Anti-Aliasing

[jmaustpc]

But I have no idea if disabling glDepthRange causes a regression somewhere and thus I think we should do the change after 0.14 is out so that we have more time for testing.

But then 0.14 will be left with a pretty ugly bug, wouldn't it be better to include this, test for week or two, if it does show up any regressions then remove the patch if you want to before releasing 0.14?

[quick61]

ickby, your anti-aliasing patch works a treat. I hardly notice any performance hit on my system until I hit MASS 8X. Between None and MSAA 4X, there is very little difference in my FPS. With 8X, there is about a 50% drop. With little visual difference between 4X and 8X, for me, 4X seems to be a good setting with no noticeable reduction in performance "feel". Thanks for this.

I have noticed that this improved visual quality does not seem to translate to the Tools > Save picture... function. Is this because of the compression level of the saved image, or is the image being taken from a process before the anti-aliasing? Any fix for this part (Save picture... tool)

Both the pics below were taken with MSAA set at 8X one with FreeCAD's tool, the other with KSnapshot

Screenshot taken with FreeCAD > Tools > Save picture...

X8_FreeCAD_Save_picture.png (13.37 KiB) Viewed 3955 times

Screenshot taken with KSnapshot

X8_KSnapshot.png (27.97 KiB) Viewed 3955 times

Mark

Nvidia GForce 9500 GT (G96)
1 GB RAM
Nvidia 331.67 driver
1.375 Ghz GPU

OS: Ubuntu 13.10
Platform: 64-bit
Version: 0.14.3562 (Git)
Branch: master
Hash: 320fabfc987d750303f3fe777238c67ae980acd9
Python version: 2.7.5+
Qt version: 4.8.4
Coin version: 4.0.0a
SoQt version: 1.5.0
OCC version: 6.7.0

[wmayer]

But then 0.14 will be left with a pretty ugly bug, wouldn't it be better to include this, test for week or two, if it does show up any regressions then remove the patch if you want to before releasing 0.14?

OK, I have tested many things and didn't find a regression. So, the fix is now active.

However, I found a regression when anti-aliasing is on. Whenever you draw into the front buffer (happens with the mesh cutting/trimming or the rubberband selection in the sketcher) the repainting doesn't work properly any more. So, when doing so I temporarily deactivate the anti-aliasing.

[jmaustpc]

But then 0.14 will be left with a pretty ugly bug, wouldn't it be better to include this, test for week or two, if it does show up any regressions then remove the patch if you want to before releasing 0.14?

OK, I have tested many things and didn't find a regression. So, the fix is now active.

However, I found a regression when anti-aliasing is on. Whenever you draw into the front buffer (happens with the mesh cutting/trimming or the rubberband selection in the sketcher) the repainting doesn't work properly any more. So, when doing so I temporarily deactivate the anti-aliasing.

Thanks, it looks great with this simple test.

transparencytest.jpg (20.18 KiB) Viewed 3920 times

Although the tools, save a picture, above does not look as good as the real thing in my display as shown with a screen capture. Look at the cylinder in side the transparent cube. The above is from the Tools menu, save picture.....and the below is from a KDE screen capture application.

transparencytestscreenshot.png (21.94 KiB) Viewed 3920 times

[detlet]

The anti aliasing is great! But I stumbled over two problems. First: If you have a image plan and anti aliasing is not activated, the image is shown as a grey rectangle. This may be confusing. Second: In a project, where I use an image plane (PNG 1412x1242) I got this error on the console:

Code: Select all

Coin error in SoNode::GLRenderS(): GL error: 'GL_OUT_OF_MEMORY', nodetype: Separator (set envvar COIN_GLERROR_DEBUGGING=1 and re-run to get more information)

. It was at a point, where the Sketcher had a problem to solve the sketch. I couldn't reproduce it till now, with the environment variable set. From the error message I think, that graphics card was gone out of RAM.

AMD Radon HD 5450
512 MB

OS: Debian GNU/Linux testing (jessie)
Platform: 64-bit
Version: 0.14.3563 (Git)
Branch: master
Hash: a5efc53bfcbdaeeec09bef30dd3203c6b4cfdb6f
Python version: 2.7.6
Qt version: 4.8.6
Coin version: 3.1.3
SoQt version: 1.5.0
OCC version: 6.7.0

[jmaustpc]

The anti aliasing is great! But I stumbled over two problems. First: If you have a image plan and anti aliasing is not activated, the image is shown as a grey rectangle. This may be confusing. Second: In a project, where I use an image plane (PNG 1412x1242) I got this error on the console:

Code: Select all

Coin error in SoNode::GLRenderS(): GL error: 'GL_OUT_OF_MEMORY', nodetype: Separator (set envvar COIN_GLERROR_DEBUGGING=1 and re-run to get more information)

. It was at a point, where the Sketcher had a problem to solve the sketch. I couldn't reproduce it till now, with the environment variable set. From the error message I think, that graphics card was gone out of RAM.

AMD Radon HD 5450
512 MB

OS: Debian GNU/Linux testing (jessie)
Platform: 64-bit
Version: 0.14.3563 (Git)
Branch: master
Hash: a5efc53bfcbdaeeec09bef30dd3203c6b4cfdb6f
Python version: 2.7.6
Qt version: 4.8.6
Coin version: 3.1.3
SoQt version: 1.5.0
OCC version: 6.7.0

Can you post a file that demonstrates these issues?

[kwahoo]

From the error message I think, that graphics card was gone out of RAM.

FYI: If you are using open source drivers, you can draw some graphs (to monitor VRAM usage) inside the FreeCAD GL viewport:

Code: Select all

GALLIUM_HUD=VRAM-usage,requested-VRAM freecad

MSAAx8, 300MiB memory usage:


And 1412x1242 PNG image MSAAx8, 436MiB:


OS: Ubuntu 14.04 LTS
Platform: 64-bit
Version: 0.14.3559 (Git)
Branch: master
Hash: 1fffc978b4c2873da5826c1de12207f3b292c4cf
Python version: 2.7.6
Qt version: 4.8.6
Coin version: 4.0.0a
SoQt version: 1.6.0a
OCC version: 6.7.0

[cblt2l]

Maybe someone else can confirm this but none of the MSAA types seem to work on my Ivy Bridge HD4000 chipset. The console gives this warning when I switch between them:

Code: Select all

Coin warning in SoQtGLWidgetP::buildGLWidget(): wanted sample buffers, but that is not supported by the OpenGL driver

OS: Debian GNU/Linux 7.5 (wheezy)
Platform: 64-bit
Version: 0.14.3606 (Git)
Branch: master
Hash: 1816c22f8d1fe4f348c40ae9976845d69fc5fc53
Python version: 2.7.3
Qt version: 4.8.2
Coin version: 3.1.3
SoQt version: 1.5.0
OCC version: 6.7.0

[kwahoo]

Maybe someone else can confirm this but none of the MSAA types seem to work on my Ivy Bridge HD4000 chipset.

Works fine for me.

Code: Select all

glxinfo | grep OpenGL
OpenGL vendor string: Intel Open Source Technology Center
OpenGL renderer string: Mesa DRI Intel(R) Ivybridge Mobile x86/MMX/SSE2
OpenGL core profile version string: 3.3 (Core Profile) Mesa 10.3.0-devel
OpenGL core profile shading language version string: 3.30

OS: Ubuntu 14.04 LTS
Platform: 64-bit
Version: 0.14.3559 (Git)
Branch: master
Hash: 1fffc978b4c2873da5826c1de12207f3b292c4cf
Python version: 2.7.6
Qt version: 4.8.6
Coin version: 4.0.0a
SoQt version: 1.6.0a

[ulrich1a]

Code: Select all
Coin warning in SoQtGLWidgetP::buildGLWidget(): wanted sample buffers, but that is not supported by the OpenGL driver



OS: Debian GNU/Linux 7.5 (wheezy)
Platform: 64-bit
Version: 0.14.3606 (Git)

It may be a Debian wheezy issue. I got the same message with a Radeon HD4670.

Code: Select all

OpenGL vendor string: X.Org
OpenGL renderer string: Gallium 0.4 on AMD RV730
OpenGL version string: 2.1 Mesa 8.0.5
OpenGL shading language version string: 1.20

OS: Debian GNU/Linux 7.5 (wheezy)
Platform: 32-bit
Version: 0.14.3564 (Git)
Branch: master
Hash: 3f4cbc421766b841cebcdf57a3de9a2c8211070c
Python version: 2.7.3
Qt version: 4.8.2
Coin version: 3.1.3
SoQt version: 1.5.0
OCC version: 6.7.0

Ulrich