Merge branch 'release/v2_2_0'

This commit is contained in:
manuelk 2013-09-11 18:42:10 -07:00
commit e34122223c
45 changed files with 4791 additions and 107 deletions

View File

@ -131,6 +131,12 @@ if(CMAKE_COMPILER_IS_GNUCC OR CMAKE_COMPILER_IS_CLANGCC OR CMAKE_COMPILER_IS_ICC
foreach (ICC_LIB iomp5 irng intlc)
if(CMAKE_SIZEOF_VOID_P MATCHES "8")
list(APPEND ICC_LIB_ARCH "intel64")
elseif(CMAKE_SIZEOF_VOID_P MATCHES "4")
list(APPEND ICC_LIB_ARCH "ia32")
endif()
find_library( ICC_${ICC_LIB}
NAMES
${ICC_LIB}
@ -138,9 +144,7 @@ if(CMAKE_COMPILER_IS_GNUCC OR CMAKE_COMPILER_IS_CLANGCC OR CMAKE_COMPILER_IS_ICC
${ICC_LOCATION}/lib
/opt/intel/lib/
PATH_SUFFIXES
intel64/
ia32/
mic/
${ICC_LIB_ARCH}
)
if (ICC_${ICC_LIB})

View File

@ -71,6 +71,7 @@ if (DOCUTILS_FOUND AND PYTHONINTERP_FOUND)
far_overview.rst
getting_started.rst
glviewer.rst
glstencilviewer.rst
glbatchviewer.rst
hbr_overview.rst
intro.rst

View File

@ -1,4 +1,4 @@
..
..
Copyright 2013 Pixar
Licensed under the Apache License, Version 2.0 (the "License");
@ -21,7 +21,7 @@
either express or implied. See the License for the specific
language governing permissions and limitations under the
License.
Code Examples
-------------
@ -29,7 +29,7 @@ Code Examples
Standalone Viewers
==================
OpenSubdiv builds a number of standalone viewers that demonstrate various aspects
OpenSubdiv builds a number of standalone viewers that demonstrate various aspects
of the software.
----
@ -37,17 +37,18 @@ of the software.
.. list-table:: **OpenGL examples**
:class: quickref
:widths: 50 50
* - | `glViewer <glviewer.html>`_
| `glBatchViewer <glbatchviewer.html>`_
| `glStencilViewer <glstencilviewer.html>`_
- | `limitEval <limiteval.html>`_
| `paintTest <painttest.html>`_
| `ptexViewer <ptexviewer.html>`_
- | `paintTest <painttest.html>`_
| `limitEval <limiteval.html>`_
.. list-table:: **DirectX examples**
:class: quickref
:widths: 50 50
* - | `dxViewer <dxviewer.html>`_
- |
@ -64,8 +65,8 @@ of the software.
**Note:**
the Maya plugins are currently unsupported and they may fail to compile
or work with current versions of OpenSubdiv. These were originally written for
the sole purpose of live demonstrations and the code is provided only as an
or work with current versions of OpenSubdiv. These were originally written for
the sole purpose of live demonstrations and the code is provided only as an
implementation example.
|
@ -76,13 +77,13 @@ Common Keyboard Controls
========================
.. code:: c++
Left mouse button drag : orbit camera
Middle mouse button drag : pan camera
Right mouse button : dolly camera
n, p : next/prev model
1, 2, 3, ..., 9, 0 : specify adaptive isolation or uniform refinment level
+, - : increase / decrease tessellation
+, - : increase / decrease tessellation
Tab : toggle full-screen
Esc : turn on / off the HUD
w : switch display mode

Binary file not shown.

View File

@ -65,6 +65,7 @@ SEE ALSO
`Code Examples <code_examples.html>`__, \
`glViewer <glviewer.html>`__, \
`glBatchViewer <glbatchviewer.html>`__, \
`glStencilViewer <glstencilviewer.html>`__, \
`ptexViewer <ptexviewer.html>`__, \
`paintTest <painttest.html>`__, \
`limitEval <limiteval.html>`__, \

View File

@ -1,4 +1,4 @@
..
..
Copyright 2013 Pixar
Licensed under the Apache License, Version 2.0 (the "License");
@ -21,7 +21,7 @@
either express or implied. See the License for the specific
language governing permissions and limitations under the
License.
FAR Overview
------------
@ -33,13 +33,13 @@ FAR Overview
Feature Adaptive Representation (Far)
=====================================
Far is a serialized topoloigcal data representation.Far uses hbr to create and
cache fast run time data structures for table driven subdivision of vertices and
cubic patches for limit surface evaluation. `Feature-adaptive <subdivision_surfaces.html#feature-adaptive-subdivision>`__
refinement logic is used to adaptively refine coarse topology near features like
extraordinary vertices and creases in order to make the topology amenable to
cubic patch evaluation. Far is also a generic, templated algorithmic base API
that clients in higher levels instantiate and use by providing an implementation
Far is a serialized topoloigcal data representation.Far uses hbr to create and
cache fast run time data structures for table driven subdivision of vertices and
cubic patches for limit surface evaluation. `Feature-adaptive <subdivision_surfaces.html#feature-adaptive-subdivision>`__
refinement logic is used to adaptively refine coarse topology near features like
extraordinary vertices and creases in order to make the topology amenable to
cubic patch evaluation. Far is also a generic, templated algorithmic base API
that clients in higher levels instantiate and use by providing an implementation
of a vertex class. It supports these subdivision schemes:
Factories & Tables
@ -50,3 +50,146 @@ Subdivision Tables
Patch Tables
============
Stencil Tables
==============
Stencils are the most direct method of evaluation of specific locations on the
limit of a subdivision surface starting from the coarse vertices of the control
cage.
.. image:: images/far_stencil0.png
:align: center
Sample Location
***************
Each stencil is associated with a singular parametric location on the coarse
mesh. The paramatric location is defined as face location and local [0.0 - 1.0]
(u,v) triplet:
In the case of a non-coarse quad face, the parametric sub-face quadrant needs to
be identified. This can be done either explicitly or implicitly by using the
unique ptex face indices for instance.
.. image:: images/far_stencil6.png
:align: center
Principles
**********
Iterative subdivision algorithms such as the one used in `FarSubdivisionTables <#subdivision-tables>`__
converge towards the limit surface by sucessively refining the vertices of the
coarse control cage.
.. image:: images/far_stencil4.png
:align: center
Each step is dependent upon the previous subidivion step being completed, and a
substantial number of steps may be required in order approximate the limit. Since
each subdivision step incurs an O(4 :superscript:`n`) growing amount of
computations, the accrued number of interpolations can be quite large.
However, every intermediate subdivided vertex can be expressed as a linear
interpolation of vertice from the previous step. So, eventually, every point at
on the limit surface can be expressed as a weighted average of the set of coarse
control vertices from the one-ring surrounding the face that the point is in:
.. image:: images/far_stencil3.png
:align: center
Where:
.. image:: images/far_stencil2.png
:align: center
Stencils are created by combining the list of control vertices of the 1-ring
to a set of interpolation weights obtained by successive accumulation of
subdivision interpolation weights.
The weight accumulation process is made efficient by adaptively subdividing the
control cage only around extraordinary locations, and otherwise reverting to fast
bi-cubic bspline patch evaluation. The use of bi-cubic patches also allows the
accumulation of analytical derivatives.
API Architecture
****************
The base container for stencil data is the FarStencilTables class. As with most
other Far entities, it has an associated FarStencilTablesFactory that requires
an HbrMesh:
.. image:: images/far_stencil5.png
:align: center
Assuming a properly qualified HbrMesh:
.. code:: c++
HMesh<OpenSubdiv::FarStencilFactoryVertex> * mesh;
FarStencilTables controlStencils;
OpenSubdiv::FarStencilTablesFactory<> factory(mesh);
for (int i=0; i<nfaces; ++i) {
HFace * f = mesh->GetFace(i);
int nv = f->GetNumVertices();
if (nv!=4) {
// if the face is not a quad, we have to iterate over sub-quad(rants)
for (int j=0; j<f->GetNumVertices(); ++j) {
factory.SetCurrentFace(i,j);
factory.AppendStencils( &controlStencils, nsamples/nv, u, v, reflevel );
}
} else {
factory.SetCurrentFace(i);
factory.AppendStencils( &controlStencils, g_nsamples, u, v, reflevel );
}
}
When the control vertices (controlPoints) move in space, the limit locations can
be very efficiently recomputed simply by applying the blending weights to the
series of coarse control vertices:
.. code:: c++
class StencilType {
public:
void Clear() {
memset( &x, 0, sizeof(StencilType));
}
void AddWithWeight( StencilType const & cv, float weight ) {
x += cv.x * weight;
y += cv.y * weight;
z += cv.z * weight;
}
float x,y,z;
};
std::vector<StencilType> controlPoints,
points,
utan,
vtan;
// Uppdate points by applying stencils
controlStencils.UpdateValues<StencilType>( reinterpret_cast<StencilType const *>(
&controlPoints[0]), &points[0] );
// Uppdate tangents by applying derivative stencils
controlStencils.UpdateDerivs<StencilType>( reinterpret_cast<StencilType const *>(
&controlPoints[0]), &utan[0], &vtan[0] );

View File

@ -81,7 +81,7 @@ SEE ALSO
`Code Examples <code_examples.html>`__, \
`glViewer <glviewer.html>`__, \
`glBatchViewer <glbatchviewer.html>`__, \
`glStencilViewer <glstencilviewer.html>`__, \
`ptexViewer <ptexviewer.html>`__, \
`paintTest <painttest.html>`__, \
`limitEval <limiteval.html>`__, \

View File

@ -0,0 +1,74 @@
..
Copyright 2013 Pixar
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License
and the following modification to it: Section 6 Trademarks.
deleted and replaced with:
6. Trademarks. This License does not grant permission to use the
trade names, trademarks, service marks, or product names of the
Licensor and its affiliates, except as required for reproducing
the content of the NOTICE file.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing,
software distributed under the License is distributed on an
"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
either express or implied. See the License for the specific
language governing permissions and limitations under the
License.
glStencilViewer
---------------
.. contents::
:local:
:backlinks: none
SYNOPSIS
========
.. parsed-literal::
:class: codefhead
**glStencilViewer** [**-d** *isolation level*] [**-f**] *objfile(s)*
DESCRIPTION
===========
``glStencilViewer`` is a stand-alone application that showcases the application of
pre-computed stencil tables to a collection of geometric test shapes. Multiple
controls are available to experiment with the algorithms.
.. image:: images/glstencilviewer.png
:width: 400px
:align: center
:target: images/glstencilviewer.png
OPTIONS
=======
**-d** *isolation level*
Select the desired isolation level of adaptive feature isolation. This can be
useful when trying to load large pieces of geometry.
**-f**
Launches the application in full-screen mode (if is supported by GLFW on the
OS)
SEE ALSO
========
`Code Examples <code_examples.html>`__, \
`glViewer <glviewer.html>`__, \
`glBatchViewer <glbatchviewer.html>`__, \
`ptexViewer <ptexviewer.html>`__, \
`paintTest <painttest.html>`__, \
`limitEval <limiteval.html>`__, \
`dxViewer <dxviewer.html>`__, \

View File

@ -69,6 +69,7 @@ SEE ALSO
`Code Examples <code_examples.html>`__, \
`glBatchViewer <glbatchviewer.html>`__, \
`glStencilViewer <glstencilviewer.html>`__, \
`ptexViewer <ptexviewer.html>`__, \
`paintTest <painttest.html>`__, \
`limitEval <limiteval.html>`__, \

Binary file not shown.

After

Width:  |  Height:  |  Size: 32 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 34 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 24 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 15 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 16 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 24 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 14 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 331 KiB

View File

@ -51,6 +51,11 @@
<ul>
<li><a href="hbr_overview.html">Hbr</a></li>
<li><a href="far_overview.html">Far</a></li>
<ul>
<li><a href="far_overview.html#subdivision-tables">Subdivision Tables</a></li>
<li><a href="far_overview.html#patch-tables">Patch Tables</a></li>
<li><a href="far_overview.html#stencil-tables">Stencil Tables</a></li>
</ul>
<li><a href="osd_overview.html">Osd</a></li>
</ul>
</li>

View File

@ -79,6 +79,7 @@ SEE ALSO
`Code Examples <code_examples.html>`__, \
`glViewer <glviewer.html>`__, \
`glBatchViewer <glbatchviewer.html>`__, \
`glStencilViewer <glstencilviewer.html>`__, \
`ptexViewer <ptexviewer.html>`__, \
`limitEval <limiteval.html>`__, \
`dxViewer <dxviewer.html>`__, \

View File

@ -134,6 +134,7 @@ SEE ALSO
`Code Examples <code_examples.html>`__, \
`glViewer <glviewer.html>`__, \
`glBatchViewer <glbatchviewer.html>`__, \
`glStencilViewer <glstencilviewer.html>`__, \
`paintTest <painttest.html>`__, \
`limitEval <limiteval.html>`__, \
`dxViewer <dxviewer.html>`__, \

View File

@ -32,6 +32,18 @@ Release Notes
----
Release 2.2.0
=============
**New Features**
- Added subdivision stencil functionality (Far & OsdEval)
**Bug Fixes**
- Fix D3D11DrawContext to check for NULL pointers
- Fix cpuEvalLimitController crash bug
- Fixed search path suffixes for ICC libs
- Fixed invalid initialization of glslTransformFeedback kernel.
Release 2.1.0
=============

View File

@ -26,6 +26,7 @@
if( OPENGL_FOUND AND (GLEW_FOUND AND GLFW_FOUND) OR (APPLE AND GLFW_FOUND))
add_subdirectory(glViewer)
add_subdirectory(glBatchViewer)
add_subdirectory(glStencilViewer)
add_subdirectory(simpleCpu)
add_subdirectory(limitEval)
if(OPENGL_4_3_FOUND AND (NOT APPLE))

View File

@ -27,6 +27,13 @@
#include <cmath>
inline void
cross(float *n, const float *v1, const float *v2) {
n[0] = v1[1]*v2[2]-v1[2]*v2[1];
n[1] = v1[2]*v2[0]-v1[0]*v2[2];
n[2] = v1[0]*v2[1]-v1[1]*v2[0];
}
inline void
cross(float *n, const float *p0, const float *p1, const float *p2) {

View File

@ -0,0 +1,56 @@
#
# Copyright 2013 Pixar
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License
# and the following modification to it: Section 6 Trademarks.
# deleted and replaced with:
#
# 6. Trademarks. This License does not grant permission to use the
# trade names, trademarks, service marks, or product names of the
# Licensor and its affiliates, except as required for reproducing
# the content of the NOTICE file.
#
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
# either express or implied. See the License for the specific
# language governing permissions and limitations under the
# License.
#
# *** glStencilViewer ***
set(PLATFORM_LIBRARIES
${OSD_LINK_TARGET}
${GLFW_LIBRARIES}
)
include_directories(
${PROJECT_SOURCE_DIR}/opensubdiv
${PROJECT_SOURCE_DIR}/regression
${GLFW_INCLUDE_DIR}
)
if ( GLEW_FOUND )
include_directories(${GLEW_INCLUDE_DIR})
list(APPEND PLATFORM_LIBRARIES ${GLEW_LIBRARY})
endif()
_add_possibly_cuda_executable(glStencilViewer
main.cpp
../common/font_image.cpp
../common/hud.cpp
../common/gl_hud.cpp
${INC_FILES}
)
target_link_libraries(glStencilViewer
${PLATFORM_LIBRARIES}
)
install(TARGETS glStencilViewer DESTINATION ${CMAKE_BINDIR_BASE})

File diff suppressed because it is too large Load Diff

View File

@ -41,6 +41,8 @@ set(PUBLIC_HEADER_FILES
patchMap.h
patchTables.h
patchTablesFactory.h
stencilTablesFactory.h
stencilTables.h
subdivisionTables.h
subdivisionTablesFactory.h
vertexEditTables.h

View File

@ -0,0 +1,249 @@
//
// Copyright 2013 Pixar
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License
// and the following modification to it: Section 6 Trademarks.
// deleted and replaced with:
//
// 6. Trademarks. This License does not grant permission to use the
// trade names, trademarks, service marks, or product names of the
// Licensor and its affiliates, except as required for reproducing
// the content of the NOTICE file.
//
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific
// language governing permissions and limitations under the
// License.
//
#ifndef FAR_STENCILTABLES_H
#define FAR_STENCILTABLES_H
#include "../version.h"
#include <vector>
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
/// \brief Stencil descriptor
///
/// Allows access and manipulation of a single stencil in a FarStencilTables.
///
class FarStencil {
public:
/// Returns the size of the stencil (number of control vertices)
int GetSize() const {
return *_size;
}
/// Returns the control vertices indices
int const * GetVertexIndices() const {
return _indices;
}
/// Returns the interpolation weights
float const * GetValueWeights() const {
return _point;
}
/// Returns U derivative interpolation weights
float const * GetUDerivWeights() const {
return _uderiv;
}
/// Returns V derivative interpolation weights
float const * GetVDerivWeights() const {
return _vderiv;
}
/// Increment to the next stencil in the array
/// Note : there is no array boundary check !
void Increment() {
int stride = *_size;
++_size;
_indices += stride;
_point += stride;
_uderiv += stride;
_vderiv += stride;
}
private:
FarStencil( int * size,
int * indices,
float * point,
float * uderiv,
float * vderiv )
: _size(size),
_indices(indices),
_point(point),
_uderiv(uderiv),
_vderiv(vderiv) {
}
friend class FarStencilTables;
template <class T> friend class FarStencilTablesFactory;
int * _size,
* _indices;
float * _point,
* _uderiv,
* _vderiv;
};
/// \brief Table of subdivision stencils.
///
/// Stencils are the most direct methods of evaluation of locations on the limit
/// of a surface. Every point of a limit surface can be computed by linearly
/// blending a collection of coarse control vertices.
///
/// A stencil assigns a series of control vertex indices with a blending weight
/// that corresponds to a unique parametric location of the limit surface. When
/// the control vertices move in space, the limit location can be very efficiently
/// recomputed simply by applying the blending weights to the series of coarse
/// control vertices.
///
class FarStencilTables {
public:
/// \brief Clears the stencils from the table
void Clear() {
_sizes.clear();
_offsets.clear();
_indices.clear();
_point.clear();
_uderiv.clear();
_vderiv.clear();
}
/// \brief Returns the number of stencils in the tables
int GetNumStencils() const {
return (int)_sizes.size();
}
/// \brief Updates point values based on the control values
///
/// \note The values array is assumed to be at least as big as the
/// result of \c GetNumStencils().
template <class T>
void UpdateValues( T const *controlValues, T *values, int stride=0 ) const {
_Update( controlValues, &_point.at(0), values, stride );
}
/// \brief Updates derivative values based on the control values
///
/// \note The values array is assumed to be at least as big as the
/// result of \c GetNumStencils().
template <class T>
void UpdateDerivs( T const *controlValues, T *uderivs,
T *vderivs, int stride=0 ) const {
_Update( controlValues, &_uderiv.at(0), uderivs, stride );
_Update( controlValues, &_vderiv.at(0), vderivs, stride );
}
/// \brief Returns a FarStencil at index i in the tables
FarStencil GetStencil(int i) const;
/// \brief Returns the number of control vertices of each stencil in the table
std::vector<int> const & GetSizes() const {
return _sizes;
}
/// \brief Returns the offset to a given stencil
std::vector<int> const & GetOffsets() const {
return _offsets;
}
/// \brief Returns the indices of the control vertices
std::vector<int> const & GetControlIndices() const {
return _indices;
}
/// \brief Returns the stencils interpolation weights
std::vector<float> const & GetWeights() const {
return _point;
}
/// \brief Returns the stencils U deriv weights
std::vector<float> const & GetDuWeights() const {
return _uderiv;
}
/// \brief Returns the stencils U deriv weights
std::vector<float> const & GetDvWeights() const {
return _vderiv;
}
private:
template <class T> friend class FarStencilTablesFactory;
// Update values by appling cached stencil weights to new control values
template <class T> void _Update( T const *controlValues,
float const * weights,
T *values,
int stride ) const;
std::vector<int> _sizes; // number of coeffiecient for each stencil
std::vector<int> _offsets; // offset to the start of each stencil
std::vector<int> _indices;
std::vector<float> _point, // weight coefficients (value & derivatives)
_uderiv,
_vderiv;
};
template <class T> void
FarStencilTables::_Update( T const *controlValues,
float const * weights,
T *values,
int stride ) const {
int const * index = &_indices.at(0);
for (int i=0; i<GetNumStencils(); ++i) {
// Zero out the result accumulators
values[i].Clear();
// For each element in the array, add the coefs contribution
for (int j=0; j<_sizes[i]; ++j, ++index, ++weights) {
values[i].AddWithWeight( controlValues[*index], *weights );
}
}
}
inline FarStencil
FarStencilTables::GetStencil(int i) const {
int ofs = _offsets[i];
return FarStencil( const_cast<int *>(&_sizes[i]),
const_cast<int *>(&_indices[ofs]),
const_cast<float *>(&_point[ofs]),
const_cast<float *>(&_uderiv[ofs]),
const_cast<float *>(&_vderiv[ofs]) );
}
} // end namespace OPENSUBDIV_VERSION
using namespace OPENSUBDIV_VERSION;
} // end namespace OpenSubdiv
#endif // FAR_STENCILTABLES_H

File diff suppressed because it is too large Load Diff

View File

@ -258,19 +258,16 @@ public:
bool HasCreaseEdits() const { return hasCreaseEdits; }
void Unrefine(int numCoarseVerts, int numCoarseFaces) {
static int oldMaxFaceID = 0;
if(oldMaxFaceID == 0) {
oldMaxFaceID = numCoarseFaces;
}
for (int i = numCoarseFaces; i < maxFaceID; ++i) {
HbrFace<T>* f = GetFace(i);
if(f and not f->IsCoarse())
DeleteFace(f);
}
//oldMaxFaceID = maxFaceID;
maxFaceID = numCoarseFaces;
for(int i=numCoarseVerts; (int)vertices.size(); ++i ) {
for(int i=numCoarseVerts; i<(int)vertices.size(); ++i ) {
HbrVertex<T>* v = GetVertex(i);
if(v and not v->IsReferenced())
DeleteVertex(v);

View File

@ -53,6 +53,8 @@ set(CPU_SOURCE_FILES
cpuEvalLimitContext.cpp
cpuEvalLimitController.cpp
cpuEvalLimitKernel.cpp
cpuEvalStencilsContext.cpp
cpuEvalStencilsController.cpp
cpuVertexBuffer.cpp
error.cpp
evalLimitContext.cpp
@ -81,6 +83,8 @@ set(PUBLIC_HEADER_FILES
cpuComputeController.h
cpuEvalLimitContext.h
cpuEvalLimitController.h
cpuEvalStencilsContext.h
cpuEvalStencilsController.h
cpuVertexBuffer.h
error.h
evalLimitContext.h
@ -148,12 +152,14 @@ list(APPEND DOXY_HEADER_FILES
set(OPENMP_PUBLIC_HEADERS
ompKernel.h
ompComputeController.h
ompEvalStencilsController.h
)
if( OPENMP_FOUND )
list(APPEND CPU_SOURCE_FILES
ompKernel.cpp
ompComputeController.cpp
ompEvalStencilsController.cpp
)
list(APPEND PUBLIC_HEADER_FILES ${OPENMP_PUBLIC_HEADERS})
@ -175,10 +181,12 @@ if( TBB_FOUND )
list(APPEND CPU_SOURCE_FILES
tbbKernel.cpp
tbbComputeController.cpp
tbbEvalStencilsController.cpp
)
list(APPEND PUBLIC_HEADER_FILES
tbbKernel.h
tbbComputeController.h
tbbEvalStencilsController.h
)
if (CMAKE_COMPILER_IS_GNUCXX)
list(APPEND PLATFORM_CPU_LIBRARIES

View File

@ -66,69 +66,62 @@ OsdCpuEvalLimitController::_EvalLimitSample( OpenSubdiv::OsdEvalCoords const & c
int offset = vertexData.outDesc.stride * index;
// Based on patch type - go execute interpolation
switch( parray.GetDescriptor().GetType() ) {
if (vertexData.IsBound()) {
float * out = vertexData.out.GetData()+offset,
* outDu = vertexData.outDu.IsBound() ? vertexData.outDu.GetData()+offset : 0,
* outDv = vertexData.outDv.IsBound() ? vertexData.outDv.GetData()+offset : 0;
// Based on patch type - go execute interpolation
switch( parray.GetDescriptor().GetType() ) {
case FarPatchTables::REGULAR : if (vertexData.IsBound()) {
evalBSpline( v, u, cvs,
case FarPatchTables::REGULAR : evalBSpline( v, u, cvs,
vertexData.inDesc,
vertexData.in.GetData(),
vertexData.outDesc,
vertexData.out.GetData()+offset,
vertexData.outDu.GetData()+offset,
vertexData.outDv.GetData()+offset );
} break;
out, outDu, outDv );
break;
case FarPatchTables::BOUNDARY : if (vertexData.IsBound()) {
evalBoundary( v, u, cvs,
case FarPatchTables::BOUNDARY : evalBoundary( v, u, cvs,
vertexData.inDesc,
vertexData.in.GetData(),
vertexData.outDesc,
vertexData.out.GetData()+offset,
vertexData.outDu.GetData()+offset,
vertexData.outDv.GetData()+offset );
} break;
out, outDu, outDv );
break;
case FarPatchTables::CORNER : if (vertexData.IsBound()) {
evalCorner( v, u, cvs,
case FarPatchTables::CORNER : evalCorner( v, u, cvs,
vertexData.inDesc,
vertexData.in.GetData(),
vertexData.outDesc,
vertexData.out.GetData()+offset,
vertexData.outDu.GetData()+offset,
vertexData.outDv.GetData()+offset );
} break;
out, outDu, outDv );
break;
case FarPatchTables::GREGORY : if (vertexData.IsBound()) {
evalGregory( v, u, cvs,
case FarPatchTables::GREGORY : evalGregory( v, u, cvs,
&context->GetVertexValenceTable()[0],
&context->GetQuadOffsetTable()[ parray.GetQuadOffsetIndex() + handle->vertexOffset ],
&context->GetQuadOffsetTable()[ parray.GetQuadOffsetIndex() + handle->vertexOffset ],
context->GetMaxValence(),
vertexData.inDesc,
vertexData.in.GetData(),
vertexData.outDesc,
vertexData.out.GetData()+offset,
vertexData.outDu.GetData()+offset,
vertexData.outDv.GetData()+offset );
} break;
out, outDu, outDv );
break;
case FarPatchTables::GREGORY_BOUNDARY :
if (vertexData.IsBound()) {
evalGregoryBoundary(v, u, cvs,
&context->GetVertexValenceTable()[0],
&context->GetQuadOffsetTable()[ parray.GetQuadOffsetIndex() + handle->vertexOffset ],
context->GetMaxValence(),
vertexData.inDesc,
vertexData.in.GetData(),
vertexData.outDesc,
vertexData.out.GetData()+offset,
vertexData.outDu.GetData()+offset,
vertexData.outDv.GetData()+offset );
} break;
case FarPatchTables::GREGORY_BOUNDARY :
evalGregoryBoundary( v, u, cvs,
&context->GetVertexValenceTable()[0],
&context->GetQuadOffsetTable()[ parray.GetQuadOffsetIndex() + handle->vertexOffset ],
context->GetMaxValence(),
vertexData.inDesc,
vertexData.in.GetData(),
vertexData.outDesc,
out, outDu, outDv );
break;
default:
assert(0);
default:
assert(0);
}
}
}

View File

@ -0,0 +1,45 @@
//
// Copyright 2013 Pixar
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License
// and the following modification to it: Section 6 Trademarks.
// deleted and replaced with:
//
// 6. Trademarks. This License does not grant permission to use the
// trade names, trademarks, service marks, or product names of the
// Licensor and its affiliates, except as required for reproducing
// the content of the NOTICE file.
//
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific
// language governing permissions and limitations under the
// License.
//
#include "../osd/cpuEvalStencilsContext.h"
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
OsdCpuEvalStencilsContext::OsdCpuEvalStencilsContext(FarStencilTables const *stencils) :
_stencils(stencils),
_controlData(0),
_outputData(0),
_outputUDeriv(0),
_outputVDeriv(0) {
}
OsdCpuEvalStencilsContext *
OsdCpuEvalStencilsContext::Create(FarStencilTables const *stencils) {
return new OsdCpuEvalStencilsContext(stencils);
}
} // end namespace OPENSUBDIV_VERSION
} // end namespace OpenSubdiv

View File

@ -0,0 +1,193 @@
//
// Copyright 2013 Pixar
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License
// and the following modification to it: Section 6 Trademarks.
// deleted and replaced with:
//
// 6. Trademarks. This License does not grant permission to use the
// trade names, trademarks, service marks, or product names of the
// Licensor and its affiliates, except as required for reproducing
// the content of the NOTICE file.
//
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific
// language governing permissions and limitations under the
// License.
//
#ifndef FAR_CPU_EVALSTENCILS_CONTEXT_H
#define FAR_CPU_EVALSTENCILS_CONTEXT_H
#include "../version.h"
#include "../far/stencilTables.h"
#include "../osd/vertexDescriptor.h"
#include "../osd/nonCopyable.h"
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
///
/// \brief CPU stencils evaluation context
///
///
class OsdCpuEvalStencilsContext : OsdNonCopyable<OsdCpuEvalStencilsContext> {
public:
/// \brief Creates an OsdCpuEvalStencilsContext instance
///
/// @param stencils a pointer to the FarStencilTables
///
static OsdCpuEvalStencilsContext * Create(FarStencilTables const *stencils);
/// \brief Binds control vertex data buffer
///
/// Binding ensures that data buffers are properly inter-operated between
/// Contexts and Controllers operating across multiple devices.
///
/// @param controlDataDesc
///
/// @param controlData
///
template<class VERTEX_BUFFER>
void BindControlData(OsdVertexBufferDescriptor const & controlDataDesc, VERTEX_BUFFER *controlData ) {
_controlData = controlData ? controlData->BindCpuBuffer() : 0;
_controlDataDesc = controlDataDesc;
}
/// \brief Binds output vertex data buffer
///
/// Binding ensures that data buffers are properly inter-operated between
/// Contexts and Controllers operating across multiple devices.
///
/// @param outputDataDesc
///
/// @param outputData
///
template<class VERTEX_BUFFER>
void BindOutputData( OsdVertexBufferDescriptor const & outputDataDesc, VERTEX_BUFFER *outputData ) {
_outputData = outputData ? outputData->BindCpuBuffer() : 0;
_outputDataDesc = outputDataDesc;
}
/// \brief Binds output derivative vertex data buffer
///
/// Binding ensures that data buffers are properly inter-operated between
/// Contexts and Controllers operating across multiple devices.
///
/// @param controlDataDesc
///
/// @param controlData
///
/// @param outputDataDesc
///
/// @param outputData
///
template<class VERTEX_BUFFER>
void BindOutputDerivData( OsdVertexBufferDescriptor const & outputDuDesc, VERTEX_BUFFER *outputDu,
OsdVertexBufferDescriptor const & outputDvDesc, VERTEX_BUFFER *outputDv ) {
_outputUDeriv = outputDu ? outputDu ->BindCpuBuffer() : 0;
_outputVDeriv = outputDv ? outputDv->BindCpuBuffer() : 0;
_outputDuDesc = outputDuDesc;
_outputDvDesc = outputDvDesc;
}
/// \brief Unbinds any previously bound vertex and varying data buffers.
void Unbind() {
_controlData = 0;
_controlDataDesc.Reset();
_outputData = 0;
_outputDataDesc.Reset();
_outputUDeriv = 0;
_outputDuDesc.Reset();
_outputVDeriv = 0;
_outputDvDesc.Reset();
}
/// \brief returns a pointer to the control vertex data
float const * GetControlData() const {
return _controlData;
}
/// \brief returns a pointer to the output vertex data
float * GetOutputData() {
return _outputData;
}
/// \brief returns a pointer to the output u-derivative vertex data
float * GetOutputUDerivData() {
return _outputUDeriv;
}
/// \brief returns a pointer to the output v-derivative vertex data
float * GetOutputVDerivData() {
return _outputVDeriv;
}
/// \brief Returns the vertex data descriptor for the control vertices buffer
OsdVertexBufferDescriptor GetControlDataDescriptor() const {
return _controlDataDesc;
}
/// \brief Returns the vertex data descriptor for the output vertices buffer
OsdVertexBufferDescriptor GetOutputDataDescriptor() const {
return _outputDataDesc;
}
/// \brief Returns the vertex data descriptor for the U derivatives data buffer
OsdVertexBufferDescriptor GetDuDataDescriptor() const {
return _outputDuDesc;
}
/// \brief Returns the vertex data descriptor for the V derivatives data buffer
OsdVertexBufferDescriptor GetDvDataDescriptor() const {
return _outputDvDesc;
}
/// \brief Returns the FarStencilTables applied
FarStencilTables const * GetStencilTables() const {
return _stencils;
}
protected:
OsdCpuEvalStencilsContext(FarStencilTables const *stencils);
private:
FarStencilTables const * _stencils;
OsdVertexBufferDescriptor _controlDataDesc,
_outputDataDesc,
_outputDuDesc,
_outputDvDesc;
float * _controlData,
* _outputData,
* _outputUDeriv,
* _outputVDeriv;
};
} // end namespace OPENSUBDIV_VERSION
using namespace OPENSUBDIV_VERSION;
} // end namespace OpenSubdiv
#endif // FAR_CPU_EVALSTENCILS_CONTEXT_H

View File

@ -0,0 +1,147 @@
//
// Copyright 2013 Pixar
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License
// and the following modification to it: Section 6 Trademarks.
// deleted and replaced with:
//
// 6. Trademarks. This License does not grant permission to use the
// trade names, trademarks, service marks, or product names of the
// Licensor and its affiliates, except as required for reproducing
// the content of the NOTICE file.
//
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific
// language governing permissions and limitations under the
// License.
//
#include "../osd/cpuEvalStencilsController.h"
#include <cassert>
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
OsdCpuEvalStencilsController::OsdCpuEvalStencilsController() {
}
OsdCpuEvalStencilsController::~OsdCpuEvalStencilsController() {
}
int
OsdCpuEvalStencilsController::_UpdateValues( OsdCpuEvalStencilsContext * context ) {
int result=0;
FarStencilTables const * stencils = context->GetStencilTables();
int nstencils = stencils->GetNumStencils();
if (not nstencils)
return result;
OsdVertexBufferDescriptor ctrlDesc = context->GetControlDataDescriptor(),
outDesc = context->GetOutputDataDescriptor();
// make sure that we have control data to work with
if (not ctrlDesc.CanEval(outDesc))
return 0;
float const * ctrl = context->GetControlData() + ctrlDesc.offset;
float * out = context->GetOutputData() + outDesc.offset;
if ((not ctrl) or (not out))
return result;
int const * sizes = &stencils->GetSizes().at(0),
* index = &stencils->GetControlIndices().at(0);
float const * weight = &stencils->GetWeights().at(0);
for (int i=0; i<nstencils; ++i) {
memset(out, 0, outDesc.length*sizeof(float));
for (int j=0; j<sizes[i]; ++j, ++index, ++weight) {
float const * cv = ctrl + (*index)*ctrlDesc.stride;
for (int k=0; k<outDesc.length; ++k) {
out[k] += cv[k] * (*weight);
}
}
out += outDesc.stride;
}
return nstencils;
}
int
OsdCpuEvalStencilsController::_UpdateDerivs( OsdCpuEvalStencilsContext * context ) {
int result=0;
FarStencilTables const * stencils = context->GetStencilTables();
int nstencils = stencils->GetNumStencils();
if (not nstencils)
return result;
OsdVertexBufferDescriptor ctrlDesc = context->GetControlDataDescriptor(),
duDesc = context->GetDuDataDescriptor(),
dvDesc = context->GetDvDataDescriptor();
// make sure that we have control data to work with
if (not (ctrlDesc.CanEval(duDesc) and ctrlDesc.CanEval(dvDesc)))
return 0;
float const * ctrl = context->GetControlData() + ctrlDesc.offset;
float * du = context->GetOutputUDerivData() + duDesc.offset,
* dv = context->GetOutputVDerivData() + dvDesc.offset;
if ((not ctrl) or (not du) or (not dv))
return result;
int const * sizes = &stencils->GetSizes().at(0),
* index = &stencils->GetControlIndices().at(0);
float const * duweight = &stencils->GetDuWeights().at(0),
* dvweight = &stencils->GetDvWeights().at(0);
for (int i=0; i<nstencils; ++i) {
memset(du, 0, duDesc.length*sizeof(float));
memset(dv, 0, dvDesc.length*sizeof(float));
for (int j=0; j<sizes[i]; ++j, ++index, ++duweight, ++dvweight) {
float const * cv = ctrl + (*index)*ctrlDesc.stride;
for (int k=0; k<duDesc.length; ++k) {
du[k] += cv[k] * (*duweight);
dv[k] += cv[k] * (*dvweight);
}
}
du += duDesc.stride;
dv += dvDesc.stride;
}
return nstencils;
}
void
OsdCpuEvalStencilsController::Synchronize() {
}
} // end namespace OPENSUBDIV_VERSION
} // end namespace OpenSubdiv

View File

@ -0,0 +1,144 @@
//
// Copyright 2013 Pixar
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License
// and the following modification to it: Section 6 Trademarks.
// deleted and replaced with:
//
// 6. Trademarks. This License does not grant permission to use the
// trade names, trademarks, service marks, or product names of the
// Licensor and its affiliates, except as required for reproducing
// the content of the NOTICE file.
//
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific
// language governing permissions and limitations under the
// License.
//
#ifndef FAR_CPU_EVALSTENCILS_CONTROLLER_H
#define FAR_CPU_EVALSTENCILS_CONTROLLER_H
#include "../version.h"
#include "../osd/cpuEvalStencilsContext.h"
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
///
/// \brief CPU stencils evaluation controller
///
/// OsdCpuStencilsController is a compute controller class to launch
/// single threaded CPU stencil evalution kernels.
///
/// Controller entities execute requests from Context instances that they share
/// common interfaces with. Controllers are attached to discrete compute devices
/// and share the devices resources with Context entities.
///
class OsdCpuEvalStencilsController {
public:
/// Constructor.
OsdCpuEvalStencilsController();
/// Destructor.
~OsdCpuEvalStencilsController();
/// \brief Applies stencil weights to the control vertex data
///
/// Applies the stencil weights to the control vertex data to evaluate the
/// interpolated limit positions at the parametric locations of the stencils
///
/// @param context the OsdCpuEvalStencilsContext with the stencil weights
///
/// @param controlDataDesc vertex buffer descriptor for the control vertex data
///
/// @param controlVertices vertex buffer with the control vertices data
///
/// @param outputDataDesc vertex buffer descriptor for the output vertex data
///
/// @param outputData vertex buffer where the vertex data will be output
///
template<class CONTROL_BUFFER, class OUTPUT_BUFFER>
int UpdateValues( OsdCpuEvalStencilsContext * context,
OsdVertexBufferDescriptor const & controlDataDesc, CONTROL_BUFFER *controlVertices,
OsdVertexBufferDescriptor const & outputDataDesc, OUTPUT_BUFFER *outputData ) {
if (not context->GetStencilTables()->GetNumStencils())
return 0;
context->BindControlData( controlDataDesc, controlVertices );
context->BindOutputData( outputDataDesc, outputData );
int n = _UpdateValues( context );
context->Unbind();
return n;
}
/// \brief Applies derivative stencil weights to the control vertex data
///
/// Computes the U and V derivative stencils to the control vertex data at
/// the parametric locations contained in each stencil
///
/// @param context the OsdCpuEvalStencilsContext with the stencil weights
///
/// @param controlDataDesc vertex buffer descriptor for the control vertex data
///
/// @param controlVertices vertex buffer with the control vertices data
///
/// @param outputDuDesc vertex buffer descriptor for the U derivative output data
///
/// @param outputDuData output vertex buffer for the U derivative data
///
/// @param outputDvDesc vertex buffer descriptor for the V deriv output data
///
/// @param outputDvData output vertex buffer for the V derivative data
///
template<class CONTROL_BUFFER, class OUTPUT_BUFFER>
int UpdateDerivs( OsdCpuEvalStencilsContext * context,
OsdVertexBufferDescriptor const & controlDataDesc, CONTROL_BUFFER *controlVertices,
OsdVertexBufferDescriptor const & outputDuDesc, OUTPUT_BUFFER *outputDuData,
OsdVertexBufferDescriptor const & outputDvDesc, OUTPUT_BUFFER *outputDvData ) {
if (not context->GetStencilTables()->GetNumStencils())
return 0;
context->BindControlData( controlDataDesc, controlVertices );
context->BindOutputDerivData( outputDuDesc, outputDuData, outputDvDesc, outputDvData );
int n = _UpdateDerivs( context );
context->Unbind();
return n;
}
/// Waits until all running subdivision kernels finish.
void Synchronize();
private:
int _UpdateValues( OsdCpuEvalStencilsContext * context );
int _UpdateDerivs( OsdCpuEvalStencilsContext * context );
};
} // end namespace OPENSUBDIV_VERSION
using namespace OPENSUBDIV_VERSION;
} // end namespace OpenSubdiv
#endif // FAR_CPU_EVALSTENCILS_CONTROLLER_H

View File

@ -86,10 +86,11 @@ public:
///
template<class VERTEX_BUFFER>
void UpdateVertexTexture(VERTEX_BUFFER *vbo, ID3D11DeviceContext *pd3d11DeviceContext) {
updateVertexTexture(vbo->BindD3D11Buffer(pd3d11DeviceContext),
pd3d11DeviceContext,
vbo->GetNumVertices(),
vbo->GetNumElements());
if (vbo)
updateVertexTexture(vbo->BindD3D11Buffer(pd3d11DeviceContext),
pd3d11DeviceContext,
vbo->GetNumVertices(),
vbo->GetNumElements());
}
ID3D11Buffer *patchIndexBuffer;

View File

@ -70,7 +70,8 @@ public:
///
template<class VERTEX_BUFFER>
void UpdateVertexTexture(VERTEX_BUFFER *vbo) {
updateVertexTexture(vbo->BindVBO(), vbo->GetNumElements());
if (vbo)
updateVertexTexture(vbo->BindVBO(), vbo->GetNumElements());
}
/// true if the GL version detected supports shader tessellation

View File

@ -40,6 +40,7 @@
#include "../osd/opengl.h"
#include <cassert>
#include <string>
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
@ -94,20 +95,30 @@ OsdGLSLTransformFeedbackKernelBundle::Compile(int numVertexElements, int numVary
glCompileShader(shader);
glAttachShader(_program, shader);
const char *outputs[4];
int nOutputs = 0;
std::vector<std::string> outputs;
// position and custom vertex data are stored same buffer whereas varying data
// exists on another buffer. "gl_NextBuffer" identifier helps to split them.
if (numVertexElements > 0)
outputs[nOutputs++] = "outVertexData";
if (numVaryingElements > 0) {
if (nOutputs > 0)
outputs[nOutputs++] = "gl_NextBuffer";
outputs[nOutputs++] = "outVaryingData";
for (int i = 0; i < numVertexElements; ++i) {
char attrName[32];
snprintf(attrName, 32, "outVertexData[%d]", i);
outputs.push_back(attrName);
}
for (int i = 0; i < numVaryingElements; ++i) {
if (i == 0 and (not outputs.empty())) {
outputs.push_back("gl_NextBuffer");
}
char attrName[32];
snprintf(attrName, 32, "outVaryingData[%d]", i);
outputs.push_back(attrName);
}
std::vector<const char *> pOutputs;
for (size_t i = 0; i < outputs.size(); ++i) {
pOutputs.push_back(&outputs[i][0]);
}
glTransformFeedbackVaryings(_program, nOutputs, outputs, GL_INTERLEAVED_ATTRIBS);
glTransformFeedbackVaryings(_program, (GLsizei)outputs.size(),
&pOutputs[0], GL_INTERLEAVED_ATTRIBS);
OSD_DEBUG_CHECK_GL_ERROR("Transform feedback initialize\n");

View File

@ -0,0 +1,152 @@
//
// Copyright 2013 Pixar
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License
// and the following modification to it: Section 6 Trademarks.
// deleted and replaced with:
//
// 6. Trademarks. This License does not grant permission to use the
// trade names, trademarks, service marks, or product names of the
// Licensor and its affiliates, except as required for reproducing
// the content of the NOTICE file.
//
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific
// language governing permissions and limitations under the
// License.
//
#include "../osd/ompEvalStencilsController.h"
#include <cassert>
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
OsdOmpEvalStencilsController::OsdOmpEvalStencilsController(int numThreads) {
_numThreads = (numThreads == -1) ? omp_get_num_procs() : numThreads;
}
OsdOmpEvalStencilsController::~OsdOmpEvalStencilsController() {
}
int
OsdOmpEvalStencilsController::_UpdateValues( OsdCpuEvalStencilsContext * context ) {
int result=0;
FarStencilTables const * stencils = context->GetStencilTables();
int nstencils = stencils->GetNumStencils();
if (not nstencils)
return result;
OsdVertexBufferDescriptor ctrlDesc = context->GetControlDataDescriptor(),
outDesc = context->GetOutputDataDescriptor();
// make sure that we have control data to work with
if (not ctrlDesc.CanEval(outDesc))
return 0;
float const * ctrl = context->GetControlData() + ctrlDesc.offset;
if (not ctrl)
return result;
#pragma omp parallel for
for (int i=0; i<nstencils; ++i) {
int size = stencils->GetSizes()[i],
offset = stencils->GetOffsets()[i];
int const * index = &stencils->GetControlIndices().at(offset);
float const * weight = &stencils->GetWeights().at(offset);
float * out = context->GetOutputData() + i * outDesc.stride + outDesc.offset;
memset(out, 0, outDesc.length*sizeof(float));
for (int j=0; j<size; ++j, ++index, ++weight) {
float const * cv = ctrl + (*index)*ctrlDesc.stride;
for (int k=0; k<outDesc.length; ++k) {
out[k] += cv[k] * (*weight);
}
}
}
return nstencils;
}
int
OsdOmpEvalStencilsController::_UpdateDerivs( OsdCpuEvalStencilsContext * context ) {
int result=0;
FarStencilTables const * stencils = context->GetStencilTables();
int nstencils = stencils->GetNumStencils();
if (not nstencils)
return result;
OsdVertexBufferDescriptor ctrlDesc = context->GetControlDataDescriptor(),
duDesc = context->GetDuDataDescriptor(),
dvDesc = context->GetDvDataDescriptor();
// make sure that we have control data to work with
if (not (ctrlDesc.CanEval(duDesc) and ctrlDesc.CanEval(dvDesc)))
return 0;
float const * ctrl = context->GetControlData() + ctrlDesc.offset;
if (not ctrl)
return result;
#pragma omp parallel for
for (int i=0; i<nstencils; ++i) {
int size = stencils->GetSizes()[i],
offset = stencils->GetOffsets()[i];
int const * index = &stencils->GetControlIndices().at(offset);
float const * duweight = &stencils->GetDuWeights().at(offset),
* dvweight = &stencils->GetDvWeights().at(offset);
float * du = context->GetOutputUDerivData() + i * duDesc.stride + duDesc.offset,
* dv = context->GetOutputVDerivData() + i * dvDesc.stride + dvDesc.offset;
memset(du, 0, duDesc.length*sizeof(float));
memset(dv, 0, dvDesc.length*sizeof(float));
for (int j=0; j<size; ++j, ++index, ++duweight, ++dvweight) {
float const * cv = ctrl + (*index)*ctrlDesc.stride;
for (int k=0; k<duDesc.length; ++k) {
du[k] += cv[k] * (*duweight);
dv[k] += cv[k] * (*dvweight);
}
}
}
return nstencils;
}
void
OsdOmpEvalStencilsController::Synchronize() {
}
} // end namespace OPENSUBDIV_VERSION
} // end namespace OpenSubdiv

View File

@ -0,0 +1,154 @@
//
// Copyright 2013 Pixar
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License
// and the following modification to it: Section 6 Trademarks.
// deleted and replaced with:
//
// 6. Trademarks. This License does not grant permission to use the
// trade names, trademarks, service marks, or product names of the
// Licensor and its affiliates, except as required for reproducing
// the content of the NOTICE file.
//
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific
// language governing permissions and limitations under the
// License.
//
#ifndef FAR_OMP_EVALSTENCILS_CONTROLLER_H
#define FAR_OMP_EVALSTENCILS_CONTROLLER_H
#include "../version.h"
#include "../osd/cpuEvalStencilsContext.h"
#ifdef OPENSUBDIV_HAS_OPENMP
#include <omp.h>
#endif
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
///
/// \brief CPU stencils evaluation controller
///
/// OsdCpuStencilsController is a compute controller class to launch
/// single threaded CPU stencil evalution kernels.
///
/// Controller entities execute requests from Context instances that they share
/// common interfaces with. Controllers are attached to discrete compute devices
/// and share the devices resources with Context entities.
///
class OsdOmpEvalStencilsController {
public:
/// \brief Constructor.
///
/// @param numThreads specifies how many openmp parallel threads to use.
/// -1 attempts to use all available processors.
///
OsdOmpEvalStencilsController(int numThreads=-1);
/// \brief Destructor.
~OsdOmpEvalStencilsController();
/// \brief Applies stencil weights to the control vertex data
///
/// Applies the stencil weights to the control vertex data to evaluate the
/// interpolated limit positions at the parametric locations of the stencils
///
/// @param context the OsdCpuEvalStencilsContext with the stencil weights
///
/// @param controlDataDesc vertex buffer descriptor for the control vertex data
///
/// @param controlVertices vertex buffer with the control vertices data
///
/// @param outputDataDesc vertex buffer descriptor for the output vertex data
///
/// @param outputData output vertex buffer for the interpolated data
///
template<class CONTROL_BUFFER, class OUTPUT_BUFFER>
int UpdateValues( OsdCpuEvalStencilsContext * context,
OsdVertexBufferDescriptor const & controlDataDesc, CONTROL_BUFFER *controlVertices,
OsdVertexBufferDescriptor const & outputDataDesc, OUTPUT_BUFFER *outputData ) {
if (not context->GetStencilTables()->GetNumStencils())
return 0;
omp_set_num_threads(_numThreads);
context->BindControlData( controlDataDesc, controlVertices );
context->BindOutputData( outputDataDesc, outputData );
int n = _UpdateValues( context );
context->Unbind();
return n;
}
/// \brief Applies derivative stencil weights to the control vertex data
///
/// Computes the U and V derivative stencils to the control vertex data at
/// the parametric locations contained in each stencil
///
/// @param context the OsdCpuEvalStencilsContext with the stencil weights
///
/// @param controlDataDesc vertex buffer descriptor for the control vertex data
///
/// @param controlVertices vertex buffer with the control vertices data
///
/// @param outputDuDesc vertex buffer descriptor for the U derivative output data
///
/// @param outputDuData output vertex buffer for the U derivative data
///
/// @param outputDvDesc vertex buffer descriptor for the V deriv output data
///
/// @param outputDvData output vertex buffer for the V derivative data
///
template<class CONTROL_BUFFER, class OUTPUT_BUFFER>
int UpdateDerivs( OsdCpuEvalStencilsContext * context,
OsdVertexBufferDescriptor const & controlDataDesc, CONTROL_BUFFER *controlVertices,
OsdVertexBufferDescriptor const & outputDuDesc, OUTPUT_BUFFER *outputDuData,
OsdVertexBufferDescriptor const & outputDvDesc, OUTPUT_BUFFER *outputDvData ) {
if (not context->GetStencilTables()->GetNumStencils())
return 0;
context->BindControlData( controlDataDesc, controlVertices );
context->BindOutputDerivData( outputDuDesc, outputDuData, outputDvDesc, outputDvData );
int n = _UpdateDerivs( context );
context->Unbind();
return n;
}
/// Waits until all running subdivision kernels finish.
void Synchronize();
private:
int _UpdateValues( OsdCpuEvalStencilsContext * context );
int _UpdateDerivs( OsdCpuEvalStencilsContext * context );
int _numThreads;
};
} // end namespace OPENSUBDIV_VERSION
using namespace OPENSUBDIV_VERSION;
} // end namespace OpenSubdiv
#endif // FAR_OMP_EVALSTENCILS_CONTROLLER_H

View File

@ -0,0 +1,197 @@
//
// Copyright 2013 Pixar
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License
// and the following modification to it: Section 6 Trademarks.
// deleted and replaced with:
//
// 6. Trademarks. This License does not grant permission to use the
// trade names, trademarks, service marks, or product names of the
// Licensor and its affiliates, except as required for reproducing
// the content of the NOTICE file.
//
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific
// language governing permissions and limitations under the
// License.
//
#include "../osd/tbbEvalStencilsController.h"
#include <tbb/parallel_for.h>
#include <tbb/task_scheduler_init.h>
#include <cassert>
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
#define grain_size 200
OsdTbbEvalStencilsController::OsdTbbEvalStencilsController(int numThreads) {
_numThreads = numThreads > 0 ? numThreads : tbb::task_scheduler_init::automatic;
tbb::task_scheduler_init init(numThreads);
}
OsdTbbEvalStencilsController::~OsdTbbEvalStencilsController() {
}
class StencilKernel {
public:
enum Mode { UNDEFINED, POINT, U_DERIV, V_DERIV };
StencilKernel( FarStencilTables const * stencils,
OsdVertexBufferDescriptor ctrlDesc,
float const * ctrlData ) :
_stencils(stencils),
_mode(UNDEFINED),
_ctrlDesc(ctrlDesc),
_length(0),
_outStride(0),
_outData(0) {
_ctrlData = ctrlData + ctrlDesc.offset;
}
bool SetOutput(Mode mode, OsdVertexBufferDescriptor outDesc, float * outData) {
if (_ctrlDesc.CanEval(outDesc)) {
_mode = mode;
_length = outDesc.length;
_outStride = outDesc.stride;
_outData = outData + outDesc.offset;
return true;
}
return false;
}
void operator() (tbb::blocked_range<int> const &r) const {
assert(_stencils and _ctrlData and _length and _outStride and _outData);
int offset = _stencils->GetOffsets()[r.begin()];
int const * sizes = &_stencils->GetSizes()[r.begin()],
* index = &_stencils->GetControlIndices()[offset];
float const * weight;
switch (_mode) {
case POINT : weight = &_stencils->GetWeights()[offset]; break;
case U_DERIV : weight = &_stencils->GetDuWeights()[offset]; break;
case V_DERIV : weight = &_stencils->GetDvWeights()[offset]; break;
default:
return;
}
assert( weight);
float * out = _outData + r.begin() * _outStride;
for (int i=r.begin(); i<r.end(); ++i, ++sizes) {
memset( out, 0, _length * sizeof(float) );
for (int j=0; j<(*sizes); ++j, ++index, ++weight) {
float const * cv = _ctrlData + (*index)*_ctrlDesc.stride;
for (int k=0; k<_length; ++k) {
out[k] += cv[k] * (*weight);
}
}
out+=_outStride;
}
}
private:
FarStencilTables const * _stencils;
Mode _mode;
OsdVertexBufferDescriptor _ctrlDesc;
float const * _ctrlData;
int _length,
_outStride;
float * _outData;
};
int
OsdTbbEvalStencilsController::_UpdateValues( OsdCpuEvalStencilsContext * context ) {
FarStencilTables const * stencils = context->GetStencilTables();
if (not stencils)
return 0;
int nstencils = stencils->GetNumStencils();
if (not nstencils)
return 0;
StencilKernel kernel( stencils, context->GetControlDataDescriptor(),
context->GetControlData() );
if (not kernel.SetOutput( StencilKernel::POINT,
context->GetOutputDataDescriptor(),
context->GetOutputData() ))
return 0;
tbb::blocked_range<int> range(0, nstencils, grain_size);
tbb::parallel_for(range, kernel);
return nstencils;
}
int
OsdTbbEvalStencilsController::_UpdateDerivs( OsdCpuEvalStencilsContext * context ) {
FarStencilTables const * stencils = context->GetStencilTables();
if (not stencils)
return 0;
int nstencils = stencils->GetNumStencils();
if (not nstencils)
return 0;
tbb::blocked_range<int> range(0, nstencils, grain_size);
StencilKernel kernel( stencils, context->GetControlDataDescriptor(),
context->GetControlData() );
if (not kernel.SetOutput( StencilKernel::U_DERIV,
context->GetDuDataDescriptor(),
context->GetOutputUDerivData() ) )
return 0;
tbb::parallel_for(range, kernel);
if (not kernel.SetOutput( StencilKernel::V_DERIV,
context->GetDvDataDescriptor(),
context->GetOutputVDerivData() ) )
return 0;
tbb::parallel_for(range, kernel);
return nstencils;
}
void
OsdTbbEvalStencilsController::Synchronize() {
}
} // end namespace OPENSUBDIV_VERSION
} // end namespace OpenSubdiv

View File

@ -0,0 +1,149 @@
//
// Copyright 2013 Pixar
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License
// and the following modification to it: Section 6 Trademarks.
// deleted and replaced with:
//
// 6. Trademarks. This License does not grant permission to use the
// trade names, trademarks, service marks, or product names of the
// Licensor and its affiliates, except as required for reproducing
// the content of the NOTICE file.
//
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
// either express or implied. See the License for the specific
// language governing permissions and limitations under the
// License.
//
#ifndef FAR_TBB_EVALSTENCILS_CONTROLLER_H
#define FAR_TBB_EVALSTENCILS_CONTROLLER_H
#include "../version.h"
#include "../osd/cpuEvalStencilsContext.h"
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
///
/// \brief CPU stencils evaluation controller
///
/// OsdCpuStencilsController is a compute controller class to launch
/// single threaded CPU stencil evalution kernels.
///
/// Controller entities execute requests from Context instances that they share
/// common interfaces with. Controllers are attached to discrete compute devices
/// and share the devices resources with Context entities.
///
class OsdTbbEvalStencilsController {
public:
/// \brief Constructor.
///
/// @param numThreads specifies how many openmp parallel threads to use.
/// -1 attempts to use all available processors.
///
OsdTbbEvalStencilsController(int numThreads=-1);
/// \brief Destructor.
~OsdTbbEvalStencilsController();
/// \brief Applies stencil weights to the control vertex data
///
/// Applies the stencil weights to the control vertex data to evaluate the
/// interpolated limit positions at the parametric locations of the stencils
///
/// @param context the OsdCpuEvalStencilsContext with the stencil weights
///
/// @param controlDataDesc vertex buffer descriptor for the control vertex data
///
/// @param controlVertices vertex buffer with the control vertices data
///
/// @param outputDataDesc vertex buffer descriptor for the output vertex data
///
/// @param outputData output vertex buffer for the interpolated data
///
template<class CONTROL_BUFFER, class OUTPUT_BUFFER>
int UpdateValues( OsdCpuEvalStencilsContext * context,
OsdVertexBufferDescriptor const & controlDataDesc, CONTROL_BUFFER *controlVertices,
OsdVertexBufferDescriptor const & outputDataDesc, OUTPUT_BUFFER *outputData ) {
if (not context->GetStencilTables()->GetNumStencils())
return 0;
context->BindControlData( controlDataDesc, controlVertices );
context->BindOutputData( outputDataDesc, outputData );
int n = _UpdateValues( context );
context->Unbind();
return n;
}
/// \brief Applies derivative stencil weights to the control vertex data
///
/// Computes the U and V derivative stencils to the control vertex data at
/// the parametric locations contained in each stencil
///
/// @param context the OsdCpuEvalStencilsContext with the stencil weights
///
/// @param controlDataDesc vertex buffer descriptor for the control vertex data
///
/// @param controlVertices vertex buffer with the control vertices data
///
/// @param outputDuDesc vertex buffer descriptor for the U derivative output data
///
/// @param outputDuData output vertex buffer for the U derivative data
///
/// @param outputDvDesc vertex buffer descriptor for the V deriv output data
///
/// @param outputDvData output vertex buffer for the V derivative data
///
template<class CONTROL_BUFFER, class OUTPUT_BUFFER>
int UpdateDerivs( OsdCpuEvalStencilsContext * context,
OsdVertexBufferDescriptor const & controlDataDesc, CONTROL_BUFFER *controlVertices,
OsdVertexBufferDescriptor const & outputDuDesc, OUTPUT_BUFFER *outputDuData,
OsdVertexBufferDescriptor const & outputDvDesc, OUTPUT_BUFFER *outputDvData ) {
if (not context->GetStencilTables()->GetNumStencils())
return 0;
context->BindControlData( controlDataDesc, controlVertices );
context->BindOutputDerivData( outputDuDesc, outputDuData, outputDvDesc, outputDvData );
int n = _UpdateDerivs( context );
context->Unbind();
return n;
}
/// Waits until all running subdivision kernels finish.
void Synchronize();
private:
int _UpdateValues( OsdCpuEvalStencilsContext * context );
int _UpdateDerivs( OsdCpuEvalStencilsContext * context );
int _numThreads;
};
} // end namespace OPENSUBDIV_VERSION
using namespace OPENSUBDIV_VERSION;
} // end namespace OpenSubdiv
#endif // FAR_TBB_EVALSTENCILS_CONTROLLER_H

View File

@ -56,12 +56,12 @@ struct OsdVertexDescriptor {
numVertexElements = numVertexElem;
numVaryingElements = numVaryingElem;
}
/// Resets the descriptor
void Reset() {
numVertexElements = numVaryingElements = 0;
}
/// Returns the total number of elements (vertex + varying)
int GetNumElements() const {
return numVertexElements + numVaryingElements;
@ -75,7 +75,7 @@ struct OsdVertexDescriptor {
/// Resets the contents of vertex & varying primvar data buffers for a given
/// vertex.
///
/// @param vertex The float array containing the vertex-interpolated primvar
/// @param vertex The float array containing the vertex-interpolated primvar
/// data that needs to be reset.
///
/// @param varying The float array containing the varying-interpolated primvar
@ -85,15 +85,15 @@ struct OsdVertexDescriptor {
///
void Clear(float *vertex, float *varying, int index) const {
if (vertex) {
memset(vertex+index*numVertexElements, 0, sizeof(float)*numVertexElements);
memset(vertex+index*numVertexElements, 0, sizeof(float)*numVertexElements);
}
if (varying) {
memset(varying+index*numVaryingElements, 0, sizeof(float)*numVaryingElements);
memset(varying+index*numVaryingElements, 0, sizeof(float)*numVaryingElements);
}
}
/// Applies "dst += src*weight" to "vertex" primvar data in a vertex buffer.
///
/// @param vertex The VertexData buffer
@ -104,14 +104,14 @@ struct OsdVertexDescriptor {
///
/// @param weight Weight applied to the primvar data.
///
inline
inline
void AddWithWeight(float *vertex, int dstIndex, int srcIndex, float weight) const {
int d = dstIndex * numVertexElements;
int s = srcIndex * numVertexElements;
int s = srcIndex * numVertexElements;
#if defined ( __INTEL_COMPILER ) or defined ( __ICC )
#pragma ivdep
#pragma ivdep
#pragma vector aligned
#endif
#endif
for (int i = 0; i < numVertexElements; ++i)
vertex[d++] += vertex[s++] * weight;
}
@ -131,9 +131,9 @@ struct OsdVertexDescriptor {
int d = dstIndex * numVaryingElements;
int s = srcIndex * numVaryingElements;
#if defined ( __INTEL_COMPILER ) or defined ( __ICC )
#pragma ivdep
#pragma ivdep
#pragma vector aligned
#endif
#endif
for (int i = 0; i < numVaryingElements; ++i)
varying[d++] += varying[s++] * weight;
}
@ -191,16 +191,15 @@ struct OsdVertexBufferDescriptor {
/// True if the descriptor values are internally consistent
bool IsValid() const {
return (length>0) and (offset<length) and (stride>=length);
return ((length>0) and (offset<stride) and (length<=stride-offset));
}
/// True if the 'other' descriptor can be used as a destination for
/// data evaluations.
bool CanEval( OsdVertexBufferDescriptor const & other ) const {
return IsValid() and
other.IsValid() and
(length==other.length) and
(other.length <= (stride-offset));
return (IsValid() and
other.IsValid() and
(length==other.length));
}
/// Resets the descriptor to default

View File

@ -26,6 +26,6 @@
#ifndef OPENSUBDIV_VERSION_H
#define OPENSUBDIV_VERSION_H
#define OPENSUBDIV_VERSION v2_1_0
#define OPENSUBDIV_VERSION v2_2_0
#endif /* OPENSUBDIV_VERSION_H */

View File

@ -51,7 +51,7 @@ if ( OPENCL_FOUND )
endif()
add_executable(osd_regression
_add_possibly_cuda_executable(osd_regression
${SOURCE_FILES}
)