Realm
A distributed, event-based tasking library
Loading...
Searching...
No Matches
Realm::Cuda Namespace Reference

Classes

struct  GPUInfo
class  GPUContextManager
class  GPUCompletionNotification
class  GPUWorkFence
class  GPUWorkStart
class  GPUStream
class  GPUWorker
class  GPUEventPool
class  ContextSynchronizer
class  GPU
class  AutoGPUContext
class  GPUProcessor
class  CudaDeviceMemoryInfo
class  GPUFBMemory
class  GPUDynamicFBMemory
class  GPUZCMemory
class  GPUFBIBMemory
class  GPUCompletionEvent
class  GPURequest
class  GPUIndirectTransferCompletion
class  GPUTransferCompletion
class  MemSpecificCudaArray
class  AddressInfoCudaArray
class  GPUXferDes
class  GPUIndirectXferDes
class  GPUIndirectChannel
class  GPUIndirectRemoteChannelInfo
class  GPUIndirectRemoteChannel
class  GPUChannel
class  GPURemoteChannelInfo
class  GPURemoteChannel
class  GPUfillXferDes
class  GPUfillChannel
struct  KernelVariantDesc
class  GPUreduceXferDes
class  GPUreduceChannel
class  GPUreduceRemoteChannelInfo
class  GPUreduceRemoteChannel
struct  CudaIpcImportRequest
class  GPUReplHeapListener
class  GPUAllocation
 Class for managing the lifetime of a given gpu allocation. As instances of this class own an underlying resource they are not copyable and must be std::move'd (thus invalidating the original variable) or references made. More...
struct  AffineSubRect
struct  AffineCopyPair
struct  AffineFillRect
struct  AffineFillInfo
struct  AffineLargeFillInfo
struct  AffineCopyInfo
struct  MemcpyTransposeInfo
struct  MemcpyIndirectInfo
class  CudaModuleConfig
struct  CudaRedOpDesc
class  CudaModule
struct  MemReducInfo
struct  AffineReducSubRect
struct  AffineReducPair
struct  AffineReducInfo
struct  FatBin
struct  RegisteredModule
struct  RegisteredFunction
struct  RegisteredVariable
class  GlobalRegistrations

Typedefs

typedef enum Realm::Cuda::nvmlIntNvLinkDeviceType_enum nvmlIntNvLinkDeviceType_t
typedef void(* StreamAwareTaskFuncPtr) (const void *args, size_t arglen, const void *user_data, size_t user_data_len, Processor proc, CUstream_st *stream)
typedef char Uuid[UUID_SIZE]

Enumerations

enum  GPUMemcpyKind { GPU_MEMCPY_HOST_TO_DEVICE , GPU_MEMCPY_DEVICE_TO_HOST , GPU_MEMCPY_DEVICE_TO_DEVICE , GPU_MEMCPY_PEER_TO_PEER }
enum  nvmlIntNvLinkDeviceType_enum { NVML_NVLINK_DEVICE_TYPE_GPU = 0x00 , NVML_NVLINK_DEVICE_TYPE_IBMNPU = 0x01 , NVML_NVLINK_DEVICE_TYPE_SWITCH = 0x02 , NVML_NVLINK_DEVICE_TYPE_UNKNOWN = 0xFF }

Functions

CUresult cuGetProcAddress (const char *, void **, int, int)
CUresult cuCtxRecordEvent (CUcontext hctx, CUevent event)
 CUDA_DRIVER_APIS (DECL_FNPTR_EXTERN)
nvmlReturn_t nvmlDeviceGetNvLinkRemoteDeviceType (nvmlDevice_t device, unsigned int link, nvmlIntNvLinkDeviceType_t *pNvLinkDeviceType)
REALM_PUBLIC_API CUstream_st * get_task_cuda_stream ()
REALM_PUBLIC_API void set_task_ctxsync_required (bool is_required)
REALM_PUBLIC_API bool get_cuda_device_uuid (Processor p, Uuid *uuid)
REALM_PUBLIC_API bool get_cuda_device_id (Processor p, int *device)

Variables

CudaModule * cuda_module_singleton
bool cudart_hijack_active
int cudart_hijack_nongpu_sync

Typedef Documentation

◆ nvmlIntNvLinkDeviceType_t

◆ StreamAwareTaskFuncPtr

typedef void(* Realm::Cuda::StreamAwareTaskFuncPtr) (const void *args, size_t arglen, const void *user_data, size_t user_data_len, Processor proc, CUstream_st *stream)

◆ Uuid

typedef char Realm::Cuda::Uuid[UUID_SIZE]

Enumeration Type Documentation

◆ GPUMemcpyKind

Enumerator
GPU_MEMCPY_HOST_TO_DEVICE 
GPU_MEMCPY_DEVICE_TO_HOST 
GPU_MEMCPY_DEVICE_TO_DEVICE 
GPU_MEMCPY_PEER_TO_PEER 

◆ nvmlIntNvLinkDeviceType_enum

Enumerator
NVML_NVLINK_DEVICE_TYPE_GPU 
NVML_NVLINK_DEVICE_TYPE_IBMNPU 
NVML_NVLINK_DEVICE_TYPE_SWITCH 
NVML_NVLINK_DEVICE_TYPE_UNKNOWN 

Function Documentation

◆ cuCtxRecordEvent()

CUresult Realm::Cuda::cuCtxRecordEvent ( CUcontext hctx,
CUevent event )

◆ CUDA_DRIVER_APIS()

Realm::Cuda::CUDA_DRIVER_APIS ( DECL_FNPTR_EXTERN )

◆ cuGetProcAddress()

CUresult Realm::Cuda::cuGetProcAddress ( const char * ,
void ** ,
int ,
int  )

◆ get_cuda_device_id()

REALM_PUBLIC_API bool Realm::Cuda::get_cuda_device_id ( Processor p,
int * device )

◆ get_cuda_device_uuid()

REALM_PUBLIC_API bool Realm::Cuda::get_cuda_device_uuid ( Processor p,
Uuid * uuid )

◆ get_task_cuda_stream()

REALM_PUBLIC_API CUstream_st * Realm::Cuda::get_task_cuda_stream ( )

◆ nvmlDeviceGetNvLinkRemoteDeviceType()

nvmlReturn_t Realm::Cuda::nvmlDeviceGetNvLinkRemoteDeviceType ( nvmlDevice_t device,
unsigned int link,
nvmlIntNvLinkDeviceType_t * pNvLinkDeviceType )

◆ set_task_ctxsync_required()

REALM_PUBLIC_API void Realm::Cuda::set_task_ctxsync_required ( bool is_required)

Variable Documentation

◆ cuda_module_singleton

CudaModule* Realm::Cuda::cuda_module_singleton
extern

◆ cudart_hijack_active

bool Realm::Cuda::cudart_hijack_active
extern

◆ cudart_hijack_nongpu_sync

int Realm::Cuda::cudart_hijack_nongpu_sync
extern