mirror of
https://gitlab.kitware.com/vtk/vtk-m
synced 2024-09-19 18:45:43 +00:00
8d62bf1286
1. Use cudaPerThreadStream instead of the default streams 2. Since there have been changes to ArrayHandle code, the API to create ArrayHandle from a device pointer has changed.
204 lines
6.0 KiB
C++
204 lines
6.0 KiB
C++
//============================================================================
|
|
// Copyright (c) Kitware, Inc.
|
|
// All rights reserved.
|
|
// See LICENSE.txt for details.
|
|
//
|
|
// This software is distributed WITHOUT ANY WARRANTY; without even
|
|
// the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
|
|
// PURPOSE. See the above copyright notice for more information.
|
|
//============================================================================
|
|
#ifndef vtkm_interop_cuda_internal_TransferToOpenGL_h
|
|
#define vtkm_interop_cuda_internal_TransferToOpenGL_h
|
|
|
|
#include <vtkm/cont/ErrorBadAllocation.h>
|
|
#include <vtkm/cont/ErrorExecution.h>
|
|
|
|
#include <vtkm/cont/cuda/internal/DeviceAdapterTagCuda.h>
|
|
|
|
#include <vtkm/interop/internal/TransferToOpenGL.h>
|
|
|
|
// Disable warnings we check vtkm for but Thrust does not.
|
|
#include <vtkm/exec/cuda/internal/ThrustPatches.h>
|
|
VTKM_THIRDPARTY_PRE_INCLUDE
|
|
#include <thrust/copy.h>
|
|
#include <thrust/device_ptr.h>
|
|
#include <thrust/system/cuda/execution_policy.h>
|
|
#include <vtkm/exec/cuda/internal/ExecutionPolicy.h>
|
|
VTKM_THIRDPARTY_POST_INCLUDE
|
|
|
|
namespace vtkm
|
|
{
|
|
namespace interop
|
|
{
|
|
namespace internal
|
|
{
|
|
|
|
/// \brief cuda backend and opengl interop resource management
|
|
///
|
|
/// \c TransferResource manages cuda resource binding for a given buffer
|
|
///
|
|
///
|
|
class CudaTransferResource : public vtkm::interop::internal::TransferResource
|
|
{
|
|
public:
|
|
CudaTransferResource()
|
|
: vtkm::interop::internal::TransferResource()
|
|
{
|
|
this->Registered = false;
|
|
}
|
|
|
|
~CudaTransferResource()
|
|
{
|
|
//unregister the buffer
|
|
if (this->Registered)
|
|
{
|
|
cudaGraphicsUnregisterResource(this->CudaResource);
|
|
}
|
|
}
|
|
|
|
bool IsRegistered() const { return Registered; }
|
|
|
|
void Register(GLuint* handle)
|
|
{
|
|
if (this->Registered)
|
|
{
|
|
//If you don't release the cuda resource before re-registering you
|
|
//will leak memory on the OpenGL side.
|
|
cudaGraphicsUnregisterResource(this->CudaResource);
|
|
}
|
|
|
|
this->Registered = true;
|
|
cudaError_t cError =
|
|
cudaGraphicsGLRegisterBuffer(&this->CudaResource, *handle, cudaGraphicsMapFlagsWriteDiscard);
|
|
if (cError != cudaSuccess)
|
|
{
|
|
throw vtkm::cont::ErrorExecution("Could not register the OpenGL buffer handle to CUDA.");
|
|
}
|
|
}
|
|
|
|
void Map()
|
|
{
|
|
//map the resource into cuda, so we can copy it
|
|
cudaError_t cError = cudaGraphicsMapResources(1, &this->CudaResource, cudaStreamPerThread);
|
|
if (cError != cudaSuccess)
|
|
{
|
|
throw vtkm::cont::ErrorBadAllocation(
|
|
"Could not allocate enough memory in CUDA for OpenGL interop.");
|
|
}
|
|
}
|
|
|
|
template <typename ValueType>
|
|
ValueType* GetMappedPoiner(vtkm::Int64 desiredSize)
|
|
{
|
|
//get the mapped pointer
|
|
std::size_t cuda_size;
|
|
ValueType* pointer = nullptr;
|
|
cudaError_t cError =
|
|
cudaGraphicsResourceGetMappedPointer((void**)&pointer, &cuda_size, this->CudaResource);
|
|
|
|
if (cError != cudaSuccess)
|
|
{
|
|
throw vtkm::cont::ErrorExecution("Unable to get pointers to CUDA memory for OpenGL buffer.");
|
|
}
|
|
|
|
//assert that cuda_size is the same size as the buffer we created in OpenGL
|
|
VTKM_ASSERT(cuda_size >= static_cast<std::size_t>(desiredSize));
|
|
return pointer;
|
|
}
|
|
|
|
void UnMap() { cudaGraphicsUnmapResources(1, &this->CudaResource, cudaStreamPerThread); }
|
|
|
|
private:
|
|
bool Registered;
|
|
cudaGraphicsResource_t CudaResource;
|
|
};
|
|
|
|
/// \brief Manages transferring an ArrayHandle to opengl .
|
|
///
|
|
/// \c TransferToOpenGL manages to transfer the contents of an ArrayHandle
|
|
/// to OpenGL as efficiently as possible.
|
|
///
|
|
template <typename ValueType>
|
|
class TransferToOpenGL<ValueType, vtkm::cont::DeviceAdapterTagCuda>
|
|
{
|
|
using DeviceAdapterTag = vtkm::cont::DeviceAdapterTagCuda;
|
|
|
|
public:
|
|
VTKM_CONT explicit TransferToOpenGL(BufferState& state)
|
|
: State(state)
|
|
, Resource(nullptr)
|
|
{
|
|
if (!this->State.HasType())
|
|
{
|
|
this->State.DeduceAndSetType(ValueType());
|
|
}
|
|
|
|
this->Resource =
|
|
dynamic_cast<vtkm::interop::internal::CudaTransferResource*>(state.GetResource());
|
|
if (!this->Resource)
|
|
{
|
|
vtkm::interop::internal::CudaTransferResource* cudaResource =
|
|
new vtkm::interop::internal::CudaTransferResource();
|
|
|
|
//reset the resource to be a cuda resource
|
|
this->State.SetResource(cudaResource);
|
|
this->Resource = cudaResource;
|
|
}
|
|
}
|
|
|
|
template <typename StorageTag>
|
|
VTKM_CONT void Transfer(const vtkm::cont::ArrayHandle<ValueType, StorageTag>& handle) const
|
|
{
|
|
//make a buffer for the handle if the user has forgotten too
|
|
if (!glIsBuffer(*this->State.GetHandle()))
|
|
{
|
|
glGenBuffers(1, this->State.GetHandle());
|
|
}
|
|
|
|
//bind the buffer to the given buffer type
|
|
glBindBuffer(this->State.GetType(), *this->State.GetHandle());
|
|
|
|
//Determine if we need to reallocate the buffer
|
|
const vtkm::Int64 size =
|
|
static_cast<vtkm::Int64>(sizeof(ValueType)) * handle.GetNumberOfValues();
|
|
this->State.SetSize(size);
|
|
const bool resize = this->State.ShouldRealloc(size);
|
|
if (resize)
|
|
{
|
|
//Allocate the memory and set it as GL_DYNAMIC_DRAW draw
|
|
glBufferData(this->State.GetType(), static_cast<GLsizeiptr>(size), 0, GL_DYNAMIC_DRAW);
|
|
|
|
this->State.SetCapacity(size);
|
|
}
|
|
|
|
if (!this->Resource->IsRegistered() || resize)
|
|
{
|
|
//register the buffer as being used by cuda. This needs to be done everytime
|
|
//we change the size of the buffer. That is why we only change the buffer
|
|
//size as infrequently as possible
|
|
this->Resource->Register(this->State.GetHandle());
|
|
}
|
|
|
|
this->Resource->Map();
|
|
|
|
ValueType* beginPointer = this->Resource->GetMappedPoiner<ValueType>(size);
|
|
vtkm::cont::ArrayHandleBasic<ValueType> deviceMemory(
|
|
beginPointer, handle.GetNumberOfValues(), DeviceAdapterTag{}, [](void*) {});
|
|
|
|
//Do a device to device memory copy
|
|
vtkm::cont::DeviceAdapterAlgorithm<DeviceAdapterTag>::Copy(handle, deviceMemory);
|
|
|
|
//unmap the resource
|
|
this->Resource->UnMap();
|
|
}
|
|
|
|
private:
|
|
vtkm::interop::BufferState& State;
|
|
vtkm::interop::internal::CudaTransferResource* Resource;
|
|
};
|
|
}
|
|
}
|
|
} //namespace vtkm::interop::cuda::internal
|
|
|
|
#endif
|