summaryrefslogtreecommitdiff
path: root/indra/llmath
diff options
context:
space:
mode:
Diffstat (limited to 'indra/llmath')
-rw-r--r--indra/llmath/CMakeLists.txt1
-rw-r--r--indra/llmath/llcalcparser.h2
-rw-r--r--indra/llmath/llcamera.h12
-rw-r--r--indra/llmath/llcoord.h117
-rw-r--r--indra/llmath/llmath.h6
-rw-r--r--indra/llmath/llmatrix3a.h2
-rw-r--r--indra/llmath/llmatrix4a.h2
-rw-r--r--indra/llmath/lloctree.h159
-rw-r--r--indra/llmath/llplane.h4
-rw-r--r--indra/llmath/llsimdmath.h3
-rw-r--r--indra/llmath/llsimdtypes.inl2
-rw-r--r--indra/llmath/llvector4a.cpp8
-rw-r--r--indra/llmath/llvector4a.h6
-rw-r--r--indra/llmath/llvector4a.inl1
-rw-r--r--indra/llmath/llvector4logical.h2
-rw-r--r--indra/llmath/llvolume.cpp436
-rw-r--r--indra/llmath/llvolume.h18
-rw-r--r--indra/llmath/llvolumemgr.cpp3
-rw-r--r--indra/llmath/llvolumeoctree.cpp6
-rw-r--r--indra/llmath/llvolumeoctree.h35
-rw-r--r--indra/llmath/m4math.cpp38
-rw-r--r--indra/llmath/tests/alignment_test.cpp141
-rw-r--r--indra/llmath/v3color.h1
23 files changed, 515 insertions, 490 deletions
diff --git a/indra/llmath/CMakeLists.txt b/indra/llmath/CMakeLists.txt
index b5e59c1ca3..5865ae030c 100644
--- a/indra/llmath/CMakeLists.txt
+++ b/indra/llmath/CMakeLists.txt
@@ -117,6 +117,7 @@ if (LL_TESTS)
# INTEGRATION TESTS
set(test_libs llmath llcommon ${LLCOMMON_LIBRARIES} ${WINDOWS_LIBRARIES})
# TODO: Some of these need refactoring to be proper Unit tests rather than Integration tests.
+ LL_ADD_INTEGRATION_TEST(alignment "" "${test_libs}")
LL_ADD_INTEGRATION_TEST(llbbox llbbox.cpp "${test_libs}")
LL_ADD_INTEGRATION_TEST(llquaternion llquaternion.cpp "${test_libs}")
LL_ADD_INTEGRATION_TEST(mathmisc "" "${test_libs}")
diff --git a/indra/llmath/llcalcparser.h b/indra/llmath/llcalcparser.h
index bd9c8c2519..e0ad270266 100644
--- a/indra/llmath/llcalcparser.h
+++ b/indra/llmath/llcalcparser.h
@@ -174,7 +174,7 @@ private:
F32 _log(const F32& a) const { return log(a); }
F32 _exp(const F32& a) const { return exp(a); }
F32 _fabs(const F32& a) const { return fabs(a); }
- F32 _floor(const F32& a) const { return llfloor(a); }
+ F32 _floor(const F32& a) const { return (F32)llfloor(a); }
F32 _ceil(const F32& a) const { return llceil(a); }
F32 _atan2(const F32& a,const F32& b) const { return atan2(a,b); }
diff --git a/indra/llmath/llcamera.h b/indra/llmath/llcamera.h
index ec67b91d05..0b591be622 100644
--- a/indra/llmath/llcamera.h
+++ b/indra/llmath/llcamera.h
@@ -60,7 +60,7 @@ static const F32 MAX_FIELD_OF_VIEW = 175.f * DEG_TO_RAD;
// roll(), pitch(), yaw()
// etc...
-
+LL_ALIGN_PREFIX(16)
class LLCamera
: public LLCoordFrame
{
@@ -108,7 +108,7 @@ public:
};
private:
- LLPlane mAgentPlanes[7]; //frustum planes in agent space a la gluUnproject (I'm a bastard, I know) - DaveP
+ LL_ALIGN_16(LLPlane mAgentPlanes[7]); //frustum planes in agent space a la gluUnproject (I'm a bastard, I know) - DaveP
U8 mPlaneMask[8]; // 8 for alignment
F32 mView; // angle between top and bottom frustum planes in radians.
@@ -116,13 +116,13 @@ private:
S32 mViewHeightInPixels; // for ViewHeightInPixels() only
F32 mNearPlane;
F32 mFarPlane;
- LLPlane mLocalPlanes[4];
+ LL_ALIGN_16(LLPlane mLocalPlanes[4]);
F32 mFixedDistance; // Always return this distance, unless < 0
LLVector3 mFrustCenter; // center of frustum and radius squared for ultra-quick exclusion test
F32 mFrustRadiusSquared;
- LLPlane mWorldPlanes[PLANE_NUM];
- LLPlane mHorizPlanes[HORIZ_PLANE_NUM];
+ LL_ALIGN_16(LLPlane mWorldPlanes[PLANE_NUM]);
+ LL_ALIGN_16(LLPlane mHorizPlanes[HORIZ_PLANE_NUM]);
U32 mPlaneCount; //defaults to 6, if setUserClipPlane is called, uses user supplied clip plane in
@@ -208,7 +208,7 @@ protected:
void calculateFrustumPlanes(F32 left, F32 right, F32 top, F32 bottom);
void calculateFrustumPlanesFromWindow(F32 x1, F32 y1, F32 x2, F32 y2);
void calculateWorldFrustumPlanes();
-};
+} LL_ALIGN_POSTFIX(16);
#endif
diff --git a/indra/llmath/llcoord.h b/indra/llmath/llcoord.h
index 706ad92787..9b76268afd 100644
--- a/indra/llmath/llcoord.h
+++ b/indra/llmath/llcoord.h
@@ -26,80 +26,87 @@
#ifndef LL_LLCOORD_H
#define LL_LLCOORD_H
+template<typename> class LLCoord;
+struct LL_COORD_TYPE_GL;
+struct LL_COORD_TYPE_WINDOW;
+struct LL_COORD_TYPE_SCREEN;
+
+typedef LLCoord<LL_COORD_TYPE_GL> LLCoordGL;
+typedef LLCoord<LL_COORD_TYPE_WINDOW> LLCoordWindow;
+typedef LLCoord<LL_COORD_TYPE_SCREEN> LLCoordScreen;
+
+struct LLCoordCommon
+{
+ LLCoordCommon(S32 x, S32 y) : mX(x), mY(y) {}
+ LLCoordCommon() : mX(0), mY(0) {}
+ S32 mX;
+ S32 mY;
+};
+
// A two-dimensional pixel value
-class LLCoord
+template<typename COORD_FRAME>
+class LLCoord : protected COORD_FRAME
{
public:
- S32 mX;
- S32 mY;
+ typedef LLCoord<COORD_FRAME> self_t;
+ typename COORD_FRAME::value_t mX;
+ typename COORD_FRAME::value_t mY;
LLCoord(): mX(0), mY(0)
{}
- LLCoord(S32 x, S32 y): mX(x), mY(y)
- {}
- virtual ~LLCoord()
+ LLCoord(typename COORD_FRAME::value_t x, typename COORD_FRAME::value_t y): mX(x), mY(y)
{}
- virtual void set(S32 x, S32 y) { mX = x; mY = y; }
-};
+ LLCoord(const LLCoordCommon& other)
+ {
+ COORD_FRAME::convertFromCommon(other);
+ }
+ LLCoordCommon convert() const
+ {
+ return COORD_FRAME::convertToCommon();
+ }
-// GL coordinates start in the client region of a window,
-// with left, bottom = 0, 0
-class LLCoordGL : public LLCoord
-{
-public:
- LLCoordGL() : LLCoord()
- {}
- LLCoordGL(S32 x, S32 y) : LLCoord(x, y)
- {}
- bool operator==(const LLCoordGL& other) const { return mX == other.mX && mY == other.mY; }
- bool operator!=(const LLCoordGL& other) const { return !(*this == other); }
-};
+ void set(typename COORD_FRAME::value_t x, typename COORD_FRAME::value_t y) { mX = x; mY = y;}
+ bool operator==(const self_t& other) const { return mX == other.mX && mY == other.mY; }
+ bool operator!=(const self_t& other) const { return !(*this == other); }
-//bool operator ==(const LLCoordGL& a, const LLCoordGL& b);
+ static const self_t& getTypedCoords(const COORD_FRAME& self) { return static_cast<const self_t&>(self); }
+ static self_t& getTypedCoords(COORD_FRAME& self) { return static_cast<self_t&>(self); }
+};
-// Window coords include things like window borders,
-// menu regions, etc.
-class LLCoordWindow : public LLCoord
+struct LL_COORD_TYPE_GL
{
-public:
- LLCoordWindow() : LLCoord()
- {}
- LLCoordWindow(S32 x, S32 y) : LLCoord(x, y)
- {}
- bool operator==(const LLCoordWindow& other) const { return mX == other.mX && mY == other.mY; }
- bool operator!=(const LLCoordWindow& other) const { return !(*this == other); }
-};
+ typedef S32 value_t;
+ LLCoordCommon convertToCommon() const
+ {
+ const LLCoordGL& self = LLCoordGL::getTypedCoords(*this);
+ return LLCoordCommon(self.mX, self.mY);
+ }
-// Screen coords start at left, top = 0, 0
-class LLCoordScreen : public LLCoord
+ void convertFromCommon(const LLCoordCommon& from)
+ {
+ LLCoordGL& self = LLCoordGL::getTypedCoords(*this);
+ self.mX = from.mX;
+ self.mY = from.mY;
+ }
+};
+
+struct LL_COORD_TYPE_WINDOW
{
-public:
- LLCoordScreen() : LLCoord()
- {}
- LLCoordScreen(S32 x, S32 y) : LLCoord(x, y)
- {}
- bool operator==(const LLCoordScreen& other) const { return mX == other.mX && mY == other.mY; }
- bool operator!=(const LLCoordScreen& other) const { return !(*this == other); }
+ typedef S32 value_t;
+
+ LLCoordCommon convertToCommon() const;
+ void convertFromCommon(const LLCoordCommon& from);
};
-class LLCoordFont : public LLCoord
+struct LL_COORD_TYPE_SCREEN
{
-public:
- F32 mZ;
-
- LLCoordFont() : LLCoord(), mZ(0.f)
- {}
- LLCoordFont(S32 x, S32 y, F32 z = 0) : LLCoord(x,y), mZ(z)
- {}
-
- void set(S32 x, S32 y) { LLCoord::set(x,y); mZ = 0.f; }
- void set(S32 x, S32 y, F32 z) { mX = x; mY = y; mZ = z; }
- bool operator==(const LLCoordFont& other) const { return mX == other.mX && mY == other.mY; }
- bool operator!=(const LLCoordFont& other) const { return !(*this == other); }
+ typedef S32 value_t;
+
+ LLCoordCommon convertToCommon() const;
+ void convertFromCommon(const LLCoordCommon& from);
};
-
#endif
diff --git a/indra/llmath/llmath.h b/indra/llmath/llmath.h
index 9297bcbac2..b93f89d674 100644
--- a/indra/llmath/llmath.h
+++ b/indra/llmath/llmath.h
@@ -85,7 +85,7 @@ const F32 F_ALMOST_ONE = 1.0f - F_ALMOST_ZERO;
const F32 FP_MAG_THRESHOLD = 0.0000001f;
// TODO: Replace with logic like is_approx_equal
-inline BOOL is_approx_zero( F32 f ) { return (-F_APPROXIMATELY_ZERO < f) && (f < F_APPROXIMATELY_ZERO); }
+inline bool is_approx_zero( F32 f ) { return (-F_APPROXIMATELY_ZERO < f) && (f < F_APPROXIMATELY_ZERO); }
// These functions work by interpreting sign+exp+mantissa as an unsigned
// integer.
@@ -111,13 +111,13 @@ inline BOOL is_approx_zero( F32 f ) { return (-F_APPROXIMATELY_ZERO < f) && (f <
// WARNING: Infinity is comparable with F32_MAX and negative
// infinity is comparable with F32_MIN
-inline BOOL is_approx_equal(F32 x, F32 y)
+inline bool is_approx_equal(F32 x, F32 y)
{
const S32 COMPARE_MANTISSA_UP_TO_BIT = 0x02;
return (std::abs((S32) ((U32&)x - (U32&)y) ) < COMPARE_MANTISSA_UP_TO_BIT);
}
-inline BOOL is_approx_equal(F64 x, F64 y)
+inline bool is_approx_equal(F64 x, F64 y)
{
const S64 COMPARE_MANTISSA_UP_TO_BIT = 0x02;
return (std::abs((S32) ((U64&)x - (U64&)y) ) < COMPARE_MANTISSA_UP_TO_BIT);
diff --git a/indra/llmath/llmatrix3a.h b/indra/llmath/llmatrix3a.h
index adb7e3389d..9916cfd2da 100644
--- a/indra/llmath/llmatrix3a.h
+++ b/indra/llmath/llmatrix3a.h
@@ -111,7 +111,7 @@ public:
protected:
- LLVector4a mColumns[3];
+ LL_ALIGN_16(LLVector4a mColumns[3]);
};
diff --git a/indra/llmath/llmatrix4a.h b/indra/llmath/llmatrix4a.h
index 27cf5b79f6..c4cefdb4fa 100644
--- a/indra/llmath/llmatrix4a.h
+++ b/indra/llmath/llmatrix4a.h
@@ -34,7 +34,7 @@
class LLMatrix4a
{
public:
- LLVector4a mMatrix[4];
+ LL_ALIGN_16(LLVector4a mMatrix[4]);
inline void clear()
{
diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h
index 3c1ae45d68..7348904c61 100644
--- a/indra/llmath/lloctree.h
+++ b/indra/llmath/lloctree.h
@@ -31,7 +31,6 @@
#include "v3math.h"
#include "llvector4a.h"
#include <vector>
-#include <set>
#define OCT_ERRS LL_WARNS("OctreeErrors")
@@ -79,16 +78,18 @@ public:
typedef LLOctreeTraveler<T> oct_traveler;
typedef LLTreeTraveler<T> tree_traveler;
- typedef typename std::set<LLPointer<T> > element_list;
- typedef typename std::set<LLPointer<T> >::iterator element_iter;
- typedef typename std::set<LLPointer<T> >::const_iterator const_element_iter;
+ typedef std::vector< LLPointer<T> > element_list; // note: don't remove the whitespace between "> >"
+ typedef LLPointer<T>* element_iter;
+ typedef const LLPointer<T>* const_element_iter;
typedef typename std::vector<LLTreeListener<T>*>::iterator tree_listener_iter;
- typedef typename std::vector<LLOctreeNode<T>* > child_list;
+ typedef LLOctreeNode<T>** child_list;
+ typedef LLOctreeNode<T>** child_iter;
+
typedef LLTreeNode<T> BaseType;
typedef LLOctreeNode<T> oct_node;
typedef LLOctreeListener<T> oct_listener;
- /*void* operator new(size_t size)
+ void* operator new(size_t size)
{
return ll_aligned_malloc_16(size);
}
@@ -96,7 +97,7 @@ public:
void operator delete(void* ptr)
{
ll_aligned_free_16(ptr);
- }*/
+ }
LLOctreeNode( const LLVector4a& center,
const LLVector4a& size,
@@ -105,6 +106,10 @@ public:
: mParent((oct_node*)parent),
mOctant(octant)
{
+ //always keep a NULL terminated list to avoid out of bounds exceptions in debug builds
+ mData.push_back(NULL);
+ mDataEnd = &mData[0];
+
mCenter = center;
mSize = size;
@@ -114,6 +119,8 @@ public:
mOctant = ((oct_node*) mParent)->getOctant(mCenter);
}
+ mElementCount = 0;
+
clearChildren();
}
@@ -121,6 +128,16 @@ public:
{
BaseType::destroyListeners();
+ for (U32 i = 0; i < mElementCount; ++i)
+ {
+ mData[i]->setBinIndex(-1);
+ mData[i] = NULL;
+ }
+
+ mData.clear();
+ mData.push_back(NULL);
+ mDataEnd = &mData[0];
+
for (U32 i = 0; i < getChildCount(); i++)
{
delete getChild(i);
@@ -217,13 +234,18 @@ public:
}
void accept(oct_traveler* visitor) { visitor->visit(this); }
- virtual bool isLeaf() const { return mChild.empty(); }
+ virtual bool isLeaf() const { return mChildCount == 0; }
- U32 getElementCount() const { return mData.size(); }
+ U32 getElementCount() const { return mElementCount; }
+ bool isEmpty() const { return mElementCount == 0; }
element_list& getData() { return mData; }
const element_list& getData() const { return mData; }
-
- U32 getChildCount() const { return mChild.size(); }
+ element_iter getDataBegin() { return &mData[0]; }
+ element_iter getDataEnd() { return mDataEnd; }
+ const_element_iter getDataBegin() const { return &mData[0]; }
+ const_element_iter getDataEnd() const { return mDataEnd; }
+
+ U32 getChildCount() const { return mChildCount; }
oct_node* getChild(U32 index) { return mChild[index]; }
const oct_node* getChild(U32 index) const { return mChild[index]; }
child_list& getChildren() { return mChild; }
@@ -287,7 +309,7 @@ public:
virtual bool insert(T* data)
{
- if (data == NULL)
+ if (data == NULL || data->getBinIndex() != -1)
{
OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE BRANCH !!!" << llendl;
return false;
@@ -300,16 +322,11 @@ public:
if ((getElementCount() < gOctreeMaxCapacity && contains(data->getBinRadius()) ||
(data->getBinRadius() > getSize()[0] && parent && parent->getElementCount() >= gOctreeMaxCapacity)))
{ //it belongs here
-#if LL_OCTREE_PARANOIA_CHECK
- //if this is a redundant insertion, error out (should never happen)
- if (mData.find(data) != mData.end())
- {
- llwarns << "Redundant octree insertion detected. " << data << llendl;
- return false;
- }
-#endif
-
- mData.insert(data);
+ mData.push_back(NULL);
+ mData[mElementCount] = data;
+ mElementCount++;
+ mDataEnd = &mData[mElementCount];
+ data->setBinIndex(mElementCount-1);
BaseType::insert(data);
return true;
}
@@ -344,7 +361,11 @@ public:
if( lt == 0x7 )
{
- mData.insert(data);
+ mData.push_back(NULL);
+ mData[mElementCount] = data;
+ mElementCount++;
+ mDataEnd = &mData[mElementCount];
+ data->setBinIndex(mElementCount-1);
BaseType::insert(data);
return true;
}
@@ -394,22 +415,58 @@ public:
return false;
}
+ void _remove(T* data, S32 i)
+ { //precondition -- mElementCount > 0, idx is in range [0, mElementCount)
+
+ mElementCount--;
+ data->setBinIndex(-1);
+
+ if (mElementCount > 0)
+ {
+ if (mElementCount != i)
+ {
+ mData[i] = mData[mElementCount]; //might unref data, do not access data after this point
+ mData[i]->setBinIndex(i);
+ }
+
+ mData[mElementCount] = NULL;
+ mData.pop_back();
+ mDataEnd = &mData[mElementCount];
+ }
+ else
+ {
+ mData.clear();
+ mData.push_back(NULL);
+ mDataEnd = &mData[0];
+ }
+
+ notifyRemoval(data);
+ checkAlive();
+ }
+
bool remove(T* data)
{
- if (mData.find(data) != mData.end())
- { //we have data
- mData.erase(data);
- notifyRemoval(data);
- checkAlive();
- return true;
- }
- else if (isInside(data))
+ S32 i = data->getBinIndex();
+
+ if (i >= 0 && i < mElementCount)
+ {
+ if (mData[i] == data)
+ { //found it
+ _remove(data, i);
+ llassert(data->getBinIndex() == -1);
+ return true;
+ }
+ }
+
+ if (isInside(data))
{
oct_node* dest = getNodeAt(data);
if (dest != this)
{
- return dest->remove(data);
+ bool ret = dest->remove(data);
+ llassert(data->getBinIndex() == -1);
+ return ret;
}
}
@@ -426,20 +483,22 @@ public:
}
//node is now root
- llwarns << "!!! OCTREE REMOVING FACE BY ADDRESS, SEVERE PERFORMANCE PENALTY |||" << llendl;
+ llwarns << "!!! OCTREE REMOVING ELEMENT BY ADDRESS, SEVERE PERFORMANCE PENALTY |||" << llendl;
node->removeByAddress(data);
+ llassert(data->getBinIndex() == -1);
return true;
}
void removeByAddress(T* data)
{
- if (mData.find(data) != mData.end())
+ for (U32 i = 0; i < mElementCount; ++i)
{
- mData.erase(data);
- notifyRemoval(data);
- llwarns << "FOUND!" << llendl;
- checkAlive();
- return;
+ if (mData[i] == data)
+ { //we have data
+ _remove(data, i);
+ llwarns << "FOUND!" << llendl;
+ return;
+ }
}
for (U32 i = 0; i < getChildCount(); i++)
@@ -451,7 +510,7 @@ public:
void clearChildren()
{
- mChild.clear();
+ mChildCount = 0;
U32* foo = (U32*) mChildMap;
foo[0] = foo[1] = 0xFFFFFFFF;
@@ -512,9 +571,10 @@ public:
}
#endif
- mChildMap[child->getOctant()] = (U8) mChild.size();
+ mChildMap[child->getOctant()] = mChildCount;
- mChild.push_back(child);
+ mChild[mChildCount] = child;
+ ++mChildCount;
child->setParent(this);
if (!silent)
@@ -534,21 +594,23 @@ public:
oct_listener* listener = getOctListener(i);
listener->handleChildRemoval(this, getChild(index));
}
-
-
if (destroy)
{
mChild[index]->destroy();
delete mChild[index];
}
- mChild.erase(mChild.begin() + index);
+
+ --mChildCount;
+
+ mChild[index] = mChild[mChildCount];
+
//rebuild child map
U32* foo = (U32*) mChildMap;
foo[0] = foo[1] = 0xFFFFFFFF;
- for (U32 i = 0; i < mChild.size(); ++i)
+ for (U32 i = 0; i < mChildCount; ++i)
{
mChildMap[mChild[i]->getOctant()] = i;
}
@@ -599,10 +661,13 @@ protected:
oct_node* mParent;
U8 mOctant;
- child_list mChild;
+ LLOctreeNode<T>* mChild[8];
U8 mChildMap[8];
+ U32 mChildCount;
element_list mData;
+ element_iter mDataEnd;
+ U32 mElementCount;
};
diff --git a/indra/llmath/llplane.h b/indra/llmath/llplane.h
index a611894721..3c32441b11 100644
--- a/indra/llmath/llplane.h
+++ b/indra/llmath/llplane.h
@@ -36,6 +36,8 @@
// The plane normal = [A, B, C]
// The closest approach = D / sqrt(A*A + B*B + C*C)
+
+LL_ALIGN_PREFIX(16)
class LLPlane
{
public:
@@ -94,7 +96,7 @@ public:
private:
LLVector4a mV;
-};
+} LL_ALIGN_POSTFIX(16);
diff --git a/indra/llmath/llsimdmath.h b/indra/llmath/llsimdmath.h
index c7cdf7b32c..01458521ec 100644
--- a/indra/llmath/llsimdmath.h
+++ b/indra/llmath/llsimdmath.h
@@ -67,11 +67,10 @@ template <typename T> T* LL_NEXT_ALIGNED_ADDRESS_64(T* address)
#define LL_ALIGN_16(var) LL_ALIGN_PREFIX(16) var LL_ALIGN_POSTFIX(16)
-
-
#include <xmmintrin.h>
#include <emmintrin.h>
+#include "llmemory.h"
#include "llsimdtypes.h"
#include "llsimdtypes.inl"
diff --git a/indra/llmath/llsimdtypes.inl b/indra/llmath/llsimdtypes.inl
index 712239e425..e905c84954 100644
--- a/indra/llmath/llsimdtypes.inl
+++ b/indra/llmath/llsimdtypes.inl
@@ -62,6 +62,7 @@ inline LLSimdScalar operator/(const LLSimdScalar& a, const LLSimdScalar& b)
inline LLSimdScalar operator-(const LLSimdScalar& a)
{
static LL_ALIGN_16(const U32 signMask[4]) = {0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+ ll_assert_aligned(signMask,16);
return _mm_xor_ps(*reinterpret_cast<const LLQuad*>(signMask), a);
}
@@ -146,6 +147,7 @@ inline LLSimdScalar& LLSimdScalar::operator/=(const LLSimdScalar& rhs)
inline LLSimdScalar LLSimdScalar::getAbs() const
{
static const LL_ALIGN_16(U32 F_ABS_MASK_4A[4]) = { 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF };
+ ll_assert_aligned(F_ABS_MASK_4A,16);
return _mm_and_ps( mQ, *reinterpret_cast<const LLQuad*>(F_ABS_MASK_4A));
}
diff --git a/indra/llmath/llvector4a.cpp b/indra/llmath/llvector4a.cpp
index b66b7a7076..6edeb0fefe 100644
--- a/indra/llmath/llvector4a.cpp
+++ b/indra/llmath/llvector4a.cpp
@@ -24,6 +24,7 @@
* $/LicenseInfo$
*/
+#include "llmemory.h"
#include "llmath.h"
#include "llquantize.h"
@@ -44,7 +45,10 @@ extern const LLVector4a LL_V4A_EPSILON = reinterpret_cast<const LLVector4a&> ( F
assert(dst != NULL);
assert(bytes > 0);
assert((bytes % sizeof(F32))== 0);
-
+ ll_assert_aligned(src,16);
+ ll_assert_aligned(dst,16);
+ assert(bytes%16==0);
+
F32* end = dst + (bytes / sizeof(F32) );
if (bytes > 64)
@@ -189,6 +193,8 @@ void LLVector4a::quantize16( const LLVector4a& low, const LLVector4a& high )
LLVector4a oneOverDelta;
{
static LL_ALIGN_16( const F32 F_TWO_4A[4] ) = { 2.f, 2.f, 2.f, 2.f };
+ ll_assert_aligned(F_TWO_4A,16);
+
LLVector4a two; two.load4a( F_TWO_4A );
// Here we use _mm_rcp_ps plus one round of newton-raphson
diff --git a/indra/llmath/llvector4a.h b/indra/llmath/llvector4a.h
index 596082509d..0526793d3a 100644
--- a/indra/llmath/llvector4a.h
+++ b/indra/llmath/llvector4a.h
@@ -32,6 +32,7 @@ class LLRotation;
#include <assert.h>
#include "llpreprocessor.h"
+#include "llmemory.h"
///////////////////////////////////
// FIRST TIME USERS PLEASE READ
@@ -46,6 +47,7 @@ class LLRotation;
// LLVector3/LLVector4.
/////////////////////////////////
+LL_ALIGN_PREFIX(16)
class LLVector4a
{
public:
@@ -82,6 +84,7 @@ public:
}
// Copy words 16-byte blocks from src to dst. Source and destination must not overlap.
+ // Source and dest must be 16-byte aligned and size must be multiple of 16.
static void memcpyNonAliased16(F32* __restrict dst, const F32* __restrict src, size_t bytes);
////////////////////////////////////
@@ -90,6 +93,7 @@ public:
LLVector4a()
{ //DO NOT INITIALIZE -- The overhead is completely unnecessary
+ ll_assert_aligned(this,16);
}
LLVector4a(F32 x, F32 y, F32 z, F32 w = 0.f)
@@ -313,7 +317,7 @@ public:
private:
LLQuad mQ;
-};
+} LL_ALIGN_POSTFIX(16);
inline void update_min_max(LLVector4a& min, LLVector4a& max, const LLVector4a& p)
{
diff --git a/indra/llmath/llvector4a.inl b/indra/llmath/llvector4a.inl
index 7ad22a5631..7c52ffef21 100644
--- a/indra/llmath/llvector4a.inl
+++ b/indra/llmath/llvector4a.inl
@@ -475,6 +475,7 @@ inline void LLVector4a::setLerp(const LLVector4a& lhs, const LLVector4a& rhs, F3
inline LLBool32 LLVector4a::isFinite3() const
{
static LL_ALIGN_16(const U32 nanOrInfMask[4]) = { 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 };
+ ll_assert_aligned(nanOrInfMask,16);
const __m128i nanOrInfMaskV = *reinterpret_cast<const __m128i*> (nanOrInfMask);
const __m128i maskResult = _mm_and_si128( _mm_castps_si128(mQ), nanOrInfMaskV );
const LLVector4Logical equalityCheck = _mm_castsi128_ps(_mm_cmpeq_epi32( maskResult, nanOrInfMaskV ));
diff --git a/indra/llmath/llvector4logical.h b/indra/llmath/llvector4logical.h
index dd66b09d43..c5698f7cea 100644
--- a/indra/llmath/llvector4logical.h
+++ b/indra/llmath/llvector4logical.h
@@ -27,6 +27,7 @@
#ifndef LL_VECTOR4LOGICAL_H
#define LL_VECTOR4LOGICAL_H
+#include "llmemory.h"
////////////////////////////
// LLVector4Logical
@@ -77,6 +78,7 @@ public:
inline LLVector4Logical& invert()
{
static const LL_ALIGN_16(U32 allOnes[4]) = { 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF };
+ ll_assert_aligned(allOnes,16);
mQ = _mm_andnot_ps( mQ, *(LLQuad*)(allOnes) );
return *this;
}
diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp
index da0fa32963..02c8d2b86f 100644
--- a/indra/llmath/llvolume.cpp
+++ b/indra/llmath/llvolume.cpp
@@ -35,7 +35,6 @@
#include <cmath>
#include "llerror.h"
-#include "llmemtype.h"
#include "llvolumemgr.h"
#include "v2math.h"
@@ -95,17 +94,6 @@ const S32 SCULPT_MIN_AREA_DETAIL = 1;
extern BOOL gDebugGL;
-void assert_aligned(void* ptr, uintptr_t alignment)
-{
-#if 0
- uintptr_t t = (uintptr_t) ptr;
- if (t%alignment != 0)
- {
- llerrs << "Alignment check failed." << llendl;
- }
-#endif
-}
-
BOOL check_same_clock_dir( const LLVector3& pt1, const LLVector3& pt2, const LLVector3& pt3, const LLVector3& norm)
{
LLVector3 test = (pt2-pt1)%(pt3-pt2);
@@ -328,16 +316,16 @@ public:
LLVector4a& min = node->mExtents[0];
LLVector4a& max = node->mExtents[1];
- if (!branch->getData().empty())
+ if (!branch->isEmpty())
{ //node has data, find AABB that binds data set
- const LLVolumeTriangle* tri = *(branch->getData().begin());
+ const LLVolumeTriangle* tri = *(branch->getDataBegin());
//initialize min/max to first available vertex
min = *(tri->mV[0]);
max = *(tri->mV[0]);
for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter =
- branch->getData().begin(); iter != branch->getData().end(); ++iter)
+ branch->getDataBegin(); iter != branch->getDataEnd(); ++iter)
{ //for each triangle in node
//stretch by triangles in node
@@ -352,7 +340,7 @@ public:
max.setMax(max, *tri->mV[2]);
}
}
- else if (!branch->getChildren().empty())
+ else if (!branch->isLeaf())
{ //no data, but child nodes exist
LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(0)->getListener(0);
@@ -389,8 +377,6 @@ public:
LLProfile::Face* LLProfile::addCap(S16 faceID)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
Face *face = vector_append(mFaces, 1);
face->mIndex = 0;
@@ -403,8 +389,6 @@ LLProfile::Face* LLProfile::addCap(S16 faceID)
LLProfile::Face* LLProfile::addFace(S32 i, S32 count, F32 scaleU, S16 faceID, BOOL flat)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
Face *face = vector_append(mFaces, 1);
face->mIndex = i;
@@ -420,7 +404,6 @@ LLProfile::Face* LLProfile::addFace(S32 i, S32 count, F32 scaleU, S16 faceID, BO
//static
S32 LLProfile::getNumNGonPoints(const LLProfileParams& params, S32 sides, F32 offset, F32 bevel, F32 ang_scale, S32 split)
{ // this is basically LLProfile::genNGon stripped down to only the operations that influence the number of points
- LLMemType m1(LLMemType::MTYPE_VOLUME);
S32 np = 0;
// Generate an n-sided "circular" path.
@@ -486,8 +469,6 @@ S32 LLProfile::getNumNGonPoints(const LLProfileParams& params, S32 sides, F32 of
// filleted and chamfered corners
void LLProfile::genNGon(const LLProfileParams& params, S32 sides, F32 offset, F32 bevel, F32 ang_scale, S32 split)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
// Generate an n-sided "circular" path.
// 0 is (1,0), and we go counter-clockwise along a circular path from there.
const F32 tableScale[] = { 1, 1, 1, 0.5f, 0.707107f, 0.53f, 0.525f, 0.5f };
@@ -741,8 +722,6 @@ LLProfile::Face* LLProfile::addHole(const LLProfileParams& params, BOOL flat, F3
S32 LLProfile::getNumPoints(const LLProfileParams& params, BOOL path_open,F32 detail, S32 split,
BOOL is_sculpted, S32 sculpt_size)
{ // this is basically LLProfile::generate stripped down to only operations that influence the number of points
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
if (detail < MIN_LOD)
{
detail = MIN_LOD;
@@ -853,8 +832,6 @@ S32 LLProfile::getNumPoints(const LLProfileParams& params, BOOL path_open,F32 de
BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detail, S32 split,
BOOL is_sculpted, S32 sculpt_size)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
if ((!mDirty) && (!is_sculpted))
{
return FALSE;
@@ -1127,8 +1104,6 @@ BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detai
BOOL LLProfileParams::importFile(LLFILE *fp)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
const S32 BUFSIZE = 16384;
char buffer[BUFSIZE]; /* Flawfinder: ignore */
// *NOTE: changing the size or type of these buffers will require
@@ -1204,8 +1179,6 @@ BOOL LLProfileParams::exportFile(LLFILE *fp) const
BOOL LLProfileParams::importLegacyStream(std::istream& input_stream)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
const S32 BUFSIZE = 16384;
char buffer[BUFSIZE]; /* Flawfinder: ignore */
// *NOTE: changing the size or type of these buffers will require
@@ -1297,7 +1270,6 @@ bool LLProfileParams::fromLLSD(LLSD& sd)
void LLProfileParams::copyParams(const LLProfileParams &params)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
setCurveType(params.getCurveType());
setBegin(params.getBegin());
setEnd(params.getEnd());
@@ -1514,8 +1486,6 @@ const LLVector2 LLPathParams::getEndScale() const
S32 LLPath::getNumPoints(const LLPathParams& params, F32 detail)
{ // this is basically LLPath::generate stripped down to only the operations that influence the number of points
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
if (detail < MIN_LOD)
{
detail = MIN_LOD;
@@ -1565,8 +1535,6 @@ S32 LLPath::getNumPoints(const LLPathParams& params, F32 detail)
BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split,
BOOL is_sculpted, S32 sculpt_size)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
if ((!mDirty) && (!is_sculpted))
{
return FALSE;
@@ -1694,8 +1662,6 @@ BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split,
BOOL LLDynamicPath::generate(const LLPathParams& params, F32 detail, S32 split,
BOOL is_sculpted, S32 sculpt_size)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
mOpen = TRUE; // Draw end caps
if (getPathLength() == 0)
{
@@ -1717,8 +1683,6 @@ BOOL LLDynamicPath::generate(const LLPathParams& params, F32 detail, S32 split,
BOOL LLPathParams::importFile(LLFILE *fp)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
const S32 BUFSIZE = 16384;
char buffer[BUFSIZE]; /* Flawfinder: ignore */
// *NOTE: changing the size or type of these buffers will require
@@ -1863,8 +1827,6 @@ BOOL LLPathParams::exportFile(LLFILE *fp) const
BOOL LLPathParams::importLegacyStream(std::istream& input_stream)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
const S32 BUFSIZE = 16384;
char buffer[BUFSIZE]; /* Flawfinder: ignore */
// *NOTE: changing the size or type of these buffers will require
@@ -2072,12 +2034,11 @@ S32 LLVolume::sNumMeshPoints = 0;
LLVolume::LLVolume(const LLVolumeParams &params, const F32 detail, const BOOL generate_single_face, const BOOL is_unique)
: mParams(params)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
mUnique = is_unique;
mFaceMask = 0x0;
mDetail = detail;
mSculptLevel = -2;
+ mSurfaceArea = 1.f; //only calculated for sculpts, defaults to 1 for all other prims
mIsMeshAssetLoaded = FALSE;
mLODScaleBias.setVec(1,1,1);
mHullPoints = NULL;
@@ -2144,7 +2105,6 @@ LLVolume::~LLVolume()
BOOL LLVolume::generate()
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
llassert_always(mProfilep);
//Added 10.03.05 Dave Parks
@@ -2740,8 +2700,6 @@ S32 LLVolume::getNumFaces() const
void LLVolume::createVolumeFaces()
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
if (mGenerateSingleFace)
{
// do nothing
@@ -2903,7 +2861,7 @@ F32 LLVolume::sculptGetSurfaceArea()
// compute the area of the quad by taking the length of the cross product of the two triangles
LLVector3 cross1 = (p1 - p2) % (p1 - p3);
LLVector3 cross2 = (p4 - p2) % (p4 - p3);
- area += (cross1.magVec() + cross2.magVec()) / 2.0;
+ area += (cross1.magVec() + cross2.magVec()) / 2.f;
}
}
@@ -2913,8 +2871,6 @@ F32 LLVolume::sculptGetSurfaceArea()
// create placeholder shape
void LLVolume::sculptGeneratePlaceholder()
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
S32 sizeS = mPathp->mPath.size();
S32 sizeT = mProfilep->mProfile.size();
@@ -2951,9 +2907,6 @@ void LLVolume::sculptGenerateMapVertices(U16 sculpt_width, U16 sculpt_height, S8
BOOL sculpt_mirror = sculpt_type & LL_SCULPT_FLAG_MIRROR;
BOOL reverse_horizontal = (sculpt_invert ? !sculpt_mirror : sculpt_mirror); // XOR
-
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
S32 sizeS = mPathp->mPath.size();
S32 sizeT = mProfilep->mProfile.size();
@@ -3102,7 +3055,6 @@ void sculpt_calc_mesh_resolution(U16 width, U16 height, U8 type, F32 detail, S32
// sculpt replaces generate() for sculpted surfaces
void LLVolume::sculpt(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data, S32 sculpt_level)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
U8 sculpt_type = mParams.getSculptType();
BOOL data_is_empty = FALSE;
@@ -3144,6 +3096,8 @@ void LLVolume::sculpt(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components,
{
F32 area = sculptGetSurfaceArea();
+ mSurfaceArea = area;
+
const F32 SCULPT_MAX_AREA = 384.f;
if (area < SCULPT_MIN_AREA || area > SCULPT_MAX_AREA)
@@ -3237,7 +3191,6 @@ bool LLVolumeParams::operator<(const LLVolumeParams &params) const
void LLVolumeParams::copyParams(const LLVolumeParams &params)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
mProfileParams.copyParams(params.mProfileParams);
mPathParams.copyParams(params.mPathParams);
mSculptID = params.getSculptID();
@@ -3609,8 +3562,6 @@ bool LLVolumeParams::validate(U8 prof_curve, F32 prof_begin, F32 prof_end, F32 h
S32 *LLVolume::getTriangleIndices(U32 &num_indices) const
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
S32 expected_num_triangle_indices = getNumTriangleIndices();
if (expected_num_triangle_indices > MAX_VOLUME_TRIANGLE_INDICES)
{
@@ -4338,8 +4289,6 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices,
const LLMatrix3& norm_mat_in,
S32 face_mask)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
LLMatrix4a mat;
mat.loadu(mat_in);
@@ -4617,18 +4566,83 @@ S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& en
genBinormals(i);
}
- if (!face.mOctree)
- {
- face.createOctree();
- }
-
- //LLVector4a* p = (LLVector4a*) face.mPositions;
+ if (isUnique())
+ { //don't bother with an octree for flexi volumes
+ U32 tri_count = face.mNumIndices/3;
+
+ for (U32 j = 0; j < tri_count; ++j)
+ {
+ U16 idx0 = face.mIndices[j*3+0];
+ U16 idx1 = face.mIndices[j*3+1];
+ U16 idx2 = face.mIndices[j*3+2];
+
+ const LLVector4a& v0 = face.mPositions[idx0];
+ const LLVector4a& v1 = face.mPositions[idx1];
+ const LLVector4a& v2 = face.mPositions[idx2];
+
+ F32 a,b,t;
- LLOctreeTriangleRayIntersect intersect(start, dir, &face, &closest_t, intersection, tex_coord, normal, bi_normal);
- intersect.traverse(face.mOctree);
- if (intersect.mHitFace)
+ if (LLTriangleRayIntersect(v0, v1, v2,
+ start, dir, a, b, t))
+ {
+ if ((t >= 0.f) && // if hit is after start
+ (t <= 1.f) && // and before end
+ (t < closest_t)) // and this hit is closer
+ {
+ closest_t = t;
+ hit_face = i;
+
+ if (intersection != NULL)
+ {
+ LLVector4a intersect = dir;
+ intersect.mul(closest_t);
+ intersect.add(start);
+ intersection->set(intersect.getF32ptr());
+ }
+
+
+ if (tex_coord != NULL)
+ {
+ LLVector2* tc = (LLVector2*) face.mTexCoords;
+ *tex_coord = ((1.f - a - b) * tc[idx0] +
+ a * tc[idx1] +
+ b * tc[idx2]);
+
+ }
+
+ if (normal!= NULL)
+ {
+ LLVector4* norm = (LLVector4*) face.mNormals;
+
+ *normal = ((1.f - a - b) * LLVector3(norm[idx0]) +
+ a * LLVector3(norm[idx1]) +
+ b * LLVector3(norm[idx2]));
+ }
+
+ if (bi_normal != NULL)
+ {
+ LLVector4* binormal = (LLVector4*) face.mBinormals;
+ *bi_normal = ((1.f - a - b) * LLVector3(binormal[idx0]) +
+ a * LLVector3(binormal[idx1]) +
+ b * LLVector3(binormal[idx2]));
+ }
+ }
+ }
+ }
+ }
+ else
{
- hit_face = i;
+ if (!face.mOctree)
+ {
+ face.createOctree();
+ }
+
+ LLOctreeTriangleRayIntersect intersect(start, dir, &face, &closest_t, intersection, tex_coord, normal, bi_normal);
+ intersect.traverse(face.mOctree);
+ if (intersect.mHitFace)
+ {
+ hit_face = i;
+ }
}
}
}
@@ -4736,241 +4750,8 @@ BOOL equalTriangle(const S32 *a, const S32 *b)
return FALSE;
}
-BOOL LLVolume::cleanupTriangleData( const S32 num_input_vertices,
- const std::vector<Point>& input_vertices,
- const S32 num_input_triangles,
- S32 *input_triangles,
- S32 &num_output_vertices,
- LLVector3 **output_vertices,
- S32 &num_output_triangles,
- S32 **output_triangles)
-{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
- /* Testing: avoid any cleanup
- static BOOL skip_cleanup = TRUE;
- if ( skip_cleanup )
- {
- num_output_vertices = num_input_vertices;
- num_output_triangles = num_input_triangles;
-
- *output_vertices = new LLVector3[num_input_vertices];
- for (S32 index = 0; index < num_input_vertices; index++)
- {
- (*output_vertices)[index] = input_vertices[index].mPos;
- }
-
- *output_triangles = new S32[num_input_triangles*3];
- memcpy(*output_triangles, input_triangles, 3*num_input_triangles*sizeof(S32)); // Flawfinder: ignore
- return TRUE;
- }
- */
-
- // Here's how we do this:
- // Create a structure which contains the original vertex index and the
- // LLVector3 data.
- // "Sort" the data by the vectors
- // Create an array the size of the old vertex list, with a mapping of
- // old indices to new indices.
- // Go through triangles, shift so the lowest index is first
- // Sort triangles by first index
- // Remove duplicate triangles
- // Allocate and pack new triangle data.
-
- //LLTimer cleanupTimer;
- //llinfos << "In vertices: " << num_input_vertices << llendl;
- //llinfos << "In triangles: " << num_input_triangles << llendl;
-
- S32 i;
- typedef std::multiset<LLVertexIndexPair*, lessVertex> vertex_set_t;
- vertex_set_t vertex_list;
-
- LLVertexIndexPair *pairp = NULL;
- for (i = 0; i < num_input_vertices; i++)
- {
- LLVertexIndexPair *new_pairp = new LLVertexIndexPair(input_vertices[i].mPos, i);
- vertex_list.insert(new_pairp);
- }
-
- // Generate the vertex mapping and the list of vertices without
- // duplicates. This will crash if there are no vertices.
- llassert(num_input_vertices > 0); // check for no vertices!
- S32 *vertex_mapping = new S32[num_input_vertices];
- LLVector3 *new_vertices = new LLVector3[num_input_vertices];
- LLVertexIndexPair *prev_pairp = NULL;
-
- S32 new_num_vertices;
-
- new_num_vertices = 0;
- for (vertex_set_t::iterator iter = vertex_list.begin(),
- end = vertex_list.end();
- iter != end; iter++)
- {
- pairp = *iter;
- if (!prev_pairp || ((pairp->mVertex - prev_pairp->mVertex).magVecSquared() >= VERTEX_SLOP_SQRD))
- {
- new_vertices[new_num_vertices] = pairp->mVertex;
- //llinfos << "Added vertex " << new_num_vertices << " : " << pairp->mVertex << llendl;
- new_num_vertices++;
- // Update the previous
- prev_pairp = pairp;
- }
- else
- {
- //llinfos << "Removed duplicate vertex " << pairp->mVertex << ", distance magVecSquared() is " << (pairp->mVertex - prev_pairp->mVertex).magVecSquared() << llendl;
- }
- vertex_mapping[pairp->mIndex] = new_num_vertices - 1;
- }
-
- // Iterate through triangles and remove degenerates, re-ordering vertices
- // along the way.
- S32 *new_triangles = new S32[num_input_triangles * 3];
- S32 new_num_triangles = 0;
-
- for (i = 0; i < num_input_triangles; i++)
- {
- S32 v1 = i*3;
- S32 v2 = v1 + 1;
- S32 v3 = v1 + 2;
-
- //llinfos << "Checking triangle " << input_triangles[v1] << ":" << input_triangles[v2] << ":" << input_triangles[v3] << llendl;
- input_triangles[v1] = vertex_mapping[input_triangles[v1]];
- input_triangles[v2] = vertex_mapping[input_triangles[v2]];
- input_triangles[v3] = vertex_mapping[input_triangles[v3]];
-
- if ((input_triangles[v1] == input_triangles[v2])
- || (input_triangles[v1] == input_triangles[v3])
- || (input_triangles[v2] == input_triangles[v3]))
- {
- //llinfos << "Removing degenerate triangle " << input_triangles[v1] << ":" << input_triangles[v2] << ":" << input_triangles[v3] << llendl;
- // Degenerate triangle, skip
- continue;
- }
-
- if (input_triangles[v1] < input_triangles[v2])
- {
- if (input_triangles[v1] < input_triangles[v3])
- {
- // (0 < 1) && (0 < 2)
- new_triangles[new_num_triangles*3] = input_triangles[v1];
- new_triangles[new_num_triangles*3+1] = input_triangles[v2];
- new_triangles[new_num_triangles*3+2] = input_triangles[v3];
- }
- else
- {
- // (0 < 1) && (2 < 0)
- new_triangles[new_num_triangles*3] = input_triangles[v3];
- new_triangles[new_num_triangles*3+1] = input_triangles[v1];
- new_triangles[new_num_triangles*3+2] = input_triangles[v2];
- }
- }
- else if (input_triangles[v2] < input_triangles[v3])
- {
- // (1 < 0) && (1 < 2)
- new_triangles[new_num_triangles*3] = input_triangles[v2];
- new_triangles[new_num_triangles*3+1] = input_triangles[v3];
- new_triangles[new_num_triangles*3+2] = input_triangles[v1];
- }
- else
- {
- // (1 < 0) && (2 < 1)
- new_triangles[new_num_triangles*3] = input_triangles[v3];
- new_triangles[new_num_triangles*3+1] = input_triangles[v1];
- new_triangles[new_num_triangles*3+2] = input_triangles[v2];
- }
- new_num_triangles++;
- }
-
- if (new_num_triangles == 0)
- {
- llwarns << "Created volume object with 0 faces." << llendl;
- delete[] new_triangles;
- delete[] vertex_mapping;
- delete[] new_vertices;
- return FALSE;
- }
-
- typedef std::set<S32*, lessTriangle> triangle_set_t;
- triangle_set_t triangle_list;
-
- for (i = 0; i < new_num_triangles; i++)
- {
- triangle_list.insert(&new_triangles[i*3]);
- }
-
- // Sort through the triangle list, and delete duplicates
-
- S32 *prevp = NULL;
- S32 *curp = NULL;
-
- S32 *sorted_tris = new S32[new_num_triangles*3];
- S32 cur_tri = 0;
- for (triangle_set_t::iterator iter = triangle_list.begin(),
- end = triangle_list.end();
- iter != end; iter++)
- {
- curp = *iter;
- if (!prevp || !equalTriangle(prevp, curp))
- {
- //llinfos << "Added triangle " << *curp << ":" << *(curp+1) << ":" << *(curp+2) << llendl;
- sorted_tris[cur_tri*3] = *curp;
- sorted_tris[cur_tri*3+1] = *(curp+1);
- sorted_tris[cur_tri*3+2] = *(curp+2);
- cur_tri++;
- prevp = curp;
- }
- else
- {
- //llinfos << "Skipped triangle " << *curp << ":" << *(curp+1) << ":" << *(curp+2) << llendl;
- }
- }
-
- *output_vertices = new LLVector3[new_num_vertices];
- num_output_vertices = new_num_vertices;
- for (i = 0; i < new_num_vertices; i++)
- {
- (*output_vertices)[i] = new_vertices[i];
- }
-
- *output_triangles = new S32[cur_tri*3];
- num_output_triangles = cur_tri;
- memcpy(*output_triangles, sorted_tris, 3*cur_tri*sizeof(S32)); /* Flawfinder: ignore */
-
- /*
- llinfos << "Out vertices: " << num_output_vertices << llendl;
- llinfos << "Out triangles: " << num_output_triangles << llendl;
- for (i = 0; i < num_output_vertices; i++)
- {
- llinfos << i << ":" << (*output_vertices)[i] << llendl;
- }
- for (i = 0; i < num_output_triangles; i++)
- {
- llinfos << i << ":" << (*output_triangles)[i*3] << ":" << (*output_triangles)[i*3+1] << ":" << (*output_triangles)[i*3+2] << llendl;
- }
- */
-
- //llinfos << "Out vertices: " << num_output_vertices << llendl;
- //llinfos << "Out triangles: " << num_output_triangles << llendl;
- delete[] vertex_mapping;
- vertex_mapping = NULL;
- delete[] new_vertices;
- new_vertices = NULL;
- delete[] new_triangles;
- new_triangles = NULL;
- delete[] sorted_tris;
- sorted_tris = NULL;
- triangle_list.clear();
- std::for_each(vertex_list.begin(), vertex_list.end(), DeletePointer());
- vertex_list.clear();
-
- return TRUE;
-}
-
-
BOOL LLVolumeParams::importFile(LLFILE *fp)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
//llinfos << "importing volume" << llendl;
const S32 BUFSIZE = 16384;
char buffer[BUFSIZE]; /* Flawfinder: ignore */
@@ -5025,8 +4806,6 @@ BOOL LLVolumeParams::exportFile(LLFILE *fp) const
BOOL LLVolumeParams::importLegacyStream(std::istream& input_stream)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
//llinfos << "importing volume" << llendl;
const S32 BUFSIZE = 16384;
// *NOTE: changing the size or type of this buffer will require
@@ -5066,8 +4845,6 @@ BOOL LLVolumeParams::importLegacyStream(std::istream& input_stream)
BOOL LLVolumeParams::exportLegacyStream(std::ostream& output_stream) const
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
output_stream <<"\tshape 0\n";
output_stream <<"\t{\n";
mPathParams.exportLegacyStream(output_stream);
@@ -5822,7 +5599,7 @@ F32 find_vertex_score(LLVCacheVertexData& data)
}
//bonus points for having low valence
- F32 valence_boost = powf(data.mActiveTriangles, -FindVertexScore_ValenceBoostPower);
+ F32 valence_boost = powf((F32)data.mActiveTriangles, -FindVertexScore_ValenceBoostPower);
score += FindVertexScore_ValenceBoostScale * valence_boost;
return score;
@@ -6283,8 +6060,6 @@ void LerpPlanarVertex(LLVolumeFace::VertexData& v0,
BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
const std::vector<LLVolume::Point>& mesh = volume->getMesh();
const std::vector<LLVector3>& profile = volume->getProfile().mProfile;
S32 max_s = volume->getProfile().getTotal();
@@ -6435,8 +6210,6 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build)
BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
if (!(mTypeMask & HOLLOW_MASK) &&
!(mTypeMask & OPEN_MASK) &&
((volume->getParams().getPathParams().getBegin()==0.0f)&&
@@ -6823,8 +6596,6 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build)
void LLVolumeFace::createBinormals()
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
if (!mBinormals)
{
allocateBinormals(mNumVertices);
@@ -6894,14 +6665,14 @@ void LLVolumeFace::resizeVertices(S32 num_verts)
if (num_verts)
{
mPositions = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts);
- assert_aligned(mPositions, 16);
+ ll_assert_aligned(mPositions, 16);
mNormals = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts);
- assert_aligned(mNormals, 16);
+ ll_assert_aligned(mNormals, 16);
//pad texture coordinate block end to allow for QWORD reads
S32 size = ((num_verts*sizeof(LLVector2)) + 0xF) & ~0xF;
mTexCoords = (LLVector2*) ll_aligned_malloc_16(size);
- assert_aligned(mTexCoords, 16);
+ ll_assert_aligned(mTexCoords, 16);
}
else
{
@@ -6922,17 +6693,21 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con
{
S32 new_verts = mNumVertices+1;
S32 new_size = new_verts*16;
-// S32 old_size = mNumVertices*16;
+ S32 old_size = mNumVertices*16;
//positions
- mPositions = (LLVector4a*) realloc(mPositions, new_size);
+ mPositions = (LLVector4a*) ll_aligned_realloc_16(mPositions, new_size, old_size);
+ ll_assert_aligned(mPositions,16);
//normals
- mNormals = (LLVector4a*) realloc(mNormals, new_size);
-
+ mNormals = (LLVector4a*) ll_aligned_realloc_16(mNormals, new_size, old_size);
+ ll_assert_aligned(mNormals,16);
+
//tex coords
new_size = ((new_verts*8)+0xF) & ~0xF;
- mTexCoords = (LLVector2*) realloc(mTexCoords, new_size);
+ old_size = ((mNumVertices*8)+0xF) & ~0xF;
+ mTexCoords = (LLVector2*) ll_aligned_realloc_16(mTexCoords, new_size, old_size);
+ ll_assert_aligned(mTexCoords,16);
//just clear binormals
@@ -6985,7 +6760,8 @@ void LLVolumeFace::pushIndex(const U16& idx)
S32 old_size = ((mNumIndices*2)+0xF) & ~0xF;
if (new_size != old_size)
{
- mIndices = (U16*) realloc(mIndices, new_size);
+ mIndices = (U16*) ll_aligned_realloc_16(mIndices, new_size, old_size);
+ ll_assert_aligned(mIndices,16);
}
mIndices[mNumIndices++] = idx;
@@ -7026,12 +6802,12 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat
}
//allocate new buffer space
- mPositions = (LLVector4a*) realloc(mPositions, new_count*sizeof(LLVector4a));
- assert_aligned(mPositions, 16);
- mNormals = (LLVector4a*) realloc(mNormals, new_count*sizeof(LLVector4a));
- assert_aligned(mNormals, 16);
- mTexCoords = (LLVector2*) realloc(mTexCoords, (new_count*sizeof(LLVector2)+0xF) & ~0xF);
- assert_aligned(mTexCoords, 16);
+ mPositions = (LLVector4a*) ll_aligned_realloc_16(mPositions, new_count*sizeof(LLVector4a), mNumVertices*sizeof(LLVector4a));
+ ll_assert_aligned(mPositions, 16);
+ mNormals = (LLVector4a*) ll_aligned_realloc_16(mNormals, new_count*sizeof(LLVector4a), mNumVertices*sizeof(LLVector4a));
+ ll_assert_aligned(mNormals, 16);
+ mTexCoords = (LLVector2*) ll_aligned_realloc_16(mTexCoords, (new_count*sizeof(LLVector2)+0xF) & ~0xF, (mNumVertices*sizeof(LLVector2)+0xF) & ~0xF);
+ ll_assert_aligned(mTexCoords, 16);
mNumVertices = new_count;
@@ -7077,7 +6853,7 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat
new_count = mNumIndices + face.mNumIndices;
//allocate new index buffer
- mIndices = (U16*) realloc(mIndices, (new_count*sizeof(U16)+0xF) & ~0xF);
+ mIndices = (U16*) ll_aligned_realloc_16(mIndices, (new_count*sizeof(U16)+0xF) & ~0xF, (mNumIndices*sizeof(U16)+0xF) & ~0xF);
//get destination address into new index buffer
U16* dst_idx = mIndices+mNumIndices;
@@ -7091,8 +6867,6 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat
BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
-
BOOL flat = mTypeMask & FLAT_MASK;
U8 sculpt_type = volume->getParams().getSculptType();
diff --git a/indra/llmath/llvolume.h b/indra/llmath/llvolume.h
index afd1ec5eed..c845556557 100644
--- a/indra/llmath/llvolume.h
+++ b/indra/llmath/llvolume.h
@@ -54,6 +54,7 @@ class LLVolumeTriangle;
#include "llstrider.h"
#include "v4coloru.h"
#include "llrefcount.h"
+#include "llpointer.h"
#include "llfile.h"
//============================================================================
@@ -919,6 +920,10 @@ public:
LLVector2* mTexCoords;
U16* mIndices;
+ //vertex buffer filled in by LLFace to cache this volume face geometry in vram
+ // (declared as a LLPointer to LLRefCount to avoid dependency on LLVertexBuffer)
+ mutable LLPointer<LLRefCount> mVertexBuffer;
+
std::vector<S32> mEdge;
//list of skin weights for rigged volumes
@@ -963,6 +968,7 @@ public:
S32 getNumFaces() const;
S32 getNumVolumeFaces() const { return mVolumeFaces.size(); }
F32 getDetail() const { return mDetail; }
+ F32 getSurfaceArea() const { return mSurfaceArea; }
const LLVolumeParams& getParams() const { return mParams; }
LLVolumeParams getCopyOfParams() const { return mParams; }
const LLProfile& getProfile() const { return *mProfilep; }
@@ -1017,17 +1023,6 @@ public:
LLVector3* normal = NULL,
LLVector3* bi_normal = NULL);
- // The following cleans up vertices and triangles,
- // getting rid of degenerate triangles and duplicate vertices,
- // and allocates new arrays with the clean data.
- static BOOL cleanupTriangleData( const S32 num_input_vertices,
- const std::vector<Point> &input_vertices,
- const S32 num_input_triangles,
- S32 *input_triangles,
- S32 &num_output_vertices,
- LLVector3 **output_vertices,
- S32 &num_output_triangles,
- S32 **output_triangles);
LLFaceID generateFaceMask();
BOOL isFaceMaskValid(LLFaceID face_mask);
@@ -1065,6 +1060,7 @@ public:
BOOL mUnique;
F32 mDetail;
S32 mSculptLevel;
+ F32 mSurfaceArea; //unscaled surface area
BOOL mIsMeshAssetLoaded;
LLVolumeParams mParams;
diff --git a/indra/llmath/llvolumemgr.cpp b/indra/llmath/llvolumemgr.cpp
index c60b750088..9083273ee5 100644
--- a/indra/llmath/llvolumemgr.cpp
+++ b/indra/llmath/llvolumemgr.cpp
@@ -26,7 +26,6 @@
#include "linden_common.h"
#include "llvolumemgr.h"
-#include "llmemtype.h"
#include "llvolume.h"
@@ -182,7 +181,6 @@ void LLVolumeMgr::insertGroup(LLVolumeLODGroup* volgroup)
// protected
LLVolumeLODGroup* LLVolumeMgr::createNewGroup(const LLVolumeParams& volume_params)
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
LLVolumeLODGroup* volgroup = new LLVolumeLODGroup(volume_params);
insertGroup(volgroup);
return volgroup;
@@ -297,7 +295,6 @@ LLVolume* LLVolumeLODGroup::refLOD(const S32 detail)
mRefs++;
if (mVolumeLODs[detail].isNull())
{
- LLMemType m1(LLMemType::MTYPE_VOLUME);
mVolumeLODs[detail] = new LLVolume(mVolumeParams, mDetailScales[detail]);
}
mLODRefs[detail]++;
diff --git a/indra/llmath/llvolumeoctree.cpp b/indra/llmath/llvolumeoctree.cpp
index b5a935c2b5..cc83cb7235 100644
--- a/indra/llmath/llvolumeoctree.cpp
+++ b/indra/llmath/llvolumeoctree.cpp
@@ -131,7 +131,7 @@ void LLOctreeTriangleRayIntersect::traverse(const LLOctreeNode<LLVolumeTriangle>
void LLOctreeTriangleRayIntersect::visit(const LLOctreeNode<LLVolumeTriangle>* node)
{
for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter =
- node->getData().begin(); iter != node->getData().end(); ++iter)
+ node->getDataBegin(); iter != node->getDataEnd(); ++iter)
{
const LLVolumeTriangle* tri = *iter;
@@ -236,8 +236,8 @@ void LLVolumeOctreeValidate::visit(const LLOctreeNode<LLVolumeTriangle>* branch)
}
//children fit, check data
- for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = branch->getData().begin();
- iter != branch->getData().end(); ++iter)
+ for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = branch->getDataBegin();
+ iter != branch->getDataEnd(); ++iter)
{
const LLVolumeTriangle* tri = *iter;
diff --git a/indra/llmath/llvolumeoctree.h b/indra/llmath/llvolumeoctree.h
index 688d91dc40..9ae34a0c4e 100644
--- a/indra/llmath/llvolumeoctree.h
+++ b/indra/llmath/llvolumeoctree.h
@@ -37,9 +37,19 @@
class LLVolumeTriangle : public LLRefCount
{
public:
+ void* operator new(size_t size)
+ {
+ return ll_aligned_malloc_16(size);
+ }
+
+ void operator delete(void* ptr)
+ {
+ ll_aligned_free_16(ptr);
+ }
+
LLVolumeTriangle()
{
-
+ mBinIndex = -1;
}
LLVolumeTriangle(const LLVolumeTriangle& rhs)
@@ -58,21 +68,38 @@ public:
}
- LLVector4a mPositionGroup;
+ LL_ALIGN_16(LLVector4a mPositionGroup);
const LLVector4a* mV[3];
U16 mIndex[3];
F32 mRadius;
+ mutable S32 mBinIndex;
+
virtual const LLVector4a& getPositionGroup() const;
virtual const F32& getBinRadius() const;
+
+ S32 getBinIndex() const { return mBinIndex; }
+ void setBinIndex(S32 idx) const { mBinIndex = idx; }
+
+
};
class LLVolumeOctreeListener : public LLOctreeListener<LLVolumeTriangle>
{
public:
+ void* operator new(size_t size)
+ {
+ return ll_aligned_malloc_16(size);
+ }
+
+ void operator delete(void* ptr)
+ {
+ ll_aligned_free_16(ptr);
+ }
+
LLVolumeOctreeListener(LLOctreeNode<LLVolumeTriangle>* node);
~LLVolumeOctreeListener();
@@ -99,8 +126,8 @@ public:
public:
- LLVector4a mBounds[2]; // bounding box (center, size) of this node and all its children (tight fit to objects)
- LLVector4a mExtents[2]; // extents (min, max) of this node and all its children
+ LL_ALIGN_16(LLVector4a mBounds[2]); // bounding box (center, size) of this node and all its children (tight fit to objects)
+ LL_ALIGN_16(LLVector4a mExtents[2]); // extents (min, max) of this node and all its children
};
class LLOctreeTriangleRayIntersect : public LLOctreeTraveler<LLVolumeTriangle>
diff --git a/indra/llmath/m4math.cpp b/indra/llmath/m4math.cpp
index bad4deb4de..6a1b4143cf 100644
--- a/indra/llmath/m4math.cpp
+++ b/indra/llmath/m4math.cpp
@@ -858,25 +858,25 @@ LLSD LLMatrix4::getValue() const
void LLMatrix4::setValue(const LLSD& data)
{
- mMatrix[0][0] = data[0].asReal();
- mMatrix[0][1] = data[1].asReal();
- mMatrix[0][2] = data[2].asReal();
- mMatrix[0][3] = data[3].asReal();
-
- mMatrix[1][0] = data[4].asReal();
- mMatrix[1][1] = data[5].asReal();
- mMatrix[1][2] = data[6].asReal();
- mMatrix[1][3] = data[7].asReal();
-
- mMatrix[2][0] = data[8].asReal();
- mMatrix[2][1] = data[9].asReal();
- mMatrix[2][2] = data[10].asReal();
- mMatrix[2][3] = data[11].asReal();
-
- mMatrix[3][0] = data[12].asReal();
- mMatrix[3][1] = data[13].asReal();
- mMatrix[3][2] = data[14].asReal();
- mMatrix[3][3] = data[15].asReal();
+ mMatrix[0][0] = (F32)data[0].asReal();
+ mMatrix[0][1] = (F32)data[1].asReal();
+ mMatrix[0][2] = (F32)data[2].asReal();
+ mMatrix[0][3] = (F32)data[3].asReal();
+
+ mMatrix[1][0] = (F32)data[4].asReal();
+ mMatrix[1][1] = (F32)data[5].asReal();
+ mMatrix[1][2] = (F32)data[6].asReal();
+ mMatrix[1][3] = (F32)data[7].asReal();
+
+ mMatrix[2][0] = (F32)data[8].asReal();
+ mMatrix[2][1] = (F32)data[9].asReal();
+ mMatrix[2][2] = (F32)data[10].asReal();
+ mMatrix[2][3] = (F32)data[11].asReal();
+
+ mMatrix[3][0] = (F32)data[12].asReal();
+ mMatrix[3][1] = (F32)data[13].asReal();
+ mMatrix[3][2] = (F32)data[14].asReal();
+ mMatrix[3][3] = (F32)data[15].asReal();
}
diff --git a/indra/llmath/tests/alignment_test.cpp b/indra/llmath/tests/alignment_test.cpp
new file mode 100644
index 0000000000..5ee3c45502
--- /dev/null
+++ b/indra/llmath/tests/alignment_test.cpp
@@ -0,0 +1,141 @@
+/**
+ * @file v3dmath_test.cpp
+ * @author Vir
+ * @date 2011-12
+ * @brief v3dmath test cases.
+ *
+ * $LicenseInfo:firstyear=2011&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2011, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA
+ * $/LicenseInfo$
+ */
+
+// Tests related to allocating objects with alignment constraints, particularly for SSE support.
+
+#include "linden_common.h"
+#include "../test/lltut.h"
+#include "../llmath.h"
+#include "../llsimdmath.h"
+#include "../llvector4a.h"
+
+namespace tut
+{
+
+#define is_aligned(ptr,alignment) ((reinterpret_cast<uintptr_t>(ptr))%(alignment)==0)
+#define is_aligned_relative(ptr,base_ptr,alignment) ((reinterpret_cast<uintptr_t>(ptr)-reinterpret_cast<uintptr_t>(base_ptr))%(alignment)==0)
+
+struct alignment_test {};
+
+typedef test_group<alignment_test> alignment_test_t;
+typedef alignment_test_t::object alignment_test_object_t;
+tut::alignment_test_t tut_alignment_test("LLAlignment");
+
+LL_ALIGN_PREFIX(16)
+class MyVector4a
+{
+public:
+ void* operator new(size_t size)
+ {
+ return ll_aligned_malloc_16(size);
+ }
+
+ void operator delete(void *p)
+ {
+ ll_aligned_free_16(p);
+ }
+
+ void* operator new[](size_t count)
+ { // try to allocate count bytes for an array
+ return ll_aligned_malloc_16(count);
+ }
+
+ void operator delete[](void *p)
+ {
+ ll_aligned_free_16(p);
+ }
+
+ LLQuad mQ;
+} LL_ALIGN_POSTFIX(16);
+
+
+// Verify that aligned allocators perform as advertised.
+template<> template<>
+void alignment_test_object_t::test<1>()
+{
+# ifdef LL_DEBUG
+// skip("This test fails on Windows when compiled in debug mode.");
+# endif
+
+ const int num_tests = 7;
+ void *align_ptr;
+ for (int i=0; i<num_tests; i++)
+ {
+ align_ptr = ll_aligned_malloc_16(sizeof(MyVector4a));
+ ensure("ll_aligned_malloc_16 failed", is_aligned(align_ptr,16));
+
+ align_ptr = ll_aligned_realloc_16(align_ptr,2*sizeof(MyVector4a), sizeof(MyVector4a));
+ ensure("ll_aligned_realloc_16 failed", is_aligned(align_ptr,16));
+
+ ll_aligned_free_16(align_ptr);
+
+ align_ptr = ll_aligned_malloc_32(sizeof(MyVector4a));
+ ensure("ll_aligned_malloc_32 failed", is_aligned(align_ptr,32));
+ ll_aligned_free_32(align_ptr);
+ }
+}
+
+// In-place allocation of objects and arrays.
+template<> template<>
+void alignment_test_object_t::test<2>()
+{
+ MyVector4a vec1;
+ ensure("LLAlignment vec1 unaligned", is_aligned(&vec1,16));
+
+ MyVector4a veca[12];
+ ensure("LLAlignment veca unaligned", is_aligned(veca,16));
+}
+
+// Heap allocation of objects and arrays.
+template<> template<>
+void alignment_test_object_t::test<3>()
+{
+# ifdef LL_DEBUG
+// skip("This test fails on Windows when compiled in debug mode.");
+# endif
+
+ const int ARR_SIZE = 7;
+ for(int i=0; i<ARR_SIZE; i++)
+ {
+ MyVector4a *vecp = new MyVector4a;
+ ensure("LLAlignment vecp unaligned", is_aligned(vecp,16));
+ delete vecp;
+ }
+
+ MyVector4a *veca = new MyVector4a[ARR_SIZE];
+ //std::cout << "veca base is " << (S32) veca << std::endl;
+ ensure("LLAligment veca base", is_aligned(veca,16));
+ for(int i=0; i<ARR_SIZE; i++)
+ {
+ std::cout << "veca[" << i << "]" << std::endl;
+ ensure("LLAlignment veca member unaligned", is_aligned(&veca[i],16));
+ }
+ delete [] veca;
+}
+
+}
diff --git a/indra/llmath/v3color.h b/indra/llmath/v3color.h
index 56cb2ae73e..daf3a6857b 100644
--- a/indra/llmath/v3color.h
+++ b/indra/llmath/v3color.h
@@ -33,6 +33,7 @@ class LLVector4;
#include "llerror.h"
#include "llmath.h"
#include "llsd.h"
+#include <string.h>
// LLColor3 = |r g b|
generated by cgit v1.2.3 (git 2.25.1) at 2025-06-04 02:23:29 +0000