User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b6) Gecko/20100101 Firefox/4.0b6 Build Identifier: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b8pre) Gecko/20101008 Firefox/4.0b8pre When resizing the FishIE demo, FPS drops from about 40 to 30 when resizing from fullscreen to non fullscreen. FPS drop when resizing from smaller to bigger would be understandable, but this FPS drop happens when resizing from bigger to smaller. Reproducible: Always Steps to Reproduce: 1. Open the FishIE demo 2. Go fullscreen (F11) 3. Click the 1000 fishes link and wait for it to load fully 4. Point your mouse to the top of the screen so that the address bar shows up, or press F11 to exit fullscreen 5. Move your mouse out of the address bar or press F11 to go fullscreen again Actual Results: It becomes a slower (lower fps) when it's non-fullscreen, and becomes faster again when fullscreen. Expected Results: FPS should stay relatively same. Adapter Description ATI Radeon HD 4600 Series Vendor ID 1002 Device ID 9498 Adapter RAM 1024 Adapter Drivers aticfx64 aticfx64 aticfx32 aticfx32 atiumd64 atidxx64 atiumdag atidxx32 atiumdva atiumd6a atitmm64 Driver Version 8.773.0.0 Driver Date 9-7-2010 Direct2D Enabled true DirectWrite Enabled true GPU Accelerated Windows 1/1 Direct3D 10

There is no fullscreen->nonfullscreen fps drop with D3D9, but it is always slower.

I think this is expected, resizing with hardware accel is expensive.

It's not only when resizing. FPS is always lower when there is chrome visible.

I can reproduce this, but it's not surprising -- when we go full screen (with no chrome visible), the layer tree is much simpler -- essentially consisting of just the canvas and possibly a thebes layer on top. So it's faster to composite everything to the screen; when the UI is visible, there are layers for the UI components visible that have to be composited or clipped out. Might be interesting to examine what gets repainted per frame with/without UI visible to see if there's something extra that's being drawn.

(In reply to comment #4) > I can reproduce this, but it's not surprising -- when we go full screen (with > no chrome visible), the layer tree is much simpler -- essentially consisting of > just the canvas and possibly a thebes layer on top. So it's faster to > composite everything to the screen; when the UI is visible, there are layers > for the UI components visible that have to be composited or clipped out. > > Might be interesting to examine what gets repainted per frame with/without UI > visible to see if there's something extra that's being drawn. I wonder if that should be so significant. You're right that we should investigate if something's needlessly being repainted.

On IE9 there's 43 fps with full screen and 43fps with chrome visible. No drops.

(In reply to comment #6) > On IE9 there's 43 fps with full screen and 43fps with chrome visible. No drops. IE9 does not use D2D to draw chrome, so it makes sense that they're unaffected by whether it's visible. We do need to confirm nothing is needlessly invalidated though!

Try switching to Aero Basic (i.e. no glass) to see if that makes the fps drop go away? Maybe it's a problem triggered by the glass.

You could test the chrome theory by bringing up FishIE in a custom window with no chrome elements (e.g. use window.open with titlebar=no,toolbar=no,location=no --- https://developer.mozilla.org/en/DOM/window.open).

You are running on drivers from september. Try to update them to more recent version from this year. http://sites.amd.com/us/game/downloads/Pages/radeon_win7-64.aspx

also does firefox 4 final help?