BOINC 7.6.23 for Linux Recongnizes OpenCL AMD/ATi GPU via MESA, But No Apps From SETI?

Message boards : Number crunching : BOINC 7.6.23 for Linux Recongnizes OpenCL AMD/ATi GPU via MESA, But No Apps From SETI?
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Paul

Send message
Joined: 17 May 99
Posts: 72
Credit: 42,977,964
RAC: 43
United States
Message 1762780 - Posted: 6 Feb 2016, 6:38:17 UTC

I'm sure those in this community know that AMD proprietary drivers have been trouble for a long time, especially very recently. I thought it was a good time to try open-source drivers and to my amazement...they work!

I've tried this over the years many times while in the middle of re-installing Catalyst and BOINC never found my GPU. Turns out the official BOINC for Linux release (7.2.42) hasn't been updated in years (long story, problem=wxWidgets), but the code has and if you build it, it will recognize your GPUs just fine via MESA.

But, I'm not getting any GPU apps or tasks from SETI (or Einstein). Is that because the new SETI v8 opencl AMD app we got last week will not work with OpenCL AMD GPU via MESA, only Catalyst? Please confirm.

(Also, does someone know why? Obviously, it's not supported, but it has to be asked, what part of OpenCL is missing in the open-source drivers. I mean, I can install MESA development libs if that's all that's needed. Or, does anyone know if there are beta apps for this "platform"? I'd be willing to test that. I checked at Lunatics and I don't see any Linux GPU apps anymore.)
ID: 1762780 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1762952 - Posted: 6 Feb 2016, 19:01:00 UTC

I wonder if the OpenCL version supported by your driver is to low.
[2] AMD AMD PITCAIRN (DRM 2.43.0, LLVM 3.7.0) (1024MB) OpenCL: 1.1

I'm not sure, but the server might be checking for at least OpenCL 1.2 for ATI GPUs. Like it does for iGPUs.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1762952 · Report as offensive
Juha
Volunteer tester

Send message
Joined: 7 Mar 04
Posts: 388
Credit: 1,857,738
RAC: 0
Finland
Message 1762962 - Posted: 6 Feb 2016, 19:41:09 UTC - in response to Message 1762780.  

Could you post <coprocs> section from sched_request_setiathome.berkeley.edu.xml ?

The ATI plan classes here have been set up with Catalyst in mind and I'm thinking none of them match MESA clover.

You could download the app manually and make an app_info.xml file. Someone else has to tell the right download URL.
ID: 1762962 · Report as offensive
Paul

Send message
Joined: 17 May 99
Posts: 72
Credit: 42,977,964
RAC: 43
United States
Message 1762964 - Posted: 6 Feb 2016, 19:52:18 UTC

Hey HAL!

Yes, that's my suspicion, also. It's too bad I can't check old logs to see what OCL version was detected with fglrx; I think it was 1.2, IIRC. I know PITCARN is 1.2 capable, in any case. Think it would be worth could trying new MESA? I see Phoronix uses MESA development branch for his beading edge tests. I wonder how to tell before I try it if it will help?

Also, I would happily upgrade my card if that would help! That's worth not having to go back to fglrx. But, still, how can tell?

Looks like Einstein *did* just send me OCL apps, but tasks are failing with computation error, so, that's not a good sign. In my experience, the focus with BOINC projects has been exclusively on proprietary drivers, by which I mean this level support detail (i.e. minimum OCL version support or specific OCL features required) isn't reported; projects just say CAL/CUDA/OCL is supported for AMD/nVidia/Intel, and then instructions say to get proprietary drivers.

You are a tester and been all over this form. Can I assume you would know about a specific beta for MESA drivers, if it existed?

I feel like I'm really close. And, I'm really motivated not to have to go back to proprietary drivers. :-) As always, I'm willing to try or test anything. Just let me know if you have ideas.
ID: 1762964 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1762969 - Posted: 6 Feb 2016, 20:38:06 UTC - in response to Message 1762964.  

It's too bad I can't check old logs to see what OCL version was detected with fglrx; I think it was 1.2, IIRC.

You might be able to recover that information from a file called 'stdoutdae.txt' in your BOINC data directory.
ID: 1762969 · Report as offensive
Juha
Volunteer tester

Send message
Joined: 7 Mar 04
Posts: 388
Credit: 1,857,738
RAC: 0
Finland
Message 1762985 - Posted: 6 Feb 2016, 21:40:47 UTC - in response to Message 1762964.  

I'm pretty sure Raistmer has said OpenCL 1.2 is not required. Anyway, there's plenty of plan classes with min_opencl_version=100 . Your problem is most likely the Mesa version 11.1.0 when the minimum required Catalyst versions are higher than that.

And why there is no plan class for Mesa? As far as I know you are the first person to test it here and the second person BOINC wide. The first person has been testing at Einstein. You should probably note his comment about missing modf function.
ID: 1762985 · Report as offensive
Paul

Send message
Joined: 17 May 99
Posts: 72
Credit: 42,977,964
RAC: 43
United States
Message 1763262 - Posted: 7 Feb 2016, 22:06:34 UTC - in response to Message 1762969.  

You might be able to recover that information from a file called 'stdoutdae.txt' in your BOINC data directory.

Thanks for the suggestion. I remember looking in that file many times log ago, but can't find it any more; doesn't exist.
ID: 1763262 · Report as offensive
Paul

Send message
Joined: 17 May 99
Posts: 72
Credit: 42,977,964
RAC: 43
United States
Message 1763267 - Posted: 7 Feb 2016, 22:26:24 UTC - in response to Message 1762985.  

I'm pretty sure Raistmer has said OpenCL 1.2 is not required. Anyway, there's plenty of plan classes with min_opencl_version=100 . Your problem is most likely the Mesa version 11.1.0 when the minimum required Catalyst versions are higher than that.

Cool! Thanks, never seen that before. "plan class"...got it. Now I see how the app matching works, okay.

What do you mean by "Catalyst version"...Okay, I think I get it. MESA and Catalyst both provide the OCL drivers, so it's trying to match condition <min_opencl_driver_revision>112420 (11.2.420), where as I have only 11.1.0? Is that right? Sorry, I don't know what I'm looking at here, really.

And why there is no plan class for Mesa? As far as I know you are the first person to test it here and the second person BOINC wide. The first person has been testing at Einstein. You should probably note his comment about missing modf function.

:-) Yep, I found that thread! OP figured out there is a bug in libclc and is sending me the fix. Seems like I'm going to get Einstein going, at least. And, I'll push for the fix with Fedora--BOINC and libclc.
ID: 1763267 · Report as offensive
Paul

Send message
Joined: 17 May 99
Posts: 72
Credit: 42,977,964
RAC: 43
United States
Message 1763380 - Posted: 8 Feb 2016, 5:26:41 UTC

Quick update... Thanks to Paulie over at Einstein@Home, I have GPU apps from that project working with 7.6.23 + Mesa. I'm thrilled. Thanks again to everyone here for the help. A manual patch to libclc was required, so it should be noted that it's not going to work on Fedora without that.

Regarding requirements and matching plan class, I think I got it wrong and I still don't understand how to use the plan class specs.
ID: 1763380 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1763472 - Posted: 8 Feb 2016, 16:20:06 UTC - in response to Message 1763380.  

I still don't understand how to use the plan class specs.

If you write your own app_info.xml the plan_class will not matter
The only thing hard to find is the name of the executable (and any libs that may go with it if any)
Someone with Linux have to tell the executable names for Linux ATI AMD OpenCL apps
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1763472 · Report as offensive
Paul

Send message
Joined: 17 May 99
Posts: 72
Credit: 42,977,964
RAC: 43
United States
Message 1765649 - Posted: 17 Feb 2016, 7:00:34 UTC

Sorry, I've been playing with E@H...

Now I'm back to looking at SETI. I'm still confused by what Juha said, which was that my Mesa version didn't satisfy the Catalyst requirement. Can you point me to that in the plan class spec file? What version do I need?

I've looked at it twice now and I still don't understand what I'm seeing. I'm confused by the version number encoding, too. I found this:
https://boinc.berkeley.edu/trac/wiki/AppPlanSpec, but the number of digits doesn't match, so it's ambiguous. Why do some apps with name "_nocal" have "_cal_" attributes?

After I understand what's wrong, BilBg is saying I can work around it by going to anonymous platform and writing my own app_info.xml file. But, I still need to get the app binaries, correct? I need to build them myself from SVN repo, right?
ID: 1765649 · Report as offensive
Juha
Volunteer tester

Send message
Joined: 7 Mar 04
Posts: 388
Credit: 1,857,738
RAC: 0
Finland
Message 1765856 - Posted: 17 Feb 2016, 22:56:01 UTC - in response to Message 1765649.  

I was going to link to the plan class doc and explain how it works but there was always something else. Sorry about that.

Ok, you are looking for a plan class that has gpu_type=amd and has opencl tag. Your driver shouldn't support CAL so I think the server sees it as cal_target=0 . That means you can rule out all plan classes that have min_cal_target>0 .

Then the opencl driver version. I think yours is seen as opencl_driver_revision=110100 . That rules out quite a few plan classes. And you can leave out all plan classes that have min_os_version. They match Macs and Windows and only Macs have apps with those plan classes.

I said none of the plan classes matches Mesa but I think I was wrong. opencl_ati_100 looks like it should match and there is an Astropulse app with that plan class.


Anyway, I think we need some mechanism to tell apart Catalyst and Mesa and I'm not seeing anything useful in the plan class doc. Before I go and ask developers to add such mechanism it would be nice if you posted the <coproc> section so that we could see what information is available. It would also make debugging the plan class selection easier.
ID: 1765856 · Report as offensive
Paul

Send message
Joined: 17 May 99
Posts: 72
Credit: 42,977,964
RAC: 43
United States
Message 1765887 - Posted: 18 Feb 2016, 1:02:48 UTC

Ah, got it. Thanks, Juha, that helps becuase I thought that one was a fit, too, which was confusing.

Here's my corpoc file:
    <coprocs>
   <ati_opencl>
      <name>AMD PITCAIRN (DRM 2.43.0, LLVM 3.7.0)</name>
      <vendor>AMD</vendor>
      <vendor_id>4098</vendor_id>
      <available>1</available>
      <half_fp_config>0</half_fp_config>
      <single_fp_config>6</single_fp_config>
      <double_fp_config>63</double_fp_config>
      <endian_little>1</endian_little>
      <execution_capabilities>1</execution_capabilities>
      <extensions>cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_fp64</extensions>
      <global_mem_size>1073741824</global_mem_size>
      <local_mem_size>32768</local_mem_size>
      <max_clock_frequency>1050</max_clock_frequency>
      <max_compute_units>20</max_compute_units>
      <nv_compute_capability_major>0</nv_compute_capability_major>
      <nv_compute_capability_minor>0</nv_compute_capability_minor>
      <amd_simd_per_compute_unit>0</amd_simd_per_compute_unit>
      <amd_simd_width>0</amd_simd_width>
      <amd_simd_instruction_width>0</amd_simd_instruction_width>
      <opencl_platform_version>OpenCL 1.1 MESA 11.1.0</opencl_platform_version>
      <opencl_device_version>OpenCL 1.1 MESA 11.1.0</opencl_device_version>
      <opencl_driver_version>11.1.0</opencl_driver_version>
      <device_num>0</device_num>
      <peak_flops>1680000000000.000000</peak_flops>
      <opencl_available_ram>1073741824.000000</opencl_available_ram>
      <opencl_device_index>0</opencl_device_index>
      <warn_bad_cuda>0</warn_bad_cuda>
   </ati_opencl>
   <ati_opencl>
      <name>AMD PITCAIRN (DRM 2.43.0, LLVM 3.7.0)</name>
      <vendor>AMD</vendor>
      <vendor_id>4098</vendor_id>
      <available>1</available>
      <half_fp_config>0</half_fp_config>
      <single_fp_config>6</single_fp_config>
      <double_fp_config>63</double_fp_config>
      <endian_little>1</endian_little>
      <execution_capabilities>1</execution_capabilities>
      <extensions>cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_fp64</extensions>
      <global_mem_size>1073741824</global_mem_size>
      <local_mem_size>32768</local_mem_size>
      <max_clock_frequency>1050</max_clock_frequency>
      <max_compute_units>20</max_compute_units>
      <nv_compute_capability_major>0</nv_compute_capability_major>
      <nv_compute_capability_minor>0</nv_compute_capability_minor>
      <amd_simd_per_compute_unit>0</amd_simd_per_compute_unit>
      <amd_simd_width>0</amd_simd_width>
      <amd_simd_instruction_width>0</amd_simd_instruction_width>
      <opencl_platform_version>OpenCL 1.1 MESA 11.1.0</opencl_platform_version>
      <opencl_device_version>OpenCL 1.1 MESA 11.1.0</opencl_device_version>
      <opencl_driver_version>11.1.0</opencl_driver_version>
      <device_num>1</device_num>
      <peak_flops>1680000000000.000000</peak_flops>
      <opencl_available_ram>1073741824.000000</opencl_available_ram>
      <opencl_device_index>1</opencl_device_index>
      <warn_bad_cuda>0</warn_bad_cuda>
   </ati_opencl>
<opencl_cpu_prop>
   <platform_vendor>The pocl project</platform_vendor>
   <opencl_cpu_info>
      <name>pthread-AMD FX(tm)-8350 Eight-Core Processor</name>
      <vendor>AuthenticAMD</vendor>
      <vendor_id>4098</vendor_id>
      <available>1</available>
      <half_fp_config>0</half_fp_config>
      <single_fp_config>6</single_fp_config>
      <double_fp_config>6</double_fp_config>
      <endian_little>1</endian_little>
      <execution_capabilities>3</execution_capabilities>
      <extensions>cl_khr_fp64 cl_khr_byte_addressable_store cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics</extensions>
      <global_mem_size>33651023872</global_mem_size>
      <local_mem_size>33651023872</local_mem_size>
      <max_clock_frequency>4013</max_clock_frequency>
      <max_compute_units>8</max_compute_units>
      <nv_compute_capability_major>0</nv_compute_capability_major>
      <nv_compute_capability_minor>0</nv_compute_capability_minor>
      <amd_simd_per_compute_unit>0</amd_simd_per_compute_unit>
      <amd_simd_width>0</amd_simd_width>
      <amd_simd_instruction_width>0</amd_simd_instruction_width>
      <opencl_platform_version>OpenCL 1.2 pocl 0.12</opencl_platform_version>
      <opencl_device_version>OpenCL 1.2 pocl</opencl_device_version>
      <opencl_driver_version>0.12</opencl_driver_version>
   </opencl_cpu_info>
</opencl_cpu_prop>
<warning>NVIDIA: libcuda.so: cannot open shared object file: No such file or directory</warning>
<warning>ATI: libaticalrt.so: cannot open shared object file: No such file or directory</warning>
    </coprocs>

ID: 1765887 · Report as offensive
Paul

Send message
Joined: 17 May 99
Posts: 72
Credit: 42,977,964
RAC: 43
United States
Message 1767583 - Posted: 25 Feb 2016, 21:44:58 UTC

Whoa! I was away for a few days and just rebooted my system. When I started BOINC again, I saw four SETI AP GPU tasks waiting to run.

I don't know what happened because I can't see any log data prior to reboot, but I'm sure it wasn't running any SET GPU work at the end of last week.

The odd thing is, they don't seem to run correctly; they keep getting postponed and no reason is given.

Thu 25 Feb 2016 01:33:45 PM PST | SETI@home | Sending scheduler request: To fetch work.
Thu 25 Feb 2016 01:33:45 PM PST | SETI@home | Reporting 1 completed tasks
Thu 25 Feb 2016 01:33:45 PM PST | SETI@home | Requesting new tasks for CPU and AMD/ATI GPU
Thu 25 Feb 2016 01:33:48 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:33:48 PM PST | SETI@home | Scheduler request completed: got 31 new tasks
Thu 25 Feb 2016 01:33:49 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:33:50 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:33:50 PM PST | SETI@home | Started download of 20dc10ac.17853.6639.9.36.194.vlar
Thu 25 Feb 2016 01:33:50 PM PST | SETI@home | Started download of 20dc10ac.17853.6639.9.36.200.vlar
Thu 25 Feb 2016 01:33:51 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:33:52 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:34:08 PM PST | SETI@home | Finished download of 20dc10ac.17853.6639.9.36.194.vlar
Thu 25 Feb 2016 01:34:08 PM PST | SETI@home | Started download of 20dc10ac.17853.6639.9.36.205.vlar
Thu 25 Feb 2016 01:34:09 PM PST | SETI@home | Finished download of 20dc10ac.17853.6639.9.36.200.vlar
Thu 25 Feb 2016 01:34:09 PM PST | SETI@home | Started download of 20dc10ac.17853.6639.9.36.225.vlar
Thu 25 Feb 2016 01:34:26 PM PST | SETI@home | Finished download of 20dc10ac.17853.6639.9.36.205.vlar
Thu 25 Feb 2016 01:34:26 PM PST | SETI@home | Started download of 20dc10ac.17853.6639.9.36.211.vlar
Thu 25 Feb 2016 01:34:27 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:34:28 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:34:28 PM PST | SETI@home | Finished download of 20dc10ac.17853.6639.9.36.225.vlar
Thu 25 Feb 2016 01:34:28 PM PST | SETI@home | Started download of 20dc10ac.17853.6639.9.36.217.vlar
Thu 25 Feb 2016 01:34:30 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:34:31 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:34:32 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:34:45 PM PST | SETI@home | Finished download of 20dc10ac.17853.6639.9.36.211.vlar
Thu 25 Feb 2016 01:34:45 PM PST | SETI@home | Started download of 20dc10ac.18940.5412.10.37.241.vlar
Thu 25 Feb 2016 01:34:47 PM PST | SETI@home | Finished download of 20dc10ac.17853.6639.9.36.217.vlar
Thu 25 Feb 2016 01:34:47 PM PST | SETI@home | Started download of 20dc10ac.17853.6639.9.36.223.vlar
Thu 25 Feb 2016 01:35:04 PM PST | SETI@home | Finished download of 20dc10ac.18940.5412.10.37.241.vlar
Thu 25 Feb 2016 01:35:04 PM PST | SETI@home | Started download of 20dc10ac.17122.18909.7.34.5.vlar
Thu 25 Feb 2016 01:35:05 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:35:06 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:35:06 PM PST | SETI@home | Finished download of 20dc10ac.17853.6639.9.36.223.vlar
Thu 25 Feb 2016 01:35:06 PM PST | SETI@home | Started download of 20dc10ac.17853.6639.9.36.233.vlar
Thu 25 Feb 2016 01:35:07 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:35:08 PM PST | SETI@home | task postponed 30.000000 sec:
Thu 25 Feb 2016 01:35:09 PM PST | SETI@home | task postponed 30.000000 sec:


If I suspect E@H project, these AP tasks get scheduled to run, but instantly go back to "waiting to run" status.

Strange, but, okay. These tasks aren't failing, so that's probably a good sign, at least. If it would be useful for me to turn on some more debugging, just let me know.
ID: 1767583 · Report as offensive
Juha
Volunteer tester

Send message
Joined: 7 Mar 04
Posts: 388
Credit: 1,857,738
RAC: 0
Finland
Message 1767594 - Posted: 25 Feb 2016, 22:35:03 UTC - in response to Message 1765887.  

Thanks for the coproc_info.xml file.

Right now you don't have any reported AstroPulse tasks but when I looked earlier all the tasks had failed. The reason may be an old bug in BOINC API where the app can't handle multiple OpenCL platforms. You could try removing pocl to see if it's the API bug or something else.

To try running MB v8 tasks via anonymous platform mechanism download the following files and put them in projects/setiathome.berkeley.edu . Remember to mark the program file executable.

http://boinc2.ssl.berkeley.edu/beta/download/setiathome_8.04_x86_64-pc-linux-gnu__opencl_ati5_sah
http://boinc2.ssl.berkeley.edu/beta/download/MultiBeam_Kernels_r3316.cl

Then create file projects/setiathome.berkeley.edu/app_info.xml with the following content.

<app_info> 
    <app>
        <name>setiathome_v8</name>
    </app>
    <file_info>
        <name>setiathome_8.04_x86_64-pc-linux-gnu__opencl_ati5_sah</name>
        <executable/>
    </file_info>
    <app_version>
        <app_name>setiathome_v8</app_name>
        <version_num>804</version_num>
        <platform>x86_64-pc-linux-gnu</platform>
        <avg_ncpus>0.04</avg_ncpus>
        <max_ncpus>0.2</max_ncpus>
        <plan_class>opencl_ati5_sah</plan_class>
        <cmdline></cmdline>     
        <coproc>
            <type>ATI</type>
            <count>1</count>
        </coproc>
        <file_ref>
            <file_name>setiathome_8.04_x86_64-pc-linux-gnu__opencl_ati5_sah</file_name>
            <main_program/>
        </file_ref>
    </app_version>
</app_info>


Note that this file contains application for MB v8 only and only for ATI GPU. If you have anything else from Seti cached BOINC will trash it. You'll want to run the cache down first. It is also completely untested. Set your cache to minimum, otherwise you risk blowing your entire daily quota in an instant.

You need to restart the client for it to see changes in app_info.xml . If/when you want to go back to stock apps remove app_info.xml file and restart the client.
ID: 1767594 · Report as offensive
Juha
Volunteer tester

Send message
Joined: 7 Mar 04
Posts: 388
Credit: 1,857,738
RAC: 0
Finland
Message 1767596 - Posted: 25 Feb 2016, 22:39:00 UTC - in response to Message 1767583.  

Oh, you posted while I was typing my post.

Strange, but, okay. These tasks aren't failing, so that's probably a good sign, at least.


Well, not good. See the previous post.
ID: 1767596 · Report as offensive
Paul

Send message
Joined: 17 May 99
Posts: 72
Credit: 42,977,964
RAC: 43
United States
Message 1767675 - Posted: 26 Feb 2016, 4:42:00 UTC

Hi Juha!

Thanks for the advice! I will try anon ASAP. I reduced my cache size in preparation.

As for removing POCL, I did that, and now I get a reason for the suspended applications. At least, I think these messages are related to that, specifically:

Thu 25 Feb 2016 08:27:44 PM PST | SETI@home | Task postponed: CL file build failure
Thu 25 Feb 2016 08:27:45 PM PST | SETI@home | Task postponed: CL file build failure
Thu 25 Feb 2016 08:27:46 PM PST | SETI@home | Task postponed: CL file build failure
Thu 25 Feb 2016 08:27:47 PM PST | SETI@home | Task postponed: CL file build failure
Thu 25 Feb 2016 08:27:48 PM PST | SETI@home | Task postponed: CL file build failure

ID: 1767675 · Report as offensive
Juha
Volunteer tester

Send message
Joined: 7 Mar 04
Posts: 388
Credit: 1,857,738
RAC: 0
Finland
Message 1767903 - Posted: 26 Feb 2016, 22:00:48 UTC - in response to Message 1767675.  

CL build failure could be similar to what Einstein BRP had, Mesa/libclc missing an OpenCL function it should have had. You could peek into the stderr.txt file in one of the slot directories to see if there is a better explanation.

(Mesa/libclc claims to implement OpenCL 1.1. It was missing modf function which, as far as I can tell, was already in OpenCL 1.0. ??? )
ID: 1767903 · Report as offensive
Paul

Send message
Joined: 17 May 99
Posts: 72
Credit: 42,977,964
RAC: 43
United States
Message 1768030 - Posted: 27 Feb 2016, 8:09:55 UTC

Cool, thanks yet again.

I found a really large stderr.txt file that seemed to be from one of these SETI OCL apps. At the top, it was complaining about POCL. But later, the output was very different. I assume that change occured when I uninstalled pocl. The following is a copy of the last set of lines in the file, which repeat many times before it:

...
OpenCL platform detected: Mesa
Number of OpenCL devices found : 2 
BOINC assigns slot on device #0.
Info: BOINC provided OpenCL device ID used
Used GPU device parameters are:
	Number of compute units: 20
	Single buffer allocation size: 256MB
	Total device global memory: 1024MB
	max WG size: 256
	local mem type: Real
	-unroll default value used: 18
	-ffa_block default value used: 5120
	-ffa_block_fetch default value used: 2560
AstroPulse v7.08
Linux 64 bit, Rev 2751, OpenCL version by Raistmer, GPU mode
 V7, by Raistmer ported to Linux   by Lunatics.kwsn.net team. 
oclFFT fix for ATI GPUs by Urs Echternacht
ffa threshold mods by Joe Segur
SSE3 dechirping by JDWhale using SSE3 emulation

Build features: Non-graphics OpenCL OCL_ZERO_COPY OPENCL_WRITE COMBINED_DECHIRP_KERNEL SMALL_CHIRP_TABLE TWIN_FFA FFTW BLANKIT USE_INCREASED_PRECISION SSE2 64bit 
 System: Linux  x86_64  Kernel: 4.3.5-300.fc23.x86_64
 CPU   : AMD FX(tm)-8350 Eight-Core Processor
 8 core(s), Speed :  4013.815 MHz
 L1 : 64 KB, Cache : 2048 KB

Number of OpenCL platforms:				 1


 OpenCL Platform Name:					 Clover
Number of devices:				 2
  Max compute units:				 20
  Max work group size:				 256
  Max clock frequency:				 1050Mhz
  Max memory allocation:			 268435456
  Cache type:					 None
  Cache line size:				 0
  Cache size:					 0
  Global memory size:				 1073741824
  Constant buffer size:				 268435456
  Max number of constant args:			 16
  Local memory type:				 Scratchpad
  Local memory size:				 32768
  Queue properties:				 
    Out-of-Order:				 No
  Name:						 AMD PITCAIRN (DRM 2.43.0, LLVM 3.7.0)
  Vendor:					 AMD
  Driver version:				 11.1.0
  Version:					 OpenCL 1.1 MESA 11.1.0
  Extensions:					 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_fp64
  Max compute units:				 20
  Max work group size:				 256
  Max clock frequency:				 1050Mhz
  Max memory allocation:			 268435456
  Cache type:					 None
  Cache line size:				 0
  Cache size:					 0
  Global memory size:				 1073741824
  Constant buffer size:				 268435456
  Max number of constant args:			 16
  Local memory type:				 Scratchpad
  Local memory size:				 32768
  Queue properties:				 
    Out-of-Order:				 No
  Name:						 AMD PITCAIRN (DRM 2.43.0, LLVM 3.7.0)
  Vendor:					 AMD
  Driver version:				 11.1.0
  Version:					 OpenCL 1.1 MESA 11.1.0
  Extensions:					 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_fp64


INFO: can't open binary kernel file: /home/pdestefa/local/BOINC/projects/setiathome.berkeley.edu/AstroPulse_Kernels_r2751.cl_AMDPITCAIRNDRM2430LLVM370.bin_V7_TWIN_FFA_1110, continue with recompile...
Error : Building Program (binary, clBuildProgram):main kernels: not OK code -43
CL file build log on device AMD PITCAIRN (DRM 2.43.0, LLVM 3.7.0)


I don't see any error messages that are useful to me.

Hmm...When I uninstalled POCL, it also uninstalled llvm (and some other stuff). Does BOINC need llvm installed to do this compilation? ...Hmm, no, I just installed llvm and that didn't help when BOINC tried to run these tasks again by itself. I wonder if one of the other packages is also needed. Mesa doesn't depend on them, so the package manager just removed them, but they were all OCL related.

...Nope, I reinstalled everything except pocl and it didn't change anything. I don't need to restart BOINC, do I?
ID: 1768030 · Report as offensive
Juha
Volunteer tester

Send message
Joined: 7 Mar 04
Posts: 388
Credit: 1,857,738
RAC: 0
Finland
Message 1768204 - Posted: 27 Feb 2016, 21:59:25 UTC - in response to Message 1768030.  

Error : Building Program (binary, clBuildProgram):main kernels: not OK code -43
CL file build log on device AMD PITCAIRN (DRM 2.43.0, LLVM 3.7.0)


-43 is CL_INVALID_BUILD_OPTIONS. The compiler is probably choking on some ATI specific compiler option. Too bad it doesn't actually include the build log even though it says so.

As the app isn't going to work I suppose you could just abort the AstroPulse tasks and move to anon platform to test the Multibeam app. I'm a bit afraid you'll have similar results with it.

To tell the truth I would edit the program file with hex editor and blank the bad compiler options just to see how far I can push it. That would make it necessary to switch to anon platform. BOINC doesn't like when files it has downloaded are changed.


Mesa uses LLVM and libclc uses Clang but those would be in library form. I think we can trust the packager to have done his/her job right and you have all the necessary dependencies installed.
ID: 1768204 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : BOINC 7.6.23 for Linux Recongnizes OpenCL AMD/ATi GPU via MESA, But No Apps From SETI?


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.