OpenSubdiv/opensubdiv/osd/glComputeEvaluator.cpp

546 lines
20 KiB
C++
Raw Normal View History

Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
//
// Copyright 2015 Pixar
//
// Licensed under the Apache License, Version 2.0 (the "Apache License")
// with the following modification; you may not use this file except in
// compliance with the Apache License and the following modification to it:
// Section 6. Trademarks. is 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 to comply with Section 4(c) of
// the License and to reproduce the content of the NOTICE file.
//
// You may obtain a copy of the Apache License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the Apache License with the above modification is
// distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, either express or implied. See the Apache License for the specific
// language governing permissions and limitations under the Apache License.
//
#include "../osd/glComputeEvaluator.h"
#include "../osd/glslPatchShaderSource.h"
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
#include <cassert>
#include <sstream>
#include <string>
#include <vector>
#include "../far/error.h"
#include "../far/stencilTable.h"
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
namespace Osd {
static const char *shaderSource =
#include "../osd/glslComputeKernel.gen.h"
;
template <class T> GLuint
createSSBO(std::vector<T> const & src) {
if (src.empty()) {
return 0;
}
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
GLuint devicePtr = 0;
glGenBuffers(1, &devicePtr);
#if defined(GL_EXT_direct_state_access)
if (glNamedBufferDataEXT) {
glNamedBufferDataEXT(devicePtr, src.size()*sizeof(T),
&src.at(0), GL_STATIC_DRAW);
} else {
#else
{
#endif
GLint prev = 0;
glGetIntegerv(GL_SHADER_STORAGE_BUFFER_BINDING, &prev);
glBindBuffer(GL_SHADER_STORAGE_BUFFER, devicePtr);
glBufferData(GL_SHADER_STORAGE_BUFFER, src.size()*sizeof(T),
&src.at(0), GL_STATIC_DRAW);
glBindBuffer(GL_SHADER_STORAGE_BUFFER, prev);
}
return devicePtr;
}
GLStencilTableSSBO::GLStencilTableSSBO(
Far::StencilTable const *stencilTable) {
_numStencils = stencilTable->GetNumStencils();
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
if (_numStencils > 0) {
_sizes = createSSBO(stencilTable->GetSizes());
_offsets = createSSBO(stencilTable->GetOffsets());
_indices = createSSBO(stencilTable->GetControlIndices());
_weights = createSSBO(stencilTable->GetWeights());
_duWeights = _dvWeights = 0;
_duuWeights = _duvWeights = _dvvWeights = 0;
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
} else {
_sizes = _offsets = _indices = _weights = 0;
_duWeights = _dvWeights = 0;
_duuWeights = _duvWeights = _dvvWeights = 0;
}
}
GLStencilTableSSBO::GLStencilTableSSBO(
Far::LimitStencilTable const *limitStencilTable) {
_numStencils = limitStencilTable->GetNumStencils();
if (_numStencils > 0) {
_sizes = createSSBO(limitStencilTable->GetSizes());
_offsets = createSSBO(limitStencilTable->GetOffsets());
_indices = createSSBO(limitStencilTable->GetControlIndices());
_weights = createSSBO(limitStencilTable->GetWeights());
_duWeights = createSSBO(limitStencilTable->GetDuWeights());
_dvWeights = createSSBO(limitStencilTable->GetDvWeights());
_duuWeights = createSSBO(limitStencilTable->GetDuuWeights());
_duvWeights = createSSBO(limitStencilTable->GetDuvWeights());
_dvvWeights = createSSBO(limitStencilTable->GetDvvWeights());
} else {
_sizes = _offsets = _indices = _weights = 0;
_duWeights = _dvWeights = 0;
_duuWeights = _duvWeights = _dvvWeights = 0;
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
}
}
GLStencilTableSSBO::~GLStencilTableSSBO() {
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
if (_sizes) glDeleteBuffers(1, &_sizes);
if (_offsets) glDeleteBuffers(1, &_offsets);
if (_indices) glDeleteBuffers(1, &_indices);
if (_weights) glDeleteBuffers(1, &_weights);
if (_duWeights) glDeleteBuffers(1, &_duWeights);
if (_dvWeights) glDeleteBuffers(1, &_dvWeights);
if (_duuWeights) glDeleteBuffers(1, &_duuWeights);
if (_duvWeights) glDeleteBuffers(1, &_duvWeights);
if (_dvvWeights) glDeleteBuffers(1, &_dvvWeights);
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
}
// ---------------------------------------------------------------------------
GLComputeEvaluator::GLComputeEvaluator() : _workGroupSize(64) {
memset (&_stencilKernel, 0, sizeof(_stencilKernel));
memset (&_patchKernel, 0, sizeof(_patchKernel));
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
}
GLComputeEvaluator::~GLComputeEvaluator() {
}
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
static GLuint
compileKernel(BufferDescriptor const &srcDesc,
BufferDescriptor const &dstDesc,
BufferDescriptor const & duDesc,
BufferDescriptor const & dvDesc,
BufferDescriptor const & duuDesc,
BufferDescriptor const & duvDesc,
BufferDescriptor const & dvvDesc,
const char *kernelDefine,
int workGroupSize) {
GLuint program = glCreateProgram();
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
GLuint shader = glCreateShader(GL_COMPUTE_SHADER);
std::string patchBasisShaderSource =
GLSLPatchShaderSource::GetPatchBasisShaderSource();
const char *patchBasisShaderSourceDefine = "#define OSD_PATCH_BASIS_GLSL\n";
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
std::ostringstream defines;
defines << "#define LENGTH " << srcDesc.length << "\n"
<< "#define SRC_STRIDE " << srcDesc.stride << "\n"
<< "#define DST_STRIDE " << dstDesc.stride << "\n"
<< "#define WORK_GROUP_SIZE " << workGroupSize << "\n"
<< kernelDefine << "\n"
<< patchBasisShaderSourceDefine << "\n";
bool deriv1 = (duDesc.length > 0 || dvDesc.length > 0);
bool deriv2 = (duuDesc.length > 0 || duvDesc.length > 0 || dvvDesc.length > 0);
if (deriv1) {
defines << "#define OPENSUBDIV_GLSL_COMPUTE_USE_1ST_DERIVATIVES\n";
}
if (deriv2) {
defines << "#define OPENSUBDIV_GLSL_COMPUTE_USE_2ND_DERIVATIVES\n";
}
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
std::string defineStr = defines.str();
const char *shaderSources[4] = {"#version 430\n", 0, 0, 0};
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
shaderSources[1] = defineStr.c_str();
shaderSources[2] = patchBasisShaderSource.c_str();
shaderSources[3] = shaderSource;
glShaderSource(shader, 4, shaderSources, NULL);
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
glCompileShader(shader);
glAttachShader(program, shader);
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
GLint linked = 0;
glLinkProgram(program);
glGetProgramiv(program, GL_LINK_STATUS, &linked);
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
if (linked == GL_FALSE) {
char buffer[1024];
glGetShaderInfoLog(shader, 1024, NULL, buffer);
Far::Error(Far::FAR_RUNTIME_ERROR, buffer);
glGetProgramInfoLog(program, 1024, NULL, buffer);
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
Far::Error(Far::FAR_RUNTIME_ERROR, buffer);
glDeleteProgram(program);
return 0;
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
}
glDeleteShader(shader);
return program;
}
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
bool
GLComputeEvaluator::Compile(BufferDescriptor const &srcDesc,
BufferDescriptor const &dstDesc,
BufferDescriptor const &duDesc,
BufferDescriptor const &dvDesc,
BufferDescriptor const &duuDesc,
BufferDescriptor const &duvDesc,
BufferDescriptor const &dvvDesc) {
// create a stencil kernel
if (!_stencilKernel.Compile(srcDesc, dstDesc,
duDesc, dvDesc,
duuDesc, duvDesc, dvvDesc,
_workGroupSize)) {
return false;
}
// create a patch kernel
if (!_patchKernel.Compile(srcDesc, dstDesc,
duDesc, dvDesc,
duuDesc, duvDesc, dvvDesc,
_workGroupSize)) {
return false;
}
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
return true;
}
/* static */
void
GLComputeEvaluator::Synchronize(void * /*kernel*/) {
// XXX: this is currently just for the performance measuring purpose.
// need to be reimplemented by fence and sync.
glFinish();
}
bool
GLComputeEvaluator::EvalStencils(
GLuint srcBuffer, BufferDescriptor const &srcDesc,
GLuint dstBuffer, BufferDescriptor const &dstDesc,
GLuint duBuffer, BufferDescriptor const &duDesc,
GLuint dvBuffer, BufferDescriptor const &dvDesc,
GLuint sizesBuffer,
GLuint offsetsBuffer,
GLuint indicesBuffer,
GLuint weightsBuffer,
GLuint duWeightsBuffer,
GLuint dvWeightsBuffer,
int start, int end) const {
return EvalStencils(srcBuffer, srcDesc,
dstBuffer, dstDesc,
duBuffer, duDesc,
dvBuffer, dvDesc,
0, BufferDescriptor(),
0, BufferDescriptor(),
0, BufferDescriptor(),
sizesBuffer, offsetsBuffer, indicesBuffer,
weightsBuffer,
duWeightsBuffer, dvWeightsBuffer,
0, 0, 0,
start, end);
}
bool
GLComputeEvaluator::EvalStencils(
GLuint srcBuffer, BufferDescriptor const &srcDesc,
GLuint dstBuffer, BufferDescriptor const &dstDesc,
GLuint duBuffer, BufferDescriptor const &duDesc,
GLuint dvBuffer, BufferDescriptor const &dvDesc,
GLuint duuBuffer, BufferDescriptor const &duuDesc,
GLuint duvBuffer, BufferDescriptor const &duvDesc,
GLuint dvvBuffer, BufferDescriptor const &dvvDesc,
GLuint sizesBuffer,
GLuint offsetsBuffer,
GLuint indicesBuffer,
GLuint weightsBuffer,
GLuint duWeightsBuffer,
GLuint dvWeightsBuffer,
GLuint duuWeightsBuffer,
GLuint duvWeightsBuffer,
GLuint dvvWeightsBuffer,
int start, int end) const {
if (!_stencilKernel.program) return false;
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
int count = end - start;
if (count <= 0) {
return true;
}
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, srcBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, dstBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 2, duBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 3, dvBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 10, duuBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 11, duvBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 12, dvvBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 4, sizesBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 5, offsetsBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 6, indicesBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 7, weightsBuffer);
if (duWeightsBuffer)
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 8, duWeightsBuffer);
if (dvWeightsBuffer)
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 9, dvWeightsBuffer);
if (duuWeightsBuffer)
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 13, duuWeightsBuffer);
if (duvWeightsBuffer)
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 14, duvWeightsBuffer);
if (dvvWeightsBuffer)
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 15, dvvWeightsBuffer);
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
glUseProgram(_stencilKernel.program);
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
glUniform1i(_stencilKernel.uniformStart, start);
glUniform1i(_stencilKernel.uniformEnd, end);
glUniform1i(_stencilKernel.uniformSrcOffset, srcDesc.offset);
glUniform1i(_stencilKernel.uniformDstOffset, dstDesc.offset);
if (_stencilKernel.uniformDuDesc > 0) {
glUniform3i(_stencilKernel.uniformDuDesc,
duDesc.offset, duDesc.length, duDesc.stride);
}
if (_stencilKernel.uniformDvDesc > 0) {
glUniform3i(_stencilKernel.uniformDvDesc,
dvDesc.offset, dvDesc.length, dvDesc.stride);
}
if (_stencilKernel.uniformDuuDesc > 0) {
glUniform3i(_stencilKernel.uniformDuuDesc,
duuDesc.offset, duuDesc.length, duuDesc.stride);
}
if (_stencilKernel.uniformDuvDesc > 0) {
glUniform3i(_stencilKernel.uniformDuvDesc,
duvDesc.offset, duvDesc.length, duvDesc.stride);
}
if (_stencilKernel.uniformDvvDesc > 0) {
glUniform3i(_stencilKernel.uniformDvvDesc,
dvvDesc.offset, dvvDesc.length, dvvDesc.stride);
}
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
glDispatchCompute((count + _workGroupSize - 1) / _workGroupSize, 1, 1);
glUseProgram(0);
glMemoryBarrier(GL_TEXTURE_FETCH_BARRIER_BIT);
for (int i = 0; i < 16; ++i) {
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, i, 0);
}
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
return true;
}
bool
GLComputeEvaluator::EvalPatches(
GLuint srcBuffer, BufferDescriptor const &srcDesc,
GLuint dstBuffer, BufferDescriptor const &dstDesc,
GLuint duBuffer, BufferDescriptor const &duDesc,
GLuint dvBuffer, BufferDescriptor const &dvDesc,
int numPatchCoords,
GLuint patchCoordsBuffer,
const PatchArrayVector &patchArrays,
GLuint patchIndexBuffer,
GLuint patchParamsBuffer) const {
return EvalPatches(srcBuffer, srcDesc,
dstBuffer, dstDesc,
duBuffer, duDesc,
dvBuffer, dvDesc,
0, BufferDescriptor(),
0, BufferDescriptor(),
0, BufferDescriptor(),
numPatchCoords,
patchCoordsBuffer,
patchArrays,
patchIndexBuffer,
patchParamsBuffer);
}
bool
GLComputeEvaluator::EvalPatches(
GLuint srcBuffer, BufferDescriptor const &srcDesc,
GLuint dstBuffer, BufferDescriptor const &dstDesc,
GLuint duBuffer, BufferDescriptor const &duDesc,
GLuint dvBuffer, BufferDescriptor const &dvDesc,
GLuint duuBuffer, BufferDescriptor const &duuDesc,
GLuint duvBuffer, BufferDescriptor const &duvDesc,
GLuint dvvBuffer, BufferDescriptor const &dvvDesc,
int numPatchCoords,
GLuint patchCoordsBuffer,
const PatchArrayVector &patchArrays,
GLuint patchIndexBuffer,
GLuint patchParamsBuffer) const {
if (!_patchKernel.program) return false;
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, srcBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, dstBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 2, duBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 3, dvBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 10, duuBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 11, duvBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 12, dvvBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 4, patchCoordsBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 5, patchIndexBuffer);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 6, patchParamsBuffer);
glUseProgram(_patchKernel.program);
glUniform1i(_patchKernel.uniformSrcOffset, srcDesc.offset);
glUniform1i(_patchKernel.uniformDstOffset, dstDesc.offset);
glUniform4iv(_patchKernel.uniformPatchArray, (int)patchArrays.size(),
(const GLint*)&patchArrays[0]);
if (_patchKernel.uniformDuDesc > 0) {
glUniform3i(_patchKernel.uniformDuDesc,
duDesc.offset, duDesc.length, duDesc.stride);
}
if (_patchKernel.uniformDvDesc > 0) {
glUniform3i(_patchKernel.uniformDvDesc,
dvDesc.offset, dvDesc.length, dvDesc.stride);
}
if (_patchKernel.uniformDuuDesc > 0) {
glUniform3i(_patchKernel.uniformDuuDesc,
duuDesc.offset, duuDesc.length, duuDesc.stride);
}
if (_patchKernel.uniformDuvDesc > 0) {
glUniform3i(_patchKernel.uniformDuvDesc,
duvDesc.offset, duvDesc.length, duvDesc.stride);
}
if (_patchKernel.uniformDvvDesc > 0) {
glUniform3i(_patchKernel.uniformDvvDesc,
dvvDesc.offset, dvvDesc.length, dvvDesc.stride);
}
glDispatchCompute((numPatchCoords + _workGroupSize - 1) / _workGroupSize, 1, 1);
glUseProgram(0);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, 0);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, 0);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 2, 0);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 3, 0);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 4, 0);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 5, 0);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 6, 0);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 10, 0);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 11, 0);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 12, 0);
return true;
}
// ---------------------------------------------------------------------------
GLComputeEvaluator::_StencilKernel::_StencilKernel() : program(0) {
}
GLComputeEvaluator::_StencilKernel::~_StencilKernel() {
if (program) {
glDeleteProgram(program);
}
}
bool
GLComputeEvaluator::_StencilKernel::Compile(BufferDescriptor const &srcDesc,
BufferDescriptor const &dstDesc,
BufferDescriptor const &duDesc,
BufferDescriptor const &dvDesc,
BufferDescriptor const &duuDesc,
BufferDescriptor const &duvDesc,
BufferDescriptor const &dvvDesc,
int workGroupSize) {
// create stencil kernel
if (program) {
glDeleteProgram(program);
}
const char * kernelDefine =
"#define OPENSUBDIV_GLSL_COMPUTE_KERNEL_EVAL_STENCILS\n";
program = compileKernel(srcDesc, dstDesc,
duDesc, dvDesc, duuDesc, duvDesc, dvvDesc,
kernelDefine, workGroupSize);
if (program == 0) return false;
// cache uniform locations (TODO: use uniform block)
uniformStart = glGetUniformLocation(program, "batchStart");
uniformEnd = glGetUniformLocation(program, "batchEnd");
uniformSrcOffset = glGetUniformLocation(program, "srcOffset");
uniformDstOffset = glGetUniformLocation(program, "dstOffset");
uniformDuDesc = glGetUniformLocation(program, "duDesc");
uniformDvDesc = glGetUniformLocation(program, "dvDesc");
uniformDuuDesc = glGetUniformLocation(program, "duuDesc");
uniformDuvDesc = glGetUniformLocation(program, "duvDesc");
uniformDvvDesc = glGetUniformLocation(program, "dvvDesc");
return true;
}
// ---------------------------------------------------------------------------
GLComputeEvaluator::_PatchKernel::_PatchKernel() : program(0) {
}
GLComputeEvaluator::_PatchKernel::~_PatchKernel() {
if (program) {
glDeleteProgram(program);
}
}
bool
GLComputeEvaluator::_PatchKernel::Compile(BufferDescriptor const &srcDesc,
BufferDescriptor const &dstDesc,
BufferDescriptor const &duDesc,
BufferDescriptor const &dvDesc,
BufferDescriptor const &duuDesc,
BufferDescriptor const &duvDesc,
BufferDescriptor const &dvvDesc,
int workGroupSize) {
// create stencil kernel
if (program) {
glDeleteProgram(program);
}
const char * kernelDefine =
"#define OPENSUBDIV_GLSL_COMPUTE_KERNEL_EVAL_PATCHES\n";
program = compileKernel(srcDesc, dstDesc,
duDesc, dvDesc, duuDesc, duvDesc, dvvDesc,
kernelDefine, workGroupSize);
if (program == 0) return false;
// cache uniform locations
uniformSrcOffset = glGetUniformLocation(program, "srcOffset");
uniformDstOffset = glGetUniformLocation(program, "dstOffset");
uniformPatchArray = glGetUniformLocation(program, "patchArray");
uniformDuDesc = glGetUniformLocation(program, "duDesc");
uniformDvDesc = glGetUniformLocation(program, "dvDesc");
uniformDuuDesc = glGetUniformLocation(program, "duuDesc");
uniformDuvDesc = glGetUniformLocation(program, "duvDesc");
uniformDvvDesc = glGetUniformLocation(program, "dvvDesc");
return true;
}
Refurbish osd layer API. In OpenSubdiv 2.x, we encapsulated subdivision tables into compute context in osd layer since those tables are order-dependent and have to be applied in a certain manner. In 3.0, we adopted stencil table based refinement. It's more simple and such an encapsulation is no longer needed. Also 2.0 API has several ownership issues of GPU kernel caching, and forces unnecessary instantiation of controllers even though the cpu kernels typically don't need instances unlike GPU ones. This change completely revisit osd client facing APIs. All contexts and controllers were replaced with device-specific tables and evaluators. While we can still use consistent API across various device backends, unnecessary complexities have been removed. For example, cpu evaluator is just a set of static functions and also there's no need to replicate FarStencilTables to ComputeContext. Also the new API delegates the ownership of compiled GPU kernels to clients, for the better management of resources especially in multiple GPU environment. In addition to integrating ComputeController and EvalStencilController into a single function Evaluator::EvalStencils(), EvalLimit API is also added into Evaluator. This is working but still in progress, and we'll make a followup change for the complete implementation. -some naming convention changes: GLSLTransformFeedback to GLXFBEvaluator GLSLCompute to GLComputeEvaluator -move LimitLocation struct into examples/glEvalLimit. We're still discussing patch evaluation interface. Basically we'd like to tease all ptex-specific parametrization out of far/osd layer. TODO: -implments EvalPatches() in the right way -derivative evaluation API is still interim. -VertexBufferDescriptor needs a better API to advance its location -synchronization mechanism is not ideal (too global). -OsdMesh class is hacky. need to fix it.
2015-05-09 00:31:26 +00:00
} // end namespace Osd
} // end namespace OPENSUBDIV_VERSION
} // end namespace OpenSubdiv