Support for (integrated) Intel GPUs (Ivy Bridge and later)

Raistmer*
Raistmer*
Joined: 20 Feb 05
Posts: 208
Credit: 181407464
RAC: 8747

Thanks, command line was

Thanks, command line was crucial for success.
Now will try to run under profiler.

Raistmer*
Raistmer*
Joined: 20 Feb 05
Posts: 208
Credit: 181407464
RAC: 8747

It just consumes full CPU

It just consumes full CPU core under CodeXL GPU profiler and makes no progress. Even stderr.txt didn't created. will leave it running through nigh but most probably no profile results will be gathered.

Oliver Behnke
Oliver Behnke
Moderator
Administrator
Joined: 4 Sep 07
Posts: 984
Credit: 25171376
RAC: 43

RE: Can't run offline, app

Quote:

Can't run offline, app exits after few seconds...

Should I supply some command line options for it?


Well, yes, of course :-) That's why I mentioned it earlier. Thanks Richard for stepping in!

Quote:

Thanks, command line was crucial for success.


So you can confirm you could run it from start to finish on the machine you're going to use for profiling? stderr.txt and results.cand* do exist at the end of the run?

Quote:

It just consumes full CPU core under CodeXL GPU profiler and makes no progress.


Hm, not sure how CodeXL instruments an app in order to profile it...

Oliver

Einstein@Home Project

Oliver Behnke
Oliver Behnke
Moderator
Administrator
Joined: 4 Sep 07
Posts: 984
Credit: 25171376
RAC: 43

RE: [19:54:38][3092][ERROR

Quote:

[19:54:38][3092][ERROR] Error in OpenCL context: Out of device memory.
[19:54:38][3092][ERROR] Error in OpenCL context: Out of device memory.

(and is still running). Pretty much the same as the live result for task 406867971.

Now that's an interesting feature: the was an error but the app finished anyway? Do you see these regularly and, if so, do the affected tasks validate?

Update: this "error" is thrown by the OpenCL runtime (via a callback) during context creation. Interestingly enough though, the call to create the context itself doesn't return an error. Doesn't seem to matter...

Best,
Oliver

Einstein@Home Project

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2143
Credit: 2955539861
RAC: 717388

RE: RE: [19:54:38][3092][

Quote:
Quote:
[19:54:38][3092][ERROR] Error in OpenCL context: Out of device memory.
[19:54:38][3092][ERROR] Error in OpenCL context: Out of device memory.

(and is still running). Pretty much the same as the live result for task 406867971.

Now that's an interesting feature: the was an error but the app finished anyway? Do you see these regularly and, if so, do the affected tasks validate?

Update: this "error" is thrown by the OpenCL runtime (via a callback) during context creation. Interestingly enough though, the call to create the context itself doesn't return an error. Doesn't seem to matter...

Best,
Oliver


Yes, spot-check any you like from

Valid tasks for computer 5744895
Valid tasks for computer 8864187

The OpenCL error is reported in every task I've looked at, but the overall failure rate (either error or invalid) is extraordinarily low - in fact, the Haswell (8864187) is showing no failures at all - the overclocked i7 is a little more temperamental.

Holmis
Joined: 4 Jan 05
Posts: 1118
Credit: 1055935564
RAC: 0

RE: [19:54:38][3092][ERROR]

Quote:
[19:54:38][3092][ERROR] Error in OpenCL context: Out of device memory.
[19:54:38][3092][ERROR] Error in OpenCL context: Out of device memory.


The first mention of this OpenCL error was during testing over at Albert@home in this message by Bikeman. It was in response to me posting a stderr.txt from the first test of the app outside the lab. So it's been there from the start.

Raistmer*
Raistmer*
Joined: 20 Feb 05
Posts: 208
Credit: 181407464
RAC: 8747

RE: So you can confirm you

Quote:


So you can confirm you could run it from start to finish on the machine you're going to use for profiling? stderr.txt and results.cand* do exist at the end of the run?

Hm, not sure how CodeXL instruments an app in order to profile it...

Oliver

1) didn't run it from start to finish, just checked that it starts OK and writes into stderr.txt about it.
Running under profiler didn't result even in creating stderr.txt at all.

2) No additional instrumentation on compilation/linkage stages required.
So I used stock binary downloaded from project.

PS. Here is thread on AMD CodeXL forum about this issue, for reference.
http://devgurus.amd.com/thread/167648
There are similar issues in other threads on that forum too. I found no solution so far.

PPS: threads with similar issues (app hangs on start):
http://devgurus.amd.com/message/1300495#1300495
http://devgurus.amd.com/message/1295856#1295856
http://devgurus.amd.com/message/1287984#1287984

Last one looks especially similar, some conversation with AMD staff... but no useful answer still.

Oliver Behnke
Oliver Behnke
Moderator
Administrator
Joined: 4 Sep 07
Posts: 984
Credit: 25171376
RAC: 43

Just a rough guess, but maybe

Just a rough guess, but maybe CodeXL can't handle cross-compiled MinGW binaries and expects MS builds...?

Oliver

Einstein@Home Project

Oliver Behnke
Oliver Behnke
Moderator
Administrator
Joined: 4 Sep 07
Posts: 984
Credit: 25171376
RAC: 43

RE: http://devgurus.amd.com

Funny, I know Gergely and could ask him whether he could solve his issue. Are the symptoms the same in your case or did you indeed just list "similar" ones...?

Einstein@Home Project

Raistmer*
Raistmer*
Joined: 20 Feb 05
Posts: 208
Credit: 181407464
RAC: 8747

RE: Just a rough guess, but

Quote:

Just a rough guess, but maybe CodeXL can't handle cross-compiled MinGW binaries and expects MS builds...?

Oliver


Viable guess. I had issues with AMD profilers (non-solvable) when used ICC for MultiBeam ATi GPU app building. So far only MSVC-compiled binary was able to be profiled...

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.