On 04/13/2017 08:34 AM, Igor Mironchik wrote:
Hi,
Strange, I launch 3Dtree on Linux under valgrind:
==4321== HEAP SUMMARY:
==4321== in use at exit: 439,965 bytes in 6,185 blocks
==4321== total heap usage: 2,207,719 allocs, 2,201,534 frees, 1,241,469,509
bytes allocated
==4321==
==4321== LEAK SUMMARY:
==4321== definitely lost: 1,080 bytes in 10 blocks
==4321== indirectly lost: 7,609 bytes in 55 blocks
==4321== possibly lost: 416 bytes in 1 blocks
==4321== still reachable: 430,860 bytes in 6,119 blocks
==4321== suppressed: 0 bytes in 0 blocks
Check the virtual image size on Linux,
just to make sure it is not growing as well.
That is the bottom line.
But on my Windows OS after each restart of tree I receive plus ~100MB additional
memory.
As I can see this memory still reachable. But why 3Dtree allocates more and more
memory and doesn't use freed one?
13.04.2017 9:58, Sean Harmer пишет:
Hi,
On 13/04/2017 07:09, Igor Mironchik wrote:
Hello,
3Dtree has been updated. Now autumn is animated. Removed hand-made
classes of leaf geometry and mesh. Now example uses ready mesh prepared
in Blender. Modified a little constants of the tree. Example looks nice.
This is a benchmark of your video card, because leafs and branches are
stand alone objects (QEntity). For example 5 years tree has ~ 900
objects. My Intel Pentium with Intel HD graphics normally paint only 5
years tree, 6 years tree starts to slow down.
Yes, at present each entity with a material/geometry renderer combo gets
translated to an OpenGL draw call. We are looking to add batching (similar to
how Qt Quick 2 works) but it's not there yet.
If you don't need to address each individual leaf/branch in your object model.
E.g. if you're just rendering them, then you can get *much* better performance
by using instanced rendering.
Essentially you have one entity representing all leaves. This entity contains
the material and geometry renderer as normal. The positions and any other
custom properties that vary between instances (rotation, leaf size, leaf
colour etc) should be placed into an additional attribute/buffer and provided
to the geometry renderer. On this attribute, set the divisor property to 1
meaning that 1 piece of data maps to 1 instance of the leave in the scene.
With this approach you will be able to render 10,000's leaves in a single draw
call. It maps through to a call to glDrawElementsInstanced() in case you want
to read up on it. Essentially you're moving the for loop over each leaf on to
the GPU.
Typically, you'd have your data for the leaves in an array in C++ and use this
to populate the leaf instance buffer.
The plan is to have a batcher that uses this instancing facility.
Cheers,
Sean
And one more - now you can rotate the tree with the left mouse button.
11.04.2017 12:15, Igor Mironchik пишет:
Hi,
I fixed a little 3Dtree. Now branches positions are correct. And I
know that not all leafs is visible (this is because leaf geometry is
implemented as plain which renders only on one side).
10.04.2017 13:20, Igor Mironchik пишет:
Hello,
I guess that Qt3D has huge memory leaks.
You can check it on this example: https://github.com/igormironchik/3Dtree
It's 3D tree, that grows year by year.
By default tree will grow 5 years (5 minutes).
When tree grown you can restart tree. And here I delete all resources:
void
MainWindowPrivate::createTree()
{
if(m_tree)
{
for(constauto&e:m_rootEntity->childNodes())
e->deleteLater();
}
m_tree=newBranch(m_startPos,m_endPos,c_startBranchRadius,
true,m_rootEntity);
m_tree->setAge(0.0f);
m_tree->updatePosition();
m_tree->placeLeafs();
}
But a lot of memory are eaten.
_______________________________________________
Interest mailing list
Interest@qt-project.org
http://lists.qt-project.org/mailman/listinfo/interest
_______________________________________________
Interest mailing list
Interest@qt-project.org
http://lists.qt-project.org/mailman/listinfo/interest
_______________________________________________
Interest mailing list
Interest@qt-project.org
http://lists.qt-project.org/mailman/listinfo/interest
