NVIDIACUDA Toolkit Documentation
Search In:
CUDA Toolkit v9.0.176
CUPTI
  • 1. Usage
    • 1.1. CUPTI Compatibility and Requirements
    • 1.2. CUPTI Initialization
    • 1.3. CUPTI Activity API
      • 1.3.1. SASS Source Correlation
      • 1.3.2. PC Sampling
      • 1.3.3. NVLink
      • 1.3.4. OpenACC
      • 1.3.5. External Correlation
    • 1.4. CUPTI Callback API
      • 1.4.1. Driver and Runtime API Callbacks
      • 1.4.2. Resource Callbacks
      • 1.4.3. Synchronization Callbacks
      • 1.4.4. NVIDIA Tools Extension Callbacks
    • 1.5. CUPTI Event API
      • 1.5.1. Collecting Kernel Execution Events
      • 1.5.2. Sampling Events
    • 1.6. CUPTI Metric API
      • 1.6.1. Metrics Reference
        • 1.6.1.1. Metrics for Capability 3.x
        • 1.6.1.2. Metrics for Capability 5.x
        • 1.6.1.3. Metrics for Capability 6.x
        • 1.6.1.4. Metrics for Capability 7.0
    • 1.7. Samples
  • 2. Modules
    • 2.1. CUPTI Version
    • 2.2. CUPTI Result Codes
    • 2.3. CUPTI Activity API
    • 2.4. CUPTI Callback API
    • 2.5. CUPTI Event API
    • 2.6. CUPTI Metric API
  • 3. Data Structures
    • 3.1. CUpti_Activity
    • 3.2. CUpti_ActivityAPI
    • 3.3. CUpti_ActivityAutoBoostState
    • 3.4. CUpti_ActivityBranch
    • 3.5. CUpti_ActivityBranch2
    • 3.6. CUpti_ActivityCdpKernel
    • 3.7. CUpti_ActivityContext
    • 3.8. CUpti_ActivityCudaEvent
    • 3.9. CUpti_ActivityDevice
    • 3.10. CUpti_ActivityDevice2
    • 3.11. CUpti_ActivityDeviceAttribute
    • 3.12. CUpti_ActivityEnvironment
    • 3.13. CUpti_ActivityEvent
    • 3.14. CUpti_ActivityEventInstance
    • 3.15. CUpti_ActivityExternalCorrelation
    • 3.16. CUpti_ActivityFunction
    • 3.17. CUpti_ActivityGlobalAccess
    • 3.18. CUpti_ActivityGlobalAccess2
    • 3.19. CUpti_ActivityGlobalAccess3
    • 3.20. CUpti_ActivityInstantaneousEvent
    • 3.21. CUpti_ActivityInstantaneousEventInstance
    • 3.22. CUpti_ActivityInstantaneousMetric
    • 3.23. CUpti_ActivityInstantaneousMetricInstance
    • 3.24. CUpti_ActivityInstructionCorrelation
    • 3.25. CUpti_ActivityInstructionExecution
    • 3.26. CUpti_ActivityKernel
    • 3.27. CUpti_ActivityKernel2
    • 3.28. CUpti_ActivityKernel3
    • 3.29. CUpti_ActivityKernel4
    • 3.30. CUpti_ActivityMarker
    • 3.31. CUpti_ActivityMarker2
    • 3.32. CUpti_ActivityMarkerData
    • 3.33. CUpti_ActivityMemcpy
    • 3.34. CUpti_ActivityMemcpy2
    • 3.35. CUpti_ActivityMemory
    • 3.36. CUpti_ActivityMemset
    • 3.37. CUpti_ActivityMetric
    • 3.38. CUpti_ActivityMetricInstance
    • 3.39. CUpti_ActivityModule
    • 3.40. CUpti_ActivityName
    • 3.41. CUpti_ActivityNvLink
    • 3.42. CUpti_ActivityNvLink2
    • 3.43. CUpti_ActivityObjectKindId
    • 3.44. CUpti_ActivityOpenAcc
    • 3.45. CUpti_ActivityOpenAccData
    • 3.46. CUpti_ActivityOpenAccLaunch
    • 3.47. CUpti_ActivityOpenAccOther
    • 3.48. CUpti_ActivityOverhead
    • 3.49. CUpti_ActivityPCSampling
    • 3.50. CUpti_ActivityPCSampling2
    • 3.51. CUpti_ActivityPCSampling3
    • 3.52. CUpti_ActivityPCSamplingConfig
    • 3.53. CUpti_ActivityPCSamplingRecordInfo
    • 3.54. CUpti_ActivityPreemption
    • 3.55. CUpti_ActivitySharedAccess
    • 3.56. CUpti_ActivitySourceLocator
    • 3.57. CUpti_ActivityStream
    • 3.58. CUpti_ActivitySynchronization
    • 3.59. CUpti_ActivityUnifiedMemoryCounter
    • 3.60. CUpti_ActivityUnifiedMemoryCounter2
    • 3.61. CUpti_ActivityUnifiedMemoryCounterConfig
    • 3.62. CUpti_CallbackData
    • 3.63. CUpti_EventGroupSet
    • 3.64. CUpti_EventGroupSets
    • 3.65. CUpti_MetricValue
    • 3.66. CUpti_ModuleResourceData
    • 3.67. CUpti_NvtxData
    • 3.68. CUpti_ResourceData
    • 3.69. CUpti_SynchronizeData
  • 4. Data Fields
  • 5. Limitations
  • 6. Changelog
  • Notices

    Search Results

      < Previous | Next >
      CUPTI (PDF) - v9.0.176 (older) - Last updated September 2, 2017 - Send Feedback

      3.41. CUpti_ActivityNvLink Struct Reference

      [CUPTI Activity API]

      This structure gives capabilities of each logical NVLink connection between two devices, gpu<->gpu or gpu<->CPU which can be used to understand the topology. NVLink information are now reported using the CUpti_ActivityNvLink2 activity record.

      Public Variables

      uint64_t  bandwidth
      uint32_t  domainId
      uint32_t  flag
      CUpti_ActivityNvLink::@17  idDev0
      CUpti_ActivityNvLink::@18  idDev1
      uint32_t  index
      CUpti_ActivityKind kind
      uint32_t  nvlinkVersion
      uint32_t  physicalNvLinkCount
      int8_t  portDev0[4]
      int8_t  portDev1[4]
      CUpti_DevType typeDev0
      CUpti_DevType typeDev1

      Variables

      uint64_t CUpti_ActivityNvLink::bandwidth [inherited]

      Banwidth of NVLink in kbytes/sec

      uint32_t CUpti_ActivityNvLink::domainId [inherited]

      Domain ID of NPU. On Linux, this can be queried using lspci.

      uint32_t CUpti_ActivityNvLink::flag [inherited]

      Flag gives capabilities of the link

      See also:

      CUpti_LinkFlag

      CUpti_ActivityNvLink::@17 CUpti_ActivityNvLink::idDev0 [inherited]

      If typeDev0 is CUPTI_DEV_TYPE_GPU, UUID for device 0. CUpti_ActivityDevice2. If typeDev0 is CUPTI_DEV_TYPE_NPU, struct npu for NPU.

      CUpti_ActivityNvLink::@18 CUpti_ActivityNvLink::idDev1 [inherited]

      If typeDev1 is CUPTI_DEV_TYPE_GPU, UUID for device 1. CUpti_ActivityDevice2. If typeDev1 is CUPTI_DEV_TYPE_NPU, struct npu for NPU.

      uint32_t CUpti_ActivityNvLink::index [inherited]

      Index of the NPU. First index will always be zero.

      CUpti_ActivityKindCUpti_ActivityNvLink::kind [inherited]

      The activity record kind, must be CUPTI_ACTIVITY_KIND_NVLINK.

      uint32_t CUpti_ActivityNvLink::nvlinkVersion [inherited]

      NVLink version.

      uint32_t CUpti_ActivityNvLink::physicalNvLinkCount [inherited]

      Number of physical NVLinks present between two devices.

      int8_t CUpti_ActivityNvLink::portDev0[4] [inherited]

      Port numbers for maximum 4 NVLinks connected to device 0. If typeDev0 is CUPTI_DEV_TYPE_NPU, ignore this field. In case of invalid/unknown port number, this field will be set to value CUPTI_NVLINK_INVALID_PORT. This will be used to correlate the metric values to individual physical link and attribute traffic to the logical NVLink in the topology.

      int8_t CUpti_ActivityNvLink::portDev1[4] [inherited]

      Port numbers for maximum 4 NVLinks connected to device 1. If typeDev1 is CUPTI_DEV_TYPE_NPU, ignore this field. In case of invalid/unknown port number, this field will be set to value CUPTI_NVLINK_INVALID_PORT. This will be used to correlate the metric values to individual physical link and attribute traffic to the logical NVLink in the topology.

      CUpti_DevTypeCUpti_ActivityNvLink::typeDev0 [inherited]

      Type of device 0 CUpti_DevType

      CUpti_DevTypeCUpti_ActivityNvLink::typeDev1 [inherited]

      Type of device 1 CUpti_DevType