the "General Settings" you are showing seem to be alright.

But these debug error dialogs / messages CodeXL throws  can be quite
deceiving....dont pay too much attention to em.

Were hard to believe a DLL is misplaced or even Windows 8 resp. U14.04 are
substiantially unsupportive. Prob. probably resides elsewhere.


1. which CPU / GPU do you plan on using ?

2. pls. add a screenshot of your CodeXL project settings showing whats
selected under "analyze -> Kernel/Shader Build Options". Practically you'ld
end up with No_Optimization + Debug, i.e. -O0 -g

3. under "Profile -> GPU Profile: Perf. Counters" select the device you
want to debug on.

4. Are you using some printf extension from within kernels ? If yes,
intentionally add "#pragma OPENCL EXTENSION cl_amd_printf: enable" in order
to keep the OCL compiler from doing something odd instead, which works nice
and easy for standalone / commandline calls of python & pyopencl but
mysteriously makes CodeXL fail if initiating the debug run.

5. Additionally, add following three lines to the import section of your
pyopencl script:

import os
os.environ['PYOPENCL_COMPILER_OUTPUT'] = '1'     # rather cool but important
os.environ['PYOPENCL_NO_CACHE'] = '1'            # obsoletes relics
which can negatively impact CodeXL

6. double check when doing "device = platform.get_devices()[0]" that you
indeed get the right device and this device is GPU-debug capable.
Check " https://en.wikipedia.org/wiki/AMD_Radeon_Rx_200_series ", seek the
table and search your device. You're off best using cards featuring GCN1.1
or newer.


Pls. feed back if / what measure did the trick for you.

Otherwise, strip your code down to a manageable portion still throwing that
exception and mail that snippet for me to try out on my machine.

Nice day !
Kai


On Fri, Nov 20, 2015 at 7:53 AM, Yu Yang <[email protected]> wrote:

> I've tried the method you given in windows8.1 and ubuntu 14.04, but no
> luck,
>
> ​when run in cmdline, it works, it means APPSDK,python,pyopencl are
> installed correctly, but when use CodeXL, it gives me an warning, and I
> can't step into the kernel
>
> ​this is my CodeXL settings
>
> ​Any advice? thx
>
> 2015-11-19 17:27 GMT+08:00 Dude Dudish <[email protected]>:
>
>> Yu, All,
>> yes, there's a way.
>> CodeXL (AMD) tool is capable of debugging single-step source-level wise
>> through ur kernel code at runtime, however, while working relatively
>> straightforward for CPUs certain prerequisites apply to cases of using GPUs
>> (primarily regarding the HW).
>>
>> Pls check out Link AND ATTACHMENT (pdf) on that site:
>>
>> http://wiki.tiker.net/PyOpenCL/Debugging
>>
>> I uploaded text and pdf including the screenshot and config instructions
>> lil while ago.
>> Joyfully looking forward to some other user-experiences.... :-)
>>
>> BR from Frankfurt,
>> Kai
>>
>> On Thu, Nov 19, 2015 at 9:29 AM, Yu Yang <[email protected]> wrote:
>>
>>> PyOpenCL is very convenient, but it seems that CodeXL can't debug
>>> PyOpenCL project, any solutions? thx
>>>
>>> _______________________________________________
>>> PyOpenCL mailing list
>>> [email protected]
>>> http://lists.tiker.net/listinfo/pyopencl
>>>
>>>
>>
>
_______________________________________________
PyOpenCL mailing list
[email protected]
http://lists.tiker.net/listinfo/pyopencl

Reply via email to