Hello,<div><br></div><div><br></div><div>I am trying to improve the rendering speed of web-pages using webkit with gtk-x11 backend.</div><div>In order to make use of the acceleration provided by EXA, i have hooked the solid, copy and composite hooks provided by EXA with my hardware implementation.</div>
<div>And when i tested the performance using some X11 application that creates a simple pixmap and copies it to a window multiple times</div><div>using XCopyArea(), then the performance gain is quite good. I have got 6-8x performance gain.</div>
<div><br></div><div>When we consider the web-page scrolling scenario, i guess it also must be making a call to XCopyArea() only finally to bitblit the entire window contents during scrolling.</div><div>But when i test the performance using some auto scrolling webpage on my web-browser app, if EXA is enabled its getting worse.</div>
<div><br></div><div>According to my observation, if the pixmap migration is explicitly made OFF, then the performance is at par with that of EXA-Disabled case.</div><div>This suggests that during the cases in which EXA fallbacks, most of the time consumed in system to video memory copy is killing the performance.</div>
<div><br></div><div>My question is, Is EXA really provides some performance gain in cases like web-page rendering?? </div><div>If yes, am i missing on some implementation part??</div><div>Can the migration mechanism of EXA be improved to suit the web-page rendering scenarios??</div>
<div>What kind of practical applications are considered to be improved by EXA??</div><div><br></div>