About Athens and Cairo
Hi! I have some random questions about the Cairo binding supported by Athens. - the rendering is done by the graphic card isnât it? - Once rendered, is the produced bitmap copied by the CPU in Pharo? With Roassal3d, we are facing a problem of performance because there is an unnecessary copy between the graphic card and the Form displayed by Pharo. I guess we have the same problem with Cairo. - Is there an example of clipping in Cairo? - Any idea how efficient the clipping is? I mean, will I gain a lot by not displaying elements located outside the visual part or should I leave the clipping do the job? Cheers, Alexandre -- _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;: Alexandre Bergel http://www.bergel.eu ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;.
Le 14/12/2013 09:58, Alexandre Bergel a écrit :
- the rendering is done by the graphic card isnât it?
I think it is sofware based, by the libcairo. Hilaire -- Dr. Geo http://drgeo.eu
Le 14/12/2013 09:58, Alexandre Bergel a écrit :
- the rendering is done by the graphic card isnât it?
I think it is sofware based, by the libcairo.
Yes, but on http://cairographics.org it is written: Cairo is designed to produce consistent output on all output media while taking advantage of display hardware acceleration when available (eg. through the X Render Extension). Cheers, Alexandre -- _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;: Alexandre Bergel http://www.bergel.eu ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;.
No cairo is not GPU accelerated , rendering is done by your CPU , but of course the graphic card is the one that display things. here is a detailed explanation -> http://lists.cairographics.org/archives/cairo/2012-October/023609.html please note that even in the case of opengl backend I very much disagree that this is true hardware acceleration. What the opengl backend is doing is to render cairo surfaces as opengl textures. So even though the texture themselves are hardware accelerated and will even use the the memory of the GPU the process of rendering to the texture will happen solely on the CPU. So its more like a CPU library trying to be hardware accelerated. But even using the OS backends like Quartz is not quite the same. Because you still call C code and C code is executed by your CPU and that C code suppose to call GPU functions. Opengl speed is for the very reason of shaders meaning that it compiles programs that are run solely on the gpu hence true GPU acceleration. Saying that you should not be experiencing slow downs unless you were doing something really intense. I have seen benchmarks of Cairo and they are blazing fast. Usually slow down are primarily because of bad code. I have seen animation in athens take out 50% of my dual core and others only a 5%. Cant say I am sure what happens under the surface maybe Igor may offer a more detailed look into this. I do feel however that slow speed are 99% to be blamed on Pharo way of rendering things. Something similar happens for browsers, SVG is GPU accelerated, Javascript is super fast but it only takes the slowness of DOM (HTML) itself to bring the whole system on its knees. It only take one slow part to ruin the experience. I think a solution to this would be to bypass pharo rendering completely, create a new window and render directly to that window running Pharo headless. On Sat, Dec 14, 2013 at 10:58 AM, Alexandre Bergel <alexandre.bergel@me.com>wrote:
Hi!
I have some random questions about the Cairo binding supported by Athens.
- the rendering is done by the graphic card isnât it?
- Once rendered, is the produced bitmap copied by the CPU in Pharo? With Roassal3d, we are facing a problem of performance because there is an unnecessary copy between the graphic card and the Form displayed by Pharo. I guess we have the same problem with Cairo.
- Is there an example of clipping in Cairo?
- Any idea how efficient the clipping is? I mean, will I gain a lot by not displaying elements located outside the visual part or should I leave the clipping do the job?
Cheers, Alexandre -- _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;: Alexandre Bergel http://www.bergel.eu ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;.
Thanks Kilon for the explanation. Alexandre On Dec 14, 2013, at 10:56 AM, kilon alios <kilon.alios@gmail.com> wrote:
No cairo is not GPU accelerated , rendering is done by your CPU , but of course the graphic card is the one that display things.
here is a detailed explanation -> http://lists.cairographics.org/archives/cairo/2012-October/023609.html
please note that even in the case of opengl backend I very much disagree that this is true hardware acceleration. What the opengl backend is doing is to render cairo surfaces as opengl textures. So even though the texture themselves are hardware accelerated and will even use the the memory of the GPU the process of rendering to the texture will happen solely on the CPU. So its more like a CPU library trying to be hardware accelerated.
But even using the OS backends like Quartz is not quite the same. Because you still call C code and C code is executed by your CPU and that C code suppose to call GPU functions. Opengl speed is for the very reason of shaders meaning that it compiles programs that are run solely on the gpu hence true GPU acceleration.
Saying that you should not be experiencing slow downs unless you were doing something really intense. I have seen benchmarks of Cairo and they are blazing fast. Usually slow down are primarily because of bad code. I have seen animation in athens take out 50% of my dual core and others only a 5%. Cant say I am sure what happens under the surface maybe Igor may offer a more detailed look into this. I do feel however that slow speed are 99% to be blamed on Pharo way of rendering things.
Something similar happens for browsers, SVG is GPU accelerated, Javascript is super fast but it only takes the slowness of DOM (HTML) itself to bring the whole system on its knees. It only take one slow part to ruin the experience. I think a solution to this would be to bypass pharo rendering completely, create a new window and render directly to that window running Pharo headless.
On Sat, Dec 14, 2013 at 10:58 AM, Alexandre Bergel <alexandre.bergel@me.com> wrote: Hi!
I have some random questions about the Cairo binding supported by Athens.
- the rendering is done by the graphic card isnât it?
- Once rendered, is the produced bitmap copied by the CPU in Pharo? With Roassal3d, we are facing a problem of performance because there is an unnecessary copy between the graphic card and the Form displayed by Pharo. I guess we have the same problem with Cairo.
- Is there an example of clipping in Cairo?
- Any idea how efficient the clipping is? I mean, will I gain a lot by not displaying elements located outside the visual part or should I leave the clipping do the job?
Cheers, Alexandre -- _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;: Alexandre Bergel http://www.bergel.eu ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;.
-- _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;: Alexandre Bergel http://www.bergel.eu ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;.
For the moment athens use the form buffer so there is a copy that igor would like to kill. Now igor got killed by the textEditor. I hope that he will come back in good shape. The best things to improve the situation is that people with interest like you really invest to understand the underlying library and approach because you cannot thing that they will be able somebody below knowing everything and answering. On Dec 14, 2013, at 9:58 AM, Alexandre Bergel <alexandre.bergel@me.com> wrote:
Hi!
I have some random questions about the Cairo binding supported by Athens.
- the rendering is done by the graphic card isnât it?
- Once rendered, is the produced bitmap copied by the CPU in Pharo?
kind of. Yes there is a copy in the FormBuffer.
With Roassal3d, we are facing a problem of performance because there is an unnecessary copy between the graphic card and the Form displayed by Pharo. I guess we have the same problem with Cairo.
Yes Now if people would help Igor could focus on fixing that. Now without textEditor = no Athens.
- Is there an example of clipping in Cairo?
I thikn that cairo already handle it.
- Any idea how efficient the clipping is? I mean, will I gain a lot by not displaying elements located outside the visual part or should I leave the clipping do the job?
Cheers, Alexandre -- _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;: Alexandre Bergel http://www.bergel.eu ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;.
participants (4)
-
Alexandre Bergel -
Hilaire Fernandes -
kilon alios -
Stéphane Ducasse