We ran into performance problems on a 24bpp device and it turns out MOZ_GFX_OPTIMIZE_MOBILE forces 16bit. It also does some other things that make things look **** on mobile. Lets reevaluate whether those things really cost more performance than we can afford, in particular on current devices (e.g. font hinting!).

See also bug 687835

NB: bug 688115 added another of these #ifdef MOZ_GFX_OPTIMIZE_MOBILE // Prefer r5g6b5 for potential savings in memory bandwidth. // This needs to be reevaluated for newer devices. { gfxASurface::ImageFormatRGB16_565, kEGLConfigAttribsRGB16 }, #endif

We likely need some way to switch between 16 bit and 24/32 bit colour based on device. I don't think that we'll have benefit from switching wholesale to 24-bit, but we definitely need to do tests on that.

note that bug 499853 looks like a legit use of MOZ_GFX_OPTIMIZE_MOBILE, so I'm not filing a bug to review it.

bug 634759 recently landed and added some MOZ_GFX_OPTIMIZE_MOBILE checks, I asked on the bug if the fix is considered permanent or if there should be follow up work (it seems the bug is not fully understood). Other than that, and bug 499853 noted above, all uses of MOZ_GFX_OPTIMIZE_MOBILE have bugs filed against them.

++blassey, thanks a lot

I think I disagree with bug 499853. We should not behave differently on mobile vs desktop on this. Is performance a concern here? How often do we get missing glyphs? At most I would go with a pref. These hard-coded defines should really die.

bug 499853 says that putting the hex codes in the missing glyphs *was* a performance problem. I'd have no objection to making that a pref though.

The bug was 2 years ago, on lesser average hardware, and a less optimized stack. I would like to remeasure this, and also understand how often this happens (e.g. do you know any site I can use for testing?)

http://www.bbc.co.uk/persian/ assuming you haven't installed persian fonts on your device

cool, thanks, thats a great test case

Closing all opened bug in a graveyard component