OpenSubdiv/opensubdiv/osd/cudaEvaluator.cpp

386 lines
14 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/cudaEvaluator.h"
#include <cuda_runtime.h>
#include <vector>
#include "../far/stencilTable.h"
#include "../osd/types.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
extern "C" {
void CudaEvalStencils(const float *src,
float *dst,
int length,
int srcStride,
int dstStride,
const int * sizes,
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
const int * offsets,
const int * indices,
const float * weights,
int start,
int end);
void CudaEvalPatches(
const float *src, float *dst,
int length, int srcStride, int dstStride,
int numPatchCoords,
const void *patchCoords,
const void *patchArrays,
const int *patchIndices,
const void *patchParams);
void CudaEvalPatchesWithDerivatives(
const float *src, float *dst,
float *du, float *dv,
float *duu, float *duv, float *dvv,
int length, int srcStride, int dstStride,
int duStride, int dvStride,
int duuStride, int duvStride, int dvvStride,
int numPatchCoords,
const void *patchCoords,
const void *patchArrays,
const int *patchIndices,
const void *patchParams);
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 {
template <class T> void *
createCudaBuffer(std::vector<T> const & src) {
if (src.empty()) {
return 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
void * devicePtr = 0;
size_t size = src.size()*sizeof(T);
cudaError_t err = cudaMalloc(&devicePtr, size);
if (err != cudaSuccess) {
return devicePtr;
}
err = cudaMemcpy(devicePtr, &src.at(0), size, cudaMemcpyHostToDevice);
if (err != cudaSuccess) {
cudaFree(devicePtr);
return 0;
}
return devicePtr;
}
// ----------------------------------------------------------------------------
CudaStencilTable::CudaStencilTable(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 = createCudaBuffer(stencilTable->GetSizes());
_offsets = createCudaBuffer(stencilTable->GetOffsets());
_indices = createCudaBuffer(stencilTable->GetControlIndices());
_weights = createCudaBuffer(stencilTable->GetWeights());
_duWeights = _dvWeights = NULL;
_duuWeights = _duvWeights = _dvvWeights = 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
} else {
_sizes = _offsets = _indices = _weights = NULL;
_duWeights = _dvWeights = NULL;
_duuWeights = _duvWeights = _dvvWeights = NULL;
}
}
CudaStencilTable::CudaStencilTable(Far::LimitStencilTable const *limitStencilTable) {
_numStencils = limitStencilTable->GetNumStencils();
if (_numStencils > 0) {
_sizes = createCudaBuffer(limitStencilTable->GetSizes());
_offsets = createCudaBuffer(limitStencilTable->GetOffsets());
_indices = createCudaBuffer(limitStencilTable->GetControlIndices());
_weights = createCudaBuffer(limitStencilTable->GetWeights());
_duWeights = createCudaBuffer(limitStencilTable->GetDuWeights());
_dvWeights = createCudaBuffer(limitStencilTable->GetDvWeights());
_duuWeights = createCudaBuffer(limitStencilTable->GetDuuWeights());
_duvWeights = createCudaBuffer(limitStencilTable->GetDuvWeights());
_dvvWeights = createCudaBuffer(limitStencilTable->GetDvvWeights());
} else {
_sizes = _offsets = _indices = _weights = NULL;
_duWeights = _dvWeights = NULL;
_duuWeights = _duvWeights = _dvvWeights = 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
}
}
CudaStencilTable::~CudaStencilTable() {
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) cudaFree(_sizes);
if (_offsets) cudaFree(_offsets);
if (_indices) cudaFree(_indices);
if (_weights) cudaFree(_weights);
if (_duWeights) cudaFree(_duWeights);
if (_dvWeights) cudaFree(_dvWeights);
if (_duuWeights) cudaFree(_duuWeights);
if (_duvWeights) cudaFree(_duvWeights);
if (_dvvWeights) cudaFree(_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
}
// ---------------------------------------------------------------------------
/* static */
bool
CudaEvaluator::EvalStencils(const float *src, BufferDescriptor const &srcDesc,
float *dst, BufferDescriptor const &dstDesc,
const int * sizes,
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
const int * offsets,
const int * indices,
const float * weights,
int start,
int end) {
if (dst == NULL) 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
CudaEvalStencils(src + srcDesc.offset,
dst + dstDesc.offset,
srcDesc.length,
srcDesc.stride,
dstDesc.stride,
sizes, offsets, indices, weights,
start, end);
return true;
}
/* static */
bool
CudaEvaluator::EvalStencils(const float *src, BufferDescriptor const &srcDesc,
float *dst, BufferDescriptor const &dstDesc,
float *du, BufferDescriptor const &duDesc,
float *dv, BufferDescriptor const &dvDesc,
const int * sizes,
const int * offsets,
const int * indices,
const float * weights,
const float * duWeights,
const float * dvWeights,
int start,
int end) {
// PERFORMANCE: need to combine 3 launches together
if (dst) {
CudaEvalStencils(src + srcDesc.offset,
dst + dstDesc.offset,
srcDesc.length,
srcDesc.stride,
dstDesc.stride,
sizes, offsets, indices, weights,
start, end);
}
if (du) {
CudaEvalStencils(src + srcDesc.offset,
du + duDesc.offset,
srcDesc.length,
srcDesc.stride,
duDesc.stride,
sizes, offsets, indices, duWeights,
start, end);
}
if (dv) {
CudaEvalStencils(src + srcDesc.offset,
dv + dvDesc.offset,
srcDesc.length,
srcDesc.stride,
dvDesc.stride,
sizes, offsets, indices, dvWeights,
start, end);
}
return true;
}
/* static */
bool
CudaEvaluator::EvalStencils(const float *src, BufferDescriptor const &srcDesc,
float *dst, BufferDescriptor const &dstDesc,
float *du, BufferDescriptor const &duDesc,
float *dv, BufferDescriptor const &dvDesc,
float *duu, BufferDescriptor const &duuDesc,
float *duv, BufferDescriptor const &duvDesc,
float *dvv, BufferDescriptor const &dvvDesc,
const int * sizes,
const int * offsets,
const int * indices,
const float * weights,
const float * duWeights,
const float * dvWeights,
const float * duuWeights,
const float * duvWeights,
const float * dvvWeights,
int start,
int end) {
// PERFORMANCE: need to combine 3 launches together
if (dst) {
CudaEvalStencils(src + srcDesc.offset,
dst + dstDesc.offset,
srcDesc.length,
srcDesc.stride,
dstDesc.stride,
sizes, offsets, indices, weights,
start, end);
}
if (du) {
CudaEvalStencils(src + srcDesc.offset,
du + duDesc.offset,
srcDesc.length,
srcDesc.stride,
duDesc.stride,
sizes, offsets, indices, duWeights,
start, end);
}
if (dv) {
CudaEvalStencils(src + srcDesc.offset,
dv + dvDesc.offset,
srcDesc.length,
srcDesc.stride,
dvDesc.stride,
sizes, offsets, indices, dvWeights,
start, end);
}
if (duu) {
CudaEvalStencils(src + srcDesc.offset,
duu + duuDesc.offset,
srcDesc.length,
srcDesc.stride,
duuDesc.stride,
sizes, offsets, indices, duuWeights,
start, end);
}
if (duv) {
CudaEvalStencils(src + srcDesc.offset,
duv + duvDesc.offset,
srcDesc.length,
srcDesc.stride,
duvDesc.stride,
sizes, offsets, indices, duvWeights,
start, end);
}
if (dvv) {
CudaEvalStencils(src + srcDesc.offset,
dvv + dvvDesc.offset,
srcDesc.length,
srcDesc.stride,
dvvDesc.stride,
sizes, offsets, indices, dvvWeights,
start, end);
}
return true;
}
/* static */
bool
CudaEvaluator::EvalPatches(const float *src,
BufferDescriptor const &srcDesc,
float *dst,
BufferDescriptor const &dstDesc,
int numPatchCoords,
const PatchCoord *patchCoords,
const PatchArray *patchArrays,
const int *patchIndices,
const PatchParam *patchParams) {
if (src) src += srcDesc.offset;
if (dst) dst += dstDesc.offset;
CudaEvalPatches(src, dst,
srcDesc.length, srcDesc.stride, dstDesc.stride,
numPatchCoords, patchCoords, patchArrays, patchIndices, patchParams);
return true;
}
/* static */
bool
CudaEvaluator::EvalPatches(
const float *src, BufferDescriptor const &srcDesc,
float *dst, BufferDescriptor const &dstDesc,
float *du, BufferDescriptor const &duDesc,
float *dv, BufferDescriptor const &dvDesc,
int numPatchCoords,
const PatchCoord *patchCoords,
const PatchArray *patchArrays,
const int *patchIndices,
const PatchParam *patchParams) {
if (src) src += srcDesc.offset;
if (dst) dst += dstDesc.offset;
if (du) du += duDesc.offset;
if (dv) dv += dvDesc.offset;
CudaEvalPatchesWithDerivatives(
src, dst, du, dv, NULL, NULL, NULL,
srcDesc.length, srcDesc.stride, dstDesc.stride,
duDesc.stride, dvDesc.stride, 0, 0, 0,
numPatchCoords, patchCoords, patchArrays, patchIndices, patchParams);
return true;
}
/* static */
bool
CudaEvaluator::EvalPatches(
const float *src, BufferDescriptor const &srcDesc,
float *dst, BufferDescriptor const &dstDesc,
float *du, BufferDescriptor const &duDesc,
float *dv, BufferDescriptor const &dvDesc,
float *duu, BufferDescriptor const &duuDesc,
float *duv, BufferDescriptor const &duvDesc,
float *dvv, BufferDescriptor const &dvvDesc,
int numPatchCoords,
const PatchCoord *patchCoords,
const PatchArray *patchArrays,
const int *patchIndices,
const PatchParam *patchParams) {
if (src) src += srcDesc.offset;
if (dst) dst += dstDesc.offset;
if (du) du += duDesc.offset;
if (dv) dv += dvDesc.offset;
if (duu) duu += duuDesc.offset;
if (duv) duv += duvDesc.offset;
if (dvv) dvv += dvvDesc.offset;
CudaEvalPatchesWithDerivatives(
src, dst, du, dv, duu, duv, dvv,
srcDesc.length, srcDesc.stride, dstDesc.stride,
duDesc.stride, dvDesc.stride,
duuDesc.stride, duvDesc.stride, dvvDesc.stride,
numPatchCoords, patchCoords, patchArrays, patchIndices, patchParams);
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
/* static */
void
CudaEvaluator::Synchronize(void * /*deviceContext*/) {
cudaThreadSynchronize();
}
} // end namespace Osd
} // end namespace OPENSUBDIV_VERSION
} // end namespace OpenSubdiv