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.
|
|
|
|
//
|
|
|
|
|
2015-05-19 18:22:37 +00:00
|
|
|
#ifndef OPENSUBDIV_OPENSUBDIV3_OSD_CL_EVALUATOR_H
|
|
|
|
#define OPENSUBDIV_OPENSUBDIV3_OSD_CL_EVALUATOR_H
|
2015-05-09 00:31:26 +00:00
|
|
|
|
|
|
|
#include "../version.h"
|
|
|
|
|
|
|
|
#include "../osd/opencl.h"
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
#include "../osd/types.h"
|
2015-05-09 00:31:26 +00:00
|
|
|
#include "../osd/vertexDescriptor.h"
|
|
|
|
|
|
|
|
namespace OpenSubdiv {
|
|
|
|
namespace OPENSUBDIV_VERSION {
|
|
|
|
|
|
|
|
namespace Far {
|
2015-05-22 18:50:01 +00:00
|
|
|
class StencilTable;
|
2015-05-09 00:31:26 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
namespace Osd {
|
|
|
|
|
2015-05-22 18:50:01 +00:00
|
|
|
/// \brief OpenCL stencil table
|
2015-05-09 00:31:26 +00:00
|
|
|
///
|
2015-05-22 18:50:01 +00:00
|
|
|
/// This class is an OpenCL buffer representation of Far::StencilTable.
|
2015-05-09 00:31:26 +00:00
|
|
|
///
|
|
|
|
/// CLCompute consumes this table to apply stencils
|
|
|
|
///
|
|
|
|
///
|
2015-05-22 18:50:01 +00:00
|
|
|
class CLStencilTable {
|
2015-05-09 00:31:26 +00:00
|
|
|
public:
|
|
|
|
template <typename DEVICE_CONTEXT>
|
2015-05-22 18:50:01 +00:00
|
|
|
static CLStencilTable *Create(Far::StencilTable const *stencilTable,
|
|
|
|
DEVICE_CONTEXT context) {
|
|
|
|
return new CLStencilTable(stencilTable, context->GetContext());
|
2015-05-09 00:31:26 +00:00
|
|
|
}
|
|
|
|
|
2015-05-28 00:23:36 +00:00
|
|
|
template <typename DEVICE_CONTEXT>
|
|
|
|
static CLStencilTable *Create(
|
|
|
|
Far::LimitStencilTable const *limitStencilTable,
|
|
|
|
DEVICE_CONTEXT context) {
|
|
|
|
return new CLStencilTable(limitStencilTable, context->GetContext());
|
|
|
|
}
|
|
|
|
|
2015-05-22 18:50:01 +00:00
|
|
|
CLStencilTable(Far::StencilTable const *stencilTable,
|
|
|
|
cl_context clContext);
|
2015-05-28 00:23:36 +00:00
|
|
|
CLStencilTable(Far::LimitStencilTable const *limitStencilTable,
|
|
|
|
cl_context clContext);
|
2015-05-22 18:50:01 +00:00
|
|
|
~CLStencilTable();
|
2015-05-09 00:31:26 +00:00
|
|
|
|
|
|
|
// interfaces needed for CLComputeKernel
|
|
|
|
cl_mem GetSizesBuffer() const { return _sizes; }
|
|
|
|
cl_mem GetOffsetsBuffer() const { return _offsets; }
|
|
|
|
cl_mem GetIndicesBuffer() const { return _indices; }
|
|
|
|
cl_mem GetWeightsBuffer() const { return _weights; }
|
2015-05-28 00:23:36 +00:00
|
|
|
cl_mem GetDuWeightsBuffer() const { return _duWeights; }
|
|
|
|
cl_mem GetDvWeightsBuffer() const { return _dvWeights; }
|
2015-05-09 00:31:26 +00:00
|
|
|
int GetNumStencils() const { return _numStencils; }
|
|
|
|
|
|
|
|
private:
|
|
|
|
cl_mem _sizes;
|
|
|
|
cl_mem _offsets;
|
|
|
|
cl_mem _indices;
|
|
|
|
cl_mem _weights;
|
2015-05-28 00:23:36 +00:00
|
|
|
cl_mem _duWeights;
|
|
|
|
cl_mem _dvWeights;
|
2015-05-09 00:31:26 +00:00
|
|
|
int _numStencils;
|
|
|
|
};
|
|
|
|
|
|
|
|
// ---------------------------------------------------------------------------
|
|
|
|
|
|
|
|
class CLEvaluator {
|
|
|
|
public:
|
|
|
|
typedef bool Instantiatable;
|
|
|
|
/// Constructor.
|
|
|
|
CLEvaluator(cl_context context, cl_command_queue queue);
|
|
|
|
|
|
|
|
/// Desctructor.
|
|
|
|
~CLEvaluator();
|
|
|
|
|
|
|
|
/// Generic creator template.
|
|
|
|
template <typename DEVICE_CONTEXT>
|
|
|
|
static CLEvaluator *Create(VertexBufferDescriptor const &srcDesc,
|
|
|
|
VertexBufferDescriptor const &dstDesc,
|
2015-05-28 00:23:36 +00:00
|
|
|
VertexBufferDescriptor const &duDesc,
|
|
|
|
VertexBufferDescriptor const &dvDesc,
|
2015-05-09 00:31:26 +00:00
|
|
|
DEVICE_CONTEXT deviceContext) {
|
2015-05-28 00:23:36 +00:00
|
|
|
return Create(srcDesc, dstDesc, duDesc, dvDesc,
|
2015-05-09 00:31:26 +00:00
|
|
|
deviceContext->GetContext(),
|
|
|
|
deviceContext->GetCommandQueue());
|
|
|
|
}
|
|
|
|
|
|
|
|
static CLEvaluator * Create(VertexBufferDescriptor const &srcDesc,
|
|
|
|
VertexBufferDescriptor const &dstDesc,
|
2015-05-28 00:23:36 +00:00
|
|
|
VertexBufferDescriptor const &duDesc,
|
|
|
|
VertexBufferDescriptor const &dvDesc,
|
2015-05-09 00:31:26 +00:00
|
|
|
cl_context clContext,
|
|
|
|
cl_command_queue clCommandQueue) {
|
|
|
|
CLEvaluator *kernel = new CLEvaluator(clContext, clCommandQueue);
|
2015-05-28 00:23:36 +00:00
|
|
|
if (kernel->Compile(srcDesc, dstDesc, duDesc, dvDesc)) return kernel;
|
2015-05-09 00:31:26 +00:00
|
|
|
delete kernel;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
/// ----------------------------------------------------------------------
|
|
|
|
///
|
|
|
|
/// Stencil evaluations with StencilTable
|
|
|
|
///
|
|
|
|
/// ----------------------------------------------------------------------
|
|
|
|
|
2015-05-09 00:31:26 +00:00
|
|
|
/// \brief Generic static compute function. This function has a same
|
|
|
|
/// signature as other device kernels have so that it can be called
|
|
|
|
/// transparently from OsdMesh template interface.
|
|
|
|
///
|
|
|
|
/// @param srcBuffer Input primvar buffer.
|
|
|
|
/// must have BindCLBuffer() method returning a
|
|
|
|
/// const float pointer for read
|
|
|
|
///
|
|
|
|
/// @param srcDesc vertex buffer descriptor for the input buffer
|
|
|
|
///
|
|
|
|
/// @param dstBuffer Output primvar buffer
|
|
|
|
/// must have BindCLBuffer() method returning a
|
|
|
|
/// float pointer for write
|
|
|
|
///
|
|
|
|
/// @param dstDesc vertex buffer descriptor for the output buffer
|
|
|
|
///
|
2015-05-22 18:50:01 +00:00
|
|
|
/// @param stencilTable stencil table to be applied. The table must have
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
/// SSBO interfaces.
|
2015-05-09 00:31:26 +00:00
|
|
|
///
|
|
|
|
/// @param instance cached compiled instance. Clients are supposed to
|
|
|
|
/// pre-compile an instance of this class and provide
|
|
|
|
/// to this function. If it's null the kernel still
|
|
|
|
/// compute by instantiating on-demand kernel although
|
|
|
|
/// it may cause a performance problem.
|
|
|
|
///
|
|
|
|
/// @param deviceContext client providing context class which supports
|
|
|
|
/// cL_context GetContext()
|
|
|
|
/// cl_command_queue GetCommandQueue()
|
|
|
|
/// methods.
|
|
|
|
///
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
template <typename SRC_BUFFER, typename DST_BUFFER,
|
|
|
|
typename STENCIL_TABLE, typename DEVICE_CONTEXT>
|
|
|
|
static bool EvalStencils(
|
|
|
|
SRC_BUFFER *srcBuffer, VertexBufferDescriptor const &srcDesc,
|
|
|
|
DST_BUFFER *dstBuffer, VertexBufferDescriptor const &dstDesc,
|
|
|
|
STENCIL_TABLE const *stencilTable,
|
|
|
|
CLEvaluator const *instance,
|
|
|
|
DEVICE_CONTEXT deviceContext) {
|
|
|
|
|
2015-05-09 00:31:26 +00:00
|
|
|
if (instance) {
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
return instance->EvalStencils(srcBuffer, srcDesc,
|
|
|
|
dstBuffer, dstDesc,
|
2015-05-09 00:31:26 +00:00
|
|
|
stencilTable);
|
|
|
|
} else {
|
|
|
|
// Create an instance on demand (slow)
|
2015-05-28 00:23:36 +00:00
|
|
|
instance = Create(srcDesc, dstDesc,
|
|
|
|
VertexBufferDescriptor(),
|
|
|
|
VertexBufferDescriptor(),
|
|
|
|
deviceContext);
|
2015-05-09 00:31:26 +00:00
|
|
|
if (instance) {
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
bool r = instance->EvalStencils(srcBuffer, srcDesc,
|
|
|
|
dstBuffer, dstDesc,
|
2015-05-09 00:31:26 +00:00
|
|
|
stencilTable);
|
|
|
|
delete instance;
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-05-28 00:23:36 +00:00
|
|
|
/// \brief Generic static compute function. This function has a same
|
|
|
|
/// signature as other device kernels have so that it can be called
|
|
|
|
/// transparently from OsdMesh template interface.
|
|
|
|
///
|
|
|
|
/// @param srcBuffer Input primvar buffer.
|
|
|
|
/// must have BindCLBuffer() method returning a
|
|
|
|
/// const float pointer for read
|
|
|
|
///
|
|
|
|
/// @param srcDesc vertex buffer descriptor for the input buffer
|
|
|
|
///
|
|
|
|
/// @param dstBuffer Output primvar buffer
|
|
|
|
/// must have BindCLBuffer() method returning a
|
|
|
|
/// float pointer for write
|
|
|
|
///
|
|
|
|
/// @param dstDesc vertex buffer descriptor for the output buffer
|
|
|
|
///
|
|
|
|
/// @param stencilTable stencil table to be applied. The table must have
|
|
|
|
/// SSBO interfaces.
|
|
|
|
///
|
|
|
|
/// @param instance cached compiled instance. Clients are supposed to
|
|
|
|
/// pre-compile an instance of this class and provide
|
|
|
|
/// to this function. If it's null the kernel still
|
|
|
|
/// compute by instantiating on-demand kernel although
|
|
|
|
/// it may cause a performance problem.
|
|
|
|
///
|
|
|
|
/// @param deviceContext client providing context class which supports
|
|
|
|
/// cL_context GetContext()
|
|
|
|
/// cl_command_queue GetCommandQueue()
|
|
|
|
/// methods.
|
|
|
|
///
|
|
|
|
template <typename SRC_BUFFER, typename DST_BUFFER,
|
|
|
|
typename STENCIL_TABLE, typename DEVICE_CONTEXT>
|
|
|
|
static bool EvalStencils(
|
|
|
|
SRC_BUFFER *srcBuffer, VertexBufferDescriptor const &srcDesc,
|
|
|
|
DST_BUFFER *dstBuffer, VertexBufferDescriptor const &dstDesc,
|
|
|
|
DST_BUFFER *duBuffer, VertexBufferDescriptor const &duDesc,
|
|
|
|
DST_BUFFER *dvBuffer, VertexBufferDescriptor const &dvDesc,
|
|
|
|
STENCIL_TABLE const *stencilTable,
|
|
|
|
CLEvaluator const *instance,
|
|
|
|
DEVICE_CONTEXT deviceContext) {
|
|
|
|
|
|
|
|
if (instance) {
|
|
|
|
return instance->EvalStencils(srcBuffer, srcDesc,
|
|
|
|
dstBuffer, dstDesc,
|
|
|
|
duBuffer, duDesc,
|
|
|
|
dvBuffer, dvDesc,
|
|
|
|
stencilTable);
|
|
|
|
} else {
|
|
|
|
// Create an instance on demand (slow)
|
|
|
|
instance = Create(srcDesc, dstDesc, duDesc, dvDesc,
|
|
|
|
deviceContext);
|
|
|
|
if (instance) {
|
|
|
|
bool r = instance->EvalStencils(srcBuffer, srcDesc,
|
|
|
|
dstBuffer, dstDesc,
|
|
|
|
duBuffer, duDesc,
|
|
|
|
dvBuffer, dvDesc,
|
|
|
|
stencilTable);
|
|
|
|
delete instance;
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/// Generic compute function.
|
|
|
|
/// Dispatch the CL compute kernel asynchronously.
|
|
|
|
/// Returns false if the kernel hasn't been compiled yet.
|
|
|
|
template <typename SRC_BUFFER, typename DST_BUFFER, typename STENCIL_TABLE>
|
|
|
|
bool EvalStencils(
|
|
|
|
SRC_BUFFER *srcBuffer, VertexBufferDescriptor const &srcDesc,
|
|
|
|
DST_BUFFER *dstBuffer, VertexBufferDescriptor const &dstDesc,
|
|
|
|
STENCIL_TABLE const *stencilTable) const {
|
|
|
|
return EvalStencils(srcBuffer->BindCLBuffer(_clCommandQueue), srcDesc,
|
|
|
|
dstBuffer->BindCLBuffer(_clCommandQueue), dstDesc,
|
|
|
|
stencilTable->GetSizesBuffer(),
|
|
|
|
stencilTable->GetOffsetsBuffer(),
|
|
|
|
stencilTable->GetIndicesBuffer(),
|
|
|
|
stencilTable->GetWeightsBuffer(),
|
|
|
|
0,
|
|
|
|
stencilTable->GetNumStencils());
|
|
|
|
}
|
|
|
|
|
2015-05-09 00:31:26 +00:00
|
|
|
/// Generic compute function.
|
|
|
|
/// Dispatch the CL compute kernel asynchronously.
|
|
|
|
/// Returns false if the kernel hasn't been compiled yet.
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
template <typename SRC_BUFFER, typename DST_BUFFER, typename STENCIL_TABLE>
|
|
|
|
bool EvalStencils(
|
|
|
|
SRC_BUFFER *srcBuffer, VertexBufferDescriptor const &srcDesc,
|
|
|
|
DST_BUFFER *dstBuffer, VertexBufferDescriptor const &dstDesc,
|
2015-05-28 00:23:36 +00:00
|
|
|
DST_BUFFER *duBuffer, VertexBufferDescriptor const &duDesc,
|
|
|
|
DST_BUFFER *dvBuffer, VertexBufferDescriptor const &dvDesc,
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
STENCIL_TABLE const *stencilTable) const {
|
2015-05-28 00:23:36 +00:00
|
|
|
return EvalStencils(srcBuffer->BindCLBuffer(_clCommandQueue), srcDesc,
|
|
|
|
dstBuffer->BindCLBuffer(_clCommandQueue), dstDesc,
|
|
|
|
duBuffer->BindCLBuffer(_clCommandQueue), duDesc,
|
|
|
|
dvBuffer->BindCLBuffer(_clCommandQueue), dvDesc,
|
2015-05-09 00:31:26 +00:00
|
|
|
stencilTable->GetSizesBuffer(),
|
|
|
|
stencilTable->GetOffsetsBuffer(),
|
|
|
|
stencilTable->GetIndicesBuffer(),
|
|
|
|
stencilTable->GetWeightsBuffer(),
|
2015-05-28 00:23:36 +00:00
|
|
|
stencilTable->GetDuWeightsBuffer(),
|
|
|
|
stencilTable->GetDvWeightsBuffer(),
|
2015-05-09 00:31:26 +00:00
|
|
|
0,
|
|
|
|
stencilTable->GetNumStencils());
|
|
|
|
}
|
|
|
|
|
|
|
|
/// Dispatch the CL compute kernel asynchronously.
|
|
|
|
/// returns false if the kernel hasn't been compiled yet.
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
bool EvalStencils(cl_mem src, VertexBufferDescriptor const &srcDesc,
|
|
|
|
cl_mem dst, VertexBufferDescriptor const &dstDesc,
|
2015-05-09 00:31:26 +00:00
|
|
|
cl_mem sizes,
|
|
|
|
cl_mem offsets,
|
|
|
|
cl_mem indices,
|
|
|
|
cl_mem weights,
|
|
|
|
int start,
|
|
|
|
int end) const;
|
|
|
|
|
2015-05-28 00:23:36 +00:00
|
|
|
/// Dispatch the CL compute kernel asynchronously.
|
|
|
|
/// returns false if the kernel hasn't been compiled yet.
|
|
|
|
bool EvalStencils(cl_mem src, VertexBufferDescriptor const &srcDesc,
|
|
|
|
cl_mem dst, VertexBufferDescriptor const &dstDesc,
|
|
|
|
cl_mem du, VertexBufferDescriptor const &duDesc,
|
|
|
|
cl_mem dv, VertexBufferDescriptor const &dvDesc,
|
|
|
|
cl_mem sizes,
|
|
|
|
cl_mem offsets,
|
|
|
|
cl_mem indices,
|
|
|
|
cl_mem weights,
|
|
|
|
cl_mem duWeights,
|
|
|
|
cl_mem dvWeights,
|
|
|
|
int start,
|
|
|
|
int end) const;
|
|
|
|
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
/// ----------------------------------------------------------------------
|
|
|
|
///
|
|
|
|
/// Limit evaluations with PatchTable
|
|
|
|
///
|
|
|
|
/// ----------------------------------------------------------------------
|
|
|
|
///
|
|
|
|
/// \brief Generic limit eval function. This function has a same
|
|
|
|
/// signature as other device kernels have so that it can be called
|
|
|
|
/// in the same way.
|
|
|
|
///
|
|
|
|
/// @param srcBuffer Input primvar buffer.
|
|
|
|
/// must have BindCLBuffer() method returning a CL
|
|
|
|
/// buffer object of source data
|
|
|
|
///
|
|
|
|
/// @param srcDesc vertex buffer descriptor for the input buffer
|
|
|
|
///
|
|
|
|
/// @param dstBuffer Output primvar buffer
|
|
|
|
/// must have BindCLBuffer() method returning a CL
|
|
|
|
/// buffer object of destination data
|
|
|
|
///
|
|
|
|
/// @param dstDesc vertex buffer descriptor for the output buffer
|
|
|
|
///
|
|
|
|
/// @param numPatchCoords number of patchCoords.
|
|
|
|
///
|
|
|
|
/// @param patchCoords array of locations to be evaluated.
|
|
|
|
/// must have BindCLBuffer() method returning an
|
|
|
|
/// array of PatchCoord struct.
|
|
|
|
///
|
|
|
|
/// @param patchTable CLPatchTable or equivalent
|
|
|
|
///
|
|
|
|
/// @param instance cached compiled instance. Clients are supposed to
|
|
|
|
/// pre-compile an instance of this class and provide
|
|
|
|
/// to this function. If it's null the kernel still
|
|
|
|
/// compute by instantiating on-demand kernel although
|
|
|
|
/// it may cause a performance problem.
|
|
|
|
///
|
|
|
|
/// @param deviceContext client providing context class which supports
|
|
|
|
/// cL_context GetContext()
|
|
|
|
/// cl_command_queue GetCommandQueue()
|
|
|
|
/// methods.
|
|
|
|
///
|
|
|
|
template <typename SRC_BUFFER, typename DST_BUFFER,
|
|
|
|
typename PATCHCOORD_BUFFER, typename PATCH_TABLE,
|
|
|
|
typename DEVICE_CONTEXT>
|
|
|
|
static bool EvalPatches(
|
|
|
|
SRC_BUFFER *srcBuffer, VertexBufferDescriptor const &srcDesc,
|
|
|
|
DST_BUFFER *dstBuffer, VertexBufferDescriptor const &dstDesc,
|
|
|
|
int numPatchCoords,
|
|
|
|
PATCHCOORD_BUFFER *patchCoords,
|
|
|
|
PATCH_TABLE *patchTable,
|
|
|
|
CLEvaluator const *instance,
|
|
|
|
DEVICE_CONTEXT deviceContext) {
|
|
|
|
|
|
|
|
if (instance) {
|
|
|
|
return instance->EvalPatches(srcBuffer, srcDesc,
|
|
|
|
dstBuffer, dstDesc,
|
|
|
|
numPatchCoords, patchCoords,
|
|
|
|
patchTable);
|
|
|
|
} else {
|
|
|
|
// Create an instance on demand (slow)
|
|
|
|
(void)deviceContext; // unused
|
2015-05-28 00:23:36 +00:00
|
|
|
instance = Create(srcDesc, dstDesc,
|
|
|
|
VertexBufferDescriptor(),
|
|
|
|
VertexBufferDescriptor(),
|
|
|
|
deviceContext);
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
if (instance) {
|
|
|
|
bool r = instance->EvalPatches(srcBuffer, srcDesc,
|
|
|
|
dstBuffer, dstDesc,
|
|
|
|
numPatchCoords, patchCoords,
|
|
|
|
patchTable);
|
|
|
|
delete instance;
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/// \brief Generic limit eval function. This function has a same
|
|
|
|
/// signature as other device kernels have so that it can be called
|
|
|
|
/// in the same way.
|
|
|
|
///
|
|
|
|
/// @param srcBuffer Input primvar buffer.
|
|
|
|
/// must have BindCLBuffer() method returning a CL
|
|
|
|
/// buffer object of source data
|
|
|
|
///
|
|
|
|
/// @param srcDesc vertex buffer descriptor for the input buffer
|
|
|
|
///
|
|
|
|
/// @param dstBuffer Output primvar buffer
|
|
|
|
/// must have BindCLBuffer() method returning a CL
|
|
|
|
/// buffer object of destination data
|
|
|
|
///
|
|
|
|
/// @param dstDesc vertex buffer descriptor for the output buffer
|
|
|
|
///
|
|
|
|
/// @param duBuffer
|
|
|
|
///
|
|
|
|
/// @param duDesc
|
|
|
|
///
|
|
|
|
/// @param dvBuffer
|
|
|
|
///
|
|
|
|
/// @param dvDesc
|
|
|
|
///
|
|
|
|
/// @param numPatchCoords number of patchCoords.
|
|
|
|
///
|
|
|
|
/// @param patchCoords array of locations to be evaluated.
|
|
|
|
/// must have BindCLBuffer() method returning an
|
|
|
|
/// array of PatchCoord struct
|
|
|
|
///
|
|
|
|
/// @param patchTable CLPatchTable or equivalent
|
|
|
|
///
|
|
|
|
/// @param instance cached compiled instance. Clients are supposed to
|
|
|
|
/// pre-compile an instance of this class and provide
|
|
|
|
/// to this function. If it's null the kernel still
|
|
|
|
/// compute by instantiating on-demand kernel although
|
|
|
|
/// it may cause a performance problem.
|
|
|
|
///
|
|
|
|
/// @param deviceContext client providing context class which supports
|
|
|
|
/// cL_context GetContext()
|
|
|
|
/// cl_command_queue GetCommandQueue()
|
|
|
|
/// methods.
|
|
|
|
///
|
|
|
|
template <typename SRC_BUFFER, typename DST_BUFFER,
|
|
|
|
typename PATCHCOORD_BUFFER, typename PATCH_TABLE,
|
|
|
|
typename DEVICE_CONTEXT>
|
|
|
|
static bool EvalPatches(
|
|
|
|
SRC_BUFFER *srcBuffer, VertexBufferDescriptor const &srcDesc,
|
|
|
|
DST_BUFFER *dstBuffer, VertexBufferDescriptor const &dstDesc,
|
|
|
|
DST_BUFFER *duBuffer, VertexBufferDescriptor const &duDesc,
|
|
|
|
DST_BUFFER *dvBuffer, VertexBufferDescriptor const &dvDesc,
|
|
|
|
int numPatchCoords,
|
|
|
|
PATCHCOORD_BUFFER *patchCoords,
|
|
|
|
PATCH_TABLE *patchTable,
|
|
|
|
CLEvaluator const *instance,
|
|
|
|
DEVICE_CONTEXT deviceContext) {
|
|
|
|
|
|
|
|
if (instance) {
|
|
|
|
return instance->EvalPatches(srcBuffer, srcDesc,
|
|
|
|
dstBuffer, dstDesc,
|
|
|
|
duBuffer, duDesc,
|
|
|
|
dvBuffer, dvDesc,
|
|
|
|
numPatchCoords, patchCoords,
|
|
|
|
patchTable);
|
|
|
|
} else {
|
|
|
|
// Create an instance on demand (slow)
|
|
|
|
(void)deviceContext; // unused
|
2015-05-28 00:23:36 +00:00
|
|
|
instance = Create(srcDesc, dstDesc, duDesc, dvDesc, deviceContext);
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
if (instance) {
|
|
|
|
bool r = instance->EvalPatches(srcBuffer, srcDesc,
|
|
|
|
dstBuffer, dstDesc,
|
|
|
|
duBuffer, duDesc,
|
|
|
|
dvBuffer, dvDesc,
|
|
|
|
numPatchCoords, patchCoords,
|
|
|
|
patchTable);
|
|
|
|
delete instance;
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/// \brief Generic limit eval function. This function has a same
|
|
|
|
/// signature as other device kernels have so that it can be called
|
|
|
|
/// in the same way.
|
|
|
|
///
|
|
|
|
/// @param srcBuffer Input primvar buffer.
|
|
|
|
/// must have BindCLBuffer() method returning a CL
|
|
|
|
/// buffer object of source data
|
|
|
|
///
|
|
|
|
/// @param srcDesc vertex buffer descriptor for the input buffer
|
|
|
|
///
|
|
|
|
/// @param dstBuffer Output primvar buffer
|
|
|
|
/// must have BindCLBuffer() method returning a CL
|
|
|
|
/// buffer object of destination data
|
|
|
|
///
|
|
|
|
/// @param dstDesc vertex buffer descriptor for the output buffer
|
|
|
|
///
|
|
|
|
/// @param numPatchCoords number of patchCoords.
|
|
|
|
///
|
|
|
|
/// @param patchCoords array of locations to be evaluated.
|
|
|
|
/// must have BindCLBuffer() method returning an
|
|
|
|
/// array of PatchCoord struct.
|
|
|
|
///
|
|
|
|
/// @param patchTable CLPatchTable or equivalent
|
|
|
|
///
|
|
|
|
template <typename SRC_BUFFER, typename DST_BUFFER,
|
|
|
|
typename PATCHCOORD_BUFFER, typename PATCH_TABLE>
|
|
|
|
bool EvalPatches(
|
|
|
|
SRC_BUFFER *srcBuffer, VertexBufferDescriptor const &srcDesc,
|
|
|
|
DST_BUFFER *dstBuffer, VertexBufferDescriptor const &dstDesc,
|
|
|
|
int numPatchCoords,
|
|
|
|
PATCHCOORD_BUFFER *patchCoords,
|
|
|
|
PATCH_TABLE *patchTable) const {
|
|
|
|
|
|
|
|
return EvalPatches(srcBuffer->BindCLBuffer(_clCommandQueue), srcDesc,
|
|
|
|
dstBuffer->BindCLBuffer(_clCommandQueue), dstDesc,
|
|
|
|
0, VertexBufferDescriptor(),
|
|
|
|
0, VertexBufferDescriptor(),
|
|
|
|
numPatchCoords,
|
|
|
|
patchCoords->BindCLBuffer(_clCommandQueue),
|
|
|
|
patchTable->GetPatchArrayBuffer(),
|
|
|
|
patchTable->GetPatchIndexBuffer(),
|
|
|
|
patchTable->GetPatchParamBuffer());
|
|
|
|
}
|
|
|
|
|
|
|
|
/// \brief Generic limit eval function with derivatives. This function has
|
|
|
|
/// a same signature as other device kernels have so that it can be
|
|
|
|
/// called in the same way.
|
|
|
|
///
|
|
|
|
/// @param srcBuffer Input primvar buffer.
|
|
|
|
/// must have BindCLBuffer() method returning a CL
|
|
|
|
/// buffer object of source data
|
|
|
|
///
|
|
|
|
/// @param srcDesc vertex buffer descriptor for the input buffer
|
|
|
|
///
|
|
|
|
/// @param dstBuffer Output primvar buffer
|
|
|
|
/// must have BindCLBuffer() method returning a CL
|
|
|
|
/// buffer object of destination data
|
|
|
|
///
|
|
|
|
/// @param dstDesc vertex buffer descriptor for the output buffer
|
|
|
|
///
|
|
|
|
/// @param duBuffer Output U-derivatives buffer
|
|
|
|
/// must have BindCLBuffer() method returning a CL
|
|
|
|
/// buffer object of destination data of Du
|
|
|
|
///
|
|
|
|
/// @param duDesc vertex buffer descriptor for the duBuffer
|
|
|
|
///
|
|
|
|
/// @param dvBuffer Output V-derivatives buffer
|
|
|
|
/// must have BindCLBuffer() method returning a CL
|
|
|
|
/// buffer object of destination data of Dv
|
|
|
|
///
|
|
|
|
/// @param dvDesc vertex buffer descriptor for the dvBuffer
|
|
|
|
///
|
|
|
|
/// @param numPatchCoords number of patchCoords.
|
|
|
|
///
|
|
|
|
/// @param patchCoords array of locations to be evaluated.
|
|
|
|
///
|
|
|
|
/// @param patchTable CLPatchTable or equivalent
|
|
|
|
///
|
|
|
|
template <typename SRC_BUFFER, typename DST_BUFFER,
|
|
|
|
typename PATCHCOORD_BUFFER, typename PATCH_TABLE>
|
|
|
|
bool EvalPatches(
|
|
|
|
SRC_BUFFER *srcBuffer, VertexBufferDescriptor const &srcDesc,
|
|
|
|
DST_BUFFER *dstBuffer, VertexBufferDescriptor const &dstDesc,
|
|
|
|
DST_BUFFER *duBuffer, VertexBufferDescriptor const &duDesc,
|
|
|
|
DST_BUFFER *dvBuffer, VertexBufferDescriptor const &dvDesc,
|
|
|
|
int numPatchCoords,
|
|
|
|
PATCHCOORD_BUFFER *patchCoords,
|
|
|
|
PATCH_TABLE *patchTable) const {
|
|
|
|
|
|
|
|
return EvalPatches(srcBuffer->BindCLBuffer(_clCommandQueue), srcDesc,
|
|
|
|
dstBuffer->BindCLBuffer(_clCommandQueue), dstDesc,
|
|
|
|
duBuffer->BindCLBuffer(_clCommandQueue), duDesc,
|
|
|
|
dvBuffer->BindCLBuffer(_clCommandQueue), dvDesc,
|
|
|
|
numPatchCoords,
|
|
|
|
patchCoords->BindCLBuffer(_clCommandQueue),
|
|
|
|
patchTable->GetPatchArrayBuffer(),
|
|
|
|
patchTable->GetPatchIndexBuffer(),
|
|
|
|
patchTable->GetPatchParamBuffer());
|
|
|
|
}
|
|
|
|
|
|
|
|
bool EvalPatches(cl_mem src, VertexBufferDescriptor const &srcDesc,
|
|
|
|
cl_mem dst, VertexBufferDescriptor const &dstDesc,
|
|
|
|
cl_mem du, VertexBufferDescriptor const &duDesc,
|
|
|
|
cl_mem dv, VertexBufferDescriptor const &dvDesc,
|
|
|
|
int numPatchCoords,
|
|
|
|
cl_mem patchCoordsBuffer,
|
|
|
|
cl_mem patchArrayBuffer,
|
|
|
|
cl_mem patchIndexBuffer,
|
|
|
|
cl_mem patchParamsBuffer) const;
|
|
|
|
|
|
|
|
/// ----------------------------------------------------------------------
|
|
|
|
///
|
|
|
|
/// Other methods
|
|
|
|
///
|
|
|
|
/// ----------------------------------------------------------------------
|
|
|
|
|
2015-05-09 00:31:26 +00:00
|
|
|
/// Configure OpenCL kernel.
|
|
|
|
/// Returns false if it fails to compile the kernel.
|
|
|
|
bool Compile(VertexBufferDescriptor const &srcDesc,
|
2015-05-28 00:23:36 +00:00
|
|
|
VertexBufferDescriptor const &dstDesc,
|
|
|
|
VertexBufferDescriptor const &duDesc,
|
|
|
|
VertexBufferDescriptor const &dvDesc);
|
2015-05-09 00:31:26 +00:00
|
|
|
|
|
|
|
/// Wait the OpenCL kernels finish.
|
|
|
|
template <typename DEVICE_CONTEXT>
|
|
|
|
static void Synchronize(DEVICE_CONTEXT deviceContext) {
|
|
|
|
Synchronize(deviceContext->GetCommandQueue());
|
|
|
|
}
|
|
|
|
|
|
|
|
static void Synchronize(cl_command_queue queue);
|
|
|
|
|
|
|
|
private:
|
|
|
|
cl_context _clContext;
|
|
|
|
cl_command_queue _clCommandQueue;
|
|
|
|
cl_program _program;
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
cl_kernel _stencilKernel;
|
2015-05-28 00:23:36 +00:00
|
|
|
cl_kernel _stencilDerivKernel;
|
Osd API refactor: EvalStencils and EvalPatches
Add EvalStencils and EvalPatches API for most of CPU and GPU evaluators.
with this change, Eval API in the osd layer consists of following parts:
- Evaluators (Cpu, Omp, Tbb, Cuda, CL, GLXFB, GLCompute, D3D11Compute)
implements EvalStencils and EvalPatches(*). Both supports derivatives
(not fully implemented though)
- Interop vertex buffer classes (optional, same as before)
Note that these classes are not necessary to use Evaluators.
All evaluators have EvalStencils/Patches which take device-specific
buffer objects. For example, GLXFBEvaluator can take GLuint directly
for both stencil tables and input primvars. Although using these
interop classes makes it easy to integrate osd into relatively
simple applications.
- device-dependent StencilTable and PatchTable (optional)
These are also optional, but can be used simply a substitute of
Far::StencilTable and Far::PatchTable for osd evaluators.
- PatchArray, PatchCoord, PatchParam
They are tiny structs used for GPU based patch evaluation.
(*) TODO and known issues:
- CLEvaluator and D3D11Evaluator's EvalPatches() have not been implemented.
- GPU Gregory patch evaluation has not been implemented in EvalPatches().
- CudaEvaluator::EvalPatches() is very unstable.
- All patch evaluation kernels have not been well optimized.
- Currently GLXFB kernel doesn't support derivative evaluation.
There's a technical difficulty for the multi-stream output.
2015-05-26 04:51:55 +00:00
|
|
|
cl_kernel _patchKernel;
|
2015-05-09 00:31:26 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
} // end namespace Osd
|
|
|
|
|
|
|
|
} // end namespace OPENSUBDIV_VERSION
|
|
|
|
using namespace OPENSUBDIV_VERSION;
|
|
|
|
|
|
|
|
} // end namespace OpenSubdiv
|
|
|
|
|
|
|
|
|
2015-05-19 18:22:37 +00:00
|
|
|
#endif // OPENSUBDIV_OPENSUBDIV3_OSD_CL_EVALUATOR_H
|