Topics

Topics#

Rocprofiler SDK Developer API: Topics

Rocprofiler SDK Developer API 0.5.0

ROCm Profiling API and tools

Here is a list of all topics with brief descriptions:

[detail level 12]

Agent Information	Needs brief description
Thread Trace Service	Provides API calls to enable and handle thread trace data
Buffer Handling	Creation, destruction, and flushing of buffers populated with data from rocprofiler
Asynchronous Tracing Service	Receive callbacks for batches of records from an internal (background) thread
Synchronous Tracing Services	Receive immediate callbacks on the calling thread
Context Handling	Associate services with a handle. This handle is used to activate/deactivate the services during the application runtime
Hardware counters Information	Query functions related to hardware counters
Symbol Versions	The names used for the shared library versioned symbols
Agent Profile Counting Service	Needs brief description
Dispatch Profile Counting Service	Per-dispatch hardware counter collection service
External Correlation IDs	User-defined correlation identifiers to supplement rocprofiler generated correlation ids
Basic data types	Basic data types and typedefs
Intercept table for runtime libraries	Enable tools to wrap the runtime API function calls of HIP, HSA, and ROCTx before and after the "real" implementation is called
Internal Thread Handling	Callbacks before and after threads created internally by libraries
PC Sampling	Enabling PC (Program Counter) Sampling for GPU Activity
Profile Configurations	Group one or more hardware counters into a unique handle
Result Codes	The various result codes that RCCL API calls may return
▼Communicator Configuration	Structure that allows for customizing Communicator behavior via ncclCommInitRankConfig
configuration	Users can assign value to attributes to specify the behavior of a communicator
Version Information	API call that returns RCCL version
Communicator Initialization/Destruction	API calls that operate on communicators. Communicators objects are used to launch collective communication operations. Unique ranks between 0 and N-1 must be assigned to each HIP device participating in the same Communicator. Using the same HIP device for multiple ranks of the same Communicator is not supported at this time
Error Checking Calls	API calls that check for errors
Communicator Information	API calls that query communicator information
API Enumerations	Enumerations used by collective communication calls
Custom Reduction Operator	API calls relating to creation/destroying custom reduction operator that pre-multiplies local source arrays prior to reduction
Collective Communication Operations	Collective communication operations must be called separately for each communicator in a communicator clique
MSCCL Algorithm	API calls relating to the optional MSCCL algorithm datapath
Group semantics	When managing multiple GPUs from a single thread, and since RCCL collective calls may perform inter-CPU synchronization, we need to "group" calls for different ranks/devices into a single call
Tool registration	Data types and functions for tool registration with rocprofiler
Library Versioning	Version information about the interface and the associated installed library
Miscellaneous Utility Functions	Utility functions for library
SPM Service	Streaming Performance Monitoring