Skip to content
No description, website, or topics provided.
C++ Shell Makefile Other
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
ResStudio/Prototype
.gitattributes
.gitignore
README.md
aa.txt

README.md

DuctTeip

DuctTeip is a framework written in C++ for task based parallel programming in distributed environments. It uses MPI for communication and inter node parallelism, and SuperGlue framework for shared memory multi-core parallelism. SuperGlue is another C++ task based parallel programming framework that can be found in SuperGlue git repository.

Installation and Build

  • Clone both the DuctTeip and SuperGlue repository to local paths.
  • Edit the Makefile in the Prototype folder such that the SUPERGLUE_DIR points to the local path where the SuperGlue framework is cloned.
  • Ensure that MPI environment settings are available. (Load the module, if you compile in cluster environments)
  • Run make where the Makkefile exists. Two debug and release versions of DuctTeip library are produced at ./lib and ./lib/release directories, corrspondingly.

Build the examples

To build and run the examples that exist in the sub-directories of the cloned repository:

  • Edit the Makefile such that the DUCTTEIP_DIR and SUPERGLUE_DIR point to the places where these frameworks are copied.
  • Run make to get the compiled output

Runtime command line flags used by DuctTeip

Flags (prepended with a '-') Meaning
P Total number of processors
p No. of processor in rows, in grid topology
q No. of processor in columns, in grid topology
M No. of elements of input matrix in row
N No. of elements of input matrix in columns
B No. of blocks at DuctTeip level
b No. of blocks at SuperGlue level
t number of threads at each computing node
DLB Use Dynamic Load Balancing for task execution at DuctTeip level

In progress and Future Works

  • Improving Dynamic Load Balancing
  • Including the Unified Task Programming
  • Implementing 3D Fast Multipole Method in distributed environment
You can’t perform that action at this time.