Dear Vikram,
Thank you again for the pointer (and the patience)! I used the PerfLog
class and compared. As you mentioned, the assembly time change is quite
small (1.8772 vs 1.8845 sec).
And I appreciate your suggestion for checking directly on the type of
element as well!
Have a great week ahead,
Shawn
PS the complete timing report:
Before including the qrule creation in the loop:
------------------------------------------------------------------------------------------------------------
| Assemble Performance: Alive time=1.88073, Active time=1.87716
|
------------------------------------------------------------------------------------------------------------
| Event nCalls Total Time Avg Time Total
Time Avg Time % of Active Time |
| w/o Sub w/o Sub With
Sub With Sub w/o S With S |
|------------------------------------------------------------------------------------------------------------|
|
|
| Entire assembly 1 1.8772 1.877163 1.8772
1.877163 100.00 100.00 |
------------------------------------------------------------------------------------------------------------
| Totals: 1 1.8772
100.00 |
------------------------------------------------------------------------------------------------------------
After including the qrule creation in the loop:
------------------------------------------------------------------------------------------------------------
| Assemble Performance: Alive time=1.88795, Active time=1.88449
|
------------------------------------------------------------------------------------------------------------
| Event nCalls Total Time Avg Time Total
Time Avg Time % of Active Time |
| w/o Sub w/o Sub With
Sub With Sub w/o S With S |
|------------------------------------------------------------------------------------------------------------|
|
|
| Entire assembly 1 1.8845 1.884490 1.8845
1.884490 100.00 100.00 |
------------------------------------------------------------------------------------------------------------
| Totals: 1 1.8845
100.00 |
------------------------------------------------------------------------------------------------------------
On Sun, Apr 14, 2019 at 4:11 PM Vikram Garg <vikram.v.g...@gmail.com> wrote:
> Hi Yuxiang, glad to hear you are getting the expected results. This change
> looks correct for your case and should not affect the assembly time by a
> significant amount. You could also check for the actual type of element
> (QUAD4 or TRI) instead of checking indirectly via the number of nodes.
>
> You can check the performance change by checking the performance log which
> is included at the end of the program's output. Look for something like:
>
> | System
> |
> | assemble() 1 0.0014 0.001422
> 0.0028 0.002767 17.59 34.22 |
>
> Thanks.
> Vikram Garg
>
> vikramvgarg.github.io/
>
>
> On Sun, Apr 14, 2019 at 5:22 PM Yuxiang Wang <yw...@virginia.edu> wrote:
>
>> Hi Vikram,
>>
>> Thank you for your response!
>>
>> I didn't think about that - I tried to add that inside the loop over
>> elements and got the correct results!
>>
>> Question - do you think that will significantly slow down the assembly,
>> because we are now creating qrule and attaching it for every element loop?
>>
>> What I did was the following:
>>
>> for (const auto & elem : mesh.active_local_element_ptr_range()) {
>>
>> const unsigned int n_nodes = elem->n_nodes();
>>
>> Order qrule_order;
>>
>> switch (n_nodes)
>> {
>> case 3:
>> qrule_order = SECOND;
>> break;
>>
>> case 4:
>> qrule_order = THIRD;
>> break;
>>
>> default:
>> qrule_order = fe_type.default_quadrature_order();
>> break;
>> }
>>
>> QGauss qrule (dim, qrule_order);
>> fe->attach_quadrature_rule (&qrule);
>>
>> Thanks!
>>
>> Best,
>> Shawn
>>
>> On Sun, Apr 14, 2019 at 2:53 PM Vikram Garg <vikram.v.g...@gmail.com>
>> wrote:
>>
>>> Hi Yuxiang,
>>> It should be possible to move the construction of the
>>> finite element and the specification of the quadrature rule inside the loop
>>> over elements. Did this give any errors, or is otherwise unfeasible for
>>> performance reasons ?
>>>
>>> Thanks.
>>> Vikram Garg
>>>
>>> vikramvgarg.github.io/
>>>
>>>
>>> On Sun, Apr 14, 2019 at 4:30 PM Yuxiang Wang <yw...@virginia.edu> wrote:
>>>
>>>> Dear all,
>>>>
>>>> Sorry for the spam.
>>>>
>>>> I have a mix of QUAD4 and TRI3 elements in my mesh, and would like to
>>>> use
>>>> different quadrature rule orders for each type. Specifically, I'd like
>>>> to
>>>> have order THIRD for QUAD4 and SECOND for TRI3. By default, the orders
>>>> are
>>>> both THIRD.
>>>>
>>>> In most examples, I noticed that the specification of qrule is only for
>>>> the
>>>> entire FEType and not dependent on the element type (e.g. QUAD4 vs
>>>> TRI3).
>>>> Could you please share with me some hints on how to make
>>>> element-dependent
>>>> qrule order selection happen?
>>>>
>>>> Best,
>>>> Shawn
>>>>
>>>>
>>>> --
>>>> Yuxiang "Shawn" Wang, PhD
>>>> yw...@virginia.edu
>>>> +1 (434) 284-0836
>>>>
>>>> _______________________________________________
>>>> Libmesh-users mailing list
>>>> Libmesh-users@lists.sourceforge.net
>>>> https://lists.sourceforge.net/lists/listinfo/libmesh-users
>>>>
>>>
>>
>> --
>> Yuxiang "Shawn" Wang, PhD
>> yw...@virginia.edu
>> +1 (434) 284-0836
>>
>
--
Yuxiang "Shawn" Wang, PhD
yw...@virginia.edu
+1 (434) 284-0836
_______________________________________________
Libmesh-users mailing list
Libmesh-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libmesh-users
