make sure aligned structures/classes use the btAlignedAlloc/btAlignedFree, by overriding the operator new/delete for that struct/class.

integrated some contributions from IBM Germany for libspe2

src/LinearMath/btPoolAllocator.h

@@ -17,6 +17,7 @@ subject to the following restrictions:
 #define _BT_POOL_ALLOCATOR_H
 
 #include "btScalar.h"
+#include "btAlignedAllocator.h"
 
 class btPoolAllocator
 {
@@ -32,7 +33,7 @@ public:
 		:m_elemSize(elemSize),
 		m_maxElements(maxElements)
 	{
-		m_pool = new unsigned char[m_elemSize*m_maxElements];
+		m_pool = (unsigned char*) btAlignedAlloc(m_elemSize*m_maxElements,16);
 
 		unsigned char* p = m_pool;
         m_firstFree = p;
@@ -47,7 +48,7 @@ public:
 
 	~btPoolAllocator()
 	{
-		delete m_pool;
+		btAlignedFree( m_pool);
 	}
 
 	void*	allocate(int size)

src/LinearMath/btScalar.h

@@ -25,11 +25,13 @@ subject to the following restrictions:
 
 
 
+
 #ifdef WIN32
 
 		#if defined(__MINGW32__) || defined(__CYGWIN__) || (defined (_MSC_VER) && _MSC_VER < 1300)
 			#define SIMD_FORCE_INLINE inline
 			#define ATTRIBUTE_ALIGNED16(a) a
+			#define ATTRIBUTE_ALIGNED128(a) a
 		#else
 			#define BT_HAS_ALIGNED_ALOCATOR
 			#pragma warning(disable:4530)
@@ -37,6 +39,7 @@ subject to the following restrictions:
 			#pragma warning(disable:4786)
 			#define SIMD_FORCE_INLINE __forceinline
 			#define ATTRIBUTE_ALIGNED16(a) __declspec(align(16)) a
+			#define ATTRIBUTE_ALIGNED128(a) __declspec (align(128)) a
 		#ifdef _XBOX
 			#define BT_USE_VMX128
 
@@ -52,29 +55,61 @@ subject to the following restrictions:
 		#define btAssert assert
 		//btFullAssert is optional, slows down a lot
 		#define btFullAssert(x)
+
+		#define btLikely(_c)  _c
+		#define btUnlikely(_c) _c
+
 #else
 	
 #if defined	(__CELLOS_LV2__)
 		#define SIMD_FORCE_INLINE inline
 		#define ATTRIBUTE_ALIGNED16(a) a __attribute__ ((aligned (16)))
+		#define ATTRIBUTE_ALIGNED128(a) a __attribute__ ((aligned (128)))
 		#ifndef assert
 		#include <assert.h>
 		#endif
 		#define btAssert assert
 		//btFullAssert is optional, slows down a lot
 		#define btFullAssert(x)
+
+		#define btLikely(_c)  _c
+		#define btUnlikely(_c) _c
+
 #else
 
+#ifdef USE_LIBSPE2
+
+		#define SIMD_FORCE_INLINE __inline
+		#define ATTRIBUTE_ALIGNED16(a) a __attribute__ ((aligned (16)))
+		#define ATTRIBUTE_ALIGNED128(a) a __attribute__ ((aligned (128)))
+		#ifndef assert
+		#include <assert.h>
+		#endif
+		#define btAssert assert
+		//btFullAssert is optional, slows down a lot
+		#define btFullAssert(x)
+
+
+		#define btLikely(_c)   __builtin_expect((_c), 1)
+		#define btUnlikely(_c) __builtin_expect((_c), 0)
+		
+
+#else
 	//non-windows systems
 
 		#define SIMD_FORCE_INLINE inline
 		#define ATTRIBUTE_ALIGNED16(a) a
+		#define ATTRIBUTE_ALIGNED128(a) a
 		#ifndef assert
 		#include <assert.h>
 		#endif
 		#define btAssert assert
 		//btFullAssert is optional, slows down a lot
 		#define btFullAssert(x)
+
+
+#endif // LIBSPE2
+
 #endif	//__CELLOS_LV2__
 #endif
 
@@ -94,6 +129,14 @@ typedef float btScalar;
 #endif
 
 
+#define BT_DECLARE_ALIGNED_ALLOCATOR() \
+	SIMD_FORCE_INLINE void* operator new(size_t sizeInBytes)	{ return btAlignedAlloc(sizeInBytes,16); }	\
+	SIMD_FORCE_INLINE void  operator delete(void* ptr)			{ btAlignedFree(ptr); }	\
+	SIMD_FORCE_INLINE void* operator new(size_t, void* ptr)	{ return ptr; }	\
+	SIMD_FORCE_INLINE void  operator delete(void*, void*)		{ }	\
+
+
+
 #if defined(BT_USE_DOUBLE_PRECISION) || defined(BT_FORCE_DOUBLE_FUNCTIONS)
 		
 SIMD_FORCE_INLINE btScalar btSqrt(btScalar x) { return sqrt(x); }

src/LinearMath/btStackAlloc.h

@@ -21,6 +21,7 @@ Nov.2006
 #define BT_STACK_ALLOC
 
 #include "btScalar.h" //for btAssert
+#include "btAlignedAllocator.h"
 
 struct btBlock
 {
@@ -39,7 +40,7 @@ public:
 	inline void		create(unsigned int size)
 	{
 		destroy();
-		data		=	new unsigned char[size];
+		data		=  (unsigned char*) btAlignedAlloc(size,16);
 		totalsize	=	size;
 	}
 	inline void		destroy()
@@ -49,7 +50,9 @@ public:
 
 		if(usedsize==0)
 		{
-			if(!ischild)		delete[] data;
+			if(!ischild)		
+				btAlignedFree(data);
+
 			data				=	0;
 			usedsize			=	0;
 		}