CUpti_ActivityKernel4 Struct Reference
[CUPTI Activity API]

The activity record for a kernel (CUDA 9.0(with sm_70 support) onwards). (deprecated in CUDA 11.0). More...


Data Fields

int32_t blockX
int32_t blockY
int32_t blockZ
union {
cacheConfig
uint64_t completed
uint32_t contextId
uint32_t correlationId
uint32_t deviceId
int32_t dynamicSharedMemory
uint64_t end
int64_t gridId
int32_t gridX
int32_t gridY
int32_t gridZ
uint8_t isSharedMemoryCarveoutRequested
CUpti_ActivityKind kind
uint8_t launchType
uint32_t localMemoryPerThread
uint32_t localMemoryTotal
const char * name
uint8_t padding
CUpti_ActivityPartitionedGlobalCacheConfig partitionedGlobalCacheExecuted
CUpti_ActivityPartitionedGlobalCacheConfig partitionedGlobalCacheRequested
uint64_t queued
uint16_t registersPerThread
void * reserved0
uint8_t sharedMemoryCarveoutRequested
uint8_t sharedMemoryConfig
uint32_t sharedMemoryExecuted
uint64_t start
int32_t staticSharedMemory
uint32_t streamId
uint64_t submitted
uint8_t executed:4
uint8_t requested:4


Detailed Description

This activity record represents a kernel execution (CUPTI_ACTIVITY_KIND_KERNEL and CUPTI_ACTIVITY_KIND_CONCURRENT_KERNEL). Kernel activities are now reported using the CUpti_ActivityKernel6 activity record.

Field Documentation

The X-dimension block size for the kernel.

The Y-dimension block size for the kernel.

The Z-dimension grid size for the kernel.

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

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.

The ID of the context where the kernel is executing.

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.

The ID of the device where the kernel is executing.

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

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.

The cache configuration used for the kernel. The value is one of the CUfunc_cache enumeration values from cuda.h.

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

The X-dimension grid size for the kernel.

The Y-dimension grid size for the kernel.

The Z-dimension grid size for the kernel.

This indicates if CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT was updated for the kernel launch

The activity record kind, must be CUPTI_ACTIVITY_KIND_KERNEL or CUPTI_ACTIVITY_KIND_CONCURRENT_KERNEL.

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

See also:
CUpti_ActivityLaunchType

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

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

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

Undefined. Reserved for internal use.

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.

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.

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 asynchrnous 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.

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

The cache configuration requested by the kernel. The value is one of the CUfunc_cache enumeration values from cuda.h.

Undefined. Reserved for internal use.

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.

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

Shared memory size set by the driver.

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.

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

The ID of the stream where the kernel is executing.

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.


Generated on Sun Feb 14 21:14:24 2021 for Cupti by  doxygen 1.5.8