MPI: The Complete Reference

Next: Contents

• Contents
• Introduction
• The Goals of MPI
• Who Should Use This Standard?
• What Platforms are Targets for Implementation?
• What is Included in MPI?
• What is Not Included in MPI?
• Version of MPI
• MPI Conventions and Design Choices
• Document Notation
• Procedure Specification
• Semantic Terms
• Processes
• Types of MPI Calls
• Opaque Objects
• Named Constants
• Choice Arguments
• Language Binding
• Fortran 77 Binding Issues
• C Binding Issues
• Point-to-Point Communication
• Introduction and Overview
• Blocking Send and Receive Operations
• Blocking Send
• Send Buffer and Message Data
• Message Envelope
• Comments on Send
• Blocking Receive
• Receive Buffer
• Message Selection
• Return Status
• Comments on Receive
• Datatype Matching and Data Conversion
• Type Matching Rules
• Type MPI_CHARACTER
• Data Conversion
• Comments on Data Conversion
• Semantics of Blocking Point-to-point
• Buffering and Safety
• Multithreading
• Order
• Progress
• Fairness
• Example - Jacobi iteration
• Send-Receive
• Null Processes
• Nonblocking Communication
• Request Objects
• Posting Operations
• Completion Operations
• Examples
• Freeing Requests
• Semantics of Nonblocking Communications
• Order
• Progress
• Fairness
• Buffering and resource limitations
• Comments on Semantics of Nonblocking Communications
• Multiple Completions
• Probe and Cancel
• Persistent Communication Requests
• Communication-Complete Calls with Null Request Handles
• Communication Modes
• Blocking Calls
• Nonblocking Calls
• Persistent Requests
• Buffer Allocation and Usage
• Model Implementation of Buffered Mode
• Comments on Communication Modes
• User-Defined Datatypes and Packing
• Introduction
• Introduction to User-Defined Datatypes
• Datatype Constructors
• Contiguous
• Vector
• Hvector
• Indexed
• Hindexed
• Struct
• Use of Derived Datatypes
• Commit
• Deallocation
• Relation to count
• Type Matching
• Message Length
• Address Function
• Lower-bound and Upper-bound Markers
• Absolute Addresses
• Pack and Unpack
• Derived Datatypes vs Pack/Unpack
• Collective Communications
• Introduction and Overview
• Operational Details
• Communicator Argument
• Barrier Synchronization
• Broadcast
• Example Using MPI_BCAST
• Gather
• Examples Using MPI_GATHER
• Gather, Vector Variant
• Examples Using MPI_GATHERV
• Scatter
• An Example Using MPI_SCATTER
• Scatter: Vector Variant
• Examples Using MPI_SCATTERV
• Gather to All
• An Example Using MPI_ALLGATHER
• Gather to All: Vector Variant
• All to All Scatter/Gather
• All to All: Vector Variant
• Global Reduction Operations
• Reduce
• Predefined Reduce Operations
• MINLOC and MAXLOC
• All Reduce
• Reduce-Scatter
• Scan
• User-Defined Operations for Reduce and Scan
• The Semantics of Collective Communications
• Communicators
• Introduction
• Division of Processes
• Avoiding Message Conflicts Between Modules
• Extensibility by Users
• Safety
• Overview
• Groups
• Communicator
• Communication Domains
• Compatibility with Current Practice
• Group Management
• Group Accessors
• Group Constructors
• Group Destructors
• Communicator Management
• Communicator Accessors
• Communicator Constructors
• Communicator Destructor
• Safe Parallel Libraries
• Caching
• Introduction
• Caching Functions
• Intercommunication
• Introduction
• Intercommunicator Accessors
• Intercommunicator Constructors
• Process Topologies
• Introduction
• Virtual Topologies
• Overlapping Topologies
• Embedding in MPI
• Cartesian Topology Functions
• Cartesian Constructor Function
• Cartesian Convenience Function: MPI_DIMS_CREATE
• Cartesian Inquiry Functions
• Cartesian Translator Functions
• Cartesian Shift Function
• Cartesian Partition Function
• Cartesian Low-level Functions
• Graph Topology Functions
• Graph Constructor Function
• Graph Inquiry Functions
• Graph Information Functions
• Low-level Graph Functions
• Topology Inquiry Functions
• An Application Example
• Environmental Management
• Implementation Information
• Environmental Inquiries
• Tag Values
• Host Rank
• I/O Rank
• Clock Synchronization
• Timers and Synchronization
• Initialization and Exit
• Error Handling
• Error Handlers
• Error Codes
• Interaction with Executing Environment
• Independence of Basic Runtime Routines
• Interaction with Signals in POSIX
• The MPI Profiling Interface
• Requirements
• Discussion
• Logic of the Design
• Miscellaneous Control of Profiling
• Examples
• Profiler Implementation
• MPI Library Implementation
• Systems With Weak symbols
• Systems without Weak Symbols
• Complications
• Multiple Counting
• Linker Oddities
• Multiple Levels of Interception
• Conclusions
• Design Issues
• Why is MPI so big?
• Should we be concerned about the size of MPI?
• Why does MPI not guarantee buffering?
• Portable Programming with MPI
• Dependency on Buffering
• Collective Communication and Synchronization
• Ambiguous Communications and Portability
• Heterogeneous Computing with MPI
• MPI Implementations
• Extensions to MPI
• References
• About this document ...




---

Jack Dongarra
Fri Sep 1 06:16:55 EDT 1995