OpenSubdiv/opensubdiv/osd/d3d11ComputeEvaluator.cpp

383 lines
13 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/d3d11ComputeEvaluator.h"
#include <cassert>
#include <sstream>
#include <string>
#include <vector>
#define INITGUID // for IID_ID3D11ShaderReflection
#include <D3D11.h>
#include <D3D11shader.h>
#include <D3Dcompiler.h>
#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 {
#define SAFE_RELEASE(p) { if(p) { (p)->Release(); (p)=NULL; } }
static const char *shaderSource =
#include "../osd/hlslComputeKernel.gen.h"
;
// ----------------------------------------------------------------------------
// must match constant buffer declaration in hlslComputeKernel.hlsl
__declspec(align(16))
struct KernelUniformArgs {
int start; // batch
int end;
int srcOffset;
int dstOffset;
};
// ----------------------------------------------------------------------------
template <typename T>
static ID3D11Buffer *createBuffer(std::vector<T> const &src,
ID3D11Device *device) {
size_t size = src.size()*sizeof(T);
ID3D11Buffer *buffer = NULL;
D3D11_BUFFER_DESC bd;
bd.ByteWidth = (unsigned int)size;
bd.Usage = D3D11_USAGE_IMMUTABLE;
bd.BindFlags = D3D11_BIND_SHADER_RESOURCE;
bd.CPUAccessFlags = 0;
bd.MiscFlags = 0;
bd.StructureByteStride = 0;
D3D11_SUBRESOURCE_DATA initData;
initData.pSysMem = &src.at(0);
HRESULT hr = device->CreateBuffer(&bd, &initData, &buffer);
if (FAILED(hr)) {
Far::Error(Far::FAR_RUNTIME_ERROR,
"Error creating compute table buffer\n");
return NULL;
}
return buffer;
}
static ID3D11ShaderResourceView *createSRV(ID3D11Buffer *buffer,
DXGI_FORMAT format,
ID3D11Device *device,
size_t size) {
ID3D11ShaderResourceView *srv = NULL;
D3D11_SHADER_RESOURCE_VIEW_DESC srvd;
ZeroMemory(&srvd, sizeof(srvd));
srvd.Format = format;
srvd.ViewDimension = D3D11_SRV_DIMENSION_BUFFER;
srvd.Buffer.FirstElement = 0;
srvd.Buffer.NumElements = (unsigned int)size;
HRESULT hr = device->CreateShaderResourceView(buffer, &srvd, &srv);
if (FAILED(hr)) {
Far::Error(Far::FAR_RUNTIME_ERROR,
"Error creating compute table shader resource view\n");
return NULL;
}
return srv;
}
D3D11StencilTable::D3D11StencilTable(Far::StencilTable const *stencilTable,
ID3D11DeviceContext *deviceContext)
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
{
ID3D11Device *device = NULL;
deviceContext->GetDevice(&device);
assert(device);
_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) {
std::vector<int> const &sizes = stencilTable->GetSizes();
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
_sizesBuffer = createBuffer(sizes, device);
_offsetsBuffer = createBuffer(stencilTable->GetOffsets(), device);
_indicesBuffer = createBuffer(stencilTable->GetControlIndices(), device);
_weightsBuffer = createBuffer(stencilTable->GetWeights(), device);
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
_sizes = createSRV(_sizesBuffer, DXGI_FORMAT_R32_SINT, device,
stencilTable->GetSizes().size());
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
_offsets = createSRV(_offsetsBuffer, DXGI_FORMAT_R32_SINT, device,
stencilTable->GetOffsets().size());
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
_indices = createSRV(_indicesBuffer, DXGI_FORMAT_R32_SINT, device,
stencilTable->GetControlIndices().size());
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
_weights= createSRV(_weightsBuffer, DXGI_FORMAT_R32_FLOAT, device,
stencilTable->GetWeights().size());
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;
_sizesBuffer = _offsetsBuffer = _indicesBuffer = _weightsBuffer = NULL;
}
}
D3D11StencilTable::~D3D11StencilTable() {
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
SAFE_RELEASE(_sizes);
SAFE_RELEASE(_sizesBuffer);
SAFE_RELEASE(_offsets);
SAFE_RELEASE(_offsetsBuffer);
SAFE_RELEASE(_indices);
SAFE_RELEASE(_indicesBuffer);
SAFE_RELEASE(_weights);
SAFE_RELEASE(_weightsBuffer);
}
// ---------------------------------------------------------------------------
D3D11ComputeEvaluator::D3D11ComputeEvaluator() :
_computeShader(NULL),
_classLinkage(NULL),
_singleBufferKernel(NULL),
_separateBufferKernel(NULL),
_uniformArgs(NULL),
_workGroupSize(64) {
}
D3D11ComputeEvaluator *
D3D11ComputeEvaluator::Create(BufferDescriptor const &srcDesc,
BufferDescriptor const &dstDesc,
BufferDescriptor const &duDesc,
BufferDescriptor const &dvDesc,
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
ID3D11DeviceContext *deviceContext) {
(void)deviceContext; // not used
// TODO: implements derivatives
(void)duDesc;
(void)dvDesc;
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
D3D11ComputeEvaluator *instance = new D3D11ComputeEvaluator();
if (instance->Compile(srcDesc, dstDesc, deviceContext)) return instance;
delete instance;
return NULL;
}
D3D11ComputeEvaluator::~D3D11ComputeEvaluator() {
SAFE_RELEASE(_computeShader);
SAFE_RELEASE(_classLinkage);
SAFE_RELEASE(_singleBufferKernel);
SAFE_RELEASE(_separateBufferKernel);
SAFE_RELEASE(_uniformArgs);
}
bool
D3D11ComputeEvaluator::Compile(BufferDescriptor const &srcDesc,
BufferDescriptor const &dstDesc,
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
ID3D11DeviceContext *deviceContext) {
if (srcDesc.length > dstDesc.length) {
Far::Error(Far::FAR_RUNTIME_ERROR,
"srcDesc length must be less than or equal to "
"dstDesc length.\n");
return false;
}
DWORD dwShaderFlags = D3DCOMPILE_ENABLE_STRICTNESS;
#if defined(D3D10_SHADER_RESOURCES_MAY_ALIAS)
dwShaderFlags |= D3D10_SHADER_RESOURCES_MAY_ALIAS;
#endif
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
#ifdef _DEBUG
dwShaderFlags |= D3DCOMPILE_DEBUG;
#endif
std::ostringstream ss;
ss << srcDesc.length; std::string lengthValue(ss.str()); ss.str("");
ss << srcDesc.stride; std::string srcStrideValue(ss.str()); ss.str("");
ss << dstDesc.stride; std::string dstStrideValue(ss.str()); ss.str("");
ss << _workGroupSize; std::string workgroupSizeValue(ss.str()); ss.str("");
D3D_SHADER_MACRO defines[] =
{ "LENGTH", lengthValue.c_str(),
"SRC_STRIDE", srcStrideValue.c_str(),
"DST_STRIDE", dstStrideValue.c_str(),
"WORK_GROUP_SIZE", workgroupSizeValue.c_str(),
0, 0 };
ID3DBlob * computeShaderBuffer = NULL;
ID3DBlob * errorBuffer = NULL;
HRESULT hr = D3DCompile(shaderSource, strlen(shaderSource),
NULL, &defines[0], NULL,
"cs_main", "cs_5_0",
dwShaderFlags, 0,
&computeShaderBuffer, &errorBuffer);
if (FAILED(hr)) {
if (errorBuffer != NULL) {
Far::Error(Far::FAR_RUNTIME_ERROR,
"Error compiling HLSL shader: %s\n",
(CHAR*)errorBuffer->GetBufferPointer());
errorBuffer->Release();
return false;
}
}
ID3D11Device *device = NULL;
deviceContext->GetDevice(&device);
assert(device);
device->CreateClassLinkage(&_classLinkage);
assert(_classLinkage);
device->CreateComputeShader(computeShaderBuffer->GetBufferPointer(),
computeShaderBuffer->GetBufferSize(),
_classLinkage,
&_computeShader);
assert(_computeShader);
ID3D11ShaderReflection *reflector;
D3DReflect(computeShaderBuffer->GetBufferPointer(),
computeShaderBuffer->GetBufferSize(),
IID_ID3D11ShaderReflection, (void**) &reflector);
assert(reflector);
assert(reflector->GetNumInterfaceSlots() == 1);
reflector->Release();
computeShaderBuffer->Release();
_classLinkage->GetClassInstance("singleBufferCompute", 0, &_singleBufferKernel);
assert(_singleBufferKernel);
_classLinkage->GetClassInstance("separateBufferCompute", 0, &_separateBufferKernel);
assert(_separateBufferKernel);
D3D11_BUFFER_DESC cbDesc;
ZeroMemory(&cbDesc, sizeof(cbDesc));
cbDesc.Usage = D3D11_USAGE_DYNAMIC;
cbDesc.BindFlags = D3D11_BIND_CONSTANT_BUFFER;
cbDesc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
cbDesc.MiscFlags = 0;
cbDesc.ByteWidth = sizeof(KernelUniformArgs);
device->CreateBuffer(&cbDesc, NULL, &_uniformArgs);
return true;
}
/* static */
void
D3D11ComputeEvaluator::Synchronize(ID3D11DeviceContext *deviceContext) {
// XXX: this is currently just for the performance measuring purpose.
// XXXFIXME!
ID3D11Query *query = NULL;
ID3D11Device *device = NULL;
deviceContext->GetDevice(&device);
assert(device);
D3D11_QUERY_DESC desc;
desc.Query = D3D11_QUERY_EVENT;
desc.MiscFlags = 0;
device->CreateQuery(&desc, &query);
deviceContext->Flush();
deviceContext->End(query);
while (S_OK != deviceContext->GetData(query, NULL, 0, 0));
SAFE_RELEASE(query);
}
bool
D3D11ComputeEvaluator::EvalStencils(ID3D11UnorderedAccessView *srcUAV,
BufferDescriptor const &srcDesc,
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
ID3D11UnorderedAccessView *dstUAV,
BufferDescriptor const &dstDesc,
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
ID3D11ShaderResourceView *sizesSRV,
ID3D11ShaderResourceView *offsetsSRV,
ID3D11ShaderResourceView *indicesSRV,
ID3D11ShaderResourceView *weightsSRV,
int start,
int end,
ID3D11DeviceContext *deviceContext) const {
assert(deviceContext);
int count = end - start;
if (count <= 0) return true;
KernelUniformArgs args;
args.start = start;
args.end = end;
args.srcOffset = srcDesc.offset;
args.dstOffset = dstDesc.offset;
D3D11_MAPPED_SUBRESOURCE mappedResource;
deviceContext->Map(_uniformArgs, 0, D3D11_MAP_WRITE_DISCARD, 0, &mappedResource);
CopyMemory(mappedResource.pData, &args, sizeof(KernelUniformArgs));
deviceContext->Unmap(_uniformArgs, 0);
deviceContext->CSSetConstantBuffers(0, 1, &_uniformArgs); // b0
// Unbind the vertexBuffer from the input assembler
ID3D11Buffer *NULLBuffer = 0;
UINT voffset = 0, vstride = 0;
deviceContext->IASetVertexBuffers(0, 1, &NULLBuffer, &voffset, &vstride);
ID3D11ShaderResourceView *NULLSRV = 0;
deviceContext->VSSetShaderResources(0, 1, &NULLSRV);
// bind UAV
ID3D11UnorderedAccessView *UAViews[] = { srcUAV, dstUAV };
ID3D11ShaderResourceView *SRViews[] = {
sizesSRV, offsetsSRV, indicesSRV, weightsSRV };
// bind source vertex and stencil table
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
deviceContext->CSSetShaderResources(1, 4, SRViews); // t1-t4
if (srcUAV == dstUAV) {
deviceContext->CSSetUnorderedAccessViews(0, 1, UAViews, 0); // u0
// Dispatch src == dst buffer
deviceContext->CSSetShader(_computeShader, &_singleBufferKernel, 1);
deviceContext->Dispatch((count + _workGroupSize - 1) / _workGroupSize, 1, 1);
} else {
deviceContext->CSSetUnorderedAccessViews(0, 2, UAViews, 0); // u0, u1
// Dispatch src != dst buffer
deviceContext->CSSetShader(_computeShader, &_separateBufferKernel, 1);
deviceContext->Dispatch((count + _workGroupSize - 1) / _workGroupSize, 1, 1);
}
// unbind stencil table and vertexbuffers
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
SRViews[0] = SRViews[1] = SRViews[2] = SRViews[3] = NULL;
deviceContext->CSSetShaderResources(1, 4, SRViews);
UAViews[0] = UAViews[1] = NULL;
deviceContext->CSSetUnorderedAccessViews(0, 2, UAViews, 0);
return true;
}
} // end namespace Osd
} // end namespace OPENSUBDIV_VERSION
} // end namespace OpenSubdiv