From a766e26db46c7c054bae1021470dbe365f2a3cb3 Mon Sep 17 00:00:00 2001 From: William Todd Stinson Date: Mon, 10 Sep 2012 10:37:02 -0700 Subject: Backing out the changes contributing to DRTVWR-167 and DRTVWR-179 from the repository. --- indra/llmath/CMakeLists.txt | 1 - indra/llmath/llcamera.h | 12 ++-- indra/llmath/llmatrix3a.h | 2 +- indra/llmath/llmatrix4a.h | 2 +- indra/llmath/lloctree.h | 147 +++++++++++----------------------------- indra/llmath/llplane.h | 4 +- indra/llmath/llsimdmath.h | 3 +- indra/llmath/llsimdtypes.inl | 2 - indra/llmath/llvector4a.cpp | 8 +-- indra/llmath/llvector4a.h | 6 +- indra/llmath/llvector4a.inl | 1 - indra/llmath/llvector4logical.h | 2 - indra/llmath/llvolume.cpp | 53 ++++++++------- indra/llmath/llvolumeoctree.cpp | 6 +- indra/llmath/llvolumeoctree.h | 35 ++-------- 15 files changed, 89 insertions(+), 195 deletions(-) (limited to 'indra/llmath') diff --git a/indra/llmath/CMakeLists.txt b/indra/llmath/CMakeLists.txt index 5865ae030c..b5e59c1ca3 100644 --- a/indra/llmath/CMakeLists.txt +++ b/indra/llmath/CMakeLists.txt @@ -117,7 +117,6 @@ 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 0b591be622..ec67b91d05 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: - LL_ALIGN_16(LLPlane mAgentPlanes[7]); //frustum planes in agent space a la gluUnproject (I'm a bastard, I know) - DaveP + 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; - LL_ALIGN_16(LLPlane mLocalPlanes[4]); + 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; - LL_ALIGN_16(LLPlane mWorldPlanes[PLANE_NUM]); - LL_ALIGN_16(LLPlane mHorizPlanes[HORIZ_PLANE_NUM]); + LLPlane mWorldPlanes[PLANE_NUM]; + 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 9916cfd2da..adb7e3389d 100644 --- a/indra/llmath/llmatrix3a.h +++ b/indra/llmath/llmatrix3a.h @@ -111,7 +111,7 @@ public: protected: - LL_ALIGN_16(LLVector4a mColumns[3]); + LLVector4a mColumns[3]; }; diff --git a/indra/llmath/llmatrix4a.h b/indra/llmath/llmatrix4a.h index c4cefdb4fa..27cf5b79f6 100644 --- a/indra/llmath/llmatrix4a.h +++ b/indra/llmath/llmatrix4a.h @@ -34,7 +34,7 @@ class LLMatrix4a { public: - LL_ALIGN_16(LLVector4a mMatrix[4]); + LLVector4a mMatrix[4]; inline void clear() { diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h index c3f6f7de2a..1b11e83b4a 100644 --- a/indra/llmath/lloctree.h +++ b/indra/llmath/lloctree.h @@ -31,6 +31,7 @@ #include "v3math.h" #include "llvector4a.h" #include +#include #define OCT_ERRS LL_WARNS("OctreeErrors") @@ -78,18 +79,16 @@ public: typedef LLOctreeTraveler oct_traveler; typedef LLTreeTraveler tree_traveler; - typedef LLPointer* element_list; - typedef LLPointer* element_iter; - typedef const LLPointer* const_element_iter; + typedef typename std::set > element_list; + typedef typename element_list::iterator element_iter; + typedef typename element_list::const_iterator const_element_iter; typedef typename std::vector*>::iterator tree_listener_iter; - typedef LLOctreeNode** child_list; - typedef LLOctreeNode** child_iter; - + typedef typename std::vector* > child_list; typedef LLTreeNode BaseType; typedef LLOctreeNode oct_node; typedef LLOctreeListener oct_listener; - void* operator new(size_t size) + /*void* operator new(size_t size) { return ll_aligned_malloc_16(size); } @@ -97,7 +96,7 @@ public: void operator delete(void* ptr) { ll_aligned_free_16(ptr); - } + }*/ LLOctreeNode( const LLVector4a& center, const LLVector4a& size, @@ -106,9 +105,6 @@ public: : mParent((oct_node*)parent), mOctant(octant) { - mData = NULL; - mDataEnd = NULL; - mCenter = center; mSize = size; @@ -127,16 +123,6 @@ public: { BaseType::destroyListeners(); - for (U32 i = 0; i < mElementCount; ++i) - { - mData[i]->setBinIndex(-1); - mData[i] = NULL; - } - - free(mData); - mData = NULL; - mDataEnd = NULL; - for (U32 i = 0; i < getChildCount(); i++) { delete getChild(i); @@ -233,17 +219,12 @@ public: } void accept(oct_traveler* visitor) { visitor->visit(this); } - virtual bool isLeaf() const { return mChildCount == 0; } + virtual bool isLeaf() const { return mChild.empty(); } 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; } - element_iter getDataEnd() { return mDataEnd; } - const_element_iter getDataBegin() const { return mData; } - 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]; } @@ -308,7 +289,7 @@ public: virtual bool insert(T* data) { - if (data == NULL || data->getBinIndex() != -1) + if (data == NULL) { OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE BRANCH !!!" << llendl; return false; @@ -321,16 +302,13 @@ public: if ((getElementCount() < gOctreeMaxCapacity && contains(data->getBinRadius()) || (data->getBinRadius() > getSize()[0] && parent && parent->getElementCount() >= gOctreeMaxCapacity))) { //it belongs here - mElementCount++; - mData = (element_list) realloc(mData, sizeof(LLPointer)*mElementCount); - - //avoid unref on uninitialized memory - memset(mData+mElementCount-1, 0, sizeof(LLPointer)); + //if this is a redundant insertion, error out (should never happen) + llassert(mData.find(data) == mData.end()); - mData[mElementCount-1] = data; - mDataEnd = mData + mElementCount; - data->setBinIndex(mElementCount-1); + mData.insert(data); BaseType::insert(data); + + mElementCount = mData.size(); return true; } else @@ -364,16 +342,10 @@ public: if( lt == 0x7 ) { - mElementCount++; - mData = (element_list) realloc(mData, sizeof(LLPointer)*mElementCount); - - //avoid unref on uninitialized memory - memset(mData+mElementCount-1, 0, sizeof(LLPointer)); - - mData[mElementCount-1] = data; - mDataEnd = mData + mElementCount; - data->setBinIndex(mElementCount-1); + mData.insert(data); BaseType::insert(data); + + mElementCount = mData.size(); return true; } @@ -422,59 +394,23 @@ 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; //needed for unref - mData = (element_list) realloc(mData, sizeof(LLPointer)*mElementCount); - mDataEnd = mData+mElementCount; - } - else - { - mData[0] = NULL; //needed for unref - free(mData); - mData = NULL; - mDataEnd = NULL; - } - - notifyRemoval(data); - checkAlive(); - } - bool remove(T* 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)) + if (mData.find(data) != mData.end()) + { //we have data + mData.erase(data); + mElementCount = mData.size(); + notifyRemoval(data); + checkAlive(); + return true; + } + else if (isInside(data)) { oct_node* dest = getNodeAt(data); if (dest != this) { - bool ret = dest->remove(data); - llassert(data->getBinIndex() == -1); - return ret; + return dest->remove(data); } } @@ -493,20 +429,19 @@ public: //node is now root llwarns << "!!! OCTREE REMOVING FACE BY ADDRESS, SEVERE PERFORMANCE PENALTY |||" << llendl; node->removeByAddress(data); - llassert(data->getBinIndex() == -1); return true; } void removeByAddress(T* data) { - for (U32 i = 0; i < mElementCount; ++i) + if (mData.find(data) != mData.end()) { - if (mData[i] == data) - { //we have data - _remove(data, i); - llwarns << "FOUND!" << llendl; - return; - } + mData.erase(data); + mElementCount = mData.size(); + notifyRemoval(data); + llwarns << "FOUND!" << llendl; + checkAlive(); + return; } for (U32 i = 0; i < getChildCount(); i++) @@ -518,8 +453,8 @@ public: void clearChildren() { + mChild.clear(); mChildCount = 0; - U32* foo = (U32*) mChildMap; foo[0] = foo[1] = 0xFFFFFFFF; } @@ -581,7 +516,7 @@ public: mChildMap[child->getOctant()] = mChildCount; - mChild[mChildCount] = child; + mChild.push_back(child); ++mChildCount; child->setParent(this); @@ -608,12 +543,9 @@ 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; @@ -669,12 +601,11 @@ protected: oct_node* mParent; U8 mOctant; - LLOctreeNode* mChild[8]; + child_list mChild; 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 3c32441b11..a611894721 100644 --- a/indra/llmath/llplane.h +++ b/indra/llmath/llplane.h @@ -36,8 +36,6 @@ // The plane normal = [A, B, C] // The closest approach = D / sqrt(A*A + B*B + C*C) - -LL_ALIGN_PREFIX(16) class LLPlane { public: @@ -96,7 +94,7 @@ public: private: LLVector4a mV; -} LL_ALIGN_POSTFIX(16); +}; diff --git a/indra/llmath/llsimdmath.h b/indra/llmath/llsimdmath.h index 01458521ec..c7cdf7b32c 100644 --- a/indra/llmath/llsimdmath.h +++ b/indra/llmath/llsimdmath.h @@ -67,10 +67,11 @@ template T* LL_NEXT_ALIGNED_ADDRESS_64(T* address) #define LL_ALIGN_16(var) LL_ALIGN_PREFIX(16) var LL_ALIGN_POSTFIX(16) + + #include #include -#include "llmemory.h" #include "llsimdtypes.h" #include "llsimdtypes.inl" diff --git a/indra/llmath/llsimdtypes.inl b/indra/llmath/llsimdtypes.inl index e905c84954..712239e425 100644 --- a/indra/llmath/llsimdtypes.inl +++ b/indra/llmath/llsimdtypes.inl @@ -62,7 +62,6 @@ 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(signMask), a); } @@ -147,7 +146,6 @@ 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(F_ABS_MASK_4A)); } diff --git a/indra/llmath/llvector4a.cpp b/indra/llmath/llvector4a.cpp index 6edeb0fefe..b66b7a7076 100644 --- a/indra/llmath/llvector4a.cpp +++ b/indra/llmath/llvector4a.cpp @@ -24,7 +24,6 @@ * $/LicenseInfo$ */ -#include "llmemory.h" #include "llmath.h" #include "llquantize.h" @@ -45,10 +44,7 @@ extern const LLVector4a LL_V4A_EPSILON = reinterpret_cast ( 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) @@ -193,8 +189,6 @@ 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 0526793d3a..596082509d 100644 --- a/indra/llmath/llvector4a.h +++ b/indra/llmath/llvector4a.h @@ -32,7 +32,6 @@ class LLRotation; #include #include "llpreprocessor.h" -#include "llmemory.h" /////////////////////////////////// // FIRST TIME USERS PLEASE READ @@ -47,7 +46,6 @@ class LLRotation; // LLVector3/LLVector4. ///////////////////////////////// -LL_ALIGN_PREFIX(16) class LLVector4a { public: @@ -84,7 +82,6 @@ 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); //////////////////////////////////// @@ -93,7 +90,6 @@ 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) @@ -317,7 +313,7 @@ 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 7c52ffef21..7ad22a5631 100644 --- a/indra/llmath/llvector4a.inl +++ b/indra/llmath/llvector4a.inl @@ -475,7 +475,6 @@ 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 (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 c5698f7cea..dd66b09d43 100644 --- a/indra/llmath/llvector4logical.h +++ b/indra/llmath/llvector4logical.h @@ -27,7 +27,6 @@ #ifndef LL_VECTOR4LOGICAL_H #define LL_VECTOR4LOGICAL_H -#include "llmemory.h" //////////////////////////// // LLVector4Logical @@ -78,7 +77,6 @@ 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 53d56e96da..cc9744756f 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -95,6 +95,17 @@ 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); @@ -317,16 +328,16 @@ public: LLVector4a& min = node->mExtents[0]; LLVector4a& max = node->mExtents[1]; - if (!branch->isEmpty()) + if (!branch->getData().empty()) { //node has data, find AABB that binds data set - const LLVolumeTriangle* tri = *(branch->getDataBegin()); + const LLVolumeTriangle* tri = *(branch->getData().begin()); //initialize min/max to first available vertex min = *(tri->mV[0]); max = *(tri->mV[0]); for (LLOctreeNode::const_element_iter iter = - branch->getDataBegin(); iter != branch->getDataEnd(); ++iter) + branch->getData().begin(); iter != branch->getData().end(); ++iter) { //for each triangle in node //stretch by triangles in node @@ -341,7 +352,7 @@ public: max.setMax(max, *tri->mV[2]); } } - else if (!branch->isLeaf()) + else if (!branch->getChildren().empty()) { //no data, but child nodes exist LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(0)->getListener(0); @@ -6951,14 +6962,14 @@ void LLVolumeFace::resizeVertices(S32 num_verts) if (num_verts) { mPositions = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); - ll_assert_aligned(mPositions, 16); + assert_aligned(mPositions, 16); mNormals = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); - ll_assert_aligned(mNormals, 16); + 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); - ll_assert_aligned(mTexCoords, 16); + assert_aligned(mTexCoords, 16); } else { @@ -6982,17 +6993,14 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con // S32 old_size = mNumVertices*16; //positions - mPositions = (LLVector4a*) ll_aligned_realloc_16(mPositions, new_size); - ll_assert_aligned(mPositions,16); + mPositions = (LLVector4a*) realloc(mPositions, new_size); //normals - mNormals = (LLVector4a*) ll_aligned_realloc_16(mNormals, new_size); - ll_assert_aligned(mNormals,16); - + mNormals = (LLVector4a*) realloc(mNormals, new_size); + //tex coords new_size = ((new_verts*8)+0xF) & ~0xF; - mTexCoords = (LLVector2*) ll_aligned_realloc_16(mTexCoords, new_size); - ll_assert_aligned(mTexCoords,16); + mTexCoords = (LLVector2*) realloc(mTexCoords, new_size); //just clear binormals @@ -7045,8 +7053,7 @@ void LLVolumeFace::pushIndex(const U16& idx) S32 old_size = ((mNumIndices*2)+0xF) & ~0xF; if (new_size != old_size) { - mIndices = (U16*) ll_aligned_realloc_16(mIndices, new_size); - ll_assert_aligned(mIndices,16); + mIndices = (U16*) realloc(mIndices, new_size); } mIndices[mNumIndices++] = idx; @@ -7087,12 +7094,12 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat } //allocate new buffer space - mPositions = (LLVector4a*) ll_aligned_realloc_16(mPositions, new_count*sizeof(LLVector4a)); - ll_assert_aligned(mPositions, 16); - mNormals = (LLVector4a*) ll_aligned_realloc_16(mNormals, new_count*sizeof(LLVector4a)); - ll_assert_aligned(mNormals, 16); - mTexCoords = (LLVector2*) ll_aligned_realloc_16(mTexCoords, (new_count*sizeof(LLVector2)+0xF) & ~0xF); - ll_assert_aligned(mTexCoords, 16); + 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); mNumVertices = new_count; @@ -7138,7 +7145,7 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat new_count = mNumIndices + face.mNumIndices; //allocate new index buffer - mIndices = (U16*) ll_aligned_realloc_16(mIndices, (new_count*sizeof(U16)+0xF) & ~0xF); + mIndices = (U16*) realloc(mIndices, (new_count*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 cc83cb7235..b5a935c2b5 100644 --- a/indra/llmath/llvolumeoctree.cpp +++ b/indra/llmath/llvolumeoctree.cpp @@ -131,7 +131,7 @@ void LLOctreeTriangleRayIntersect::traverse(const LLOctreeNode void LLOctreeTriangleRayIntersect::visit(const LLOctreeNode* node) { for (LLOctreeNode::const_element_iter iter = - node->getDataBegin(); iter != node->getDataEnd(); ++iter) + node->getData().begin(); iter != node->getData().end(); ++iter) { const LLVolumeTriangle* tri = *iter; @@ -236,8 +236,8 @@ void LLVolumeOctreeValidate::visit(const LLOctreeNode* branch) } //children fit, check data - for (LLOctreeNode::const_element_iter iter = branch->getDataBegin(); - iter != branch->getDataEnd(); ++iter) + for (LLOctreeNode::const_element_iter iter = branch->getData().begin(); + iter != branch->getData().end(); ++iter) { const LLVolumeTriangle* tri = *iter; diff --git a/indra/llmath/llvolumeoctree.h b/indra/llmath/llvolumeoctree.h index 9ae34a0c4e..688d91dc40 100644 --- a/indra/llmath/llvolumeoctree.h +++ b/indra/llmath/llvolumeoctree.h @@ -37,19 +37,9 @@ 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) @@ -68,38 +58,21 @@ public: } - LL_ALIGN_16(LLVector4a mPositionGroup); + 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 { 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* node); ~LLVolumeOctreeListener(); @@ -126,8 +99,8 @@ public: public: - 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 + 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 }; class LLOctreeTriangleRayIntersect : public LLOctreeTraveler -- cgit v1.3 From 97d969a338c1e4f973eb817ba7701aff51a02ccb Mon Sep 17 00:00:00 2001 From: Oz Linden Date: Wed, 12 Sep 2012 14:36:37 -0400 Subject: initial attempt to restore changes that make removing tcmalloc possible; not tested --- indra/cmake/Copy3rdPartyLibs.cmake | 9 ++++--- indra/cmake/GooglePerfTools.cmake | 40 +++++++++++++++++++++-------- indra/cmake/LLAddBuildTest.cmake | 9 +++++++ indra/llcommon/llallocator.cpp | 2 +- indra/llcommon/llmemory.cpp | 12 +++++++++ indra/llcommon/llmemory.h | 45 +++++++++++++++++++++++---------- indra/llmath/CMakeLists.txt | 1 + indra/llmath/llcamera.h | 12 ++++----- indra/llmath/llmatrix3a.h | 2 +- indra/llmath/llmatrix4a.h | 2 +- indra/llmath/lloctree.h | 4 +-- indra/llmath/llplane.h | 4 ++- indra/llmath/llsimdmath.h | 3 +-- indra/llmath/llsimdtypes.inl | 2 ++ indra/llmath/llvector4a.cpp | 8 +++++- indra/llmath/llvector4a.h | 6 ++++- indra/llmath/llvector4a.inl | 1 + indra/llmath/llvector4logical.h | 2 ++ indra/llmath/llvolume.cpp | 45 ++++++++++++++------------------- indra/llmath/llvolumeoctree.h | 26 ++++++++++++++++--- indra/llprimitive/llmodel.cpp | 3 ++- indra/newview/CMakeLists.txt | 7 +++-- indra/newview/llappviewerwin32.cpp | 2 ++ indra/newview/lldrawable.h | 17 ++++++++++--- indra/newview/lldynamictexture.h | 12 ++++++++- indra/newview/llface.h | 11 ++++++++ indra/newview/llfloatermodelpreview.cpp | 3 ++- indra/newview/llpolymesh.cpp | 24 +++++++++--------- indra/newview/llspatialpartition.cpp | 3 +++ indra/newview/llspatialpartition.h | 36 ++++++++++++++++++++------ indra/newview/llviewercamera.h | 13 +++++++++- indra/newview/llviewerjointmesh.cpp | 6 +++-- indra/newview/llvoavatar.cpp | 2 +- indra/newview/llvoavatar.h | 12 ++++++++- indra/newview/llvoavatarself.h | 10 ++++++++ indra/newview/viewer_manifest.py | 10 +++++++- 36 files changed, 298 insertions(+), 108 deletions(-) (limited to 'indra/llmath') diff --git a/indra/cmake/Copy3rdPartyLibs.cmake b/indra/cmake/Copy3rdPartyLibs.cmake index 224e0a8b51..9f05c4cff2 100644 --- a/indra/cmake/Copy3rdPartyLibs.cmake +++ b/indra/cmake/Copy3rdPartyLibs.cmake @@ -57,10 +57,10 @@ if(WINDOWS) libhunspell.dll ) - if(USE_GOOGLE_PERFTOOLS) + if(USE_TCMALLOC) set(debug_files ${debug_files} libtcmalloc_minimal-debug.dll) set(release_files ${release_files} libtcmalloc_minimal.dll) - endif(USE_GOOGLE_PERFTOOLS) + endif(USE_TCMALLOC) if (FMOD) set(debug_files ${debug_files} fmod.dll) @@ -272,13 +272,16 @@ elseif(LINUX) libopenal.so libopenjpeg.so libssl.so - libtcmalloc_minimal.so libuuid.so.16 libuuid.so.16.0.22 libssl.so.1.0.0 libfontconfig.so.1.4.4 ) + if (USE_TCMALLOC) + set(release_files ${release_files} "libtcmalloc_minimal.so") + endif (USE_TCMALLOC) + if (FMOD) set(release_files ${release_files} "libfmod-3.75.so") endif (FMOD) diff --git a/indra/cmake/GooglePerfTools.cmake b/indra/cmake/GooglePerfTools.cmake index d9f91193be..09501e0406 100644 --- a/indra/cmake/GooglePerfTools.cmake +++ b/indra/cmake/GooglePerfTools.cmake @@ -1,20 +1,34 @@ # -*- cmake -*- include(Prebuilt) +# If you want to enable or disable TCMALLOC in viewer builds, this is the place. +# set ON or OFF as desired. +set (USE_TCMALLOC ON) + if (STANDALONE) include(FindGooglePerfTools) else (STANDALONE) if (WINDOWS) - use_prebuilt_binary(tcmalloc) - set(TCMALLOC_LIBRARIES - debug libtcmalloc_minimal-debug - optimized libtcmalloc_minimal) + if (USE_TCMALLOC) + use_prebuilt_binary(tcmalloc) + set(TCMALLOC_LIBRARIES + debug libtcmalloc_minimal-debug + optimized libtcmalloc_minimal) + set(TCMALLOC_LINK_FLAGS "/INCLUDE:__tcmalloc") + else (USE_TCMALLOC) + set(TCMALLOC_LIBRARIES) + set(TCMALLOC_LINK_FLAGS) + endif (USE_TCMALLOC) set(GOOGLE_PERFTOOLS_FOUND "YES") endif (WINDOWS) if (LINUX) - use_prebuilt_binary(tcmalloc) - set(TCMALLOC_LIBRARIES - tcmalloc) + if (USE_TCMALLOC) + use_prebuilt_binary(tcmalloc) + set(TCMALLOC_LIBRARIES + tcmalloc) + else (USE_TCMALLOC) + set(TCMALLOC_LIBRARIES) + endif (USE_TCMALLOC) set(PROFILER_LIBRARIES profiler) set(GOOGLE_PERFTOOLS_INCLUDE_DIR ${LIBS_PREBUILT_DIR}/include) @@ -29,13 +43,19 @@ if (GOOGLE_PERFTOOLS_FOUND) endif (GOOGLE_PERFTOOLS_FOUND) if (WINDOWS) - set(USE_GOOGLE_PERFTOOLS ON) + set(USE_GOOGLE_PERFTOOLS ON) endif (WINDOWS) if (USE_GOOGLE_PERFTOOLS) - set(TCMALLOC_FLAG -ULL_USE_TCMALLOC=1) + if (USE_TCMALLOC) + set(TCMALLOC_FLAG -DLL_USE_TCMALLOC=1) + else (USE_TCMALLOC) + set(TCMALLOC_FLAG -ULL_USE_TCMALLOC) + endif (USE_TCMALLOC) +endif (USE_GOOGLE_PERFTOOLS) + +if (USE_GOOGLE_PERFTOOLS) include_directories(${GOOGLE_PERFTOOLS_INCLUDE_DIR}) set(GOOGLE_PERFTOOLS_LIBRARIES ${TCMALLOC_LIBRARIES} ${STACKTRACE_LIBRARIES} ${PROFILER_LIBRARIES}) else (USE_GOOGLE_PERFTOOLS) - set(TCMALLOC_FLAG -ULL_USE_TCMALLOC) endif (USE_GOOGLE_PERFTOOLS) diff --git a/indra/cmake/LLAddBuildTest.cmake b/indra/cmake/LLAddBuildTest.cmake index 03ce46781c..543075db5b 100755 --- a/indra/cmake/LLAddBuildTest.cmake +++ b/indra/cmake/LLAddBuildTest.cmake @@ -214,6 +214,15 @@ FUNCTION(LL_ADD_INTEGRATION_TEST SET_TARGET_PROPERTIES(INTEGRATION_TEST_${testname} PROPERTIES COMPILE_FLAGS -I"${TUT_INCLUDE_DIR}") endif(STANDALONE) + if (WINDOWS) + SET_TARGET_PROPERTIES(INTEGRATION_TEST_${testname} + PROPERTIES + LINK_FLAGS "/debug /NODEFAULTLIB:LIBCMT /SUBSYSTEM:WINDOWS ${TCMALLOC_LINK_FLAGS}" + LINK_FLAGS_DEBUG "/NODEFAULTLIB:\"LIBCMT;LIBCMTD;MSVCRT\" /INCREMENTAL:NO" + LINK_FLAGS_RELEASE "" + ) + endif (WINDOWS) + # Add link deps to the executable if(TEST_DEBUG) message(STATUS "TARGET_LINK_LIBRARIES(INTEGRATION_TEST_${testname} ${libraries})") diff --git a/indra/llcommon/llallocator.cpp b/indra/llcommon/llallocator.cpp index 6f6abefc67..87654b5b97 100644 --- a/indra/llcommon/llallocator.cpp +++ b/indra/llcommon/llallocator.cpp @@ -27,7 +27,7 @@ #include "linden_common.h" #include "llallocator.h" -#if LL_USE_TCMALLOC +#if (LL_USE_TCMALLOC && LL_USE_HEAP_PROFILER) #include "google/heap-profiler.h" #include "google/commandlineflags_public.h" diff --git a/indra/llcommon/llmemory.cpp b/indra/llcommon/llmemory.cpp index 3b9758f996..afaf366668 100644 --- a/indra/llcommon/llmemory.cpp +++ b/indra/llcommon/llmemory.cpp @@ -61,6 +61,18 @@ BOOL LLMemory::sEnableMemoryFailurePrevention = FALSE; LLPrivateMemoryPoolManager::mem_allocation_info_t LLPrivateMemoryPoolManager::sMemAllocationTracker; #endif +void ll_assert_aligned_func(uintptr_t ptr,U32 alignment) +{ +#ifdef SHOW_ASSERT + // Redundant, place to set breakpoints. + if (ptr%alignment!=0) + { + llwarns << "alignment check failed" << llendl; + } + llassert(ptr%alignment==0); +#endif +} + //static void LLMemory::initClass() { diff --git a/indra/llcommon/llmemory.h b/indra/llcommon/llmemory.h index bbbdaa6497..9dd776ff57 100644 --- a/indra/llcommon/llmemory.h +++ b/indra/llcommon/llmemory.h @@ -27,7 +27,6 @@ #define LLMEMORY_H #include "llmemtype.h" -#if LL_DEBUG inline void* ll_aligned_malloc( size_t size, int align ) { void* mem = malloc( size + (align - 1) + sizeof(void*) ); @@ -43,10 +42,11 @@ inline void ll_aligned_free( void* ptr ) free( ((void**)ptr)[-1] ); } +#if !LL_USE_TCMALLOC inline void* ll_aligned_malloc_16(size_t size) // returned hunk MUST be freed with ll_aligned_free_16(). { #if defined(LL_WINDOWS) - return _mm_malloc(size, 16); + return _aligned_malloc(size, 16); #elif defined(LL_DARWIN) return malloc(size); // default osx malloc is 16 byte aligned. #else @@ -58,21 +58,38 @@ inline void* ll_aligned_malloc_16(size_t size) // returned hunk MUST be freed wi #endif } +inline void* ll_aligned_realloc_16(void* ptr, size_t size) // returned hunk MUST be freed with ll_aligned_free_16(). +{ +#if defined(LL_WINDOWS) + return _aligned_realloc(ptr, size, 16); +#elif defined(LL_DARWIN) + return realloc(ptr,size); // default osx malloc is 16 byte aligned. +#else + return realloc(ptr,size); // FIXME not guaranteed to be aligned. +#endif +} + inline void ll_aligned_free_16(void *p) { #if defined(LL_WINDOWS) - _mm_free(p); + _aligned_free(p); #elif defined(LL_DARWIN) return free(p); #else free(p); // posix_memalign() is compatible with heap deallocator #endif } +#else // USE_TCMALLOC +// ll_aligned_foo_16 are not needed with tcmalloc +#define ll_aligned_malloc_16 malloc +#define ll_aligned_realloc_16 realloc +#define ll_aligned_free_16 free +#endif // USE_TCMALLOC inline void* ll_aligned_malloc_32(size_t size) // returned hunk MUST be freed with ll_aligned_free_32(). { #if defined(LL_WINDOWS) - return _mm_malloc(size, 32); + return _aligned_malloc(size, 32); #elif defined(LL_DARWIN) return ll_aligned_malloc( size, 32 ); #else @@ -87,22 +104,13 @@ inline void* ll_aligned_malloc_32(size_t size) // returned hunk MUST be freed wi inline void ll_aligned_free_32(void *p) { #if defined(LL_WINDOWS) - _mm_free(p); + _aligned_free(p); #elif defined(LL_DARWIN) ll_aligned_free( p ); #else free(p); // posix_memalign() is compatible with heap deallocator #endif } -#else // LL_DEBUG -// ll_aligned_foo are noops now that we use tcmalloc everywhere (tcmalloc aligns automatically at appropriate intervals) -#define ll_aligned_malloc( size, align ) malloc(size) -#define ll_aligned_free( ptr ) free(ptr) -#define ll_aligned_malloc_16 malloc -#define ll_aligned_free_16 free -#define ll_aligned_malloc_32 malloc -#define ll_aligned_free_32 free -#endif // LL_DEBUG #ifndef __DEBUG_PRIVATE_MEM__ #define __DEBUG_PRIVATE_MEM__ 0 @@ -512,4 +520,13 @@ void LLPrivateMemoryPoolTester::operator delete[](void* addr) // LLSingleton moved to llsingleton.h +LL_COMMON_API void ll_assert_aligned_func(uintptr_t ptr,U32 alignment); + +#ifdef SHOW_ASSERT +#define ll_assert_aligned(ptr,alignment) ll_assert_aligned_func(reinterpret_cast(ptr),((U32)alignment)) +#else +#define ll_assert_aligned(ptr,alignment) +#endif + + #endif 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..6c7744cdf1 100644 --- a/indra/llmath/lloctree.h +++ b/indra/llmath/lloctree.h @@ -88,7 +88,7 @@ public: typedef LLOctreeNode oct_node; typedef LLOctreeListener oct_listener; - /*void* operator new(size_t size) + void* operator new(size_t size) { return ll_aligned_malloc_16(size); } @@ -96,7 +96,7 @@ public: void operator delete(void* ptr) { ll_aligned_free_16(ptr); - }*/ + } LLOctreeNode( const LLVector4a& center, const LLVector4a& size, 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 T* LL_NEXT_ALIGNED_ADDRESS_64(T* address) #define LL_ALIGN_16(var) LL_ALIGN_PREFIX(16) var LL_ALIGN_POSTFIX(16) - - #include #include +#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(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(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 ( 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 #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 (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..11fa7080ce 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); @@ -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 { @@ -6993,14 +6982,17 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con // S32 old_size = mNumVertices*16; //positions - mPositions = (LLVector4a*) realloc(mPositions, new_size); + mPositions = (LLVector4a*) ll_aligned_realloc_16(mPositions, new_size); + ll_assert_aligned(mPositions,16); //normals - mNormals = (LLVector4a*) realloc(mNormals, new_size); - + mNormals = (LLVector4a*) ll_aligned_realloc_16(mNormals, new_size); + ll_assert_aligned(mNormals,16); + //tex coords new_size = ((new_verts*8)+0xF) & ~0xF; - mTexCoords = (LLVector2*) realloc(mTexCoords, new_size); + mTexCoords = (LLVector2*) ll_aligned_realloc_16(mTexCoords, new_size); + ll_assert_aligned(mTexCoords,16); //just clear binormals @@ -7053,7 +7045,8 @@ 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); + ll_assert_aligned(mIndices,16); } mIndices[mNumIndices++] = idx; @@ -7094,12 +7087,12 @@ 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)); + ll_assert_aligned(mPositions, 16); + mNormals = (LLVector4a*) ll_aligned_realloc_16(mNormals, new_count*sizeof(LLVector4a)); + ll_assert_aligned(mNormals, 16); + mTexCoords = (LLVector2*) ll_aligned_realloc_16(mTexCoords, (new_count*sizeof(LLVector2)+0xF) & ~0xF); + ll_assert_aligned(mTexCoords, 16); mNumVertices = new_count; @@ -7145,7 +7138,7 @@ 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); //get destination address into new index buffer U16* dst_idx = mIndices+mNumIndices; diff --git a/indra/llmath/llvolumeoctree.h b/indra/llmath/llvolumeoctree.h index 688d91dc40..dac97b14b5 100644 --- a/indra/llmath/llvolumeoctree.h +++ b/indra/llmath/llvolumeoctree.h @@ -37,6 +37,16 @@ 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() { @@ -58,7 +68,7 @@ public: } - LLVector4a mPositionGroup; + LL_ALIGN_16(LLVector4a mPositionGroup); const LLVector4a* mV[3]; U16 mIndex[3]; @@ -73,6 +83,16 @@ class LLVolumeOctreeListener : public LLOctreeListener { 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* node); ~LLVolumeOctreeListener(); @@ -99,8 +119,8 @@ 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 diff --git a/indra/llprimitive/llmodel.cpp b/indra/llprimitive/llmodel.cpp index cb32a510b8..28ed051c55 100644 --- a/indra/llprimitive/llmodel.cpp +++ b/indra/llprimitive/llmodel.cpp @@ -1026,7 +1026,8 @@ void LLModel::setVolumeFaceData( if (tc.get()) { - LLVector4a::memcpyNonAliased16((F32*) face.mTexCoords, (F32*) tc.get(), num_verts*2*sizeof(F32)); + U32 tex_size = (num_verts*2*sizeof(F32)+0xF)&~0xF; + LLVector4a::memcpyNonAliased16((F32*) face.mTexCoords, (F32*) tc.get(), tex_size); } else { diff --git a/indra/newview/CMakeLists.txt b/indra/newview/CMakeLists.txt index c0a252637f..2ae162a623 100644 --- a/indra/newview/CMakeLists.txt +++ b/indra/newview/CMakeLists.txt @@ -1565,8 +1565,7 @@ if (WINDOWS) set_target_properties(${VIEWER_BINARY_NAME} PROPERTIES # *TODO -reenable this once we get server usage sorted out - #LINK_FLAGS "/debug /NODEFAULTLIB:LIBCMT /SUBSYSTEM:WINDOWS /INCLUDE:\"__tcmalloc\"" - LINK_FLAGS "/debug /NODEFAULTLIB:LIBCMT /SUBSYSTEM:WINDOWS /INCLUDE:__tcmalloc " + LINK_FLAGS "/debug /NODEFAULTLIB:LIBCMT /SUBSYSTEM:WINDOWS " LINK_FLAGS_DEBUG "/NODEFAULTLIB:\"LIBCMT;LIBCMTD;MSVCRT\" /INCREMENTAL:NO" LINK_FLAGS_RELEASE "/FORCE:MULTIPLE /MAP\"secondlife-bin.MAP\" /OPT:REF" ) @@ -1585,7 +1584,7 @@ if (WINDOWS) # In the meantime, if you have any ideas on how to easily maintain one list, either here or in viewer_manifest.py # and have the build deps get tracked *please* tell me about it. - if(USE_GOOGLE_PERFTOOLS) + if(USE_TCMALLOC) # Configure a var for tcmalloc location, if used. # Note the need to specify multiple names explicitly. set(GOOGLE_PERF_TOOLS_SOURCE @@ -1593,7 +1592,7 @@ if (WINDOWS) ${SHARED_LIB_STAGING_DIR}/RelWithDebInfo/libtcmalloc_minimal.dll ${SHARED_LIB_STAGING_DIR}/Debug/libtcmalloc_minimal-debug.dll ) - endif(USE_GOOGLE_PERFTOOLS) + endif(USE_TCMALLOC) set(COPY_INPUT_DEPENDENCIES diff --git a/indra/newview/llappviewerwin32.cpp b/indra/newview/llappviewerwin32.cpp index bad60a9757..53c77fa22e 100644 --- a/indra/newview/llappviewerwin32.cpp +++ b/indra/newview/llappviewerwin32.cpp @@ -130,6 +130,8 @@ int APIENTRY WINMAIN(HINSTANCE hInstance, // This results in a 2-3x improvement in opening a new Inventory window (which uses a large numebr of allocations) // Note: This won't work when running from the debugger unless the _NO_DEBUG_HEAP environment variable is set to 1 + // Enable to get mem debugging within visual studio. + //_CrtSetDbgFlag(_CRTDBG_ALLOC_MEM_DF | _CRTDBG_LEAK_CHECK_DF); _CrtSetDbgFlag(0); // default, just making explicit ULONG ulEnableLFH = 2; diff --git a/indra/newview/lldrawable.h b/indra/newview/lldrawable.h index e2064b79f8..8c7db61502 100644 --- a/indra/newview/lldrawable.h +++ b/indra/newview/lldrawable.h @@ -59,6 +59,7 @@ class LLViewerTexture; const U32 SILHOUETTE_HIGHLIGHT = 0; // All data for new renderer goes into this class. +LL_ALIGN_PREFIX(16) class LLDrawable : public LLRefCount { public: @@ -75,6 +76,16 @@ public: static void initClass(); + void* operator new(size_t size) + { + return ll_aligned_malloc_16(size); + } + + void operator delete(void* ptr) + { + ll_aligned_free_16(ptr); + } + LLDrawable() { init(); } MEM_TYPE_NEW(LLMemType::MTYPE_DRAWABLE); @@ -281,8 +292,8 @@ public: } EDrawableFlags; private: //aligned members - LLVector4a mExtents[2]; - LLVector4a mPositionGroup; + LL_ALIGN_16(LLVector4a mExtents[2]); + LL_ALIGN_16(LLVector4a mPositionGroup); public: LLXformMatrix mXform; @@ -323,7 +334,7 @@ private: static U32 sNumZombieDrawables; static LLDynamicArrayPtr > sDeadList; -}; +} LL_ALIGN_POSTFIX(16); inline LLFace* LLDrawable::getFace(const S32 i) const diff --git a/indra/newview/lldynamictexture.h b/indra/newview/lldynamictexture.h index e18090545d..c51e7d1e1a 100644 --- a/indra/newview/lldynamictexture.h +++ b/indra/newview/lldynamictexture.h @@ -36,6 +36,16 @@ class LLViewerDynamicTexture : public LLViewerTexture { public: + void* operator new(size_t size) + { + return ll_aligned_malloc_16(size); + } + + void operator delete(void* ptr) + { + ll_aligned_free_16(ptr); + } + enum { LL_VIEWER_DYNAMIC_TEXTURE = LLViewerTexture::DYNAMIC_TEXTURE, @@ -85,7 +95,7 @@ protected: protected: BOOL mClamp; LLCoordGL mOrigin; - LLCamera mCamera; + LL_ALIGN_16(LLCamera mCamera); typedef std::set instance_list_t; static instance_list_t sInstances[ LLViewerDynamicTexture::ORDER_COUNT ]; diff --git a/indra/newview/llface.h b/indra/newview/llface.h index 76ea5c853a..5dca27487f 100644 --- a/indra/newview/llface.h +++ b/indra/newview/llface.h @@ -59,6 +59,17 @@ class LLFace { public: + void* operator new(size_t size) + { + return ll_aligned_malloc_16(size); + } + + void operator delete(void* ptr) + { + ll_aligned_free_16(ptr); + } + + LLFace(const LLFace& rhs) { *this = rhs; diff --git a/indra/newview/llfloatermodelpreview.cpp b/indra/newview/llfloatermodelpreview.cpp index 3fe535cbe8..a071f338ba 100755 --- a/indra/newview/llfloatermodelpreview.cpp +++ b/indra/newview/llfloatermodelpreview.cpp @@ -4774,7 +4774,8 @@ void LLModelPreview::genBuffers(S32 lod, bool include_skin_weights) if (vf.mTexCoords) { vb->getTexCoord0Strider(tc_strider); - LLVector4a::memcpyNonAliased16((F32*) tc_strider.get(), (F32*) vf.mTexCoords, num_vertices*2*sizeof(F32)); + S32 tex_size = (num_vertices*2*sizeof(F32)+0xF) & ~0xF; + LLVector4a::memcpyNonAliased16((F32*) tc_strider.get(), (F32*) vf.mTexCoords, tex_size); } if (vf.mNormals) diff --git a/indra/newview/llpolymesh.cpp b/indra/newview/llpolymesh.cpp index 450f9b2be7..0860506086 100644 --- a/indra/newview/llpolymesh.cpp +++ b/indra/newview/llpolymesh.cpp @@ -129,22 +129,22 @@ void LLPolyMeshSharedData::freeMeshData() { mNumVertices = 0; - delete [] mBaseCoords; + ll_aligned_free_16(mBaseCoords); mBaseCoords = NULL; - delete [] mBaseNormals; + ll_aligned_free_16(mBaseNormals); mBaseNormals = NULL; - delete [] mBaseBinormals; + ll_aligned_free_16(mBaseBinormals); mBaseBinormals = NULL; - delete [] mTexCoords; + ll_aligned_free_16(mTexCoords); mTexCoords = NULL; - delete [] mDetailTexCoords; + ll_aligned_free_16(mDetailTexCoords); mDetailTexCoords = NULL; - delete [] mWeights; + ll_aligned_free_16(mWeights); mWeights = NULL; } @@ -229,12 +229,12 @@ U32 LLPolyMeshSharedData::getNumKB() BOOL LLPolyMeshSharedData::allocateVertexData( U32 numVertices ) { U32 i; - mBaseCoords = new LLVector3[ numVertices ]; - mBaseNormals = new LLVector3[ numVertices ]; - mBaseBinormals = new LLVector3[ numVertices ]; - mTexCoords = new LLVector2[ numVertices ]; - mDetailTexCoords = new LLVector2[ numVertices ]; - mWeights = new F32[ numVertices ]; + mBaseCoords = (LLVector3*) ll_aligned_malloc_16(numVertices*sizeof(LLVector3)); + mBaseNormals = (LLVector3*) ll_aligned_malloc_16(numVertices*sizeof(LLVector3)); + mBaseBinormals = (LLVector3*) ll_aligned_malloc_16(numVertices*sizeof(LLVector3)); + mTexCoords = (LLVector2*) ll_aligned_malloc_16(numVertices*sizeof(LLVector2)); + mDetailTexCoords = (LLVector2*) ll_aligned_malloc_16(numVertices*sizeof(LLVector2)); + mWeights = (F32*) ll_aligned_malloc_16(numVertices*sizeof(F32)); for (i = 0; i < numVertices; i++) { mWeights[i] = 0.f; diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp index 325a2d3004..45ef8f1a6d 100644 --- a/indra/newview/llspatialpartition.cpp +++ b/indra/newview/llspatialpartition.cpp @@ -529,6 +529,7 @@ void LLSpatialGroup::setVisible() void LLSpatialGroup::validate() { + ll_assert_aligned(this,64); #if LL_OCTREE_PARANOIA_CHECK sg_assert(!isState(DIRTY)); @@ -1195,6 +1196,8 @@ LLSpatialGroup::LLSpatialGroup(OctreeNode* node, LLSpatialPartition* part) : mCurUpdatingSlotp(NULL), mCurUpdatingTexture (NULL) { + ll_assert_aligned(this,16); + sNodeCount++; LLMemType mt(LLMemType::MTYPE_SPACE_PARTITION); diff --git a/indra/newview/llspatialpartition.h b/indra/newview/llspatialpartition.h index f0e4f15a83..7968c28900 100644 --- a/indra/newview/llspatialpartition.h +++ b/indra/newview/llspatialpartition.h @@ -68,6 +68,16 @@ protected: ~LLDrawInfo(); public: + void* operator new(size_t size) + { + return ll_aligned_malloc_16(size); + } + + void operator delete(void* ptr) + { + ll_aligned_free_16(ptr); + } + LLDrawInfo(const LLDrawInfo& rhs) { @@ -106,7 +116,7 @@ public: F32 mPartSize; F32 mVSize; LLSpatialGroup* mGroup; - LLFace* mFace; //associated face + LL_ALIGN_16(LLFace* mFace); //associated face F32 mDistance; U32 mDrawMode; @@ -181,7 +191,7 @@ public: }; }; -LL_ALIGN_PREFIX(64) +LL_ALIGN_PREFIX(16) class LLSpatialGroup : public LLOctreeListener { friend class LLSpatialPartition; @@ -193,6 +203,16 @@ public: *this = rhs; } + void* operator new(size_t size) + { + return ll_aligned_malloc_16(size); + } + + void operator delete(void* ptr) + { + ll_aligned_free_16(ptr); + } + const LLSpatialGroup& operator=(const LLSpatialGroup& rhs) { llerrs << "Illegal operation!" << llendl; @@ -370,12 +390,12 @@ public: V4_COUNT = 10 } eV4Index; - 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 - LLVector4a mObjectExtents[2]; // extents (min, max) of objects in this node - LLVector4a mObjectBounds[2]; // bounding box (center, size) of objects in this node - LLVector4a mViewAngle; - LLVector4a mLastUpdateViewAngle; + 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 + LL_ALIGN_16(LLVector4a mObjectExtents[2]); // extents (min, max) of objects in this node + LL_ALIGN_16(LLVector4a mObjectBounds[2]); // bounding box (center, size) of objects in this node + LL_ALIGN_16(LLVector4a mViewAngle); + LL_ALIGN_16(LLVector4a mLastUpdateViewAngle); F32 mObjectBoxSize; //cached mObjectBounds[1].getLength3() diff --git a/indra/newview/llviewercamera.h b/indra/newview/llviewercamera.h index 184033de42..b857c7fe89 100644 --- a/indra/newview/llviewercamera.h +++ b/indra/newview/llviewercamera.h @@ -51,9 +51,19 @@ const BOOL NOT_FOR_SELECTION = FALSE; extern template class LLViewerCamera* LLSingleton::getInstance(); #endif +LL_ALIGN_PREFIX(16) class LLViewerCamera : public LLCamera, public LLSingleton { public: + void* operator new(size_t size) + { + return ll_aligned_malloc_16(size); + } + + void operator delete(void* ptr) + { + ll_aligned_free_16(ptr); + } typedef enum { @@ -137,6 +147,7 @@ protected: S16 mZoomSubregion; public: -}; +} LL_ALIGN_POSTFIX(16); + #endif // LL_LLVIEWERCAMERA_H diff --git a/indra/newview/llviewerjointmesh.cpp b/indra/newview/llviewerjointmesh.cpp index f029ae5302..5d1aa870a3 100644 --- a/indra/newview/llviewerjointmesh.cpp +++ b/indra/newview/llviewerjointmesh.cpp @@ -729,8 +729,10 @@ void LLViewerJointMesh::updateFaceData(LLFace *face, F32 pixel_area, BOOL damp_w F32* vw = (F32*) vertex_weightsp.get(); F32* cw = (F32*) clothing_weightsp.get(); - LLVector4a::memcpyNonAliased16(tc, (F32*) mMesh->getTexCoords(), num_verts*2*sizeof(F32)); - LLVector4a::memcpyNonAliased16(vw, (F32*) mMesh->getWeights(), num_verts*sizeof(F32)); + S32 tc_size = (num_verts*2*sizeof(F32)+0xF) & ~0xF; + LLVector4a::memcpyNonAliased16(tc, (F32*) mMesh->getTexCoords(), tc_size); + S32 vw_size = (num_verts*sizeof(F32)+0xF) & ~0xF; + LLVector4a::memcpyNonAliased16(vw, (F32*) mMesh->getWeights(), vw_size); LLVector4a::memcpyNonAliased16(cw, (F32*) mMesh->getClothingWeights(), num_verts*4*sizeof(F32)); } diff --git a/indra/newview/llvoavatar.cpp b/indra/newview/llvoavatar.cpp index 33dc12c473..1b7009a5c2 100644 --- a/indra/newview/llvoavatar.cpp +++ b/indra/newview/llvoavatar.cpp @@ -2692,7 +2692,7 @@ void LLVOAvatar::idleUpdateMisc(bool detailed_update) if (isImpostor() && !mNeedsImpostorUpdate) { - LLVector4a ext[2]; + LL_ALIGN_16(LLVector4a ext[2]); F32 distance; LLVector3 angle; diff --git a/indra/newview/llvoavatar.h b/indra/newview/llvoavatar.h index 6fb56a4c0b..4081a1408d 100644 --- a/indra/newview/llvoavatar.h +++ b/indra/newview/llvoavatar.h @@ -93,6 +93,16 @@ protected: **/ public: + void* operator new(size_t size) + { + return ll_aligned_malloc_16(size); + } + + void operator delete(void* ptr) + { + ll_aligned_free_16(ptr); + } + LLVOAvatar(const LLUUID &id, const LLPCode pcode, LLViewerRegion *regionp); virtual void markDead(); static void initClass(); // Initialize data that's only init'd once per class. @@ -215,7 +225,7 @@ public: bool isBuilt() const { return mIsBuilt; } private: //aligned members - LLVector4a mImpostorExtents[2]; + LL_ALIGN_16(LLVector4a mImpostorExtents[2]); private: BOOL mSupportsAlphaLayers; // For backwards compatibility, TRUE for 1.23+ clients diff --git a/indra/newview/llvoavatarself.h b/indra/newview/llvoavatarself.h index 543891ca63..2b273e616c 100644 --- a/indra/newview/llvoavatarself.h +++ b/indra/newview/llvoavatarself.h @@ -49,6 +49,16 @@ class LLVOAvatarSelf : **/ public: + void* operator new(size_t size) + { + return ll_aligned_malloc_16(size); + } + + void operator delete(void* ptr) + { + ll_aligned_free_16(ptr); + } + LLVOAvatarSelf(const LLUUID &id, const LLPCode pcode, LLViewerRegion *regionp); virtual ~LLVOAvatarSelf(); virtual void markDead(); diff --git a/indra/newview/viewer_manifest.py b/indra/newview/viewer_manifest.py index 7c6b5403e1..894d2f0925 100644 --- a/indra/newview/viewer_manifest.py +++ b/indra/newview/viewer_manifest.py @@ -1080,7 +1080,15 @@ class Linux_i686Manifest(LinuxManifest): # previous call did, without having to explicitly state the # version number. self.path("libfontconfig.so.*.*") - self.path("libtcmalloc.so*") #formerly called google perf tools + try: + self.path("libtcmalloc.so", "libtcmalloc.so") #formerly called google perf tools + self.path("libtcmalloc.so.0", "libtcmalloc.so.0") #formerly called google perf tools + self.path("libtcmalloc.so.0.1.0", "libtcmalloc.so.0.1.0") #formerly called google perf tools + pass + except: + print "tcmalloc files not found, skipping" + pass + try: self.path("libfmod-3.75.so") pass -- cgit v1.3 From 9ec263d2f1f51b334949281393ef2446796f1404 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 12 Sep 2012 15:44:36 -0500 Subject: MAINT-1503 Minimal set of changes needed to safely disable tcmalloc --- indra/cmake/GooglePerfTools.cmake | 2 +- indra/llcommon/llmemory.h | 24 ++++++++++++++++-------- indra/llmath/llvolume.cpp | 19 ++++++++++--------- indra/llmath/tests/alignment_test.cpp | 2 +- 4 files changed, 28 insertions(+), 19 deletions(-) (limited to 'indra/llmath') diff --git a/indra/cmake/GooglePerfTools.cmake b/indra/cmake/GooglePerfTools.cmake index 09501e0406..73b3642ae6 100644 --- a/indra/cmake/GooglePerfTools.cmake +++ b/indra/cmake/GooglePerfTools.cmake @@ -3,7 +3,7 @@ include(Prebuilt) # If you want to enable or disable TCMALLOC in viewer builds, this is the place. # set ON or OFF as desired. -set (USE_TCMALLOC ON) +set (USE_TCMALLOC OFF) if (STANDALONE) include(FindGooglePerfTools) diff --git a/indra/llcommon/llmemory.h b/indra/llcommon/llmemory.h index 9dd776ff57..f37e8459ff 100644 --- a/indra/llcommon/llmemory.h +++ b/indra/llcommon/llmemory.h @@ -58,27 +58,35 @@ inline void* ll_aligned_malloc_16(size_t size) // returned hunk MUST be freed wi #endif } -inline void* ll_aligned_realloc_16(void* ptr, size_t size) // returned hunk MUST be freed with ll_aligned_free_16(). +inline void ll_aligned_free_16(void *p) { #if defined(LL_WINDOWS) - return _aligned_realloc(ptr, size, 16); + _aligned_free(p); #elif defined(LL_DARWIN) - return realloc(ptr,size); // default osx malloc is 16 byte aligned. + return free(p); #else - return realloc(ptr,size); // FIXME not guaranteed to be aligned. + free(p); // posix_memalign() is compatible with heap deallocator #endif } -inline void ll_aligned_free_16(void *p) +inline void* ll_aligned_realloc_16(void* ptr, size_t size, size_t old_size) // returned hunk MUST be freed with ll_aligned_free_16(). { #if defined(LL_WINDOWS) - _aligned_free(p); + return _aligned_realloc(ptr, size, 16); #elif defined(LL_DARWIN) - return free(p); + return realloc(ptr,size); // default osx malloc is 16 byte aligned. #else - free(p); // posix_memalign() is compatible with heap deallocator + //FIXME: memcpy is SLOW + void* ret = ll_aligned_malloc_16(size); + if (ptr) + { + memcpy(ret, ptr, old_size); + ll_aligned_free_16(ptr); + } + return ret; #endif } + #else // USE_TCMALLOC // ll_aligned_foo_16 are not needed with tcmalloc #define ll_aligned_malloc_16 malloc diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 11fa7080ce..ea09a3afe7 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -6979,19 +6979,20 @@ 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*) ll_aligned_realloc_16(mPositions, new_size); + mPositions = (LLVector4a*) ll_aligned_realloc_16(mPositions, new_size, old_size); ll_assert_aligned(mPositions,16); //normals - mNormals = (LLVector4a*) ll_aligned_realloc_16(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*) ll_aligned_realloc_16(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); @@ -7045,7 +7046,7 @@ void LLVolumeFace::pushIndex(const U16& idx) S32 old_size = ((mNumIndices*2)+0xF) & ~0xF; if (new_size != old_size) { - mIndices = (U16*) ll_aligned_realloc_16(mIndices, new_size); + mIndices = (U16*) ll_aligned_realloc_16(mIndices, new_size, old_size); ll_assert_aligned(mIndices,16); } @@ -7087,11 +7088,11 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat } //allocate new buffer space - mPositions = (LLVector4a*) ll_aligned_realloc_16(mPositions, new_count*sizeof(LLVector4a)); + 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)); + 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); + 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; @@ -7138,7 +7139,7 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat new_count = mNumIndices + face.mNumIndices; //allocate new index buffer - mIndices = (U16*) ll_aligned_realloc_16(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/tests/alignment_test.cpp b/indra/llmath/tests/alignment_test.cpp index ac0c45ae6f..5c426c57c3 100644 --- a/indra/llmath/tests/alignment_test.cpp +++ b/indra/llmath/tests/alignment_test.cpp @@ -78,7 +78,7 @@ 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); -- cgit v1.3 From 8f7bb8f6d58dfa1d0b160948b2ca790cd6d4ae5c Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 13 Sep 2012 20:31:22 -0500 Subject: Port over Simon's fix for alignment test --- indra/llmath/tests/alignment_test.cpp | 31 +++++++++++++++++++++---------- 1 file changed, 21 insertions(+), 10 deletions(-) (limited to 'indra/llmath') diff --git a/indra/llmath/tests/alignment_test.cpp b/indra/llmath/tests/alignment_test.cpp index 5c426c57c3..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); -- cgit v1.3 From cf98064700a736f73a6c21ce899b186919cbeb64 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 20 Sep 2012 09:48:55 -0400 Subject: reapply 52b6c9168974: MAINT-646 Factor std::set out of lloctree --- indra/llcharacter/llcharacter.cpp | 7 -- indra/llcharacter/lleditingmotion.cpp | 3 - indra/llcharacter/llkeyframewalkmotion.cpp | 4 - indra/llcharacter/llmotioncontroller.cpp | 29 ++----- indra/llmath/lloctree.h | 125 ++++++++++++++++++++++------- indra/llmath/llvolume.cpp | 6 +- indra/llmath/llvolumeoctree.cpp | 6 +- indra/llmath/llvolumeoctree.h | 9 ++- indra/newview/lldrawable.cpp | 4 + indra/newview/lldrawable.h | 4 + indra/newview/llface.cpp | 13 +-- indra/newview/llspatialpartition.cpp | 42 +++++----- indra/newview/llspatialpartition.h | 5 ++ indra/newview/llvograss.cpp | 2 +- indra/newview/llvopartgroup.cpp | 2 +- indra/newview/llvovolume.cpp | 10 +-- indra/newview/pipeline.cpp | 14 ++-- 17 files changed, 168 insertions(+), 117 deletions(-) (limited to 'indra/llmath') diff --git a/indra/llcharacter/llcharacter.cpp b/indra/llcharacter/llcharacter.cpp index 0a6a8f9fa6..07f0baae51 100644 --- a/indra/llcharacter/llcharacter.cpp +++ b/indra/llcharacter/llcharacter.cpp @@ -187,20 +187,14 @@ void LLCharacter::requestStopMotion( LLMotion* motion) //----------------------------------------------------------------------------- // updateMotions() //----------------------------------------------------------------------------- -static LLFastTimer::DeclareTimer FTM_UPDATE_ANIMATION("Update Animation"); -static LLFastTimer::DeclareTimer FTM_UPDATE_HIDDEN_ANIMATION("Update Hidden Anim"); -static LLFastTimer::DeclareTimer FTM_UPDATE_MOTIONS("Update Motions"); - void LLCharacter::updateMotions(e_update_t update_type) { if (update_type == HIDDEN_UPDATE) { - LLFastTimer t(FTM_UPDATE_HIDDEN_ANIMATION); mMotionController.updateMotionsMinimal(); } else { - LLFastTimer t(FTM_UPDATE_ANIMATION); // unpause if the number of outstanding pause requests has dropped to the initial one if (mMotionController.isPaused() && mPauseRequest->getNumRefs() == 1) { @@ -208,7 +202,6 @@ void LLCharacter::updateMotions(e_update_t update_type) } bool force_update = (update_type == FORCE_UPDATE); { - LLFastTimer t(FTM_UPDATE_MOTIONS); mMotionController.updateMotions(force_update); } } diff --git a/indra/llcharacter/lleditingmotion.cpp b/indra/llcharacter/lleditingmotion.cpp index 830e323476..66b3c2bd25 100644 --- a/indra/llcharacter/lleditingmotion.cpp +++ b/indra/llcharacter/lleditingmotion.cpp @@ -158,14 +158,11 @@ BOOL LLEditingMotion::onActivate() return TRUE; } -static LLFastTimer::DeclareTimer FTM_EDITING_MOTION("Editing Motion"); - //----------------------------------------------------------------------------- // LLEditingMotion::onUpdate() //----------------------------------------------------------------------------- BOOL LLEditingMotion::onUpdate(F32 time, U8* joint_mask) { - LLFastTimer t(FTM_EDITING_MOTION); LLVector3 focus_pt; LLVector3* pointAtPt = (LLVector3*)mCharacter->getAnimationData("PointAtPoint"); diff --git a/indra/llcharacter/llkeyframewalkmotion.cpp b/indra/llcharacter/llkeyframewalkmotion.cpp index ea63f9f8dc..d52eb89a5c 100644 --- a/indra/llcharacter/llkeyframewalkmotion.cpp +++ b/indra/llcharacter/llkeyframewalkmotion.cpp @@ -196,15 +196,11 @@ BOOL LLWalkAdjustMotion::onActivate() return TRUE; } -static LLFastTimer::DeclareTimer FTM_WALK_ADJUST_MOTION("Walk Adjust"); - //----------------------------------------------------------------------------- // LLWalkAdjustMotion::onUpdate() //----------------------------------------------------------------------------- BOOL LLWalkAdjustMotion::onUpdate(F32 time, U8* joint_mask) { - LLFastTimer t(FTM_WALK_ADJUST_MOTION); - // delta_time is guaranteed to be non zero F32 delta_time = llclamp(time - mLastTime, TIME_EPSILON, MAX_TIME_DELTA); mLastTime = time; diff --git a/indra/llcharacter/llmotioncontroller.cpp b/indra/llcharacter/llmotioncontroller.cpp index e7ec83efc0..236b8c616d 100644 --- a/indra/llcharacter/llmotioncontroller.cpp +++ b/indra/llcharacter/llmotioncontroller.cpp @@ -539,14 +539,11 @@ void LLMotionController::updateIdleActiveMotions() } } -static LLFastTimer::DeclareTimer FTM_UPDATE_MOTIONS_BY_TYPE("Update Motions By Type"); - //----------------------------------------------------------------------------- // updateMotionsByType() //----------------------------------------------------------------------------- void LLMotionController::updateMotionsByType(LLMotion::LLMotionBlendType anim_type) { - LLFastTimer t(FTM_UPDATE_MOTIONS_BY_TYPE); BOOL update_result = TRUE; U8 last_joint_signature[LL_CHARACTER_MAX_JOINTS]; @@ -798,9 +795,6 @@ void LLMotionController::updateLoadingMotions() // call updateMotion() or updateMotionsMinimal() every frame //----------------------------------------------------------------------------- -static LLFastTimer::DeclareTimer FTM_UPDATE_MOTION_PURGE_EXCESS("Purge Excess Motions"); -static LLFastTimer::DeclareTimer FTM_UPDATE_LOADING_MOTIONS("Update Loading Motions"); - //----------------------------------------------------------------------------- // updateMotion() //----------------------------------------------------------------------------- @@ -814,12 +808,9 @@ void LLMotionController::updateMotions(bool force_update) mPrevTimerElapsed = cur_time; mLastTime = mAnimTime; - { - LLFastTimer t(FTM_UPDATE_MOTION_PURGE_EXCESS); - // Always cap the number of loaded motions - purgeExcessMotions(); - } - + // Always cap the number of loaded motions + purgeExcessMotions(); + // Update timing info for this time step. if (!mPaused) { @@ -840,11 +831,8 @@ void LLMotionController::updateMotions(bool force_update) mLastInterp = interp; } - { - LLFastTimer t(FTM_UPDATE_LOADING_MOTIONS); - updateLoadingMotions(); - } - + updateLoadingMotions(); + return; } @@ -862,11 +850,8 @@ void LLMotionController::updateMotions(bool force_update) } } - { - LLFastTimer t(FTM_UPDATE_LOADING_MOTIONS); - updateLoadingMotions(); - } - + updateLoadingMotions(); + resetJointSignatures(); if (mPaused && !force_update) diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h index 6c7744cdf1..a7e176fd51 100644 --- a/indra/llmath/lloctree.h +++ b/indra/llmath/lloctree.h @@ -79,9 +79,9 @@ public: typedef LLOctreeTraveler oct_traveler; typedef LLTreeTraveler tree_traveler; - typedef typename std::set > element_list; - typedef typename element_list::iterator element_iter; - typedef typename element_list::const_iterator const_element_iter; + typedef LLPointer* element_list; + typedef LLPointer* element_iter; + typedef const LLPointer* const_element_iter; typedef typename std::vector*>::iterator tree_listener_iter; typedef typename std::vector* > child_list; typedef LLTreeNode BaseType; @@ -105,6 +105,9 @@ public: : mParent((oct_node*)parent), mOctant(octant) { + mData = NULL; + mDataEnd = NULL; + mCenter = center; mSize = size; @@ -123,6 +126,16 @@ public: { BaseType::destroyListeners(); + for (U32 i = 0; i < mElementCount; ++i) + { + mData[i]->setBinIndex(-1); + mData[i] = NULL; + } + + free(mData); + mData = NULL; + mDataEnd = NULL; + for (U32 i = 0; i < getChildCount(); i++) { delete getChild(i); @@ -222,9 +235,14 @@ public: virtual bool isLeaf() const { return mChild.empty(); } 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; } + element_iter getDataEnd() { return mDataEnd; } + const_element_iter getDataBegin() const { return mData; } + 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 +307,7 @@ 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 +320,16 @@ 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()); + mElementCount++; + mData = (element_list) realloc(mData, sizeof(LLPointer)*mElementCount); - mData.insert(data); - BaseType::insert(data); + //avoid unref on uninitialized memory + memset(mData+mElementCount-1, 0, sizeof(LLPointer)); - mElementCount = mData.size(); + mData[mElementCount-1] = data; + mDataEnd = mData + mElementCount; + data->setBinIndex(mElementCount-1); + BaseType::insert(data); return true; } else @@ -342,10 +363,16 @@ public: if( lt == 0x7 ) { - mData.insert(data); - BaseType::insert(data); + mElementCount++; + mData = (element_list) realloc(mData, sizeof(LLPointer)*mElementCount); + + //avoid unref on uninitialized memory + memset(mData+mElementCount-1, 0, sizeof(LLPointer)); - mElementCount = mData.size(); + mData[mElementCount-1] = data; + mDataEnd = mData + mElementCount; + data->setBinIndex(mElementCount-1); + BaseType::insert(data); return true; } @@ -394,23 +421,59 @@ 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; //needed for unref + mData = (element_list) realloc(mData, sizeof(LLPointer)*mElementCount); + mDataEnd = mData+mElementCount; + } + else + { + mData[0] = NULL; //needed for unref + free(mData); + mData = NULL; + mDataEnd = NULL; + } + + 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; } } @@ -429,19 +492,20 @@ public: //node is now root llwarns << "!!! OCTREE REMOVING FACE 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++) @@ -606,6 +670,7 @@ protected: U32 mChildCount; element_list mData; + element_iter mDataEnd; U32 mElementCount; }; diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index ea09a3afe7..919079fa88 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -317,16 +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::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 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 void LLOctreeTriangleRayIntersect::visit(const LLOctreeNode* node) { for (LLOctreeNode::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* branch) } //children fit, check data - for (LLOctreeNode::const_element_iter iter = branch->getData().begin(); - iter != branch->getData().end(); ++iter) + for (LLOctreeNode::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 dac97b14b5..9ae34a0c4e 100644 --- a/indra/llmath/llvolumeoctree.h +++ b/indra/llmath/llvolumeoctree.h @@ -49,7 +49,7 @@ public: LLVolumeTriangle() { - + mBinIndex = -1; } LLVolumeTriangle(const LLVolumeTriangle& rhs) @@ -74,9 +74,16 @@ public: 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 diff --git a/indra/newview/lldrawable.cpp b/indra/newview/lldrawable.cpp index 108ec92f6d..4435e04b6c 100644 --- a/indra/newview/lldrawable.cpp +++ b/indra/newview/lldrawable.cpp @@ -108,6 +108,8 @@ void LLDrawable::init() mGeneration = -1; mBinRadius = 1.f; + mBinIndex = -1; + mSpatialBridge = NULL; } @@ -1087,6 +1089,8 @@ LLSpatialBridge::LLSpatialBridge(LLDrawable* root, BOOL render_by_group, U32 dat mDrawable = root; root->setSpatialBridge(this); + mBinIndex = -1; + mRenderType = mDrawable->mRenderType; mDrawableType = mDrawable->mRenderType; diff --git a/indra/newview/lldrawable.h b/indra/newview/lldrawable.h index 8c7db61502..08511bb9fc 100644 --- a/indra/newview/lldrawable.h +++ b/indra/newview/lldrawable.h @@ -120,6 +120,9 @@ public: F32 getIntensity() const { return llmin(mXform.getScale().mV[0], 4.f); } S32 getLOD() const { return mVObjp ? mVObjp->getLOD() : 1; } F32 getBinRadius() const { return mBinRadius; } + S32 getBinIndex() const { return mBinIndex; } + void setBinIndex(S32 index) const { mBinIndex = index; } + void getMinMax(LLVector3& min,LLVector3& max) const { mXform.getMinMax(min,max); } LLXformMatrix* getXform() { return &mXform; } @@ -326,6 +329,7 @@ private: mutable U32 mVisible; F32 mRadius; F32 mBinRadius; + mutable S32 mBinIndex; S32 mGeneration; LLVector3 mCurrentScale; diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 02a8d1244f..eea5161538 100755 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -163,15 +163,10 @@ void LLFace::init(LLDrawable* drawablep, LLViewerObject* objp) mGeomCount = 0; mGeomIndex = 0; mIndicesCount = 0; - if (drawablep->getRenderType() == LLPipeline::RENDER_TYPE_PARTICLES || - drawablep->getRenderType() == LLPipeline::RENDER_TYPE_HUD_PARTICLES) - { //indicate to LLParticlePartition that this particle is uninitialized - mIndicesIndex = 0xFFFFFFFF; - } - else - { - mIndicesIndex = 0; - } + + //special value to indicate uninitialized position + mIndicesIndex = 0xFFFFFFFF; + mIndexInTex = 0; mTexture = NULL; mTEOffset = -1; diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp index ccc5dc7bea..a71529a3bd 100644 --- a/indra/newview/llspatialpartition.cpp +++ b/indra/newview/llspatialpartition.cpp @@ -542,7 +542,7 @@ void LLSpatialGroup::validate() validateDrawMap(); - for (element_iter i = getData().begin(); i != getData().end(); ++i) + for (element_iter i = getDataBegin(); i != getDataEnd(); ++i) { LLDrawable* drawable = *i; sg_assert(drawable->getSpatialGroup() == this); @@ -759,7 +759,7 @@ BOOL LLSpatialGroup::boundObjects(BOOL empty, LLVector4a& minOut, LLVector4a& ma { const OctreeNode* node = mOctreeNode; - if (node->getData().empty()) + if (node->isEmpty()) { //don't do anything if there are no objects if (empty && mOctreeNode->getParent()) { //only root is allowed to be empty @@ -776,14 +776,14 @@ BOOL LLSpatialGroup::boundObjects(BOOL empty, LLVector4a& minOut, LLVector4a& ma clearState(OBJECT_DIRTY); //initialize bounding box to first element - OctreeNode::const_element_iter i = node->getData().begin(); + OctreeNode::const_element_iter i = node->getDataBegin(); LLDrawable* drawablep = *i; const LLVector4a* minMax = drawablep->getSpatialExtents(); newMin = minMax[0]; newMax = minMax[1]; - for (++i; i != node->getData().end(); ++i) + for (++i; i != node->getDataEnd(); ++i) { drawablep = *i; minMax = drawablep->getSpatialExtents(); @@ -1247,7 +1247,7 @@ void LLSpatialGroup::updateDistance(LLCamera &camera) llerrs << "Spatial group dirty on distance update." << llendl; } #endif - if (!getData().empty()) + if (!isEmpty()) { mRadius = mSpatialPartition->mRenderByGroup ? mObjectBounds[1].getLength3().getF32() : (F32) mOctreeNode->getSize().getLength3().getF32(); @@ -1398,7 +1398,7 @@ void LLSpatialGroup::handleDestruction(const TreeNode* node) LLMemType mt(LLMemType::MTYPE_SPACE_PARTITION); setState(DEAD); - for (element_iter i = getData().begin(); i != getData().end(); ++i) + for (element_iter i = getDataBegin(); i != getDataEnd(); ++i) { LLDrawable* drawable = *i; if (drawable->getSpatialGroup() == this) @@ -1485,7 +1485,7 @@ void LLSpatialGroup::destroyGL(bool keep_occlusion) } - for (LLSpatialGroup::element_iter i = getData().begin(); i != getData().end(); ++i) + for (LLSpatialGroup::element_iter i = getDataBegin(); i != getDataEnd(); ++i) { LLDrawable* drawable = *i; for (S32 j = 0; j < drawable->getNumFaces(); j++) @@ -2117,7 +2117,7 @@ public: virtual void processGroup(LLSpatialGroup* group) { - llassert(!group->isState(LLSpatialGroup::DIRTY) && !group->getData().empty()) + llassert(!group->isState(LLSpatialGroup::DIRTY) && !group->isEmpty()) if (mRes < 2) { @@ -2184,7 +2184,7 @@ public: { LLSpatialGroup::OctreeNode* branch = group->mOctreeNode; - for (LLSpatialGroup::OctreeNode::const_element_iter i = branch->getData().begin(); i != branch->getData().end(); ++i) + for (LLSpatialGroup::OctreeNode::const_element_iter i = branch->getDataBegin(); i != branch->getDataEnd(); ++i) { LLDrawable* drawable = *i; @@ -2308,7 +2308,7 @@ public: LLSpatialGroup* group = (LLSpatialGroup*) state->getListener(0); group->destroyGL(); - for (LLSpatialGroup::element_iter i = group->getData().begin(); i != group->getData().end(); ++i) + for (LLSpatialGroup::element_iter i = group->getDataBegin(); i != group->getDataEnd(); ++i) { LLDrawable* drawable = *i; if (drawable->getVObj().notNull() && !group->mSpatialPartition->mRenderByGroup) @@ -2618,7 +2618,7 @@ void renderOctree(LLSpatialGroup* group) gGL.flush(); glLineWidth(1.f); gGL.flush(); - for (LLSpatialGroup::element_iter i = group->getData().begin(); i != group->getData().end(); ++i) + for (LLSpatialGroup::element_iter i = group->getDataBegin(); i != group->getDataEnd(); ++i) { LLDrawable* drawable = *i; if (!group->mSpatialPartition->isBridge()) @@ -2664,7 +2664,7 @@ void renderOctree(LLSpatialGroup* group) } else { - if (group->mBufferUsage == GL_STATIC_DRAW_ARB && !group->getData().empty() + if (group->mBufferUsage == GL_STATIC_DRAW_ARB && !group->isEmpty() && group->mSpatialPartition->mRenderByGroup) { col.setVec(0.8f, 0.4f, 0.1f, 0.1f); @@ -2732,7 +2732,7 @@ void renderVisibility(LLSpatialGroup* group, LLCamera* camera) glPolygonMode(GL_FRONT_AND_BACK, GL_LINE); BOOL render_objects = (!LLPipeline::sUseOcclusion || !group->isOcclusionState(LLSpatialGroup::OCCLUDED)) && group->isVisible() && - !group->getData().empty(); + !group->isEmpty(); if (render_objects) { @@ -3463,7 +3463,7 @@ void renderPhysicsShape(LLDrawable* drawable, LLVOVolume* volume) void renderPhysicsShapes(LLSpatialGroup* group) { - for (LLSpatialGroup::OctreeNode::const_element_iter i = group->getData().begin(); i != group->getData().end(); ++i) + for (LLSpatialGroup::OctreeNode::const_element_iter i = group->getDataBegin(); i != group->getDataEnd(); ++i) { LLDrawable* drawable = *i; LLVOVolume* volume = drawable->getVOVolume(); @@ -3812,7 +3812,7 @@ public: LLVector3 center, size; - if (branch->getData().empty()) + if (branch->isEmpty()) { gGL.diffuseColor3f(1.f,0.2f,0.f); center.set(branch->getCenter().getF32ptr()); @@ -3848,8 +3848,8 @@ public: } gGL.begin(LLRender::TRIANGLES); - for (LLOctreeNode::const_element_iter iter = branch->getData().begin(); - iter != branch->getData().end(); + for (LLOctreeNode::const_element_iter iter = branch->getDataBegin(); + iter != branch->getDataEnd(); ++iter) { const LLVolumeTriangle* tri = *iter; @@ -4086,7 +4086,7 @@ public: if (gPipeline.hasRenderDebugMask(LLPipeline::RENDER_DEBUG_BBOXES)) { - if (!group->getData().empty()) + if (!group->isEmpty()) { gGL.diffuseColor3f(0,0,1); drawBoxOutline(group->mObjectBounds[0], @@ -4094,7 +4094,7 @@ public: } } - for (LLSpatialGroup::OctreeNode::const_element_iter i = branch->getData().begin(); i != branch->getData().end(); ++i) + for (LLSpatialGroup::OctreeNode::const_element_iter i = branch->getDataBegin(); i != branch->getDataEnd(); ++i) { LLDrawable* drawable = *i; @@ -4283,7 +4283,7 @@ public: return; } - for (LLSpatialGroup::OctreeNode::const_element_iter i = branch->getData().begin(); i != branch->getData().end(); ++i) + for (LLSpatialGroup::OctreeNode::const_element_iter i = branch->getDataBegin(); i != branch->getDataEnd(); ++i) { LLDrawable* drawable = *i; @@ -4507,7 +4507,7 @@ public: virtual void visit(const LLSpatialGroup::OctreeNode* branch) { - for (LLSpatialGroup::OctreeNode::const_element_iter i = branch->getData().begin(); i != branch->getData().end(); ++i) + for (LLSpatialGroup::OctreeNode::const_element_iter i = branch->getDataBegin(); i != branch->getDataEnd(); ++i) { check(*i); } diff --git a/indra/newview/llspatialpartition.h b/indra/newview/llspatialpartition.h index d935c1f140..f050df2b39 100644 --- a/indra/newview/llspatialpartition.h +++ b/indra/newview/llspatialpartition.h @@ -345,8 +345,13 @@ public: void dirtyGeom() { setState(GEOM_DIRTY); } void dirtyMesh() { setState(MESH_DIRTY); } + + //octree wrappers to make code more readable element_list& getData() { return mOctreeNode->getData(); } + element_iter getDataBegin() { return mOctreeNode->getDataBegin(); } + element_iter getDataEnd() { return mOctreeNode->getDataEnd(); } U32 getElementCount() const { return mOctreeNode->getElementCount(); } + bool isEmpty() const { return mOctreeNode->isEmpty(); } void drawObjectBox(LLColor4 col); diff --git a/indra/newview/llvograss.cpp b/indra/newview/llvograss.cpp index 44968342bf..5ad9ccc9af 100644 --- a/indra/newview/llvograss.cpp +++ b/indra/newview/llvograss.cpp @@ -622,7 +622,7 @@ void LLGrassPartition::addGeometryCount(LLSpatialGroup* group, U32& vertex_count mFaceList.clear(); LLViewerCamera* camera = LLViewerCamera::getInstance(); - for (LLSpatialGroup::element_iter i = group->getData().begin(); i != group->getData().end(); ++i) + for (LLSpatialGroup::element_iter i = group->getDataBegin(); i != group->getDataEnd(); ++i) { LLDrawable* drawablep = *i; diff --git a/indra/newview/llvopartgroup.cpp b/indra/newview/llvopartgroup.cpp index e21358b65a..02282fd6a6 100644 --- a/indra/newview/llvopartgroup.cpp +++ b/indra/newview/llvopartgroup.cpp @@ -554,7 +554,7 @@ void LLParticlePartition::addGeometryCount(LLSpatialGroup* group, U32& vertex_co mFaceList.clear(); LLViewerCamera* camera = LLViewerCamera::getInstance(); - for (LLSpatialGroup::element_iter i = group->getData().begin(); i != group->getData().end(); ++i) + for (LLSpatialGroup::element_iter i = group->getDataBegin(); i != group->getDataEnd(); ++i) { LLDrawable* drawablep = *i; diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index 0abe73d55e..d1c668f984 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -4251,7 +4251,7 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group) LLFastTimer t(FTM_REBUILD_VOLUME_FACE_LIST); //get all the faces into a list - for (LLSpatialGroup::element_iter drawable_iter = group->getData().begin(); drawable_iter != group->getData().end(); ++drawable_iter) + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { LLDrawable* drawablep = *drawable_iter; @@ -4657,7 +4657,7 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group) if (!LLPipeline::sDelayVBUpdate) { //drawables have been rebuilt, clear rebuild status - for (LLSpatialGroup::element_iter drawable_iter = group->getData().begin(); drawable_iter != group->getData().end(); ++drawable_iter) + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { LLDrawable* drawablep = *drawable_iter; drawablep->clearState(LLDrawable::REBUILD_ALL); @@ -4697,7 +4697,7 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group) std::set mapped_buffers; - for (LLSpatialGroup::element_iter drawable_iter = group->getData().begin(); drawable_iter != group->getData().end(); ++drawable_iter) + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { LLDrawable* drawablep = *drawable_iter; @@ -4759,7 +4759,7 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group) if(num_mapped_veretx_buffer != LLVertexBuffer::sMappedCount) { llwarns << "Not all mapped vertex buffers are unmapped!" << llendl ; - for (LLSpatialGroup::element_iter drawable_iter = group->getData().begin(); drawable_iter != group->getData().end(); ++drawable_iter) + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { LLDrawable* drawablep = *drawable_iter; for (S32 i = 0; i < drawablep->getNumFaces(); ++i) @@ -5227,7 +5227,7 @@ void LLGeometryManager::addGeometryCount(LLSpatialGroup* group, U32 &vertex_coun //for each drawable - for (LLSpatialGroup::element_iter drawable_iter = group->getData().begin(); drawable_iter != group->getData().end(); ++drawable_iter) + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { LLDrawable* drawablep = *drawable_iter; diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index ec0e200ae0..49a9b31c14 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -1371,7 +1371,7 @@ public: { LLSpatialGroup* group = (LLSpatialGroup*) node->getListener(0); - if (!group->isState(LLSpatialGroup::GEOM_DIRTY) && !group->getData().empty()) + if (!group->isState(LLSpatialGroup::GEOM_DIRTY) && !group->isEmpty()) { for (LLSpatialGroup::draw_map_t::iterator i = group->mDrawMap.begin(); i != group->mDrawMap.end(); ++i) { @@ -1976,7 +1976,7 @@ void LLPipeline::clearReferences() void check_references(LLSpatialGroup* group, LLDrawable* drawable) { - for (LLSpatialGroup::element_iter i = group->getData().begin(); i != group->getData().end(); ++i) + for (LLSpatialGroup::element_iter i = group->getDataBegin(); i != group->getDataEnd(); ++i) { if (drawable == *i) { @@ -1998,7 +1998,7 @@ void check_references(LLDrawable* drawable, LLFace* face) void check_references(LLSpatialGroup* group, LLFace* face) { - for (LLSpatialGroup::element_iter i = group->getData().begin(); i != group->getData().end(); ++i) + for (LLSpatialGroup::element_iter i = group->getDataBegin(); i != group->getDataEnd(); ++i) { LLDrawable* drawable = *i; check_references(drawable, face); @@ -2374,7 +2374,7 @@ void LLPipeline::updateCull(LLCamera& camera, LLCullResult& result, S32 water_cl void LLPipeline::markNotCulled(LLSpatialGroup* group, LLCamera& camera) { - if (group->getData().empty()) + if (group->isEmpty()) { return; } @@ -3004,7 +3004,7 @@ void LLPipeline::stateSort(LLCamera& camera, LLCullResult &result) else { group->setVisible(); - for (LLSpatialGroup::element_iter i = group->getData().begin(); i != group->getData().end(); ++i) + for (LLSpatialGroup::element_iter i = group->getDataBegin(); i != group->getDataEnd(); ++i) { markVisible(*i, camera); } @@ -3092,7 +3092,7 @@ void LLPipeline::stateSort(LLSpatialGroup* group, LLCamera& camera) LLMemType mt(LLMemType::MTYPE_PIPELINE_STATE_SORT); if (group->changeLOD()) { - for (LLSpatialGroup::element_iter i = group->getData().begin(); i != group->getData().end(); ++i) + for (LLSpatialGroup::element_iter i = group->getDataBegin(); i != group->getDataEnd(); ++i) { LLDrawable* drawablep = *i; stateSort(drawablep, camera); @@ -3215,7 +3215,7 @@ void forAllDrawables(LLCullResult::sg_iterator begin, { for (LLCullResult::sg_iterator i = begin; i != end; ++i) { - for (LLSpatialGroup::element_iter j = (*i)->getData().begin(); j != (*i)->getData().end(); ++j) + for (LLSpatialGroup::element_iter j = (*i)->getDataBegin(); j != (*i)->getDataEnd(); ++j) { func(*j); } -- cgit v1.3 From 4127f3e7fcc725dbbf500c52605c5b950073c828 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 20 Sep 2012 09:48:56 -0400 Subject: reapply efcec3eb374f: MAINT-646 Factor std::vector out of lloctree --- indra/llmath/lloctree.h | 18 +++++++++++------- indra/llmath/llvolume.cpp | 2 +- 2 files changed, 12 insertions(+), 8 deletions(-) (limited to 'indra/llmath') diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h index a7e176fd51..c3f6f7de2a 100644 --- a/indra/llmath/lloctree.h +++ b/indra/llmath/lloctree.h @@ -31,7 +31,6 @@ #include "v3math.h" #include "llvector4a.h" #include -#include #define OCT_ERRS LL_WARNS("OctreeErrors") @@ -83,7 +82,9 @@ public: typedef LLPointer* element_iter; typedef const LLPointer* const_element_iter; typedef typename std::vector*>::iterator tree_listener_iter; - typedef typename std::vector* > child_list; + typedef LLOctreeNode** child_list; + typedef LLOctreeNode** child_iter; + typedef LLTreeNode BaseType; typedef LLOctreeNode oct_node; typedef LLOctreeListener oct_listener; @@ -232,7 +233,7 @@ 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; } @@ -517,8 +518,8 @@ public: void clearChildren() { - mChild.clear(); mChildCount = 0; + U32* foo = (U32*) mChildMap; foo[0] = foo[1] = 0xFFFFFFFF; } @@ -580,7 +581,7 @@ public: mChildMap[child->getOctant()] = mChildCount; - mChild.push_back(child); + mChild[mChildCount] = child; ++mChildCount; child->setParent(this); @@ -607,9 +608,12 @@ 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; @@ -665,7 +669,7 @@ protected: oct_node* mParent; U8 mOctant; - child_list mChild; + LLOctreeNode* mChild[8]; U8 mChildMap[8]; U32 mChildCount; diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 919079fa88..85ea14f9bc 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -341,7 +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); -- cgit v1.3