[petsc-users] Questions about setting values for GPU based matrices

[Fredrik Heffer Valdmanis]

2011/11/29 Matthew Knepley <knepley at gmail.com>

> On Tue, Nov 29, 2011 at 10:37 AM, Fredrik Heffer Valdmanis <
> fredva at ifi.uio.no> wrote:
>
>> 2011/11/29 Matthew Knepley <knepley at gmail.com>
>>
>>> On Tue, Nov 29, 2011 at 2:38 AM, Fredrik Heffer Valdmanis <
>>> fredva at ifi.uio.no> wrote:
>>>
>>>> 2011/10/28 Matthew Knepley <knepley at gmail.com>
>>>>
>>>>> On Fri, Oct 28, 2011 at 10:24 AM, Fredrik Heffer Valdmanis <
>>>>> fredva at ifi.uio.no> wrote:
>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> I am working on integrating the new GPU based vectors and matrices
>>>>>> into FEniCS. Now, I'm looking at the possibility for getting some speedup
>>>>>> during finite element assembly, specifically when inserting the local
>>>>>> element matrix into the global element matrix. In that regard, I have a few
>>>>>> questions I hope you can help me out with:
>>>>>>
>>>>>> - When calling MatSetValues with a MATSEQAIJCUSP matrix as parameter,
>>>>>> what exactly is it that happens? As far as I can see, MatSetValues is not
>>>>>> implemented for GPU based matrices, neither is the mat->ops->setvalues set
>>>>>> to point at any function for this Mat type.
>>>>>>
>>>>>
>>>>> Yes, MatSetValues always operates on the CPU side. It would not make
>>>>> sense to do individual operations on the GPU.
>>>>>
>>>>> I have written batched of assembly for element matrices that are all
>>>>> the same size:
>>>>>
>>>>>
>>>>> http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-current/docs/manualpages/Mat/MatSetValuesBatch.html
>>>>>
>>>>>
>>>>>> - Is it such that matrices are assembled in their entirety on the
>>>>>> CPU, and then copied over to the GPU (after calling MatAssemblyBegin)? Or
>>>>>> are values copied over to the GPU each time you call MatSetValues?
>>>>>>
>>>>>
>>>>> That function assembles the matrix on the GPU and then copies to the
>>>>> CPU. The only time you do not want this copy is when
>>>>> you are running in serial and never touch the matrix afterwards, so I
>>>>> left it in.
>>>>>
>>>>>
>>>>>> - Can we expect to see any speedup from using MatSetValuesBatch over
>>>>>> MatSetValues, or is the batch version simply a utility function? This
>>>>>> question goes for both CPU- and GPU-based matrices.
>>>>>>
>>>>>
>>>>> CPU: no
>>>>>
>>>>> GPU: yes, I see about the memory bandwidth ratio
>>>>>
>>>>>
>>>>> Hi,
>>>>
>>>> I have now integrated MatSetValuesBatch in our existing PETSc wrapper
>>>> layer. I have tested matrix assembly with Poisson's equation on different
>>>> meshes with elements of varying order. I have timed the single call to
>>>> MatSetValuesBatch and compared that to the total time consumed by the
>>>> repeated calls to MatSetValues in the old implementation. I have the
>>>> following results:
>>>>
>>>> Poisson on 1000x1000 unit square, 1st order Lagrange elements:
>>>> MatSetValuesBatch: 0.88576 s
>>>> repeated calls to MatSetValues: 0.76654 s
>>>>
>>>> Poisson on 500x500 unit square, 2nd order Lagrange elements:
>>>> MatSetValuesBatch: 0.9324 s
>>>> repeated calls to MatSetValues: 0.81644 s
>>>>
>>>> Poisson on 300x300 unit square, 3rd order Lagrange elements:
>>>> MatSetValuesBatch: 0.93988 s
>>>> repeated calls to MatSetValues: 1.03884 s
>>>>
>>>> As you can see, the two methods take almost the same amount of time.
>>>> What behavior and performance should we expect? Is there any way to
>>>> optimize the performance of batched assembly?
>>>>
>>>
>>> Almost certainly it is not dispatching to the CUDA version. The regular
>>> version just calls MatSetValues() in a loop. Are you
>>> using a SEQAIJCUSP matrix?
>>>
>>  Yes. The same matrices yields a speedup of 4-6x when solving the system
>> on the GPU.
>>
>
> Please confirm that the correct routine by running wth -info and sending
> the output.
>
> Please send the output of -log_summary so I can confirm the results.
>
> You can run KSP ex4 and reproduce my results where I see a 5.5x speedup on
> the GTX285
>
> I am not sure what to look for in those outputs. I have uploaded the
output of running my assembly program with -info and -log_summary, and the
output of running ex4 with -log_summary. See

http://folk.uio.no/fredva/assembly_info.txt
http://folk.uio.no/fredva/assembly_log_summary.txt
http://folk.uio.no/fredva/ex4_log_summary.txt

Trying this on a different machine now, I actually see some speedup. 3rd
order Poisson on 300x300 assembles in 0.211 sec on GPU and 0.4232 sec on
CPU. For 1st order and 1000x1000 mesh, I go from 0.31 sec to 0.205 sec.
I have tried to increase the mesh size to see if the speedup increases, but
I hit the bad_alloc error pretty quick.

For a problem of that size, should I expect even more speedup? Please let
me know if you need any more output from test runs on my machine.

-- 
Fredrik
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20111201/460366cb/attachment.htm>