veyepar: Making code run fast on all the things (with OpenCL)

Author(s):	Beau Johnston
Location	OGGB3
Date	jan Wed 14	Days Raw Files
Start	10:40	First Raw Start	10:30
Duration	0:45:00	Offset	0:09:03
End	11:25	Last Raw End	11:29
Chapters	00:00
Total cuts_time	41 min.

http://lca2015.linux.org.au/schedule/30202/view_talk
raw-playlist raw-mp4-playlist encoded-files-playlist host public tweet mp4 svg png

assets release.pdf Making_code_run_fast_on_all_the_things_with_OpenCL.json
logs
Admin: episode episode list cut list raw files day marks day marks day image_files

State:

Locked:

clear this to unlock

Locked by:

user/process that locked.

Start:

initially scheduled time from master, adjusted to match reality

Duration:

length in hh:mm:ss

Name:

Video Title (shows in video search results)

Emails:

email(s) of the presenter(s)

Released:

has someone authorised pubication

Normalise:

Channelcopy:

m=mono, 01=copy left to right, 10=right to left, 00=ignore.

Thumbnail:

filename.png

Description:

markdown

Modern computer systems are becoming more heterogenous. This can be seen with
the increased use of accelerator devices and co-processors for both industrial
and personal use. Devices such as Graphical Processing Units (GPUs) offer
improved performance (both in taking a shorter duration to compute and
energy efficiency) for a range of applications. Other tasks are better suited
to the Digital Signal Processor (DSP), Field Programmable Gate Array (FPGA),
another Central Processor Unit (CPU) or any alternative co-processor. All of
these devices are significantly different in architecture and programming
language support. Developing an application for all of these devices results in
much redundancy, fragmentation and frustration by requiring the developer to
write potentially several alternative implementations. This fragmentation can
be really bad for programmers and the Open Source community in general as each
vendor provides their own API's, proprietary tools and restrict our freedoms by
locking us into a third party ecosystem.

Thankfully a language and Open Source standard exists to save us from vendor
hell, the answer is Open Computing Language (OpenCL)!

During this talk I will discuss how to leverage this hardware agnostic language
(from a programmers perspective) and recap on some tips on how to write flexible
kernels once and have them run efficiently on all platforms. Additionally I'll focus on proper utilisation of time... our time (as developers) and the clients. Too often do we put off learning a new language or standard because of the initial time we need to sink into it before we have working code. I'll do this work for you, namely by comparing a few different language implementations of OpenCL with a major emphasis being placed on development time and comparing it against execution time (and what cost does it have your sanity). You'll also be given some running OpenCL boilerplate code in each of these languages, so you can hit the ground running!

Please attend my talk, as I need your help in increasing it's adoption which
will in turn increase vendor support for the lastest, greatest and more
flexible versions of the OpenCL standard.

Comment:

production notes