From fc835e25a28de37f32b3699f33824a16afd3ced3 Mon Sep 17 00:00:00 2001
From: Matthias Hochsteger <mhochsteger@cerbsim.com>
Date: Mon, 7 Dec 2020 10:58:39 +0100
Subject: [PATCH] merge amd64 and arm code

---
 CMakeLists.txt                      |  1 -
 libsrc/core/ngcore_api.hpp          |  8 ++++++++
 libsrc/core/utils.hpp               | 16 ++++++++++++----
 libsrc/general/ngsimd.hpp           | 27 +++++++++++----------------
 libsrc/interface/nginterface_v2.cpp |  3 +--
 5 files changed, 32 insertions(+), 23 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 723be197..9cd4da39 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -197,7 +197,6 @@ if(HAVE_DLFCN_H)
 endif()
 
 include_directories(BEFORE ${CMAKE_CURRENT_BINARY_DIR})
-include_directories(BEFORE /usr/aarch64-linux-gnu/include)
 
 #######################################################################
 # platform specific definitions and flags
diff --git a/libsrc/core/ngcore_api.hpp b/libsrc/core/ngcore_api.hpp
index b6412157..c74f9841 100644
--- a/libsrc/core/ngcore_api.hpp
+++ b/libsrc/core/ngcore_api.hpp
@@ -67,6 +67,14 @@
   #endif
 #endif
 
+#if defined(__amd64__) || defined(_M_AMD64)
+#define NETGEN_ARCH_AMD64
+#endif
+
+#if defined(__arm__) || defined(_M_ARM)
+#define NETGEN_ARCH_ARM
+#endif
+
 #ifdef __MAC_OS_X_VERSION_MIN_REQUIRED
 #if __MAC_OS_X_VERSION_MIN_REQUIRED < 101400
 // The c++ standard library on MacOS 10.13 and earlier has no aligned new operator,
diff --git a/libsrc/core/utils.hpp b/libsrc/core/utils.hpp
index db5b65d1..c201c9c6 100644
--- a/libsrc/core/utils.hpp
+++ b/libsrc/core/utils.hpp
@@ -7,7 +7,10 @@
 #include <ostream>
 #include <sstream>
 #include <string>
+
+#if defined(__APPLE__) && defined(NETGEN_ARCH_ARM)
 #include <mach/mach_time.h>
+#endif
 
 /*
 #ifdef WIN32
@@ -55,11 +58,16 @@ namespace ngcore
 
   inline TTimePoint GetTimeCounter() noexcept
   {
+#if defined(__APPLE__) && defined(NETGEN_ARCH_ARM)
     return mach_absolute_time();
-    // return __builtin_readcyclecounter();
-    // return TTimePoint(__rdtsc());
-      // return TTimePoint(__builtin_readcyclecounter());
-      // return TTimePoint(0);
+#elif defined(NETGEN_ARCH_AMD64)
+    return __rdtsc();
+#elif defined(NETGEN_ARCH_ARM)
+    return __builtin_readcyclecounter();
+#else
+#warning "Unsupported CPU architecture"
+    return 0;
+#endif
   }
 
   template <class T>
diff --git a/libsrc/general/ngsimd.hpp b/libsrc/general/ngsimd.hpp
index 87d6ba8a..69e1502f 100644
--- a/libsrc/general/ngsimd.hpp
+++ b/libsrc/general/ngsimd.hpp
@@ -6,7 +6,6 @@
 /* Date:   25. Mar. 16                                                    */
 /**************************************************************************/
 
-// #include <immintrin.h>
 #include <tuple>
 #include <ostream>
 #include <stdexcept>
@@ -15,6 +14,10 @@
 
 #include <core/utils.hpp>
 
+#ifdef NETGEN_ARCH_AMD64
+#include <immintrin.h>
+#endif // NETGEN_ARCH_AMD64
+
 #ifdef WIN32
 #ifndef AVX_OPERATORS_DEFINED
 #define AVX_OPERATORS_DEFINED
@@ -50,23 +53,12 @@ NG_INLINE __m256d operator/= (__m256d &a, __m256d b) { return a = a/b; }
 
 namespace ngsimd
 {
-  // MSVC does not define SSE. It's always present on 64bit cpus
-#if (defined(_M_AMD64) || defined(_M_X64) || defined(__AVX__))
-#ifndef __SSE__
-#define __SSE__
-#endif
-#ifndef __SSE2__
-#define __SSE2__
-#endif
-#endif
-
-  
   constexpr int GetDefaultSIMDSize() {
 #if defined __AVX512F__
     return 8;
 #elif defined __AVX__
     return 4;
-#elif defined __SSE__
+#elif defined NETGEN_ARCH_AMD64
     return 2;
 #else
     return 1;
@@ -74,12 +66,15 @@ namespace ngsimd
   }
 
 #if defined __AVX512F__
+    #define NETGEN_HAVE_SIMD
     typedef __m512 tAVX;
     typedef __m512d tAVXd;
 #elif defined __AVX__
+    #define NETGEN_HAVE_SIMD
     typedef __m256 tAVX;
     typedef __m256d tAVXd;
-#elif defined __SSE__
+#elif defined NETGEN_ARCH_AMD64
+    #define NETGEN_HAVE_SIMD
     typedef __m128 tAVX;
     typedef __m128d tAVXd;
 #endif
@@ -256,7 +251,7 @@ using std::fabs;
 /////////////////////////////////////////////////////////////////////////////
 // SSE - Simd width 2
 /////////////////////////////////////////////////////////////////////////////
-#ifdef __SSE__
+#ifdef NETGEN_ARCH_AMD64
   template<>
   class alignas(16) SIMD<double,2>
   {
@@ -349,7 +344,7 @@ using std::fabs;
     return SIMD<double,4>(hsum);
   }
   */
-#endif // __SSE__
+#endif // NETGEN_ARCH_AMD64
 
   
 
diff --git a/libsrc/interface/nginterface_v2.cpp b/libsrc/interface/nginterface_v2.cpp
index a276f376..dce172d3 100644
--- a/libsrc/interface/nginterface_v2.cpp
+++ b/libsrc/interface/nginterface_v2.cpp
@@ -779,8 +779,7 @@ namespace netgen
 
 
 
-#ifdef __SSE__
-#include <immintrin.h>
+#ifdef NETGEN_HAVE_SIMD
   
   template<> DLL_HEADER void Ngx_Mesh :: 
   MultiElementTransformation<1,1> (int elnr, int npts,