Content

*) HOW TO INSTALL OPENCL (AMD Radeon HD + Linux)

*) IN THIS DIRECTORY AND ON THE WEB

*) OTHER MATERIAL

===========================================================================

HOW TO INSTALL OPENCL (AMD Radeon HD + Linux)

You may ask for an account on my own server, where OpenCL is already installed.

Otherwise, installing the OpenCL library and includes is a matter of minutes.

I've tried with the one that comes along with AMD APP (formerly ATI Stream):
http://developer.amd.com/sdks/AMDAPPSDK/Pages/default.aspx
Follow the "downloads" link on the leftmost vertical column
(ignore all the links in the upper horizontal bar).

If you have an OpenCL-capable AMD/ATI videocard chipset (I've a humble Radeon
HD 5450), you might also want to install the AMD proprietary driver; follow
the same "downloads" link as above.  This way, the card can be detected as an
OpenCL device of type GPU.
Otherwise I think the host CPU will be the only detected OpenCL device,
of type CPU of course (not GPU).

First, install the graphic proprietary driver.  Then, configure X to work with
that driver (no "framebuffer").  Last, install the remainder of the SDK
including OpenCL.

Use the utility "clinfo" for detecting all the OpenCL-capable devices in your
platform.  You should be detected a device of type CPU, at least, plus a
a device of type GPU.

THE X SERVER MUST BE ACTIVE:  if you want your videocard to be detected as
an OpenCL device, it is --mandatory-- that the X server is active and switched
to the video output (so the screen looks in "graphic mode").  If X is off,
the GPU is not detected.  If X is on, but you switch to a testual VT console,
the "clinfo" (as well as any other OpenCL program) will hang on GPU detection.

USING THE GPU FROM REMOTE is possible, but again, the X server of the GPU host
must be active and switched to video output.  Look out for any firewall
settings on the GPU host, that might prevent X connections from establishing
in localhost.

If the user who needs to use the GPU is different from the user who owns the
running X process, then the owner of X has to invoke the "xhost" command for
granting access privileges, and the user has to invoke
"DISPLAY=:0.0 && export DISPLAY" for getting connected to the X process.
This is not necessary if the remote user is the same as the owner of the
running X process.

The SDK provides C bindings (include CL/cl.h),
and C++ bindings (include CL/cl.hpp).
An experimental Java binding can be found at http://www.jocl.org/

===========================================================================

IN THIS DIRECTORY AND ON THE WEB

OpenCL_tutorial.pdf
	introductory slides, some minor mistakes, good as the very first
	doc to read

OpenCL_intro.pdf
	more detailed introductory slides, interesting considerations about
	performance on different hardware platforms

Introduction_to_OpenCL_Programming_Training_Guide.pdf
Introduction_to_OpenCL_Programming_Exercises/
	a good tutorial (except the "installing" section, that is outdated)

OpenCL-1-1-quick-reference-card.pdf
	the name is self-explaining

http://www.it.uu.se/katalog/davbl791/presentations/Introduction%20to%20OpenCL.pdf
	a more extensive step-by-step tutorial

http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/
	online man pages (OpenCL 1.1)

http://www.khronos.org/opencl/
	a comprehensive site

Aaftab Munshi, Benedict R. Gaster, Timothy G. Mattson, James Fung, Dan Ginsburg
"OpenCL Programming Guide", Addison-Wesley, 2012
	a book ;-) ask around for a copy...

===========================================================================

OTHER MATERIAL

other/AMD_OpenCL_Tutorial_SAAHPC2010.pdf
	for those who like the pics; complete, but not the right choice
	for a novice IMHO; worth a look

other/OpenCL_Best_Practices_Guide.pdf
	performance evaluation and optimization hints; not for novices,
	yet quite easy to follow

other/OpenCL_Programming_Guide.pdf
	more specific for the CUDA implementation of OpenCL, but the
	introduction is worth a look; matrix multiplication example;
	some guidelines for program optimization; low-level details
	related to nVidia gpus

http://opencl.codeplex.com/wikipage?title=OpenCL Tutorials - 1
http://www.codeproject.com/KB/showcase/Portable-Parallelism.aspx
http://www.codeproject.com/KB/showcase/Memory-Spaces.aspx
http://www.codeproject.com/KB/showcase/Work-Groups-Sync.aspx