Issues with make check.

[thyrgle]

Hi,

When I run make check as:

HARLAN_DEVICE=gpu make check

I get:

Some tests failed:
test/raytrace.kfc
test/kernel-even-odd.kfc

(Outputs at bottom)

I am using:

Mac OS X 10.9.2
2.7 GHz Intel Core i7
Intel HD Graphics 4000 1024 MB

if that matters.

Any ideas or suggestions?

Thanks.

(raytrace.kfc.out)
OpenCL Platform Name: Apple
OpenCL Platform Vendor: Apple
OpenCL Platform Version: OpenCL 1.2 (Dec 8 2013 21:07:05)
found 3 devices
Creating queue for Intel(R) Core(TM) i7-3740QM CPU @ 2.70GHz
OpenCL Error: OpenCL Build Warning : Compiler build log:
<program source>:228:19: warning: no previous prototype for function 'miss'
ray$dresult_24388 miss() {
^
<program source>:257:13: warning: unused variable 'tag_25252'
int tag_25252 = extract_tag(m_25253);
^
<program source>:261:13: warning: unused variable 'tag_25250'
int tag_25250 = extract_tag(m_25251);
^
<program source>:273:13: warning: unused variable 'tag_25248'
int tag_25248 = extract_tag(m_25249);
^
<program source>:277:13: warning: unused variable 'tag_25246'
int tag_25246 = extract_tag(m_25247);
^
<program source>:289:13: warning: unused variable 'tag_25244'
int tag_25244 = extract_tag(m_25245);
^
<program source>:293:13: warning: unused variable 'tag_25242'
int tag_25242 = extract_tag(m_25243);
^
<program source>:304:13: warning: unused variable 'tag_25240'
int tag_25240 = extract_tag(m_25241);
^
<program source>:315:13: warning: unused variable 'tag_25238'
int tag_25238 = extract_tag(m_25239);
^
<program source>:334:13: warning: unused variable 'tag_25236'
int tag_25236 = extract_tag(m_25237);
^
<program source>:361:42: warning: unused variable '__24349_24432'
point3f$dt_24389 __24349_24432 = m_25233.data.hit.f1;
^
<program source>:352:34: warning: unused variable '__24347_24430'
point3f$dt_24389 __24347_24430 = m_25235.data.hit.f1;
^
<program source>:435:25: warning: expression result unused
42;
^~
<program source>:578:13: warning: unused variable 'tag_25214'
int tag_25214 = extract_tag(m_25215);
^
<program source>:654:36: warning: unused variable 'ktemp_25290'
region_ptr ktemp_25290 = x_25332;
^
<program source>:706:31: warning: unused variable 'dist_24383_24424'
float dist_24383_24424 = m_25231.data.hit.f0;
^
<program source>:607:20: warning: unused variable 'ktemp_25291'
region_ptr ktemp_25291 = x_25338;
^

(point3f 0.4082482755 0.4082482755 0.816496551)
(point3f 0.4082482755 0.4082482755 0.816496551)
(point3f 0.4082482755 0.4082482755 0.816496551)
JITTIME: 0.002417804

(kernel-even-odd.kfc.out)
(OpenCL Platform Name: Apple
OpenCL Platform Vendor: Apple
OpenCL Platform Version: OpenCL 1.2 (Dec 8 2013 21:07:05)
found 3 devices
Creating queue for Intel(R) Core(TM) i7-3740QM CPU @ 2.70GHz
OpenCL Error: OpenCL Build Warning : Compiler build log:
<program source>:144:25: warning: expression result unused
42;
^~
<program source>:301:25: warning: expression result unused
42;
^~

[1 0 1 0 1]
JITTIME: 0.000101291

[Eric Holk]

Hi,

I've been seeing similar issues on my end as well. It depends a lot on which OpenCL implementation you're using. It seems that Harlan generates some code that crashes some OpenCL compilers. Does it work any better if you use HARLAN_DEVICE=cpu?

Actually, it looks like the output you posted is using the CPU device (since it says "Creating queue for Intel(R) Core(TM) i7-3740QM CPU @ 2.70GHz"), and the program seems to be running normally. What if you directly run the test.bin/kernel-even-odd.kfc.bin program, which HARLAN_DEVICE=gpu?

Feel free to ignore these failures for now and keep playing around with the language.

Thanks for trying it out!

-Eric

[Christopher Sumnicht]

Hi,

Running with HARLAN_DEVICE=cpu gives:

Some tests failed:

    test/nested-kernels.kfc

    test/dmm.kfc

    test/dmm-hand-optimized.kfc

Running kernel-even-odd independently gives:

(HARLAN_DEVICE=gpu ./kernel-even-odd.kfc.bin)

OpenCL Platform Name:    Apple

OpenCL Platform Vendor:  Apple

OpenCL Platform Version: OpenCL 1.2 (Dec  8 2013 21:07:05)

found 2 devices

Creating queue for HD Graphics 4000

[1]    3563 segmentation fault  HARLAN_DEVICE=gpu ./kernel-even-odd.kfc.bin

Thanks for the reply. 

[Eric Holk]

Thanks for the extra information.

I have access to a couple of machines that are showing similar failures, so I will investigate further here.

-Eric

--

You received this message because you are subscribed to the Google Groups "Harlan Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to harlan-dev+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Eric Holk]

Cristopher,

I've done a little more investigation and I made some changes to Harlan that might help. If you try it with the latest version, do you see any difference?

On my Mac, the three failures you saw on the CPU are gone, but I added a new test that still breaks.

-Eric

On Monday, May 5, 2014 2:23:47 PM UTC-4, Eric Holk wrote:

Thanks for the extra information.

I have access to a couple of machines that are showing similar failures, so I will investigate further here.

-Eric

On Fri, May 2, 2014 at 9:30 PM, Christopher Sumnicht <extremeobjcprogrammer@gmail.com> wrote:

Hi,

Running with HARLAN_DEVICE=cpu gives:

Some tests failed:

    test/nested-kernels.kfc

    test/dmm.kfc

    test/dmm-hand-optimized.kfc

Running kernel-even-odd independently gives:

(HARLAN_DEVICE=gpu ./kernel-even-odd.kfc.bin)

OpenCL Platform Name:    Apple

OpenCL Platform Vendor:  Apple

OpenCL Platform Version: OpenCL 1.2 (Dec  8 2013 21:07:05)

found 2 devices

Creating queue for HD Graphics 4000

[1]    3563 segmentation fault  HARLAN_DEVICE=gpu ./kernel-even-odd.kfc.bin

Thanks for the reply. 

--

You received this message because you are subscribed to the Google Groups "Harlan Developers" group.

To unsubscribe from this group and stop receiving emails from it, send an email to harlan-dev+unsubscribe@googlegroups.com.

[Сергей Антоненко]

I have same tests failed on latest version

Some tests failed:

    test/raytrace.kfc

    test/kernel-even-odd.kfc

sh: line 1: 95513 Segmentation fault: 11  ./test.bin/raytrace.kfc.bin >> ./test.bin/raytrace.kfc.out 2>&1

OpenCL Platform Name:    Apple

OpenCL Platform Vendor:  Apple

OpenCL Platform Version: OpenCL 1.2 (Dec  8 2013 21:07:05)

found 1 devices

Creating queue for Iris Pro

Running test...FAILED

пятница, 2 мая 2014 г., 13:07:19 UTC+7 пользователь thyrgle написал:

[Eric Holk]

Is there anything in ./test.bin/raytrace.kfc.out or ./test.bin/kernel-even-odd.kfc? If so, would you sharing the contents?

If not, could you run ./test.bin/raytrace.kfc.bin and test.bin/kernel-even-odd.kfc and share the output?

If you set the environment variable HARLAN_DEVICE=cpu do you see something different? Is the Iris Pro the only GPU in your machine?

Thanks,

Eric

--

You received this message because you are subscribed to the Google Groups "Harlan Developers" group.

To unsubscribe from this group and stop receiving emails from it, send an email to harlan-dev+...@googlegroups.com.

[Сергей Антоненко]

Thanks for your reply.

$ cat ./test.bin/raytrace.kfc.out

OpenCL Platform Name:    Apple

OpenCL Platform Vendor:  Apple

OpenCL Platform Version: OpenCL 1.2 (Dec  8 2013 21:07:05)

found 1 devices

Creating queue for Iris Pro

$ cat test.bin/kernel-even-odd.kfc.out

OpenCL Platform Name:    Apple

OpenCL Platform Vendor:  Apple

OpenCL Platform Version: OpenCL 1.2 (Dec  8 2013 21:07:05)

found 1 devices

Creating queue for Iris Pro

> If you set the environment variable HARLAN_DEVICE=cpu do you see something different? Is the Iris Pro the only GPU in your machine?

Yes, it fails differently:

Some tests failed:

    test/nested-kernels2.kfc

Successes: 146; Failures: 1; Ignored: 27; Total: 174

$ cat test.bin//nested-kernels2.kfc.out 

OpenCL Platform Name:    Apple

OpenCL Platform Vendor:  Apple

OpenCL Platform Version: OpenCL 1.2 (Dec  8 2013 21:07:05)

found 1 devices

Creating queue for Intel(R) Core(TM) i7-4750HQ CPU @ 2.00GHz

OpenCL Error: [CL_DEVICE_NOT_AVAILABLE] : OpenCL Error : Error: build program driver returned (-2)

OpenCL Error: OpenCL Warning : clBuildProgram failed: could not build program for 0xffffffff (Intel(R) Core(TM) i7-4750HQ CPU @ 2.00GHz) (err:-2)

OpenCL Error: [CL_BUILD_ERROR] : OpenCL Build Error : Compiler build log:

CVMS_ERROR_COMPILER_FAILURE: CVMS compiler has crashed or hung building an element.

CVMS_ERROR_COMPILER_FAILURE: CVMS compiler has crashed or hung building an element.

clBuildProgram failed with error CL_BUILD_PROGRAM_FAILURE (-11) in cl++.cpp:227

--
You received this message because you are subscribed to a topic in the Google Groups "Harlan Developers" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/harlan-dev/6o8xQoqjlG0/unsubscribe.
To unsubscribe from this group and all its topics, send an email to harlan-dev+...@googlegroups.com.

[Eric Holk]

Thanks!

The nested-kernels2 failure seems to be due to a bug in Apple's OpenCL implementation. I've opened a bug with them.

I suspect a similar thing is happening with the first two failures. I don't know if I have access to an Iris Pro machine to try to reduce the test cases, but it might be the same bug. I've been having trouble with Intel OpenCLs lately.

-Eric

[Eric Holk]

It'd be worth updating to Mavericks 10.9.3. It looks like one of the OpenCL bugs I had found there was fixed. Maybe that fixed some of your issues as well.

[Сергей Антоненко]

I've updated OS X up to 10.9.3 so as everything that were updatable. Still same tests are failing with same errors. Don't know, what other versions can I check? Or would you plz refer me to git commit supposed to fix this problems, so I could try to do my best figuring that out?

Thank you.

[Eric Holk]

The latest Harlan commit is 07b473bbdef1b0f1f6bac29d6ca232d15725e817, which has all the fixes I've made.

These issues are likely due to bugs in Apple's OpenCL, since I do not see them on other platforms. I've got a few failing tests on my 10.9.3 machine that I'm going to try to reduce down to OpenCL test cases and submit them to Apple if I cannot see a problem on Harlan's side.

In the meantime, you could try setting the HARLAN_DISABLE_OPENCL_OPTS environment variable to 1 and see if that helps. This turns off the OpenCL optimizer, which will probably hurt your performance but may avoid OpenCL bugs.

If you have access to a Linux machine, you might try there. Intel, NVIDIA and AMD all have OpenCL's that work for me.

Finally, since this is only a few cases that fail, you should be fine to just ignore these for now and keep playing around with Harlan.

-Eric

To unsubscribe from this group and stop receiving emails from it, send an email to harlan-dev+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to a topic in the Google Groups "Harlan Developers" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/harlan-dev/6o8xQoqjlG0/unsubscribe.

To unsubscribe from this group and all its topics, send an email to harlan-dev+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Harlan Developers" group.

To unsubscribe from this group and stop receiving emails from it, send an email to harlan-dev+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to a topic in the Google Groups "Harlan Developers" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/harlan-dev/6o8xQoqjlG0/unsubscribe.

To unsubscribe from this group and all its topics, send an email to harlan-dev+unsubscribe@googlegroups.com.