Parallel Computing With OpenMP

Host Site:

Texas Advanced Computing Center

Host site URL:

https://www.tacc.utexas.edu/

The course will cover features that are becoming more important to HPC as shared-memory microarchitectures are becoming more complex, and the need for directing the parallel execution of even more complex coding paradigms is becoming paramount to achievement of efficient programming and optimal performance.

Three major topics will be explored to address these challenges.

1.)OpenMP Tasking basics:
Syntax and operation, synchronization, data-sharing model, dependencies and prioritization. Examples and lab will show how to generate tasks and control tasking for while loops, task loops, pointer chasing and tree searchers as well as show how to set up task dependences among tasks.

2.) OpenMP SIMD basics:
SIMD syntax and combination with parallel loop directives.
SIMD directives are used to advise compilers about execution of loops within the SIMD execution units (vectorization). For instance, this includes use of linear variables, alignment of arrays and forced SIMD execution. Moreover, the declare SIMD directive provides mechanism for creating vectorized functions. Examples and lab will illustrate vectorization of loops and functions for optimally executing code in SIMD logical units.

3.) OpenMP Thread Affinity basics:
Declaration of places and bind operations. OpenMP now provides a general, non-vendor-specific syntax and runtime control to place threads of a work-sharing loop (and nested loops). The control can be directed to abstract places (cores, sockets, etc.) or numbered places (cpu ids). Also, binding policies describe mapping distributions. Examples and exercises will illustrate situations where affinity may be needed for testing and/or obtaining optimal thread binding.

After presentation topics labs will provide experience for using syntax and concepts, in simple-to-understand examples; and experiments on codes will test understanding.Late afternoons will be used for lab reviews and providing assistance with users who are interested in implementing new OpenMP features in their code. (The latter is limited to availability of the three instructors.) Additional OpenMP features (including proposed 5.0 features) may be discussed in the late afternoon if time permits.

This training will be webcast to remote participants. Local participants are strongly encouraged to attend in person.

More information: https://www.tacc.utexas.edu/

Sessions:

In person (Texas Advanced Computing Center)

03/31/2016 09:00 - 04/01/2016 17:00 CDT (SESSION HAS ENDED)
View Session Details →
Registration CLOSED
Registration open date
03/10/2016 14:52 CST
Registration close date
03/29/2016 12:00 CDT
Class size restriction
30 registrants

(19 spots left)

Waitlist

0 registrants

Contact Information
Contact
Jason Allison
Contact phone
512-475-9238
Contact email
jasona@tacc.utexas.edu
Location
Name
Texas Advanced Computing Center
Address
J.J. Pickle Research Campus
10100 Burnet Rd.
Austin, TX 78758
Phone
512-475-9238

Webcast

03/31/2016 09:00 - 04/01/2016 17:00 CDT (SESSION HAS ENDED)
View Session Details →
Registration CLOSED
Registration open date
03/10/2016 14:57 CST
Registration close date
03/29/2016 12:00 CDT
Class size restriction
30 registrants

(0 spots left)

Waitlist

0 registrants

Contact Information
Contact
Jason Allison
Contact phone
512-475-9238
Contact email
jasona@tacc.utexas.edu
Location
Name
Texas Advanced Computing Center
Phone
512-475-9238
URL
https://www.tacc.utexas.edu
Posted: 03/10/2016 20:50 UTC