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/llcamera.h12
-rw-r--r--indra/llmath/llmatrix3a.h2
-rw-r--r--indra/llmath/llmatrix4a.h2
-rw-r--r--indra/llmath/lloctree.h143
-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.cpp56
-rw-r--r--indra/llmath/llvolumeoctree.cpp6
-rw-r--r--indra/llmath/llvolumeoctree.h35
-rw-r--r--indra/llmath/tests/alignment_test.cpp33
16 files changed, 213 insertions, 103 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/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/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 1b11e83b4a..4ac1e55cfc 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 element_list::iterator element_iter;
- typedef typename element_list::const_iterator const_element_iter;
+ typedef std::vector<LLPointer<T> > element_list;
+ 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;
@@ -123,6 +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);
@@ -219,12 +234,17 @@ 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 mElementCount; }
+ bool isEmpty() const { return mElementCount == 0; }
element_list& getData() { return mData; }
const element_list& getData() const { return mData; }
-
+ 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]; }
@@ -289,7 +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;
@@ -302,13 +322,12 @@ public:
if ((getElementCount() < gOctreeMaxCapacity && contains(data->getBinRadius()) ||
(data->getBinRadius() > getSize()[0] && parent && parent->getElementCount() >= gOctreeMaxCapacity)))
{ //it belongs here
- //if this is a redundant insertion, error out (should never happen)
- llassert(mData.find(data) == mData.end());
-
- mData.insert(data);
+ mData.push_back(NULL);
+ mData[mElementCount] = data;
+ mElementCount++;
+ mDataEnd = &mData[mElementCount];
+ data->setBinIndex(mElementCount-1);
BaseType::insert(data);
-
- mElementCount = mData.size();
return true;
}
else
@@ -342,10 +361,12 @@ 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);
-
- mElementCount = mData.size();
return true;
}
@@ -394,23 +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);
- mElementCount = mData.size();
- 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;
}
}
@@ -427,21 +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);
- mElementCount = mData.size();
- 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++)
@@ -453,8 +510,8 @@ public:
void clearChildren()
{
- mChild.clear();
mChildCount = 0;
+
U32* foo = (U32*) mChildMap;
foo[0] = foo[1] = 0xFFFFFFFF;
}
@@ -516,7 +573,7 @@ public:
mChildMap[child->getOctant()] = mChildCount;
- mChild.push_back(child);
+ mChild[mChildCount] = child;
++mChildCount;
child->setParent(this);
@@ -543,9 +600,12 @@ public:
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;
@@ -601,11 +661,12 @@ 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 cc9744756f..85ea14f9bc 100644
--- a/indra/llmath/llvolume.cpp
+++ b/indra/llmath/llvolume.cpp
@@ -95,17 +95,6 @@ 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 +317,16 @@ 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 +341,7 @@ 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);
@@ -6962,14 +6951,14 @@ 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
{
@@ -6990,17 +6979,21 @@ 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
@@ -7053,7 +7046,8 @@ 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;
@@ -7094,12 +7088,12 @@ 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;
@@ -7145,7 +7139,7 @@ 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;
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/tests/alignment_test.cpp b/indra/llmath/tests/alignment_test.cpp
index ac0c45ae6f..9105b1c1fd 100644
--- a/indra/llmath/tests/alignment_test.cpp
+++ b/indra/llmath/tests/alignment_test.cpp
@@ -34,16 +34,6 @@
#include "../llsimdmath.h"
#include "../llvector4a.h"
-void* operator new(size_t size)
-{
- return ll_aligned_malloc_16(size);
-}
-
-void operator delete(void *p)
-{
- ll_aligned_free_16(p);
-}
-
namespace tut
{
@@ -59,6 +49,27 @@ 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);
@@ -78,7 +89,7 @@ void alignment_test_object_t::test<1>()
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));
+ 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);
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-24 20:15:54 +0000