include/cudamanager/PxCudaContextManager.h
File members: include/cudamanager/PxCudaContextManager.h
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
// * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
// * Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution.
// * Neither the name of NVIDIA CORPORATION nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ''AS IS'' AND ANY
// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
//
// Copyright (c) 2008-2023 NVIDIA Corporation. All rights reserved.
#ifndef PX_CUDA_CONTEXT_MANAGER_H
#define PX_CUDA_CONTEXT_MANAGER_H
#include "foundation/PxPreprocessor.h"
#if PX_SUPPORT_GPU_PHYSX
#include "foundation/PxSimpleTypes.h"
#include "foundation/PxErrorCallback.h"
#include "foundation/PxFlags.h"
#include "PxCudaTypes.h"
#if !PX_DOXYGEN
namespace physx
{
#endif
class PxCudaContext;
struct PxCudaInteropRegisterFlag
{
enum Enum
{
eNONE = 0x00,
eREAD_ONLY = 0x01,
eWRITE_DISCARD = 0x02,
eSURFACE_LDST = 0x04,
eTEXTURE_GATHER = 0x08
};
};
class PxDeviceAllocatorCallback
{
public:
virtual bool memAlloc(void** ptr, size_t size) = 0;
virtual bool memFree(void* ptr) = 0;
protected:
virtual ~PxDeviceAllocatorCallback() {}
};
typedef PxFlags<PxCudaInteropRegisterFlag::Enum, uint32_t> PxCudaInteropRegisterFlags;
PX_FLAGS_OPERATORS(PxCudaInteropRegisterFlag::Enum, uint32_t)
class PxCudaContextManagerDesc
{
public:
CUcontext* ctx;
void* graphicsDevice;
const char* appGUID;
PxDeviceAllocatorCallback* deviceAllocator;
PX_INLINE PxCudaContextManagerDesc() :
ctx (NULL),
graphicsDevice (NULL),
appGUID (NULL),
deviceAllocator (NULL)
{
}
};
struct PxKernelIndex
{
PxU32 moduleIndex;
const char* functionName;
};
class PxCudaContextManager
{
public:
template<typename T>
void clearDeviceBufferAsync(T* deviceBuffer, PxU32 numElements, CUstream stream, PxI32 value = 0)
{
clearDeviceBufferAsyncInternal(deviceBuffer, numElements * sizeof(T), stream, value);
}
template<typename T>
void copyDToH(T* hostBuffer, const T* deviceBuffer, PxU32 numElements)
{
copyDToHInternal(hostBuffer, deviceBuffer, numElements * sizeof(T));
}
template<typename T>
void copyHToD(T* deviceBuffer, const T* hostBuffer, PxU32 numElements)
{
copyHToDInternal(deviceBuffer, hostBuffer, numElements * sizeof(T));
}
template<typename T>
void copyDToHAsync(T* hostBuffer, const T* deviceBuffer, PxU32 numElements, CUstream stream)
{
copyDToHAsyncInternal(hostBuffer, deviceBuffer, numElements * sizeof(T), stream);
}
template<typename T>
void copyHToDAsync(T* deviceBuffer, const T* hostBuffer, PxU32 numElements, CUstream stream)
{
copyHToDAsyncInternal(deviceBuffer, hostBuffer, numElements * sizeof(T), stream);
}
template<typename T>
void copyDToDAsync(T* dstDeviceBuffer, const T* srcDeviceBuffer, PxU32 numElements, CUstream stream)
{
copyDToDAsyncInternal(dstDeviceBuffer, srcDeviceBuffer, numElements * sizeof(T), stream);
}
template<typename T>
void memsetAsync(T* dstDeviceBuffer, const T& value, PxU32 numElements, CUstream stream)
{
PX_COMPILE_TIME_ASSERT(sizeof(value) == sizeof(PxU32) || sizeof(value) == sizeof(PxU8));
if (sizeof(value) == sizeof(PxU32))
memsetD32AsyncInternal(dstDeviceBuffer, reinterpret_cast<const PxU32&>(value), numElements, stream);
else
memsetD8AsyncInternal(dstDeviceBuffer, reinterpret_cast<const PxU8&>(value), numElements, stream);
}
template<typename T>
void allocDeviceBuffer(T*& deviceBuffer, PxU32 numElements, const char* filename = __FILE__, PxI32 line = __LINE__)
{
void* ptr = allocDeviceBufferInternal(PxU64(numElements) * sizeof(T), filename, line);
deviceBuffer = reinterpret_cast<T*>(ptr);
}
template<typename T>
T* allocDeviceBuffer(PxU32 numElements, const char* filename = __FILE__, PxI32 line = __LINE__)
{
void* ptr = allocDeviceBufferInternal(PxU64(numElements) * sizeof(T), filename, line);
return reinterpret_cast<T*>(ptr);
}
template<typename T>
void freeDeviceBuffer(T*& deviceBuffer)
{
freeDeviceBufferInternal(deviceBuffer);
deviceBuffer = NULL;
}
template<typename T>
void allocPinnedHostBuffer(T*& pinnedHostBuffer, PxU32 numElements, const char* filename = __FILE__, PxI32 line = __LINE__)
{
void* ptr = allocPinnedHostBufferInternal(PxU64(numElements) * sizeof(T), filename, line);
pinnedHostBuffer = reinterpret_cast<T*>(ptr);
}
template<typename T>
T* allocPinnedHostBuffer(PxU32 numElements, const char* filename = __FILE__, PxI32 line = __LINE__)
{
void* ptr = allocPinnedHostBufferInternal(PxU64(numElements) * sizeof(T), filename, line);
return reinterpret_cast<T*>(ptr);
}
template<typename T>
void freePinnedHostBuffer(T*& pinnedHostBuffer)
{
freePinnedHostBufferInternal(pinnedHostBuffer);
pinnedHostBuffer = NULL;
}
virtual CUdeviceptr getMappedDevicePtr(void* pinnedHostBuffer) = 0;
virtual void acquireContext() = 0;
virtual void releaseContext() = 0;
virtual CUcontext getContext() = 0;
virtual PxCudaContext* getCudaContext() = 0;
virtual bool contextIsValid() const = 0;
/* Query CUDA context and device properties, without acquiring context */
virtual bool supportsArchSM10() const = 0;
virtual bool supportsArchSM11() const = 0;
virtual bool supportsArchSM12() const = 0;
virtual bool supportsArchSM13() const = 0;
virtual bool supportsArchSM20() const = 0;
virtual bool supportsArchSM30() const = 0;
virtual bool supportsArchSM35() const = 0;
virtual bool supportsArchSM50() const = 0;
virtual bool supportsArchSM52() const = 0;
virtual bool supportsArchSM60() const = 0;
virtual bool isIntegrated() const = 0;
virtual bool canMapHostMemory() const = 0;
virtual int getDriverVersion() const = 0;
virtual size_t getDeviceTotalMemBytes() const = 0;
virtual int getMultiprocessorCount() const = 0;
virtual unsigned int getClockRate() const = 0;
virtual int getSharedMemPerBlock() const = 0;
virtual int getSharedMemPerMultiprocessor() const = 0;
virtual unsigned int getMaxThreadsPerBlock() const = 0;
virtual const char *getDeviceName() const = 0;
virtual CUdevice getDevice() const = 0;
virtual void setUsingConcurrentStreams(bool) = 0;
virtual bool getUsingConcurrentStreams() const = 0;
/* End query methods that don't require context to be acquired */
virtual void getDeviceMemoryInfo(size_t& free, size_t& total) const = 0;
virtual int usingDedicatedGPU() const = 0;
virtual CUmodule* getCuModules() = 0;
virtual void release() = 0;
protected:
virtual ~PxCudaContextManager() {}
virtual void* allocDeviceBufferInternal(PxU64 numBytes, const char* filename = NULL, PxI32 line = -1) = 0;
virtual void* allocPinnedHostBufferInternal(PxU64 numBytes, const char* filename = NULL, PxI32 line = -1) = 0;
virtual void freeDeviceBufferInternal(void* deviceBuffer) = 0;
virtual void freePinnedHostBufferInternal(void* pinnedHostBuffer) = 0;
virtual void clearDeviceBufferAsyncInternal(void* deviceBuffer, PxU32 numBytes, CUstream stream, PxI32 value) = 0;
virtual void copyDToHAsyncInternal(void* hostBuffer, const void* deviceBuffer, PxU32 numBytes, CUstream stream) = 0;
virtual void copyHToDAsyncInternal(void* deviceBuffer, const void* hostBuffer, PxU32 numBytes, CUstream stream) = 0;
virtual void copyDToDAsyncInternal(void* dstDeviceBuffer, const void* srcDeviceBuffer, PxU32 numBytes, CUstream stream) = 0;
virtual void copyDToHInternal(void* hostBuffer, const void* deviceBuffer, PxU32 numBytes) = 0;
virtual void copyHToDInternal(void* deviceBuffer, const void* hostBuffer, PxU32 numBytes) = 0;
virtual void memsetD8AsyncInternal(void* dstDeviceBuffer, const PxU8& value, PxU32 numBytes, CUstream stream) = 0;
virtual void memsetD32AsyncInternal(void* dstDeviceBuffer, const PxU32& value, PxU32 numIntegers, CUstream stream) = 0;
};
#define PX_DEVICE_ALLOC(cudaContextManager, deviceBuffer, numElements) cudaContextManager->allocDeviceBuffer(deviceBuffer, numElements, PX_FL)
#define PX_DEVICE_ALLOC_T(T, cudaContextManager, numElements) cudaContextManager->allocDeviceBuffer<T>(numElements, PX_FL)
#define PX_DEVICE_FREE(cudaContextManager, deviceBuffer) cudaContextManager->freeDeviceBuffer(deviceBuffer);
#define PX_PINNED_HOST_ALLOC(cudaContextManager, pinnedHostBuffer, numElements) cudaContextManager->allocPinnedHostBuffer(pinnedHostBuffer, numElements, PX_FL)
#define PX_PINNED_HOST_ALLOC_T(T, cudaContextManager, numElements) cudaContextManager->allocPinnedHostBuffer<T>(numElements, PX_FL)
#define PX_PINNED_HOST_FREE(cudaContextManager, pinnedHostBuffer) cudaContextManager->freePinnedHostBuffer(pinnedHostBuffer);
class PxScopedCudaLock
{
public:
PxScopedCudaLock(PxCudaContextManager& ctx) : mCtx(&ctx)
{
mCtx->acquireContext();
}
~PxScopedCudaLock()
{
mCtx->releaseContext();
}
protected:
PxCudaContextManager* mCtx;
};
#if !PX_DOXYGEN
} // namespace physx
#endif
#endif // PX_SUPPORT_GPU_PHYSX
#endif