bindings

The cupti.cupti module exposes 1:1 Python bindings for the CUPTI C API. For most users the higher-level cupti.pm_sampling and cupti.profiler_host modules are the recommended entry points; the bindings on this page are intended for the Activity and Callback APIs (which do not yet have a pythonic wrapper), and as raw-access fallbacks for PM Sampling and Profiler Host.

Known limitations

• The members of the inner struct/union classes (_py_anon_pod*) are not adequately documented. Refer to the CUPTI C documentation for member-level details.

• The kind member of Python classes has type int. Use cupti.cupti.ActivityKind to interpret its value.

• Some parts of the API still cite C enums / data structures by their C names instead of mapping them to their Python counterparts.

Activity API

Functions

cupti.cupti.activity_configure_unified_memory_counter(config: int, count: int)

Set Unified Memory Counter configuration.

Parameters:

• config (intptr_t) – A pointer to CUpti_ActivityUnifiedMemoryCounterConfig structures containing Unified Memory counter configuration.

• count (uint32_t) – Number of Unified Memory counter configuration structures.

See also

cuptiActivityConfigureUnifiedMemoryCounter

cupti.cupti.activity_disable(kind: int)

Disable collection of a specific kind of activity record.

Parameters:

kind (ActivityKind) – The kind of activity record to stop collecting.

See also

cuptiActivityDisable

cupti.cupti.activity_disable_context(context: int, kind: int)

Disable collection of a specific kind of activity record for a context.

Parameters:

• context (intptr_t) – The context for which activity is to be disabled.

• kind (ActivityKind) – The kind of activity record to stop collecting.

See also

cuptiActivityDisableContext

cupti.cupti.activity_enable(kind: int)

Enable collection of a specific kind of activity record.

Parameters:

kind (ActivityKind) – The kind of activity record to collect.

See also

cuptiActivityEnable

cupti.cupti.activity_enable_all_sync_records(enable: int)

Enables collecting records for all synchronization operations.

Parameters:

enable (uint8_t) – is a boolean, denoting whether to enable or disable the collection of all CUDA event query and stream query records.

See also

cuptiActivityEnableAllSyncRecords

cupti.cupti.activity_enable_allocation_source(enable: int)

Enables tracking the source library for memory allocation requests.

Parameters:

enable (uint8_t) – is a boolean, denoting whether the source library of the memory allocation request needs to be tracked.

See also

cuptiActivityEnableAllocationSource

cupti.cupti.activity_enable_and_dump(kind: int)

Enable collection of a specific kind of activity record. For certain activity kinds it dumps existing records.

Parameters:

kind (ActivityKind) – The kind of activity record to collect.

See also

cuptiActivityEnableAndDump

cupti.cupti.activity_enable_context(context: int, kind: int)

Enable collection of a specific kind of activity record for a context.

Parameters:

• context (intptr_t) – The context for which activity is to be enabled.

• kind (ActivityKind) – The kind of activity record to collect.

See also

cuptiActivityEnableContext

cupti.cupti.activity_enable_cuda_event_device_timestamps(enable: int)

Enable/Disable collecting device timestamp for CUPTI_ACTIVITY_KIND_CUDA_EVENT record.

Parameters:

enable (uint8_t) – is a boolean, denoting whether to enable or disable the collection of CUDA event device timestamps.

See also

cuptiActivityEnableCudaEventDeviceTimestamps

cupti.cupti.activity_enable_device_graph(enable: int)

Controls the collection of records for device launched graphs.

Parameters:

enable (uint8_t) – is a boolean, denoting whether these records should be collected.

See also

cuptiActivityEnableDeviceGraph

cupti.cupti.activity_enable_driver_api(cbid: int, enable: int)

Controls the collection of activity records for specific CUDA Driver APIs.

Parameters:

• cbid (uint32_t) – callback id of the CUDA Driver API. This can be found in the header cupti_driver_cbid.h.

• enable (uint8_t) – is a boolean, denoting whether to enable or disable the collection.

See also

cuptiActivityEnableDriverApi

cupti.cupti.activity_enable_hw_trace(enable: int)

Enables CUDA kernel timestamp collection via Hardware Event System (HES).

Parameters:

enable (uint8_t) – is a boolean flag to enable (true) HES-based timestamp collection.

See also

cuptiActivityEnableHWTrace

cupti.cupti.activity_enable_latency_timestamps(enable: int)

Controls the collection of queued and submitted timestamps for kernels.

Parameters:

enable (uint8_t) – is a boolean, denoting whether these timestamps should be collected.

See also

cuptiActivityEnableLatencyTimestamps

cupti.cupti.activity_enable_launch_attributes(enable: int)

Controls the collection of launch attributes for kernels.

Parameters:

enable (uint8_t) – is a boolean denoting whether these launch attributes should be collected.

See also

cuptiActivityEnableLaunchAttributes

cupti.cupti.activity_enable_runtime_api(cbid: int, enable: int)

Controls the collection of activity records for specific CUDA Runtime APIs.

Parameters:

• cbid (uint32_t) – callback id of the CUDA Runtime API. This can be found in the header cupti_runtime_cbid.h.

• enable (uint8_t) – is a boolean, denoting whether to enable or disable the collection.

See also

cuptiActivityEnableRuntimeApi

cupti.cupti.activity_flush_all(flag: int)

Request to deliver activity records via the buffer completion callback.

Parameters:

flag (uint32_t) – The flag can be set to indicate a forced flush. See CUpti_ActivityFlag.

See also

cuptiActivityFlushAll

cupti.cupti.activity_flush_period(time: int)

Sets the flush period for the worker thread.

Parameters:

time (uint32_t) – flush period in milliseconds (ms).

See also

cuptiActivityFlushPeriod

cupti.cupti.activity_get_attribute(attr: int, value_size: int, value: int)

Read an activity API attribute.

Parameters:

• attr (ActivityAttribute) – The attribute to read.

• value_size (intptr_t) – Size of buffer pointed by the value, and returns the number of bytes written to value.

• value (intptr_t) – Returns the value of the attribute.

See also

cuptiActivityGetAttribute

cupti.cupti.activity_get_num_dropped_records(

context: int,

stream_id: int,

dropped: int,

)

Get the number of activity records that were dropped of insufficient buffer space.

Parameters:

• context (intptr_t) – The context, or NULL to get dropped count from global queue.

• stream_id (uint32_t) – The stream ID.

• dropped (intptr_t) – The number of records that were dropped since the last call to this function.

See also

cuptiActivityGetNumDroppedRecords

cupti.cupti.activity_pop_external_correlation_id(kind: int) → int

Pop an external correlation id for the calling thread.

Parameters:

kind (ExternalCorrelationKind) – The kind of external API activities should be correlated with.

Returns:

If the function returns successful, contains the last external correlation id for this kind, can be NULL.

Return type:

uint64_t

See also

cuptiActivityPopExternalCorrelationId

cupti.cupti.activity_push_external_correlation_id(kind: int, id: int)

Push an external correlation id for the calling thread.

Parameters:

• kind (ExternalCorrelationKind) – The kind of external API activities should be correlated with.

• id (uint64_t) – External correlation id.

See also

cuptiActivityPushExternalCorrelationId

cupti.cupti.activity_register_callbacks(

func_buffer_requested,

func_buffer_completed,

)

Registers callback functions with CUPTI for activity buffer handling.

Parameters:

• func_buffer_requested (object) – callback which is invoked when an empty buffer is requested by CUPTI.

• func_buffer_completed (object) – callback which is invoked when a buffer containing activity records is available from CUPTI.

See also

cuptiActivityRegisterCallbacks

cupti.cupti.activity_register_timestamp_callback(func_timestamp)

Registers callback function with CUPTI for providing timestamp.

Parameters:

func_timestamp (object) – callback which is invoked when a timestamp is needed by CUPTI.

See also

cuptiActivityRegisterTimestampCallback

cupti.cupti.activity_set_attribute(attr: int, value_size: int, value: int)

Write an activity API attribute.

Parameters:

• attr (ActivityAttribute) – The attribute to write.

• value_size (intptr_t) – The size, in bytes, of the value.

• value (intptr_t) – The attribute value to write.

See also

cuptiActivitySetAttribute

cupti.cupti.compute_capability_supported(major: int, minor: int) → int

Check support for a compute capability.

Parameters:

• major (int) – The major revision number of the compute capability.

• minor (int) – The minor revision number of the compute capability.

Returns:

Pointer to an integer to return the support status.

Return type:

int

See also

cuptiComputeCapabilitySupported

cupti.cupti.device_supported(dev: int) → int

Check support for a compute device.

Parameters:

dev (int) – The device handle returned by CUDA Driver API cuDeviceGet.

Returns:

Pointer to an integer to return the support status.

Return type:

int

See also

cuptiDeviceSupported

cupti.cupti.device_virtualization_mode(dev: int) → int

Query the virtualization mode of the device.

Parameters:

dev (int) – The device handle returned by CUDA Driver API cuDeviceGet.

Returns:

Pointer to an CUpti_DeviceVirtualizationMode to return the virtualization mode.

Return type:

int

See also

cuptiDeviceVirtualizationMode

cupti.cupti.finalize()

Detach CUPTI from the running process.

See also

cuptiFinalize

cupti.cupti.get_auto_boost_state(context: int, state: int)

Get auto boost state.

Parameters:

• context (intptr_t) – A valid CUcontext.

• state (intptr_t) – A pointer to CUpti_ActivityAutoBoostState structure which contains the current state and the id of the process that has requested the current state.

See also

cuptiGetAutoBoostState

cupti.cupti.get_context_id(context: int) → int

Get the ID of a context.

Parameters:

context (intptr_t) – The context.

Returns:

Returns a process-unique ID for the context.

Return type:

uint32_t

See also

cuptiGetContextId

cupti.cupti.get_device_id(context: int) → int

Get the ID of a device.

Parameters:

context (intptr_t) – The context, or NULL to indicate the current context.

Returns:

Returns the ID of the device that is current for the calling thread.

Return type:

uint32_t

See also

cuptiGetDeviceId

cupti.cupti.get_graph_exec_id(graph_exec: int) → int

Get the unique ID of executable graph.

Parameters:

graph_exec (intptr_t) – The executable graph.

Returns:

Returns the unique ID of the executable graph.

Return type:

uint32_t

See also

cuptiGetGraphExecId

cupti.cupti.get_graph_id(graph: int) → int

Get the unique ID of graph.

Parameters:

graph (intptr_t) – The graph.

Returns:

Returns the unique ID of the graph.

Return type:

uint32_t

See also

cuptiGetGraphId

cupti.cupti.get_graph_node_id(node: int) → int

Get the unique ID of a graph node.

Parameters:

node (intptr_t) – The graph node.

Returns:

Returns the unique ID of the node.

Return type:

uint64_t

See also

cuptiGetGraphNodeId

cupti.cupti.get_last_error() → int

Returns the last error from a cupti call or callback.

See also

cuptiGetLastError

cupti.cupti.get_stream_id_ex(

context: int,

stream: int,

per_thread_stream: int,

) → int

Get the ID of a stream.

Parameters:

• context (intptr_t) – If non-NULL then the stream is checked to ensure that it belongs to this context. Typically this parameter should be null.

• stream (intptr_t) – The stream.

• per_thread_stream (uint8_t) – Flag to indicate if program is compiled for per-thread streams.

Returns:

Returns a context-unique ID for the stream.

Return type:

uint32_t

See also

cuptiGetStreamIdEx

cupti.cupti.get_thread_id_type() → int

Get the thread-id type.

Returns:

.

Return type:

int

See also

cuptiGetThreadIdType

cupti.cupti.get_timestamp() → int

Get the CUPTI timestamp.

Returns:

Returns the CUPTI timestamp.

Return type:

uint64_t

See also

cuptiGetTimestamp

cupti.cupti.is_tracing_session_running() → int

Check whether a CUPTI tracing session is still running. This API returns true if a CUPTI library is already loaded by another CUPTI user by using any CUPTI API. This API returns false when CUPTI is finalized and there is no CUPTI currently loaded and active. Can be used to determine if it is safe to unload your CUPTI based tool. Note that this API itself does not load the CUPTI library, it merely checks if a CUPTI library is already loaded by another CUPTI user.

Returns:

Returns whether the tracing session is still running.

Return type:

uint8_t

See also

cuptiIsTracingSessionRunning

cupti.cupti.set_thread_id_type(type: int)

Set the thread-id type.

Parameters:

type (ActivityThreadIdType) –

.

See also

cuptiSetThreadIdType

Enums

class cupti.cupti.ActivityAttribute(value)

Bases: IntEnum

Activity attributes.These attributes are used to control the behavior of the activity API.

See CUpti_ActivityAttribute.

ATTR_CIG_MODE = 2147483649

ATTR_DEVICE_BUFFER_FORCE_INT = 2147483647

ATTR_DEVICE_BUFFER_POOL_LIMIT = 2

ATTR_DEVICE_BUFFER_PRE_ALLOCATE_VALUE = 6

ATTR_DEVICE_BUFFER_SIZE = 0

ATTR_DEVICE_BUFFER_SIZE_CDP = 1

ATTR_DEVICE_BUFFER_SIZE_DEVICE_GRAPHS = 10

ATTR_ENABLE_ALLOCATION_SOURCE_TRACKING = 14

ATTR_ENABLE_ALL_SYNC_RECORDS = 16

ATTR_ENABLE_CIG_MODE = 23

ATTR_ENABLE_CUDA_EVENT_DEVICE_TIMESTAMPS = 17

ATTR_ENABLE_DEVICE_GRAPH_TRACE = 19

ATTR_ENABLE_HES = 13

ATTR_ENABLE_KERNEL_LATENCY_TIMESTAMPS = 15

ATTR_ENABLE_KERNEL_LAUNCH_ATTRIBUTES = 18

ATTR_ENABLE_MULTI_SUBSCRIBER_GRAPH_LEVEL_TRACE = 2147483648

ATTR_ENABLE_MULTI_SUBSCRIBER_GRAPH_TRACE = 20

ATTR_MEM_ALLOCATION_TYPE_HOST_PINNED = 8

ATTR_MULTIPLE_SUBSCRIBER_STATE = 12

ATTR_PER_THREAD_BUFFER = 9

ATTR_PROFILING_SEMAPHORE_POOL_LIMIT = 4

ATTR_PROFILING_SEMAPHORE_POOL_SIZE = 3

ATTR_PROFILING_SEMAPHORE_PRE_ALLOCATE_VALUE = 7

ATTR_THREAD_ID_TYPE = 21

ATTR_TIMESTAMP_CALLBACK = 22

ATTR_USER_DEFINED_RECORDS = 11

ATTR_ZEROED_OUT_BUFFER = 5

class cupti.cupti.ActivityComputeApiKind(value)

Bases: IntEnum

The kind of a compute API.

See CUpti_ActivityComputeApiKind.

CUDA = 1

CUDA_MPS = 2

FORCE_INT = 2147483647

UNKNOWN = 0

class cupti.cupti.ActivityEnvironmentKind(value)

Bases: IntEnum

The kind of environment data. Used to indicate what type of data is being reported by an environment activity record.

See CUpti_ActivityEnvironmentKind.

COOLING = 4

COUNT = 5

FORCE_INT = 2147483647

POWER = 3

SPEED = 1

TEMPERATURE = 2

UNKNOWN = 0

class cupti.cupti.ActivityFlag(value)

Bases: IntEnum

Flags associated with activity records.Activity record flags. Flags can be combined by bitwise OR to associated multiple flags with an activity record. Each flag is specific to a certain activity kind, as noted below.

See CUpti_ActivityFlag.

DEVICE_ATTRIBUTE_CUDEVICE = 1

DEVICE_CONCURRENT_KERNELS = 1

FLUSH_FORCED = 1

FORCE_INT = 2147483647

GLOBAL_ACCESS_KIND_CACHED = 512

GLOBAL_ACCESS_KIND_LOAD = 256

GLOBAL_ACCESS_KIND_SIZE_MASK = 255

INSTRUCTION_CLASS_MASK = 510

INSTRUCTION_VALUE_INVALID = 1

MARKER_COLOR_ARGB = 2

MARKER_COLOR_NONE = 1

MARKER_INSTANTANEOUS = 1

MARKER_START = 2

MARKER_SYNC_ACQUIRE = 8

MARKER_SYNC_ACQUIRE_FAILED = 32

MARKER_SYNC_ACQUIRE_SUCCESS = 16

MARKER_SYNC_RELEASE = 64

MEMCPY_ASYNC = 1

MEMSET_ASYNC = 1

METRIC_OVERFLOWED = 1

METRIC_VALUE_INVALID = 2

NONE = 0

SHARED_ACCESS_KIND_LOAD = 256

SHARED_ACCESS_KIND_SIZE_MASK = 255

THRASHING_IN_CPU = 1

THROTTLING_IN_CPU = 1

class cupti.cupti.ActivityInstructionClass(value)

Bases: IntEnum

SASS instruction classification.The sass instruction are broadly divided into different class. Each enum represents a classification.

See CUpti_ActivityInstructionClass.

BARRIER = 17

BIT_CONVERSION = 4

CONSTANT = 11

CONTROL_FLOW = 5

FP_16 = 19

FP_32 = 1

FP_64 = 2

GENERIC = 9

GLOBAL = 6

GLOBAL_ATOMIC = 13

INTEGER = 3

INTER_THREAD_COMMUNICATION = 16

KIND_FORCE_INT = 2147483647

LOCAL = 8

MISCELLANEOUS = 18

SHARED = 7

SHARED_ATOMIC = 14

SURFACE = 10

SURFACE_ATOMIC = 15

TEXTURE = 12

UNIFORM = 20

UNKNOWN = 0

class cupti.cupti.ActivityJitEntryType(value)

Bases: IntEnum

The types of JIT entry.To be used in CUpti_ActivityJit.

See CUpti_ActivityJitEntryType.

FORCE_INT = 2147483647

INVALID = 0

NVVM_IR_TO_PTX = 2

PTX_TO_CUBIN = 1

class cupti.cupti.ActivityJitOperationType(value)

Bases: IntEnum

The types of JIT compilation operations.To be used in CUpti_ActivityJit.

See CUpti_ActivityJitOperationType.

CACHE_LOAD = 1

CACHE_STORE = 2

COMPILE = 3

FORCE_INT = 2147483647

INVALID = 0

class cupti.cupti.ActivityKind(value)

Bases: IntEnum

The kinds of activity records.Each activity record kind represents information about a GPU or an activity occurring on a CPU or GPU. Each kind is associated with a activity record structure that holds the information associated with the kind.

See CUpti_ActivityKind.

BRANCH = 16

CDP_KERNEL = 18

COMPUTE_ENGINE_CTX_SWITCH = 57

CONCURRENT_KERNEL = 10

CONTEXT = 9

COUNT = 60

CUDA_EVENT = 36

DEVICE = 8

DEVICE_ATTRIBUTE = 28

DEVICE_GRAPH_TRACE = 53

DRIVER = 4

ENVIRONMENT = 20

EVENT = 6

EVENT_INSTANCE = 21

EXTERNAL_CORRELATION = 39

FORCE_INT = 2147483647

FUNCTION = 26

GLOBAL_ACCESS = 15

GRAPH_HOST_NODE = 56

GRAPH_TRACE = 51

GREEN_CONTEXT = 59

HOST_LAUNCH = 58

INSTANTANEOUS_EVENT = 41

INSTANTANEOUS_EVENT_INSTANCE = 42

INSTANTANEOUS_METRIC = 43

INSTANTANEOUS_METRIC_INSTANCE = 44

INSTRUCTION_CORRELATION = 32

INSTRUCTION_EXECUTION = 24

INTERNAL_LAUNCH_API = 48

INVALID = 0

JIT = 52

KERNEL = 3

MARKER = 12

MARKER_DATA = 13

MEMCPY = 1

MEMCPY2 = 22

MEMORY = 45

MEMORY2 = 49

MEMORY_POOL = 50

MEMSET = 2

MEM_DECOMPRESS = 54

METRIC = 7

METRIC_INSTANCE = 23

MODULE = 27

NAME = 11

NVLINK = 40

OPENACC_DATA = 33

OPENACC_LAUNCH = 34

OPENACC_OTHER = 35

OPENMP = 47

OVERHEAD = 17

PCIE = 46

PC_SAMPLING = 30

PC_SAMPLING_RECORD_INFO = 31

PREEMPTION = 19

ROTATION = 55

RUNTIME = 5

SHARED_ACCESS = 29

SOURCE_LOCATOR = 14

STREAM = 37

SYNCHRONIZATION = 38

UNIFIED_MEMORY_COUNTER = 25

class cupti.cupti.ActivityLaunchType(value)

Bases: IntEnum

The type of the CUDA kernel launch.

See CUpti_ActivityLaunchType.

CBL_COMMANDLIST = 3

COOPERATIVE_MULTI_DEVICE = 2

COOPERATIVE_SINGLE_DEVICE = 1

REGULAR = 0

class cupti.cupti.ActivityMemcpyKind(value)

Bases: IntEnum

The kind of a memory copy, indicating the source and destination targets of the copy.Each kind represents the source and destination targets of a memory copy. Targets are host, device, and array.

See CUpti_ActivityMemcpyKind.

ATOA = 5

ATOD = 6

ATOH = 4

DTOA = 7

DTOD = 8

DTOH = 2

FORCE_INT = 2147483647

HTOA = 3

HTOD = 1

HTOH = 9

PTOP = 10

UNKNOWN = 0

class cupti.cupti.ActivityMemoryKind(value)

Bases: IntEnum

The kinds of memory accessed by a memory operation/copy.Each kind represents the type of the memory accessed by a memory operation/copy.

See CUpti_ActivityMemoryKind.

ARRAY = 4

DEVICE = 3

DEVICE_STATIC = 6

FORCE_INT = 2147483647

MANAGED = 5

MANAGED_STATIC = 7

PAGEABLE = 1

PINNED = 2

UNKNOWN = 0

class cupti.cupti.ActivityMemoryOperationType(value)

Bases: IntEnum

Memory operation types.Describes the type of memory operation, to be used with CUpti_ActivityMemory4.

See CUpti_ActivityMemoryOperationType.

ALLOCATION = 1

FORCE_INT = 2147483647

INVALID = 0

RELEASE = 2

class cupti.cupti.ActivityMemoryPoolOperationType(value)

Bases: IntEnum

Memory pool operation types.Describes the type of memory pool operation, to be used with CUpti_ActivityMemoryPool2.

See CUpti_ActivityMemoryPoolOperationType.

CREATED = 1

DESTROYED = 2

FORCE_INT = 2147483647

INVALID = 0

TRIMMED = 3

class cupti.cupti.ActivityMemoryPoolType(value)

Bases: IntEnum

Memory pool types.Describes the type of memory pool, to be used with CUpti_ActivityMemory4.

See CUpti_ActivityMemoryPoolType.

FORCE_INT = 2147483647

IMPORTED = 2

INVALID = 0

LOCAL = 1

class cupti.cupti.ActivityObjectKind(value)

Bases: IntEnum

The kinds of activity objects.

See CUpti_ActivityObjectKind.

CONTEXT = 4

DEVICE = 3

FORCE_INT = 2147483647

PROCESS = 1

STREAM = 5

THREAD = 2

UNKNOWN = 0

class cupti.cupti.ActivityOverheadKind(value)

Bases: IntEnum

The kinds of activity overhead.

See CUpti_ActivityOverheadKind.

ACTIVITY_BUFFER_REQUEST = 458752

COMMAND_BUFFER_FULL = 393216

CUPTI_BUFFER_FLUSH = 65536

CUPTI_INSTRUMENTATION = 131072

CUPTI_RESOURCE = 196608

DRIVER_COMPILER = 1

FORCE_INT = 2147483647

LAZY_FUNCTION_LOADING = 327680

RUNTIME_TRIGGERED_MODULE_LOADING = 262144

UNKNOWN = 0

UVM_ACTIVITY_INIT = 524288

class cupti.cupti.ActivityPCSamplingPeriod(value)

Bases: IntEnum

Sampling period for PC sampling method.Sampling period can be set using cuptiActivityConfigurePCSampling

See CUpti_ActivityPCSamplingPeriod.

FORCE_INT = 2147483647

HIGH = 4

INVALID = 0

LOW = 2

MAX = 5

MID = 3

MIN = 1

class cupti.cupti.ActivityPCSamplingStallReason(value)

Bases: IntEnum

The stall reason for PC sampling activity.

See CUpti_ActivityPCSamplingStallReason.

CONSTANT_MEMORY_DEPENDENCY = 7

EXEC_DEPENDENCY = 3

FORCE_INT = 2147483647

INST_FETCH = 2

INVALID = 0

MEMORY_DEPENDENCY = 4

MEMORY_THROTTLE = 9

NONE = 1

NOT_SELECTED = 10

OTHER = 11

PIPE_BUSY = 8

SLEEPING = 12

SYNC = 6

TEXTURE = 5

class cupti.cupti.ActivityPartitionedGlobalCacheConfig(value)

Bases: IntEnum

Partitioned global caching option.

See CUpti_ActivityPartitionedGlobalCacheConfig.

FORCE_INT = 2147483647

NOT_SUPPORTED = 1

OFF = 2

ON = 3

UNKNOWN = 0

class cupti.cupti.ActivityPreemptionKind(value)

Bases: IntEnum

The kind of a preemption activity.

See CUpti_ActivityPreemptionKind.

FORCE_INT = 2147483647

RESTORE = 2

SAVE = 1

UNKNOWN = 0

class cupti.cupti.ActivityStreamFlag(value)

Bases: IntEnum

stream type.The types of stream to be used with CUpti_ActivityStream.

See CUpti_ActivityStreamFlag.

FLAG_DEFAULT = 1

FLAG_FORCE_INT = 2147483647

FLAG_NON_BLOCKING = 2

FLAG_NULL = 3

FLAG_UNKNOWN = 0

MASK = 65535

class cupti.cupti.ActivitySynchronizationType(value)

Bases: IntEnum

Synchronization type.The types of synchronization to be used with CUpti_ActivitySynchronization2.

See CUpti_ActivitySynchronizationType.

CONTEXT_SYNCHRONIZE = 4

EVENT_SYNCHRONIZE = 1

FORCE_INT = 2147483647

STREAM_SYNCHRONIZE = 3

STREAM_WAIT_EVENT = 2

UNKNOWN = 0

class cupti.cupti.ActivityThreadIdType(value)

Bases: IntEnum

Thread-Id types.CUPTI uses different methods to obtain the thread-id depending on the support and the underlying platform. This enum documents these methods for each type. APIs cuptiSetThreadIdType and cuptiGetThreadIdType can be used to set and get the thread-id type.

See CUpti_ActivityThreadIdType.

DEFAULT = 0

FORCE_INT = 2147483647

SIZE = 2

SYSTEM = 1

class cupti.cupti.ActivityUnifiedMemoryAccessType(value)

Bases: IntEnum

Memory access type for unified memory page faults.This is valid for CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_GPU_PAGE_FAULT and CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_CPU_PAGE_FAULT_COUNT

See CUpti_ActivityUnifiedMemoryAccessType.

ATOMIC = 3

PREFETCH = 4

READ = 1

UNKNOWN = 0

WRITE = 2

class cupti.cupti.ActivityUnifiedMemoryCounterKind(value)

Bases: IntEnum

Kind of the Unified Memory counter.Many activities are associated with Unified Memory mechanism; among them are transfers from host to device, device to host, page fault at host side.

See CUpti_ActivityUnifiedMemoryCounterKind.

BYTES_TRANSFER_DTOD = 8

BYTES_TRANSFER_DTOH = 2

BYTES_TRANSFER_HTOD = 1

COUNT = 9

CPU_PAGE_FAULT_COUNT = 3

FORCE_INT = 2147483647

GPU_PAGE_FAULT = 4

REMOTE_MAP = 7

THRASHING = 5

THROTTLING = 6

UNKNOWN = 0

class cupti.cupti.ActivityUnifiedMemoryCounterScope(value)

Bases: IntEnum

Scope of the unified memory counter (deprecated in CUDA 7.0).

See CUpti_ActivityUnifiedMemoryCounterScope.

COUNT = 3

FORCE_INT = 2147483647

PROCESS_ALL_DEVICES = 2

PROCESS_SINGLE_DEVICE = 1

UNKNOWN = 0

class cupti.cupti.ActivityUnifiedMemoryMigrationCause(value)

Bases: IntEnum

Migration cause of the Unified Memory counter.This is valid for CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_HTOD and CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_DTOH and CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_DTOD

See CUpti_ActivityUnifiedMemoryMigrationCause.

ACCESS_COUNTERS = 5

COHERENCE = 2

EVICTION = 4

PREFETCH = 3

UNKNOWN = 0

USER = 1

class cupti.cupti.ActivityUnifiedMemoryRemoteMapCause(value)

Bases: IntEnum

Remote memory map cause of the Unified Memory counter.This is valid for CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_REMOTE_MAP

See CUpti_ActivityUnifiedMemoryRemoteMapCause.

COHERENCE = 1

EVICTION = 5

OUT_OF_MEMORY = 4

POLICY = 3

THRASHING = 2

UNKNOWN = 0

class cupti.cupti.ChannelType(value)

Bases: IntEnum

See CUpti_ChannelType.

ASYNC_MEMCPY = 2

COMPUTE = 1

DECOMP = 3

FORCE_INT = 2147483647

INVALID = 0

class cupti.cupti.ComputeEngineCtxSwitchOperationType(value)

Bases: IntEnum

The operation type of CUDA context switch event records.

See CUpti_ComputeEngineCtxSwitchOperationType.

COUNT = 2147483647

INVALID = 0

START = 1

class cupti.cupti.ConfidentialComputeRotationEventType(value)

Bases: IntEnum

See CUpti_ConfidentialComputeRotationEventType.

EVENT_TYPE_FORCE_INT = 2147483647

INVALID_ROTATION_EVENT = 0

KEY_ROTATION_ACKNOWLEGED = 2

KEY_ROTATION_CHANNEL_BLOCKED = 1

KEY_ROTATION_CHANNEL_DRAINED = 2

KEY_ROTATION_CHANNEL_UNBLOCKED = 3

KEY_ROTATION_COMPLETED = 4

KEY_ROTATION_REQUESTED = 1

KEY_ROTATION_STARTED = 3

class cupti.cupti.ContextCigMode(value)

Bases: IntEnum

CIG (CUDA in Graphics) Modes.Describes the CIG modes associated with the CUDA context.

See CUpti_ContextCigMode.

CIG = 1

CIG_FALLBACK = 2

FORCE_INT = 2147483647

NONE = 0

class cupti.cupti.DevType(value)

Bases: IntEnum

The device type for device connected to NVLink.

See CUpti_DevType.

FORCE_INT = 2147483647

GPU = 1

INVALID = 0

NPU = 2

class cupti.cupti.DeviceAttribute(value)

Bases: IntEnum

Device attributes.CUPTI device attributes. These attributes can be read using cuptiDeviceGetAttribute.

See CUpti_DeviceAttribute.

ATTR_CLASS = 10

ATTR_FLOP_DP_PER_CYCLE = 12

ATTR_FLOP_HP_PER_CYCLE = 17

ATTR_FLOP_SP_PER_CYCLE = 11

ATTR_FORCE_INT = 2147483647

ATTR_GLOBAL_MEMORY_BANDWIDTH = 3

ATTR_GPU_CPU_NVLINK_BW = 19

ATTR_INSTRUCTION_PER_CYCLE = 4

ATTR_INSTRUCTION_THROUGHPUT_SINGLE_PRECISION = 5

ATTR_MAX_EVENT_DOMAIN_ID = 2

ATTR_MAX_EVENT_ID = 1

ATTR_MAX_FRAME_BUFFERS = 6

ATTR_MAX_L2_UNITS = 13

ATTR_MAX_SHARED_MEMORY_CACHE_CONFIG_PREFER_EQUAL = 16

ATTR_MAX_SHARED_MEMORY_CACHE_CONFIG_PREFER_L1 = 15

ATTR_MAX_SHARED_MEMORY_CACHE_CONFIG_PREFER_SHARED = 14

ATTR_NVLINK_PRESENT = 18

ATTR_NVSWITCH_PRESENT = 20

ATTR_PCIE_GEN = 9

ATTR_PCIE_LINK_RATE = 7

ATTR_PCIE_LINK_WIDTH = 8

class cupti.cupti.DeviceVirtualizationMode(value)

Bases: IntEnum

This indicates the virtualization mode in which CUDA device is running

See CUpti_DeviceVirtualizationMode.

FORCE_INT = 2147483647

NONE = 0

PASS_THROUGH = 1

VIRTUAL_GPU = 2

class cupti.cupti.EnvironmentClocksThrottleReason(value)

Bases: IntEnum

Reasons for clock throttling.The possible reasons that a clock can be throttled. There can be more than one reason that a clock is being throttled so these types can be combined by bitwise OR. These are used in the clocksThrottleReason field in the Environment Activity Record.

See CUpti_EnvironmentClocksThrottleReason.

FORCE_INT = 2147483647

GPU_IDLE = 1

HW_SLOWDOWN = 8

NONE = 0

SW_POWER_CAP = 4

UNKNOWN = 2147483648

UNSUPPORTED = 1073741824

USER_DEFINED_CLOCKS = 2

class cupti.cupti.ExternalCorrelationKind(value)

Bases: IntEnum

The kind of external APIs supported for correlation.Custom correlation kinds are reserved for usage in external tools.

See CUpti_ExternalCorrelationKind.

CUSTOM0 = 3

CUSTOM1 = 4

CUSTOM2 = 5

FORCE_INT = 2147483647

INVALID = 0

OPENACC = 2

SIZE = 6

UNKNOWN = 1

class cupti.cupti.FuncExecutionModel(value)

Bases: IntEnum

The execution model of a kernel function. This should be used to set executionModel field in kernel records.

See CUpti_FuncExecutionModel.

FORCE_INT = 2147483647

SIMT = 1

SIZE = 3

TILE = 2

UNKNOWN = 0

class cupti.cupti.FuncShmemLimitConfig(value)

Bases: IntEnum

The shared memory limit per block config for a kernel This should be used to set ‘cudaOccFuncShmemConfig’ field in occupancy calculator API.

See CUpti_FuncShmemLimitConfig.

DEFAULT = 0

FORCE_INT = 2147483647

OPTIN = 1

class cupti.cupti.OpenAccConstructKind(value)

Bases: IntEnum

The OpenAcc parent construct kind for OpenAcc activity records.

See CUpti_OpenAccConstructKind.

ATOMIC = 8

DATA = 4

DECLARE = 9

ENTER_DATA = 5

EXIT_DATA = 6

FORCE_INT = 2147483647

HOST_DATA = 7

INIT = 10

KERNELS = 2

LOOP = 3

PARALLEL = 1

ROUTINE = 14

RUNTIME_API = 16

SET = 12

SHUTDOWN = 11

UNKNOWN = 0

UPDATE = 13

WAIT = 15

class cupti.cupti.OpenAccEventKind(value)

Bases: IntEnum

The OpenAcc event kind for OpenAcc activity records.

See CUpti_OpenAccEventKind.

ALLOC = 15

COMPUTE_CONSTRUCT = 9

CREATE = 13

DELETE = 14

DEVICE_INIT = 1

DEVICE_SHUTDOWN = 2

ENQUEUE_DOWNLOAD = 6

ENQUEUE_LAUNCH = 4

ENQUEUE_UPLOAD = 5

ENTER_DATA = 11

EXIT_DATA = 12

FORCE_INT = 2147483647

FREE = 16

IMPLICIT_WAIT = 8

INVALID = 0

RUNTIME_SHUTDOWN = 3

UPDATE = 10

WAIT = 7

class cupti.cupti.OpenMpEventKind(value)

Bases: IntEnum

See CUpti_OpenMpEventKind.

FORCE_INT = 2147483647

IDLE = 4

INVALID = 0

PARALLEL = 1

TASK = 2

THREAD = 3

WAIT_BARRIER = 5

WAIT_TASKWAIT = 6

class cupti.cupti.PcieDeviceType(value)

Bases: IntEnum

Field to differentiate whether PCIE Activity record is of a GPU or a PCI Bridge

See CUpti_PcieDeviceType.

BRIDGE = 1

FORCE_INT = 2147483647

GPU = 0

Classes

class cupti.cupti.ActivityAPI

Bases: object

Empty-initialize an instance of CUpti_ActivityAPI.

See also

CUpti_ActivityAPI

cbid

The ID of the driver or runtime function.

Type:

int

correlation_id

The correlation ID of the driver or runtime CUDA function. Each function invocation is assigned a unique correlation ID that is identical to the correlation ID in the memcpy, memset, or kernel activity record that is associated with this function.

Type:

int

end

The end timestamp for the function, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the function.

Type:

int

static from_buffer(buffer)

Create an ActivityAPI instance with the memory from the given buffer.

static from_data(data)

Create an ActivityAPI instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_api_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityAPI instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_DRIVER, CUPTI_ACTIVITY_KIND_RUNTIME, or CUPTI_ACTIVITY_KIND_INTERNAL_LAUNCH_API.

Type:

int

process_id

The ID of the process where the driver or runtime CUDA function is executing.

Type:

int

ptr

Get the pointer address to the data as Python int.

return_value

The return value for the function. For a CUDA driver function with will be a CUresult value, and for a CUDA runtime function this will be a cudaError_t value.

Type:

int

start

The start timestamp for the function, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the function.

Type:

int

thread_id

The ID of the thread where the driver or runtime CUDA function is executing.

Type:

int

class cupti.cupti.ActivityAutoBoostState

Bases: object

Empty-initialize an instance of CUpti_ActivityAutoBoostState.

See also

CUpti_ActivityAutoBoostState

enabled

Returned auto boost state. 1 is returned in case auto boost is enabled, 0 otherwise

Type:

int

static from_buffer(buffer)

Create an ActivityAutoBoostState instance with the memory from the given buffer.

static from_data(data)

Create an ActivityAutoBoostState instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_auto_boost_state_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityAutoBoostState instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

pid

Id of process that has set the current boost state. The value will be CUPTI_AUTO_BOOST_INVALID_CLIENT_PID if the user does not have the permission to query process ids or there is an error in querying the process id.

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityCdpKernel

Bases: object

Empty-initialize an instance of CUpti_ActivityCdpKernel.

See also

CUpti_ActivityCdpKernel

block_x

The X-dimension block size for the kernel.

Type:

int

block_y

The Y-dimension block size for the kernel.

Type:

int

block_z

The Z-dimension grid size for the kernel.

Type:

int

cache_config

_py_anon_pod9:

completed

The timestamp when kernel is marked as completed, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the completion time is unknown.

Type:

int

context_id

The ID of the context where the kernel is executing.

Type:

int

correlation_id

The correlation ID of the kernel. Each kernel execution is assigned a unique correlation ID that is identical to the correlation ID in the driver API activity record that launched the kernel.

Type:

int

device_id

The ID of the device where the kernel is executing.

Type:

int

dynamic_shared_memory

The dynamic shared memory reserved for the kernel, in bytes.

Type:

int

end

The end timestamp for the kernel execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the kernel.

Type:

int

static from_buffer(buffer)

Create an ActivityCdpKernel instance with the memory from the given buffer.

static from_data(data)

Create an ActivityCdpKernel instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_cdp_kernel_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityCdpKernel instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

grid_id

The grid ID of the kernel. Each kernel execution is assigned a unique grid ID.

Type:

int

grid_x

The X-dimension grid size for the kernel.

Type:

int

grid_y

The Y-dimension grid size for the kernel.

Type:

int

grid_z

The Z-dimension grid size for the kernel.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_CDP_KERNEL

Type:

int

local_memory_per_thread

The amount of local memory reserved for each thread, in bytes.

Type:

int

local_memory_total

The total amount of local memory reserved for the kernel, in bytes.

Type:

int

name

The name of the kernel. This name is shared across all activity records representing the same kernel, and so should not be modified.

Type:

str

pad

Undefined. Reserved for internal use.

Type:

int

parent_block_x

The X-dimension of the parent block.

Type:

int

parent_block_y

The Y-dimension of the parent block.

Type:

int

parent_block_z

The Z-dimension of the parent block.

Type:

int

parent_grid_id

The grid ID of the parent kernel.

Type:

int

ptr

Get the pointer address to the data as Python int.

queued

The timestamp when kernel is queued up, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the queued time is unknown.

Type:

int

registers_per_thread

The number of registers required for each thread executing the kernel.

Type:

int

shared_memory_config

The shared memory configuration used for the kernel. The value is one of the CUsharedconfig enumeration values from cuda.h.

Type:

int

start

The start timestamp for the kernel execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the kernel.

Type:

int

static_shared_memory

The static shared memory allocated for the kernel, in bytes.

Type:

int

stream_id

The ID of the stream where the kernel is executing.

Type:

int

submitted

The timestamp when kernel is submitted to the gpu, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the submission time is unknown.

Type:

int

class cupti.cupti.ActivityComputeEngineCtxSwitch

Bases: object

Empty-initialize an instance of CUpti_ActivityComputeEngineCtxSwitch.

See also

CUpti_ActivityComputeEngineCtxSwitch

context_id

The ID of the CUDA context.

Type:

int

static from_buffer(buffer)

Create an ActivityComputeEngineCtxSwitch instance with the memory from the given buffer.

static from_data(data)

Create an ActivityComputeEngineCtxSwitch instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_compute_engine_ctx_switch_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityComputeEngineCtxSwitch instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_COMPUTE_ENGINE_CTX_SWITCH

Type:

int

operation_type

The type of the Compute Engine Context switch operation. CUPTI_COMPUTE_ENGINE_CTX_SWITCH_OPERATION_START indicates the start of the context switch operation. CUPTI_COMPUTE_ENGINE_CTX_SWITCH_OPERATION_END indicates the end of the context switch operation.

Type:

int

padding

int:

ptr

Get the pointer address to the data as Python int.

timestamp

The timestamp at which the CUpti_ComputeEngineCtxSwitchOperationType occurs.

Type:

int

class cupti.cupti.ActivityConfidentialComputeRotation

Bases: object

Empty-initialize an instance of CUpti_ActivityConfidentialComputeRotation.

See also

CUpti_ActivityConfidentialComputeRotation

channel_id

Channel ID

Type:

int

channel_type

Channel Type CUpti_ChannelType

Type:

int

context_id

Context ID

Type:

int

device_id

Device ID

Type:

int

event_type

Type of event CUpti_ConfidentialComputeRotationEventType

Type:

int

static from_buffer(buffer)

Create an ActivityConfidentialComputeRotation instance with the memory from the given buffer.

static from_data(data)

Create an ActivityConfidentialComputeRotation instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_confidential_compute_rotation_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityConfidentialComputeRotation instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_CONFIDENTIAL_COMPUTE_ROTATION.

Type:

int

ptr

Get the pointer address to the data as Python int.

timestamp

Timestamp in ns

Type:

int

class cupti.cupti.ActivityContext4

Bases: object

Empty-initialize an instance of CUpti_ActivityContext4.

See also

CUpti_ActivityContext4

cig_mode

This field indicates the CIG mode

Type:

int

compute_api_kind

The compute API kind. CUpti_ActivityComputeApiKind

Type:

int

context_id

The context ID.

Type:

int

device_id

The device ID.

Type:

int

static from_buffer(buffer)

Create an ActivityContext4 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityContext4 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_context4_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityContext4 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

is_green_context

This field indicates whether the context is a green context

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_CONTEXT.

Type:

int

null_stream_id

The ID for the NULL stream in this context

Type:

int

num_multiprocessors

Number of multiprocessors assigned to the green context Invalid if the field ‘isGreenContext’ is 0

Type:

int

padding

Undefined. Reserved for internal use.

Type:

int

padding2

Undefined. Reserved for internal use.

Type:

int

parent_context_id

The ID of the parent context. It would be 0 if context does not have parent

Type:

int

process_id

The ID of the process associated with the context.

Type:

int

ptr

Get the pointer address to the data as Python int.

Deprecated since version 13.2.0.

class cupti.cupti.ActivityContext3

Bases: object

Empty-initialize an instance of CUpti_ActivityContext3.

See also

CUpti_ActivityContext3

cig_mode

This field indicates the CIG mode

Type:

int

compute_api_kind

The compute API kind. CUpti_ActivityComputeApiKind

Type:

int

context_id

The context ID.

Type:

int

device_id

The device ID.

Type:

int

static from_buffer(buffer)

Create an ActivityContext3 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityContext3 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_context3_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityContext3 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

is_green_context

This field indicates whether the context is a green context

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_CONTEXT.

Type:

int

null_stream_id

The ID for the NULL stream in this context

Type:

int

num_multiprocessors

Number of multiprocessors assigned to the green context Invalid if the field ‘isGreenContext’ is 0

Type:

int

padding

int:

padding2

int:

parent_context_id

The ID of the parent context. It would be 0 if context does not have parent

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityCudaEvent2

Bases: object

Empty-initialize an instance of CUpti_ActivityCudaEvent2.

See also

CUpti_ActivityCudaEvent2

context_id

The ID of the context where the event was recorded.

Type:

int

correlation_id

The correlation ID of the API to which this result is associated.

Type:

int

cuda_event_sync_id

A unique ID to associate event synchronization records with the latest CUDA Event record. Similar field is added in CUpti_ActivitySynchronization2 to associate CUDA Event record to the synchronization record. The same CUDA event can be used multiple times, so the event id will not be unique to correlate the synchronization record with the latest CUDA Event record. This field will be unique and can be used to do the required correlation.

Type:

int

device_id

The ID of the device where the event was recorded.

Type:

int

device_timestamp

The device-side timestamp on CUDA event record. Timestamp is in nanoseconds. Collection of this field is disabled by default. It can be enabled by calling CUPTI API cuptiActivityEnableCudaEventDeviceTimestamps

Type:

int

event_id

A unique event ID to identify the event record.

Type:

int

static from_buffer(buffer)

Create an ActivityCudaEvent2 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityCudaEvent2 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_cuda_event2_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityCudaEvent2 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_CUDA_EVENT.

Type:

int

pad

Undefined. Reserved for internal use.

Type:

int

pad2

Undefined. Reserved for internal use.

Type:

int

ptr

Get the pointer address to the data as Python int.

reserved0

Undefined. Reserved for internal use.

Type:

int

stream_id

The compute stream where the event was recorded.

Type:

int

Deprecated since version 12.8.0.

class cupti.cupti.ActivityCudaEvent

Bases: object

Empty-initialize an instance of CUpti_ActivityCudaEvent.

See also

CUpti_ActivityCudaEvent

context_id

int:

correlation_id

int:

event_id

int:

static from_buffer(buffer)

Create an ActivityCudaEvent instance with the memory from the given buffer.

static from_data(data)

Create an ActivityCudaEvent instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_cuda_event_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityCudaEvent instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

int:

pad

int:

ptr

Get the pointer address to the data as Python int.

stream_id

int:

class cupti.cupti.ActivityDevice6

Bases: object

Empty-initialize an instance of CUpti_ActivityDevice6.

See also

CUpti_ActivityDevice6

compute_capability_major

Compute capability for the device, major number.

Type:

int

compute_capability_minor

Compute capability for the device, minor number.

Type:

int

compute_instance_id

Compute Instance id for MIG enabled devices. If mig mode is disabled value is set to UINT32_MAX

Type:

int

constant_memory_size

The amount of constant memory on the device, in bytes.

Type:

int

core_clock_rate

The core clock rate of the device, in kHz.

Type:

int

ecc_enabled

ECC enabled flag for device

Type:

int

flags_

The flags associated with the device. CUpti_ActivityFlag

Type:

int

static from_buffer(buffer)

Create an ActivityDevice6 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityDevice6 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_device6_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityDevice6 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

global_memory_bandwidth

The global memory bandwidth available on the device, in kBytes/sec.

Type:

int

global_memory_size

The amount of global memory on the device, in bytes.

Type:

int

gpu_instance_id

GPU Instance id for MIG enabled devices. If mig mode is disabled value is set to UINT32_MAX

Type:

int

id

The device ID.

Type:

int

is_cuda_visible

Flag to indicate whether the device is visible to CUDA. Users can set the device visibility using CUDA_VISIBLE_DEVICES environment

Type:

int

is_mig_enabled

MIG enabled flag for device

Type:

int

is_numa_node

Numa (Non-uniform memory access) information for device GPU is a NUMA node or not

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_DEVICE.

Type:

int

l2cache_size

The size of the L2 cache on the device, in bytes.

Type:

int

max_block_dim_x

Maximum allowed X dimension for a block.

Type:

int

max_block_dim_y

Maximum allowed Y dimension for a block.

Type:

int

max_block_dim_z

Maximum allowed Z dimension for a block.

Type:

int

max_blocks_per_multiprocessor

Maximum number of blocks that can be present on a multiprocessor at any given time.

Type:

int

max_grid_dim_x

Maximum allowed X dimension for a grid.

Type:

int

max_grid_dim_y

Maximum allowed Y dimension for a grid.

Type:

int

max_grid_dim_z

Maximum allowed Z dimension for a grid.

Type:

int

max_ipc

The maximum “instructions per cycle” possible on each device multiprocessor.

Type:

int

max_registers_per_block

Maximum number of registers that can be allocated to a block.

Type:

int

max_registers_per_multiprocessor

Maximum number of 32-bit registers available per multiprocessor.

Type:

int

max_shared_memory_per_block

Maximum amount of shared memory that can be assigned to a block, in bytes.

Type:

int

max_shared_memory_per_multiprocessor

Maximum amount of shared memory available per multiprocessor, in bytes.

Type:

int

max_threads_per_block

Maximum number of threads allowed in a block.

Type:

int

max_warps_per_multiprocessor

Maximum number of warps that can be present on a multiprocessor at any given time.

Type:

int

mig_uuid

The MIG UUID. This value is the globally unique immutable alphanumeric identifier of the device.

Type:

cuda.bindings.driver.CUuuid

name

The device name. Client is responsible for freeing this memory using the free function when done.

Type:

str

num_memcpy_engines

Number of memory copy engines on the device.

Type:

int

num_multiprocessors

Number of multiprocessors on the device.

Type:

int

num_threads_per_warp

The number of threads per warp on the device.

Type:

int

num_tpcs

Number of TPCs on the device.

Type:

int

numa_id

Numa (Non-uniform memory access) information for device NUMA node ID of the GPU memory if GPU is not a NUMA node, it returns invalidNumaId

Type:

int

ptr

Get the pointer address to the data as Python int.

reserved0

Undefined. Reserved for internal use to maintain 8-byte alignment.

Type:

int

uuid

The device UUID. This value is the globally unique immutable alphanumeric identifier of the device.

Type:

cuda.bindings.driver.CUuuid

Deprecated since version 13.2.0.

class cupti.cupti.ActivityDevice5

Bases: object

Empty-initialize an instance of CUpti_ActivityDevice5.

See also

CUpti_ActivityDevice5

compute_capability_major

Compute capability for the device, major number.

Type:

int

compute_capability_minor

Compute capability for the device, minor number.

Type:

int

compute_instance_id

Compute Instance id for MIG enabled devices. If mig mode is disabled value is set to UINT32_MAX

Type:

int

constant_memory_size

The amount of constant memory on the device, in bytes.

Type:

int

core_clock_rate

The core clock rate of the device, in kHz.

Type:

int

ecc_enabled

ECC enabled flag for device

Type:

int

flags_

The flags associated with the device. CUpti_ActivityFlag

Type:

int

static from_buffer(buffer)

Create an ActivityDevice5 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityDevice5 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_device5_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityDevice5 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

global_memory_bandwidth

The global memory bandwidth available on the device, in kBytes/sec.

Type:

int

global_memory_size

The amount of global memory on the device, in bytes.

Type:

int

gpu_instance_id

GPU Instance id for MIG enabled devices. If mig mode is disabled value is set to UINT32_MAX

Type:

int

id

The device ID.

Type:

int

is_cuda_visible

Flag to indicate whether the device is visible to CUDA. Users can set the device visibility using CUDA_VISIBLE_DEVICES environment

Type:

int

is_mig_enabled

MIG enabled flag for device

Type:

int

is_numa_node

Numa (Non-uniform memory access) information for device GPU is a NUMA node or not

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_DEVICE.

Type:

int

l2cache_size

The size of the L2 cache on the device, in bytes.

Type:

int

max_block_dim_x

Maximum allowed X dimension for a block.

Type:

int

max_block_dim_y

Maximum allowed Y dimension for a block.

Type:

int

max_block_dim_z

Maximum allowed Z dimension for a block.

Type:

int

max_blocks_per_multiprocessor

Maximum number of blocks that can be present on a multiprocessor at any given time.

Type:

int

max_grid_dim_x

Maximum allowed X dimension for a grid.

Type:

int

max_grid_dim_y

Maximum allowed Y dimension for a grid.

Type:

int

max_grid_dim_z

Maximum allowed Z dimension for a grid.

Type:

int

max_ipc

The maximum “instructions per cycle” possible on each device multiprocessor.

Type:

int

max_registers_per_block

Maximum number of registers that can be allocated to a block.

Type:

int

max_registers_per_multiprocessor

Maximum number of 32-bit registers available per multiprocessor.

Type:

int

max_shared_memory_per_block

Maximum amount of shared memory that can be assigned to a block, in bytes.

Type:

int

max_shared_memory_per_multiprocessor

Maximum amount of shared memory available per multiprocessor, in bytes.

Type:

int

max_threads_per_block

Maximum number of threads allowed in a block.

Type:

int

max_warps_per_multiprocessor

Maximum number of warps that can be present on a multiprocessor at any given time.

Type:

int

mig_uuid

The MIG UUID. This value is the globally unique immutable alphanumeric identifier of the device.

Type:

cuda.bindings.driver.CUuuid

name

The device name. Client is responsible for freeing this memory using the free function when done.

Type:

str

num_memcpy_engines

Number of memory copy engines on the device.

Type:

int

num_multiprocessors

Number of multiprocessors on the device.

Type:

int

num_threads_per_warp

The number of threads per warp on the device.

Type:

int

numa_id

Numa (Non-uniform memory access) information for device NUMA node ID of the GPU memory if GPU is not a NUMA node, it returns invalidNumaId

Type:

int

ptr

Get the pointer address to the data as Python int.

uuid

The device UUID. This value is the globally unique immutable alphanumeric identifier of the device.

Type:

cuda.bindings.driver.CUuuid

class cupti.cupti.ActivityDeviceAttribute

Bases: object

Empty-initialize an instance of CUpti_ActivityDeviceAttribute.

See also

CUpti_ActivityDeviceAttribute

attribute

The attribute, either a CUpti_DeviceAttribute or CUdevice_attribute. Flag CUPTI_ACTIVITY_FLAG_DEVICE_ATTRIBUTE_CUDEVICE is used to indicate what kind of attribute this is. If CUPTI_ACTIVITY_FLAG_DEVICE_ATTRIBUTE_CUDEVICE is 1 then CUdevice_attribute field is value, otherwise CUpti_DeviceAttribute field is valid.

Type:

_py_anon_pod11

device_id

The ID of the device that this attribute applies to.

Type:

int

flags_

The flags associated with the device. CUpti_ActivityFlag

Type:

int

static from_buffer(buffer)

Create an ActivityDeviceAttribute instance with the memory from the given buffer.

static from_data(data)

Create an ActivityDeviceAttribute instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_device_attribute_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityDeviceAttribute instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_DEVICE_ATTRIBUTE.

Type:

int

ptr

Get the pointer address to the data as Python int.

value

The value for the attribute. See CUpti_DeviceAttribute and CUdevice_attribute for the type of the value for a given attribute.

Type:

_py_anon_pod12

class cupti.cupti.ActivityDeviceGraphTrace

Bases: object

Empty-initialize an instance of CUpti_ActivityDeviceGraphTrace.

See also

CUpti_ActivityDeviceGraphTrace

context_id

The ID of the context where the first node of the graph is executed.

Type:

int

device_id

The ID of the device where the first node of the graph is executed.

Type:

int

device_launch_mode

The type of launch. See CUpti_DeviceGraphLaunchMode

Type:

int

end

The end timestamp for the graph execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the graph.

Type:

int

static from_buffer(buffer)

Create an ActivityDeviceGraphTrace instance with the memory from the given buffer.

static from_data(data)

Create an ActivityDeviceGraphTrace instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_device_graph_trace_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityDeviceGraphTrace instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

The unique ID of the graph that is launched.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_DEVICE_GRAPH_TRACE

Type:

int

launcher_graph_id

The unique ID of the graph that has launched this graph.

Type:

int

ptr

Get the pointer address to the data as Python int.

start

The start timestamp for the graph execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the graph.

Type:

int

stream_id

The ID of the stream where the graph is being launched.

Type:

int

class cupti.cupti.ActivityEnvironment

Bases: object

Empty-initialize an instance of CUpti_ActivityEnvironment.

See also

CUpti_ActivityEnvironment

data_

_py_anon_pod13:

device_id

The ID of the device

Type:

int

environment_kind

The kind of data reported in this record.

Type:

int

static from_buffer(buffer)

Create an ActivityEnvironment instance with the memory from the given buffer.

static from_data(data)

Create an ActivityEnvironment instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_environment_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityEnvironment instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_ENVIRONMENT.

Type:

int

ptr

Get the pointer address to the data as Python int.

timestamp

The timestamp when this sample was retrieved, in ns. A value of 0 indicates that timestamp information could not be collected for the marker.

Type:

int

class cupti.cupti.ActivityEnvironmentCooling

Bases: object

Empty-initialize an instance of CUpti_ActivityEnvironmentCooling.

See also

CUpti_ActivityEnvironmentCooling

fan_speed

The fan speed as percentage of maximum.

Type:

int

static from_buffer(buffer)

Create an ActivityEnvironmentCooling instance with the memory from the given buffer.

static from_data(data)

Create an ActivityEnvironmentCooling instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_environment_cooling_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityEnvironmentCooling instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityEnvironmentPower

Bases: object

Empty-initialize an instance of CUpti_ActivityEnvironmentPower.

See also

CUpti_ActivityEnvironmentPower

static from_buffer(buffer)

Create an ActivityEnvironmentPower instance with the memory from the given buffer.

static from_data(data)

Create an ActivityEnvironmentPower instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_environment_power_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityEnvironmentPower instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

power

int:

power_limit

int:

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityEnvironmentSpeed

Bases: object

Empty-initialize an instance of CUpti_ActivityEnvironmentSpeed.

See also

CUpti_ActivityEnvironmentSpeed

clocks_throttle_reasons

The clocks throttle reasons.

Type:

int

static from_buffer(buffer)

Create an ActivityEnvironmentSpeed instance with the memory from the given buffer.

static from_data(data)

Create an ActivityEnvironmentSpeed instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_environment_speed_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityEnvironmentSpeed instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

memory_clock

The memory frequency in MHz

Type:

int

pcie_link_gen

The PCIe link generation.

Type:

int

pcie_link_width

The PCIe link width.

Type:

int

ptr

Get the pointer address to the data as Python int.

sm_clock

The SM frequency in MHz

Type:

int

class cupti.cupti.ActivityEnvironmentTemperature

Bases: object

Empty-initialize an instance of CUpti_ActivityEnvironmentTemperature.

See also

CUpti_ActivityEnvironmentTemperature

static from_buffer(buffer)

Create an ActivityEnvironmentTemperature instance with the memory from the given buffer.

static from_data(data)

Create an ActivityEnvironmentTemperature instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_environment_temperature_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityEnvironmentTemperature instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

gpu_temperature

The GPU temperature in degrees C.

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityExternalCorrelation

Bases: object

Empty-initialize an instance of CUpti_ActivityExternalCorrelation.

See also

CUpti_ActivityExternalCorrelation

correlation_id

The correlation ID of the associated CUDA driver or runtime API record.

Type:

int

external_id

The correlation ID of the associated non-CUDA API record. The exact field in the associated external record depends on that record’s activity kind (externalKind).

Type:

int

external_kind

The kind of external API this record correlated to.

Type:

int

static from_buffer(buffer)

Create an ActivityExternalCorrelation instance with the memory from the given buffer.

static from_data(data)

Create an ActivityExternalCorrelation instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_external_correlation_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityExternalCorrelation instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The kind of this activity.

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityFunction

Bases: object

Empty-initialize an instance of CUpti_ActivityFunction.

See also

CUpti_ActivityFunction

context_id

The ID of the context where the function is launched.

Type:

int

static from_buffer(buffer)

Create an ActivityFunction instance with the memory from the given buffer.

static from_data(data)

Create an ActivityFunction instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_function_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityFunction instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

function_index

The function’s unique symbol index in the module.

Type:

int

id

ID to uniquely identify the record

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_FUNCTION.

Type:

int

module_id

The module ID in which this global/device function is present.

Type:

int

name

The name of the function. This name is shared across all activity records representing the same kernel, and so should not be modified.

Type:

str

pad

Undefined. Reserved for internal use.

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityGraphHostNode

Bases: object

Empty-initialize an instance of CUpti_ActivityGraphHostNode.

See also

CUpti_ActivityGraphHostNode

context_id

The ID of the CUDA context to which the waiting CUDA stream belongs.

Type:

int

correlation_id

The correlation ID of the graph host node operation. Each operation is assigned a unique correlation ID that is identical to the correlation ID in the driver API activity record that launched the operation.

Type:

int

device_id

The ID of the CUDA device to which the CUDA context and stream belong. The graph host node is executing on the CPU, but it is associated with the CUDA device through the CUDA context and stream.

Type:

int

end

The end timestamp. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the start time is unknown.

Type:

int

static from_buffer(buffer)

Create an ActivityGraphHostNode instance with the memory from the given buffer.

static from_data(data)

Create an ActivityGraphHostNode instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_graph_host_node_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityGraphHostNode instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

The unique ID of the graph that executed this host node through graph launch.

Type:

int

graph_node_id

The unique ID of the graph node that executed this host node through graph launch.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_GRAPH_HOST_NODE

Type:

int

process_id

The ID of the process where the host node is executing.

Type:

int

ptr

Get the pointer address to the data as Python int.

start

The start timestamp. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the start time is unknown.

Type:

int

stream_id

The ID of the stream that waits for the graph host node to finish.

Type:

int

thread_id

The ID of the thread where the host node is executing.

Type:

int

class cupti.cupti.ActivityGraphTrace2

Bases: object

Empty-initialize an instance of CUpti_ActivityGraphTrace2.

See also

CUpti_ActivityGraphTrace2

context_id

The ID of the context where the first node of the graph is executed. If this is INT_MAX, then the start is on the host.

Type:

int

correlation_id

The correlation ID of the graph launch. Each graph launch is assigned a unique correlation ID that is identical to the correlation ID in the driver API activity record that launched the graph.

Type:

int

device_id

The ID of the device where the first node of the graph is executed. If this is INT_MAX, then the start is on the host.

Type:

int

end

The end timestamp for the graph execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the graph.

Type:

int

end_context_id

The ID of the context where the last node of the graph is executed.

Type:

int

end_device_id

The ID of the device where last node of the graph is executed

Type:

int

static from_buffer(buffer)

Create an ActivityGraphTrace2 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityGraphTrace2 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_graph_trace2_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityGraphTrace2 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

The unique ID of the graph that is launched.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_GRAPH_TRACE

Type:

int

ptr

Get the pointer address to the data as Python int.

start

The start timestamp for the graph execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the graph.

Type:

int

stream_id

The ID of the stream where the graph is being launched.

Type:

int

class cupti.cupti.ActivityGreenContext2

Bases: object

Empty-initialize an instance of CUpti_ActivityGreenContext2.

See also

CUpti_ActivityGreenContext2

context_id

The context ID of the green context.

Type:

int

device_id

The device ID associated with the green context.

Type:

int

static from_buffer(buffer)

Create an ActivityGreenContext2 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityGreenContext2 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_green_context2_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityGreenContext2 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_GREEN_CONTEXT.

Type:

int

logical_tpc_mask

Interpreting the TPC Mask** For a green context with numTpcs=3, logicalTpcMaskSize=8, and logicalTpcMask={66,8,0,0,0,0,0,0}: - Word 0: Value 66 (binary: …01000010, showing lowest 8 bits of 32-bit word) indicates TPCs 1 and 6 - Word 1: Value 8 (binary: …00001000) indicates TPC 35 (bit 3 of word 1, i.e., global TPC ID 32+3) - Each word is 32 bits covering TPCs 0-31 (word 0), 32-63 (word 1), etc. - To check if TPC N is allocated: ``(logicalTpcMask[N/32] & (1U << (N%32))) - logicalTpcMaskSize indicates how many words in the array contain valid data

Type:

uint32

Type:

(array of length 32).Bitset of allocated TPC IDs represented as 32-bit words. Valid words are specified by logicalTpcMaskSize; unused entries are zeroed. The array supports up to 1024 TPCs (32 words × 32 bits). Each bit represents a logical TPC ID assigned to the green context. For example, if bit k is set, logical TPC ID k is assigned to the context. **Example

logical_tpc_mask_size

The size (in 32-bit words) of the logical TPC mask stored in logicalTpcMask[].

Type:

int

num_multiprocessors

The number of multiprocessors (SMs) allocated to the green context.

Type:

int

num_tpcs

The number of TPCs allocated to the green context.

Type:

int

padding

Undefined. Reserved for internal use.

Type:

int

parent_context_id

The ID of the parent context.

Type:

int

ptr

Get the pointer address to the data as Python int.

workqueue_concurrency_limit

The concurrency limit for the work queue associated with the green context. This defines the maximum number of concurrent operations allowed on this work queue.

Type:

int

workqueue_resource_id

The work queue resource ID associated with the green context. This is a unique identifier for the work queue resource used by this green context.

Type:

int

workqueue_sharing_scope

The sharing scope for the work queue associated with the green context. This defines how the work queue can be shared across different contexts or processes. Refer CUdevWorkqueueConfigScope in cuda.h for possible values.

Type:

int

Deprecated since version 13.3.0.

class cupti.cupti.ActivityGreenContext

Bases: object

Empty-initialize an instance of CUpti_ActivityGreenContext.

See also

CUpti_ActivityGreenContext

context_id

The context ID of the green context.

Type:

int

device_id

The device ID associated with the green context.

Type:

int

static from_buffer(buffer)

Create an ActivityGreenContext instance with the memory from the given buffer.

static from_data(data)

Create an ActivityGreenContext instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_green_context_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityGreenContext instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_GREEN_CONTEXT.

Type:

int

logical_tpc_mask

Interpreting the TPC Mask** For a green context with numTpcs=3, logicalTpcMaskSize=8, and logicalTpcMask={66,8,0,0,0,0,0,0}: - Word 0: Value 66 (binary: …01000010, showing lowest 8 bits of 32-bit word) indicates TPCs 1 and 6 - Word 1: Value 8 (binary: …00001000) indicates TPC 35 (bit 3 of word 1, i.e., global TPC ID 32+3) - Each word is 32 bits covering TPCs 0-31 (word 0), 32-63 (word 1), etc. - To check if TPC N is allocated: ``(logicalTpcMask[N/32] & (1U << (N%32))) - logicalTpcMaskSize indicates how many words in the array contain valid data

Type:

uint32

Type:

(array of length 32).Bitset of allocated TPC IDs represented as 32-bit words. Valid words are specified by logicalTpcMaskSize; unused entries are zeroed. The array supports up to 1024 TPCs (32 words × 32 bits). Each bit represents a logical TPC ID assigned to the green context. For example, if bit k is set, logical TPC ID k is assigned to the context. **Example

logical_tpc_mask_size

The size (in 32-bit words) of the logical TPC mask stored in logicalTpcMask[].

Type:

int

num_multiprocessors

The number of multiprocessors (SMs) allocated to the green context.

Type:

int

num_tpcs

The number of TPCs allocated to the green context.

Type:

int

padding

Undefined. Reserved for internal use.

Type:

int

parent_context_id

The ID of the parent context.

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityHostLaunch

Bases: object

Empty-initialize an instance of CUpti_ActivityHostLaunch.

See also

CUpti_ActivityHostLaunch

context_id

The ID of the CUDA context to which the waiting CUDA stream belongs.

Type:

int

correlation_id

The correlation ID of the host launch operation. Each operation is assigned a unique correlation ID that is identical to the correlation ID in the driver API activity record that launched the operation.

Type:

int

device_id

The ID of the CUDA device to which the CUDA context and stream belong. The host function is executing on the CPU, but it is associated with the CUDA device through the CUDA context and stream.

Type:

int

end

The end timestamp. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the start time is unknown.

Type:

int

static from_buffer(buffer)

Create an ActivityHostLaunch instance with the memory from the given buffer.

static from_data(data)

Create an ActivityHostLaunch instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_host_launch_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityHostLaunch instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_HOST_LAUNCH

Type:

int

padding

int:

process_id

The ID of the process where the host function is executing.

Type:

int

ptr

Get the pointer address to the data as Python int.

start

The start timestamp. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the start time is unknown.

Type:

int

stream_id

The ID of the CUDA context to which the waiting CUDA stream belongs.

Type:

int

thread_id

The ID of the thread where the host function is executing.

Type:

int

class cupti.cupti.ActivityJit2

Bases: object

Empty-initialize an instance of CUpti_ActivityJit2.

See also

CUpti_ActivityJit2

cache_path

The path where the fat binary is cached.

Type:

str

cache_size

The size of compute cache.

Type:

int

correlation_id

The correlation ID of the JIT operation to which records belong to. Each JIT operation is assigned a unique correlation ID that is identical to the correlation ID in the driver or runtime API activity record that launched the JIT operation.

Type:

int

device_id

The device ID.

Type:

int

end

The end timestamp for the JIT operation, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the JIT operation.

Type:

int

static from_buffer(buffer)

Create an ActivityJit2 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityJit2 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_jit2_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityJit2 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

jit_entry_type

The JIT entry type.

Type:

int

jit_operation_correlation_id

The correlation ID to correlate JIT compilation, load and store operations. Each JIT compilation unit is assigned a unique correlation ID at the time of the JIT compilation. This correlation id can be used to find the matching JIT cache load/store records.

Type:

int

jit_operation_type

The JIT operation type.

Type:

int

kind

The activity record kind must be CUPTI_ACTIVITY_KIND_JIT.

Type:

int

padding

Internal use.

Type:

int

process_id

The ID of the process where the JIT operation is executing.

Type:

int

ptr

Get the pointer address to the data as Python int.

start

The start timestamp for the JIT operation, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the JIT operation.

Type:

int

thread_id

The ID of the thread where the JIT operation is executing.

Type:

int

class cupti.cupti.ActivityKernel12

Bases: object

Empty-initialize an instance of CUpti_ActivityKernel12.

See also

CUpti_ActivityKernel12

block_x

The X-dimension block size for the kernel.

Type:

int

block_y

The Y-dimension block size for the kernel.

Type:

int

block_z

The Z-dimension grid size for the kernel.

Type:

int

cache_config

For devices with compute capability 7.5+ cacheConfig values are not updated in case field isSharedMemoryCarveoutRequested is set

Type:

_py_anon_pod56

channel_id

The ID of the HW channel on which the kernel is launched.

Type:

int

channel_type

The type of the channel

Type:

int

cluster_scheduling_policy

The cluster scheduling policy for the kernel. Refer CUclusterSchedulingPolicy Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_x

The X-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_y

The Y-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_z

The Z-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

completed

The completed timestamp for the kernel execution, in ns. It represents the completion of all it’s child kernels and the kernel itself. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the completion time is unknown.

Type:

int

context_id

The ID of the context where the kernel is executing.

Type:

int

correlation_id

The correlation ID of the kernel. Each kernel execution is assigned a unique correlation ID that is identical to the correlation ID in the driver or runtime API activity record that launched the kernel.

Type:

int

device_id

The ID of the device where the kernel is executing.

Type:

int

dynamic_shared_memory

The dynamic shared memory reserved for the kernel, in bytes.

Type:

int

end

The end timestamp for the kernel execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the kernel.

Type:

int

execution_model

The execution model of the kernel function. CUpti_FuncExecutionModel

Type:

int

static from_buffer(buffer)

Create an ActivityKernel12 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityKernel12 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_kernel12_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityKernel12 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

The unique ID of the graph that launched this kernel through graph launch APIs. This field will be 0 if the kernel is not launched through graph launch APIs.

Type:

int

graph_node_id

The unique ID of the graph node that launched this kernel through graph launch APIs. This field will be 0 if the kernel is not launched through graph launch APIs.

Type:

int

grid_id

The grid ID of the kernel. Each kernel is assigned a unique grid ID at runtime.

Type:

int

grid_x

The X-dimension grid size for the kernel.

Type:

int

grid_y

The Y-dimension grid size for the kernel.

Type:

int

grid_z

The Z-dimension grid size for the kernel.

Type:

int

is_device_launched

This field is set to 1 if the kernel is part of a device launched graph.

Type:

int

is_shared_memory_carveout_requested

This indicates if CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT was updated for the kernel launch

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_KERNEL or CUPTI_ACTIVITY_KIND_CONCURRENT_KERNEL.

Type:

int

launch_type

The indicates if the kernel was executed via a regular launch or via a single/multi device cooperative launch. CUpti_ActivityLaunchType

Type:

int

local_memory_per_thread

The amount of local memory reserved for each thread, in bytes.

Type:

int

local_memory_total

The total amount of local memory reserved for the kernel, in bytes (deprecated in CUDA 11.8). Refer field localMemoryTotal_v2

Type:

int

local_memory_total_v2

The total amount of local memory reserved for the kernel, in bytes.

Type:

int

max_active_clusters

The maximum clusters that could co-exist on the target device for the kernel

Type:

int

max_potential_cluster_size

The maximum cluster size for the kernel

Type:

int

name

The name of the kernel. This name is shared across all activity records representing the same kernel, and so should not be modified.

Type:

str

p_access_policy_window

The pointer to the access policy window. The structure CUaccessPolicyWindow is defined in cuda.h.

Type:

int

padding

Undefined. Reserved for internal use.

Type:

int

padding3

(array of length 7).

Type:

uint8

padding4

int:

padding5

int:

partitioned_global_cache_executed

The partitioned global caching executed for the kernel. Partitioned global caching is required to enable caching on certain chips, such as devices with compute capability 5.2. Partitioned global caching can be automatically disabled if the occupancy requirement of the launch cannot support caching.

Type:

int

partitioned_global_cache_requested

The partitioned global caching requested for the kernel. Partitioned global caching is required to enable caching on certain chips, such as devices with compute capability 5.2.

Type:

int

priority

The launch priority of the kernel.

Type:

int

ptr

Get the pointer address to the data as Python int.

queued

The timestamp when the kernel is queued up in the command buffer, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the queued time could not be collected for the kernel. This timestamp is not collected by default. Use API cuptiActivityEnableLatencyTimestamps() to enable collection. Command buffer is a buffer written by CUDA driver to send commands like kernel launch, memory copy etc to the GPU. All launches of CUDA kernels are asynchronous with respect to the host, the host requests the launch by writing commands into the command buffer, then returns without checking the GPU’s progress.

Type:

int

registers_per_thread

The number of registers required for each thread executing the kernel.

Type:

int

reserved0

Undefined. Reserved for internal use.

Type:

int

shared_memory_carveout_requested

Shared memory carveout value requested for the function in percentage of the total resource. The value will be updated only if field isSharedMemoryCarveoutRequested is set.

Type:

int

shared_memory_config

The shared memory configuration used for the kernel. The value is one of the CUsharedconfig enumeration values from cuda.h.

Type:

int

shared_memory_executed

Shared memory size set by the driver.

Type:

int

shmem_limit_config

The shared memory limit config for the kernel. This field shows whether user has opted for a higher per block limit of dynamic shared memory.

Type:

int

start

The start timestamp for the kernel execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the kernel.

Type:

int

static_shared_memory

The static shared memory allocated for the kernel, in bytes.

Type:

int

stream_id

The ID of the stream where the kernel is executing.

Type:

int

submitted

The timestamp when the command buffer containing the kernel launch is submitted to the GPU, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the submitted time could not be collected for the kernel. This timestamp is not collected by default. Use API cuptiActivityEnableLatencyTimestamps() to enable collection.

Type:

int

Deprecated since version 13.3.0.

class cupti.cupti.ActivityKernel11

Bases: object

Empty-initialize an instance of CUpti_ActivityKernel11.

See also

CUpti_ActivityKernel11

block_x

The X-dimension block size for the kernel.

Type:

int

block_y

The Y-dimension block size for the kernel.

Type:

int

block_z

The Z-dimension grid size for the kernel.

Type:

int

cache_config

For devices with compute capability 7.5+ cacheConfig values are not updated in case field isSharedMemoryCarveoutRequested is set

Type:

_py_anon_pod52

channel_id

The ID of the HW channel on which the kernel is launched.

Type:

int

channel_type

The type of the channel

Type:

int

cluster_scheduling_policy

The cluster scheduling policy for the kernel. Refer CUclusterSchedulingPolicy Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_x

The X-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_y

The Y-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_z

The Z-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

completed

The completed timestamp for the kernel execution, in ns. It represents the completion of all it’s child kernels and the kernel itself. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the completion time is unknown.

Type:

int

context_id

The ID of the context where the kernel is executing.

Type:

int

correlation_id

The correlation ID of the kernel. Each kernel execution is assigned a unique correlation ID that is identical to the correlation ID in the driver or runtime API activity record that launched the kernel.

Type:

int

device_id

The ID of the device where the kernel is executing.

Type:

int

dynamic_shared_memory

The dynamic shared memory reserved for the kernel, in bytes.

Type:

int

end

The end timestamp for the kernel execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the kernel.

Type:

int

static from_buffer(buffer)

Create an ActivityKernel11 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityKernel11 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_kernel11_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityKernel11 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

The unique ID of the graph that launched this kernel through graph launch APIs. This field will be 0 if the kernel is not launched through graph launch APIs.

Type:

int

graph_node_id

The unique ID of the graph node that launched this kernel through graph launch APIs. This field will be 0 if the kernel is not launched through graph launch APIs.

Type:

int

grid_id

The grid ID of the kernel. Each kernel is assigned a unique grid ID at runtime.

Type:

int

grid_x

The X-dimension grid size for the kernel.

Type:

int

grid_y

The Y-dimension grid size for the kernel.

Type:

int

grid_z

The Z-dimension grid size for the kernel.

Type:

int

is_device_launched

This field is set to 1 if the kernel is part of a device launched graph.

Type:

int

is_shared_memory_carveout_requested

This indicates if CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT was updated for the kernel launch

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_KERNEL or CUPTI_ACTIVITY_KIND_CONCURRENT_KERNEL.

Type:

int

launch_type

The indicates if the kernel was executed via a regular launch or via a single/multi device cooperative launch. CUpti_ActivityLaunchType

Type:

int

local_memory_per_thread

The amount of local memory reserved for each thread, in bytes.

Type:

int

local_memory_total

The total amount of local memory reserved for the kernel, in bytes (deprecated in CUDA 11.8). Refer field localMemoryTotal_v2

Type:

int

local_memory_total_v2

The total amount of local memory reserved for the kernel, in bytes.

Type:

int

max_active_clusters

The maximum clusters that could co-exist on the target device for the kernel

Type:

int

max_potential_cluster_size

The maximum cluster size for the kernel

Type:

int

name

The name of the kernel. This name is shared across all activity records representing the same kernel, and so should not be modified.

Type:

str

p_access_policy_window

The pointer to the access policy window. The structure CUaccessPolicyWindow is defined in cuda.h.

Type:

int

padding

Undefined. Reserved for internal use.

Type:

int

padding3

(array of length 7).

Type:

uint8

padding4

int:

partitioned_global_cache_executed

The partitioned global caching executed for the kernel. Partitioned global caching is required to enable caching on certain chips, such as devices with compute capability 5.2. Partitioned global caching can be automatically disabled if the occupancy requirement of the launch cannot support caching.

Type:

int

partitioned_global_cache_requested

The partitioned global caching requested for the kernel. Partitioned global caching is required to enable caching on certain chips, such as devices with compute capability 5.2.

Type:

int

priority

The launch priority of the kernel.

Type:

int

ptr

Get the pointer address to the data as Python int.

queued

The timestamp when the kernel is queued up in the command buffer, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the queued time could not be collected for the kernel. This timestamp is not collected by default. Use API cuptiActivityEnableLatencyTimestamps() to enable collection. Command buffer is a buffer written by CUDA driver to send commands like kernel launch, memory copy etc to the GPU. All launches of CUDA kernels are asynchronous with respect to the host, the host requests the launch by writing commands into the command buffer, then returns without checking the GPU’s progress.

Type:

int

registers_per_thread

The number of registers required for each thread executing the kernel.

Type:

int

reserved0

Undefined. Reserved for internal use.

Type:

int

shared_memory_carveout_requested

Shared memory carveout value requested for the function in percentage of the total resource. The value will be updated only if field isSharedMemoryCarveoutRequested is set.

Type:

int

shared_memory_config

The shared memory configuration used for the kernel. The value is one of the CUsharedconfig enumeration values from cuda.h.

Type:

int

shared_memory_executed

Shared memory size set by the driver.

Type:

int

shmem_limit_config

The shared memory limit config for the kernel. This field shows whether user has opted for a higher per block limit of dynamic shared memory.

Type:

int

start

The start timestamp for the kernel execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the kernel.

Type:

int

static_shared_memory

The static shared memory allocated for the kernel, in bytes.

Type:

int

stream_id

The ID of the stream where the kernel is executing.

Type:

int

submitted

The timestamp when the command buffer containing the kernel launch is submitted to the GPU, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the submitted time could not be collected for the kernel. This timestamp is not collected by default. Use API cuptiActivityEnableLatencyTimestamps() to enable collection.

Type:

int

Deprecated since version 13.2.0.

class cupti.cupti.ActivityKernel10

Bases: object

Empty-initialize an instance of CUpti_ActivityKernel10.

See also

CUpti_ActivityKernel10

block_x

The X-dimension block size for the kernel.

Type:

int

block_y

The Y-dimension block size for the kernel.

Type:

int

block_z

The Z-dimension grid size for the kernel.

Type:

int

cache_config

For devices with compute capability 7.5+ cacheConfig values are not updated in case field isSharedMemoryCarveoutRequested is set

Type:

_py_anon_pod50

channel_id

The ID of the HW channel on which the kernel is launched.

Type:

int

channel_type

The type of the channel

Type:

int

cluster_scheduling_policy

The cluster scheduling policy for the kernel. Refer CUclusterSchedulingPolicy Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_x

The X-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_y

The Y-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_z

The Z-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

completed

The completed timestamp for the kernel execution, in ns. It represents the completion of all it’s child kernels and the kernel itself. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the completion time is unknown.

Type:

int

context_id

The ID of the context where the kernel is executing.

Type:

int

correlation_id

The correlation ID of the kernel. Each kernel execution is assigned a unique correlation ID that is identical to the correlation ID in the driver or runtime API activity record that launched the kernel.

Type:

int

device_id

The ID of the device where the kernel is executing.

Type:

int

dynamic_shared_memory

The dynamic shared memory reserved for the kernel, in bytes.

Type:

int

end

The end timestamp for the kernel execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the kernel.

Type:

int

static from_buffer(buffer)

Create an ActivityKernel10 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityKernel10 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_kernel10_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityKernel10 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

The unique ID of the graph that launched this kernel through graph launch APIs. This field will be 0 if the kernel is not launched through graph launch APIs.

Type:

int

graph_node_id

The unique ID of the graph node that launched this kernel through graph launch APIs. This field will be 0 if the kernel is not launched through graph launch APIs.

Type:

int

grid_id

The grid ID of the kernel. Each kernel is assigned a unique grid ID at runtime.

Type:

int

grid_x

The X-dimension grid size for the kernel.

Type:

int

grid_y

The Y-dimension grid size for the kernel.

Type:

int

grid_z

The Z-dimension grid size for the kernel.

Type:

int

is_device_launched

This field is set to 1 if the kernel is part of a device launched graph.

Type:

int

is_shared_memory_carveout_requested

This indicates if CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT was updated for the kernel launch

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_KERNEL or CUPTI_ACTIVITY_KIND_CONCURRENT_KERNEL.

Type:

int

launch_type

The indicates if the kernel was executed via a regular launch or via a single/multi device cooperative launch. CUpti_ActivityLaunchType

Type:

int

local_memory_per_thread

The amount of local memory reserved for each thread, in bytes.

Type:

int

local_memory_total

The total amount of local memory reserved for the kernel, in bytes (deprecated in CUDA 11.8). Refer field localMemoryTotal_v2

Type:

int

local_memory_total_v2

The total amount of local memory reserved for the kernel, in bytes.

Type:

int

max_active_clusters

The maximum clusters that could co-exist on the target device for the kernel

Type:

int

max_potential_cluster_size

The maximum cluster size for the kernel

Type:

int

name

The name of the kernel. This name is shared across all activity records representing the same kernel, and so should not be modified.

Type:

str

p_access_policy_window

The pointer to the access policy window. The structure CUaccessPolicyWindow is defined in cuda.h.

Type:

int

padding

Undefined. Reserved for internal use.

Type:

int

padding3

(array of length 7).

Type:

uint8

partitioned_global_cache_executed

The partitioned global caching executed for the kernel. Partitioned global caching is required to enable caching on certain chips, such as devices with compute capability 5.2. Partitioned global caching can be automatically disabled if the occupancy requirement of the launch cannot support caching.

Type:

int

partitioned_global_cache_requested

The partitioned global caching requested for the kernel. Partitioned global caching is required to enable caching on certain chips, such as devices with compute capability 5.2.

Type:

int

ptr

Get the pointer address to the data as Python int.

queued

The timestamp when the kernel is queued up in the command buffer, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the queued time could not be collected for the kernel. This timestamp is not collected by default. Use API cuptiActivityEnableLatencyTimestamps() to enable collection. Command buffer is a buffer written by CUDA driver to send commands like kernel launch, memory copy etc to the GPU. All launches of CUDA kernels are asynchronous with respect to the host, the host requests the launch by writing commands into the command buffer, then returns without checking the GPU’s progress.

Type:

int

registers_per_thread

The number of registers required for each thread executing the kernel.

Type:

int

reserved0

Undefined. Reserved for internal use.

Type:

int

shared_memory_carveout_requested

Shared memory carveout value requested for the function in percentage of the total resource. The value will be updated only if field isSharedMemoryCarveoutRequested is set.

Type:

int

shared_memory_config

The shared memory configuration used for the kernel. The value is one of the CUsharedconfig enumeration values from cuda.h.

Type:

int

shared_memory_executed

Shared memory size set by the driver.

Type:

int

shmem_limit_config

The shared memory limit config for the kernel. This field shows whether user has opted for a higher per block limit of dynamic shared memory.

Type:

int

start

The start timestamp for the kernel execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the kernel.

Type:

int

static_shared_memory

The static shared memory allocated for the kernel, in bytes.

Type:

int

stream_id

The ID of the stream where the kernel is executing.

Type:

int

submitted

The timestamp when the command buffer containing the kernel launch is submitted to the GPU, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the submitted time could not be collected for the kernel. This timestamp is not collected by default. Use API cuptiActivityEnableLatencyTimestamps() to enable collection.

Type:

int

Deprecated since version 13.0.0.

class cupti.cupti.ActivityKernel9

Bases: object

Empty-initialize an instance of CUpti_ActivityKernel9.

See also

CUpti_ActivityKernel9

block_x

The X-dimension block size for the kernel.

Type:

int

block_y

The Y-dimension block size for the kernel.

Type:

int

block_z

The Z-dimension grid size for the kernel.

Type:

int

cache_config

For devices with compute capability 7.0+ cacheConfig values are not updated in case field isSharedMemoryCarveoutRequested is set

Type:

_py_anon_pod7

channel_id

The ID of the HW channel on which the kernel is launched.

Type:

int

channel_type

The type of the channel

Type:

int

cluster_scheduling_policy

The cluster scheduling policy for the kernel. Refer CUclusterSchedulingPolicy Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_x

The X-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_y

The Y-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

cluster_z

The Z-dimension cluster size for the kernel. Field is valid for devices with compute capability 9.0 and higher

Type:

int

completed

The completed timestamp for the kernel execution, in ns. It represents the completion of all it’s child kernels and the kernel itself. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the completion time is unknown.

Type:

int

context_id

The ID of the context where the kernel is executing.

Type:

int

correlation_id

The correlation ID of the kernel. Each kernel execution is assigned a unique correlation ID that is identical to the correlation ID in the driver or runtime API activity record that launched the kernel.

Type:

int

device_id

The ID of the device where the kernel is executing.

Type:

int

dynamic_shared_memory

The dynamic shared memory reserved for the kernel, in bytes.

Type:

int

end

The end timestamp for the kernel execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the kernel.

Type:

int

static from_buffer(buffer)

Create an ActivityKernel9 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityKernel9 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_kernel9_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityKernel9 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

The unique ID of the graph that launched this kernel through graph launch APIs. This field will be 0 if the kernel is not launched through graph launch APIs.

Type:

int

graph_node_id

The unique ID of the graph node that launched this kernel through graph launch APIs. This field will be 0 if the kernel is not launched through graph launch APIs.

Type:

int

grid_id

The grid ID of the kernel. Each kernel is assigned a unique grid ID at runtime.

Type:

int

grid_x

The X-dimension grid size for the kernel.

Type:

int

grid_y

The Y-dimension grid size for the kernel.

Type:

int

grid_z

The Z-dimension grid size for the kernel.

Type:

int

is_shared_memory_carveout_requested

This indicates if CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT was updated for the kernel launch

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_KERNEL or CUPTI_ACTIVITY_KIND_CONCURRENT_KERNEL.

Type:

int

launch_type

The indicates if the kernel was executed via a regular launch or via a single/multi device cooperative launch. CUpti_ActivityLaunchType

Type:

int

local_memory_per_thread

The amount of local memory reserved for each thread, in bytes.

Type:

int

local_memory_total

The total amount of local memory reserved for the kernel, in bytes (deprecated in CUDA 11.8). Refer field localMemoryTotal_v2

Type:

int

local_memory_total_v2

The total amount of local memory reserved for the kernel, in bytes.

Type:

int

max_active_clusters

The maximum clusters that could co-exist on the target device for the kernel

Type:

int

max_potential_cluster_size

The maximum cluster size for the kernel

Type:

int

name

The name of the kernel. This name is shared across all activity records representing the same kernel, and so should not be modified.

Type:

str

p_access_policy_window

The pointer to the access policy window. The structure CUaccessPolicyWindow is defined in cuda.h.

Type:

int

padding

Undefined. Reserved for internal use.

Type:

int

partitioned_global_cache_executed

The partitioned global caching executed for the kernel. Partitioned global caching is required to enable caching on certain chips, such as devices with compute capability 5.2. Partitioned global caching can be automatically disabled if the occupancy requirement of the launch cannot support caching.

Type:

int

partitioned_global_cache_requested

The partitioned global caching requested for the kernel. Partitioned global caching is required to enable caching on certain chips, such as devices with compute capability 5.2.

Type:

int

ptr

Get the pointer address to the data as Python int.

queued

The timestamp when the kernel is queued up in the command buffer, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the queued time could not be collected for the kernel. This timestamp is not collected by default. Use API cuptiActivityEnableLatencyTimestamps() to enable collection. Command buffer is a buffer written by CUDA driver to send commands like kernel launch, memory copy etc to the GPU. All launches of CUDA kernels are asynchronous with respect to the host, the host requests the launch by writing commands into the command buffer, then returns without checking the GPU’s progress.

Type:

int

registers_per_thread

The number of registers required for each thread executing the kernel.

Type:

int

reserved0

Undefined. Reserved for internal use.

Type:

int

shared_memory_carveout_requested

Shared memory carveout value requested for the function in percentage of the total resource. The value will be updated only if field isSharedMemoryCarveoutRequested is set.

Type:

int

shared_memory_config

The shared memory configuration used for the kernel. The value is one of the CUsharedconfig enumeration values from cuda.h.

Type:

int

shared_memory_executed

Shared memory size set by the driver.

Type:

int

shmem_limit_config

The shared memory limit config for the kernel. This field shows whether user has opted for a higher per block limit of dynamic shared memory.

Type:

int

start

The start timestamp for the kernel execution, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the kernel.

Type:

int

static_shared_memory

The static shared memory allocated for the kernel, in bytes.

Type:

int

stream_id

The ID of the stream where the kernel is executing.

Type:

int

submitted

The timestamp when the command buffer containing the kernel launch is submitted to the GPU, in ns. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the submitted time could not be collected for the kernel. This timestamp is not collected by default. Use API cuptiActivityEnableLatencyTimestamps() to enable collection.

Type:

int

class cupti.cupti.ActivityMarker2

Bases: object

Empty-initialize an instance of CUpti_ActivityMarker2.

See also

CUpti_ActivityMarker2

domain

The name of the domain to which this marker belongs to. This will be NULL for default domain.

Type:

str

flags_

The flags associated with the marker. CUpti_ActivityFlag

Type:

int

static from_buffer(buffer)

Create an ActivityMarker2 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMarker2 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_marker2_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityMarker2 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

id

The marker ID.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MARKER.

Type:

int

name

The marker name for an instantaneous or start marker. This will be NULL for an end marker.

Type:

str

object_id

The identifier for the activity object associated with this marker. ‘objectKind’ indicates which ID is valid for this record.

Type:

ActivityObjectKindId

object_kind

The kind of activity object associated with this marker.

Type:

int

pad

Undefined. Reserved for internal use.

Type:

int

ptr

Get the pointer address to the data as Python int.

timestamp

The timestamp for the marker, in ns. A value of 0 indicates that timestamp information could not be collected for the marker.

Type:

int

class cupti.cupti.ActivityMarkerData2

Bases: object

Empty-initialize an instance of CUpti_ActivityMarkerData2.

See also

CUpti_ActivityMarkerData2

category

The category for the marker.

Type:

int

color

The color for the marker.

Type:

int

cupti_domain_id

CUPTI maintained domain id required for NVTX extended payloads. To parse the payload correctly, the domain id must be used to identify the payload attributes as they are domain specific.

Type:

int

flags_

The flags associated with the marker. CUpti_ActivityFlag

Type:

int

static from_buffer(buffer)

Create an ActivityMarkerData2 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMarkerData2 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_marker_data2_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityMarkerData2 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

id

The marker ID.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MARKER_DATA.

Type:

int

padding

Reserved for internal use.

Type:

int

payload

The payload value.

Type:

MetricValue

payload_kind

Defines the payload format for the value associated with the marker.

Type:

int

ptr

Get the pointer address to the data as Python int.

Deprecated since version 13.2.0.

class cupti.cupti.ActivityMarkerData

Bases: object

Empty-initialize an instance of CUpti_ActivityMarkerData.

See also

CUpti_ActivityMarkerData

category

The category for the marker.

Type:

int

color

The payload value. The color for the marker.

Type:

int

flags_

The flags associated with the marker. CUpti_ActivityFlag

Type:

int

static from_buffer(buffer)

Create an ActivityMarkerData instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMarkerData instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_marker_data_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityMarkerData instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

id

The marker ID.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MARKER_DATA.

Type:

int

payload

MetricValue:

payload_kind

Defines the payload format for the value associated with the marker.

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityMemDecompress

Bases: object

Empty-initialize an instance of CUpti_ActivityMemDecompress.

See also

CUpti_ActivityMemDecompress

channel_id

The ID of the HW channel on which the memory copy is occurring.

Type:

int

channel_type

The type of the channel

Type:

int

context_id

The ID of the context.

Type:

int

correlation_id

The correlation ID of the decompression operations. Each operation is assigned a unique correlation ID that is identical to the correlation ID in the driver API activity record that launched the operation.

Type:

int

device_id

The ID of the device.

Type:

int

end

The end timestamp. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the start time is unknown.

Type:

int

static from_buffer(buffer)

Create an ActivityMemDecompress instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMemDecompress instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_mem_decompress_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityMemDecompress instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MEM_DECOMPRESS

Type:

int

number_of_operations

The number of operations in the batch.

Type:

int

ptr

Get the pointer address to the data as Python int.

reserved0

This field is reserved for internal use

Type:

int

source_bytes

The number of bytes to be read and decompressed in the batch operation.

Type:

int

start

The start timestamp. A value of CUPTI_TIMESTAMP_UNKNOWN indicates that the start time is unknown.

Type:

int

stream_id

The ID of the stream.

Type:

int

class cupti.cupti.ActivityMemcpy6

Bases: object

Empty-initialize an instance of CUpti_ActivityMemcpy6.

See also

CUpti_ActivityMemcpy6

bytes

The number of bytes transferred by the memory copy.

Type:

int

channel_id

The ID of the HW channel on which the memory copy is occurring.

Type:

int

channel_type

The type of the channel

Type:

int

context_id

The ID of the context where the memory copy is occurring.

Type:

int

copy_count

The total number of memcopy operations traced in this record. This field is valid for memcpy operations happening using MemcpyBatchAsync APIs in CUDA. In MemcpyBatchAsync APIs, multiple memcpy operations are batched together for optimization purposes based on certain heuristics. For other memcpy operations, this field will be 1.

Type:

int

copy_kind

The kind of the memory copy, stored as a byte to reduce record size. CUpti_ActivityMemcpyKind

Type:

int

correlation_id

The correlation ID of the memory copy. Each memory copy is assigned a unique correlation ID that is identical to the correlation ID in the driver API activity record that launched the memory copy.

Type:

int

device_id

The ID of the device where the memory copy is occurring.

Type:

int

dst_kind

The destination memory kind read by the memory copy, stored as a byte to reduce record size. CUpti_ActivityMemoryKind

Type:

int

end

The end timestamp for the memory copy, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the memory copy.

Type:

int

flags_

The flags associated with the memory copy. CUpti_ActivityFlag

Type:

int

static from_buffer(buffer)

Create an ActivityMemcpy6 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMemcpy6 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_memcpy6_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityMemcpy6 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

The unique ID of the graph that executed this memcpy through graph launch. This field will be 0 if the memcpy is not done through graph launch.

Type:

int

graph_node_id

The unique ID of the graph node that executed this memcpy through graph launch. This field will be 0 if the memcpy is not done through graph launch.

Type:

int

is_device_launched

This field is used to indicate if the memcpy operation is part of a device graph launch.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MEMCPY.

Type:

int

pad

Undefined. Reserved for internal use.

Type:

int

pad2

(array of length 3).Reserved for internal use.

Type:

uint8

ptr

Get the pointer address to the data as Python int.

reserved0

Undefined. Reserved for internal use.

Type:

int

runtime_correlation_id

The runtime correlation ID of the memory copy. Each memory copy is assigned a unique runtime correlation ID that is identical to the correlation ID in the runtime API activity record that launched the memory copy.

Type:

int

src_kind

The source memory kind read by the memory copy, stored as a byte to reduce record size. CUpti_ActivityMemoryKind

Type:

int

start

The start timestamp for the memory copy, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the memory copy.

Type:

int

stream_id

The ID of the stream where the memory copy is occurring.

Type:

int

Deprecated since version 12.8.0.

class cupti.cupti.ActivityMemcpy5

Bases: object

Empty-initialize an instance of CUpti_ActivityMemcpy5.

See also

CUpti_ActivityMemcpy5

bytes

int:

channel_id

int:

channel_type

int:

context_id

int:

copy_kind

int:

correlation_id

int:

device_id

int:

dst_kind

int:

end

int:

flags_

int:

static from_buffer(buffer)

Create an ActivityMemcpy5 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMemcpy5 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_memcpy5_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityMemcpy5 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

int:

graph_node_id

int:

kind

int:

pad

int:

pad2

int:

ptr

Get the pointer address to the data as Python int.

reserved0

int:

runtime_correlation_id

int:

src_kind

int:

start

int:

stream_id

int:

class cupti.cupti.ActivityMemcpyPtoP4

Bases: object

Empty-initialize an instance of CUpti_ActivityMemcpyPtoP4.

See also

CUpti_ActivityMemcpyPtoP4

bytes

The number of bytes transferred by the memory copy.

Type:

int

channel_id

The ID of the HW channel on which the memory copy is occurring.

Type:

int

channel_type

The type of the channel

Type:

int

context_id

The ID of the context where the memory copy is occurring.

Type:

int

copy_kind

The kind of the memory copy, stored as a byte to reduce record size. CUpti_ActivityMemcpyKind

Type:

int

correlation_id

The correlation ID of the memory copy. Each memory copy is assigned a unique correlation ID that is identical to the correlation ID in the driver and runtime API activity record that launched the memory copy.

Type:

int

device_id

The ID of the device where the memory copy is occurring.

Type:

int

dst_context_id

The ID of the context owning the memory being copied to.

Type:

int

dst_device_id

The ID of the device where memory is being copied to.

Type:

int

dst_kind

The destination memory kind read by the memory copy, stored as a byte to reduce record size. CUpti_ActivityMemoryKind

Type:

int

end

The end timestamp for the memory copy, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the memory copy.

Type:

int

flags_

The flags associated with the memory copy. CUpti_ActivityFlag

Type:

int

static from_buffer(buffer)

Create an ActivityMemcpyPtoP4 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMemcpyPtoP4 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_memcpy_pto_p4_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityMemcpyPtoP4 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

The unique ID of the graph that executed this memcpy through graph launch. This field will be 0 if the memcpy is not done through graph launch.

Type:

int

graph_node_id

The unique ID of the graph node that executed the memcpy through graph launch. This field will be 0 if memcpy is not done using graph launch.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MEMCPY2.

Type:

int

ptr

Get the pointer address to the data as Python int.

reserved0

Undefined. Reserved for internal use.

Type:

int

src_context_id

The ID of the context owning the memory being copied from.

Type:

int

src_device_id

The ID of the device where memory is being copied from.

Type:

int

src_kind

The source memory kind read by the memory copy, stored as a byte to reduce record size. CUpti_ActivityMemoryKind

Type:

int

start

The start timestamp for the memory copy, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the memory copy.

Type:

int

stream_id

The ID of the stream where the memory copy is occurring.

Type:

int

class cupti.cupti.ActivityMemory

Bases: object

Empty-initialize an instance of CUpti_ActivityMemory.

See also

CUpti_ActivityMemory

address

The virtual address of the allocation

Type:

int

alloc_pc

The program counter of the allocation of memory

Type:

int

bytes

The number of bytes of memory allocated.

Type:

int

context_id

The ID of the context. If context is NULL, contextId is set to CUPTI_INVALID_CONTEXT_ID.

Type:

int

device_id

The ID of the device where the memory allocation is taking place.

Type:

int

end

The end timestamp for the memory operation, i.e. the time when memory was freed, in ns. This will be 0 if memory is not freed in the application

Type:

int

free_pc

The program counter of the freeing of memory. This will be 0 if memory is not freed in the application

Type:

int

static from_buffer(buffer)

Create an ActivityMemory instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMemory instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_memory_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityMemory instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MEMORY

Type:

int

memory_kind

The memory kind requested by the user

Type:

int

name

Variable name. This name is shared across all activity records representing the same symbol, and so should not be modified.

Type:

str

pad

Undefined. Reserved for internal use.

Type:

int

process_id

The ID of the process to which this record belongs to.

Type:

int

ptr

Get the pointer address to the data as Python int.

start

The start timestamp for the memory operation, i.e. the time when memory was allocated, in ns.

Type:

int

class cupti.cupti.ActivityMemory4

Bases: object

Empty-initialize an instance of CUpti_ActivityMemory4.

See also

CUpti_ActivityMemory4

address

The virtual address of the allocation. The base address of the memory pool.

Type:

int

bytes

The number of bytes of memory allocated.

Type:

int

context_id

The ID of the context. If context is NULL, contextId is set to CUPTI_INVALID_CONTEXT_ID.

Type:

int

correlation_id

The correlation ID of the memory operation. Each memory operation is assigned a unique correlation ID that is identical to the correlation ID in the driver and runtime API activity record that launched the memory operation.

Type:

int

device_id

The ID of the device where the memory operation is taking place.

Type:

int

static from_buffer(buffer)

Create an ActivityMemory4 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMemory4 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_memory4_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityMemory4 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

is_async

isAsync is set if memory operation happens through async memory APIs.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MEMORY2

Type:

int

memory_kind

The memory kind requested by the user, CUpti_ActivityMemoryKind.

Type:

int

memory_operation_type

The memory operation requested by the user, CUpti_ActivityMemoryOperationType.

Type:

int

memory_pool_config

The memory pool configuration used for the memory operations.

Type:

_py_anon_pod5

name

Variable name. This name is shared across all activity records representing the same symbol, and so should not be modified.

Type:

str

pad1

Undefined. Reserved for internal use.

Type:

int

pc

The program counter of the memory operation.

Type:

int

process_id

int:

ptr

Get the pointer address to the data as Python int.

source

The shared object or binary that the memory allocation request comes from.

Type:

str

stream_id

The ID of the stream. If memory operation is not async, streamId is set to CUPTI_INVALID_STREAM_ID.

Type:

int

timestamp

The start timestamp for the memory operation, in ns.

Type:

int

class cupti.cupti.ActivityMemoryPool3

Bases: object

Empty-initialize an instance of CUpti_ActivityMemoryPool3.

See also

CUpti_ActivityMemoryPool3

address

The virtual address of the allocation.

Type:

int

correlation_id

The correlation ID of the memory pool operation. Each memory pool operation is assigned a unique correlation ID that is identical to the correlation ID in the driver and runtime API activity record that launched the memory operation.

Type:

int

device_id

The ID of the device where the memory pool is created.

Type:

int

static from_buffer(buffer)

Create an ActivityMemoryPool3 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMemoryPool3 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_memory_pool3_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityMemoryPool3 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

is_managed_pool

Whether the pool is of managed memory allocation or pinned memory allocation. If it is 0, it is pinned and if it is 1, the memory pool allocation is of managed memory type.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MEMORY_POOL

Type:

int

memory_pool_operation_type

The memory operation requested by the user, CUpti_ActivityMemoryPoolOperationType.

Type:

int

memory_pool_type

The type of the memory pool, CUpti_ActivityMemoryPoolType

Type:

int

min_bytes_to_keep

The minimum bytes to keep of the memory pool. minBytesToKeep is valid for CUPTI_ACTIVITY_MEMORY_POOL_OPERATION_TYPE_TRIMMED, CUpti_ActivityMemoryPoolOperationType

Type:

int

pad2

(array of length 7).Undefined. Reserved for internal use.

Type:

uint8

process_id

The ID of the process to which this record belongs to.

Type:

int

ptr

Get the pointer address to the data as Python int.

release_threshold

The release threshold of the memory pool. releaseThreshold is valid for CUPTI_ACTIVITY_MEMORY_POOL_TYPE_LOCAL, CUpti_ActivityMemoryPoolType.

Type:

int

size_

The size of the memory pool operation in bytes. size is valid for CUPTI_ACTIVITY_MEMORY_POOL_TYPE_LOCAL, CUpti_ActivityMemoryPoolType.

Type:

int

timestamp

The start timestamp for the memory operation, in ns.

Type:

int

utilized_size

The utilized size of the memory pool. utilizedSize is valid for CUPTI_ACTIVITY_MEMORY_POOL_TYPE_LOCAL, CUpti_ActivityMemoryPoolType.

Type:

int

Deprecated since version 13.0.0.

class cupti.cupti.ActivityMemoryPool2

Bases: object

Empty-initialize an instance of CUpti_ActivityMemoryPool2.

See also

CUpti_ActivityMemoryPool2

address

The virtual address of the allocation.

Type:

int

correlation_id

The correlation ID of the memory pool operation. Each memory pool operation is assigned a unique correlation ID that is identical to the correlation ID in the driver and runtime API activity record that launched the memory operation.

Type:

int

device_id

The ID of the device where the memory pool is created.

Type:

int

static from_buffer(buffer)

Create an ActivityMemoryPool2 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMemoryPool2 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_memory_pool2_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityMemoryPool2 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MEMORY_POOL

Type:

int

memory_pool_operation_type

The memory operation requested by the user, CUpti_ActivityMemoryPoolOperationType.

Type:

int

memory_pool_type

The type of the memory pool, CUpti_ActivityMemoryPoolType

Type:

int

min_bytes_to_keep

The minimum bytes to keep of the memory pool. minBytesToKeep is valid for CUPTI_ACTIVITY_MEMORY_POOL_OPERATION_TYPE_TRIMMED, CUpti_ActivityMemoryPoolOperationType

Type:

int

process_id

The ID of the process to which this record belongs to.

Type:

int

ptr

Get the pointer address to the data as Python int.

release_threshold

The release threshold of the memory pool. releaseThreshold is valid for CUPTI_ACTIVITY_MEMORY_POOL_TYPE_LOCAL, CUpti_ActivityMemoryPoolType.

Type:

int

size_

The size of the memory pool operation in bytes. size is valid for CUPTI_ACTIVITY_MEMORY_POOL_TYPE_LOCAL, CUpti_ActivityMemoryPoolType.

Type:

int

timestamp

The start timestamp for the memory operation, in ns.

Type:

int

utilized_size

The utilized size of the memory pool. utilizedSize is valid for CUPTI_ACTIVITY_MEMORY_POOL_TYPE_LOCAL, CUpti_ActivityMemoryPoolType.

Type:

int

class cupti.cupti.ActivityMemset4

Bases: object

Empty-initialize an instance of CUpti_ActivityMemset4.

See also

CUpti_ActivityMemset4

bytes

The number of bytes being set by the memory set.

Type:

int

channel_id

The ID of the HW channel on which the memory set is occurring.

Type:

int

channel_type

The type of the channel

Type:

int

context_id

The ID of the context where the memory set is occurring.

Type:

int

correlation_id

The correlation ID of the memory set. Each memory set is assigned a unique correlation ID that is identical to the correlation ID in the driver API activity record that launched the memory set.

Type:

int

device_id

The ID of the device where the memory set is occurring.

Type:

int

end

The end timestamp for the memory set, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the memory set.

Type:

int

flags_

The flags associated with the memset. CUpti_ActivityFlag

Type:

int

static from_buffer(buffer)

Create an ActivityMemset4 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityMemset4 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_memset4_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityMemset4 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph_id

The unique ID of the graph that executed this memset through graph launch. This field will be 0 if the memset is not executed through graph launch.

Type:

int

graph_node_id

The unique ID of the graph node that executed this memset through graph launch. This field will be 0 if the memset is not executed through graph launch.

Type:

int

is_device_launched

This field is used to indicate if the memset operation is part of a device graph launch.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MEMSET.

Type:

int

memory_kind

The memory kind of the memory set CUpti_ActivityMemoryKind

Type:

int

pad

Undefined. Reserved for internal use.

Type:

int

pad2

(array of length 3).Undefined. Reserved for internal use

Type:

uint8

ptr

Get the pointer address to the data as Python int.

reserved0

Undefined. Reserved for internal use.

Type:

int

start

The start timestamp for the memory set, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the memory set.

Type:

int

stream_id

The ID of the stream where the memory set is occurring.

Type:

int

value

The value being assigned to memory by the memory set.

Type:

int

class cupti.cupti.ActivityModule

Bases: object

Empty-initialize an instance of CUpti_ActivityModule.

See also

CUpti_ActivityModule

context_id

The ID of the context where the module is loaded.

Type:

int

cubin

The pointer to cubin.

Type:

int

cubin_size

The cubin size.

Type:

int

static from_buffer(buffer)

Create an ActivityModule instance with the memory from the given buffer.

static from_data(data)

Create an ActivityModule instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_module_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityModule instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

id

The module ID.

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_MODULE.

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityName

Bases: object

Empty-initialize an instance of CUpti_ActivityName.

See also

CUpti_ActivityName

static from_buffer(buffer)

Create an ActivityName instance with the memory from the given buffer.

static from_data(data)

Create an ActivityName instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_name_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityName instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_NAME.

Type:

int

name

The name.

Type:

str

object_id

The identifier for the activity object. ‘objectKind’ indicates which ID is valid for this record.

Type:

ActivityObjectKindId

object_kind

The kind of activity object being named.

Type:

int

pad

Undefined. Reserved for internal use.

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityObjectKindId

Bases: object

Empty-initialize an instance of CUpti_ActivityObjectKindId.

See also

CUpti_ActivityObjectKindId

dcs

A device object requires that we identify the device ID. A context object requires that we identify both the device and context ID. A stream object requires that we identify device, context, and stream ID.

Type:

_py_anon_pod4

static from_buffer(buffer)

Create an ActivityObjectKindId instance with the memory from the given buffer.

static from_data(data)

Create an ActivityObjectKindId instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_object_kind_id_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityObjectKindId instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

pt

A process object requires that we identify the process ID. A thread object requires that we identify both the process and thread ID.

Type:

_py_anon_pod3

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityOpenAccData

Bases: object

Empty-initialize an instance of CUpti_ActivityOpenAccData.

See also

CUpti_ActivityOpenAccData

async_

int:

async_map

int:

bytes

Number of bytes

Type:

int

cu_context_id

CUDA context id Valid only if deviceType is acc_device_nvidia.

Type:

int

cu_device_id

CUDA device id Valid only if deviceType is acc_device_nvidia.

Type:

int

cu_process_id

The ID of the process where the OpenACC activity is executing.

Type:

int

cu_stream_id

CUDA stream id Valid only if deviceType is acc_device_nvidia.

Type:

int

cu_thread_id

The ID of the thread where the OpenACC activity is executing.

Type:

int

device_number

int:

device_ptr

Device pointer if available

Type:

int

device_type

int:

end

CUPTI end timestamp

Type:

int

end_line_no

int:

event_kind

CUPTI OpenACC event kind (CUpti_OpenAccEventKind)

Type:

int

external_id

The OpenACC correlation ID. Valid only if deviceType is acc_device_nvidia. If not 0, it uniquely identifies this record. It is identical to the externalId in the preceding external correlation record of type CUPTI_EXTERNAL_CORRELATION_KIND_OPENACC.

Type:

int

static from_buffer(buffer)

Create an ActivityOpenAccData instance with the memory from the given buffer.

static from_data(data)

Create an ActivityOpenAccData instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_open_acc_data_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityOpenAccData instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

func_end_line_no

int:

func_line_no

int:

func_name

str:

host_ptr

Host pointer if available

Type:

int

implicit

int:

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_OPENACC_DATA.

Type:

int

line_no

int:

parent_construct

int:

ptr

Get the pointer address to the data as Python int.

src_file

str:

start

CUPTI start timestamp

Type:

int

thread_id

ThreadId

Type:

int

var_name

str:

version

int:

class cupti.cupti.ActivityOpenAccLaunch

Bases: object

Empty-initialize an instance of CUpti_ActivityOpenAccLaunch.

See also

CUpti_ActivityOpenAccLaunch

async_

Value of async() clause of the corresponding directive

Type:

int

async_map

Internal asynchronous queue number used

Type:

int

cu_context_id

CUDA context id Valid only if deviceType is acc_device_nvidia.

Type:

int

cu_device_id

CUDA device id Valid only if deviceType is acc_device_nvidia.

Type:

int

cu_process_id

The ID of the process where the OpenACC activity is executing.

Type:

int

cu_stream_id

CUDA stream id Valid only if deviceType is acc_device_nvidia.

Type:

int

cu_thread_id

The ID of the thread where the OpenACC activity is executing.

Type:

int

device_number

Device number

Type:

int

device_type

Device type

Type:

int

end

CUPTI end timestamp

Type:

int

end_line_no

For an OpenACC construct, this contains the line number of the end of the construct. A negative or zero value means the line number is not known.

Type:

int

event_kind

CUPTI OpenACC event kind (CUpti_OpenAccEventKind)

Type:

int

external_id

The OpenACC correlation ID. Valid only if deviceType is acc_device_nvidia. If not 0, it uniquely identifies this record. It is identical to the externalId in the preceding external correlation record of type CUPTI_EXTERNAL_CORRELATION_KIND_OPENACC.

Type:

int

static from_buffer(buffer)

Create an ActivityOpenAccLaunch instance with the memory from the given buffer.

static from_data(data)

Create an ActivityOpenAccLaunch instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_open_acc_launch_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityOpenAccLaunch instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

func_end_line_no

The last line number of the function named in func_name. A negative or zero value means the line number is not known.

Type:

int

func_line_no

The line number of the first line of the function named in func_name. A negative or zero value means the line number is not known.

Type:

int

func_name

A pointer to a null-terminated string containing the name of the function in which the event occurred.

Type:

str

implicit

1 for any implicit event, such as an implicit wait at a synchronous data construct 0 otherwise

Type:

int

kernel_name

A pointer to null-terminated string containing the name of the kernel being launched, if known, or a null pointer if not.

Type:

str

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_OPENACC_LAUNCH.

Type:

int

line_no

The line number of the directive or program construct or the starting line number of the OpenACC construct corresponding to the event. A negative or zero value means the line number is not known.

Type:

int

num_gangs

The number of gangs created for this kernel launch

Type:

int

num_workers

The number of workers created for this kernel launch

Type:

int

parent_construct

CUPTI OpenACC parent construct kind (CUpti_OpenAccConstructKind) Note that for applications using PGI OpenACC runtime < 16.1, this will always be CUPTI_OPENACC_CONSTRUCT_KIND_UNKNOWN.

Type:

int

ptr

Get the pointer address to the data as Python int.

src_file

A pointer to null-terminated string containing the name of or path to the source file, if known, or a null pointer if not.

Type:

str

start

CUPTI start timestamp

Type:

int

thread_id

ThreadId

Type:

int

vector_length

The number of vector lanes created for this kernel launch

Type:

int

version

Version number

Type:

int

class cupti.cupti.ActivityOpenAccOther

Bases: object

Empty-initialize an instance of CUpti_ActivityOpenAccOther.

See also

CUpti_ActivityOpenAccOther

async_

Value of async() clause of the corresponding directive

Type:

int

async_map

Internal asynchronous queue number used

Type:

int

cu_context_id

CUDA context id Valid only if deviceType is acc_device_nvidia.

Type:

int

cu_device_id

CUDA device id Valid only if deviceType is acc_device_nvidia.

Type:

int

cu_process_id

The ID of the process where the OpenACC activity is executing.

Type:

int

cu_stream_id

CUDA stream id Valid only if deviceType is acc_device_nvidia.

Type:

int

cu_thread_id

The ID of the thread where the OpenACC activity is executing.

Type:

int

device_number

Device number

Type:

int

device_type

Device type

Type:

int

end

CUPTI end timestamp

Type:

int

end_line_no

For an OpenACC construct, this contains the line number of the end of the construct. A negative or zero value means the line number is not known.

Type:

int

event_kind

CUPTI OpenACC event kind (CUpti_OpenAccEventKind)

Type:

int

external_id

The OpenACC correlation ID. Valid only if deviceType is acc_device_nvidia. If not 0, it uniquely identifies this record. It is identical to the externalId in the preceding external correlation record of type CUPTI_EXTERNAL_CORRELATION_KIND_OPENACC.

Type:

int

static from_buffer(buffer)

Create an ActivityOpenAccOther instance with the memory from the given buffer.

static from_data(data)

Create an ActivityOpenAccOther instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_open_acc_other_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityOpenAccOther instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

func_end_line_no

The last line number of the function named in func_name. A negative or zero value means the line number is not known.

Type:

int

func_line_no

The line number of the first line of the function named in func_name. A negative or zero value means the line number is not known.

Type:

int

func_name

A pointer to a null-terminated string containing the name of the function in which the event occurred.

Type:

str

implicit

1 for any implicit event, such as an implicit wait at a synchronous data construct 0 otherwise

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_OPENACC_OTHER.

Type:

int

line_no

The line number of the directive or program construct or the starting line number of the OpenACC construct corresponding to the event. A negative or zero value means the line number is not known.

Type:

int

parent_construct

CUPTI OpenACC parent construct kind (CUpti_OpenAccConstructKind) Note that for applications using PGI OpenACC runtime < 16.1, this will always be CUPTI_OPENACC_CONSTRUCT_KIND_UNKNOWN.

Type:

int

ptr

Get the pointer address to the data as Python int.

src_file

A pointer to null-terminated string containing the name of or path to the source file, if known, or a null pointer if not.

Type:

str

start

CUPTI start timestamp

Type:

int

thread_id

ThreadId

Type:

int

version

Version number

Type:

int

class cupti.cupti.ActivityOpenMp

Bases: object

Empty-initialize an instance of CUpti_ActivityOpenMp.

See also

CUpti_ActivityOpenMp

cu_process_id

The ID of the process where the OpenMP activity is executing.

Type:

int

cu_thread_id

The ID of the thread where the OpenMP activity is executing.

Type:

int

end

CUPTI end timestamp

Type:

int

event_kind

CUPTI OpenMP event kind (CUpti_OpenMpEventKind)

Type:

int

static from_buffer(buffer)

Create an ActivityOpenMp instance with the memory from the given buffer.

static from_data(data)

Create an ActivityOpenMp instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_open_mp_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityOpenMp instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The kind of this activity.

Type:

int

ptr

Get the pointer address to the data as Python int.

start

CUPTI start timestamp

Type:

int

thread_id

ThreadId

Type:

int

version

Version number

Type:

int

class cupti.cupti.ActivityOverhead3

Bases: object

Empty-initialize an instance of CUpti_ActivityOverhead3.

See also

CUpti_ActivityOverhead3

correlation_id

The correlation ID of the overhead operation to which records belong to. This ID is identical to the correlation ID in the driver or runtime API activity record that launched the overhead operation. In some cases, it can be zero, such as for CUPTI_ACTIVITY_OVERHEAD_CUPTI_BUFFER_FLUSH records.

Type:

int

end

The end timestamp for the overhead, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the overhead.

Type:

int

static from_buffer(buffer)

Create an ActivityOverhead3 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityOverhead3 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_overhead3_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityOverhead3 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_OVERHEAD.

Type:

int

object_id

The identifier for the activity object. ‘objectKind’ indicates which ID is valid for this record.

Type:

ActivityObjectKindId

object_kind

The kind of activity object that the overhead is associated with.

Type:

int

overhead_data

Pointer to the struct with additional details about the overhead. Refer CUpti_ActivityOverheadKind enum and the corresponding structure to typecast and access additional overhead data. Client is responsible for freeing this memory using the free function when done.

Type:

int

overhead_kind

The kind of overhead, CUPTI, DRIVER, COMPILER etc.

Type:

int

ptr

Get the pointer address to the data as Python int.

reserved0

Reserved for internal use.

Type:

int

start

The start timestamp for the overhead, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the overhead.

Type:

int

class cupti.cupti.ActivityOverheadCommandBufferFullData

Bases: object

Empty-initialize an instance of CUpti_ActivityOverheadCommandBufferFullData.

See also

CUpti_ActivityOverheadCommandBufferFullData

channel_id

The channel ID of the command buffer.

Type:

int

channel_type

The channel type of the command buffer.

Type:

int

command_buffer_length

The remaining space in the command buffer. This field will always be zero when the command buffer is full, making it not useful in such cases.

Type:

int

static from_buffer(buffer)

Create an ActivityOverheadCommandBufferFullData instance with the memory from the given buffer.

static from_data(data)

Create an ActivityOverheadCommandBufferFullData instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_overhead_command_buffer_full_data_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityOverheadCommandBufferFullData instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ActivityPreemption

Bases: object

Empty-initialize an instance of CUpti_ActivityPreemption.

See also

CUpti_ActivityPreemption

block_x

The X-dimension of the block that is preempted

Type:

int

block_y

The Y-dimension of the block that is preempted

Type:

int

block_z

The Z-dimension of the block that is preempted

Type:

int

static from_buffer(buffer)

Create an ActivityPreemption instance with the memory from the given buffer.

static from_data(data)

Create an ActivityPreemption instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_preemption_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityPreemption instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

grid_id

The grid-id of the block that is preempted

Type:

int

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_PREEMPTION

Type:

int

pad

Undefined. Reserved for internal use.

Type:

int

preemption_kind

kind of the preemption

Type:

int

ptr

Get the pointer address to the data as Python int.

timestamp

The timestamp of the preemption, in ns. A value of 0 indicates that timestamp information could not be collected for the preemption.

Type:

int

class cupti.cupti.ActivityStream

Bases: object

Empty-initialize an instance of CUpti_ActivityStream.

See also

CUpti_ActivityStream

context_id

The ID of the context where the stream was created.

Type:

int

correlation_id

The correlation ID of the API to which this result is associated.

Type:

int

flag

Flags associated with the stream.

Type:

int

static from_buffer(buffer)

Create an ActivityStream instance with the memory from the given buffer.

static from_data(data)

Create an ActivityStream instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_stream_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an ActivityStream instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_STREAM.

Type:

int

priority

The clamped priority for the stream.

Type:

int

ptr

Get the pointer address to the data as Python int.

stream_id

A unique stream ID to identify the stream.

Type:

int

class cupti.cupti.ActivitySynchronization2

Bases: object

Empty-initialize an instance of CUpti_ActivitySynchronization2.

See also

CUpti_ActivitySynchronization2

context_id

The ID of the context for which the synchronization API is called. In case of context synchronization API it is the context id for which the API is called. In case of stream/event synchronization it is the ID of the context where the stream/event was created.

Type:

int

correlation_id

The correlation ID of the API to which this result is associated.

Type:

int

cuda_event_id

The event ID for which the synchronization API is called. A CUPTI_SYNCHRONIZATION_INVALID_VALUE value indicate the field is not applicable for this record. Not valid for cuCtxSynchronize, cuStreamSynchronize.

Type:

int

cuda_event_sync_id

A unique ID to associate event synchronization records with the latest CUDA Event record. Similar field is added in CUpti_ActivityCudaEvent2 to associate synchronization record to the CUDA Event record. The same CUDA event can be used multiple times, so the event id will not be unique to correlate the synchronization record with the latest CUDA Event record. This field will be unique and can be used to do the required correlation. A CUPTI_SYNCHRONIZATION_INVALID_VALUE value indicates that the field is not applicable for this record. Valid only for synchronization records related to CUDA Events.

Type:

int

end

The end timestamp for the function, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the function.

Type:

int

static from_buffer(buffer)

Create an ActivitySynchronization2 instance with the memory from the given buffer.

static from_data(data)

Create an ActivitySynchronization2 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_synchronization2_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivitySynchronization2 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_SYNCHRONIZATION.

Type:

int

pad

Undefined. Reserved for internal use.

Type:

int

ptr

Get the pointer address to the data as Python int.

return_value

The return value for the synchronization record. Use cuptiActivityEnableAllSyncRecords API to enable/disable collection of synchronization records with return value being non-zero. This will be a CUresult value.

Type:

int

start

The start timestamp for the function, in ns. A value of 0 for both the start and end timestamps indicates that timestamp information could not be collected for the function.

Type:

int

stream_id

The compute stream for which the synchronization API is called. A CUPTI_SYNCHRONIZATION_INVALID_VALUE value indicate the field is not applicable for this record. Not valid for cuCtxSynchronize, cuEventSynchronize.

Type:

int

type

The type of record.

Type:

int

Deprecated since version 12.8.0.

class cupti.cupti.ActivitySynchronization

Bases: object

Empty-initialize an instance of CUpti_ActivitySynchronization.

See also

CUpti_ActivitySynchronization

context_id

int:

correlation_id

int:

cuda_event_id

int:

end

int:

static from_buffer(buffer)

Create an ActivitySynchronization instance with the memory from the given buffer.

static from_data(data)

Create an ActivitySynchronization instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_synchronization_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivitySynchronization instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

int:

ptr

Get the pointer address to the data as Python int.

start

int:

stream_id

int:

type

int:

class cupti.cupti.ActivityUnifiedMemoryCounter3

Bases: object

Empty-initialize an instance of CUpti_ActivityUnifiedMemoryCounter3.

See also

CUpti_ActivityUnifiedMemoryCounter3

address

This is the virtual base address of the page/s being transferred. For cpu and gpu faults, the virtual address for the page that faulted.

Type:

int

counter_kind

The Unified Memory counter kind

Type:

int

dst_id

The ID of the destination CPU/device involved in the memory transfer or remote map operation. Ignore this field if counterKind is CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_GPU_PAGE_FAULT or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_CPU_PAGE_FAULT_COUNT or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THRASHING or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THROTTLING

Type:

int

end

The end timestamp of the counter, in ns. Ignore this field if counterKind is CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_CPU_PAGE_FAULT_COUNT or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THRASHING or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_REMOTE_MAP. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_HTOD and CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_DTOH, timestamp is captured when activity finishes on GPU. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_GPU_PAGE_FAULT, timestamp is captured when CUDA driver queues the replay of faulting memory accesses on the GPU For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THROTTLING, timestamp is captured when throttling operation was finished by CUDA driver

Type:

int

flags_

The flags associated with this record. See enums CUpti_ActivityUnifiedMemoryAccessType if counterKind is CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_GPU_PAGE_FAULT and CUpti_ActivityUnifiedMemoryMigrationCause if counterKind is CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_HTOD or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_HTOD and CUpti_ActivityUnifiedMemoryRemoteMapCause if counterKind is CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_REMOTE_MAP and CUpti_ActivityFlag if counterKind is CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THRASHING or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THROTTLING

Type:

int

static from_buffer(buffer)

Create an ActivityUnifiedMemoryCounter3 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityUnifiedMemoryCounter3 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_unified_memory_counter3_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityUnifiedMemoryCounter3 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

The activity record kind, must be CUPTI_ACTIVITY_KIND_UNIFIED_MEMORY_COUNTER

Type:

int

pad

Undefined. Reserved for internal use.

Type:

int

process_id

The ID of the process to which this record belongs to.

Type:

int

processors

(array of length 5).The bitmask of devices involved in the operation. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THRASHING, it is a bitwise ORing of the device IDs fighting for the memory region. processors[0] represents the device ID of the device 0 to device 63, processors[1] represents device ID of device 64 to device 127 and so on. Ignore this field if counterKind is CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_CPU_PAGE_FAULT_COUNT or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_GPU_PAGE_FAULT or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THROTTLING or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_REMOTE_MAP or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_HTOD or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_DTOH or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_DTOD or CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_GPU_FAULT_REPLAY

Type:

uint64

ptr

Get the pointer address to the data as Python int.

src_id

The ID of the source CPU/device involved in the memory transfer, page fault, thrashing, throttling or remote map operation. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THRASHING, it is a bitwise ORing of the device IDs fighting for the memory region, ONLY if there are less than 32 devices. Ignore this field if counterKind is CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_CPU_PAGE_FAULT_COUNT

Type:

int

start

The start timestamp of the counter, in ns. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_HTOD and CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_DTOH, timestamp is captured when activity starts on GPU. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_GPU_PAGE_FAULT and CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_CPU_PAGE_FAULT_COUNT, timestamp is captured when CUDA driver started processing the fault. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THRASHING, timestamp is captured when CUDA driver detected thrashing of memory region. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THROTTLING, timestamp is captured when throttling operation was started by CUDA driver. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_REMOTE_MAP, timestamp is captured when CUDA driver has pushed all required operations to the processor specified by dstId.

Type:

int

stream_id

The ID of the stream causing the transfer. This value of this field is invalid.

Type:

int

value

Value of the counter For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_HTOD, CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_BYTES_TRANSFER_DTOH, CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_THREASHING and CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_REMOTE_MAP, it is the size of the memory region in bytes. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_GPU_PAGE_FAULT, it is the number of page fault groups for the same page. For counterKind CUPTI_ACTIVITY_UNIFIED_MEMORY_COUNTER_KIND_CPU_PAGE_FAULT_COUNT, it is the program counter for the instruction that caused fault.

Type:

int

Deprecated since version 12.8.0.

class cupti.cupti.ActivityUnifiedMemoryCounter2

Bases: object

Empty-initialize an instance of CUpti_ActivityUnifiedMemoryCounter2.

See also

CUpti_ActivityUnifiedMemoryCounter2

address

int:

counter_kind

int:

dst_id

int:

end

int:

flags_

int:

static from_buffer(buffer)

Create an ActivityUnifiedMemoryCounter2 instance with the memory from the given buffer.

static from_data(data)

Create an ActivityUnifiedMemoryCounter2 instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype activity_unified_memory_counter2_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ActivityUnifiedMemoryCounter2 instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

int:

pad

int:

process_id

int:

ptr

Get the pointer address to the data as Python int.

src_id

int:

start

int:

stream_id

int:

value

int:

class cupti.cupti.ActivityUnifiedMemoryCounterConfig

Bases: object

Empty-initialize an array of CUpti_ActivityUnifiedMemoryCounterConfig. The resulting object is of length size and of dtype activity_unified_memory_counter_config_dtype. If default-constructed, the instance represents a single struct.

Parameters:

size (int) – number of structs, default=1.

See also

CUpti_ActivityUnifiedMemoryCounterConfig

device_id

Device id of the target device. This is relevant only for single device scopes. (deprecated in CUDA 7.0)

Type:

Union[uint32, int]

enable

Control to enable/disable the counter. To enable the counter set it to non-zero value while disable is indicated by zero.

Type:

Union[uint32, int]

static from_buffer(buffer)

Create an ActivityUnifiedMemoryCounterConfig instance with the memory from the given buffer.

static from_data(data)

Create an ActivityUnifiedMemoryCounterConfig instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a 1D array of dtype activity_unified_memory_counter_config_dtype holding the data.

static from_ptr(

ptr: int,

size: int = 1,

readonly: bool = False,

)

Create an ActivityUnifiedMemoryCounterConfig instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• size (int) – number of structs, default=1.

• readonly (bool) – whether the data is read-only (to the user). default is False.

kind

Unified Memory counter Counter kind

Type:

Union[int32, int]

ptr

Get the pointer address to the data as Python int.

scope

Unified Memory counter Counter scope. (deprecated in CUDA 7.0)

Type:

Union[int32, int]

class cupti.cupti.MetricValue

Bases: object

Empty-initialize an instance of CUpti_MetricValue.

See also

CUpti_MetricValue

static from_buffer(buffer)

Create an MetricValue instance with the memory from the given buffer.

static from_data(data)

Create an MetricValue instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype metric_value_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an MetricValue instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

metric_value_double

float:

metric_value_int64

int:

metric_value_nvtx_extended_payload

Value for CUPTI_METRIC_VALUE_KIND_NVTX_EXTENDED_PAYLOAD.

Type:

int

metric_value_percent

float:

metric_value_throughput

int:

metric_value_uint64

int:

metric_value_utilization_level

int:

ptr

Get the pointer address to the data as Python int.

Callback API

Functions

cupti.cupti.enable_all_domains(enable: int, subscriber: int)

Enable or disable all callbacks in all domains.

Parameters:

• enable (uint32_t) – New enable state for all callbacks in all domain. Zero disables all callbacks, non-zero enables all callbacks.

• subscriber (intptr_t) – Handle to callback subscription.

See also

cuptiEnableAllDomains

cupti.cupti.enable_callback(enable: int, subscriber: int, domain: int, cbid: int)

Enable or disabled callbacks for a specific domain and callback ID.

Parameters:

• enable (uint32_t) – New enable state for the callback. Zero disables the callback, non-zero enables the callback.

• subscriber (intptr_t) – Handle to callback subscription.

• domain (CallbackDomain) – The domain of the callback.

• cbid (uint32_t) – The ID of the callback.

See also

cuptiEnableCallback

cupti.cupti.enable_domain(enable: int, subscriber: int, domain: int)

Enable or disabled all callbacks for a specific domain.

Parameters:

• enable (uint32_t) – New enable state for all callbacks in the domain. Zero disables all callbacks, non-zero enables all callbacks.

• subscriber (intptr_t) – Handle to callback subscription.

• domain (CallbackDomain) – The domain of the callback.

See also

cuptiEnableDomain

cupti.cupti.get_callback_name(domain: int, cbid: int)

Get the name of a callback for a specific domain and callback ID.

Parameters:

• domain (CallbackDomain) – The domain of the callback.

• cbid (uint32_t) – The ID of the callback.

Returns:

Returns name of the callback for the specified domain and callback ID

Return type:

name (str)

See also

cuptiGetCallbackName

cupti.cupti.get_callback_state(subscriber: int, domain: int, cbid: int) → int

Get the current enabled/disabled state of a callback for a specific domain and function ID.

Parameters:

• subscriber (intptr_t) – Handle to the initialize subscriber.

• domain (CallbackDomain) – The domain of the callback.

• cbid (uint32_t) – The ID of the callback.

Returns:

Returns non-zero if callback enabled, zero if not enabled.

Return type:

uint32_t

See also

cuptiGetCallbackState

cupti.cupti.subscribe(callback, userdata) → int

Initialize a callback subscriber with a callback function and user data.

Parameters:

• callback (object) – The callback function.

• userdata (object) – A pointer to user data. This data will be passed to the callback function via the userdata parameter.

Returns:

Returns handle to initialize subscriber.

Return type:

intptr_t

See also

cuptiSubscribe

cupti.cupti.subscribe_v2(callback, userdata, p_params: int) → int

Initialize a callback subscriber with a callback function and user data.

Parameters:

• callback (object) – The callback function.

• userdata (object) – A pointer to user data. This data will be passed to the callback function via the userdata parameter.

• p_params (intptr_t) – A pointer to CUpti_SubscriberParams. Can be NULL.

Returns:

Returns handle to initialize subscriber.

Return type:

intptr_t

See also

cuptiSubscribe_v2

cupti.cupti.supported_domains()

Get the available callback domains.

Returns:

List of all available callback domains

Return type:

list[CallbackDomain]

See also

cuptiSupportedDomains

cupti.cupti.unsubscribe(subscriber: int)

Unregister a callback subscriber.

Parameters:

subscriber (intptr_t) – Handle to the initialize subscriber.

See also

cuptiUnsubscribe

Enums

class cupti.cupti.ApiCallbackSite(value)

Bases: IntEnum

Specifies the point in an API call that a callback is issued.Specifies the point in an API call that a callback is issued. This value is communicated to the callback function via CUpti_CallbackData.callbackSite.

See CUpti_ApiCallbackSite.

API_CBSITE_FORCE_INT = 2147483647

API_ENTER = 0

API_EXIT = 1

class cupti.cupti.CallbackDomain(value)

Bases: IntEnum

Callback domains.Callback domains. Each domain represents callback points for a group of related API functions or CUDA driver activity.

See CUpti_CallbackDomain.

DRIVER_API = 1

FORCE_INT = 2147483647

INVALID = 0

NVTX = 5

RESOURCE = 3

RUNTIME_API = 2

SIZE = 7

STATE = 6

SYNCHRONIZE = 4

class cupti.cupti.CallbackIdResource(value)

Bases: IntEnum

Callback IDs for resource domain.Callback IDs for resource domain, CUPTI_CB_DOMAIN_RESOURCE. This value is communicated to the callback function via the cbid parameter.

See CUpti_CallbackIdResource.

CONTEXT_CREATED = 1

CONTEXT_DESTROY_STARTING = 2

CU_INIT_FINISHED = 5

FORCE_INT = 2147483647

GRAPHEXEC_CREATED = 18

GRAPHEXEC_CREATE_STARTING = 17

GRAPHEXEC_DESTROY_STARTING = 19

GRAPHNODE_CLONED = 20

GRAPHNODE_CREATED = 13

GRAPHNODE_CREATE_STARTING = 12

GRAPHNODE_DEPENDENCY_CREATED = 15

GRAPHNODE_DEPENDENCY_DESTROY_STARTING = 16

GRAPHNODE_DESTROY_STARTING = 14

GRAPH_CLONED = 11

GRAPH_CREATED = 9

GRAPH_DESTROY_STARTING = 10

GRAPH_NODE_SET_PARAMS = 23

GRAPH_NODE_UPDATED = 22

INVALID = 0

MODULE_LOADED = 6

MODULE_PROFILED = 8

MODULE_UNLOAD_STARTING = 7

SIZE = 24

STREAM_ATTRIBUTE_CHANGED = 21

STREAM_CREATED = 3

STREAM_DESTROY_STARTING = 4

class cupti.cupti.CallbackIdState(value)

Bases: IntEnum

Callback IDs for state domain.Callback IDs for state domain, CUPTI_CB_DOMAIN_STATE. This value is communicated to the callback function via the cbid parameter.

See CUpti_CallbackIdState.

ERROR = 2

FATAL_ERROR = 1

FORCE_INT = 2147483647

INVALID = 0

SIZE = 4

WARNING = 3

class cupti.cupti.CallbackIdSync(value)

Bases: IntEnum

Callback IDs for synchronization domain.Callback IDs for synchronization domain, CUPTI_CB_DOMAIN_SYNCHRONIZE. This value is communicated to the callback function via the cbid parameter.

See CUpti_CallbackIdSync.

CONTEXT_SYNCHRONIZED = 2

FORCE_INT = 2147483647

INVALID = 0

SIZE = 3

STREAM_SYNCHRONIZED = 1

class cupti.cupti.Driver_api_trace_cbid(value)

Bases: IntEnum

See CUpti_driver_api_trace_cbid.

FORCE_INT = 2147483647

INVALID = 0

SIZE = 860

cu64Array3DCreate = 230

cu64Array3DGetDescriptor = 231

cu64ArrayCreate = 228

cu64ArrayGetDescriptor = 229

cu64D3D10ResourceGetMappedPitch = 200

cu64D3D10ResourceGetMappedPointer = 198

cu64D3D10ResourceGetMappedSize = 199

cu64D3D10ResourceGetSurfaceDimensions = 201

cu64D3D9MapVertexBuffer = 206

cu64D3D9ResourceGetMappedPitch = 205

cu64D3D9ResourceGetMappedPointer = 203

cu64D3D9ResourceGetMappedSize = 204

cu64D3D9ResourceGetSurfaceDimensions = 202

cu64DeviceTotalMem = 197

cu64GLMapBufferObject = 207

cu64GLMapBufferObjectAsync = 208

cu64GraphicsResourceGetMappedPointer = 131

cu64MemAlloc = 30

cu64MemAllocPitch = 32

cu64MemFree = 34

cu64MemGetAddressRange = 36

cu64MemGetInfo = 28

cu64MemHostAlloc = 215

cu64MemHostGetDevicePointer = 41

cu64Memcpy2D = 232

cu64Memcpy2DAsync = 234

cu64Memcpy2DUnaligned = 233

cu64Memcpy3D = 59

cu64Memcpy3DAsync = 70

cu64MemcpyAtoD = 52

cu64MemcpyDtoA = 50

cu64MemcpyDtoD = 48

cu64MemcpyDtoDAsync = 65

cu64MemcpyDtoH = 46

cu64MemcpyDtoHAsync = 63

cu64MemcpyHtoD = 44

cu64MemcpyHtoDAsync = 61

cu64MemsetD16 = 74

cu64MemsetD16Async = 219

cu64MemsetD2D16 = 80

cu64MemsetD2D16Async = 225

cu64MemsetD2D32 = 82

cu64MemsetD2D32Async = 227

cu64MemsetD2D8 = 78

cu64MemsetD2D8Async = 223

cu64MemsetD32 = 76

cu64MemsetD32Async = 221

cu64MemsetD8 = 72

cu64MemsetD8Async = 217

cu64ModuleGetGlobal = 25

cu64TexRefGetAddress = 104

cu64TexRefSetAddress = 96

cu64TexRefSetAddress2D = 98

cuArray3DCreate = 90

cuArray3DCreate_v2 = 274

cuArray3DGetDescriptor = 91

cuArray3DGetDescriptor_v2 = 275

cuArrayCreate = 87

cuArrayCreate_v2 = 272

cuArrayDestroy = 89

cuArrayGetDescriptor = 88

cuArrayGetDescriptor_v2 = 273

cuArrayGetMemoryRequirements = 654

cuArrayGetPlane = 597

cuArrayGetSparseProperties = 582

cuBinaryFree = 376

cuCheckpointProcessCheckpoint = 771

cuCheckpointProcessGetRestoreThreadId = 768

cuCheckpointProcessGetState = 769

cuCheckpointProcessLock = 770

cuCheckpointProcessRestore = 772

cuCheckpointProcessUnlock = 773

cuCompilePtx = 375

cuCoredumpDeregisterCompleteCallback = 845

cuCoredumpDeregisterStartCallback = 844

cuCoredumpGetAttribute = 701

cuCoredumpGetAttributeGlobal = 702

cuCoredumpRegisterCompleteCallback = 843

cuCoredumpRegisterStartCallback = 842

cuCoredumpSetAttribute = 703

cuCoredumpSetAttributeGlobal = 704

cuCtxAttach = 12

cuCtxCreate = 10

cuCtxCreate_v2 = 235

cuCtxCreate_v3 = 645

cuCtxCreate_v4 = 757

cuCtxDestroy = 11

cuCtxDestroy_v2 = 322

cuCtxDetach = 13

cuCtxDisablePeerAccess = 314

cuCtxEnablePeerAccess = 313

cuCtxFromGreenCtx = 753

cuCtxGetApiVersion = 296

cuCtxGetCacheConfig = 299

cuCtxGetCurrent = 304

cuCtxGetDevResource = 746

cuCtxGetDevice = 16

cuCtxGetDevice_v2 = 795

cuCtxGetExecAffinity = 646

cuCtxGetFlags = 391

cuCtxGetId = 695

cuCtxGetLimit = 137

cuCtxGetSharedMemConfig = 337

cuCtxGetStreamPriorityRange = 370

cuCtxPopCurrent = 15

cuCtxPopCurrent_v2 = 324

cuCtxPushCurrent = 14

cuCtxPushCurrent_v2 = 323

cuCtxRecordEvent = 755

cuCtxResetPersistingL2Cache = 568

cuCtxSetCacheConfig = 300

cuCtxSetCurrent = 303

cuCtxSetFlags = 705

cuCtxSetLimit = 136

cuCtxSetSharedMemConfig = 336

cuCtxSynchronize = 17

cuCtxSynchronize_v2 = 800

cuCtxWaitEvent = 756

cuD3D10CtxCreate = 139

cuD3D10CtxCreateOnDevice = 212

cuD3D10CtxCreate_v2 = 236

cuD3D10GetDevice = 138

cuD3D10GetDevices = 211

cuD3D10GetDirect3DDevice = 297

cuD3D10MapResources = 143

cuD3D10RegisterResource = 141

cuD3D10ResourceGetMappedArray = 146

cuD3D10ResourceGetMappedPitch = 149

cuD3D10ResourceGetMappedPitch_v2 = 262

cuD3D10ResourceGetMappedPointer = 147

cuD3D10ResourceGetMappedPointer_v2 = 260

cuD3D10ResourceGetMappedSize = 148

cuD3D10ResourceGetMappedSize_v2 = 261

cuD3D10ResourceGetSurfaceDimensions = 150

cuD3D10ResourceGetSurfaceDimensions_v2 = 263

cuD3D10ResourceSetMapFlags = 145

cuD3D10UnmapResources = 144

cuD3D10UnregisterResource = 142

cuD3D11CtxCreate = 152

cuD3D11CtxCreateOnDevice = 210

cuD3D11CtxCreate_v2 = 237

cuD3D11GetDevice = 151

cuD3D11GetDevices = 209

cuD3D11GetDirect3DDevice = 298

cuD3D9Begin = 168

cuD3D9CtxCreate = 155

cuD3D9CtxCreateOnDevice = 214

cuD3D9CtxCreate_v2 = 238

cuD3D9End = 169

cuD3D9GetDevice = 154

cuD3D9GetDevices = 213

cuD3D9GetDirect3DDevice = 157

cuD3D9MapResources = 160

cuD3D9MapVertexBuffer = 171

cuD3D9MapVertexBuffer_v2 = 268

cuD3D9RegisterResource = 158

cuD3D9RegisterVertexBuffer = 170

cuD3D9ResourceGetMappedArray = 164

cuD3D9ResourceGetMappedPitch = 167

cuD3D9ResourceGetMappedPitch_v2 = 267

cuD3D9ResourceGetMappedPointer = 165

cuD3D9ResourceGetMappedPointer_v2 = 265

cuD3D9ResourceGetMappedSize = 166

cuD3D9ResourceGetMappedSize_v2 = 266

cuD3D9ResourceGetSurfaceDimensions = 163

cuD3D9ResourceGetSurfaceDimensions_v2 = 264

cuD3D9ResourceSetMapFlags = 162

cuD3D9UnmapResources = 161

cuD3D9UnmapVertexBuffer = 172

cuD3D9UnregisterResource = 159

cuD3D9UnregisterVertexBuffer = 173

cuDestroyExternalMemory = 488

cuDestroyExternalSemaphore = 494

cuDevResourceGenerateDesc = 748

cuDevSmResourceSplit = 822

cuDevSmResourceSplitByCount = 751

cuDeviceCanAccessPeer = 312

cuDeviceComputeCapability = 6

cuDeviceGet = 3

cuDeviceGetAttribute = 9

cuDeviceGetByPCIBusId = 331

cuDeviceGetCount = 4

cuDeviceGetDefaultMemPool = 606

cuDeviceGetDevResource = 745

cuDeviceGetExecAffinitySupport = 644

cuDeviceGetGraphMemAttribute = 641

cuDeviceGetHostAtomicCapabilities = 805

cuDeviceGetLuid = 532

cuDeviceGetMemPool = 610

cuDeviceGetName = 5

cuDeviceGetNvSciSyncAttributes = 542

cuDeviceGetP2PAtomicCapabilities = 804

cuDeviceGetP2PAttribute = 454

cuDeviceGetPCIBusId = 332

cuDeviceGetProperties = 8

cuDeviceGetTexture1DLinearMaxWidth = 579

cuDeviceGetUuid = 482

cuDeviceGetUuid_v2 = 647

cuDeviceGraphMemTrim = 640

cuDevicePrimaryCtxGetState = 392

cuDevicePrimaryCtxRelease = 387

cuDevicePrimaryCtxRelease_v2 = 544

cuDevicePrimaryCtxReset = 389

cuDevicePrimaryCtxReset_v2 = 545

cuDevicePrimaryCtxRetain = 386

cuDevicePrimaryCtxSetFlags = 388

cuDevicePrimaryCtxSetFlags_v2 = 546

cuDeviceRegisterAsyncNotification = 735

cuDeviceSetGraphMemAttribute = 642

cuDeviceSetMemPool = 609

cuDeviceTotalMem = 7

cuDeviceTotalMem_v2 = 259

cuDeviceUnregisterAsyncNotification = 736

cuDriverGetGpuCodeIsaVersion = 806

cuDriverGetVersion = 2

cuEGLStreamConsumerAcquireFrame = 395

cuEGLStreamConsumerConnect = 393

cuEGLStreamConsumerConnectWithFlags = 470

cuEGLStreamConsumerDisconnect = 394

cuEGLStreamConsumerReleaseFrame = 396

cuEGLStreamProducerConnect = 446

cuEGLStreamProducerDisconnect = 447

cuEGLStreamProducerPresentFrame = 448

cuEGLStreamProducerReturnFrame = 453

cuEventCreate = 118

cuEventCreateFromEGLSync = 479

cuEventDestroy = 122

cuEventDestroy_v2 = 325

cuEventElapsedTime = 123

cuEventElapsedTime_v2 = 780

cuEventQuery = 120

cuEventRecord = 119

cuEventRecordWithFlags = 587

cuEventRecordWithFlags_ptsz = 588

cuEventRecord_ptsz = 441

cuEventSynchronize = 121

cuExternalMemoryGetMappedBuffer = 486

cuExternalMemoryGetMappedMipmappedArray = 487

cuFlushGPUDirectRDMAWrites = 627

cuFuncGetAttribute = 85

cuFuncGetModule = 566

cuFuncGetName = 718

cuFuncGetParamCount = 835

cuFuncGetParamInfo = 733

cuFuncIsLoaded = 741

cuFuncLoad = 742

cuFuncSetAttribute = 481

cuFuncSetBlockShape = 83

cuFuncSetCacheConfig = 86

cuFuncSetSharedMemConfig = 338

cuFuncSetSharedSize = 84

cuGLCtxCreate = 174

cuGLCtxCreate_v2 = 239

cuGLGetDevices = 333

cuGLGetDevices_v2 = 385

cuGLInit = 178

cuGLMapBufferObject = 180

cuGLMapBufferObjectAsync = 184

cuGLMapBufferObjectAsync_v2 = 270

cuGLMapBufferObjectAsync_v2_ptsz = 445

cuGLMapBufferObject_v2 = 269

cuGLMapBufferObject_v2_ptds = 417

cuGLRegisterBufferObject = 179

cuGLSetBufferObjectMapFlags = 183

cuGLUnmapBufferObject = 181

cuGLUnmapBufferObjectAsync = 185

cuGLUnregisterBufferObject = 182

cuGetErrorName = 373

cuGetErrorString = 372

cuGetExportTable = 135

cuGetProcAddress = 626

cuGetProcAddress_v2 = 677

cuGraphAddBatchMemOpNode = 669

cuGraphAddChildGraphNode = 525

cuGraphAddDependencies = 518

cuGraphAddDependencies_v2 = 727

cuGraphAddEmptyNode = 526

cuGraphAddEventRecordNode = 589

cuGraphAddEventWaitNode = 590

cuGraphAddExternalSemaphoresSignalNode = 618

cuGraphAddExternalSemaphoresWaitNode = 621

cuGraphAddHostNode = 530

cuGraphAddKernelNode = 502

cuGraphAddKernelNode_v2 = 689

cuGraphAddMemAllocNode = 638

cuGraphAddMemFreeNode = 639

cuGraphAddMemcpyNode = 504

cuGraphAddMemsetNode = 506

cuGraphAddNode = 712

cuGraphAddNode_v2 = 723

cuGraphBatchMemOpNodeGetParams = 670

cuGraphBatchMemOpNodeSetParams = 671

cuGraphChildGraphNodeGetGraph = 529

cuGraphClone = 523

cuGraphConditionalHandleCreate = 722

cuGraphCreate = 501

cuGraphDebugDotPrint = 628

cuGraphDestroy = 517

cuGraphDestroyNode = 522

cuGraphEventRecordNodeGetEvent = 591

cuGraphEventRecordNodeSetEvent = 593

cuGraphEventWaitNodeGetEvent = 592

cuGraphEventWaitNodeSetEvent = 594

cuGraphExecBatchMemOpNodeSetParams = 672

cuGraphExecChildGraphNodeSetParams = 586

cuGraphExecDestroy = 516

cuGraphExecEventRecordNodeSetEvent = 595

cuGraphExecEventWaitNodeSetEvent = 596

cuGraphExecExternalSemaphoresSignalNodeSetParams = 624

cuGraphExecExternalSemaphoresWaitNodeSetParams = 625

cuGraphExecGetFlags = 658

cuGraphExecGetId = 813

cuGraphExecHostNodeSetParams = 564

cuGraphExecKernelNodeSetParams = 538

cuGraphExecKernelNodeSetParams_v2 = 692

cuGraphExecMemcpyNodeSetParams = 562

cuGraphExecMemsetNodeSetParams = 563

cuGraphExecNodeSetParams = 714

cuGraphExecUpdate = 561

cuGraphExecUpdate_v2 = 696

cuGraphExternalSemaphoresSignalNodeGetParams = 619

cuGraphExternalSemaphoresSignalNodeSetParams = 620

cuGraphExternalSemaphoresWaitNodeGetParams = 622

cuGraphExternalSemaphoresWaitNodeSetParams = 623

cuGraphGetEdges = 535

cuGraphGetEdges_v2 = 724

cuGraphGetId = 812

cuGraphGetNodes = 534

cuGraphGetRootNodes = 510

cuGraphHostNodeGetParams = 531

cuGraphHostNodeSetParams = 533

cuGraphInstantiate = 513

cuGraphInstantiateWithFlags = 643

cuGraphInstantiateWithParams = 656

cuGraphInstantiateWithParams_ptsz = 657

cuGraphInstantiate_v2 = 578

cuGraphKernelNodeCopyAttributes = 569

cuGraphKernelNodeGetAttribute = 570

cuGraphKernelNodeGetParams = 503

cuGraphKernelNodeGetParams_v2 = 690

cuGraphKernelNodeSetAttribute = 571

cuGraphKernelNodeSetParams = 521

cuGraphKernelNodeSetParams_v2 = 691

cuGraphLaunch = 514

cuGraphLaunch_ptsz = 515

cuGraphMemAllocNodeGetParams = 648

cuGraphMemFreeNodeGetParams = 649

cuGraphMemcpyNodeGetParams = 505

cuGraphMemcpyNodeSetParams = 520

cuGraphMemsetNodeGetParams = 507

cuGraphMemsetNodeSetParams = 508

cuGraphNodeFindInClone = 524

cuGraphNodeGetContainingGraph = 809

cuGraphNodeGetDependencies = 511

cuGraphNodeGetDependencies_v2 = 725

cuGraphNodeGetDependentNodes = 512

cuGraphNodeGetDependentNodes_v2 = 726

cuGraphNodeGetEnabled = 651

cuGraphNodeGetLocalId = 810

cuGraphNodeGetParams = 841

cuGraphNodeGetToolsId = 811

cuGraphNodeGetType = 509

cuGraphNodeSetEnabled = 650

cuGraphNodeSetParams = 713

cuGraphReleaseUserObject = 637

cuGraphRemoveDependencies = 519

cuGraphRemoveDependencies_v2 = 728

cuGraphRetainUserObject = 636

cuGraphUpload = 580

cuGraphUpload_ptsz = 581

cuGraphicsD3D10RegisterResource = 140

cuGraphicsD3D11RegisterResource = 153

cuGraphicsD3D9RegisterResource = 156

cuGraphicsEGLRegisterImage = 390

cuGraphicsGLRegisterBuffer = 175

cuGraphicsGLRegisterImage = 176

cuGraphicsMapResources = 133

cuGraphicsMapResources_ptsz = 443

cuGraphicsResourceGetMappedEglFrame = 449

cuGraphicsResourceGetMappedMipmappedArray = 360

cuGraphicsResourceGetMappedPointer = 130

cuGraphicsResourceGetMappedPointer_v2 = 258

cuGraphicsResourceSetMapFlags = 132

cuGraphicsResourceSetMapFlags_v2 = 380

cuGraphicsSubResourceGetMappedArray = 129

cuGraphicsUnmapResources = 134

cuGraphicsUnmapResources_ptsz = 444

cuGraphicsUnregisterResource = 128

cuGraphicsVDPAURegisterOutputSurface = 189

cuGraphicsVDPAURegisterVideoSurface = 188

cuGreenCtxCreate = 743

cuGreenCtxDestroy = 744

cuGreenCtxGetDevResource = 747

cuGreenCtxGetId = 782

cuGreenCtxRecordEvent = 749

cuGreenCtxStreamCreate = 758

cuGreenCtxWaitEvent = 750

cuImportExternalMemory = 485

cuImportExternalSemaphore = 489

cuInit = 1

cuIpcCloseMemHandle = 330

cuIpcGetEventHandle = 334

cuIpcGetMemHandle = 328

cuIpcOpenEventHandle = 335

cuIpcOpenMemHandle = 329

cuIpcOpenMemHandle_v2 = 567

cuKernelGetAttribute = 686

cuKernelGetFunction = 683

cuKernelGetLibrary = 754

cuKernelGetName = 719

cuKernelGetParamCount = 836

cuKernelGetParamInfo = 734

cuKernelSetAttribute = 687

cuKernelSetCacheConfig = 688

cuLaunch = 115

cuLaunchCooperativeKernel = 477

cuLaunchCooperativeKernelMultiDevice = 480

cuLaunchCooperativeKernel_ptsz = 478

cuLaunchGrid = 116

cuLaunchGridAsync = 117

cuLaunchHostFunc = 527

cuLaunchHostFunc_ptsz = 528

cuLaunchHostFunc_v2 = 831

cuLaunchHostFunc_v2_ptsz = 832

cuLaunchKernel = 307

cuLaunchKernelEx = 652

cuLaunchKernelEx_ptsz = 653

cuLaunchKernel_ptsz = 442

cuLibraryEnumerateKernels = 740

cuLibraryGetGlobal = 684

cuLibraryGetKernel = 681

cuLibraryGetKernelCount = 739

cuLibraryGetManaged = 685

cuLibraryGetModule = 682

cuLibraryGetUnifiedFunction = 700

cuLibraryLoadData = 678

cuLibraryLoadFromFile = 679

cuLibraryUnload = 680

cuLinkAddData = 363

cuLinkAddData_v2 = 382

cuLinkAddFile = 364

cuLinkAddFile_v2 = 383

cuLinkComplete = 365

cuLinkCreate = 362

cuLinkCreate_v2 = 381

cuLinkDestroy = 366

cuLogsCurrent = 765

cuLogsDumpToFile = 766

cuLogsDumpToMemory = 767

cuLogsRegisterCallback = 763

cuLogsUnregisterCallback = 764

cuMemAddressFree = 548

cuMemAddressReserve = 547

cuMemAdvise = 457

cuMemAdvise_v2 = 715

cuMemAlloc = 29

cuMemAllocAsync = 598

cuMemAllocAsync_ptsz = 599

cuMemAllocFromPoolAsync = 611

cuMemAllocFromPoolAsync_ptsz = 612

cuMemAllocHost = 37

cuMemAllocHost_v2 = 294

cuMemAllocManaged = 371

cuMemAllocPitch = 31

cuMemAllocPitch_v2 = 244

cuMemAlloc_v2 = 243

cuMemBatchDecompressAsync = 761

cuMemBatchDecompressAsync_ptsz = 762

cuMemCreate = 549

cuMemDiscardAndPrefetchBatchAsync = 791

cuMemDiscardAndPrefetchBatchAsync_ptsz = 792

cuMemDiscardBatchAsync = 789

cuMemDiscardBatchAsync_ptsz = 790

cuMemExportToShareableHandle = 554

cuMemFree = 33

cuMemFreeAsync = 600

cuMemFreeAsync_ptsz = 601

cuMemFreeHost = 38

cuMemFree_v2 = 245

cuMemGetAccess = 558

cuMemGetAddressRange = 35

cuMemGetAddressRange_v2 = 246

cuMemGetAllocationGranularity = 556

cuMemGetAllocationPropertiesFromHandle = 557

cuMemGetDefaultMemPool = 801

cuMemGetHandleForAddressRange = 674

cuMemGetInfo = 27

cuMemGetInfo_v2 = 242

cuMemGetMemPool = 802

cuMemHostAlloc = 39

cuMemHostAlloc_v2 = 271

cuMemHostGetDevicePointer = 40

cuMemHostGetDevicePointer_v2 = 247

cuMemHostGetFlags = 42

cuMemHostRegister = 301

cuMemHostRegister_v2 = 379

cuMemHostUnregister = 302

cuMemImportFromShareableHandle = 555

cuMemMap = 551

cuMemMapArrayAsync = 584

cuMemMapArrayAsync_ptsz = 585

cuMemPeerGetDevicePointer = 317

cuMemPeerRegister = 315

cuMemPeerUnregister = 316

cuMemPoolCreate = 607

cuMemPoolDestroy = 608

cuMemPoolExportPointer = 615

cuMemPoolExportToShareableHandle = 613

cuMemPoolGetAccess = 617

cuMemPoolGetAttribute = 604

cuMemPoolImportFromShareableHandle = 614

cuMemPoolImportPointer = 616

cuMemPoolSetAccess = 605

cuMemPoolSetAttribute = 603

cuMemPoolTrimTo = 602

cuMemPrefetchAsync = 467

cuMemPrefetchAsync_ptsz = 468

cuMemPrefetchAsync_v2 = 716

cuMemPrefetchAsync_v2_ptsz = 717

cuMemPrefetchBatchAsync = 784

cuMemPrefetchBatchAsync_ptsz = 785

cuMemRangeGetAttribute = 471

cuMemRangeGetAttributes = 472

cuMemRelease = 550

cuMemRetainAllocationHandle = 565

cuMemSetAccess = 553

cuMemSetMemPool = 803

cuMemUnmap = 552

cuMemcpy = 305

cuMemcpy2D = 56

cuMemcpy2DAsync = 68

cuMemcpy2DAsync_v2 = 289

cuMemcpy2DAsync_v2_ptsz = 424

cuMemcpy2DUnaligned = 57

cuMemcpy2DUnaligned_v2 = 288

cuMemcpy2DUnaligned_v2_ptds = 406

cuMemcpy2D_v2 = 287

cuMemcpy2D_v2_ptds = 405

cuMemcpy3D = 58

cuMemcpy3DAsync = 69

cuMemcpy3DAsync_v2 = 291

cuMemcpy3DAsync_v2_ptsz = 425

cuMemcpy3DBatchAsync = 778

cuMemcpy3DBatchAsync_ptsz = 779

cuMemcpy3DBatchAsync_v2 = 798

cuMemcpy3DBatchAsync_v2_ptsz = 799

cuMemcpy3DPeer = 320

cuMemcpy3DPeerAsync = 321

cuMemcpy3DPeerAsync_ptsz = 427

cuMemcpy3DPeer_ptds = 410

cuMemcpy3DWithAttributesAsync = 839

cuMemcpy3DWithAttributesAsync_ptsz = 840

cuMemcpy3D_v2 = 290

cuMemcpy3D_v2_ptds = 407

cuMemcpyAsync = 306

cuMemcpyAsync_ptsz = 418

cuMemcpyAtoA = 55

cuMemcpyAtoA_v2 = 286

cuMemcpyAtoA_v2_ptds = 404

cuMemcpyAtoD = 51

cuMemcpyAtoD_v2 = 284

cuMemcpyAtoD_v2_ptds = 401

cuMemcpyAtoH = 54

cuMemcpyAtoHAsync = 67

cuMemcpyAtoHAsync_v2 = 283

cuMemcpyAtoHAsync_v2_ptsz = 420

cuMemcpyAtoH_v2 = 282

cuMemcpyAtoH_v2_ptds = 403

cuMemcpyBatchAsync = 776

cuMemcpyBatchAsync_ptsz = 777

cuMemcpyBatchAsync_v2 = 796

cuMemcpyBatchAsync_v2_ptsz = 797

cuMemcpyDtoA = 49

cuMemcpyDtoA_v2 = 285

cuMemcpyDtoA_v2_ptds = 400

cuMemcpyDtoD = 47

cuMemcpyDtoDAsync = 64

cuMemcpyDtoDAsync_v2 = 281

cuMemcpyDtoDAsync_v2_ptsz = 423

cuMemcpyDtoD_v2 = 280

cuMemcpyDtoD_v2_ptds = 399

cuMemcpyDtoH = 45

cuMemcpyDtoHAsync = 62

cuMemcpyDtoHAsync_v2 = 279

cuMemcpyDtoHAsync_v2_ptsz = 422

cuMemcpyDtoH_v2 = 278

cuMemcpyDtoH_v2_ptds = 398

cuMemcpyHtoA = 53

cuMemcpyHtoAAsync = 66

cuMemcpyHtoAAsync_v2 = 293

cuMemcpyHtoAAsync_v2_ptsz = 419

cuMemcpyHtoA_v2 = 292

cuMemcpyHtoA_v2_ptds = 402

cuMemcpyHtoD = 43

cuMemcpyHtoDAsync = 60

cuMemcpyHtoDAsync_v2 = 277

cuMemcpyHtoDAsync_v2_ptsz = 421

cuMemcpyHtoD_v2 = 276

cuMemcpyHtoD_v2_ptds = 397

cuMemcpyPeer = 318

cuMemcpyPeerAsync = 319

cuMemcpyPeerAsync_ptsz = 426

cuMemcpyPeer_ptds = 409

cuMemcpyWithAttributesAsync = 837

cuMemcpyWithAttributesAsync_ptsz = 838

cuMemcpy_ptds = 408

cuMemcpy_v2 = 248

cuMemsetD16 = 73

cuMemsetD16Async = 218

cuMemsetD16Async_ptsz = 429

cuMemsetD16_v2 = 250

cuMemsetD16_v2_ptds = 412

cuMemsetD2D16 = 79

cuMemsetD2D16Async = 224

cuMemsetD2D16Async_ptsz = 432

cuMemsetD2D16_v2 = 253

cuMemsetD2D16_v2_ptds = 415

cuMemsetD2D32 = 81

cuMemsetD2D32Async = 226

cuMemsetD2D32Async_ptsz = 433

cuMemsetD2D32_v2 = 254

cuMemsetD2D32_v2_ptds = 416

cuMemsetD2D8 = 77

cuMemsetD2D8Async = 222

cuMemsetD2D8Async_ptsz = 431

cuMemsetD2D8_v2 = 252

cuMemsetD2D8_v2_ptds = 414

cuMemsetD32 = 75

cuMemsetD32Async = 220

cuMemsetD32Async_ptsz = 430

cuMemsetD32_v2 = 251

cuMemsetD32_v2_ptds = 413

cuMemsetD8 = 71

cuMemsetD8Async = 216

cuMemsetD8Async_ptsz = 428

cuMemsetD8_v2 = 249

cuMemsetD8_v2_ptds = 411

cuMipmappedArrayCreate = 347

cuMipmappedArrayDestroy = 349

cuMipmappedArrayGetLevel = 348

cuMipmappedArrayGetMemoryRequirements = 655

cuMipmappedArrayGetSparseProperties = 583

cuModuleEnumerateFunctions = 738

cuModuleGetFunction = 23

cuModuleGetFunctionCount = 737

cuModuleGetGlobal = 24

cuModuleGetGlobal_v2 = 241

cuModuleGetLoadingMode = 673

cuModuleGetSurfRef = 190

cuModuleGetTexRef = 26

cuModuleLoad = 18

cuModuleLoadData = 19

cuModuleLoadDataEx = 20

cuModuleLoadFatBinary = 21

cuModuleUnload = 22

cuMulticastAddDevice = 707

cuMulticastBindAddr = 709

cuMulticastBindAddr_v2 = 821

cuMulticastBindMem = 708

cuMulticastBindMem_v2 = 820

cuMulticastCreate = 706

cuMulticastGetGranularity = 711

cuMulticastUnbind = 710

cuNNSetAllocator = 466

cuOccupancyAvailableDynamicSMemPerBlock = 543

cuOccupancyMaxActiveBlocksPerMultiprocessor = 374

cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags = 451

cuOccupancyMaxActiveClusters = 676

cuOccupancyMaxPotentialBlockSize = 384

cuOccupancyMaxPotentialBlockSizeWithFlags = 452

cuOccupancyMaxPotentialClusterSize = 675

cuParamSetSize = 110

cuParamSetTexRef = 114

cuParamSetf = 112

cuParamSeti = 111

cuParamSetv = 113

cuPointerGetAttribute = 310

cuPointerGetAttributes = 450

cuPointerSetAttribute = 378

cuProfilerInitialize = 311

cuProfilerStart = 308

cuProfilerStop = 309

cuSignalExternalSemaphoresAsync = 490

cuSignalExternalSemaphoresAsync_ptsz = 491

cuStreamAddCallback = 346

cuStreamAddCallback_ptsz = 437

cuStreamAttachMemAsync = 377

cuStreamAttachMemAsync_ptsz = 438

cuStreamBatchMemOp = 462

cuStreamBatchMemOp_ptsz = 463

cuStreamBatchMemOp_v2 = 667

cuStreamBatchMemOp_v2_ptsz = 668

cuStreamBeginCapture = 495

cuStreamBeginCaptureToCig = 783

cuStreamBeginCaptureToCig_ptsz = 814

cuStreamBeginCaptureToGraph = 720

cuStreamBeginCaptureToGraph_ptsz = 721

cuStreamBeginCapture_ptsz = 496

cuStreamBeginCapture_v2 = 539

cuStreamBeginCapture_v2_ptsz = 540

cuStreamBeginRecaptureToGraph = 846

cuStreamBeginRecaptureToGraph_ptsz = 847

cuStreamCopyAttributes = 572

cuStreamCopyAttributes_ptsz = 573

cuStreamCreate = 124

cuStreamCreateWithPriority = 367

cuStreamDestroy = 127

cuStreamDestroy_v2 = 326

cuStreamEndCapture = 497

cuStreamEndCaptureToCig = 815

cuStreamEndCaptureToCig_ptsz = 816

cuStreamEndCapture_ptsz = 498

cuStreamGetAttribute = 574

cuStreamGetAttribute_ptsz = 575

cuStreamGetCaptureInfo = 536

cuStreamGetCaptureInfo_ptsz = 537

cuStreamGetCaptureInfo_v2 = 629

cuStreamGetCaptureInfo_v2_ptsz = 630

cuStreamGetCaptureInfo_v3 = 729

cuStreamGetCaptureInfo_v3_ptsz = 730

cuStreamGetCtx = 483

cuStreamGetCtx_ptsz = 484

cuStreamGetCtx_v2 = 759

cuStreamGetCtx_v2_ptsz = 760

cuStreamGetDevResource = 807

cuStreamGetDevResource_ptsz = 808

cuStreamGetDevice = 774

cuStreamGetDevice_ptsz = 775

cuStreamGetFlags = 369

cuStreamGetFlags_ptsz = 435

cuStreamGetGreenCtx = 752

cuStreamGetId = 693

cuStreamGetId_ptsz = 694

cuStreamGetPriority = 368

cuStreamGetPriority_ptsz = 434

cuStreamIsCapturing = 499

cuStreamIsCapturing_ptsz = 500

cuStreamQuery = 125

cuStreamQuery_ptsz = 439

cuStreamSetAttribute = 576

cuStreamSetAttribute_ptsz = 577

cuStreamSetFlags = 559

cuStreamSetFlags_ptsz = 560

cuStreamSynchronize = 126

cuStreamSynchronize_ptsz = 440

cuStreamUpdateCaptureDependencies = 631

cuStreamUpdateCaptureDependencies_ptsz = 632

cuStreamUpdateCaptureDependencies_v2 = 731

cuStreamUpdateCaptureDependencies_v2_ptsz = 732

cuStreamWaitEvent = 295

cuStreamWaitEvent_ptsz = 436

cuStreamWaitValue32 = 458

cuStreamWaitValue32_ptsz = 459

cuStreamWaitValue32_v2 = 659

cuStreamWaitValue32_v2_ptsz = 660

cuStreamWaitValue64 = 473

cuStreamWaitValue64_ptsz = 474

cuStreamWaitValue64_v2 = 661

cuStreamWaitValue64_v2_ptsz = 662

cuStreamWriteValue32 = 460

cuStreamWriteValue32_ptsz = 461

cuStreamWriteValue32_v2 = 663

cuStreamWriteValue32_v2_ptsz = 664

cuStreamWriteValue64 = 475

cuStreamWriteValue64_ptsz = 476

cuStreamWriteValue64_v2 = 665

cuStreamWriteValue64_v2_ptsz = 666

cuSurfObjectCreate = 343

cuSurfObjectDestroy = 344

cuSurfObjectGetResourceDesc = 345

cuSurfRefCreate = 191

cuSurfRefDestroy = 192

cuSurfRefGetArray = 196

cuSurfRefGetFormat = 195

cuSurfRefSetArray = 194

cuSurfRefSetFormat = 193

cuTensorMapEncodeIm2col = 698

cuTensorMapEncodeIm2colWide = 781

cuTensorMapEncodeTiled = 697

cuTensorMapReplaceAddress = 699

cuTexObjectCreate = 339

cuTexObjectDestroy = 340

cuTexObjectGetResourceDesc = 341

cuTexObjectGetResourceViewDesc = 361

cuTexObjectGetTextureDesc = 342

cuTexRefCreate = 92

cuTexRefDestroy = 93

cuTexRefGetAddress = 103

cuTexRefGetAddressMode = 106

cuTexRefGetAddress_v2 = 257

cuTexRefGetArray = 105

cuTexRefGetBorderColor = 456

cuTexRefGetFilterMode = 107

cuTexRefGetFlags = 109

cuTexRefGetFormat = 108

cuTexRefGetMaxAnisotropy = 359

cuTexRefGetMipmapFilterMode = 356

cuTexRefGetMipmapLevelBias = 357

cuTexRefGetMipmapLevelClamp = 358

cuTexRefGetMipmappedArray = 355

cuTexRefSetAddress = 95

cuTexRefSetAddress2D = 97

cuTexRefSetAddress2D_v2 = 256

cuTexRefSetAddress2D_v3 = 327

cuTexRefSetAddressMode = 100

cuTexRefSetAddress_v2 = 255

cuTexRefSetArray = 94

cuTexRefSetBorderColor = 455

cuTexRefSetFilterMode = 101

cuTexRefSetFlags = 102

cuTexRefSetFormat = 99

cuTexRefSetMaxAnisotropy = 354

cuTexRefSetMipmapFilterMode = 351

cuTexRefSetMipmapLevelBias = 352

cuTexRefSetMipmapLevelClamp = 353

cuTexRefSetMipmappedArray = 350

cuThreadExchangeStreamCaptureMode = 541

cuUserObjectCreate = 633

cuUserObjectRelease = 635

cuUserObjectRetain = 634

cuVDPAUCtxCreate = 187

cuVDPAUCtxCreate_v2 = 240

cuVDPAUGetDevice = 186

cuWGLGetDevice = 177

cuWaitExternalSemaphoresAsync = 492

cuWaitExternalSemaphoresAsync_ptsz = 493

reserved464 = 464

reserved465 = 465

reserved469 = 469

reserved786 = 786

reserved787 = 787

reserved788 = 788

reserved793 = 793

reserved794 = 794

reserved817 = 817

reserved818 = 818

reserved819 = 819

reserved823 = 823

reserved824 = 824

reserved825 = 825

reserved826 = 826

reserved827 = 827

reserved828 = 828

reserved829 = 829

reserved830 = 830

reserved833 = 833

reserved834 = 834

reserved848 = 848

reserved849 = 849

reserved850 = 850

reserved851 = 851

reserved852 = 852

reserved853 = 853

reserved854 = 854

reserved855 = 855

reserved856 = 856

reserved857 = 857

reserved858 = 858

reserved859 = 859

class cupti.cupti.Nvtx_api_trace_cbid(value)

Bases: IntEnum

See CUpti_nvtx_api_trace_cbid.

FORCE_INT = 2147483647

INVALID = 0

SIZE = 50

nvtxDomainCreateA = 41

nvtxDomainCreateW = 42

nvtxDomainDestroy = 43

nvtxDomainMarkEx = 30

nvtxDomainNameCategoryA = 37

nvtxDomainNameCategoryW = 38

nvtxDomainRangeEnd = 32

nvtxDomainRangePop = 34

nvtxDomainRangePushEx = 33

nvtxDomainRangeStartEx = 31

nvtxDomainRegisterStringA = 39

nvtxDomainRegisterStringW = 40

nvtxDomainResourceCreate = 35

nvtxDomainResourceDestroy = 36

nvtxDomainSyncUserAcquireFailed = 47

nvtxDomainSyncUserAcquireStart = 46

nvtxDomainSyncUserAcquireSuccess = 48

nvtxDomainSyncUserCreate = 44

nvtxDomainSyncUserDestroy = 45

nvtxDomainSyncUserReleasing = 49

nvtxMarkA = 1

nvtxMarkEx = 3

nvtxMarkW = 2

nvtxNameCategoryA = 12

nvtxNameCategoryW = 13

nvtxNameCuContextA = 18

nvtxNameCuContextW = 19

nvtxNameCuDeviceA = 16

nvtxNameCuDeviceW = 17

nvtxNameCuEventA = 22

nvtxNameCuEventW = 23

nvtxNameCuStreamA = 20

nvtxNameCuStreamW = 21

nvtxNameCudaDeviceA = 24

nvtxNameCudaDeviceW = 25

nvtxNameCudaEventA = 28

nvtxNameCudaEventW = 29

nvtxNameCudaStreamA = 26

nvtxNameCudaStreamW = 27

nvtxNameOsThreadA = 14

nvtxNameOsThreadW = 15

nvtxRangeEnd = 7

nvtxRangePop = 11

nvtxRangePushA = 8

nvtxRangePushEx = 10

nvtxRangePushW = 9

nvtxRangeStartA = 4

nvtxRangeStartEx = 6

nvtxRangeStartW = 5

class cupti.cupti.Runtime_api_trace_cbid(value)

Bases: IntEnum

See CUpti_runtime_api_trace_cbid.

FORCE_INT = 2147483647

INVALID = 0

SIZE = 559

cuda470_v12060 = 470

cuda471_v12060 = 471

cuda472_v12060 = 472

cuda473_v12060 = 473

cuda474_v12060 = 474

cuda475_v12060 = 475

cuda476_v12060 = 476

cuda477_v12060 = 477

cuda478_v12060 = 478

cuda479_v12060 = 479

cuda531_v13010 = 531

cuda532_v13010 = 532

cuda545_v13010 = 545

cuda546_v13010 = 546

cuda547_v13010 = 547

cuda553_v13020 = 553

cudaArrayGetInfo_v4010 = 181

cudaArrayGetMemoryRequirements_v11060 = 428

cudaArrayGetPlane_v11020 = 381

cudaArrayGetSparseProperties_v11010 = 359

cudaBindSurfaceToArray_v3020 = 61

cudaBindTexture2D_v3020 = 56

cudaBindTextureToArray_v3020 = 57

cudaBindTextureToMipmappedArray_v5000 = 195

cudaBindTexture_v3020 = 55

cudaChooseDevice_v3020 = 5

cudaConfigureCall_v3020 = 8

cudaCreateChannelDesc_v3020 = 7

cudaCreateSurfaceObject_v5000 = 189

cudaCreateTextureObject_v2_v11080 = 434

cudaCreateTextureObject_v5000 = 185

cudaCtxResetPersistingL2Cache_v11000 = 337

cudaD3D10GetDevice_v3020 = 88

cudaD3D10GetDevices_v3020 = 89

cudaD3D10GetDirect3DDevice_v3020 = 149

cudaD3D10MapResources_v3020 = 94

cudaD3D10RegisterResource_v3020 = 92

cudaD3D10ResourceGetMappedArray_v3020 = 98

cudaD3D10ResourceGetMappedPitch_v3020 = 101

cudaD3D10ResourceGetMappedPointer_v3020 = 99

cudaD3D10ResourceGetMappedSize_v3020 = 100

cudaD3D10ResourceGetSurfaceDimensions_v3020 = 97

cudaD3D10ResourceSetMapFlags_v3020 = 96

cudaD3D10SetDirect3DDevice_v3020 = 90

cudaD3D10UnmapResources_v3020 = 95

cudaD3D10UnregisterResource_v3020 = 93

cudaD3D11GetDevice_v3020 = 84

cudaD3D11GetDevices_v3020 = 85

cudaD3D11GetDirect3DDevice_v3020 = 148

cudaD3D11SetDirect3DDevice_v3020 = 86

cudaD3D9Begin_v3020 = 117

cudaD3D9End_v3020 = 118

cudaD3D9GetDevice_v3020 = 102

cudaD3D9GetDevices_v3020 = 103

cudaD3D9GetDirect3DDevice_v3020 = 105

cudaD3D9MapResources_v3020 = 109

cudaD3D9MapVertexBuffer_v3020 = 121

cudaD3D9RegisterResource_v3020 = 107

cudaD3D9RegisterVertexBuffer_v3020 = 119

cudaD3D9ResourceGetMappedArray_v3020 = 113

cudaD3D9ResourceGetMappedPitch_v3020 = 116

cudaD3D9ResourceGetMappedPointer_v3020 = 114

cudaD3D9ResourceGetMappedSize_v3020 = 115

cudaD3D9ResourceGetSurfaceDimensions_v3020 = 112

cudaD3D9ResourceSetMapFlags_v3020 = 111

cudaD3D9SetDirect3DDevice_v3020 = 104

cudaD3D9UnmapResources_v3020 = 110

cudaD3D9UnmapVertexBuffer_v3020 = 122

cudaD3D9UnregisterResource_v3020 = 108

cudaD3D9UnregisterVertexBuffer_v3020 = 120

cudaDestroyExternalMemory_v10000 = 277

cudaDestroyExternalSemaphore_v10000 = 283

cudaDestroySurfaceObject_v5000 = 190

cudaDestroyTextureObject_v5000 = 186

cudaDevResourceGenerateDesc_v13010 = 525

cudaDevSmResourceSplitByCount_v13010 = 524

cudaDevSmResourceSplit_v13010 = 549

cudaDeviceCanAccessPeer_v4000 = 154

cudaDeviceDisablePeerAccess_v4000 = 156

cudaDeviceEnablePeerAccess_v4000 = 155

cudaDeviceFlushGPUDirectRDMAWrites_v11030 = 405

cudaDeviceGetAttribute_v5000 = 200

cudaDeviceGetByPCIBusId_v4010 = 173

cudaDeviceGetCacheConfig_v3020 = 168

cudaDeviceGetDefaultMemPool_v11020 = 372

cudaDeviceGetDevResource_v13010 = 523

cudaDeviceGetExecutionCtx_v13010 = 548

cudaDeviceGetGraphMemAttribute_v11040 = 424

cudaDeviceGetHostAtomicCapabilities_v13000 = 521

cudaDeviceGetLimit_v3020 = 166

cudaDeviceGetMemPool_v11020 = 386

cudaDeviceGetNvSciSyncAttributes_v10020 = 328

cudaDeviceGetP2PAtomicCapabilities_v13000 = 522

cudaDeviceGetP2PAttribute_v8000 = 255

cudaDeviceGetPCIBusId_v4010 = 174

cudaDeviceGetSharedMemConfig_v4020 = 183

cudaDeviceGetStreamPriorityRange_v5050 = 205

cudaDeviceGetTexture1DLinearMaxWidth_v11010 = 347

cudaDeviceGraphMemTrim_v11040 = 423

cudaDeviceRegisterAsyncNotification_v12040 = 465

cudaDeviceReset_v3020 = 164

cudaDeviceSetCacheConfig_v3020 = 169

cudaDeviceSetGraphMemAttribute_v11040 = 425

cudaDeviceSetLimit_v3020 = 167

cudaDeviceSetMemPool_v11020 = 385

cudaDeviceSetSharedMemConfig_v4020 = 184

cudaDeviceSynchronize_v3020 = 165

cudaDeviceUnregisterAsyncNotification_v12040 = 466

cudaDriverGetVersion_v3020 = 1

cudaEGLStreamConsumerAcquireFrame_v7000 = 259

cudaEGLStreamConsumerConnectWithFlags_v7000 = 268

cudaEGLStreamConsumerConnect_v7000 = 257

cudaEGLStreamConsumerDisconnect_v7000 = 258

cudaEGLStreamConsumerReleaseFrame_v7000 = 260

cudaEGLStreamProducerConnect_v7000 = 261

cudaEGLStreamProducerDisconnect_v7000 = 262

cudaEGLStreamProducerPresentFrame_v7000 = 263

cudaEGLStreamProducerReturnFrame_v7000 = 264

cudaEventCreateFromEGLSync_v9000 = 271

cudaEventCreateWithFlags_v3020 = 134

cudaEventCreate_v3020 = 133

cudaEventDestroy_v3020 = 136

cudaEventElapsedTime_v12080 = 486

cudaEventElapsedTime_v2_v12080 = 486

cudaEventElapsedTime_v3020 = 139

cudaEventQuery_v3020 = 138

cudaEventRecordWithFlags_ptsz_v11010 = 371

cudaEventRecordWithFlags_v11010 = 370

cudaEventRecord_ptsz_v7000 = 242

cudaEventRecord_v3020 = 135

cudaEventSynchronize_v3020 = 137

cudaExecutionCtxDestroy_v13010 = 527

cudaExecutionCtxGetDevResource_v13010 = 528

cudaExecutionCtxGetDevice_v13010 = 529

cudaExecutionCtxGetId_v13010 = 530

cudaExecutionCtxRecordEvent_v13010 = 543

cudaExecutionCtxStreamCreate_v13010 = 533

cudaExecutionCtxSynchronize_v13010 = 534

cudaExecutionCtxWaitEvent_v13010 = 544

cudaExternalMemoryGetMappedBuffer_v10000 = 275

cudaExternalMemoryGetMappedMipmappedArray_v10000 = 276

cudaFreeArray_v3020 = 24

cudaFreeAsync_ptsz_v11020 = 376

cudaFreeAsync_v11020 = 375

cudaFreeHost_v3020 = 26

cudaFreeMipmappedArray_v5000 = 194

cudaFree_v3020 = 22

cudaFuncGetAttributes_v3020 = 15

cudaFuncGetName_v12030 = 451

cudaFuncGetParamCount_v13020 = 552

cudaFuncGetParamInfo_v12040 = 467

cudaFuncSetAttribute_v9000 = 273

cudaFuncSetCacheConfig_v3020 = 14

cudaFuncSetSharedMemConfig_v4020 = 182

cudaGLGetDevices_v4010 = 175

cudaGLMapBufferObjectAsync_v3020 = 69

cudaGLMapBufferObject_v3020 = 65

cudaGLRegisterBufferObject_v3020 = 64

cudaGLSetBufferObjectMapFlags_v3020 = 68

cudaGLSetGLDevice_v3020 = 63

cudaGLUnmapBufferObjectAsync_v3020 = 70

cudaGLUnmapBufferObject_v3020 = 66

cudaGLUnregisterBufferObject_v3020 = 67

cudaGetChannelDesc_v3020 = 6

cudaGetDeviceCount_v3020 = 3

cudaGetDeviceFlags_v7000 = 212

cudaGetDeviceProperties_v12000 = 440

cudaGetDeviceProperties_v2_v12000 = 440

cudaGetDeviceProperties_v3020 = 4

cudaGetDevice_v3020 = 17

cudaGetDriverEntryPointByVersion_ptsz_v12050 = 469

cudaGetDriverEntryPointByVersion_v12050 = 468

cudaGetDriverEntryPoint_ptsz_v11030 = 407

cudaGetDriverEntryPoint_v11030 = 406

cudaGetErrorName_v6050 = 209

cudaGetErrorString_v3020 = 12

cudaGetExportTable_v13000 = 493

cudaGetFuncBySymbol_v11000 = 336

cudaGetKernel_v12000 = 439

cudaGetLastError_v3020 = 10

cudaGetMipmappedArrayLevel_v5000 = 193

cudaGetSurfaceObjectResourceDesc_v5000 = 191

cudaGetSurfaceReference_v3020 = 62

cudaGetSymbolAddress_v3020 = 53

cudaGetSymbolSize_v3020 = 54

cudaGetTextureAlignmentOffset_v3020 = 59

cudaGetTextureObjectResourceDesc_v5000 = 187

cudaGetTextureObjectResourceViewDesc_v5000 = 199

cudaGetTextureObjectTextureDesc_v2_v11080 = 435

cudaGetTextureObjectTextureDesc_v5000 = 188

cudaGetTextureReference_v3020 = 60

cudaGraphAddChildGraphNode_v10000 = 298

cudaGraphAddDependencies_v10000 = 307

cudaGraphAddDependencies_v12030 = 458

cudaGraphAddDependencies_v2_v12030 = 458

cudaGraphAddEmptyNode_v10000 = 300

cudaGraphAddEventRecordNode_v11010 = 362

cudaGraphAddEventWaitNode_v11010 = 365

cudaGraphAddExternalSemaphoresSignalNode_v11020 = 397

cudaGraphAddExternalSemaphoresWaitNode_v11020 = 400

cudaGraphAddHostNode_v10000 = 296

cudaGraphAddKernelNode_v10000 = 289

cudaGraphAddMemAllocNode_v11040 = 419

cudaGraphAddMemFreeNode_v11040 = 421

cudaGraphAddMemcpyNode1D_v11010 = 352

cudaGraphAddMemcpyNodeFromSymbol_v11010 = 351

cudaGraphAddMemcpyNodeToSymbol_v11010 = 350

cudaGraphAddMemcpyNode_v10000 = 290

cudaGraphAddMemsetNode_v10000 = 293

cudaGraphAddNode_v12020 = 445

cudaGraphAddNode_v12030 = 460

cudaGraphAddNode_v2_v12030 = 460

cudaGraphChildGraphNodeGetGraph_v10000 = 299

cudaGraphClone_v10000 = 301

cudaGraphConditionalHandleCreate_v12030 = 454

cudaGraphConditionalHandleCreate_v2_v13010 = 535

cudaGraphCreate_v10000 = 286

cudaGraphDebugDotPrint_v11030 = 408

cudaGraphDestroyNode_v10000 = 309

cudaGraphDestroy_v10000 = 314

cudaGraphEventRecordNodeGetEvent_v11010 = 363

cudaGraphEventRecordNodeSetEvent_v11010 = 364

cudaGraphEventWaitNodeGetEvent_v11010 = 366

cudaGraphEventWaitNodeSetEvent_v11010 = 367

cudaGraphExecChildGraphNodeSetParams_v11010 = 361

cudaGraphExecDestroy_v10000 = 313

cudaGraphExecEventRecordNodeSetEvent_v11010 = 368

cudaGraphExecEventWaitNodeSetEvent_v11010 = 369

cudaGraphExecExternalSemaphoresSignalNodeSetParams_v11020 = 403

cudaGraphExecExternalSemaphoresWaitNodeSetParams_v11020 = 404

cudaGraphExecGetFlags_v12000 = 438

cudaGraphExecGetId_v13010 = 542

cudaGraphExecHostNodeSetParams_v10020 = 334

cudaGraphExecKernelNodeSetParams_v10010 = 326

cudaGraphExecMemcpyNodeSetParams1D_v11010 = 358

cudaGraphExecMemcpyNodeSetParamsFromSymbol_v11010 = 357

cudaGraphExecMemcpyNodeSetParamsToSymbol_v11010 = 356

cudaGraphExecMemcpyNodeSetParams_v10020 = 332

cudaGraphExecMemsetNodeSetParams_v10020 = 333

cudaGraphExecNodeSetParams_v12020 = 447

cudaGraphExecUpdate_v10020 = 335

cudaGraphExternalSemaphoresSignalNodeGetParams_v11020 = 398

cudaGraphExternalSemaphoresSignalNodeSetParams_v11020 = 399

cudaGraphExternalSemaphoresWaitNodeGetParams_v11020 = 401

cudaGraphExternalSemaphoresWaitNodeSetParams_v11020 = 402

cudaGraphGetEdges_v10000 = 323

cudaGraphGetEdges_v12030 = 455

cudaGraphGetEdges_v2_v12030 = 455

cudaGraphGetId_v13010 = 541

cudaGraphGetNodes_v10000 = 322

cudaGraphGetRootNodes_v10000 = 304

cudaGraphHostNodeGetParams_v10000 = 297

cudaGraphHostNodeSetParams_v10000 = 321

cudaGraphInstantiateWithFlags_v11040 = 418

cudaGraphInstantiateWithParams_ptsz_v12000 = 437

cudaGraphInstantiateWithParams_v12000 = 436

cudaGraphInstantiate_v10000 = 310

cudaGraphInstantiate_v12000 = 443

cudaGraphKernelNodeCopyAttributes_v11000 = 338

cudaGraphKernelNodeGetAttribute_v11000 = 339

cudaGraphKernelNodeGetParams_v10000 = 287

cudaGraphKernelNodeSetAttribute_v11000 = 340

cudaGraphKernelNodeSetParams_v10000 = 288

cudaGraphLaunch_ptsz_v10000 = 312

cudaGraphLaunch_v10000 = 311

cudaGraphMemAllocNodeGetParams_v11040 = 420

cudaGraphMemFreeNodeGetParams_v11040 = 422

cudaGraphMemcpyNodeGetParams_v10000 = 291

cudaGraphMemcpyNodeSetParams1D_v11010 = 355

cudaGraphMemcpyNodeSetParamsFromSymbol_v11010 = 354

cudaGraphMemcpyNodeSetParamsToSymbol_v11010 = 353

cudaGraphMemcpyNodeSetParams_v10000 = 292

cudaGraphMemsetNodeGetParams_v10000 = 294

cudaGraphMemsetNodeSetParams_v10000 = 295

cudaGraphNodeFindInClone_v10000 = 302

cudaGraphNodeGetContainingGraph_v13010 = 538

cudaGraphNodeGetDependencies_v10000 = 305

cudaGraphNodeGetDependencies_v12030 = 456

cudaGraphNodeGetDependencies_v2_v12030 = 456

cudaGraphNodeGetDependentNodes_v10000 = 306

cudaGraphNodeGetDependentNodes_v12030 = 457

cudaGraphNodeGetDependentNodes_v2_v12030 = 457

cudaGraphNodeGetEnabled_v11060 = 427

cudaGraphNodeGetLocalId_v13010 = 539

cudaGraphNodeGetParams_v13020 = 558

cudaGraphNodeGetToolsId_v13010 = 540

cudaGraphNodeGetType_v10000 = 303

cudaGraphNodeSetEnabled_v11060 = 426

cudaGraphNodeSetParams_v12020 = 446

cudaGraphReleaseUserObject_v11030 = 417

cudaGraphRemoveDependencies_v10000 = 308

cudaGraphRemoveDependencies_v12030 = 459

cudaGraphRemoveDependencies_v2_v12030 = 459

cudaGraphRetainUserObject_v11030 = 416

cudaGraphUpload_ptsz_v10000 = 349

cudaGraphUpload_v10000 = 348

cudaGraphicsD3D10RegisterResource_v3020 = 91

cudaGraphicsD3D11RegisterResource_v3020 = 87

cudaGraphicsD3D9RegisterResource_v3020 = 106

cudaGraphicsEGLRegisterImage_v7000 = 256

cudaGraphicsGLRegisterBuffer_v3020 = 73

cudaGraphicsGLRegisterImage_v3020 = 72

cudaGraphicsMapResources_v3020 = 76

cudaGraphicsResourceGetMappedEglFrame_v7000 = 265

cudaGraphicsResourceGetMappedMipmappedArray_v5000 = 196

cudaGraphicsResourceGetMappedPointer_v3020 = 78

cudaGraphicsResourceSetMapFlags_v3020 = 75

cudaGraphicsSubResourceGetMappedArray_v3020 = 79

cudaGraphicsUnmapResources_v3020 = 77

cudaGraphicsUnregisterResource_v3020 = 74

cudaGraphicsVDPAURegisterOutputSurface_v3020 = 83

cudaGraphicsVDPAURegisterVideoSurface_v3020 = 82

cudaGreenCtxCreate_v13010 = 526

cudaHostAlloc_v3020 = 27

cudaHostGetDevicePointer_v3020 = 28

cudaHostGetFlags_v3020 = 29

cudaHostRegister_v4000 = 152

cudaHostUnregister_v4000 = 153

cudaImportExternalMemory_v10000 = 274

cudaImportExternalSemaphore_v10000 = 278

cudaInitDevice_v12000 = 444

cudaIpcCloseMemHandle_v4010 = 180

cudaIpcGetEventHandle_v4010 = 176

cudaIpcGetMemHandle_v4010 = 178

cudaIpcOpenEventHandle_v4010 = 177

cudaIpcOpenMemHandle_v4010 = 179

cudaKernelSetAttributeForDevice_v12060 = 479

cudaLaunchCooperativeKernelMultiDevice_v9000 = 272

cudaLaunchCooperativeKernel_ptsz_v9000 = 270

cudaLaunchCooperativeKernel_v9000 = 269

cudaLaunchHostFunc_ptsz_v10000 = 285

cudaLaunchHostFunc_v10000 = 284

cudaLaunchHostFunc_v2_ptsz_v13020 = 551

cudaLaunchHostFunc_v2_v13020 = 550

cudaLaunchKernelExC_ptsz_v11060 = 431

cudaLaunchKernelExC_v11060 = 430

cudaLaunchKernel_ptsz_v7000 = 214

cudaLaunchKernel_v7000 = 211

cudaLaunch_ptsz_v7000 = 213

cudaLaunch_v3020 = 13

cudaLibraryEnumerateKernels_v12060 = 478

cudaLibraryGetGlobal_v12060 = 474

cudaLibraryGetKernelCount_v12060 = 477

cudaLibraryGetKernel_v12060 = 473

cudaLibraryGetManaged_v12060 = 475

cudaLibraryGetUnifiedFunction_v12060 = 476

cudaLibraryLoadData_v12060 = 470

cudaLibraryLoadFromFile_v12060 = 471

cudaLibraryUnload_v12060 = 472

cudaLogsCurrent_v13000 = 515

cudaLogsDumpToFile_v13000 = 516

cudaLogsDumpToMemory_v13000 = 517

cudaLogsRegisterCallback_v13000 = 513

cudaLogsUnregisterCallback_v13000 = 514

cudaMalloc3DArray_v3020 = 141

cudaMalloc3D_v3020 = 140

cudaMallocArray_v3020 = 23

cudaMallocAsync_ptsz_v11020 = 374

cudaMallocAsync_v11020 = 373

cudaMallocFromPoolAsync_ptsz_v11020 = 392

cudaMallocFromPoolAsync_v11020 = 391

cudaMallocHost_v3020 = 25

cudaMallocManaged_v6000 = 206

cudaMallocMipmappedArray_v5000 = 192

cudaMallocPitch_v3020 = 21

cudaMalloc_v3020 = 20

cudaMemAdvise_v12020 = 448

cudaMemAdvise_v2_v12020 = 448

cudaMemAdvise_v8000 = 254

cudaMemDiscardAndPrefetchBatchAsync_ptsz_v13000 = 492

cudaMemDiscardAndPrefetchBatchAsync_v13000 = 491

cudaMemDiscardBatchAsync_ptsz_v13000 = 490

cudaMemDiscardBatchAsync_v13000 = 489

cudaMemGetDefaultMemPool_v13000 = 518

cudaMemGetInfo_v3020 = 30

cudaMemGetMemPool_v13000 = 519

cudaMemPoolCreate_v11020 = 383

cudaMemPoolDestroy_v11020 = 384

cudaMemPoolExportPointer_v11020 = 389

cudaMemPoolExportToShareableHandle_v11020 = 387

cudaMemPoolGetAccess_v11020 = 382

cudaMemPoolGetAttribute_v11020 = 379

cudaMemPoolImportFromShareableHandle_v11020 = 388

cudaMemPoolImportPointer_v11020 = 390

cudaMemPoolSetAccess_v11020 = 380

cudaMemPoolSetAttribute_v11020 = 378

cudaMemPoolTrimTo_v11020 = 377

cudaMemPrefetchAsync_ptsz_v12020 = 450

cudaMemPrefetchAsync_ptsz_v8000 = 253

cudaMemPrefetchAsync_v12020 = 449

cudaMemPrefetchAsync_v2_ptsz_v12020 = 450

cudaMemPrefetchAsync_v2_v12020 = 449

cudaMemPrefetchAsync_v8000 = 252

cudaMemPrefetchBatchAsync_ptsz_v13000 = 488

cudaMemPrefetchBatchAsync_v13000 = 487

cudaMemRangeGetAttribute_v8000 = 266

cudaMemRangeGetAttributes_v8000 = 267

cudaMemSetMemPool_v13000 = 520

cudaMemcpy2DArrayToArray_ptds_v7000 = 222

cudaMemcpy2DArrayToArray_v3020 = 38

cudaMemcpy2DAsync_ptsz_v7000 = 228

cudaMemcpy2DAsync_v3020 = 44

cudaMemcpy2DFromArrayAsync_ptsz_v7000 = 230

cudaMemcpy2DFromArrayAsync_v3020 = 46

cudaMemcpy2DFromArray_ptds_v7000 = 220

cudaMemcpy2DFromArray_v3020 = 36

cudaMemcpy2DToArrayAsync_ptsz_v7000 = 229

cudaMemcpy2DToArrayAsync_v3020 = 45

cudaMemcpy2DToArray_ptds_v7000 = 218

cudaMemcpy2DToArray_v3020 = 34

cudaMemcpy2D_ptds_v7000 = 216

cudaMemcpy2D_v3020 = 32

cudaMemcpy3DAsync_ptsz_v7000 = 246

cudaMemcpy3DAsync_v3020 = 145

cudaMemcpy3DBatchAsync_ptsz_v12080 = 485

cudaMemcpy3DBatchAsync_ptsz_v13000 = 512

cudaMemcpy3DBatchAsync_v12080 = 484

cudaMemcpy3DBatchAsync_v13000 = 511

cudaMemcpy3DPeerAsync_ptsz_v7000 = 250

cudaMemcpy3DPeerAsync_v4000 = 163

cudaMemcpy3DPeer_ptds_v7000 = 249

cudaMemcpy3DPeer_v4000 = 162

cudaMemcpy3DWithAttributesAsync_ptsz_v13020 = 557

cudaMemcpy3DWithAttributesAsync_v13020 = 556

cudaMemcpy3D_ptds_v7000 = 245

cudaMemcpy3D_v3020 = 144

cudaMemcpyArrayToArray_ptds_v7000 = 221

cudaMemcpyArrayToArray_v3020 = 37

cudaMemcpyAsync_ptsz_v7000 = 225

cudaMemcpyAsync_v3020 = 41

cudaMemcpyBatchAsync_ptsz_v12080 = 483

cudaMemcpyBatchAsync_ptsz_v13000 = 510

cudaMemcpyBatchAsync_v12080 = 482

cudaMemcpyBatchAsync_v13000 = 509

cudaMemcpyFromArrayAsync_ptsz_v7000 = 227

cudaMemcpyFromArrayAsync_v3020 = 43

cudaMemcpyFromArray_ptds_v7000 = 219

cudaMemcpyFromArray_v3020 = 35

cudaMemcpyFromSymbolAsync_ptsz_v7000 = 232

cudaMemcpyFromSymbolAsync_v3020 = 48

cudaMemcpyFromSymbol_ptds_v7000 = 224

cudaMemcpyFromSymbol_v3020 = 40

cudaMemcpyPeerAsync_v4000 = 161

cudaMemcpyPeer_v4000 = 160

cudaMemcpyToArrayAsync_ptsz_v7000 = 226

cudaMemcpyToArrayAsync_v3020 = 42

cudaMemcpyToArray_ptds_v7000 = 217

cudaMemcpyToArray_v3020 = 33

cudaMemcpyToSymbolAsync_ptsz_v7000 = 231

cudaMemcpyToSymbolAsync_v3020 = 47

cudaMemcpyToSymbol_ptds_v7000 = 223

cudaMemcpyToSymbol_v3020 = 39

cudaMemcpyWithAttributesAsync_ptsz_v13020 = 555

cudaMemcpyWithAttributesAsync_v13020 = 554

cudaMemcpy_ptds_v7000 = 215

cudaMemcpy_v3020 = 31

cudaMemset2DAsync_ptsz_v7000 = 236

cudaMemset2DAsync_v3020 = 52

cudaMemset2D_ptds_v7000 = 234

cudaMemset2D_v3020 = 50

cudaMemset3DAsync_ptsz_v7000 = 244

cudaMemset3DAsync_v3020 = 143

cudaMemset3D_ptds_v7000 = 243

cudaMemset3D_v3020 = 142

cudaMemsetAsync_ptsz_v7000 = 235

cudaMemsetAsync_v3020 = 51

cudaMemset_ptds_v7000 = 233

cudaMemset_v3020 = 49

cudaMipmappedArrayGetMemoryRequirements_v11060 = 429

cudaMipmappedArrayGetSparseProperties_v11010 = 360

cudaOccupancyAvailableDynamicSMemPerBlock_v10200 = 329

cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags_v7000 = 251

cudaOccupancyMaxActiveBlocksPerMultiprocessor_v6000 = 207

cudaOccupancyMaxActiveBlocksPerMultiprocessor_v6050 = 210

cudaOccupancyMaxActiveClusters_v11070 = 433

cudaOccupancyMaxPotentialClusterSize_v11070 = 432

cudaPeekAtLastError_v3020 = 11

cudaPeerGetDevicePointer_v4000 = 159

cudaPeerRegister_v4000 = 157

cudaPeerUnregister_v4000 = 158

cudaPointerGetAttributes_v4000 = 151

cudaProfilerInitialize_v4000 = 170

cudaProfilerStart_v4000 = 171

cudaProfilerStop_v4000 = 172

cudaRuntimeGetVersion_v3020 = 2

cudaSetDeviceFlags_v3020 = 19

cudaSetDevice_v3020 = 16

cudaSetDoubleForDevice_v3020 = 124

cudaSetDoubleForHost_v3020 = 125

cudaSetValidDevices_v3020 = 18

cudaSetupArgument_v3020 = 9

cudaSignalExternalSemaphoresAsync_ptsz_v10000 = 280

cudaSignalExternalSemaphoresAsync_ptsz_v11020 = 394

cudaSignalExternalSemaphoresAsync_v10000 = 279

cudaSignalExternalSemaphoresAsync_v11020 = 393

cudaSignalExternalSemaphoresAsync_v2_ptsz_v11020 = 394

cudaSignalExternalSemaphoresAsync_v2_v11020 = 393

cudaStreamAddCallback_ptsz_v7000 = 248

cudaStreamAddCallback_v5000 = 197

cudaStreamAttachMemAsync_ptsz_v7000 = 241

cudaStreamAttachMemAsync_v6000 = 208

cudaStreamBeginCaptureToGraph_ptsz_v12030 = 453

cudaStreamBeginCaptureToGraph_v12030 = 452

cudaStreamBeginCapture_ptsz_v10000 = 316

cudaStreamBeginCapture_v10000 = 315

cudaStreamCopyAttributes_ptsz_v11000 = 342

cudaStreamCopyAttributes_v11000 = 341

cudaStreamCreateWithFlags_v5000 = 198

cudaStreamCreateWithPriority_v5050 = 202

cudaStreamCreate_v3020 = 129

cudaStreamDestroy_v3020 = 130

cudaStreamDestroy_v5050 = 201

cudaStreamEndCapture_ptsz_v10000 = 320

cudaStreamEndCapture_v10000 = 319

cudaStreamGetAttribute_ptsz_v11000 = 344

cudaStreamGetAttribute_v11000 = 343

cudaStreamGetCaptureInfo_ptsz_v10010 = 325

cudaStreamGetCaptureInfo_ptsz_v12030 = 462

cudaStreamGetCaptureInfo_v10010 = 324

cudaStreamGetCaptureInfo_v12030 = 461

cudaStreamGetCaptureInfo_v2_ptsz_v11030 = 410

cudaStreamGetCaptureInfo_v2_v11030 = 409

cudaStreamGetCaptureInfo_v3_ptsz_v12030 = 462

cudaStreamGetCaptureInfo_v3_v12030 = 461

cudaStreamGetDevResource_ptsz_v13010 = 537

cudaStreamGetDevResource_v13010 = 536

cudaStreamGetDevice_ptsz_v12080 = 481

cudaStreamGetDevice_v12080 = 480

cudaStreamGetFlags_ptsz_v7000 = 238

cudaStreamGetFlags_v5050 = 204

cudaStreamGetId_ptsz_v12000 = 442

cudaStreamGetId_v12000 = 441

cudaStreamGetPriority_ptsz_v7000 = 237

cudaStreamGetPriority_v5050 = 203

cudaStreamIsCapturing_ptsz_v10000 = 318

cudaStreamIsCapturing_v10000 = 317

cudaStreamQuery_ptsz_v7000 = 240

cudaStreamQuery_v3020 = 132

cudaStreamSetAttribute_ptsz_v11000 = 346

cudaStreamSetAttribute_v11000 = 345

cudaStreamSetFlags_ptsz_v10200 = 331

cudaStreamSetFlags_v10200 = 330

cudaStreamSynchronize_ptsz_v7000 = 239

cudaStreamSynchronize_v3020 = 131

cudaStreamUpdateCaptureDependencies_ptsz_v11030 = 412

cudaStreamUpdateCaptureDependencies_ptsz_v12030 = 464

cudaStreamUpdateCaptureDependencies_v11030 = 411

cudaStreamUpdateCaptureDependencies_v12030 = 463

cudaStreamUpdateCaptureDependencies_v2_ptsz_v12030 = 464

cudaStreamUpdateCaptureDependencies_v2_v12030 = 463

cudaStreamWaitEvent_ptsz_v7000 = 247

cudaStreamWaitEvent_v3020 = 147

cudaThreadExchangeStreamCaptureMode_v10010 = 327

cudaThreadExit_v3020 = 123

cudaThreadGetCacheConfig_v3020 = 150

cudaThreadGetLimit_v3020 = 127

cudaThreadSetCacheConfig_v3020 = 146

cudaThreadSetLimit_v3020 = 128

cudaThreadSynchronize_v3020 = 126

cudaUnbindTexture_v3020 = 58

cudaUserObjectCreate_v11030 = 413

cudaUserObjectRelease_v11030 = 415

cudaUserObjectRetain_v11030 = 414

cudaVDPAUGetDevice_v3020 = 80

cudaVDPAUSetVDPAUDevice_v3020 = 81

cudaWGLGetDevice_v3020 = 71

cudaWaitExternalSemaphoresAsync_ptsz_v10000 = 282

cudaWaitExternalSemaphoresAsync_ptsz_v11020 = 396

cudaWaitExternalSemaphoresAsync_v10000 = 281

cudaWaitExternalSemaphoresAsync_v11020 = 395

cudaWaitExternalSemaphoresAsync_v2_ptsz_v11020 = 396

cudaWaitExternalSemaphoresAsync_v2_v11020 = 395

cupti.cupti.driver_api_trace_cbid

alias of Driver_api_trace_cbid

cupti.cupti.runtime_api_trace_cbid

alias of Runtime_api_trace_cbid

Classes

class cupti.cupti.CallbackData

Bases: object

Empty-initialize an instance of CUpti_CallbackData.

See also

CUpti_CallbackData

callback_site

Point in the runtime or driver function from where the callback was issued.

Type:

int

context

Driver context current to the thread, or null if no context is current. This value can change from the entry to exit callback of a runtime API function if the runtime initializes a context.

Type:

int

context_uid

Unique ID for the CUDA context associated with the thread. The UIDs are assigned sequentially as contexts are created and are unique within a process.

Type:

int

correlation_data

Pointer to data shared between the entry and exit callbacks of a given runtime or drive API function invocation. This field can be used to pass 64-bit values from the entry callback to the corresponding exit callback.

Type:

int

correlation_id

The activity record correlation ID for this callback. For a driver domain callback (i.e. domain CUPTI_CB_DOMAIN_DRIVER_API) this ID will equal the correlation ID in the CUpti_ActivityAPI record corresponding to the CUDA driver function call. For a runtime domain callback (i.e. domain CUPTI_CB_DOMAIN_RUNTIME_API) this ID will equal the correlation ID in the CUpti_ActivityAPI record corresponding to the CUDA runtime function call. Within the callback, this ID can be recorded to correlate user data with the activity record. This field is new in 4.1.

Type:

int

static from_buffer(buffer)

Create an CallbackData instance with the memory from the given buffer.

static from_data(data)

Create an CallbackData instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype callback_data_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an CallbackData instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

function_name

Name of the runtime or driver API function which issued the callback. This string is a global constant and so may be accessed outside of the callback.

Type:

str

function_params

Pointer to the arguments passed to the runtime or driver API call. See generated_cuda_runtime_api_meta.h and generated_cuda_meta.h for structure definitions for the parameters for each runtime and driver API function.

Type:

int

function_return_value

Pointer to the return value of the runtime or driver API call. This field is only valid within the exit::CUPTI_API_EXIT callback. For a runtime API functionReturnValue points to a cudaError_t. For a driver API functionReturnValue points to a CUresult.

Type:

int

ptr

Get the pointer address to the data as Python int.

symbol_name

Name of the symbol operated on by the runtime or driver API function which issued the callback. This entry is valid only for driver and runtime launch callbacks, where it returns the name of the kernel.

Type:

str

class cupti.cupti.GraphData

Bases: object

Empty-initialize an instance of CUpti_GraphData.

See also

CUpti_GraphData

dependency

The dependent graph node

Type:

int

static from_buffer(buffer)

Create an GraphData instance with the memory from the given buffer.

static from_data(data)

Create an GraphData instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype graph_data_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an GraphData instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

graph

CUDA graph

Type:

int

graph_exec

CUDA executable graph

Type:

int

node

CUDA graph node

Type:

int

node_type

Type of the node

Type:

int

original_graph

The original CUDA graph from which graph is cloned

Type:

int

original_node

The original CUDA graph node from which node is cloned

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ModuleResourceData

Bases: object

Empty-initialize an instance of CUpti_ModuleResourceData.

See also

CUpti_ModuleResourceData

cubin_size

The size of the cubin.

Type:

int

static from_buffer(buffer)

Create an ModuleResourceData instance with the memory from the given buffer.

static from_data(data)

Create an ModuleResourceData instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype module_resource_data_dtype holding the data.

static from_ptr(

ptr: int,

readonly: bool = False,

owner=None,

)

Create an ModuleResourceData instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

module_id

Identifier to associate with the CUDA module.

Type:

int

p_cubin

Pointer to the associated cubin.

Type:

str

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.NvtxData

Bases: object

Empty-initialize an instance of CUpti_NvtxData.

See also

CUpti_NvtxData

static from_buffer(buffer)

Create an NvtxData instance with the memory from the given buffer.

static from_data(data)

Create an NvtxData instance wrapping the given NumPy array.

Parameters:

data (_numpy.ndarray) – a single-element array of dtype nvtx_data_dtype holding the data.

static from_ptr(ptr: int, readonly: bool = False, owner=None)

Create an NvtxData instance wrapping the given pointer.

Parameters:

• ptr (intptr_t) – pointer address as Python int to the data.

• owner (object) – The Python object that owns the pointer. If not provided, data will be copied.

• readonly (bool) – whether the data is read-only (to the user). default is False.

function_name

Name of the NVTX API function which issued the callback. This string is a global constant and so may be accessed outside of the callback.

Type:

str

function_params

Pointer to the arguments passed to the NVTX API call. See generated_nvtx_meta.h for structure definitions for the parameters for each NVTX API function.

Type:

int

function_return_value

Pointer to the return value of the NVTX API call. See nvToolsExt.h for each NVTX API function’s return value.

Type:

int

ptr

Get the pointer address to the data as Python int.

class cupti.cupti.ResourceData

Bases: object

Empty-initialize an instance of CUpti_ResourceData.

See also

CUpti_ResourceData

context

For CUPTI_CBID_RESOURCE_CONTEXT_CREATED and CUPTI_CBID_RESOURCE_CONTEXT_DESTROY_STARTING, the context being created or destroyed. For CUPTI_CBID_RESOURCE_STREAM_CREATED and CUPTI_CBID_RESOURCE_STREAM_DESTROY_STARTING, the context containing the stream being created or destroyed.

Type:

int

static from_buffer(buffer)

Create an ResourceData instance with the memory from the given buffer.

sta