There is still a segmenttion fault issue with the AVX code, disable it until this is sorted out

CMakeLists.txt

@@ -9,7 +9,16 @@ if(COMMAND cmake_policy)
    endif()
 endif()
 
-message("CMAKE Build type: ${CMAKE_BUILD_TYPE}")
+
+message(STATUS "CMAKE Build type: ${CMAKE_BUILD_TYPE}")
+# Set a default build type if none was specified
+if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
+  message(STATUS "Setting build type to 'Debug' as none was specified.")
+  set(CMAKE_BUILD_TYPE Debug CACHE STRING "Choose the type of build." FORCE)
+  # Set the possible values of build type for cmake-gui
+  set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release"
+    "MinSizeRel" "RelWithDebInfo")
+endif()
 
 include (CheckFunctionExists)
 include (CheckIncludeFile)
@@ -366,7 +375,8 @@ if (CLANG)
 
     if(ENABLE_SIMD)
       if (X86 OR X86_64)
-        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse3 -mfpmath=sse")
+        set(CMAKE_C_FLAGS_RELEASE "-O3 -msse2 -mfpmath=sse")
+        set(CMAKE_CXX_FLAGS_RELEASE "-O3 -msse2 -mfpmath=sse")
       endif()
     endif()
 endif()

simgear/math/SGVec3.hxx

@@ -451,14 +451,14 @@ template<typename T>
 inline
 T
 dist(const SGVec3<T>& v1, const SGVec3<T>& v2)
-{ return norm(v1 - v2); }
+{ return simd4::magnitude(v1.simd3() - v2.simd3()); }
 
 /// The squared euclidean distance of the two vectors
 template<typename T>
 inline
 T
 distSqr(SGVec3<T> v1, const SGVec3<T>& v2)
-{ v1 -= v2; return dot(v1, v1); }
+{ return simd4::magnitude2(v1.simd3() - v2.simd3()); }
 
 // calculate the projection of u along the direction of d.
 template<typename T>
@@ -466,7 +466,7 @@ inline
 SGVec3<T>
 projection(const SGVec3<T>& u, const SGVec3<T>& d)
 {
-  T denom = dot(d, d);
+  T denom = simd4::magnitude2(d);
   T ud = dot(u, d);
   if (SGLimits<T>::min() < denom) return u;
   else return d * (dot(u, d) / denom);

simgear/math/simd.hxx

@@ -21,6 +21,7 @@
 #include <cstdint>
 #include <cstring>
 #include <cmath>
+#include <new>
 
 #include <simgear/math/SGLimits.hxx>
 #include <simgear/math/SGMisc.hxx>
@@ -330,10 +331,12 @@ public:
     simd_aligned16() {}
     ~simd_aligned16() {}
 
-    void *operator new (size_t size) {
+    static void *operator new (size_t size) throw (std::bad_alloc) {
-        return _mm_malloc(size, 16);
+        void *p = _mm_malloc(size, 16);
+        if (!p) throw std::bad_alloc();
+        return p;
     }
-    void operator delete (void *p) {
+    static void operator delete (void *p) {
         _mm_free(p);
     }
 } ALIGN16C;
@@ -487,22 +490,12 @@ inline float dot(simd4_t<float,4> v1, const simd4_t<float,4>& v2) {
 template<>
 inline simd4_t<float,3> cross(const simd4_t<float,3>& v1, const simd4_t<float,3>& v2)
 {
-#if 1
     // http://threadlocalmutex.com/?p=8
     __m128 v41 = v1.v4(), v42 = v2.v4();
     __m128 a = _mm_shuffle_ps(v41, v41, _MM_SHUFFLE(3, 0, 2, 1));
     __m128 b = _mm_shuffle_ps(v42, v42, _MM_SHUFFLE(3, 0, 2, 1));
     __m128 c = _mm_sub_ps(_mm_mul_ps(v41, b), _mm_mul_ps(a, v42));
     return  _mm_shuffle_ps(c, c, _MM_SHUFFLE(3, 0, 2, 1));
-#else
-    v1 = _mm_sub_ps(
-            _mm_mul_ps(_mm_shuffle_ps(v1.v4(),v1.v4(),_MM_SHUFFLE(3, 0, 2, 1)),
-                       _mm_shuffle_ps(v2.v4(),v2.v4(),_MM_SHUFFLE(3, 1, 0, 2))),
-            _mm_mul_ps(_mm_shuffle_ps(v1.v4(),v1.v4(),_MM_SHUFFLE(3, 1, 0, 2)),
-                       _mm_shuffle_ps(v2.v4(),v2.v4(),_MM_SHUFFLE(3, 0, 2, 1)))
-   );
-   return v1;
-#endif
 }
 
 template<int N>
@@ -529,10 +522,8 @@ inline simd4_t<float,N>abs(simd4_t<float,N> v) {
 # endif
 
 
-# ifdef __AVX__
+# ifdef __AVX_unsupported__
-#  include <pmmintrin.h>
 #  include <immintrin.h>
-// #  include "avxintrin-emu.h"
 
 ALIGN32
 class simd_aligned32
@@ -541,10 +532,12 @@ public:
     simd_aligned32() {}
     ~simd_aligned32() {}
 
-    void *operator new (size_t size) {
+    static void *operator new (size_t size) throw (std::bad_alloc) {
-        return _mm_malloc(size, 32);
+        void *p = _mm_malloc(size, 32);
+        if (!p) throw std::bad_alloc();
+        return p;
     }
-    void operator delete (void *p) {
+    static void operator delete (void *p) {
         _mm_free(p);
     }
 } ALIGN32C;
@@ -666,7 +659,7 @@ inline simd4_t<double,2>::simd4_t(double d) {
 namespace simd4
 {
 // http://berenger.eu/blog/sseavxsimd-horizontal-sum-sum-simd-vector-intrinsic/
-inline static float hsum_pd_avx(__m256d v) {
+inline static double hsum_pd_avx(__m256d v) {
     const __m128d valupper = _mm256_extractf128_pd(v, 1);
     const __m128d vallower = _mm256_castpd256_pd128(v);
     _mm256_zeroupper();
@@ -685,7 +678,8 @@ inline double dot(simd4_t<double,4> v1, const simd4_t<double,4>& v2) {
     return hsum_pd_avx(_mm256_mul_pd(v1.v4(),v2.v4()));
 }
 
-#ifdef __AVX2__
+#  ifdef __AVX2__
+#   include <pmmintrin.h>
 template<>
 inline simd4_t<double,3> cross(const simd4_t<double,3>& v1, const simd4_t<double,3>& v2)
 {
@@ -699,7 +693,7 @@ inline simd4_t<double,3> cross(const simd4_t<double,3>& v1, const simd4_t<double
                      _mm256_permute4x64_pd(v41,_MM_SHUFFLE(3, 1, 0, 2)),
                      _mm256_permute4x64_pd(v42,_MM_SHUFFLE(3, 0, 2, 1))));
 }
-#endif
+#  endif
 
 template<int N>
 inline simd4_t<double,N> min(simd4_t<double,N> v1, const simd4_t<double,N>& v2) {

simgear/math/simd4x4.hxx

@@ -514,10 +514,9 @@ inline simd4_t<float,3> transform<float>(const simd4x4_t<float,4>& m, const simd
 # endif
 
 
-# ifdef __AVX__
+# ifdef __AVX_unsupported__
 #  include <pmmintrin.h>
 #  include <immintrin.h>
-// #  include "avxintrin-emu.h"
 
 template<>
 class simd4x4_t<double,4> : public simd_aligned32