runtime check as Yocto hard enables Crypto extension for
+#elif defined(Q_PROCESSOR_ARM) && QT_COMPILER_SUPPORTS_HERE(AES) && !defined(QHASH_AES_SANITIZER_BUILD) && !defined(QT_BOOTSTRAPPED)
+# if defined(Q_OS_LINUX)
+    // Do specific runtime-only check as Yocto hard enables Crypto extension for
     // all armv8 configs
     if (seed && (qCpuFeatures() & CpuFeatureAES))
+# else
+    if (seed && qCpuHasFeature(AES))
+# endif
         return aeshash(reinterpret_cast<const uchar *>(p), size, seed);
 #endif
     if (size <= QT_POINTER_SIZE)