Add support for ARM

- Build system fixes
- No user-mode high-precision timer by default, use clock_gettime
- Use C11 aligned_alloc pending C++17 std::aligned_alloc

Change-Id: I268365bdfd11d1e817a89584b9e086ee5b86e1dc

runtime/cmake_modules/hsa_common.cmake

@@ -57,9 +57,9 @@ if(UNIX)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fpic")
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wl,--unresolved-symbols=ignore-in-shared-libs")
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-strict-aliasing")
-  if ( IS64BIT )
+  if ( CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" )
     set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64  -msse -msse2" )
-  else ()
+  elseif ( CMAKE_SYSTEM_PROCESSOR STREQUAL "x86" )
     set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m32" )
   endif ()
   if ( "${CMAKE_BUILD_TYPE}" STREQUAL Debug )

runtime/hsa-ext-finalize/CMakeLists.txt

@@ -89,9 +89,9 @@ set ( CMAKE_CXX_FLAGS "-std=c++11 " )
 
 set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -fexceptions -fno-rtti -fvisibility=hidden -Wno-error=sign-compare -Wno-sign-compare -Wno-write-strings -Wno-deprecated-declarations -Wno-conversion-null -fno-math-errno -fno-threadsafe-statics -fmerge-all-constants -fms-extensions -Wno-error=comment -Wno-comment -Wno-error=pointer-arith -Wno-pointer-arith -fPIC" )
 
-if ( IS64BIT )
+if ( CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" )
     set  ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64  -msse -msse2" )
-else  ()
+elseif ( CMAKE_SYSTEM_PROCESSOR STREQUAL "x86" )
     set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m32" )
 endif ()
 

runtime/hsa-ext-image/CMakeLists.txt

@@ -88,9 +88,9 @@ set ( CMAKE_CXX_FLAGS "-std=c++11 " )
 
 set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -fexceptions -fno-rtti -fvisibility=hidden -Wno-error=sign-compare -Wno-sign-compare -Wno-write-strings -Wno-deprecated-declarations -Wno-conversion-null -fno-math-errno -fno-threadsafe-statics -fmerge-all-constants -fms-extensions -Wno-error=comment -Wno-comment -Wno-error=pointer-arith -Wno-pointer-arith -fPIC" )
 
-if ( IS64BIT )
+if ( CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" )
     set  ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64  -msse -msse2" )
-else  ()
+elseif ( CMAKE_SYSTEM_PROCESSOR STREQUAL "x86" )
     set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m32" )
 endif ()
 

runtime/hsa-runtime-tools/CMakeLists.txt

@@ -76,9 +76,9 @@ set ( CMAKE_CXX_FLAGS "-std=c++11 ")
 
 set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror=return-type -Werror -fexceptions -fno-rtti -fvisibility=hidden -Wno-error=sign-compare -Wno-error=enum-compare -Wno-sign-compare -Wno-write-strings -Wno-deprecated-declarations -Wno-conversion-null -fno-math-errno -fno-threadsafe-statics -fmerge-all-constants -fms-extensions -Wno-error=comment -Wno-comment -Wno-error=pointer-arith -Wno-pointer-arith -fPIC" )
 
-if (IS64BIT)
+if ( CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" )
     set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64  -msse -msse2" )
-else ()
+elseif ( CMAKE_SYSTEM_PROCESSOR STREQUAL "x86" )
     set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m32" )
 endif ()
 

runtime/hsa-runtime/cmake_modules/hsa_common.cmake

@@ -57,9 +57,9 @@ if(UNIX)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fpic")
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wl,--unresolved-symbols=ignore-in-shared-libs")
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-strict-aliasing")
-  if ( IS64BIT )
+  if ( CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" )
     set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64  -msse -msse2" )
-  else ()
+  elseif ( CMAKE_SYSTEM_PROCESSOR STREQUAL "x86" )
     set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m32" )
   endif ()
   if ( "${CMAKE_BUILD_TYPE}" STREQUAL Debug )

runtime/hsa-runtime/core/util/timer.h

@@ -46,7 +46,7 @@
 #include "core/util/utils.h"
 #include "core/util/os.h"
 #include <chrono>
-
+#include <time.h>
 #include <type_traits>
 
 namespace timer {
@@ -144,8 +144,18 @@ class fast_clock {
   typedef uint64_t raw_rep;
   typedef double raw_frequency;
 
+#ifdef __x86_64__
   static __forceinline raw_rep raw_now() { return __rdtsc(); }
   static __forceinline raw_frequency raw_freq() { return freq; }
+#endif
+#ifdef __aarch64__
+  static __forceinline raw_rep raw_now() {
+    struct timespec ts;
+    clock_gettime(CLOCK_MONOTONIC_RAW, &ts);
+    return (raw_rep(ts.tv_sec) * 1000000000 + raw_rep(ts.tv_nsec));
+  }
+  static __forceinline raw_frequency raw_freq() { return 1.e-9; }
+#endif
 
  private:
   static double period_ps;

runtime/hsa-runtime/core/util/utils.h

@@ -54,9 +54,9 @@ typedef unsigned int uint;
 typedef uint64_t uint64;
 
 #if defined(__GNUC__)
-#include "mm_malloc.h"
 #if defined(__i386__) || defined(__x86_64__)
 #include <x86intrin.h>
+#elif defined(__aarch64__)
 #else
 #error \
     "Processor or compiler not identified.  " \
@@ -70,9 +70,9 @@ typedef uint64_t uint64;
 #define __ALIGNED__(x) __attribute__((aligned(x)))
 
 static __forceinline void* _aligned_malloc(size_t size, size_t alignment) {
-  return _mm_malloc(size, alignment);
+  return aligned_alloc(alignment, size);
 }
-static __forceinline void _aligned_free(void* ptr) { return _mm_free(ptr); }
+static __forceinline void _aligned_free(void* ptr) { return free(ptr); }
 #elif defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_X64))
 #include "intrin.h"
 #define __ALIGNED__(x) __declspec(align(x))