Gromacs  2021.2
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
Typedefs | Functions
devicebuffer_ocl.h File Reference
#include "gromacs/gpu_utils/device_context.h"
#include "gromacs/gpu_utils/device_stream.h"
#include "gromacs/gpu_utils/devicebuffer_datatype.h"
#include "gromacs/gpu_utils/gpu_utils.h"
#include "gromacs/gpu_utils/gputraits_ocl.h"
#include "gromacs/gpu_utils/oclutils.h"
#include "gromacs/utility/gmxassert.h"
#include "gromacs/utility/stringutil.h"
+ Include dependency graph for devicebuffer_ocl.h:

Description

Implements the DeviceBuffer type and routines for OpenCL. Should only be included directly by the main DeviceBuffer file devicebuffer.h. TODO: the intent is for DeviceBuffer to become a class.

Author
Aleksei Iupinov a.yup.nosp@m.inov.nosp@m.@gmai.nosp@m.l.co.nosp@m.m

Typedefs

using DeviceTexture = void *
 Device texture wrapper.
 

Functions

template<typename ValueType >
void allocateDeviceBuffer (DeviceBuffer< ValueType > *buffer, size_t numValues, const DeviceContext &deviceContext)
 Allocates a device-side buffer. It is currently a caller's responsibility to call it only on not-yet allocated buffers. More...
 
template<typename DeviceBuffer >
void freeDeviceBuffer (DeviceBuffer *buffer)
 Frees a device-side buffer. This does not reset separately stored size/capacity integers, as this is planned to be a destructor of DeviceBuffer as a proper class, and no calls on buffer should be made afterwards. More...
 
template<typename ValueType >
void copyToDeviceBuffer (DeviceBuffer< ValueType > *buffer, const ValueType *hostBuffer, size_t startingOffset, size_t numValues, const DeviceStream &deviceStream, GpuApiCallBehavior transferKind, CommandEvent *timingEvent)
 Performs the host-to-device data copy, synchronous or asynchronously on request. More...
 
template<typename ValueType >
void copyFromDeviceBuffer (ValueType *hostBuffer, DeviceBuffer< ValueType > *buffer, size_t startingOffset, size_t numValues, const DeviceStream &deviceStream, GpuApiCallBehavior transferKind, CommandEvent *timingEvent)
 Performs the device-to-host data copy, synchronous or asynchronously on request. More...
 
template<typename ValueType >
void clearDeviceBufferAsync (DeviceBuffer< ValueType > *buffer, size_t startingOffset, size_t numValues, const DeviceStream &deviceStream)
 Clears the device buffer asynchronously. More...
 
template<typename T >
static bool checkDeviceBuffer (DeviceBuffer< T > buffer, int requiredSize)
 Check the validity of the device buffer. More...
 
template<typename ValueType >
void initParamLookupTable (DeviceBuffer< ValueType > *deviceBuffer, DeviceTexture *, const ValueType *hostBuffer, int numValues, const DeviceContext &deviceContext)
 Create a texture object for an array of type ValueType. More...
 
template<typename ValueType >
void destroyParamLookupTable (DeviceBuffer< ValueType > *deviceBuffer, const DeviceTexture &)
 Release the OpenCL device buffer. More...
 

Function Documentation

template<typename ValueType >
void allocateDeviceBuffer ( DeviceBuffer< ValueType > *  buffer,
size_t  numValues,
const DeviceContext &  deviceContext 
)

Allocates a device-side buffer. It is currently a caller's responsibility to call it only on not-yet allocated buffers.

Template Parameters
ValueTypeRaw value type of the buffer.
Parameters
[in,out]bufferPointer to the device-side buffer.
[in]numValuesNumber of values to accommodate.
[in]deviceContextThe buffer's device context-to-be.
template<typename T >
static bool checkDeviceBuffer ( DeviceBuffer< T >  buffer,
int  requiredSize 
)
static

Check the validity of the device buffer.

Checks if the buffer is not nullptr and if its allocation is big enough.

Parameters
[in]bufferDevice buffer to be checked.
[in]requiredSizeNumber of elements that the buffer will have to accommodate.
Returns
Whether the device buffer can be set.
template<typename ValueType >
void clearDeviceBufferAsync ( DeviceBuffer< ValueType > *  buffer,
size_t  startingOffset,
size_t  numValues,
const DeviceStream deviceStream 
)

Clears the device buffer asynchronously.

Template Parameters
ValueTypeRaw value type of the buffer.
Parameters
[in,out]bufferPointer to the device-side buffer
[in]startingOffsetOffset (in values) at the device-side buffer to start clearing at.
[in]numValuesNumber of values to clear.
[in]deviceStreamGPU stream.
template<typename ValueType >
void copyFromDeviceBuffer ( ValueType *  hostBuffer,
DeviceBuffer< ValueType > *  buffer,
size_t  startingOffset,
size_t  numValues,
const DeviceStream deviceStream,
GpuApiCallBehavior  transferKind,
CommandEvent timingEvent 
)

Performs the device-to-host data copy, synchronous or asynchronously on request.

Note that synchronous copy will not synchronize the stream in case of zero numValues because of the early return.

Template Parameters
ValueTypeRaw value type of the buffer.
Parameters
[in,out]hostBufferPointer to the raw host-side memory, also typed ValueType
[in]bufferPointer to the device-side buffer
[in]startingOffsetOffset (in values) at the device-side buffer to copy from.
[in]numValuesNumber of values to copy.
[in]deviceStreamGPU stream to perform asynchronous copy in.
[in]transferKindCopy type: synchronous or asynchronous.
[out]timingEventA pointer to the H2D copy timing event to be filled in. If the pointer is not null, the event can further be used to queue a wait for this operation or to query profiling information.
template<typename ValueType >
void copyToDeviceBuffer ( DeviceBuffer< ValueType > *  buffer,
const ValueType *  hostBuffer,
size_t  startingOffset,
size_t  numValues,
const DeviceStream deviceStream,
GpuApiCallBehavior  transferKind,
CommandEvent timingEvent 
)

Performs the host-to-device data copy, synchronous or asynchronously on request.

Note that synchronous copy will not synchronize the stream in case of zero numValues because of the early return.

Template Parameters
ValueTypeRaw value type of the buffer.
Parameters
[in,out]bufferPointer to the device-side buffer
[in]hostBufferPointer to the raw host-side memory, also typed ValueType
[in]startingOffsetOffset (in values) at the device-side buffer to copy into.
[in]numValuesNumber of values to copy.
[in]deviceStreamGPU stream to perform asynchronous copy in.
[in]transferKindCopy type: synchronous or asynchronous.
[out]timingEventA pointer to the H2D copy timing event to be filled in. If the pointer is not null, the event can further be used to queue a wait for this operation or to query profiling information.
template<typename ValueType >
void destroyParamLookupTable ( DeviceBuffer< ValueType > *  deviceBuffer,
const DeviceTexture  
)

Release the OpenCL device buffer.

Template Parameters
ValueTypeRaw data type.
Parameters
[in,out]deviceBufferDevice buffer to store data in.
template<typename DeviceBuffer >
void freeDeviceBuffer ( DeviceBuffer buffer)

Frees a device-side buffer. This does not reset separately stored size/capacity integers, as this is planned to be a destructor of DeviceBuffer as a proper class, and no calls on buffer should be made afterwards.

Parameters
[in]bufferPointer to the buffer to free.
template<typename ValueType >
void initParamLookupTable ( DeviceBuffer< ValueType > *  deviceBuffer,
DeviceTexture ,
const ValueType *  hostBuffer,
int  numValues,
const DeviceContext &  deviceContext 
)

Create a texture object for an array of type ValueType.

Creates the device buffer and copies read-only data for an array of type ValueType.

Todo:
Decide if using image2d is most efficient.
Template Parameters
ValueTypeRaw data type.
Parameters
[out]deviceBufferDevice buffer to store data in.
[in]hostBufferHost buffer to get date from.
[in]numValuesNumber of elements in the buffer.
[in]deviceContextGPU device context.