On 08/08/2011 10:04 PM, Vojtěch Král wrote:
Hi,
today I just randomly checked clinfo and it says:
FATAL: Module nvidia not found.
I have opencl-nvidia installed. This happens regardless to which libcl
I install.
I tried putting full path in /etc/OpenCL/vendors/nvidia.icd but it's
still the sam
Hi,
today I just randomly checked clinfo and it says:
FATAL: Module nvidia not found.
I have opencl-nvidia installed. This happens regardless to which libcl
I install.
I tried putting full path in /etc/OpenCL/vendors/nvidia.icd but it's
still the same...
Is it just me or is there something wrong
Thanks Ionut! So far it looks ok, I think there's no bloat.
I'll do some testing hopefully tonight.
And with Nicolas' code too...
Thanks for helping out, both of you...
~kralyk
2011/7/17 Ionut Biru :
>>
>> i've been waiting a new nvidia version and now is time to push something
>> into extra.
>
On 07/15/2011 05:25 PM, Ionut Biru wrote:
On 07/08/2011 09:49 PM, Vojtěch Král wrote:
So, what is going on? Don't you guys quit on me...
@Nicolas:
You pointed out that the nVidia ICD loader (libOpenCL.so) is only
1.0 (outdated),
that is correct, but the ICD specs only require 1.0, so 1.1 possib
On 07/08/2011 09:49 PM, Vojtěch Král wrote:
So, what is going on? Don't you guys quit on me...
@Nicolas:
You pointed out that the nVidia ICD loader (libOpenCL.so) is only
1.0 (outdated),
that is correct, but the ICD specs only require 1.0, so 1.1 possibly
should not be needed.
See specs: http
@Nicolas
I've put it up on AUR: https://aur.archlinux.org/packages.php?ID=50587
(named 'libopencl' cause 'libcl' is not allowed on aur)
Both Intel and AMD sdk now depend on it, so you can install both of
them at the same time.
The problem with this libopencl pkg is the god damned license -
-
>
> I made a simple PKGBUILD just to demonstrate:
> http://codepad.org/iK2oBo0W
>
Looks fine to me! I'll try it later on.
I think that the "libcl" pkg might very well be based off of amd's loader.
I made a simple PKGBUILD just to demonstrate:
http://codepad.org/iK2oBo0W
It contains the shared object and OpenCL info utility, it is up to date,
has no dependencies and is only 88kB in size ;-)
~kralyk
2011/7/8 Nicolas Bigaouette :
> I think it requires 1.0 to provide, well, 1.0... Version 1.1 adds many
> things. For example, there is no type3 (float3, int3, etc.) in 1.0, just in
> 1.1. So if your code uses float3, then you'll probably get some undefined
> symbols when linking to a 1.0 libOpenCL.
>
> @Nicolas:
> You pointed out that the nVidia ICD loader (libOpenCL.so) is only
> 1.0 (outdated),
> that is correct, but the ICD specs only require 1.0, so 1.1 possibly
> should not be needed.
> See specs:
> http://www.khronos.org/registry/cl/extensions/khr/cl_khr_icd.txt
>
I think it requires 1
So, what is going on? Don't you guys quit on me...
@Nicolas:
You pointed out that the nVidia ICD loader (libOpenCL.so) is only
1.0 (outdated),
that is correct, but the ICD specs only require 1.0, so 1.1 possibly
should not be needed.
See specs: http://www.khronos.org/registry/cl/extensions/khr/c
@Ionut, Lukas:
I support this proposal too and I'd agree that only the libOpenCL.so
(the ICD loader)
needs to be split off nvidia-utils. Otherwise it could just stay
intact as far as I can tell...
Also, it would probably be wise to settle for dependency names once for all,
by dependency names I
On 7 July 2011 18:34, Ionut Biru wrote:
> On 07/07/2011 05:46 PM, Nicolas Bigaouette wrote:
>
>
>
>
> OKKK
> after talking with Lukas i now have an idea about what you guys talking
> about.
>
> Right now, what can i do is to take care about nvidia's opencl
> implementation and loader.
>
> Wha
On 07/07/2011 05:46 PM, Nicolas Bigaouette wrote:
There might be just one right now, but OpenCL standard says many different
implementation should be able to install side by side, and the choice
between them is taken by the ICD loader.
OKKK
after talking with Lukas i now have an idea a
>
> Although that libclicd thingy looks outdated and/or incomplete,
> I like the testapps/list_platforms.c utility. It would be nice to have
> something
> like that in (a possible) wrapper ICD loader package (the
> /usr/lib/libOpenCL.so).
>
> AMD has a similar (more verbose) utility called 'clinfo'
2011/7/7 Nicolas Bigaouette :
> I've found somebody who tried to implement his own ICD loader:
> https://github.com/Max-E/libclicd
> It was not updated for many months, and may be missing a lot of feature.
>
> It might be easier to create a package that just provides
> /usr/lib/libOpenCL.so taken f
Thankyou Nicolas for the research, that's exactly what we need.
And yes Damjan, you are right.
I've found somebody who tried to implement his own ICD loader:
https://github.com/Max-E/libclicd
It was not updated for many months, and may be missing a lot of feature.
It might be easier to create a package that just provides
/usr/lib/libOpenCL.so taken from one of the different implementation.
> There is no conflict.
>
>> Therefore, a sollution might be to simply pick one of available ICD
>> loaders (a thin one preferably) and make it a dependency for other
>> pkgs.
>
> We only have one in our repos so there's nothing to do as far as I can tell.
>
> In case I misunderstood something,
> I'm sending this to maintainers who maintain involved pkgs.
> People that might be also interested are in Cc, please let me know
> should you find this mail too spammy/annoying :)
Next time please send this to the appropriate mailing list (arch-general
or aur-general in most cases) and if you fe
Forwarding discussion to arch-general mailing list.
Original Message
Subject: The OpenCL ICD problem
Date: Thu, 7 Jul 2011 15:48:52 +0200
From: Vojtěch Král
Hello,
I'm writing to you because of a problem which has arisen with the
OpenCL Arch packages.
To introduce myself sho
21 matches
Mail list logo
