Package Base Details: amdapp-sdk

Git Clone URL: (read-only, click to copy)
Submitter: Vi0L0
Maintainer: rigred
Last Packager: satcom886
Votes: 83
Popularity: 0.000001
First Submitted: 2013-02-11 21:03 (UTC)
Last Updated: 2019-08-16 21:33 (UTC)

Latest Comments

mirh commented on 2020-08-19 11:52 (UTC)

Download still works just fine from the original link.

satcom886 commented on 2020-03-23 16:11 (UTC)

@Martine Not sure what happened. It seems to work for me.

Martine commented on 2020-03-21 21:49 (UTC)

I get an error installing this package: var/tmp/pamac-build-martine/amdapp-sdk/PKGBUILD: line 88: cd: /var/tmp/pamac-build-martine/amdapp-sdk/src/AMD-APP-SDKInstaller-v3.0.130.136-GA-linux64: No such file or directory I've checked the folder contents of /var/tmp/pamac-build-martine/amdapp-sdk/src/AMD-APP-SDKInstaller-v3.0.130.136-GA-linux64 amdapp-sdk.conf amd_i686.icd AMD-APP-SDKInstaller-v3.0.130.136-GA-linux64.tar.bz2 amd.icd

Where does it go wrong?

satcom886 commented on 2019-08-16 22:23 (UTC)

I fixed the download URLs and the provides section and disowned the package since I don't have the time nor the knowledge required to maintain a package.

satcom886 commented on 2019-04-27 20:31 (UTC)

As @jamespharvey20 mentioned AMD no longer hosts AMD APP SDK install files. Can you please change the download URL to There are both Linux32/64 and Windows32/64 versions available.

jamespharvey20 commented on 2019-03-29 23:27 (UTC)

Upstream's download appears to be gone without much comment. It's still downloadable from <>.

enihcam commented on 2018-12-12 23:27 (UTC)

please add "opencl-driver" in provides. Thanks.

redshoe commented on 2018-02-02 16:56 (UTC)


Nevermind, I got it working. Thank you for the instruction.

redshoe commented on 2018-02-02 15:42 (UTC)


I have a question on step 2). Do you mean 'place it in /makepkg/src' by literally at /makepkg/src not in sub-working directory?

Nightbane112 commented on 2018-02-02 10:36 (UTC)

For anyone who still needs to use this package:

1) You need to edit the source URL from "" to "".

2) Download the SDK direclty from the website and place it in /makepkg/src directory. Run makepkg manually to avoid the checksum error due to incomplete download (The download gets interrupted midway when using an AUR helper).

FYI, opencl-amd ( runs OpenCL on GPU whereas this package allow running OpenCL on the CPU.

TheGuardianWolf commented on 2018-01-23 08:28 (UTC)

Anyone looking at this package should be considering this instead

ArnaudNux commented on 2017-12-25 01:52 (UTC) (edited on 2017-12-25 01:52 (UTC) by ArnaudNux)

==> Starting prepare()... ==> WARNING: Installing this package means that you have red, understood and ACCEPTED ==> WARNING: the license agreement posted at ==> WARNING: ==> WARNING: will sleep 5 sec to make sure you red it :P ==> WARNING: Download will take not much time and echo some random letters [nonce1] Loaded CA certificate '/etc/ssl/certs/ca-certificates.crt' ERROR: The certificate of '' is not trusted. ERROR: The certificate of '' has expired. The certificate has expired ==> ERROR: A failure occurred in prepare(). Aborting...

hotaru commented on 2017-12-21 18:36 (UTC)

after changing the URL as suggested by gabi.hulea:

==> Validating AMD-APP-SDKInstaller-v3.0.130.136-GA-linux64.tar.bz2 files with sha256sum ... AMD-APP-SDKInstaller-v3.0.130.136-GA-linux64.tar.bz2: FAILED sha256sum: WARNING: 1 computed checksum did NOT match

gabi.hulea commented on 2017-08-28 06:33 (UTC)

@pr0dukter: edit the package and replace "" with ""

pr0dukter commented on 2017-08-23 14:35 (UTC)

[nonce1] 2017-08-23 09:26:58 ERROR 404: Not Found.

BuddyLuvve commented on 2017-04-14 02:47 (UTC)

It's a bit late, but I just had the dangling symlink problem too... It's solved by installing the 'ocl-icd' package (I believe there are other package options for this as well as per I see that the pkgbuild lists 'opencl-icd-loader' as a dependancy, but for some reason it wasn't pulled in for installation. Any idea why, Vi0L0? I'm at a loss.

chris64 commented on 2016-08-01 10:53 (UTC) (edited on 2016-08-01 11:44 (UTC) by chris64)

I've the same issue like @unixbrain now. Did anyone else had the problem and found a solution to this? Edit: The dangling symlink problem can be solved by inserting `rm -f ${pkgdir}/${_ipath}/lib/` in line 202 before `chmod 755 ${pkgdir}/${_ipath}/lib/*`. See here:

Vi0L0 commented on 2016-07-21 19:29 (UTC) (edited on 2016-07-21 19:33 (UTC) by Vi0L0)

Checked again and I have no problems with the download. You can use my google drive's upload: Download amdapp-sdk sources snapshot and untar it, then create 'src' directory next to PKGBUILD and paste downloaded AMD-APP-SDKInstaller-v3.0.130.136-GA-linux64.tar.bz2 file into that dir. Then you can use makepkg to build package.

pbateman commented on 2016-07-19 10:02 (UTC)

The download keeps failing for me - and it has been like this continually for months... Saving to: ‘AMD-APP-SDKInstaller-v3.0.130.136-GA-linux64.tar.bz2’ AMD-APP-SDKInstaller-v3.0.130 12%[=====> ] 23.98M 1.04MB/s in 25s 2016-07-19 10:57:38 (994 KB/s) - Connection closed at byte 25149440. Retrying. It always gets to about 12%, then resets. It ultimately tries 3 times then dies. Presumably there is some cvnty stuff AMD has put in somewhere to detect something, but as I haven't seen others complain either no one is using us, or it isn't affecting them. Any ideas on this? It certainly isn't my connection, etc. as nothing else exhibits this behaviour, and as stated it has been like this for 100+ attempts over a period of months. Cheers

Vi0L0 commented on 2016-06-02 20:14 (UTC)

sorry for the delay updated, should be fine, funny thing is that lib directory structure for i686 and x86_64 is not same (thats why path in icd differs)

unixbrain commented on 2016-04-07 05:29 (UTC)

==> Starting package_amdapp-sdk-nocatalyst()... chmod: cannot operate on dangling symlink '/home/alex/amdapp-sdk/pkg/amdapp-sdk-nocatalyst/opt/AMDAPP/SDK/lib/' ==> ERROR: A failure occurred in package_amdapp-sdk-nocatalyst(). Aborting... help? I followed chris64's suggestions.

chris64 commented on 2016-03-29 17:25 (UTC) (edited on 2016-04-01 11:48 (UTC) by chris64)

There's a new release AMD-APP-SDKInstaller-v3.0.130.136-GA from 3/23/2016 with sha256sum 181fb9815e735c90ca5713acc27a6f9ed7f85135d2f2a085bed7b4c7ed157b94 (32 bit) and 0aa436acd334b686820bd3caab9f09014608741b92e3996d3642d0b148ede0f7 (64 bit). This explains the download failure. Thank you Vi0L0 for your efforts. :-) Amendment: I had to change `amd.icd` from /PATH/lib/ to /PATH/lib/sdk/ in order to make it work again.

chris64 commented on 2016-03-29 16:59 (UTC)

@Vi0L0 I'm afraid the automated download is broken. The script referenced by @wallnuss works but it's downloaded tarball doesn't match the checksum.

Vi0L0 commented on 2016-01-20 13:31 (UTC)

@wallnuss: I used suggested code in Then I started working on this package and recall that kralyk once was working on such an idea, so I took his code and used it here :). Thank you guys. I was thinking long time about it, I'm still not sure is it clear and can we really use such a download scritps. But oh well, its already popular anyway :P

vchuravy commented on 2016-01-19 03:04 (UTC)

For automated downloads you can use this script:

pbateman commented on 2016-01-09 15:49 (UTC)

Downloads are broken as AMD now requires accepting a license agreement.

HenryJia commented on 2016-01-07 19:49 (UTC) (edited on 2016-01-07 19:49 (UTC) by HenryJia)

The link for the source no longer works.

Vi0L0 commented on 2015-11-21 13:18 (UTC) (edited on 2015-11-21 13:20 (UTC) by Vi0L0)

updated to v3.0.130.135 It should be available in [catalyst] repo in minutes. Thanks ArnaudNux And sorry for the lag

ArnaudNux commented on 2015-11-01 17:09 (UTC)

Hello, The version of the file to be changed is not AMD-SDK.v3-0-124-132 but v3.0-130-135.

tuxayo commented on 2015-10-04 10:52 (UTC)

@ordoe sorry for having missed that, thanks!

5chdn commented on 2015-09-21 12:09 (UTC)

i had similar issues with a package requiring third party license agreements. my solution was a check in prepare() if the required dependency file is available... and if not spam messages to the terminal telling the user to manually download the file and exit with 1. this somehow made sure that people stopped writing "403" or "does not build" in comments. just an idea :-) @tuxayo check out first lines of PKGBUILD: # !!! NOTE !!! # Please download AMD-APP-SDKInstaller-v3.0.124.132-GA-linux<your_bits_here>.tar.bz2 from # # and paste it next to this PKGBUILD

tuxayo commented on 2015-09-21 09:55 (UTC) goes 403

Vi0L0 commented on 2015-09-06 19:25 (UTC)

thanks robbepop updated to stable 3.0: - AMD removed aparapi samples amdapp-sdk-aparapi is no longer included - it will NOT be automatically removed, thought - new header files: internal/ cl_agent_amd.h cl_debugger_amd.h cl_icd_amd.h cl_kernel_info_amd.h cl_profile_amd.h cl_thread_trace_amd.h

robbepop commented on 2015-09-02 10:12 (UTC)

Can you please update your PKGBUILD for the new version of the AMDAPP-SDK. Newest version is v3.0.124.132, got released on 25th August and is no longer beta.

Vi0L0 commented on 2015-03-08 14:00 (UTC)

check out first lines of PKGBUILD: # !!! NOTE !!! # Please download AMD-APP-SDK-v3.0-0.113.50-Beta-linux<your_bits_here>.tar.bz2 from # # and paste it next to this PKGBUILD

ShyPixie commented on 2015-01-25 21:50 (UTC)

The requested URL returned error: 403 Forbidden

Vi0L0 commented on 2014-12-14 22:45 (UTC)

I will update [catalyst] repo with this package within days (maybe tomorrow), will leave stable amdapp sdk 2.9.1 version inside [catalyst-stable]

Vi0L0 commented on 2014-12-14 22:43 (UTC)

updated to 3.0 beta: - OpenCL 2.0 support - added 1 package named amdapp-sdk-docs because amd has added some nice documentation (PDF) - some cleanup

GreenRaccoon23 commented on 2014-11-28 21:24 (UTC)

AMD really won't let us download this package directly? That's annoying.

t638403 commented on 2014-10-29 16:32 (UTC)

$ head -n4 PKGBUILD # !!! NOTE !!! # Please download AMD-APP-SDK-linux-v2.9-1.599.381-GA-<your_bits_here>.tar.bz2 from # # and paste it next to this PKGBUILD

Vi0L0 commented on 2014-09-26 19:59 (UTC)

Thanks kralyk :), I will look at it in the free time

kralyk commented on 2014-09-26 16:05 (UTC)

Hi Violo, you don't need to have users download the tarball, it can be downloaded from the PKGBUILD. See my libopencl package on how to do it...

skeksis268 commented on 2014-09-26 09:22 (UTC)

Same 403 error here

Jesin commented on 2014-09-20 20:50 (UTC)

I'm on x86_64, and I'm getting this error message when I run makepkg: curl: (22) The requested URL returned error: 403 Forbidden ==> ERROR: Failure while downloading AMD-APP-SDK-linux-v2.9-1.599.381-GA-x64.tar.bz2

Vi0L0 commented on 2014-09-14 12:37 (UTC)

updated to 2.9.1, sorry for long waiting, it was not hard, just had no time :/. Like said i created new package amdapp-sdk-nocatalyst with libs and icd file inside for non-catalyst users, it conflicts with catalyst-utils to warn catalyst users and make sure that they would not install it.

Vi0L0 commented on 2014-09-03 20:08 (UTC)

Thanks for the head up, beroal. I decided to modify 2.9.1 a bit, probably I will move libs and icd to a new splitted package (because they should be unneeded when catalyst is installed) Give me some time, depending on other works from 1 to 4 days.

beroal commented on 2014-08-31 15:24 (UTC)

The version 2.9.1 is out.

Vi0L0 commented on 2014-07-14 17:40 (UTC)

forgot to add 1 important step, lets fix it: BTW when using newer hardware: - install fresh catalyst and opencl-catalyst - install amdapp-sdk - rm /etc/OpenCL/vendors/amd.icd (otherwise code won't run on a GPU) - rm /etc/profile.d/ (as above so below)

Vi0L0 commented on 2014-07-13 09:57 (UTC)

it could provide libcl... BUT there's one problem: it doesn't support newest hardware - ie. it's not working with my 290X. It's better to use opencl-catalyst package as a libcl provider, because libs from out there always supports what catalyst supports. It would be the best to provide libcl in both amdapp-sdk and opencl-catalyst - but to let it happen I would have to remove libcl from opencl-catalyst's conflicts table and if I do this pacman will let me to install libcl package (it will fail when checking files, but its too late), or from the other side while having libcl package installed it will let me to install opencl-catalyst (and fail once again) BTW when using newer hardware: - install fresh catalyst and opencl-catalyst - install amdapp-sdk - rm /etc/OpenCL/vendors/amd.icd (otherwise code won't run on a GPU)

vchuravy commented on 2014-07-09 19:28 (UTC)

Shouldn't this package not also provide a libcl package? Since one is shipped alongside the sdk depending on the one provided by nvidia seems redundant to me. Also nvidia doesn't support OpenCL 1.2 with their libcl implementation.

Vi0L0 commented on 2013-11-27 16:39 (UTC)

Thats why there is this note on atop of the PKGBUILD: Please download AMD-APP-SDK-v2.9-lnx<your_bits_here>.tgz from and paste it next to this PKGBUILD

Timothee commented on 2013-11-27 16:20 (UTC)

Hello, I have a 403 HTTP error for the link :/

Vi0L0 commented on 2013-11-12 22:22 (UTC)

additional info to 2.9-1 from amd: "Recommandations for GPU user: Install 13.10 (or later) Catalyst driver then install SDK 2.9 to avoid the CPU RT files installation."

Vi0L0 commented on 2013-11-12 22:15 (UTC)

updated to 2.9-1, this release is different: - no docs directory, no global make, no license file(?), simple repack - no AMDAPPSDKROOT, LD_LIBRARY_PATH instead - bolt[1] samples, no sdkutils samples [1] from AMD: Bolt is a C++ template library providing GPU off-load for common parallel algorithms. We’ve added cross-OS support as well as numerous performance improvements and new functions. Bolt supports Cross vendor platforms today via CPU and multi-core fallback paths, but stay tuned, we will tackle cross-vendor acceleration next. clMath accelerates BLAS and FFT libraries accessible from Fortran, C and C++. archive tarballs are here, though you may have problems with downloading original package from amd..., anyhow: @BeepDog: added glu, thanks

BeepDog commented on 2013-09-16 18:37 (UTC)

../../../../../include/GL/glew.h:1138:20: fatal error: GL/glu.h: No such file or directory #include <GL/glu.h> ^ I'm assuming that there's a missing dependency here? I see glut, and various other GL dependencies in there, but it fails to build.

Vi0L0 commented on 2013-09-08 20:45 (UTC)

@jakogut: ok, I will look at it. @qdoe: sums are fine, please read the description atop of PKGBUILD file. You need to download it from the web site because direct download isn't allowed

5chdn commented on 2013-09-05 08:41 (UTC)

AMD-APP-SDK-v2.8.1.0-lnx64.tgz fails in sha256sum validation, could you recheck the package? thanks

jakogut commented on 2013-08-23 06:34 (UTC)

@Vi0L0: When makepkg is configured to use another build directory, it can't find the patches required by this package. You might want to change the relative paths to use ${pkgdir}.

Vi0L0 commented on 2013-07-13 20:56 (UTC)

@asd: soon I will upload 2.7 into the AUR

Vi0L0 commented on 2013-07-13 20:56 (UTC)

updated to amdapp-sdk package was split into 3 packages: amdapp-sdk :: main package amdapp-sdk-aparapi :: AparapiUtil and aparapi samples amdapp-sdk-opencv :: OpenCVUtils and opencv samples AMD-APP-SDK isn't available to be directly downloaded, you need to accept the license agreement... Please download AMD-APP-SDK-v2.8.1.0-lnx<your_bits_here>.tgz from: and paste it next to PKGBUILD Please let me know if you will find direct link

commented on 2013-07-08 18:23 (UTC)

Perhaps worth mentioning, 2.8 does not seem to work with legacy drivers (catalyst-total-hd234k in my case), but 2.7 does.

Vi0L0 commented on 2013-05-27 18:44 (UTC)

You are right veox. I will fix it with the next release (or sooner if no realease will came up in a short time)

veox commented on 2013-05-24 19:17 (UTC)

Perhaps it would be better to rename `/etc/profile.d/` to `/etc/profile.d/` or something like it, and include it in the tarball instead of generating in inside the PKGBUILD.

Vi0L0 commented on 2013-02-12 20:59 (UTC)

Known as amdstream in old days ;P... By default this package doesn't depends on amdapp-aparapi and isn't providing aparapiUtil and aparapi's samples - please set _aparapi=y on top of PKGBUILD if you want otherwise

Vi0L0 commented on 2012-12-22 22:31 (UTC)

If someone want to/can check amd app sdk 2.8 - here it is:

Vi0L0 commented on 2012-12-20 12:22 (UTC)

Anyhow i think we should wait some time for madmike...

Vi0L0 commented on 2012-12-20 12:04 (UTC)

I've been e-mailing some time ago with madmike but he didn't respond to my last e-mail. Yesterday i did mailed him one more time to propose some coporate work. Latelly I've been working on PKGBUILDs for AMD APP SDK 2.8, AMD APP Profiler and aparapi, i've got working solution, but as i'm not a dev i would like to work with some real opencl dev so he could tell is it actually good. I'm also planning to work on CodeXL, hope to end it all before x-mas.

PedsXing commented on 2012-12-20 09:50 (UTC)

This PKGBUILD has been out of date for a long time. Please update or disown it.

ptchinster commented on 2012-08-14 12:44 (UTC)

error: failed to commit transaction (conflicting files) amdstream: /usr/bin/clinfo exists in filesystem Errors occurred, no packages were upgraded. $ pacman -Qo /usr/bin/clinfo /usr/bin/clinfo is owned by catalyst-utils 12.6-1

commented on 2012-07-17 03:54 (UTC)

Here is an updated PKGBUILD file for AMD-APP-SDK 2.7: Please integrate this.

commented on 2012-07-17 03:06 (UTC)

AMD-APP-SDK 2.7 has been released. Please update PKGBUILD.

commented on 2012-05-15 19:57 (UTC)

Thanks for the fix halfie, after doing what you say works ok.

big_gie commented on 2012-05-02 16:41 (UTC)

In that case the package is broken...

commented on 2012-05-02 16:39 (UTC)

Here is a simple fix to solve the problem: sudo cp /usr/lib/ /opt/amdstream/lib After this things are working fine :-)

big_gie commented on 2012-05-02 16:36 (UTC)

There's probably something wrong with your OpenCL driver then... That would not surprise me of the catalyst driver though. I would myself try to find which function is crashing and either report it or just try to skip it entirely. clinfo does not even run a kernel. It just print the platforms/devices information.

commented on 2012-05-02 16:32 (UTC)

big_gie: Even other programs (JtR OpenCL build) are crashing with segmentation fault. So clinfo isn't alone.

big_gie commented on 2012-05-02 16:25 (UTC)

IIRC clinfo was built with OpenCL v1.1 functions while amd's driver is only 1.0, so it crashes when trying. clinfo runs fine on an nvidia system. In that case it's just a poorly written clinfo. Take a look at my oclutils:

commented on 2012-05-02 16:22 (UTC)

/opt/amdstream/bin/clinfo is crashing with segmentation fault? I re-install catalyst-hook. Any ideas to fix this problem?

commented on 2012-04-14 01:23 (UTC)

Can be adapted to ccr?

commented on 2012-04-11 23:34 (UTC)

@nplatis meh, that's a stupid bug in yaourt. In order to save time and energy i set PKGEXT to ".tar.gz". Yaourt ist ignoring this and tries to install pkgname.pkg.tar.xz. So manually issuing pacman -U /tmp/<path>/*tar.gz will install the package.

nplatis commented on 2012-04-11 23:13 (UTC)

Cannot install the package. Using yaourt, building seems to be OK, but installation fails with the message: '/tmp/yaourt-tmp-root/PKGDEST.fck/*.pkg.tar.xz': cannot find or read package

commented on 2012-04-04 00:05 (UTC)

weird readlink()-compile error worked around by including unistd.h -> 2.6-2

commented on 2012-01-11 22:10 (UTC)

kralyk has handed the package to me, so I could merge the changes into the PKGBUILD. Therefore I removed the patch files from my webserver.

commented on 2011-12-22 12:01 (UTC)

APP 2.6 is out Since there a couple of issues now I modified the PKGBUILD slightly: + include app 2.6 + include ICD reference again + include symlink for clinfo + removed empty-source-archive workaround ( isn't it simpler to just extract the sub-archive in build() ? ) + renamed OVDecode to OpenVideo + added implicit linking patch ( fixes old GL and new introduced X11 linking error ) relevant files: optional file: happy holidays everyone

giniu commented on 2011-12-19 16:04 (UTC)

I also vote for inclusion of icd, I use amd stream only for CPU as I have nVidia GPU, but I'd still would like to be able to use my Phenom with simple OpenCL apps - without installing catalyst-utils.

commented on 2011-12-17 22:35 (UTC)

Please include the creation of the icd reference in /etc/OpenCL/vendors again. Otherwise people who want to code/test on x86 only will tend to install catalyst-utils and that's just bananas.

commented on 2011-12-17 22:18 (UTC)

btw, the linking error with GL and GLU is because ld no longer recursively adds shared libraries that are marked as NEEDED. It looks like amd's code is based on the implicit inclusion of these shared libraries and therefore they have to specifically link with -lGL and -lGLU. This has to be fixed by AMD I guess. Bugreport anyone? slightly more generic workaround would be: --- make/ 2011-07-18 20:04:15.000000000 +0200 +++ /dev/null 2011-12-17 23:05:48.027376727 +0100 @@ -227,6 +227,9 @@ LD := g++ endif + # work around implicit linking error + LDFLAGS += -Xlinker --copy-dt-needed-entries + ifdef MINGW LDFLAGS += -L/usr/X11R6/lib else

kjslag commented on 2011-12-14 00:28 (UTC)

jgottula's solution works for me also. I got the same error message. pacman -Qo /usr/lib/ /usr/lib/ is owned by catalyst-utils 11.12-1

jgottula commented on 2011-12-06 09:58 (UTC)

$ pacman -Qo /usr/lib/ /usr/lib/ is owned by catalyst-utils 11.11-3 readelf:

kralyk commented on 2011-12-06 09:46 (UTC)

@jgottula: I think the main problem is this line: /usr/lib/ could not read symbols: Invalid operation What's with your What pkg is it from? Could you post 'readelf -a' of it?

jgottula commented on 2011-12-06 05:50 (UTC)

I'm getting build failures with a few source files in the 2.5-4 revision of this package (new today apparently) because the makefiles aren't telling g++ to link to GL or GLU (see /usr/bin/ld: build/debug/x86_64//FluidSimulation2D.o: undefined reference to symbol 'glEnd' /usr/bin/ld: note: 'glEnd' is defined in DSO /usr/lib/ so try adding it to the linker command line /usr/lib/ could not read symbols: Invalid operation The rough workaround is to add a line to the end of make/ that says 'LLIBS += GL GLU', and then everything compiles fine. There's probably a more correct way to fix it, but that was good enough to work.

kralyk commented on 2011-12-05 23:01 (UTC)

Updated for compliance with 11.11 Some files moved to catalyst, libopencl is no longer needed.

toffyrn commented on 2011-11-28 12:04 (UTC)

I tried removing amdstream and using only catalyst-utils+opencl-headers, see forum post: Thank you :)

kralyk commented on 2011-11-28 12:01 (UTC)

I'll have a look at it...

toffyrn commented on 2011-11-28 10:54 (UTC)

Since last catalyst(-utils) upgrade this package conflicts, and I can no longer develop OpenCL. Should one use only catalyst, and drop this one?

kralyk commented on 2011-09-24 20:30 (UTC)

Oh my god you're right. I'm so sorry, I forgot the 'makepkg --source' utility won't pack the other arch in there for me. Updated!

BluePeril commented on 2011-09-24 20:25 (UTC)

AMD-APP-SDK-v2.5-RC2-lnx32.tgz is missing in the tarball

danmilon commented on 2011-09-23 00:42 (UTC)

@kralyk Gotcha. Thank you for your help!

kralyk commented on 2011-09-22 23:02 (UTC)

@razorg: I wouldn't know about that, I personally don't use yaourt, sorry. The package works allright using standard process, and that's the furthest I can be concerned. Those zero-sized *-RC2-* files are a trick that ensures proper extraction of the sources, which AMD packs in a strange way. As all the deps are in repos, I suggest you build this pkg in a stadard way using makepkg (makepkg -sic for convenience).

danmilon commented on 2011-09-22 21:35 (UTC)

@kralyk actually i am installing through yaourt. Should the problem exist there also? Thank you.

kralyk commented on 2011-09-22 17:15 (UTC)

@razorg: You need to download the whole tarball, not just the PKGBUILD.

danmilon commented on 2011-09-22 00:27 (UTC)

gives error on building: ==> ERROR: AMD-APP-SDK-v2.5-RC2-lnx32.tgz was not found in the build directory and is not a URL. ==> ERROR: Makepkg was unable to build amdstream.

kralyk commented on 2011-09-19 16:14 (UTC)

Thanks, updated (hope it's ok).

toffyrn commented on 2011-09-19 13:24 (UTC)

The sha256sum seems to have changed. :)

kralyk commented on 2011-08-06 09:33 (UTC)

Alright, some sanity has been reached, updating. Please ignore those zero-sized .tgzs, those are needed for proper sources extraction. I hope this is reasonably ok, feel free to test.

kralyk commented on 2011-08-04 21:41 (UTC)

I know this is totally out of date, but those AMD people are doing some real weirdness with the download link and the tgz. I am definitely NOT updating this until some sanity is reached once again.

kralyk commented on 2011-08-01 12:07 (UTC)

Updated. No conflicts - libopencl also provides libcl. Use whichever you want, but be advised that extra/libcl only supports OpenCL 1.0, on the other hand it enables you to use both nVidia and AMD SDKs at the same time. Depends on use case really.

john_schaf commented on 2011-07-29 10:44 (UTC)

Should this not conflict with libcl ?

kralyk commented on 2011-07-07 02:28 (UTC)

Thanks big_gie, updated.

big_gie commented on 2011-06-15 18:07 (UTC)

I fail to get a platform on my netbook. I've found out it's because it's x86 (not x86_64) and a link is missing: ln -s /opt/amdstream/lib/$_arch/ $pkgdir/usr/lib/ Now it works!

kralyk commented on 2011-06-08 11:26 (UTC)

Yes, I'm aware of this issue, this is being resolved, we have to talk about this with other maintainers. Unfortunatelly I'm currently terribly short on free time because of exams at my uni :/

Void-995 commented on 2011-06-08 11:24 (UTC)

I've already installed NVIDIA CUDA on my PC to work with OpenCL on GPU. Aslo I want to use an amdstream to have OpenCL support on my CPU too, but installation failed with: Targets (1): amdstream-2.4-2 Total Download Size: 0.00 MB Total Installed Size: 114.60 MB Proceed with installation? [Y/n] y (1/1) checking package integrity [#########################################################] 100% (1/1) checking for file conflicts [#########################################################] 100% error: failed to commit transaction (conflicting files) amdstream: /usr/lib/ exists in filesystem amdstream: /usr/lib/ exists in filesystem ==> WARNING: Your packages are saved in /tmp/yaourt-tmp-void-995 ==> WARNING: Following packages have not been installed: amdstream

kralyk commented on 2011-05-15 16:55 (UTC)

@madmike: Well, yes, the package is not intended for in-place compiling (and compiling within system files is not recommended in general for that matter). The samples are already compiled, you'll find the binaries in /opt/amdstream/samples/opencl/bin/x86*/ If you want to make modifications to a sample or somehow customize the build process, please either modify the PKGBUILD itself or make a local copy of the source tarball somewhere in $HOME.

commented on 2011-05-15 16:49 (UTC)

There are some files missing from the original tar.gz. When I am trying to build /opt/amdstream/samples/opencl/cl/app/NBody I get: Makefile:3: ../../../../../make/ No such file or directory Makefile:38: ../../../../../make/ No such file or directory make: *** No rule to make target `../../../../../make/'. Stop. Is this intentional?

big_gie commented on 2011-05-13 12:57 (UTC)

Competition from intel: :)

kralyk commented on 2011-04-29 06:25 (UTC)

I think we don't need libatical dependency any more. Updated.

big_gie commented on 2011-04-28 18:36 (UTC)

The packagye works for me, but valgrind reports many errors. Probably bugs in amd stream... I reported the problem on amd forum.

kralyk commented on 2011-04-28 18:30 (UTC)

So what's up? Does it work for you? Any problems?

kralyk commented on 2011-04-21 22:40 (UTC)

Ok, I finally updated this. Once again sorry for the delay. Feel free to test, I seriously doubt this first pkgrel is flawless... The bad news (I know of) is that the 'clc' tool is gone. The good news is that the pkg has been cleaned by AMD somewhat, so we'll hopefully be able to avoid ld config path hacks. It's still windows-installer-like messy though... So still in /opt. Alright, that's it from me for now, I'll get back to you asap...

big_gie commented on 2011-04-21 21:15 (UTC)

You mean the kernel-only compiler? What is used to compile a kernel file before running the program? If I clBuildProgram() each time I should be fine?

commented on 2011-04-20 18:13 (UTC)

catalyst-test 11.4 and AMD APP SDK 2.4 seem to increase the performance of OpenCL apps by a bit. Please update the PKGBUILD.

commented on 2011-04-20 18:12 (UTC)

>> It fails a call to clGetPlatformIDs(). Here is a quick dirty fix: sudo ln -s /opt/amdstream/lib/x86_64/ /usr/lib/

big_gie commented on 2011-04-11 23:01 (UTC)

Here's a patch for version 2.4: Note that I can't run an OpenCL code with this new version. It fails a call to clGetPlatformIDs(). This is used to run OpenCL code on the CPU.

toffyrn commented on 2011-03-02 21:31 (UTC)

After the last pkgrel it seems to find the .so files, however a simple script containing only: "import pyopencl as cl ctx = cl.create_some_context()" crashes with: "File "/usr/lib/python2.7/site-packages/pyopencl/", line 314, in create_some_context platforms = get_platforms() pyopencl.LogicError: clGetPlatformIDs failed: invalid/unknown error code" I am also having issues when trying to run C++ applications with CL after this last pkgrel... Is it working for you? Contact me at <toffyrn at> if this is filling up the comments field.

kralyk commented on 2011-03-01 14:09 (UTC)

Ok, updated. I hope it's ok.

toffyrn commented on 2011-03-01 13:44 (UTC)

thx. waiting with a hackish symlink in the meantime :)

kralyk commented on 2011-03-01 13:27 (UTC)

Oh I see, it's the .1 thats missing. It's f***ed up since AMD ignores the linux concept of shared objects altogether. I'm gonna remove the ld conf entry, I don't wanna jeopardize glew and glut packages any more. Stay tuned, will be fixed.

toffyrn commented on 2011-02-28 22:22 (UTC)

This is found: $ cat /etc/ /opt/amdstream/lib/x86 /opt/amdstream/lib/gpu $ Try installing python2-pyopencl and run a simple python script containing only "import pyopencl". When not having symlinked as /lib/ python can not find it... Perhaps it is a bug in python2-pyopencl package?

kralyk commented on 2011-02-28 17:49 (UTC)

@toffyrn: That is strange, should be in LD path. Could you please check your /etc/ for 'amdstream.conf' and if it's content is right?

toffyrn commented on 2011-02-24 12:52 (UTC)

python2-pyopencl has a dependency on libcl, but it does not find the "", so you would perhaps need a symlink?

kralyk commented on 2011-02-08 23:01 (UTC)

Thank you for noticing me, Vi0L0! I would also say amdstream, because catalyst-utils only provides CAL runtime (on top of which OpenCL works), but no OpenCL support whatsoever as far as I know.

Vi0L0 commented on 2011-02-08 22:45 (UTC)

opencl-headers lands in [community] with optdeps: 'libcl: OpenCL library' libcl is provided only by nvidia-utils. Now i don't know which package on amd side should provide libcl - amdstream or catalyst-utils ? I would guess amdstream. What do you think kralyk?

kralyk commented on 2011-01-13 13:54 (UTC)

mrspin: updated

mrspin commented on 2011-01-12 19:40 (UTC)

Doesn't install cal_ext.h header

commented on 2011-01-01 10:13 (UTC)

works great. thanks!

kralyk commented on 2010-12-30 16:06 (UTC)

Omg I'm so absent-minded, sry, once again fixed...

BluePeril commented on 2010-12-30 07:25 (UTC)

the lines chmod 755 $pkgdir/opt/amdstream/bin/x86_64/clc chmod 755 $pkgdir/opt/amdstream/lib/x86_64/*.so should be chmod 755 $pkgdir/opt/amdstream/bin/$_arch/clc chmod 755 $pkgdir/opt/amdstream/lib/$_arch/*.so Otherwise packaging will fail with the 32bit version

kralyk commented on 2010-12-29 12:52 (UTC)


BluePeril commented on 2010-12-29 08:05 (UTC)

The checksum of the 32bit version is wrong. It should be 07fd15a51a678c4378767f61466daf9da21d0bab6088a97a70136dd8c976aca6

kralyk commented on 2010-12-28 21:20 (UTC)

Meh, thanks, fixed...

schneida commented on 2010-12-28 20:52 (UTC)

Sry, but the download for the source file is wrong. It downloads the license agreement instead of the actual file. A correct url would be:

kralyk commented on 2010-12-28 16:02 (UTC)

Updated. Pkg doesn't include the header now and has been cleaned & simplified.

big_gie commented on 2010-12-17 18:12 (UTC)

You were too fast, yes I agree that this package should remove the headers ;)

big_gie commented on 2010-12-17 18:11 (UTC)

All headers are, I think, almost the same, you can verify by yourself. I think the best generic approach would be to let opencl-headers install the headers and remove the headers from these packages.

commented on 2010-12-17 18:11 (UTC)

Or perhaps the "Arch Way" would be to not include the headers in this package, and point users to the "opencl-headers" package if they need them?

commented on 2010-12-17 18:07 (UTC)

Would it be possible to add "opencl-headers" to the "provides" array, so that a user with this package installed may build packages that makedepend on "opencl-headers", or are the headers from this package too dissimilar to those from "opencl-headers"?

commented on 2010-12-06 14:45 (UTC)

I am a robot. This is not an official message. AUR guidelines suggest to not include binaries. Examples: amdstream/more-docs.tar.gz Feel free to disregard this as you would any other comment. This robot will not post here again.

kralyk commented on 2010-09-04 10:41 (UTC)

Can you please provide wich package those files belong to? So I can put it in conflicts.

commented on 2010-09-02 23:23 (UTC)

I'm getting this error, I am pretty sure this can be ignored, but still it would be nice if this could be fixed. Proceed with installation? [Y/n] checking package integrity... (1/1) checking for file conflicts [######################] 100% error: failed to commit transaction (conflicting files) amdstream: /usr/include/CL/cl.h exists in filesystem amdstream: /usr/include/CL/cl_ext.h exists in filesystem amdstream: /usr/include/CL/cl_gl.h exists in filesystem amdstream: /usr/include/CL/cl_gl_ext.h exists in filesystem amdstream: /usr/include/CL/cl_platform.h exists in filesystem amdstream: /usr/include/CL/opencl.h exists in filesystem

commented on 2010-08-26 19:47 (UTC)

Thanks, this time works ok. Keep up the good work :)

kralyk commented on 2010-08-26 10:21 (UTC)

Thanks pfdm, updated.

commented on 2010-08-26 09:52 (UTC)

Just copy the the lib/gpu directory to /opt ! The PKGBUILD is not correct, add these lines : +cp -r ./lib/gpu $pkgdir/opt/amdstream/lib/ -echo /opt/amdstream/lib/$_arch > $pkgdir/etc/ +echo /opt/amdstream/lib/$_arch >> $pkgdir/etc/ +echo /opt/amdstream/lib/gpu >> $pkgdir/etc/ It should work!

kralyk commented on 2010-08-26 09:51 (UTC)

what does ls -l /usr/bin/clc output?

commented on 2010-08-26 00:32 (UTC)

It's not working for me (catalyst 10.8) ./NBody BUILD LOG ************************************************ Internal error: Compilation failed. ************************************************ Error: clBuildProgram failed. Error code : CL_BUILD_PROGRAM_FAILURE any help?

kralyk commented on 2010-08-18 11:42 (UTC)

Updated to 2.2. Now with OpenCL 1.1 support :) Hope everything works, feel free to test...

kralyk commented on 2010-08-17 13:42 (UTC)

The package is now updated. Sorry for the delay, I was abroad and in a deep wilderness with no internet :) And thanks Vi0L0 for your help!

commented on 2010-07-27 01:28 (UTC)

Thanks a lot ViOLO, with the first PKGBUILD works ok now :D

Vi0L0 commented on 2010-07-26 02:15 (UTC)

@Cory: thats because theres simply no llc file, and PKGBUILD of this pkg need to be updated. Both PKGBUILDs mentioned below got it - and

commented on 2010-07-25 02:44 (UTC)

What i'm doing wrong? ./FindNumDevices Supported CAL Runtime Version: 1.3.185 Found CAL Runtime Version: 1.4.696 Use -? for help CAL initialized. Finding out number of devices :- Device Count = 1 CAL shutdown successful. Press enter to exit... ./NBody sh: /opt/amdstream/bin/x86/llc: No such file or directory BUILD LOG ************************************************ Error: Compilation from LLVMIR binary to IL text failed! ************************************************ Error: clBuildProgram failed. Error code : CL_BUILD_PROGRAM_FAILURE What's the problem?

kralyk commented on 2010-07-06 15:30 (UTC)

Yes I know about those env vars I just didn't find any reason why set them up... The SDK seemed to work just as fine. I look into this and will possibly fix, if I have free time...

commented on 2010-07-02 18:45 (UTC)

The /usr conflicts were PEBCAK. Ignore. The other change I forgot to mention is to install a file in /etc/profile.d to set ATISTREAMSDKROOT and ATISTREAMSDKSAMPLEROOT, both to /opt/amdstream. This variable needs to be set according to the stream sdk installation instructions.

kralyk commented on 2010-07-01 20:01 (UTC)

mwc, what where the conflicts in /usr?

commented on 2010-07-01 19:03 (UTC)

I had to fix up a few things in this PKGBUILD in order to make this work on my system (I use a 5770 for development). 1) Remove all the symlinks into /usr to fix conflicts 2) Install llc from this package into /opt/amdstream/bin/$_arch. If this isn't done, then the examples fail complaining that it's missing. 3) Since we aren't symlinking libs into /usr, we instead set up an entry to add the SDK libs to the search path Updated PKGBUILD here:

maleadt commented on 2010-06-04 12:02 (UTC)

Having crashed my main computer, I can't test it right away, but I'll be sure have a go at it as possible. I've forced an install back then by removing the libatical dependency, which worked but obviously isn't a good solution. I've also seen an update of yaourt come along, so it might have been fixed already. Thanks for your efforts in providing this package :)

kralyk commented on 2010-06-02 13:15 (UTC)

Hi, I personally don't use yaourt, so I'm not sure if I'll be able to help. Maybe this is because catalyst package doesn't specify the version of libatical it provides. You can try to change the catalyst PKGBUILD like this: provides = (... 'libatical=10.5' ...) (version of libatical is the same as of the catalyst itself) Please let me know whether this would work...

maleadt commented on 2010-05-27 17:09 (UTC)

Weird issue: although I've got catalyst 10.5 installed (which provides libatical >= 10.4), yaourt attempts to install libatical (and fails) when installing amdstream's dependencies... ==> Install or build missing dependencies for amdstream: ==> Building and installing package ==> amdstream dependencies: - libatical (package that provides libatical already installed) - libgl (package that provides libgl already installed) - llvm (package that provides llvm already installed) - gcc-libs (already installed) - mesa (already installed) - perl (already installed) - llvm (package that provides llvm already installed) ==> Making package: amdstream 2.1-1 x86_64 (do mei 27 19:06:44 CEST 2010) ==> Checking Runtime Dependencies... ==> Installing missing dependencies... libatical>=10.4 package not found, searching for group... error: 'libatical>=10.4': not found in sync db ==> ERROR: Pacman failed to install missing dependencies. Error: Makepkg was unable to build amdstream package.

kralyk commented on 2010-05-20 00:22 (UTC)

Ok, the package is now updated. Thank you a lot for your help, cast1c, I have added you to Contributor field, hope you don't mind :) The SDK once again works for me [on x86_64]. Please feel free to test further. Please also note that amdstream is now dependant on the official LLVM package, so that we can avoid the conflict.

commented on 2010-05-19 19:37 (UTC)

Whoops, you are right. I accidently still had LD_LIBRARY_PATH set to a location where the whole SDK was unpacked. All files in /opt/amdstream/lib/<arch> except for and need to be symlinked to /usr/lib. A PKGBUILD including these changes can be found here: It is again tested on i686 only.

kralyk commented on 2010-05-18 18:49 (UTC)

Thanks for your efforts caust1c, but there is still some problem. I have already tried copying those files in /etc/OpenCL/vendors, I still cannot run OpenCL programs though. Have tried running CLInfo? (samples/opencl/bin/<platform>/CLInfo)

commented on 2010-05-18 18:37 (UTC)

OpenCL stores icd registration files differently now. I changed the PKGBUILD accordingly. The updated version can be found here: I have tested it successfully for i686, i didn't test it for x86_64 though. md5sums are updated for x86_64 as well

kralyk commented on 2010-05-07 20:07 (UTC)

Sorry guys it's taking so much time but I'm having some troubles getting the new version to work properly. I'll do my best to update as soon as possible...

kralyk commented on 2010-05-03 17:22 (UTC)

Thanks for noticing me about new version. Will be updated shortly.

kralyk commented on 2010-04-12 16:12 (UTC)

Hey. Thanks for your comments and sorry for late answer. The package has been updated. So far, it conflicts with llvm. But I'm thinking of updating it to require llvm, as it would enable users to have both llvm and amdstream.

ptb commented on 2010-04-09 20:24 (UTC)

Building fails if make is parallelized using -j e.g. -j5. Removing the -j option from the MAKEFLAGS in /etc/makepkg.conf or changing the "make || return 1" line to "make -j1 || return 1" fixed the problem for me.

Huulivoide commented on 2010-04-05 12:24 (UTC)

conflicts whit the llvm package /usr/bin/llc

Huulivoide commented on 2010-04-05 12:20 (UTC) some build error whne building the samples

kralyk commented on 2010-03-25 14:57 (UTC)

See forum topic for more info & feedback: