summaryrefslogtreecommitdiff
path: root/indra/llmath
diff options
context:
space:
mode:
Diffstat (limited to 'indra/llmath')
-rw-r--r--indra/llmath/llmatrix4a.h41
-rw-r--r--indra/llmath/lloctree.h13
-rw-r--r--indra/llmath/llrigginginfo.h29
-rw-r--r--indra/llmath/llvector4a.h11
-rw-r--r--indra/llmath/llvector4a.inl4
-rw-r--r--indra/llmath/llvolume.cpp25
-rw-r--r--indra/llmath/llvolumeoctree.h27
-rw-r--r--indra/llmath/m4math.cpp9
-rw-r--r--indra/llmath/m4math.h2
-rw-r--r--indra/llmath/v3math.cpp6
-rw-r--r--indra/llmath/v3math.h5
11 files changed, 101 insertions, 71 deletions
diff --git a/indra/llmath/llmatrix4a.h b/indra/llmath/llmatrix4a.h
index 7ba347062f..5291a05607 100644
--- a/indra/llmath/llmatrix4a.h
+++ b/indra/llmath/llmatrix4a.h
@@ -36,6 +36,26 @@ class LLMatrix4a
public:
LL_ALIGN_16(LLVector4a mMatrix[4]);
+ LLMatrix4a()
+ {
+
+ }
+
+ explicit LLMatrix4a(const LLMatrix4& val)
+ {
+ loadu(val);
+ }
+
+ inline F32* getF32ptr()
+ {
+ return (F32*) &mMatrix;
+ }
+
+ inline const F32* getF32ptr() const
+ {
+ return (F32*)&mMatrix;
+ }
+
inline void clear()
{
mMatrix[0].clear();
@@ -44,6 +64,14 @@ public:
mMatrix[3].clear();
}
+ inline void setIdentity()
+ {
+ mMatrix[0].set(1.f, 0.f, 0.f, 0.f);
+ mMatrix[1].set(0.f, 1.f, 0.f, 0.f);
+ mMatrix[2].set(0.f, 0.f, 1.f, 0.f);
+ mMatrix[3].set(0.f, 0.f, 0.f, 1.f);
+ }
+
inline void loadu(const LLMatrix4& src)
{
mMatrix[0] = _mm_loadu_ps(src.mMatrix[0]);
@@ -105,7 +133,7 @@ public:
mMatrix[3].setAdd(a.mMatrix[3],d3);
}
- inline void rotate(const LLVector4a& v, LLVector4a& res)
+ inline void rotate(const LLVector4a& v, LLVector4a& res) const
{
LLVector4a y,z;
@@ -151,6 +179,8 @@ public:
{
affineTransformSSE(v,res);
}
+
+ const LLVector4a& getTranslation() const { return mMatrix[3]; }
};
inline LLVector4a rowMul(const LLVector4a &row, const LLMatrix4a &mat)
@@ -176,6 +206,15 @@ inline void matMul(const LLMatrix4a &a, const LLMatrix4a &b, LLMatrix4a &res)
res.mMatrix[3] = row3;
}
+//Faster version of matMul wehere res must not be a or b
+inline void matMulUnsafe(const LLMatrix4a &a, const LLMatrix4a &b, LLMatrix4a &res)
+{
+ res.mMatrix[0] = rowMul(a.mMatrix[0], b);
+ res.mMatrix[1] = rowMul(a.mMatrix[1], b);
+ res.mMatrix[2] = rowMul(a.mMatrix[2], b);
+ res.mMatrix[3] = rowMul(a.mMatrix[3], b);
+}
+
inline std::ostream& operator<<(std::ostream& s, const LLMatrix4a& m)
{
s << "[" << m.mMatrix[0] << ", " << m.mMatrix[1] << ", " << m.mMatrix[2] << ", " << m.mMatrix[3] << "]";
diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h
index 0e2f62f9db..8c4a1304b4 100644
--- a/indra/llmath/lloctree.h
+++ b/indra/llmath/lloctree.h
@@ -74,8 +74,9 @@ public:
};
template <class T>
-class LLOctreeNode : public LLTreeNode<T>
+class alignas(16) LLOctreeNode : public LLTreeNode<T>
{
+ LL_ALIGN_NEW
public:
typedef LLOctreeTraveler<T> oct_traveler;
@@ -91,16 +92,6 @@ public:
typedef LLOctreeNode<T> oct_node;
typedef LLOctreeListener<T> oct_listener;
- void* operator new(size_t size)
- {
- return ll_aligned_malloc_16(size);
- }
-
- void operator delete(void* ptr)
- {
- ll_aligned_free_16(ptr);
- }
-
LLOctreeNode( const LLVector4a& center,
const LLVector4a& size,
BaseType* parent,
diff --git a/indra/llmath/llrigginginfo.h b/indra/llmath/llrigginginfo.h
index b3d6bc2d19..059c6ae082 100644
--- a/indra/llmath/llrigginginfo.h
+++ b/indra/llmath/llrigginginfo.h
@@ -34,9 +34,9 @@
// Extents are in joint space
// isRiggedTo is based on the state of all currently associated rigged meshes
-LL_ALIGN_PREFIX(16)
-class LLJointRiggingInfo
+class alignas(16) LLJointRiggingInfo
{
+ LL_ALIGN_NEW
public:
LLJointRiggingInfo();
bool isRiggedTo() const;
@@ -45,31 +45,10 @@ public:
const LLVector4a *getRiggedExtents() const;
void merge(const LLJointRiggingInfo& other);
- void* operator new(size_t size)
- {
- return ll_aligned_malloc_16(size);
- }
-
- void operator delete(void* ptr)
- {
- ll_aligned_free_16(ptr);
- }
-
- void* operator new[](size_t size)
- {
- return ll_aligned_malloc_16(size);
- }
-
- void operator delete[](void* ptr)
- {
- ll_aligned_free_16(ptr);
- }
-
-
private:
- LL_ALIGN_16(LLVector4a mRiggedExtents[2]);
+ LLVector4a mRiggedExtents[2];
bool mIsRiggedTo;
-} LL_ALIGN_POSTFIX(16);
+};
// For storing all the rigging info associated with a given avatar or
// object, keyed by joint_num.
diff --git a/indra/llmath/llvector4a.h b/indra/llmath/llvector4a.h
index 27abf39537..53c8f604f6 100644
--- a/indra/llmath/llvector4a.h
+++ b/indra/llmath/llvector4a.h
@@ -46,11 +46,10 @@ class LLRotation;
// of this writing, July 08, 2010) about getting it implemented before you resort to
// LLVector3/LLVector4.
/////////////////////////////////
-struct LLVector4a;
-LL_ALIGN_PREFIX(16)
-struct LLVector4a
+class alignas(16) LLVector4a
{
+ LL_ALIGN_NEW
public:
///////////////////////////////////
@@ -138,10 +137,10 @@ public:
// BASIC GET/SET
////////////////////////////////////
- // Return a "this" as an F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon)
+ // Return a "this" as an F32 pointer.
inline F32* getF32ptr();
- // Return a "this" as a const F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon)
+ // Return a "this" as a const F32 pointer.
inline const F32* const getF32ptr() const;
// Read-only access a single float in this vector. Do not use in proximity to any function call that manipulates
@@ -324,7 +323,7 @@ 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 69d3d01efe..8be1c1b114 100644
--- a/indra/llmath/llvector4a.inl
+++ b/indra/llmath/llvector4a.inl
@@ -58,13 +58,13 @@ inline void LLVector4a::store4a(F32* dst) const
// BASIC GET/SET
////////////////////////////////////
-// Return a "this" as an F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon)
+// Return a "this" as an F32 pointer.
F32* LLVector4a::getF32ptr()
{
return (F32*) &mQ;
}
-// Return a "this" as a const F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon)
+// Return a "this" as a const F32 pointer.
const F32* const LLVector4a::getF32ptr() const
{
return (const F32* const) &mQ;
diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp
index e085fa6ada..130f30bedc 100644
--- a/indra/llmath/llvolume.cpp
+++ b/indra/llmath/llvolume.cpp
@@ -383,6 +383,7 @@ public:
virtual void visit(const LLOctreeNode<LLVolumeTriangle>* branch)
{ //this is a depth first traversal, so it's safe to assum all children have complete
//bounding data
+ LL_PROFILE_ZONE_SCOPED
LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0);
@@ -822,6 +823,8 @@ S32 LLProfile::getNumPoints(const LLProfileParams& params, BOOL path_open,F32 de
BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detail, S32 split,
BOOL is_sculpted, S32 sculpt_size)
{
+ LL_PROFILE_ZONE_SCOPED
+
if ((!mDirty) && (!is_sculpted))
{
return FALSE;
@@ -1302,6 +1305,8 @@ S32 LLPath::getNumNGonPoints(const LLPathParams& params, S32 sides, F32 startOff
void LLPath::genNGon(const LLPathParams& params, S32 sides, F32 startOff, F32 end_scale, F32 twist_scale)
{
+ LL_PROFILE_ZONE_SCOPED
+
// Generates a circular path, starting at (1, 0, 0), counterclockwise along the xz plane.
static const F32 tableScale[] = { 1, 1, 1, 0.5f, 0.707107f, 0.53f, 0.525f, 0.5f };
@@ -1536,6 +1541,8 @@ S32 LLPath::getNumPoints(const LLPathParams& params, F32 detail)
BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split,
BOOL is_sculpted, S32 sculpt_size)
{
+ LL_PROFILE_ZONE_SCOPED
+
if ((!mDirty) && (!is_sculpted))
{
return FALSE;
@@ -2112,6 +2119,8 @@ LLVolume::~LLVolume()
BOOL LLVolume::generate()
{
+ LL_PROFILE_ZONE_SCOPED
+
LL_CHECK_MEMORY
llassert_always(mProfilep);
@@ -2370,6 +2379,8 @@ bool LLVolumeFace::VertexData::compareNormal(const LLVolumeFace::VertexData& rhs
bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size)
{
+ LL_PROFILE_ZONE_SCOPED
+
//input stream is now pointing at a zlib compressed block of LLSD
//decompress block
LLSD mdl;
@@ -2755,6 +2766,8 @@ S32 LLVolume::getNumFaces() const
void LLVolume::createVolumeFaces()
{
+ LL_PROFILE_ZONE_SCOPED
+
if (mGenerateSingleFace)
{
// do nothing
@@ -3720,6 +3733,8 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices,
const LLMatrix3& norm_mat_in,
S32 face_mask)
{
+ LL_PROFILE_ZONE_SCOPED
+
LLMatrix4a mat;
mat.loadu(mat_in);
@@ -4846,6 +4861,8 @@ void LLVolumeFace::freeData()
BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build)
{
+ LL_PROFILE_ZONE_SCOPED
+
//tree for this face is no longer valid
delete mOctree;
mOctree = NULL;
@@ -5514,6 +5531,8 @@ bool LLVolumeFace::cacheOptimize()
void LLVolumeFace::createOctree(F32 scaler, const LLVector4a& center, const LLVector4a& size)
{
+ LL_PROFILE_ZONE_SCOPED
+
if (mOctree)
{
return;
@@ -6287,6 +6306,8 @@ void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVe
void LLVolumeFace::createTangents()
{
+ LL_PROFILE_ZONE_SCOPED
+
if (!mTangents)
{
allocateTangents(mNumVertices);
@@ -6482,6 +6503,8 @@ void LLVolumeFace::fillFromLegacyData(std::vector<LLVolumeFace::VertexData>& v,
BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build)
{
+ LL_PROFILE_ZONE_SCOPED
+
LL_CHECK_MEMORY
BOOL flat = mTypeMask & FLAT_MASK;
@@ -6974,6 +6997,8 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build)
void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVector4a *normal,
const LLVector2 *texcoord, U32 triangleCount, const U16* index_array, LLVector4a *tangent)
{
+ LL_PROFILE_ZONE_SCOPED
+
//LLVector4a *tan1 = new LLVector4a[vertexCount * 2];
LLVector4a* tan1 = (LLVector4a*) ll_aligned_malloc_16(vertexCount*2*sizeof(LLVector4a));
// new(tan1) LLVector4a;
diff --git a/indra/llmath/llvolumeoctree.h b/indra/llmath/llvolumeoctree.h
index 13150028d8..b2bc440368 100644
--- a/indra/llmath/llvolumeoctree.h
+++ b/indra/llmath/llvolumeoctree.h
@@ -34,19 +34,10 @@
#include "llvolume.h"
#include "llvector4a.h"
-class LLVolumeTriangle : public LLRefCount
+class alignas(16) LLVolumeTriangle : public LLRefCount
{
+ LL_ALIGN_NEW
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;
@@ -86,20 +77,10 @@ public:
};
-class LLVolumeOctreeListener : public LLOctreeListener<LLVolumeTriangle>
+class alignas(16) LLVolumeOctreeListener : public LLOctreeListener<LLVolumeTriangle>
{
+ LL_ALIGN_NEW
public:
-
- void* operator new(size_t size)
- {
- return ll_aligned_malloc_16(size);
- }
-
- void operator delete(void* ptr)
- {
- ll_aligned_free_16(ptr);
- }
-
LLVolumeOctreeListener(LLOctreeNode<LLVolumeTriangle>* node);
~LLVolumeOctreeListener();
diff --git a/indra/llmath/m4math.cpp b/indra/llmath/m4math.cpp
index 3baf1bad18..6e40dae30b 100644
--- a/indra/llmath/m4math.cpp
+++ b/indra/llmath/m4math.cpp
@@ -32,8 +32,7 @@
#include "m4math.h"
#include "m3math.h"
#include "llquaternion.h"
-
-
+#include "llmatrix4a.h"
// LLMatrix4
@@ -115,6 +114,12 @@ LLMatrix4::LLMatrix4(const LLQuaternion &q)
*this = initRotation(q);
}
+LLMatrix4::LLMatrix4(const LLMatrix4a& mat)
+ : LLMatrix4(mat.getF32ptr())
+{
+
+}
+
LLMatrix4::LLMatrix4(const LLQuaternion &q, const LLVector4 &pos)
{
*this = initRotTrans(q, pos);
diff --git a/indra/llmath/m4math.h b/indra/llmath/m4math.h
index bf60adb9b6..b9da970cde 100644
--- a/indra/llmath/m4math.h
+++ b/indra/llmath/m4math.h
@@ -32,6 +32,7 @@
class LLVector4;
class LLMatrix3;
class LLQuaternion;
+class LLMatrix4a;
// NOTA BENE: Currently assuming a right-handed, x-forward, y-left, z-up universe
@@ -104,6 +105,7 @@ public:
explicit LLMatrix4(const F32 *mat); // Initializes Matrix to values in mat
explicit LLMatrix4(const LLMatrix3 &mat); // Initializes Matrix to values in mat and sets position to (0,0,0)
explicit LLMatrix4(const LLQuaternion &q); // Initializes Matrix with rotation q and sets position to (0,0,0)
+ explicit LLMatrix4(const LLMatrix4a& mat);
LLMatrix4(const LLMatrix3 &mat, const LLVector4 &pos); // Initializes Matrix to values in mat and pos
diff --git a/indra/llmath/v3math.cpp b/indra/llmath/v3math.cpp
index b04c67d926..93010d2250 100644
--- a/indra/llmath/v3math.cpp
+++ b/indra/llmath/v3math.cpp
@@ -316,6 +316,12 @@ LLVector3::LLVector3(const LLVector4 &vec)
mV[VZ] = (F32)vec.mV[VZ];
}
+LLVector3::LLVector3(const LLVector4a& vec)
+ : LLVector3(vec.getF32ptr())
+{
+
+}
+
LLVector3::LLVector3(const LLSD& sd)
{
setValue(sd);
diff --git a/indra/llmath/v3math.h b/indra/llmath/v3math.h
index 6f857d7061..068f489020 100644
--- a/indra/llmath/v3math.h
+++ b/indra/llmath/v3math.h
@@ -33,6 +33,7 @@
#include "llsd.h"
class LLVector2;
class LLVector4;
+class LLVector4a;
class LLMatrix3;
class LLMatrix4;
class LLVector3d;
@@ -62,7 +63,9 @@ class LLVector3
explicit LLVector3(const LLVector2 &vec); // Initializes LLVector3 to (vec[0]. vec[1], 0)
explicit LLVector3(const LLVector3d &vec); // Initializes LLVector3 to (vec[0]. vec[1], vec[2])
explicit LLVector3(const LLVector4 &vec); // Initializes LLVector4 to (vec[0]. vec[1], vec[2])
- explicit LLVector3(const LLSD& sd);
+ explicit LLVector3(const LLVector4a& vec); // Initializes LLVector4 to (vec[0]. vec[1], vec[2])
+ explicit LLVector3(const LLSD& sd);
+
LLSD getValue() const;
generated by cgit v1.2.3 (git 2.25.1) at 2025-02-15 03:37:34 +0000