pi_cuda.cpp File Reference

Implementation of CUDA Plugin. More...

#include <pi_cuda.hpp>
#include <sycl/detail/cuda_definitions.hpp>
#include <sycl/detail/defines.hpp>
#include <sycl/detail/pi.hpp>
#include <sycl/detail/pi.def>

Include dependency graph for pi_cuda.cpp:

Go to the source code of this file.

Macros
#define	_PI_API(api)    (PluginInit->PiFunctionTable).api = (decltype(&::api))(&api);

Functions
void	enableCUDATracing (cuda_tracing_context_t_ *ctx)
void	disableCUDATracing (cuda_tracing_context_t_ *ctx)
cuda_tracing_context_t_ *	createCUDATracingContext ()
void	freeCUDATracingContext (cuda_tracing_context_t_ *Ctx)
pi_result	piPlatformsGet (pi_uint32 NumEntries, pi_platform *Platforms, pi_uint32 *NumPlatforms)
pi_result	piPlatformGetInfo (pi_platform Platform, pi_platform_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piextPlatformGetNativeHandle (pi_platform Platform, pi_native_handle *NativeHandle)
	Gets the native handle of a PI platform object. More...
pi_result	piextPlatformCreateWithNativeHandle (pi_native_handle NativeHandle, pi_platform *Platform)
	Creates PI platform object from a native handle. More...
pi_result	piPluginGetLastError (char **message)
	API to get Plugin specific warning and error messages. More...
pi_result	piPluginGetBackendOption (pi_platform platform, const char *frontend_option, const char **backend_option)
	API to get backend specific option. More...
pi_result	piDevicesGet (pi_platform Platform, pi_device_type DeviceType, pi_uint32 NumEntries, pi_device *Devices, pi_uint32 *NumDevices)
pi_result	piDeviceRetain (pi_device Device)
pi_result	piDeviceRelease (pi_device Device)
pi_result	piDeviceGetInfo (pi_device Device, pi_device_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
	Returns requested info for provided native device Return PI_DEVICE_INFO_EXTENSION_DEVICELIB_ASSERT for PI_DEVICE_INFO_EXTENSIONS query when the device supports native asserts. More...
pi_result	piDevicePartition (pi_device Device, const pi_device_partition_property *Properties, pi_uint32 NumDevices, pi_device *OutDevices, pi_uint32 *OutNumDevices)
pi_result	piextDeviceSelectBinary (pi_device Device, pi_device_binary *Binaries, pi_uint32 NumBinaries, pi_uint32 *SelectedBinaryInd)
	Selects the most appropriate device binary based on runtime information and the IR characteristics. More...
pi_result	piextDeviceGetNativeHandle (pi_device Device, pi_native_handle *NativeHandle)
	Gets the native handle of a PI device object. More...
pi_result	piextDeviceCreateWithNativeHandle (pi_native_handle NativeHandle, pi_platform Platform, pi_device *Device)
	Creates PI device object from a native handle. More...
pi_result	piContextCreate (const pi_context_properties *Properties, pi_uint32 NumDevices, const pi_device *Devices, void(*PFnNotify)(const char *ErrInfo, const void *PrivateInfo, size_t CB, void *UserData), void *UserData, pi_context *RetContext)
pi_result	piContextGetInfo (pi_context Context, pi_context_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piextContextSetExtendedDeleter (pi_context Context, pi_context_extended_deleter Function, void *UserData)
pi_result	piextContextGetNativeHandle (pi_context Context, pi_native_handle *NativeHandle)
	Gets the native handle of a PI context object. More...
pi_result	piextContextCreateWithNativeHandle (pi_native_handle NativeHandle, pi_uint32 NumDevices, const pi_device *Devices, bool OwnNativeHandle, pi_context *RetContext)
	Creates PI context object from a native handle. More...
pi_result	piContextRetain (pi_context Context)
pi_result	piContextRelease (pi_context Context)
pi_result	piQueueCreate (pi_context Context, pi_device Device, pi_queue_properties Flags, pi_queue *Queue)
pi_result	piextQueueCreate (pi_context Context, pi_device Device, pi_queue_properties *Properties, pi_queue *Queue)
pi_result	piQueueGetInfo (pi_queue Queue, pi_queue_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piQueueRetain (pi_queue Queue)
pi_result	piQueueRelease (pi_queue Queue)
pi_result	piQueueFinish (pi_queue Queue)
pi_result	piQueueFlush (pi_queue Queue)
pi_result	piextQueueGetNativeHandle (pi_queue Queue, pi_native_handle *NativeHandle, int32_t *NativeHandleDesc)
	Gets the native handle of a PI queue object. More...
pi_result	piextQueueCreateWithNativeHandle (pi_native_handle NativeHandle, int32_t NativeHandleDesc, pi_context Context, pi_device Device, bool OwnNativeHandle, pi_queue_properties *Properties, pi_queue *Queue)
	Creates PI queue object from a native handle. More...
pi_result	piMemBufferCreate (pi_context Context, pi_mem_flags Flags, size_t Size, void *HostPtr, pi_mem *RetMem, const pi_mem_properties *properties)
pi_result	piMemGetInfo (pi_mem Mem, pi_mem_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piMemRetain (pi_mem Mem)
pi_result	piMemRelease (pi_mem Mem)
pi_result	piMemImageCreate (pi_context Context, pi_mem_flags Flags, const pi_image_format *ImageFormat, const pi_image_desc *ImageDesc, void *HostPtr, pi_mem *RetImage)
pi_result	piextMemGetNativeHandle (pi_mem Mem, pi_device Dev, pi_native_handle *NativeHandle)
	Gets the native handle of a PI mem object. More...
pi_result	piextMemCreateWithNativeHandle (pi_native_handle NativeHandle, pi_context Context, bool ownNativeHandle, pi_mem *Mem)
	Creates PI mem object from a native handle. More...
pi_result	piProgramCreate (pi_context Context, const void *ILBytes, size_t Length, pi_program *Program)
pi_result	piProgramCreateWithBinary (pi_context Context, pi_uint32 NumDevices, const pi_device *DeviceList, const size_t *Lengths, const unsigned char **Binaries, size_t NumMetadataEntries, const pi_device_binary_property *Metadata, pi_int32 *BinaryStatus, pi_program *Program)
	Creates a PI program for a context and loads the given binary into it. More...
pi_result	piextMemImageCreateWithNativeHandle (pi_native_handle NativeHandle, pi_context Context, bool OwnNativeHandle, const pi_image_format *ImageFormat, const pi_image_desc *ImageDesc, pi_mem *Img)
	Creates PI image object from a native handle. More...
pi_result	piProgramGetInfo (pi_program Program, pi_program_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piProgramLink (pi_context Context, pi_uint32 NumDevices, const pi_device *DeviceList, const char *Options, pi_uint32 NumInputPrograms, const pi_program *InputPrograms, void(*PFnNotify)(pi_program Program, void *UserData), void *UserData, pi_program *RetProgram)
pi_result	piProgramCompile (pi_program Program, pi_uint32 NumDevices, const pi_device *DeviceList, const char *Options, pi_uint32 NumInputHeaders, const pi_program *InputHeaders, const char **HeaderIncludeNames, void(*PFnNotify)(pi_program Program, void *UserData), void *UserData)
pi_result	piProgramBuild (pi_program Program, pi_uint32 NumDevices, const pi_device *DeviceList, const char *Options, void(*PFnNotify)(pi_program Program, void *UserData), void *UserData)
pi_result	piProgramGetBuildInfo (pi_program Program, pi_device Device, pi_program_build_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piProgramRetain (pi_program Program)
pi_result	piProgramRelease (pi_program Program)
pi_result	piextProgramGetNativeHandle (pi_program Program, pi_native_handle *NativeHandle)
	Gets the native handle of a PI program object. More...
pi_result	piextProgramCreateWithNativeHandle (pi_native_handle NativeHandle, pi_context Context, bool OwnNativeHandle, pi_program *Program)
	Creates PI program object from a native handle. More...
pi_result	piKernelCreate (pi_program Program, const char *KernelName, pi_kernel *RetKernel)
pi_result	piKernelSetArg (pi_kernel Kernel, pi_uint32 ArgIndex, size_t ArgSize, const void *ArgValue)
pi_result	piextKernelSetArgMemObj (pi_kernel Kernel, pi_uint32 ArgIndex, const pi_mem_obj_property *ArgProperties, const pi_mem *ArgValue)
pi_result	piextKernelSetArgSampler (pi_kernel Kernel, pi_uint32 ArgIndex, const pi_sampler *ArgValue)
pi_result	piKernelGetInfo (pi_kernel Kernel, pi_kernel_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piextMemImageAllocate (pi_context Context, pi_device Device, pi_image_format *ImageFormat, pi_image_desc *ImageDesc, pi_image_mem_handle *RetMem)
	API to allocate memory for bindless images. More...
pi_result	piextMemUnsampledImageCreate (pi_context Context, pi_device Device, pi_image_mem_handle ImgMem, pi_image_format *ImageFormat, pi_image_desc *ImageDesc, pi_mem *RetMem, pi_image_handle *RetHandle)
	API to create bindless image handles. More...
pi_result	piextMemSampledImageCreate (pi_context Context, pi_device Device, pi_image_mem_handle ImgMem, pi_image_format *ImageFormat, pi_image_desc *ImageDesc, pi_sampler Sampler, pi_mem *RetMem, pi_image_handle *RetHandle)
	API to create sampled bindless image handles. More...
pi_result	piextBindlessImageSamplerCreate (pi_context Context, const pi_sampler_properties *SamplerProperties, float MinMipmapLevelClamp, float MaxMipmapLevelClamp, float MaxAnisotropy, pi_sampler *RetSampler)
	API to create samplers for bindless images. More...
pi_result	piextMemMipmapGetLevel (pi_context Context, pi_device Device, pi_image_mem_handle MipMem, unsigned int Level, pi_image_mem_handle *RetMem)
	API to retrieve individual image from mipmap. More...
pi_result	piextMemImageFree (pi_context Context, pi_device Device, pi_image_mem_handle MemoryHandle)
	API to free memory for bindless images. More...
pi_result	piextMemMipmapFree (pi_context Context, pi_device Device, pi_image_mem_handle MemoryHandle)
	API to free mipmap memory for bindless images. More...
pi_result	piextMemImageCopy (pi_queue Queue, void *DstPtr, void *SrcPtr, const pi_image_format *ImageFormat, const pi_image_desc *ImageDesc, const pi_image_copy_flags Flags, pi_image_offset SrcOffset, pi_image_offset DstOffset, pi_image_region CopyExtent, pi_image_region HostExtent, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
	API to copy image data Host to Device or Device to Host. More...
pi_result	piextMemUnsampledImageHandleDestroy (pi_context Context, pi_device Device, pi_image_handle Handle)
	API to destroy bindless unsampled image handles. More...
pi_result	piextMemSampledImageHandleDestroy (pi_context Context, pi_device Device, pi_image_handle Handle)
	API to destroy bindless sampled image handles. More...
pi_result	piextMemImageGetInfo (pi_image_mem_handle MemHandle, pi_image_info ParamName, void *ParamValue, size_t *ParamValueSizeRet)
	API to query an image memory handle for specific properties. More...
pi_result	piextMemImportOpaqueFD (pi_context Context, pi_device Device, size_t Size, int FileDescriptor, pi_interop_mem_handle *RetHandle)
	API to import external memory in the form of a file descriptor. More...
pi_result	piextMemMapExternalArray (pi_context Context, pi_device Device, pi_image_format *ImageFormat, pi_image_desc *ImageDesc, pi_interop_mem_handle MemHandle, pi_image_mem_handle *RetMem)
	API to map an interop memory handle to an image memory handle. More...
pi_result	piextMemReleaseInterop (pi_context Context, pi_device Device, pi_interop_mem_handle ExtMem)
	API to destroy interop memory. More...
pi_result	piextImportExternalSemaphoreOpaqueFD (pi_context Context, pi_device Device, int FileDescriptor, pi_interop_semaphore_handle *RetHandle)
	API to import an external semaphore in the form of a file descriptor. More...
pi_result	piextDestroyExternalSemaphore (pi_context Context, pi_device Device, pi_interop_semaphore_handle SemHandle)
	API to destroy the external semaphore handle. More...
pi_result	piextWaitExternalSemaphore (pi_queue Queue, pi_interop_semaphore_handle SemHandle, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
	API to instruct the queue with a non-blocking wait on an external semaphore. More...
pi_result	piextSignalExternalSemaphore (pi_queue Queue, pi_interop_semaphore_handle SemHandle, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
	API to instruct the queue to signal the external semaphore handle once all previous commands have completed execution. More...
pi_result	piKernelGetGroupInfo (pi_kernel Kernel, pi_device Device, pi_kernel_group_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piKernelGetSubGroupInfo (pi_kernel Kernel, pi_device Device, pi_kernel_sub_group_info ParamName, size_t InputValueSize, const void *InputValue, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
	API to query information from the sub-group from a kernel. More...
pi_result	piKernelRetain (pi_kernel Kernel)
pi_result	piKernelRelease (pi_kernel Kernel)
pi_result	piEnqueueKernelLaunch (pi_queue Queue, pi_kernel Kernel, pi_uint32 WorkDim, const size_t *GlobalWorkOffset, const size_t *GlobalWorkSize, const size_t *LocalWorkSize, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *OutEvent)
pi_result	piextEnqueueCooperativeKernelLaunch (pi_queue Queue, pi_kernel Kernel, pi_uint32 WorkDim, const size_t *GlobalWorkOffset, const size_t *GlobalWorkSize, const size_t *LocalWorkSize, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *OutEvent)
pi_result	piextKernelCreateWithNativeHandle (pi_native_handle NativeHandle, pi_context Context, pi_program Program, bool OwnNativeHandle, pi_kernel *Kernel)
	Creates PI kernel object from a native handle. More...
pi_result	piextKernelGetNativeHandle (pi_kernel Kernel, pi_native_handle *NativeHandle)
	Gets the native handle of a PI kernel object. More...
pi_result	piextKernelSuggestMaxCooperativeGroupCount (pi_kernel Kernel, size_t LocalWorkSize, size_t DynamicSharedMemorySize, pi_uint32 *GroupCountRet)
	Gets the max work group count for a cooperative kernel. More...
pi_result	piEventCreate (pi_context Context, pi_event *RetEvent)
	Create PI event object in a signalled/completed state. More...
pi_result	piEventGetInfo (pi_event Event, pi_event_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piEventGetProfilingInfo (pi_event Event, pi_profiling_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piEventsWait (pi_uint32 NumEvents, const pi_event *EventList)
pi_result	piEventSetCallback (pi_event Event, pi_int32 CommandExecCallbackType, void(*PFnNotify)(pi_event Event, pi_int32 EventCommandStatus, void *UserData), void *UserData)
pi_result	piEventSetStatus (pi_event Event, pi_int32 ExecutionStatus)
pi_result	piEventRetain (pi_event Event)
pi_result	piEventRelease (pi_event Event)
pi_result	piextEventGetNativeHandle (pi_event Event, pi_native_handle *NativeHandle)
	Gets the native handle of a PI event object. More...
pi_result	piextEventCreateWithNativeHandle (pi_native_handle NativeHandle, pi_context Context, bool OwnNativeHandle, pi_event *Event)
	Creates PI event object from a native handle. More...
pi_result	piSamplerCreate (pi_context Context, const pi_sampler_properties *SamplerProperties, pi_sampler *RetSampler)
pi_result	piSamplerGetInfo (pi_sampler Sampler, pi_sampler_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piSamplerRetain (pi_sampler Sampler)
pi_result	piSamplerRelease (pi_sampler Sampler)
pi_result	piEnqueueEventsWait (pi_queue Queue, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *OutEvent)
pi_result	piEnqueueEventsWaitWithBarrier (pi_queue Queue, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *OutEvent)
pi_result	piEnqueueMemBufferRead (pi_queue Queue, pi_mem Src, pi_bool BlockingRead, size_t Offset, size_t Size, void *Dst, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piEnqueueMemBufferReadRect (pi_queue Queue, pi_mem Buffer, pi_bool BlockingRead, pi_buff_rect_offset BufferOffset, pi_buff_rect_offset HostOffset, pi_buff_rect_region Region, size_t BufferRowPitch, size_t BufferSlicePitch, size_t HostRowPitch, size_t HostSlicePitch, void *Ptr, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piEnqueueMemBufferWrite (pi_queue Queue, pi_mem Buffer, pi_bool BlockingWrite, size_t Offset, size_t Size, const void *Ptr, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piEnqueueMemBufferWriteRect (pi_queue Queue, pi_mem Buffer, pi_bool BlockingWrite, pi_buff_rect_offset BufferOffset, pi_buff_rect_offset HostOffset, pi_buff_rect_region Region, size_t BufferRowPitch, size_t BufferSlicePitch, size_t HostRowPitch, size_t HostSlicePitch, const void *Ptr, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piEnqueueMemBufferCopy (pi_queue Queue, pi_mem SrcMem, pi_mem DstMem, size_t SrcOffset, size_t DstOffset, size_t Size, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piEnqueueMemBufferCopyRect (pi_queue Queue, pi_mem SrcMem, pi_mem DstMem, pi_buff_rect_offset SrcOrigin, pi_buff_rect_offset DstOrigin, pi_buff_rect_region Region, size_t SrcRowPitch, size_t SrcSlicePitch, size_t DstRowPitch, size_t DstSlicePitch, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piEnqueueMemBufferFill (pi_queue Queue, pi_mem Buffer, const void *Pattern, size_t PatternSize, size_t Offset, size_t Size, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piEnqueueMemBufferMap (pi_queue Queue, pi_mem Mem, pi_bool BlockingMap, pi_map_flags MapFlags, size_t Offset, size_t Size, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *OutEvent, void **RetMap)
pi_result	piEnqueueMemUnmap (pi_queue Queue, pi_mem Mem, void *MappedPtr, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *OutEvent)
pi_result	piMemImageGetInfo (pi_mem Image, pi_image_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piEnqueueMemImageRead (pi_queue Queue, pi_mem Image, pi_bool BlockingRead, pi_image_offset Origin, pi_image_region Region, size_t RowPitch, size_t SlicePitch, void *Ptr, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piEnqueueMemImageWrite (pi_queue Queue, pi_mem Image, pi_bool BlockingWrite, pi_image_offset Origin, pi_image_region Region, size_t InputRowPitch, size_t InputSlicePitch, const void *Ptr, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piEnqueueMemImageCopy (pi_queue Queue, pi_mem SrcImage, pi_mem DstImage, pi_image_offset SrcOrigin, pi_image_offset DstOrigin, pi_image_region Region, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piEnqueueMemImageFill (pi_queue Queue, pi_mem Image, const void *FillColor, const size_t *Origin, const size_t *Region, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
pi_result	piMemBufferPartition (pi_mem Buffer, pi_mem_flags Flags, pi_buffer_create_type BufferCreateType, void *BufferCreateInfo, pi_mem *RetMem)
pi_result	piextGetDeviceFunctionPointer (pi_device Device, pi_program Program, const char *FunctionName, pi_uint64 *FunctionPointerRet)
	Retrieves a device function pointer to a user-defined function. More...
pi_result	piextGetGlobalVariablePointer (pi_device Device, pi_program Program, const char *GlobalVariableName, size_t *GlobalVariableSize, void **GlobalVariablePointerRet)
pi_result	piextUSMDeviceAlloc (void **ResultPtr, pi_context Context, pi_device Device, pi_usm_mem_properties *Properties, size_t Size, pi_uint32 Alignment)
	Allocates device memory. More...
pi_result	piextUSMSharedAlloc (void **ResultPtr, pi_context Context, pi_device Device, pi_usm_mem_properties *Properties, size_t Size, pi_uint32 Alignment)
	Allocates memory accessible on both host and device. More...
pi_result	piextUSMPitchedAlloc (void **ResultPtr, size_t *ResultPitch, pi_context Context, pi_device Device, pi_usm_mem_properties *Properties, size_t WidthInBytes, size_t Height, unsigned int ElementSizeBytes)
	Allocates memory accessible on device. More...
pi_result	piextUSMHostAlloc (void **ResultPtr, pi_context Context, pi_usm_mem_properties *Properties, size_t Size, pi_uint32 Alignment)
	Allocates host memory accessible by the device. More...
pi_result	piextUSMFree (pi_context Context, void *Ptr)
	Indicates that the allocated USM memory is no longer needed on the runtime side. More...
pi_result	piextKernelSetArgPointer (pi_kernel Kernel, pi_uint32 ArgIndex, size_t ArgSize, const void *ArgValue)
	Sets up pointer arguments for CL kernels. More...
pi_result	piextUSMEnqueueMemset (pi_queue Queue, void *Ptr, pi_int32 Value, size_t Count, pi_uint32 NumEventsInWaitlist, const pi_event *EventsWaitlist, pi_event *Event)
	USM Memset API. More...
pi_result	piextUSMEnqueueMemcpy (pi_queue Queue, pi_bool Blocking, void *DstPtr, const void *SrcPtr, size_t Size, pi_uint32 NumEventsInWaitlist, const pi_event *EventsWaitlist, pi_event *Event)
	USM Memcpy API. More...
pi_result	piextUSMEnqueuePrefetch (pi_queue Queue, const void *Ptr, size_t Size, pi_usm_migration_flags Flags, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *OutEvent)
	Hint to migrate memory to the device. More...
pi_result	piextUSMEnqueueMemAdvise (pi_queue Queue, const void *Ptr, size_t Length, pi_mem_advice Advice, pi_event *OutEvent)
	USM Memadvise API. More...
pi_result	piextUSMEnqueueFill2D (pi_queue Queue, void *Ptr, size_t Pitch, size_t PatternSize, const void *Pattern, size_t Width, size_t Height, pi_uint32 NumEventsWaitList, const pi_event *EventsWaitList, pi_event *Event)
	USM 2D fill API. More...
pi_result	piextUSMEnqueueMemset2D (pi_queue Queue, void *Ptr, size_t Pitch, int Value, size_t Width, size_t Height, pi_uint32 NumEventsWaitList, const pi_event *EventsWaitlist, pi_event *Event)
	USM 2D Memset API. More...
pi_result	piextUSMEnqueueMemcpy2D (pi_queue Queue, pi_bool Blocking, void *DstPtr, size_t DstPitch, const void *SrcPtr, size_t SrcPitch, size_t Width, size_t Height, pi_uint32 NumEventsInWaitlist, const pi_event *EventWaitlist, pi_event *Event)
	USM 2D Memcpy API. More...
pi_result	piextUSMGetMemAllocInfo (pi_context Context, const void *Ptr, pi_mem_alloc_info ParamName, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
	API to query information about USM allocated pointers Valid Queries: PI_MEM_ALLOC_TYPE returns host/device/shared pi_host_usm value PI_MEM_ALLOC_BASE_PTR returns the base ptr of an allocation if the queried pointer fell inside an allocation. More...
pi_result	piextUSMImport (const void *HostPtr, size_t Size, pi_context Context)
	Import host system memory into USM. More...
pi_result	piextUSMRelease (const void *HostPtr, pi_context Context)
	Release host system memory from USM. More...
pi_result	piextEnqueueDeviceGlobalVariableWrite (pi_queue Queue, pi_program Program, const char *Name, pi_bool BlockingWrite, size_t Count, size_t Offset, const void *Src, pi_uint32 NumEventsInWaitList, const pi_event *EventsWaitList, pi_event *Event)
	Device global variable. More...
pi_result	piextEnqueueDeviceGlobalVariableRead (pi_queue Queue, pi_program Program, const char *Name, pi_bool BlockingRead, size_t Count, size_t Offset, void *Dst, pi_uint32 NumEventsInWaitList, const pi_event *EventsWaitList, pi_event *Event)
	API reading data from a device global variable to host. More...
pi_result	piextEnqueueReadHostPipe (pi_queue Queue, pi_program Program, const char *PipeSymbol, pi_bool Blocking, void *Ptr, size_t Size, pi_uint32 NumEventsInWaitList, const pi_event *EventsWaitList, pi_event *Event)
	Plugin. More...
pi_result	piextEnqueueWriteHostPipe (pi_queue Queue, pi_program Program, const char *PipeSymbol, pi_bool Blocking, void *Ptr, size_t Size, pi_uint32 NumEventsInWaitList, const pi_event *EventsWaitList, pi_event *Event)
	Write to pipe of a given name. More...
pi_result	piKernelSetExecInfo (pi_kernel Kernel, pi_kernel_exec_info ParamName, size_t ParamValueSize, const void *ParamValue)
	API to set attributes controlling kernel execution. More...
pi_result	piextProgramSetSpecializationConstant (pi_program Prog, pi_uint32 SpecID, size_t Size, const void *SpecValue)
	Sets a specialization constant to a specific value. More...
pi_result	piextCommandBufferCreate (pi_context Context, pi_device Device, const pi_ext_command_buffer_desc *Desc, pi_ext_command_buffer *RetCommandBuffer)
	API to create a command-buffer. More...
pi_result	piextCommandBufferRetain (pi_ext_command_buffer CommandBuffer)
	API to increment the reference count of the command-buffer. More...
pi_result	piextCommandBufferRelease (pi_ext_command_buffer CommandBuffer)
	API to decrement the reference count of the command-buffer. More...
pi_result	piextCommandBufferFinalize (pi_ext_command_buffer CommandBuffer)
	API to stop command-buffer recording such that no more commands can be appended, and makes the command-buffer ready to enqueue on a command-queue. More...
pi_result	piextCommandBufferNDRangeKernel (pi_ext_command_buffer CommandBuffer, pi_kernel Kernel, pi_uint32 WorkDim, const size_t *GlobalWorkOffset, const size_t *GlobalWorkSize, const size_t *LocalWorkSize, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint, pi_ext_command_buffer_command *Command)
	API to append a kernel execution command to the command-buffer. More...
pi_result	piextCommandBufferMemcpyUSM (pi_ext_command_buffer CommandBuffer, void *DstPtr, const void *SrcPtr, size_t Size, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a USM memcpy command to the command-buffer. More...
pi_result	piextCommandBufferMemBufferCopy (pi_ext_command_buffer CommandBuffer, pi_mem SrcMem, pi_mem DstMem, size_t SrcOffset, size_t DstOffset, size_t Size, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a mem buffer copy command to the command-buffer. More...
pi_result	piextCommandBufferMemBufferCopyRect (pi_ext_command_buffer CommandBuffer, pi_mem SrcMem, pi_mem DstMem, pi_buff_rect_offset SrcOrigin, pi_buff_rect_offset DstOrigin, pi_buff_rect_region Region, size_t SrcRowPitch, size_t SrcSlicePitch, size_t DstRowPitch, size_t DstSlicePitch, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a rectangular mem buffer copy command to the command-buffer. More...
pi_result	piextCommandBufferMemBufferRead (pi_ext_command_buffer CommandBuffer, pi_mem Buffer, size_t Offset, size_t Size, void *Dst, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a mem buffer read command to the command-buffer. More...
pi_result	piextCommandBufferMemBufferReadRect (pi_ext_command_buffer CommandBuffer, pi_mem Buffer, pi_buff_rect_offset BufferOffset, pi_buff_rect_offset HostOffset, pi_buff_rect_region Region, size_t BufferRowPitch, size_t BufferSlicePitch, size_t HostRowPitch, size_t HostSlicePitch, void *Ptr, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a rectangular mem buffer read command to the command-buffer. More...
pi_result	piextCommandBufferMemBufferWrite (pi_ext_command_buffer CommandBuffer, pi_mem Buffer, size_t Offset, size_t Size, const void *Ptr, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a mem buffer write command to the command-buffer. More...
pi_result	piextCommandBufferMemBufferWriteRect (pi_ext_command_buffer CommandBuffer, pi_mem Buffer, pi_buff_rect_offset BufferOffset, pi_buff_rect_offset HostOffset, pi_buff_rect_region Region, size_t BufferRowPitch, size_t BufferSlicePitch, size_t HostRowPitch, size_t HostSlicePitch, const void *Ptr, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a rectangular mem buffer write command to the command-buffer. More...
pi_result	piextCommandBufferMemBufferFill (pi_ext_command_buffer CommandBuffer, pi_mem Buffer, const void *Pattern, size_t PatternSize, size_t Offset, size_t Size, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a mem buffer fill command to the command-buffer. More...
pi_result	piextCommandBufferFillUSM (pi_ext_command_buffer CommandBuffer, void *Ptr, const void *Pattern, size_t PatternSize, size_t Size, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a USM fill command to the command-buffer. More...
pi_result	piextCommandBufferPrefetchUSM (pi_ext_command_buffer CommandBuffer, const void *Ptr, size_t Size, pi_usm_migration_flags Flags, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a USM Prefetch command to the command-buffer. More...
pi_result	piextCommandBufferAdviseUSM (pi_ext_command_buffer CommandBuffer, const void *Ptr, size_t Length, pi_mem_advice Advice, pi_uint32 NumSyncPointsInWaitList, const pi_ext_sync_point *SyncPointWaitList, pi_ext_sync_point *SyncPoint)
	API to append a USM Advise command to the command-buffer. More...
pi_result	piextEnqueueCommandBuffer (pi_ext_command_buffer CommandBuffer, pi_queue Queue, pi_uint32 NumEventsInWaitList, const pi_event *EventWaitList, pi_event *Event)
	API to submit the command-buffer to queue for execution, returns an error if the command-buffer is not finalized or another instance of the same command-buffer is currently executing. More...
pi_result	piextCommandBufferUpdateKernelLaunch (pi_ext_command_buffer_command Command, pi_ext_command_buffer_update_kernel_launch_desc *Desc)
	API to update a kernel launch command inside of a command-buffer. More...
pi_result	piextCommandBufferRetainCommand (pi_ext_command_buffer_command Command)
	API to increment the reference count of a command-buffer command. More...
pi_result	piextCommandBufferReleaseCommand (pi_ext_command_buffer_command Command)
	API to decrement the reference count of a command-buffer command. More...
pi_result	piextPluginGetOpaqueData (void *opaque_data_param, void **opaque_data_return)
	API to get Plugin internal data, opaque to SYCL RT. More...
pi_result	piTearDown (void *PluginParameter)
	API to notify that the plugin should clean up its resources. More...
pi_result	piGetDeviceAndHostTimer (pi_device Device, uint64_t *DeviceTime, uint64_t *HostTime)
	Queries device for it's global timestamp in nanoseconds, and updates HostTime with the value of the host timer at the closest possible point in time to that at which DeviceTime was returned. More...
pi_result	piextEnablePeerAccess (pi_device command_device, pi_device peer_device)
pi_result	piextDisablePeerAccess (pi_device command_device, pi_device peer_device)
pi_result	piextPeerAccessGetInfo (pi_device command_device, pi_device peer_device, pi_peer_attr attr, size_t ParamValueSize, void *ParamValue, size_t *ParamValueSizeRet)
pi_result	piPluginInit (pi_plugin *PluginInit)

Variables
const char	SupportedVersion [] = _PI_CUDA_PLUGIN_VERSION_STRING

Detailed Description

Implementation of CUDA Plugin.

Definition in file pi_cuda.cpp.

Macro Definition Documentation

_PI_API

#define _PI_API

(

api

)

(PluginInit->PiFunctionTable).api = (decltype(&::api))(&api);

Function Documentation

createCUDATracingContext()

cuda_tracing_context_t_* createCUDATracingContext

(

)

Referenced by piPluginInit().

disableCUDATracing()

void disableCUDATracing

(

cuda_tracing_context_t_ *

ctx

)

enableCUDATracing()

void enableCUDATracing

(

cuda_tracing_context_t_ *

ctx

)

Referenced by piPluginInit().

freeCUDATracingContext()

void freeCUDATracingContext

(

cuda_tracing_context_t_ *

Ctx

)

piContextCreate()

pi_result piContextCreate	(	const pi_context_properties *	Properties,
		pi_uint32	NumDevices,
		const pi_device *	Devices,
		void(*)(const char *ErrInfo, const void *PrivateInfo, size_t CB, void *UserData)	PFnNotify,
		void *	UserData,
		pi_context *	RetContext
	)

Definition at line 114 of file pi_cuda.cpp.

References pi2ur::piContextCreate().

piContextGetInfo()

pi_result piContextGetInfo	(	pi_context	Context,
		pi_context_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 124 of file pi_cuda.cpp.

piContextRelease()

pi_result piContextRelease

(

pi_context

Context

)

Definition at line 157 of file pi_cuda.cpp.

piContextRetain()

pi_result piContextRetain

(

pi_context

Context

)

Definition at line 152 of file pi_cuda.cpp.

piDeviceGetInfo()

pi_result piDeviceGetInfo	(	pi_device	Device,
		pi_device_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Returns requested info for provided native device Return PI_DEVICE_INFO_EXTENSION_DEVICELIB_ASSERT for PI_DEVICE_INFO_EXTENSIONS query when the device supports native asserts.

Definition at line 78 of file pi_cuda.cpp.

piDevicePartition()

pi_result piDevicePartition	(	pi_device	Device,
		const pi_device_partition_property *	Properties,
		pi_uint32	NumDevices,
		pi_device *	OutDevices,
		pi_uint32 *	OutNumDevices
	)

Definition at line 85 of file pi_cuda.cpp.

piDeviceRelease()

pi_result piDeviceRelease

(

pi_device

Device

)

Definition at line 74 of file pi_cuda.cpp.

piDeviceRetain()

pi_result piDeviceRetain

(

pi_device

Device

)

Definition at line 70 of file pi_cuda.cpp.

piDevicesGet()

pi_result piDevicesGet	(	pi_platform	Platform,
		pi_device_type	DeviceType,
		pi_uint32	NumEntries,
		pi_device *	Devices,
		pi_uint32 *	NumDevices
	)

Definition at line 63 of file pi_cuda.cpp.

piEnqueueEventsWait()

pi_result piEnqueueEventsWait	(	pi_queue	Queue,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	OutEvent
	)

Definition at line 657 of file pi_cuda.cpp.

piEnqueueEventsWaitWithBarrier()

pi_result piEnqueueEventsWaitWithBarrier	(	pi_queue	Queue,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	OutEvent
	)

Definition at line 665 of file pi_cuda.cpp.

piEnqueueKernelLaunch()

pi_result piEnqueueKernelLaunch	(	pi_queue	Queue,
		pi_kernel	Kernel,
		pi_uint32	WorkDim,
		const size_t *	GlobalWorkOffset,
		const size_t *	GlobalWorkSize,
		const size_t *	LocalWorkSize,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	OutEvent
	)

Definition at line 537 of file pi_cuda.cpp.

piEnqueueMemBufferCopy()

pi_result piEnqueueMemBufferCopy	(	pi_queue	Queue,
		pi_mem	SrcMem,
		pi_mem	DstMem,
		size_t	SrcOffset,
		size_t	DstOffset,
		size_t	Size,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 726 of file pi_cuda.cpp.

piEnqueueMemBufferCopyRect()

pi_result piEnqueueMemBufferCopyRect	(	pi_queue	Queue,
		pi_mem	SrcMem,
		pi_mem	DstMem,
		pi_buff_rect_offset	SrcOrigin,
		pi_buff_rect_offset	DstOrigin,
		pi_buff_rect_region	Region,
		size_t	SrcRowPitch,
		size_t	SrcSlicePitch,
		size_t	DstRowPitch,
		size_t	DstSlicePitch,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 737 of file pi_cuda.cpp.

piEnqueueMemBufferFill()

pi_result piEnqueueMemBufferFill	(	pi_queue	Queue,
		pi_mem	Buffer,
		const void *	Pattern,
		size_t	PatternSize,
		size_t	Offset,
		size_t	Size,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 750 of file pi_cuda.cpp.

piEnqueueMemBufferMap()

pi_result piEnqueueMemBufferMap	(	pi_queue	Queue,
		pi_mem	Mem,
		pi_bool	BlockingMap,
		pi_map_flags	MapFlags,
		size_t	Offset,
		size_t	Size,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	OutEvent,
		void **	RetMap
	)

Definition at line 762 of file pi_cuda.cpp.

piEnqueueMemBufferRead()

pi_result piEnqueueMemBufferRead	(	pi_queue	Queue,
		pi_mem	Src,
		pi_bool	BlockingRead,
		size_t	Offset,
		size_t	Size,
		void *	Dst,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 674 of file pi_cuda.cpp.

piEnqueueMemBufferReadRect()

pi_result piEnqueueMemBufferReadRect	(	pi_queue	Queue,
		pi_mem	Buffer,
		pi_bool	BlockingRead,
		pi_buff_rect_offset	BufferOffset,
		pi_buff_rect_offset	HostOffset,
		pi_buff_rect_region	Region,
		size_t	BufferRowPitch,
		size_t	BufferSlicePitch,
		size_t	HostRowPitch,
		size_t	HostSlicePitch,
		void *	Ptr,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 686 of file pi_cuda.cpp.

piEnqueueMemBufferWrite()

pi_result piEnqueueMemBufferWrite	(	pi_queue	Queue,
		pi_mem	Buffer,
		pi_bool	BlockingWrite,
		size_t	Offset,
		size_t	Size,
		const void *	Ptr,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 700 of file pi_cuda.cpp.

piEnqueueMemBufferWriteRect()

pi_result piEnqueueMemBufferWriteRect	(	pi_queue	Queue,
		pi_mem	Buffer,
		pi_bool	BlockingWrite,
		pi_buff_rect_offset	BufferOffset,
		pi_buff_rect_offset	HostOffset,
		pi_buff_rect_region	Region,
		size_t	BufferRowPitch,
		size_t	BufferSlicePitch,
		size_t	HostRowPitch,
		size_t	HostSlicePitch,
		const void *	Ptr,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 712 of file pi_cuda.cpp.

piEnqueueMemImageCopy()

pi_result piEnqueueMemImageCopy	(	pi_queue	Queue,
		pi_mem	SrcImage,
		pi_mem	DstImage,
		pi_image_offset	SrcOrigin,
		pi_image_offset	DstOrigin,
		pi_image_region	Region,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 815 of file pi_cuda.cpp.

piEnqueueMemImageFill()

pi_result piEnqueueMemImageFill	(	pi_queue	Queue,
		pi_mem	Image,
		const void *	FillColor,
		const size_t *	Origin,
		const size_t *	Region,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 824 of file pi_cuda.cpp.

piEnqueueMemImageRead()

pi_result piEnqueueMemImageRead	(	pi_queue	Queue,
		pi_mem	Image,
		pi_bool	BlockingRead,
		pi_image_offset	Origin,
		pi_image_region	Region,
		size_t	RowPitch,
		size_t	SlicePitch,
		void *	Ptr,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 789 of file pi_cuda.cpp.

piEnqueueMemImageWrite()

pi_result piEnqueueMemImageWrite	(	pi_queue	Queue,
		pi_mem	Image,
		pi_bool	BlockingWrite,
		pi_image_offset	Origin,
		pi_image_region	Region,
		size_t	InputRowPitch,
		size_t	InputSlicePitch,
		const void *	Ptr,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	Event
	)

Definition at line 801 of file pi_cuda.cpp.

piEnqueueMemUnmap()

pi_result piEnqueueMemUnmap	(	pi_queue	Queue,
		pi_mem	Mem,
		void *	MappedPtr,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	OutEvent
	)

Definition at line 773 of file pi_cuda.cpp.

piEventCreate()

pi_result piEventCreate	(	pi_context	context,
		pi_event *	ret_event
	)

Create PI event object in a signalled/completed state.

Parameters

context	is the PI context of the event.
ret_event	is the PI even created.

Definition at line 579 of file pi_cuda.cpp.

piEventGetInfo()

pi_result piEventGetInfo	(	pi_event	Event,
		pi_event_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 583 of file pi_cuda.cpp.

piEventGetProfilingInfo()

pi_result piEventGetProfilingInfo	(	pi_event	Event,
		pi_profiling_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 590 of file pi_cuda.cpp.

piEventRelease()

pi_result piEventRelease

(

pi_event

Event

)

Definition at line 617 of file pi_cuda.cpp.

piEventRetain()

pi_result piEventRetain

(

pi_event

Event

)

Definition at line 615 of file pi_cuda.cpp.

piEventSetCallback()

pi_result piEventSetCallback	(	pi_event	Event,
		pi_int32	CommandExecCallbackType,
		void(*)(pi_event Event, pi_int32 EventCommandStatus, void *UserData)	PFnNotify,
		void *	UserData
	)

Definition at line 602 of file pi_cuda.cpp.

References pi2ur::piEventSetCallback().

piEventSetStatus()

pi_result piEventSetStatus	(	pi_event	Event,
		pi_int32	ExecutionStatus
	)

Definition at line 611 of file pi_cuda.cpp.

piEventsWait()

pi_result piEventsWait	(	pi_uint32	NumEvents,
		const pi_event *	EventList
	)

Definition at line 598 of file pi_cuda.cpp.

piextBindlessImageSamplerCreate()

pi_result piextBindlessImageSamplerCreate	(	pi_context	context,
		const pi_sampler_properties *	sampler_properties,
		float	min_mipmap_level_clamp,
		float	max_mipmap_level_clamp,
		float	max_anisotropy,
		pi_sampler *	result_sampler
	)

API to create samplers for bindless images.

Parameters

context	is the pi_context
device	is the pi_device
sampler_properties	is the pointer to the sampler properties bitfield
min_mipmap_level_clamp	is the minimum mipmap level to sample from
max_mipmap_level_clamp	is the maximum mipmap level to sample from
max_anisotropy	is the maximum anisotropic ratio
result_sampler	is the returned sampler

Definition at line 400 of file pi_cuda.cpp.

piextCommandBufferAdviseUSM()

pi_result piextCommandBufferAdviseUSM	(	pi_ext_command_buffer	command_buffer,
		const void *	ptr,
		size_t	length,
		pi_mem_advice	advice,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a USM Advise command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
ptr	is the data to be advised.
length	is the size in bytes of the memory to advise.
advice	is device specific advice.
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1204 of file pi_cuda.cpp.

piextCommandBufferCreate()

pi_result piextCommandBufferCreate	(	pi_context	context,
		pi_device	device,
		const pi_ext_command_buffer_desc *	desc,
		pi_ext_command_buffer *	ret_command_buffer
	)

API to create a command-buffer.

Parameters

context	The context to associate the command-buffer with.
device	The device to associate the command-buffer with.
desc	Descriptor for the new command-buffer.
ret_command_buffer	Pointer to fill with the address of the new command-buffer.

Definition at line 1068 of file pi_cuda.cpp.

piextCommandBufferFillUSM()

pi_result piextCommandBufferFillUSM	(	pi_ext_command_buffer	command_buffer,
		void *	ptr,
		const void *	pattern,
		size_t	pattern_size,
		size_t	size,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a USM fill command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
ptr	pointer to the USM allocation to fill.
pattern	pointer to the pattern to fill ptr with.
pattern_size	size of the pattern in bytes.
size	fill size in bytes.
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1184 of file pi_cuda.cpp.

piextCommandBufferFinalize()

pi_result piextCommandBufferFinalize

(

pi_ext_command_buffer

command_buffer

)

API to stop command-buffer recording such that no more commands can be appended, and makes the command-buffer ready to enqueue on a command-queue.

Parameters

command_buffer

The command_buffer to finalize.

Definition at line 1083 of file pi_cuda.cpp.

piextCommandBufferMemBufferCopy()

pi_result piextCommandBufferMemBufferCopy	(	pi_ext_command_buffer	command_buffer,
		pi_mem	src_buffer,
		pi_mem	dst_buffer,
		size_t	src_offset,
		size_t	dst_offset,
		size_t	size,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a mem buffer copy command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
src_buffer	is the data to be copied
dst_buffer	is the location the data will be copied
src_offset	offset into src_buffer
dst_offset	offset into dst_buffer
size	is number of bytes to copy
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1108 of file pi_cuda.cpp.

piextCommandBufferMemBufferCopyRect()

pi_result piextCommandBufferMemBufferCopyRect	(	pi_ext_command_buffer	command_buffer,
		pi_mem	src_buffer,
		pi_mem	dst_buffer,
		pi_buff_rect_offset	src_origin,
		pi_buff_rect_offset	dst_origin,
		pi_buff_rect_region	region,
		size_t	src_row_pitch,
		size_t	src_slice_pitch,
		size_t	dst_row_pitch,
		size_t	dst_slice_pitch,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a rectangular mem buffer copy command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
src_buffer	is the data to be copied
dst_buffer	is the location the data will be copied
src_origin	offset for the start of the region to copy in src_buffer
dst_origin	offset for the start of the region to copy in dst_buffer
region	The size of the region to be copied
src_row_pitch	Row pitch for the src data
src_slice_pitch	Slice pitch for the src data
dst_row_pitch	Row pitch for the dst data
dst_slice_pitch	Slice pitch for the dst data
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1118 of file pi_cuda.cpp.

piextCommandBufferMemBufferFill()

pi_result piextCommandBufferMemBufferFill	(	pi_ext_command_buffer	command_buffer,
		pi_mem	buffer,
		const void *	pattern,
		size_t	pattern_size,
		size_t	offset,
		size_t	size,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a mem buffer fill command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
buffer	is the location to fill the data.
pattern	pointer to the pattern to fill the buffer with.
pattern_size	size of the pattern in bytes.
offset	Offset into the buffer to fill from.
size	fill size in bytes.
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1174 of file pi_cuda.cpp.

piextCommandBufferMemBufferRead()

pi_result piextCommandBufferMemBufferRead	(	pi_ext_command_buffer	command_buffer,
		pi_mem	buffer,
		size_t	offset,
		size_t	size,
		void *	dst,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a mem buffer read command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
buffer	is the data to be read
offset	offset into buffer
size	is number of bytes to read
dst	is the pointer to the destination
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1130 of file pi_cuda.cpp.

piextCommandBufferMemBufferReadRect()

pi_result piextCommandBufferMemBufferReadRect	(	pi_ext_command_buffer	command_buffer,
		pi_mem	buffer,
		pi_buff_rect_offset	buffer_offset,
		pi_buff_rect_offset	host_offset,
		pi_buff_rect_region	region,
		size_t	buffer_row_pitch,
		size_t	buffer_slice_pitch,
		size_t	host_row_pitch,
		size_t	host_slice_pitch,
		void *	ptr,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a rectangular mem buffer read command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
buffer	is the data to be read
buffer_offset	offset for the start of the region to read in buffer
host_offset	offset for the start of the region to be written from ptr
region	The size of the region to read
buffer_row_pitch	Row pitch for the source buffer data
buffer_slice_pitch	Slice pitch for the source buffer data
host_row_pitch	Row pitch for the destination data ptr
host_slice_pitch	Slice pitch for the destination data ptr
ptr	is the location the data will be written
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1139 of file pi_cuda.cpp.

piextCommandBufferMemBufferWrite()

pi_result piextCommandBufferMemBufferWrite	(	pi_ext_command_buffer	command_buffer,
		pi_mem	buffer,
		size_t	offset,
		size_t	size,
		const void *	ptr,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a mem buffer write command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
buffer	is the location to write the data
offset	offset into buffer
size	is number of bytes to write
ptr	is the pointer to the source
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1152 of file pi_cuda.cpp.

piextCommandBufferMemBufferWriteRect()

pi_result piextCommandBufferMemBufferWriteRect	(	pi_ext_command_buffer	command_buffer,
		pi_mem	buffer,
		pi_buff_rect_offset	buffer_offset,
		pi_buff_rect_offset	host_offset,
		pi_buff_rect_region	region,
		size_t	buffer_row_pitch,
		size_t	buffer_slice_pitch,
		size_t	host_row_pitch,
		size_t	host_slice_pitch,
		const void *	ptr,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a rectangular mem buffer write command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
buffer	is the location to write the data
buffer_offset	offset for the start of the region to write in buffer
host_offset	offset for the start of the region to be read from ptr
region	The size of the region to write
buffer_row_pitch	Row pitch for the buffer data
buffer_slice_pitch	Slice pitch for the buffer data
host_row_pitch	Row pitch for the source data ptr
host_slice_pitch	Slice pitch for the source data ptr
ptr	is the pointer to the source
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1161 of file pi_cuda.cpp.

piextCommandBufferMemcpyUSM()

pi_result piextCommandBufferMemcpyUSM	(	pi_ext_command_buffer	command_buffer,
		void *	dst_ptr,
		const void *	src_ptr,
		size_t	size,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a USM memcpy command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
dst_ptr	is the location the data will be copied
src_ptr	is the data to be copied
size	is number of bytes to copy
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1099 of file pi_cuda.cpp.

piextCommandBufferNDRangeKernel()

pi_result piextCommandBufferNDRangeKernel	(	pi_ext_command_buffer	command_buffer,
		pi_kernel	kernel,
		pi_uint32	work_dim,
		const size_t *	global_work_offset,
		const size_t *	global_work_size,
		const size_t *	local_work_size,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point,
		pi_ext_command_buffer_command *	command
	)

API to append a kernel execution command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
kernel	The kernel to append.
work_dim	Dimension of the kernel execution.
global_work_offset	Offset to use when executing kernel.
global_work_size	Global work size to use when executing kernel.
local_work_size	Local work size to use when executing kernel.
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this kernel execution.
command	Return pointer to the command representing this kernel execution.

Definition at line 1087 of file pi_cuda.cpp.

piextCommandBufferPrefetchUSM()

pi_result piextCommandBufferPrefetchUSM	(	pi_ext_command_buffer	command_buffer,
		const void *	ptr,
		size_t	size,
		pi_usm_migration_flags	flags,
		pi_uint32	num_sync_points_in_wait_list,
		const pi_ext_sync_point *	sync_point_wait_list,
		pi_ext_sync_point *	sync_point
	)

API to append a USM Prefetch command to the command-buffer.

Parameters

command_buffer	The command-buffer to append onto.
ptr	points to the memory to migrate.
size	is the number of bytes to migrate.
flags	is a bitfield used to specify memory migration options.
num_sync_points_in_wait_list	The number of sync points in the provided wait list.
sync_point_wait_list	A list of sync points that this command must wait on.
sync_point	The sync_point associated with this memory operation.

Definition at line 1195 of file pi_cuda.cpp.

piextCommandBufferRelease()

pi_result piextCommandBufferRelease

(

pi_ext_command_buffer

command_buffer

)

API to decrement the reference count of the command-buffer.

After the command_buffer reference count becomes zero and has finished execution, the command-buffer is deleted.

Parameters

command_buffer

The command_buffer to release.

Definition at line 1079 of file pi_cuda.cpp.

piextCommandBufferReleaseCommand()

pi_result piextCommandBufferReleaseCommand

(

pi_ext_command_buffer_command

command

)

API to decrement the reference count of a command-buffer command.

After the command reference count becomes zero, the command is deleted.

Parameters

command

The command to release.

Definition at line 1234 of file pi_cuda.cpp.

piextCommandBufferRetain()

pi_result piextCommandBufferRetain

(

pi_ext_command_buffer

command_buffer

)

API to increment the reference count of the command-buffer.

Parameters

command_buffer

The command_buffer to retain.

Definition at line 1075 of file pi_cuda.cpp.

piextCommandBufferRetainCommand()

pi_result piextCommandBufferRetainCommand

(

pi_ext_command_buffer_command

command

)

API to increment the reference count of a command-buffer command.

Parameters

command

The command to release.

Definition at line 1229 of file pi_cuda.cpp.

piextCommandBufferUpdateKernelLaunch()

pi_result piextCommandBufferUpdateKernelLaunch	(	pi_ext_command_buffer_command	command,
		pi_ext_command_buffer_update_kernel_launch_desc *	desc
	)

API to update a kernel launch command inside of a command-buffer.

Parameters

command	The command to be updated.
desc	Descriptor which describes the updated parameters of the kernel launch.

Definition at line 1222 of file pi_cuda.cpp.

piextContextCreateWithNativeHandle()

pi_result piextContextCreateWithNativeHandle	(	pi_native_handle	nativeHandle,
		pi_uint32	numDevices,
		const pi_device *	devices,
		bool	pluginOwnsNativeHandle,
		pi_context *	context
	)

Creates PI context object from a native handle.

NOTE: The created PI object takes ownership of the native handle. NOTE: The number of devices and the list of devices is needed for Level Zero backend because there is no possilibity to query this information from context handle for Level Zero. If backend has API to query a list of devices from the context native handle then these parameters are ignored.

Parameters

nativeHandle	is the native handle to create PI context from.
numDevices	is the number of devices in the context. Parameter is ignored if number of devices can be queried from the context native handle for a backend.
devices	is the list of devices in the context. Parameter is ignored if devices can be queried from the context native handle for a backend.
pluginOwnsNativeHandle	Indicates whether the created PI object should take ownership of the native handle.
context	is the PI context created from the native handle.

Returns

PI_SUCCESS if successfully created pi_context from the handle. PI_ERROR_OUT_OF_HOST_MEMORY if can't allocate memory for the pi_context object. PI_ERROR_INVALID_VALUE if numDevices == 0 or devices is NULL but backend doesn't have API to query a list of devices from the context native handle. PI_UNKNOWN_ERROR in case of another error.

Definition at line 143 of file pi_cuda.cpp.

piextContextGetNativeHandle()

pi_result piextContextGetNativeHandle	(	pi_context	context,
		pi_native_handle *	nativeHandle
	)

Gets the native handle of a PI context object.

Parameters

context	is the PI context to get the native handle of.
nativeHandle	is the native handle of context.

Definition at line 138 of file pi_cuda.cpp.

piextContextSetExtendedDeleter()

pi_result piextContextSetExtendedDeleter	(	pi_context	Context,
		pi_context_extended_deleter	Function,
		void *	UserData
	)

Definition at line 132 of file pi_cuda.cpp.

piextDestroyExternalSemaphore()

pi_result piextDestroyExternalSemaphore	(	pi_context	context,
		pi_device	device,
		pi_interop_semaphore_handle	sem_handle
	)

API to destroy the external semaphore handle.

Parameters

context	is the pi_context
device	is the pi_device
sem_handle	is the interop semaphore handle to the external semaphore to be destroyed

Definition at line 487 of file pi_cuda.cpp.

piextDeviceCreateWithNativeHandle()

pi_result piextDeviceCreateWithNativeHandle	(	pi_native_handle	nativeHandle,
		pi_platform	platform,
		pi_device *	device
	)

Creates PI device object from a native handle.

NOTE: The created PI object takes ownership of the native handle.

Parameters

nativeHandle	is the native handle to create PI device from.
platform	is the platform of the device (optional).
device	is the PI device created from the native handle.

Definition at line 106 of file pi_cuda.cpp.

piextDeviceGetNativeHandle()

pi_result piextDeviceGetNativeHandle	(	pi_device	device,
		pi_native_handle *	nativeHandle
	)

Gets the native handle of a PI device object.

Parameters

device	is the PI device to get the native handle of.
nativeHandle	is the native handle of device.

Definition at line 100 of file pi_cuda.cpp.

piextDeviceSelectBinary()

pi_result piextDeviceSelectBinary	(	pi_device	Device,
		pi_device_binary *	Binaries,
		pi_uint32	NumBinaries,
		pi_uint32 *	SelectedBinaryInd
	)

Selects the most appropriate device binary based on runtime information and the IR characteristics.

Definition at line 93 of file pi_cuda.cpp.

piextDisablePeerAccess()

pi_result piextDisablePeerAccess	(	pi_device	command_device,
		pi_device	peer_device
	)

Definition at line 1258 of file pi_cuda.cpp.

piextEnablePeerAccess()

pi_result piextEnablePeerAccess	(	pi_device	command_device,
		pi_device	peer_device
	)

Definition at line 1252 of file pi_cuda.cpp.

piextEnqueueCommandBuffer()

pi_result piextEnqueueCommandBuffer	(	pi_ext_command_buffer	command_buffer,
		pi_queue	queue,
		pi_uint32	num_events_in_wait_list,
		const pi_event *	event_wait_list,
		pi_event *	event
	)

API to submit the command-buffer to queue for execution, returns an error if the command-buffer is not finalized or another instance of the same command-buffer is currently executing.

Parameters

command_buffer	The command-buffer to be submitted.
queue	The PI queue to submit on.
num_events_in_wait_list	The number of events that this execution depends on.
event_wait_list	List of pi_events to wait on.
event	The pi_event associated with this enqueue.

Definition at line 1213 of file pi_cuda.cpp.

piextEnqueueCooperativeKernelLaunch()

pi_result piextEnqueueCooperativeKernelLaunch	(	pi_queue	Queue,
		pi_kernel	Kernel,
		pi_uint32	WorkDim,
		const size_t *	GlobalWorkOffset,
		const size_t *	GlobalWorkSize,
		const size_t *	LocalWorkSize,
		pi_uint32	NumEventsInWaitList,
		const pi_event *	EventWaitList,
		pi_event *	OutEvent
	)

Definition at line 547 of file pi_cuda.cpp.

piextEnqueueDeviceGlobalVariableRead()

pi_result piextEnqueueDeviceGlobalVariableRead	(	pi_queue	queue,
		pi_program	program,
		const char *	name,
		pi_bool	blocking_read,
		size_t	count,
		size_t	offset,
		void *	dst,
		pi_uint32	num_events_in_wait_list,
		const pi_event *	event_wait_list,
		pi_event *	event
	)

API reading data from a device global variable to host.

Parameters

queue	is the queue
program	is the program containing the device global variable
blocking_read	is true if the read should block
name	is the unique identifier for the device global variable
count	is the number of bytes to copy
offset	is the byte offset into the device global variable to start copying
dst	is a pointer to where the data must be copied to
num_events_in_wait_list	is a number of events in the wait list
event_wait_list	is the wait list
event	is the resulting event

Definition at line 1002 of file pi_cuda.cpp.

piextEnqueueDeviceGlobalVariableWrite()

pi_result piextEnqueueDeviceGlobalVariableWrite	(	pi_queue	queue,
		pi_program	program,
		const char *	name,
		pi_bool	blocking_write,
		size_t	count,
		size_t	offset,
		const void *	src,
		pi_uint32	num_events_in_wait_list,
		const pi_event *	event_wait_list,
		pi_event *	event
	)

Device global variable.

API for writing data from host to a device global variable.

Parameters

queue	is the queue
program	is the program containing the device global variable
blocking_write	is true if the write should block
name	is the unique identifier for the device global variable
count	is the number of bytes to copy
offset	is the byte offset into the device global variable to start copying
src	is a pointer to where the data must be copied from
num_events_in_wait_list	is a number of events in the wait list
event_wait_list	is the wait list
event	is the resulting event

Definition at line 993 of file pi_cuda.cpp.

piextEnqueueReadHostPipe()

pi_result piextEnqueueReadHostPipe	(	pi_queue	queue,
		pi_program	program,
		const char *	pipe_symbol,
		pi_bool	blocking,
		void *	ptr,
		size_t	size,
		pi_uint32	num_events_in_waitlist,
		const pi_event *	events_waitlist,
		pi_event *	event
	)

Plugin.

Read from pipe of a given name

Parameters

queue	a valid host command-queue in which the read / write command will be queued. command_queue and program must be created with the same OpenCL context.
program	a program object with a successfully built executable.
pipe_symbol	the name of the program scope pipe global variable.
blocking	indicate if the read and write operations are blocking or non-blocking
ptr	a pointer to buffer in host memory that will hold resulting data from pipe
size	size of the memory region to read or write, in bytes.
num_events_in_waitlist	number of events in the wait list.
events_waitlist	specify events that need to complete before this particular command can be executed.
event	returns an event object that identifies this read / write command and can be used to query or queue a wait for this command to complete.

Definition at line 1013 of file pi_cuda.cpp.

piextEnqueueWriteHostPipe()

pi_result piextEnqueueWriteHostPipe	(	pi_queue	queue,
		pi_program	program,
		const char *	pipe_symbol,
		pi_bool	blocking,
		void *	ptr,
		size_t	size,
		pi_uint32	num_events_in_waitlist,
		const pi_event *	events_waitlist,
		pi_event *	event
	)

Write to pipe of a given name.

Parameters

queue	a valid host command-queue in which the read / write command will be queued. command_queue and program must be created with the same OpenCL context.
program	a program object with a successfully built executable.
pipe_symbol	the name of the program scope pipe global variable.
blocking	indicate if the read and write operations are blocking or non-blocking
ptr	a pointer to buffer in host memory that holds data to be written to host pipe.
size	size of the memory region to read or write, in bytes.
num_events_in_waitlist	number of events in the wait list.
events_waitlist	specify events that need to complete before this particular command can be executed.
event	returns an event object that identifies this read / write command and can be used to query or queue a wait for this command to complete.

Definition at line 1033 of file pi_cuda.cpp.

piextEventCreateWithNativeHandle()

pi_result piextEventCreateWithNativeHandle	(	pi_native_handle	nativeHandle,
		pi_context	context,
		bool	ownNativeHandle,
		pi_event *	event
	)

Creates PI event object from a native handle.

NOTE: The created PI object takes ownership of the native handle.

Parameters

nativeHandle	is the native handle to create PI event from.
context	is the corresponding PI context
pluginOwnsNativeHandle	Indicates whether the created PI object should take ownership of the native handle.
event	is the PI event created from the native handle.

Definition at line 627 of file pi_cuda.cpp.

piextEventGetNativeHandle()

pi_result piextEventGetNativeHandle	(	pi_event	event,
		pi_native_handle *	nativeHandle
	)

Gets the native handle of a PI event object.

Parameters

event	is the PI event to get the native handle of.
nativeHandle	is the native handle of event.

Definition at line 621 of file pi_cuda.cpp.

piextGetDeviceFunctionPointer()

pi_result piextGetDeviceFunctionPointer	(	pi_device	device,
		pi_program	program,
		const char *	function_name,
		pi_uint64 *	function_pointer_ret
	)

Retrieves a device function pointer to a user-defined function.

• function_name.
• function_pointer_ret is set to 0 if query failed.

• program must be built before calling this API.
• device must present in the list of devices returned by get_device method for
• program.

If a fallback method determines the function exists but the address is not available PI_ERROR_FUNCTION_ADDRESS_IS_NOT_AVAILABLE is returned. If the address does not exist PI_ERROR_INVALID_KERNEL_NAME is returned.

Definition at line 844 of file pi_cuda.cpp.

piextGetGlobalVariablePointer()

pi_result piextGetGlobalVariablePointer	(	pi_device	Device,
		pi_program	Program,
		const char *	GlobalVariableName,
		size_t *	GlobalVariableSize,
		void **	GlobalVariablePointerRet
	)

Definition at line 851 of file pi_cuda.cpp.

piextImportExternalSemaphoreOpaqueFD()

pi_result piextImportExternalSemaphoreOpaqueFD	(	pi_context	context,
		pi_device	device,
		int	file_descriptor,
		pi_interop_semaphore_handle *	ret_handle
	)

API to import an external semaphore in the form of a file descriptor.

Parameters

context	is the pi_context
device	is the pi_device
file_descriptor	is the file descriptor
ret_handle	is the returned interop semaphore handle to the external semaphore

Definition at line 479 of file pi_cuda.cpp.

piextKernelCreateWithNativeHandle()

pi_result piextKernelCreateWithNativeHandle	(	pi_native_handle	nativeHandle,
		pi_context	context,
		pi_program	program,
		bool	pluginOwnsNativeHandle,
		pi_kernel *	kernel
	)

Creates PI kernel object from a native handle.

NOTE: The created PI object takes ownership of the native handle.

Parameters

nativeHandle	is the native handle to create PI kernel from.
context	is the PI context of the kernel.
program	is the PI program of the kernel.
pluginOwnsNativeHandle	Indicates whether the created PI object should take ownership of the native handle.
kernel	is the PI kernel created from the native handle.

Definition at line 557 of file pi_cuda.cpp.

piextKernelGetNativeHandle()

pi_result piextKernelGetNativeHandle	(	pi_kernel	kernel,
		pi_native_handle *	nativeHandle
	)

Gets the native handle of a PI kernel object.

Parameters

kernel	is the PI kernel to get the native handle of.
nativeHandle	is the native handle of kernel.

Definition at line 567 of file pi_cuda.cpp.

piextKernelSetArgMemObj()

pi_result piextKernelSetArgMemObj	(	pi_kernel	Kernel,
		pi_uint32	ArgIndex,
		const pi_mem_obj_property *	ArgProperties,
		const pi_mem *	ArgValue
	)

Definition at line 353 of file pi_cuda.cpp.

piextKernelSetArgPointer()

pi_result piextKernelSetArgPointer	(	pi_kernel	kernel,
		pi_uint32	arg_index,
		size_t	arg_size,
		const void *	arg_value
	)

Sets up pointer arguments for CL kernels.

An extra indirection is required due to CL argument conventions.

Parameters

kernel	is the kernel to be launched
arg_index	is the index of the kernel argument
arg_size	is the size in bytes of the argument (ignored in CL)
arg_value	is the pointer argument

Definition at line 900 of file pi_cuda.cpp.

piextKernelSetArgSampler()

pi_result piextKernelSetArgSampler	(	pi_kernel	Kernel,
		pi_uint32	ArgIndex,
		const pi_sampler *	ArgValue
	)

Definition at line 360 of file pi_cuda.cpp.

piextKernelSuggestMaxCooperativeGroupCount()

pi_result piextKernelSuggestMaxCooperativeGroupCount	(	pi_kernel	kernel,
		size_t	local_work_size,
		size_t	dynamic_shared_memory_size,
		pi_uint32 *	group_count_ret
	)

Gets the max work group count for a cooperative kernel.

Parameters

kernel	is the PI kernel being queried.
local_work_size	is the number of work items in a work group that will be used when the kernel is launched.
dynamic_shared_memory_size	is the size of dynamic shared memory, for each work group, in bytes, that will be used when the kernel is launched."
group_count_ret	is a pointer to where the query result will be stored.

Definition at line 572 of file pi_cuda.cpp.

piextMemCreateWithNativeHandle()

pi_result piextMemCreateWithNativeHandle	(	pi_native_handle	nativeHandle,
		pi_context	context,
		bool	ownNativeHandle,
		pi_mem *	mem
	)

Creates PI mem object from a native handle.

NOTE: The created PI object takes ownership of the native handle.

Parameters

nativeHandle	is the native handle to create PI mem from.
context	The PI context of the memory allocation.
ownNativeHandle	Indicates if we own the native memory handle or it came from interop that asked to not transfer the ownership to SYCL RT.
mem	is the PI mem created from the native handle.

Definition at line 241 of file pi_cuda.cpp.

piextMemGetNativeHandle()

pi_result piextMemGetNativeHandle	(	pi_mem	mem,
		pi_device	dev,
		pi_native_handle *	nativeHandle
	)

Gets the native handle of a PI mem object.

Parameters

mem	is the PI mem to get the native handle of.
dev	is the PI device that the native allocation will be resident on
nativeHandle	is the native handle of mem.

Definition at line 236 of file pi_cuda.cpp.

piextMemImageAllocate()

pi_result piextMemImageAllocate	(	pi_context	context,
		pi_device	device,
		pi_image_format *	image_format,
		pi_image_desc *	image_desc,
		pi_image_mem_handle *	ret_mem
	)

API to allocate memory for bindless images.

Parameters

context	is the pi_context
device	is the pi_device
flags	are extra flags to pass (currently unused)
image_format	format of the image (channel order and data type)
image_desc	image descriptor
ret_mem	is the returning memory handle to newly allocated memory

Definition at line 374 of file pi_cuda.cpp.

piextMemImageCopy()

pi_result piextMemImageCopy	(	pi_queue	command_queue,
		void *	dst_ptr,
		void *	src_ptr,
		const pi_image_format *	image_format,
		const pi_image_desc *	image_desc,
		const pi_image_copy_flags	flags,
		pi_image_offset	src_offset,
		pi_image_offset	dst_offset,
		pi_image_region	copy_extent,
		pi_image_region	host_extent,
		pi_uint32	num_events_in_wait_list,
		const pi_event *	event_wait_list,
		pi_event *	event
	)

API to copy image data Host to Device or Device to Host.

Parameters

queue	is the queue to submit to
dst_ptr	is the location the data will be copied to
src_ptr	is the data to be copied
image_format	format of the image (channel order and data type)
image_desc	image descriptor
flags	flags describing copy direction (H2D or D2H)
src_offset	is the offset into the source image/memory
dst_offset	is the offset into the destination image/memory
copy_extent	is the extent (region) of the image/memory to copy
host_extent	is the extent (region) of the memory on the host
num_events_in_wait_list	is the number of events in the wait list
event_wait_list	is the list of events to wait on before copying
event	is the returned event representing this operation

Definition at line 427 of file pi_cuda.cpp.

piextMemImageCreateWithNativeHandle()

pi_result piextMemImageCreateWithNativeHandle	(	pi_native_handle	nativeHandle,
		pi_context	context,
		bool	ownNativeHandle,
		const pi_image_format *	ImageFormat,
		const pi_image_desc *	ImageDesc,
		pi_mem *	img
	)

Creates PI image object from a native handle.

Parameters

nativeHandle	is the native handle to create PI image from.
context	The PI context of the memory allocation.
ownNativeHandle	Indicates if we own the native memory handle or it came from interop that asked to not transfer the ownership to SYCL RT.
ImageFormat	is the pi_image_format struct that specifies the image channnel order and channel data type that match what the nativeHandle uses
ImageDesc	is the pi_image_desc struct that specifies the image dimension, pitch, slice and other information about the nativeHandle
img	is the PI img created from the native handle.

Definition at line 264 of file pi_cuda.cpp.

piextMemImageFree()

pi_result piextMemImageFree	(	pi_context	context,
		pi_device	device,
		pi_image_mem_handle	memory_handle
	)

API to free memory for bindless images.

Parameters

context	is the pi_context
device	is the pi_device
memory_handle	is the handle to image memory to be freed

Definition at line 417 of file pi_cuda.cpp.

piextMemImageGetInfo()

pi_result piextMemImageGetInfo	(	const pi_image_mem_handle	mem_handle,
		pi_image_info	param_name,
		void *	param_value,
		size_t *	param_value_size_ret
	)

API to query an image memory handle for specific properties.

Parameters

mem_handle	is the handle to the image memory
param_name	is the queried info name
param_value	is the returned query value
param_value_size_ret	is the returned query value size

Definition at line 450 of file pi_cuda.cpp.

piextMemImportOpaqueFD()

pi_result piextMemImportOpaqueFD	(	pi_context	context,
		pi_device	device,
		size_t	size,
		int	file_descriptor,
		pi_interop_mem_handle *	ret_handle
	)

API to import external memory in the form of a file descriptor.

Parameters

context	is the pi_context
device	is the pi_device
size	is the size of the external memory
file_descriptor	is the file descriptor
ret_handle	is the returned interop memory handle to the external memory

Definition at line 459 of file pi_cuda.cpp.

piextMemMapExternalArray()

pi_result piextMemMapExternalArray	(	pi_context	context,
		pi_device	device,
		pi_image_format *	image_format,
		pi_image_desc *	image_desc,
		pi_interop_mem_handle	mem_handle,
		pi_image_mem_handle *	ret_mem
	)

API to map an interop memory handle to an image memory handle.

Parameters

context	is the pi_context
device	is the pi_device
image_format	format of the image (channel order and data type)
image_desc	image descriptor
mem_handle	is the interop memory handle to the external memory
ret_mem	is the returned image memory handle to the externally allocated memory

Definition at line 465 of file pi_cuda.cpp.

piextMemMipmapFree()

pi_result piextMemMipmapFree	(	pi_context	context,
		pi_device	device,
		pi_image_mem_handle	memory_handle
	)

API to free mipmap memory for bindless images.

Parameters

context	is the pi_context
device	is the pi_device
memory_handle	is the handle to image memory to be freed

Definition at line 422 of file pi_cuda.cpp.

piextMemMipmapGetLevel()

pi_result piextMemMipmapGetLevel	(	pi_context	context,
		pi_device	device,
		pi_image_mem_handle	mip_mem,
		unsigned int	level,
		pi_image_mem_handle *	ret_mem
	)

API to retrieve individual image from mipmap.

Parameters

context	is the pi_context
device	is the pi_device
mip_mem	is the memory handle to the mipmap
level	is the requested level of the mipmap
ret_mem	is the returning memory handle to the individual image

Definition at line 409 of file pi_cuda.cpp.

piextMemReleaseInterop()

pi_result piextMemReleaseInterop	(	pi_context	context,
		pi_device	device,
		pi_interop_mem_handle	memory_handle
	)

API to destroy interop memory.

Parameters

context	is the pi_context
device	is the pi_device
memory_handle	is the handle to interop memory to be freed

Definition at line 473 of file pi_cuda.cpp.

piextMemSampledImageCreate()

pi_result piextMemSampledImageCreate	(	pi_context	context,
		pi_device	device,
		pi_image_mem_handle	img_mem,
		pi_image_format *	image_format,
		pi_image_desc *	image_desc,
		pi_sampler	sampler,
		pi_mem *	ret_mem,
		pi_image_handle *	ret_handle
	)

API to create sampled bindless image handles.

Parameters

context	is the pi_context
device	is the pi_device
img_mem	is the handle to memory from which to create the image
image_format	format of the image (channel order and data type)
image_desc	image descriptor
sampler	is the pi_sampler
ret_mem	is the returning pi_mem image object
ret_handle	is the returning memory handle to newly allocated memory

Definition at line 391 of file pi_cuda.cpp.

piextMemSampledImageHandleDestroy()

pi_result piextMemSampledImageHandleDestroy	(	pi_context	context,
		pi_device	device,
		pi_image_handle	handle
	)

API to destroy bindless sampled image handles.

Parameters

context	is the pi_context
handle	is the image handle

Definition at line 445 of file pi_cuda.cpp.

piextMemUnsampledImageCreate()

pi_result piextMemUnsampledImageCreate	(	pi_context	context,
		pi_device	device,
		pi_image_mem_handle	img_mem,
		pi_image_format *	image_format,
		pi_image_desc *	image_desc,
		pi_mem *	ret_mem,
		pi_image_handle *	ret_handle
	)

API to create bindless image handles.

Parameters

context	is the pi_context
device	is the pi_device
img_mem	is the handle to memory from which to create the image
image_format	format of the image (channel order and data type)
image_desc	image descriptor
ret_mem	is the returning pi_mem image object
ret_handle	is the returning memory handle to newly allocated memory

Definition at line 383 of file pi_cuda.cpp.

piextMemUnsampledImageHandleDestroy()

pi_result piextMemUnsampledImageHandleDestroy	(	pi_context	context,
		pi_device	device,
		pi_image_handle	handle
	)

API to destroy bindless unsampled image handles.

Parameters

context	is the pi_context
device	is the pi_device
handle	is the image handle

Definition at line 440 of file pi_cuda.cpp.

piextPeerAccessGetInfo()

pi_result piextPeerAccessGetInfo	(	pi_device	command_device,
		pi_device	peer_device,
		pi_peer_attr	attr,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 1264 of file pi_cuda.cpp.

piextPlatformCreateWithNativeHandle()

pi_result piextPlatformCreateWithNativeHandle	(	pi_native_handle	nativeHandle,
		pi_platform *	platform
	)

Creates PI platform object from a native handle.

NOTE: The created PI object takes ownership of the native handle.

Parameters

nativeHandle	is the native handle to create PI device from.
platform	is the PI platform created from the native handle.

Definition at line 47 of file pi_cuda.cpp.

piextPlatformGetNativeHandle()

pi_result piextPlatformGetNativeHandle	(	pi_platform	platform,
		pi_native_handle *	nativeHandle
	)

Gets the native handle of a PI platform object.

Parameters

platform	is the PI platform to get the native handle of.
nativeHandle	is the native handle of platform.

Definition at line 42 of file pi_cuda.cpp.

piextPluginGetOpaqueData()

pi_result piextPluginGetOpaqueData	(	void *	opaque_data_param,
		void **	opaque_data_return
	)

API to get Plugin internal data, opaque to SYCL RT.

Some devices whose device code is compiled by the host compiler (e.g. CPU emulators) may use it to access some device code functionality implemented in/behind the plugin.

Parameters

opaque_data_param	- unspecified argument, interpretation is specific to a plugin
opaque_data_return	- placeholder for the returned opaque data.

Definition at line 1238 of file pi_cuda.cpp.

piextProgramCreateWithNativeHandle()

pi_result piextProgramCreateWithNativeHandle	(	pi_native_handle	nativeHandle,
		pi_context	context,
		bool	pluginOwnsNativeHandle,
		pi_program *	program
	)

Creates PI program object from a native handle.

NOTE: The created PI object takes ownership of the native handle.

Parameters

nativeHandle	is the native handle to create PI program from.
context	is the PI context of the program.
pluginOwnsNativeHandle	Indicates whether the created PI object should take ownership of the native handle.
program	is the PI program created from the native handle.

Definition at line 333 of file pi_cuda.cpp.

piextProgramGetNativeHandle()

pi_result piextProgramGetNativeHandle	(	pi_program	program,
		pi_native_handle *	nativeHandle
	)

Gets the native handle of a PI program object.

Parameters

program	is the PI program to get the native handle of.
nativeHandle	is the native handle of program.

Definition at line 328 of file pi_cuda.cpp.

piextProgramSetSpecializationConstant()

pi_result piextProgramSetSpecializationConstant	(	pi_program	prog,
		pi_uint32	spec_id,
		size_t	spec_size,
		const void *	spec_value
	)

Sets a specialization constant to a specific value.

Note: Only used when specialization constants are natively supported (SPIR-V binaries), and not when they are emulated (AOT binaries).

Parameters

prog	the program object which will use the value
spec_id	integer ID of the constant
spec_size	size of the value
spec_value	bytes of the value

Definition at line 1060 of file pi_cuda.cpp.

piextQueueCreate()

pi_result piextQueueCreate	(	pi_context	context,
		pi_device	device,
		pi_queue_properties *	properties,
		pi_queue *	queue
	)

Parameters

properties

points to a zero-terminated array of extra data describing desired queue properties. Format is {[PROPERTY[, property-specific elements of data]*,]* 0}

Definition at line 167 of file pi_cuda.cpp.

Referenced by piQueueCreate().

piextQueueCreateWithNativeHandle()

pi_result piextQueueCreateWithNativeHandle	(	pi_native_handle	nativeHandle,
		int32_t	nativeHandleDesc,
		pi_context	context,
		pi_device	device,
		bool	pluginOwnsNativeHandle,
		pi_queue_properties *	Properties,
		pi_queue *	queue
	)

Creates PI queue object from a native handle.

NOTE: The created PI object takes ownership of the native handle.

Parameters

nativeHandle	is the native handle to create PI queue from.
nativeHandleDesc	provides additional properties of the native handle.
context	is the PI context of the queue.
device	is the PI device associated with the native device used when creating the native queue. This parameter is optional but some backends may fail to create the right PI queue if omitted.
pluginOwnsNativeHandle	Indicates whether the created PI object should take ownership of the native handle.
Properties	holds queue properties.
queue	is the PI queue created from the native handle.

Definition at line 198 of file pi_cuda.cpp.

piextQueueGetNativeHandle()

pi_result piextQueueGetNativeHandle	(	pi_queue	queue,
		pi_native_handle *	nativeHandle,
		int32_t *	nativeHandleDesc
	)

Gets the native handle of a PI queue object.

Parameters

queue	is the PI queue to get the native handle of.
nativeHandle	is the native handle of queue or commandlist.
nativeHandleDesc	provides additional properties of the native handle.

Definition at line 190 of file pi_cuda.cpp.

piextSignalExternalSemaphore()

pi_result piextSignalExternalSemaphore	(	pi_queue	command_queue,
		pi_interop_semaphore_handle	sem_handle,
		pi_uint32	num_events_in_wait_list,
		const pi_event *	event_wait_list,
		pi_event *	event
	)

API to instruct the queue to signal the external semaphore handle once all previous commands have completed execution.

Parameters

command_queue	is the queue instructed to signal
sem_handle	is the interop semaphore handle to signal
num_events_in_wait_list	is the number of events in the wait list
event_wait_list	is the list of events to wait on before this operation
event	is the returned event representing this operation

Definition at line 500 of file pi_cuda.cpp.

piextUSMDeviceAlloc()

pi_result piextUSMDeviceAlloc	(	void **	result_ptr,
		pi_context	context,
		pi_device	device,
		pi_usm_mem_properties *	properties,
		size_t	size,
		pi_uint32	alignment
	)

Allocates device memory.

Parameters

result_ptr	contains the allocated memory
context	is the pi_context
device	is the device the memory will be allocated on
properties	are optional allocation properties
size	is the size of the allocation
alignment	is the desired alignment of the allocation

Definition at line 860 of file pi_cuda.cpp.

piextUSMEnqueueFill2D()

pi_result piextUSMEnqueueFill2D	(	pi_queue	queue,
		void *	ptr,
		size_t	pitch,
		size_t	pattern_size,
		const void *	pattern,
		size_t	width,
		size_t	height,
		pi_uint32	num_events_in_waitlist,
		const pi_event *	events_waitlist,
		pi_event *	event
	)

USM 2D fill API.

Parameters

queue	is the queue to submit to
ptr	is the ptr to fill
pitch	is the total width of the destination memory including padding
pattern	is a pointer with the bytes of the pattern to set
pattern_size	is the size in bytes of the pattern
width	is width in bytes of each row to fill
height	is height the columns to fill
num_events_in_waitlist	is the number of events to wait on
events_waitlist	is an array of events to wait on
event	is the event that represents this operation

Definition at line 941 of file pi_cuda.cpp.

piextUSMEnqueueMemAdvise()

pi_result piextUSMEnqueueMemAdvise	(	pi_queue	queue,
		const void *	ptr,
		size_t	length,
		pi_mem_advice	advice,
		pi_event *	event
	)

USM Memadvise API.

Parameters

queue	is the queue to submit to
ptr	is the data to be advised
length	is the size in bytes of the memory to advise
advice	is device specific advice
event	is the event that represents this operation

Definition at line 934 of file pi_cuda.cpp.

piextUSMEnqueueMemcpy()

pi_result piextUSMEnqueueMemcpy	(	pi_queue	queue,
		pi_bool	blocking,
		void *	dst_ptr,
		const void *	src_ptr,
		size_t	size,
		pi_uint32	num_events_in_waitlist,
		const pi_event *	events_waitlist,
		pi_event *	event
	)

USM Memcpy API.

Parameters

queue	is the queue to submit to
blocking	is whether this operation should block the host
src_ptr	is the data to be copied
dst_ptr	is the location the data will be copied
size	is number of bytes to copy
num_events_in_waitlist	is the number of events to wait on
events_waitlist	is an array of events to wait on
event	is the event that represents this operation

Definition at line 913 of file pi_cuda.cpp.

piextUSMEnqueueMemcpy2D()

pi_result piextUSMEnqueueMemcpy2D	(	pi_queue	queue,
		pi_bool	blocking,
		void *	dst_ptr,
		size_t	dst_pitch,
		const void *	src_ptr,
		size_t	src_pitch,
		size_t	width,
		size_t	height,
		pi_uint32	num_events_in_waitlist,
		const pi_event *	events_waitlist,
		pi_event *	event
	)

USM 2D Memcpy API.

Parameters

queue	is the queue to submit to
blocking	is whether this operation should block the host
dst_ptr	is the location the data will be copied
dst_pitch	is the total width of the destination memory including padding
src_ptr	is the data to be copied
src_pitch	is the total width of the source memory including padding
width	is width in bytes of each row to be copied
height	is height the columns to be copied
num_events_in_waitlist	is the number of events to wait on
events_waitlist	is an array of events to wait on
event	is the event that represents this operation

Definition at line 966 of file pi_cuda.cpp.

piextUSMEnqueueMemset()

pi_result piextUSMEnqueueMemset	(	pi_queue	queue,
		void *	ptr,
		pi_int32	value,
		size_t	count,
		pi_uint32	num_events_in_waitlist,
		const pi_event *	events_waitlist,
		pi_event *	event
	)

USM Memset API.

Parameters

queue	is the queue to submit to
ptr	is the ptr to memset
value	is value to set. It is interpreted as an 8-bit value and the upper 24 bits are ignored
count	is the size in bytes to memset
num_events_in_waitlist	is the number of events to wait on
events_waitlist	is an array of events to wait on
event	is the event that represents this operation

Definition at line 905 of file pi_cuda.cpp.

piextUSMEnqueueMemset2D()

pi_result piextUSMEnqueueMemset2D	(	pi_queue	queue,
		void *	ptr,
		size_t	pitch,
		int	value,
		size_t	width,
		size_t	height,
		pi_uint32	num_events_in_waitlist,
		const pi_event *	events_waitlist,
		pi_event *	event
	)

USM 2D Memset API.

Parameters

queue	is the queue to submit to
ptr	is the ptr to fill
pitch	is the total width of the destination memory including padding
value	the value to fill into the region in
ptr
width	is width in bytes of each row to fill
height	is height the columns to fill
num_events_in_waitlist	is the number of events to wait on
events_waitlist	is an array of events to wait on
event	is the event that represents this operation

Definition at line 954 of file pi_cuda.cpp.

piextUSMEnqueuePrefetch()

pi_result piextUSMEnqueuePrefetch	(	pi_queue	queue,
		const void *	ptr,
		size_t	size,
		pi_usm_migration_flags	flags,
		pi_uint32	num_events_in_waitlist,
		const pi_event *	events_waitlist,
		pi_event *	event
	)

Hint to migrate memory to the device.

Parameters

queue	is the queue to submit to
ptr	points to the memory to migrate
size	is the number of bytes to migrate
flags	is a bitfield used to specify memory migration options
num_events_in_waitlist	is the number of events to wait on
events_waitlist	is an array of events to wait on
event	is the event that represents this operation

Definition at line 924 of file pi_cuda.cpp.

piextUSMFree()

pi_result piextUSMFree	(	pi_context	context,
		void *	ptr
	)

Indicates that the allocated USM memory is no longer needed on the runtime side.

The actual freeing of the memory may be done in a blocking or deferred manner, e.g. to avoid issues with indirect memory access from kernels.

Parameters

context	is the pi_context of the allocation
ptr	is the memory to be freed

Definition at line 895 of file pi_cuda.cpp.

piextUSMGetMemAllocInfo()

pi_result piextUSMGetMemAllocInfo	(	pi_context	context,
		const void *	ptr,
		pi_mem_alloc_info	param_name,
		size_t	param_value_size,
		void *	param_value,
		size_t *	param_value_size_ret
	)

API to query information about USM allocated pointers Valid Queries: PI_MEM_ALLOC_TYPE returns host/device/shared pi_host_usm value PI_MEM_ALLOC_BASE_PTR returns the base ptr of an allocation if the queried pointer fell inside an allocation.

Result must fit in void * PI_MEM_ALLOC_SIZE returns how big the queried pointer's allocation is in bytes. Result is a size_t. PI_MEM_ALLOC_DEVICE returns the pi_device this was allocated against

Parameters

context	is the pi_context
ptr	is the pointer to query
param_name	is the type of query to perform
param_value_size	is the size of the result in bytes
param_value	is the result
param_value_size_ret	is how many bytes were written

Definition at line 977 of file pi_cuda.cpp.

piextUSMHostAlloc()

pi_result piextUSMHostAlloc	(	void **	result_ptr,
		pi_context	context,
		pi_usm_mem_properties *	properties,
		size_t	size,
		pi_uint32	alignment
	)

Allocates host memory accessible by the device.

Parameters

result_ptr	contains the allocated memory
context	is the pi_context
properties	are optional allocation properties
size	is the size of the allocation
alignment	is the desired alignment of the allocation

Definition at line 888 of file pi_cuda.cpp.

piextUSMImport()

pi_result piextUSMImport	(	const void *	ptr,
		size_t	size,
		pi_context	context
	)

Import host system memory into USM.

Parameters

ptr	start address of memory range to import
size	is the number of bytes to import
context	is the pi_context

Definition at line 985 of file pi_cuda.cpp.

piextUSMPitchedAlloc()

pi_result piextUSMPitchedAlloc	(	void **	result_ptr,
		size_t *	result_pitch,
		pi_context	context,
		pi_device	device,
		pi_usm_mem_properties *	properties,
		size_t	width_in_bytes,
		size_t	height,
		unsigned int	element_size_bytes
	)

Allocates memory accessible on device.

Parameters

result_ptr	contains the allocated memory
result_pitch	contains the returned memory pitch
context	is the pi_context
device	is the device the memory will be allocated on
properties	are optional allocation properties
width_in_bytes	is the width of the allocation in bytes
height	is the height of the allocation in rows
element_size_bytes	is the size in bytes of an element in the allocation

Definition at line 878 of file pi_cuda.cpp.

piextUSMRelease()

pi_result piextUSMRelease	(	const void *	ptr,
		pi_context	context
	)

Release host system memory from USM.

Parameters

ptr	start address of imported memory range
context	is the pi_context

Definition at line 989 of file pi_cuda.cpp.

piextUSMSharedAlloc()

pi_result piextUSMSharedAlloc	(	void **	result_ptr,
		pi_context	context,
		pi_device	device,
		pi_usm_mem_properties *	properties,
		size_t	size,
		pi_uint32	alignment
	)

Allocates memory accessible on both host and device.

Parameters

result_ptr	contains the allocated memory
context	is the pi_context
device	is the device the memory will be allocated on
properties	are optional allocation properties
size	is the size of the allocation
alignment	is the desired alignment of the allocation

Definition at line 869 of file pi_cuda.cpp.

piextWaitExternalSemaphore()

pi_result piextWaitExternalSemaphore	(	pi_queue	command_queue,
		pi_interop_semaphore_handle	sem_handle,
		pi_uint32	num_events_in_wait_list,
		const pi_event *	event_wait_list,
		pi_event *	event
	)

API to instruct the queue with a non-blocking wait on an external semaphore.

Parameters

command_queue	is the queue instructed to wait
sem_handle	is the interop semaphore handle
num_events_in_wait_list	is the number of events in the wait list
event_wait_list	is the list of events to wait on before this operation
event	is the returned event representing this operation

Definition at line 492 of file pi_cuda.cpp.

piGetDeviceAndHostTimer()

pi_result piGetDeviceAndHostTimer	(	pi_device	Device,
		uint64_t *	DeviceTime,
		uint64_t *	HostTime
	)

Queries device for it's global timestamp in nanoseconds, and updates HostTime with the value of the host timer at the closest possible point in time to that at which DeviceTime was returned.

Parameters

Device	device to query for timestamp
DeviceTime	pointer to store device timestamp in nanoseconds. Optional argument, can be nullptr
HostTime	pointer to store host timestamp in nanoseconds. Optional argurment, can be nullptr in which case timestamp will not be written

Definition at line 1247 of file pi_cuda.cpp.

piKernelCreate()

pi_result piKernelCreate	(	pi_program	Program,
		const char *	KernelName,
		pi_kernel *	RetKernel
	)

Definition at line 341 of file pi_cuda.cpp.

piKernelGetGroupInfo()

pi_result piKernelGetGroupInfo	(	pi_kernel	Kernel,
		pi_device	Device,
		pi_kernel_group_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 508 of file pi_cuda.cpp.

piKernelGetInfo()

pi_result piKernelGetInfo	(	pi_kernel	Kernel,
		pi_kernel_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 366 of file pi_cuda.cpp.

piKernelGetSubGroupInfo()

pi_result piKernelGetSubGroupInfo	(	pi_kernel	kernel,
		pi_device	device,
		pi_kernel_sub_group_info	param_name,
		size_t	input_value_size,
		const void *	input_value,
		size_t	param_value_size,
		void *	param_value,
		size_t *	param_value_size_ret
	)

API to query information from the sub-group from a kernel.

Parameters

kernel	is the pi_kernel to query
device	is the device the kernel is executed on
param_name	is a pi_kernel_sub_group_info enum value that specifies the informtation queried for.
input_value_size	is the size of input value passed in ptr input_value param
input_value	is the ptr to the input value passed.
param_value_size	is the size of the value in bytes.
param_value	is a pointer to the value to set.
param_value_size_ret	is a pointer to return the size of data in param_value ptr.

All queries expect a return of 4 bytes in param_value_size, param_value_size_ret, and a uint32_t value should to be written in param_value ptr. Note: This behaviour differs from OpenCL. OpenCL returns size_t.

Definition at line 516 of file pi_cuda.cpp.

piKernelRelease()

pi_result piKernelRelease

(

pi_kernel

Kernel

)

Definition at line 531 of file pi_cuda.cpp.

piKernelRetain()

pi_result piKernelRetain

(

pi_kernel

Kernel

)

Definition at line 526 of file pi_cuda.cpp.

piKernelSetArg()

pi_result piKernelSetArg	(	pi_kernel	Kernel,
		pi_uint32	ArgIndex,
		size_t	ArgSize,
		const void *	ArgValue
	)

Definition at line 347 of file pi_cuda.cpp.

piKernelSetExecInfo()

pi_result piKernelSetExecInfo	(	pi_kernel	kernel,
		pi_kernel_exec_info	value_name,
		size_t	param_value_size,
		const void *	param_value
	)

API to set attributes controlling kernel execution.

Parameters

kernel	is the pi kernel to execute
param_name	is a pi_kernel_exec_info value that specifies the info passed to the kernel
param_value_size	is the size of the value in bytes
param_value	is a pointer to the value to set for the kernel

If param_name is PI_USM_INDIRECT_ACCESS, the value will be a ptr to the pi_bool value PI_TRUE If param_name is PI_USM_PTRS, the value will be an array of ptrs

Definition at line 1053 of file pi_cuda.cpp.

piMemBufferCreate()

pi_result piMemBufferCreate	(	pi_context	Context,
		pi_mem_flags	Flags,
		size_t	Size,
		void *	HostPtr,
		pi_mem *	RetMem,
		const pi_mem_properties *	properties
	)

Definition at line 210 of file pi_cuda.cpp.

piMemBufferPartition()

pi_result piMemBufferPartition	(	pi_mem	Buffer,
		pi_mem_flags	Flags,
		pi_buffer_create_type	BufferCreateType,
		void *	BufferCreateInfo,
		pi_mem *	RetMem
	)

Definition at line 836 of file pi_cuda.cpp.

piMemGetInfo()

pi_result piMemGetInfo	(	pi_mem	Mem,
		pi_mem_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 217 of file pi_cuda.cpp.

piMemImageCreate()

pi_result piMemImageCreate	(	pi_context	Context,
		pi_mem_flags	Flags,
		const pi_image_format *	ImageFormat,
		const pi_image_desc *	ImageDesc,
		void *	HostPtr,
		pi_mem *	RetImage
	)

Definition at line 227 of file pi_cuda.cpp.

piMemImageGetInfo()

pi_result piMemImageGetInfo	(	pi_mem	Image,
		pi_image_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 781 of file pi_cuda.cpp.

piMemRelease()

pi_result piMemRelease

(

pi_mem

Mem

)

Definition at line 225 of file pi_cuda.cpp.

piMemRetain()

pi_result piMemRetain

(

pi_mem

Mem

)

Definition at line 223 of file pi_cuda.cpp.

piPlatformGetInfo()

pi_result piPlatformGetInfo	(	pi_platform	Platform,
		pi_platform_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 35 of file pi_cuda.cpp.

piPlatformsGet()

pi_result piPlatformsGet	(	pi_uint32	NumEntries,
		pi_platform *	Platforms,
		pi_uint32 *	NumPlatforms
	)

Definition at line 30 of file pi_cuda.cpp.

piPluginGetBackendOption()

pi_result piPluginGetBackendOption	(	pi_platform	platform,
		const char *	frontend_option,
		const char **	backend_option
	)

API to get backend specific option.

Parameters

frontend_option	is a string that contains frontend option.
backend_option	is used to return the backend option corresponding to frontend option.

Returns

PI_SUCCESS is returned for valid frontend_option. If a valid backend option is not available, an empty string is returned.

Definition at line 56 of file pi_cuda.cpp.

piPluginGetLastError()

pi_result piPluginGetLastError

(

char **

message

)

API to get Plugin specific warning and error messages.

Parameters

message

is a returned address to the first element in the message the plugin owns the error message string. The string is thread-local. As a result, different threads may return different errors. A message is overwritten by the following error or warning that is produced within the given thread. The memory is cleaned up at the end of the thread's lifetime.

Returns

PI_SUCCESS if plugin is indicating non-fatal warning. Any other error code indicates that plugin considers this to be a fatal error and the Returns the global timestamp from

Parameters

device,and

syncronized host timestamp

Definition at line 52 of file pi_cuda.cpp.

piPluginInit()

pi_result piPluginInit

(

pi_plugin *

PluginInit

)

Definition at line 1275 of file pi_cuda.cpp.

piProgramBuild()

pi_result piProgramBuild	(	pi_program	Program,
		pi_uint32	NumDevices,
		const pi_device *	DeviceList,
		const char *	Options,
		void(*)(pi_program Program, void *UserData)	PFnNotify,
		void *	UserData
	)

Definition at line 302 of file pi_cuda.cpp.

References pi2ur::piProgramBuild().

piProgramCompile()

pi_result piProgramCompile	(	pi_program	Program,
		pi_uint32	NumDevices,
		const pi_device *	DeviceList,
		const char *	Options,
		pi_uint32	NumInputHeaders,
		const pi_program *	InputHeaders,
		const char **	HeaderIncludeNames,
		void(*)(pi_program Program, void *UserData)	PFnNotify,
		void *	UserData
	)

Definition at line 291 of file pi_cuda.cpp.

References pi2ur::piProgramCompile().

piProgramCreate()

pi_result piProgramCreate	(	pi_context	Context,
		const void *	ILBytes,
		size_t	Length,
		pi_program *	Program
	)

Definition at line 248 of file pi_cuda.cpp.

piProgramCreateWithBinary()

pi_result piProgramCreateWithBinary	(	pi_context	context,
		pi_uint32	num_devices,
		const pi_device *	device_list,
		const size_t *	lengths,
		const unsigned char **	binaries,
		size_t	num_metadata_entries,
		const pi_device_binary_property *	metadata,
		pi_int32 *	binary_status,
		pi_program *	ret_program
	)

Creates a PI program for a context and loads the given binary into it.

Parameters

context	is the PI context to associate the program with.
num_devices	is the number of devices in device_list.
device_list	is a pointer to a list of devices. These devices must all be in context.
lengths	is an array of sizes in bytes of the binary in binaries.
binaries	is a pointer to a list of program binaries.
num_metadata_entries	is the number of metadata entries in metadata.
metadata	is a pointer to a list of program metadata entries. The use of metadata entries is backend-defined.
binary_status	returns whether the program binary was loaded succesfully or not, for each device in device_list. binary_status is ignored if it is null and otherwise it must be an array of num_devices elements.
ret_program	is the PI program created from the program binaries.

Definition at line 253 of file pi_cuda.cpp.

piProgramGetBuildInfo()

pi_result piProgramGetBuildInfo	(	pi_program	Program,
		pi_device	Device,
		pi_program_build_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 310 of file pi_cuda.cpp.

piProgramGetInfo()

pi_result piProgramGetInfo	(	pi_program	Program,
		pi_program_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 272 of file pi_cuda.cpp.

piProgramLink()

pi_result piProgramLink	(	pi_context	Context,
		pi_uint32	NumDevices,
		const pi_device *	DeviceList,
		const char *	Options,
		pi_uint32	NumInputPrograms,
		const pi_program *	InputPrograms,
		void(*)(pi_program Program, void *UserData)	PFnNotify,
		void *	UserData,
		pi_program *	RetProgram
	)

Definition at line 280 of file pi_cuda.cpp.

References pi2ur::piProgramLink().

piProgramRelease()

pi_result piProgramRelease

(

pi_program

Program

)

Definition at line 324 of file pi_cuda.cpp.

piProgramRetain()

pi_result piProgramRetain

(

pi_program

Program

)

Definition at line 320 of file pi_cuda.cpp.

piQueueCreate()

pi_result piQueueCreate	(	pi_context	Context,
		pi_device	Device,
		pi_queue_properties	Flags,
		pi_queue *	Queue
	)

Definition at line 161 of file pi_cuda.cpp.

piQueueFinish()

pi_result piQueueFinish

(

pi_queue

Queue

)

Definition at line 186 of file pi_cuda.cpp.

piQueueFlush()

pi_result piQueueFlush

(

pi_queue

Queue

)

Definition at line 188 of file pi_cuda.cpp.

piQueueGetInfo()

pi_result piQueueGetInfo	(	pi_queue	Queue,
		pi_queue_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 172 of file pi_cuda.cpp.

piQueueRelease()

pi_result piQueueRelease

(

pi_queue

Queue

)

Definition at line 182 of file pi_cuda.cpp.

piQueueRetain()

pi_result piQueueRetain

(

pi_queue

Queue

)

Definition at line 180 of file pi_cuda.cpp.

piSamplerCreate()

pi_result piSamplerCreate	(	pi_context	Context,
		const pi_sampler_properties *	SamplerProperties,
		pi_sampler *	RetSampler
	)

Definition at line 635 of file pi_cuda.cpp.

piSamplerGetInfo()

pi_result piSamplerGetInfo	(	pi_sampler	Sampler,
		pi_sampler_info	ParamName,
		size_t	ParamValueSize,
		void *	ParamValue,
		size_t *	ParamValueSizeRet
	)

Definition at line 641 of file pi_cuda.cpp.

piSamplerRelease()

pi_result piSamplerRelease

(

pi_sampler

Sampler

)

Definition at line 653 of file pi_cuda.cpp.

piSamplerRetain()

pi_result piSamplerRetain

(

pi_sampler

Sampler

)

Definition at line 649 of file pi_cuda.cpp.

piTearDown()

pi_result piTearDown

(

void *

PluginParameter

)

API to notify that the plugin should clean up its resources.

No PI calls should be made until the next piPluginInit call.

Parameters

PluginParameter

placeholder for future use, currenly not used.

Definition at line 1243 of file pi_cuda.cpp.

Variable Documentation

SupportedVersion

const char SupportedVersion[] = _PI_CUDA_PLUGIN_VERSION_STRING

Definition at line 1273 of file pi_cuda.cpp.

Referenced by piPluginInit().