SSE2: Some initial CVector SSE2 math

--HG--
branch : sse2

code/nel/include/nel/misc/types_nl.h

@@ -336,14 +336,82 @@ typedef	unsigned	int			uint;			// at least 32bits (depend of processor)
 #endif
 
 #ifdef USE_SSE2
+
 extern void *operator new(size_t size) throw(std::bad_alloc);
 extern void *operator new[](size_t size) throw(std::bad_alloc);
 extern void operator delete(void *p) throw();
 extern void operator delete[](void *p) throw();
+
 #define NL_ALIGN_SSE2(nb) NL_ALIGN(nb)
+
+#	ifdef NL_COMP_VC
+
+inline void *aligned_malloc(size_t size, size_t alignment)
+{
+	return _aligned_malloc(size, alignment);
+}
+
+inline void aligned_free(void *ptr)
+{
+	_aligned_free(ptr);
+}
+
+#	else
+
+inline void *aligned_malloc(size_t size, size_t alignment)
+{
+	return memalign(alignment, size);
+}
+
+inline void aligned_free(void *ptr)
+{
+	free(ptr);
+}
+
+#	endif /* NL_COMP_ */
+
+template<class T>
+class aligned_allocator : public std::allocator<T>
+{
+public:
+	typedef size_t size_type;
+	typedef std::ptrdiff_t difference_type;
+	typedef T* pointer;
+	typedef const T* const_pointer;
+	typedef T& reference;
+	typedef const T& const_reference;
+	typedef T value_type;
+
+	template<class U>
+	struct rebind
+	{
+		typedef aligned_allocator<U> other;
+	};
+
+	aligned_allocator() : std::allocator<T>() {}
+
+	aligned_allocator(const aligned_allocator& other) : std::allocator<T>(other) {}
+
+	template<class U>
+	aligned_allocator(const aligned_allocator<U>& other) : std::allocator<T>(other) {}
+
+	~aligned_allocator() {}
+
+	pointer allocate(size_type num, const void* /*hint*/ = 0)
+	{
+		return static_cast<pointer>(aligned_malloc(NL_DEFAULT_MEMORY_ALIGNMENT, num * sizeof(T)));
+	}
+
+	void deallocate(pointer p, size_type /*num*/)
+	{
+		aligned_free(p);
+	}
+};
+
 #else
 #define NL_ALIGN_SSE2(nb) 
-#endif
+#endif /* USE_SSE2 */
+
 
 // CHashMap, CHashSet and CHashMultiMap definitions
 #if defined(_STLPORT_VERSION) // STLport detected

code/nel/include/nel/misc/vector.h

@@ -35,15 +35,24 @@ class IStream;
  * \author Lionel Berenguier
  * \author Nevrax France
  * \date 2000
+ * \author Jan Boon
+ * \date 2014
  */
 NL_ALIGN_SSE2(16)
 class CVector
 {
 public:		// Attributes.
-	float	x,y,z;
-
 #ifdef USE_SSE2
-	float	w; // Padding
+	union
+	{
+		struct
+		{
+			float x, y, z, P;
+		};
+		__m128 mm;
+	};
+#else
+	float	x,y,z;
 #endif
 
 public:		// const.
@@ -187,9 +196,21 @@ public:
 inline CVector blend(const CVector &v0, const CVector &v1, float lambda)
 {
 	float invLambda = 1.f - lambda;
+#ifdef USE_SSE2
+	CVector res;
+	__m128 mLambda = _mm_set1_ps(lambda);
+	__m128 mInvLambda = _mm_set1_ps(invLambda);
+	__m128 mv0 = v0.mm;
+	__m128 mv1 = v1.mm;
+	mv0 = _mm_mul_ps(mv0, mInvLambda);
+	mv1 = _mm_mul_ps(mv1, mLambda);
+	res.mm = _mm_add_ps(mv0, mv1);
+	return res;
+#else
 	return CVector(invLambda * v0.x + lambda * v1.x,
 		           invLambda * v0.y + lambda * v1.y,
 				   invLambda * v0.z + lambda * v1.z);
+#endif
 }
 
 

code/nel/include/nel/misc/vector_inline.h

@@ -31,23 +31,35 @@ namespace	NLMISC
 // Base Maths.
 inline	CVector	&CVector::operator+=(const CVector &v)
 {
+#ifdef USE_SSE2
+	mm = _mm_add_ps(mm, v.mm);
+#else
 	x+=v.x;
 	y+=v.y;
 	z+=v.z;
+#endif
 	return *this;
 }
 inline	CVector	&CVector::operator-=(const CVector &v)
 {
+#ifdef USE_SSE2
+	mm = _mm_sub_ps(mm, v.mm);
+#else
 	x-=v.x;
 	y-=v.y;
 	z-=v.z;
+#endif
 	return *this;
 }
 inline	CVector	&CVector::operator*=(float f)
 {
+#ifdef USE_SSE2
+	mm = _mm_mul_ps(mm, _mm_set1_ps(f));
+#else
 	x*=f;
 	y*=f;
 	z*=f;
+#endif
 	return *this;
 }
 inline	CVector	&CVector::operator/=(float f)
@@ -56,18 +68,36 @@ inline	CVector	&CVector::operator/=(float f)
 }
 inline	CVector	CVector::operator+(const CVector &v) const
 {
+#ifdef USE_SSE2
+	CVector res;
+	res.mm = _mm_add_ps(mm, v.mm);
+	return res;
+#else
 	CVector	ret(x+v.x, y+v.y, z+v.z);
 	return ret;
+#endif
 }
 inline	CVector	CVector::operator-(const CVector &v) const
 {
+#ifdef USE_SSE2
+	CVector res;
+	res.mm = _mm_sub_ps(mm, v.mm);
+	return res;
+#else
 	CVector	ret(x-v.x, y-v.y, z-v.z);
 	return ret;
+#endif
 }
 inline	CVector	CVector::operator*(float f) const
 {
+#ifdef USE_SSE2
+	CVector res;
+	res.mm = _mm_mul_ps(mm, _mm_set1_ps(f));
+	return res;
+#else
 	CVector	ret(x*f, y*f, z*f);
 	return ret;
+#endif
 }
 inline	CVector	CVector::operator/(float f) const
 {
@@ -75,12 +105,24 @@ inline	CVector	CVector::operator/(float f) const
 }
 inline	CVector	CVector::operator-() const
 {
+#ifdef USE_SSE2
+	CVector res;
+	res.mm = _mm_mul_ps(mm, _mm_set1_ps(-1.0f));
+	return res;
+#else
 	return CVector(-x,-y,-z);
+#endif
 }
 inline CVector	operator*(float f, const CVector &v)
 {
+#ifdef USE_SSE2
+	CVector res;
+	res.mm = _mm_mul_ps(v.mm, _mm_set1_ps(f));
+	return res;
+#else
 	CVector	ret(v.x*f, v.y*f, v.z*f);
 	return ret;
+#endif
 }
 
 

code/nel/src/misc/common.cpp

@@ -71,33 +71,7 @@ extern "C" long _ftol2( double dblSource ) { return _ftol( dblSource ); }
 #endif // NL_OS_WINDOWS
 
 
-#ifdef HAS_SSE2
-
-#	ifdef NL_COMP_VC
-
-inline void *aligned_malloc(size_t size, size_t alignment)
-{
-	return _aligned_malloc(size, alignment);
-}
-
-inline void aligned_free(void *p)
-{
-	_aligned_free(ptr);
-}
-
-#	else
-
-inline void *aligned_malloc(size_t size, size_t alignment)
-{
-	return memalign(alignment, size);
-}
-
-inline void aligned_free(void *ptr)
-{
-	free(ptr);
-}
-
-#	endif /* NL_COMP_ */
+#ifdef USE_SSE2
 
 void *operator new(size_t size) throw(std::bad_alloc)
 {