Merge pull request #2619 from mark-mb/master

ARM64 support

CMakeLists.txt

@@ -41,7 +41,11 @@ if ("${TARGET_ARCH}" MATCHES "x86_64|i386")
 endif()
 if ("${TARGET_ARCH}" MATCHES "arm")
     set(ARM ON)
-    add_compile_definitions(__ARM__=1)
+    if ("${TARGET_ARCH}" MATCHES "armv8")
+        add_compile_definitions(__ARM64__=1)
+    else()
+        add_compile_definitions(__ARM__=1)
+    endif()
 endif()
 if ("${TARGET_ARCH}" MATCHES "ppc")
     set(PPC ON)

cmake_modules/TargetArch.cmake

@@ -24,19 +24,40 @@
 # OF THE POSSIBILITY OF SUCH DAMAGE.
 
 # Based on the Qt 5 processor detection code, so should be very accurate
-# https://qt.gitorious.org/qt/qtbase/blobs/master/src/corelib/global/qprocessordetection.h
-# Currently handles arm (v5, v6, v7), x86 (32/64), ia64, and ppc (32/64)
+# https://github.com/qt/qtbase/blob/9a6a847/src/corelib/global/qprocessordetection.h
+# Currently handles arm / aarch64 (v5, v6, v7, v8), x86 (32/64), ia64, and ppc (32/64)
 
 # Regarding POWER/PowerPC, just as is noted in the Qt source,
 # "There are many more known variants/revisions that we do not handle/detect."
 
 set(archdetect_c_code "
-#if defined(__arm__) || defined(__TARGET_ARCH_ARM)
-    #if defined(__ARM_ARCH_7__) \\
+#define _STR(x) #x
+#define STR(x) _STR(x)
+
+#if defined(__arm__) || defined(__TARGET_ARCH_ARM) || defined(_M_ARM) || \\
+    defined(_M_ARM64) || defined(__aarch64__) || defined(__ARM64__)
+
+    #if defined(__ARM_ARCH) && __ARM_ARCH > 1
+        #pragma message \"cmake_ARCH armv\" STR(__ARM_ARCH)
+        #error
+    #elif defined(__TARGET_ARCH_ARM) && __TARGET_ARCH_ARM > 1
+        #pragma message \"cmake_ARCH armv\" STR(__TARGET_ARCH_ARM)
+        #error
+    #elif defined(_M_ARM) && _M_ARM > 1
+        #error cmake_ARCH arm ## __M_ARM
+    #elif defined(__ARM64_ARCH_8__) \\
+        || defined(__aarch64__) \\
+        || defined(__ARMv8__) \\
+        || defined(__ARMv8_A__) \\
+        || defined(_M_ARM64)
+        #error cmake_ARCH armv8
+    #elif defined(__ARM_ARCH_7__) \\
         || defined(__ARM_ARCH_7A__) \\
         || defined(__ARM_ARCH_7R__) \\
         || defined(__ARM_ARCH_7M__) \\
-        || (defined(__TARGET_ARCH_ARM) && __TARGET_ARCH_ARM-0 >= 7)
+        || defined(__ARM_ARCH_7S__) \\
+        || defined(_ARM_ARCH_7) \\
+        || defined(__CORE_CORTEXA__)
         #error cmake_ARCH armv7
     #elif defined(__ARM_ARCH_6__) \\
         || defined(__ARM_ARCH_6J__) \\
@@ -44,11 +65,10 @@ set(archdetect_c_code "
         || defined(__ARM_ARCH_6Z__) \\
         || defined(__ARM_ARCH_6K__) \\
         || defined(__ARM_ARCH_6ZK__) \\
-        || defined(__ARM_ARCH_6M__) \\
-        || (defined(__TARGET_ARCH_ARM) && __TARGET_ARCH_ARM-0 >= 6)
+        || defined(__ARM_ARCH_6M__)
         #error cmake_ARCH armv6
     #elif defined(__ARM_ARCH_5TEJ__) \\
-        || (defined(__TARGET_ARCH_ARM) && __TARGET_ARCH_ARM-0 >= 5)
+        || defined(__ARM_ARCH_5TE__)
         #error cmake_ARCH armv5
     #else
         #error cmake_ARCH arm

libraries/vecmath/src/vecmath.cpp

-#if defined(__ANDROID__)
+#if defined(__ARM__) || defined(__ARM64__)
     #include "neon_mathfun.h"
 #else
     #if !defined(__PNACL__)

openmmapi/include/openmm/internal/hardware.h

@@ -93,7 +93,8 @@ static int getNumProcessors() {
 #ifdef WIN32
 #define cpuid __cpuid
 #else
-#if !defined(__ANDROID__) && !defined(__PNACL__) && !defined(__PPC__)
+#if !defined(__ANDROID__) && !defined(__PNACL__) && !defined(__PPC__) \
+    && !defined(__ARM__) && !defined(__ARM64__)
     static void cpuid(int cpuInfo[4], int infoType){
     #ifdef __LP64__
         __asm__ __volatile__ (

openmmapi/include/openmm/internal/vectorize.h

@@ -31,8 +31,8 @@
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE  *
  * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
  * -------------------------------------------------------------------------- */
-        
-#if defined(__ARM__)
+
+#if defined(__ARM__) || defined(__ARM64__)
     #include "vectorize_neon.h"
 #elif defined(__PPC__)
     #include "vectorize_ppc.h"

openmmapi/include/openmm/internal/vectorize_neon.h

@@ -32,7 +32,12 @@
  * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
  * -------------------------------------------------------------------------- */
 
+#ifdef __ANDROID__
 #include <cpu-features.h>
+#else
+#include <sys/auxv.h>
+#include <asm/hwcap.h>
+#endif
 #include <arm_neon.h>
 #include <cmath>
 
@@ -48,8 +53,16 @@ float32x4_t log_ps(float32x4_t);
  * Determine whether ivec4 and fvec4 are supported on this processor.
  */
 static bool isVec4Supported() {
+#ifdef __ANDROID__
     uint64_t features = android_getCpuFeatures();
     return (features & ANDROID_CPU_ARM_FEATURE_NEON) != 0;
+#elif defined(__ARM__)
+    unsigned long features = getauxval(AT_HWCAP);
+    return (features & HWCAP_NEON) != 0;
+#else
+    unsigned long features = getauxval(AT_HWCAP);
+    return (features & HWCAP_ASIMD) != 0;
+#endif
 }
 
 class ivec4;