Re: Looking for '_source' kernel
by Andreas Kloeckner
Shivangi Agarwal <shivangiagarwal53(a)gmail.com> writes:
> No. It is the same error as before. I think I didn't look deep enough the
> first time. kernel_old_init is taking a name as an arg:
>
> -> kernel_old_init(self, prg, name)
> (Pdb) a
> self = <pyopencl._cl.Kernel object at 0x7f4632c17b30>
> prg = <pyopencl._cl._Program object at 0x7f463f737470>
> name = _source
>
> I tried running the program in gdb and it gave me "No stack". Is there any
> other way I can provide the bt?
I meant the Python backtrace, i.e. what Python prints if a program ends
with an error.
Andreas
10 months, 2 weeks
Looking for '_source' kernel
by Shivangi Agarwal
Hi,
I wrote a simple hello world pyocl testcase for a Xilinx accelerator card,
but I am getting kernel '_source' not found error. I stepped through the
code and it seems like the pyopencl code is trying to create a '_source'
kernel. Any ideas on how I can work around this?
“””
*-> prg.hello(commands, (1, ), (1, ), d_buf)*
*(Pdb) s*
*--Call--*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(434)__getattr__()*
*-> def __getattr__(self, attr):*
*(Pdb) n*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(435)__getattr__()*
*-> try:*
*(Pdb)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(436)__getattr__()*
*-> knl = Kernel(self, attr)*
*(Pdb)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(439)__getattr__()*
*-> knl.num_args*
*(Pdb)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(440)__getattr__()*
*-> knl._source = getattr(self, "_source", None)*
*(Pdb) a*
*self = <pyopencl.Program object at 0x7f5c52c66e90>*
*attr = hello*
*(Pdb) s*
*--Call--*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(434)__getattr__()*
*-> def __getattr__(self, attr):*
*(Pdb) n*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(435)__getattr__()*
*-> try:*
*(Pdb)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(436)__getattr__()*
*-> knl = Kernel(self, attr)*
*(Pdb) a*
*self = <pyopencl.Program object at 0x7f5c52c66e90>*
*attr = _source*
*(Pdb) n*
*[XRT] ERROR: kernel '_source' not found*
*LogicError: <pyopenc...5280b110>*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(436)__getattr__()*
*-> knl = Kernel(self, attr)*
*(Pdb) n*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(449)__getattr__()*
*-> except LogicError:*
*(Pdb)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(450)__getattr__()*
*-> raise AttributeError("'%s' was not found as a program "*
*(Pdb)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(451)__getattr__()*
*-> "info attribute or as a kernel name" % attr)*
*(Pdb)*
*AttributeError: Attribut...l name",)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(451)__getattr__()*
*-> "info attribute or as a kernel name" % attr)*
*(Pdb)*
*--Return--*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(451)__getattr__()->None*
*-> "info attribute or as a kernel name" % attr)*
*(Pdb)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(442)__getattr__()*
*-> if self._build_duration_info is not None:*
*(Pdb)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(443)__getattr__()*
*-> build_descr, was_cached, duration = self._build_duration_info*
*(Pdb)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(444)__getattr__()*
*-> if duration > 0.2:*
*(Pdb)*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(448)__getattr__()*
*-> return knl*
*(Pdb)*
*--Return--*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(448)__getattr__()-><pyopenc...52c90b90>*
*-> return knl*
*(Pdb)*
*--Call--*
*>
/usr/local/lib/python2.7/dist-packages/pyopencl/__init__.py(837)kernel_call()*
*-> def kernel_call(self, queue, global_size, local_size, *args, **kwargs):*
*(Pdb) c*
*Result: Hello World*
*“””*
Thanks,
Shivangi
10 months, 2 weeks
Memory Profiling for PyOpenCl based program
by MANISH AGRAWAL
Hello,
I am trying to implement electromagnetic simulations using PyOpenCl. The
memory that is used by the program varies and depends on user inputs. I was
looking for a way of profiling memory usage using PyOpenCl. There are tools
for querying the memory status for CUDA enabled devices but I am looking
for something which can be opted in general (AMD gpus or CPUs too).
Thanks & Regards
*Manish Agrawal*
Final Year Undergraduate
IIT Kharagpur, India
Contact : +91-8232993930 | LinkedIn <http://linkedin.com/manishhag>
10 months, 2 weeks
