PM 15 and hardware acceleration

[chumley]

Thank you moonchild for your work on PM 15. I have some good news to share.

I have a laptop with Intel 4 series integrated graphics running Win 7 and latest drivers from Intel. With previous PM versions, I experienced the following bug, which caused drop-down menus to flicker when running on a secondary monitor. The problem would go away if hardware acceleration was disabled.
https://bugzilla.mozilla.org/show_bug.cgi?id=600381

PM 15 seems to have fixed this problem, even though the bug is still open. I haven't changed any drivers recently, so the fix is due to the PM upgrade.

I thought hw acceleration might just be disabled somehow behind the scenes, so I used the following tool to test. It runs much faster when hw acceleration is enabled, so the acceleration is indeed working.
https://developer.mozilla.org/media/upl ... e/HWACCEL/

I don't know if credit for the fix goes to the FF 15 codebase or something unique about PM, but in any case I'm pleased that this little annoyance is gone.

[Moonchild]

The blocklist for Pale Moon 15 has been extended with more entries for integrated Intel graphics, because layers acceleration is simply broken in them. Pale Moon 15 selectively disables some parts of the acceleration that are poorly supported on Intel integrated graphics, while leaving overall acceleration enabled for other parts.

[chumley]

Thanks for the explanation moonchild!

I wasn't familiar with the blocklist feature. If anyone else is curious about this, here's a good explanation.

https://wiki.mozilla.org/Blocklisting/Graphics

[tribaljet]

Moonchild, do you distinguish different Intel IGPs? Because the performance found on something like a 4500MHD is quite different than what's found on a GMA 950 or on a HD 3000.

I'm running an Optimus-enabled system, that has an Intel IGP (HD 3000) to handle the system output and a Nvidia GPU (doesn't output by itself, sends the data to the Intel IGP instead) to act as a render workhorse, and given that the IGP is capable of hardware acceleration, I hope you're not considering all Intel IGPs as entirely ineligible for hardware acceleration.

[Moonchild]

There is no way to determine from the browser which task is performed by which GPU in such a setup, nor which parts are accelerated where (and IMHO doing it that way is asking for a lot of trouble!).

As for your question, the blocklist selectively enables and disables features based on manufacturer and model ID, not by generic type.

For example, an entry from the current blocklist, blocking Direct2D on certain ATI/AMD (vendor 0x1002) integrated chips (HD6310/toshiba laptops/some APUs - device 0x980*):

Code: Select all

    <gfxBlacklistEntry  blockID="g81">      <os>WINNT 6.1</os>      <vendor>0x1002</vendor>              <devices>
                      <device>0x9802</device>
                      <device>0x9803</device>
                      <device>0x9803</device>
                      <device>0x9804</device>
                      <device>0x9805</device>
                      <device>0x9806</device>
                      <device>0x9807</device>
                  </devices>
            <feature>DIRECT2D</feature>      <featureStatus>BLOCKED_DEVICE</featureStatus>      <driverVersion>1.0.0.0</driverVersion>      <driverVersionComparator>GREATER_THAN_OR_EQUAL</driverVersionComparator>    </gfxBlacklistEntry>

[tribaljet]

Glad to hear, as different models from the same vendor have drastically different features, such as proper hardware acceleration.

[satrow]

Mobile versions of the same graphics chipset are likely to be crippled in some way compared to the desktop versions.

[tribaljet]

Mobile Intel IGPs are simply downclocked versions of the desktop IGP versions. Feature wise they're exactly the same.

[chumley]

PM 15 seems to have fixed this problem ...

Whoops, it looks like I spoke too soon. It still happens sometimes, but less frequently than before. Oh well, not a big deal, especially since it only happens intermittently and only on a secondary monitor. At least I understand now that it is Intel's fault and not FF or PM.

[Moonchild]

It seems, also judging by a related recent support request, that this is a flaw in the DirectX 10 implementation of the Intel drivers, and that forcing it to use DirectX 9 can fix all of the issues.

In about:config, try setting layers.prefer-d3d9 to true

EDIT: By the way, the fact that it doesn't show up as easily in Pale Moon 15 is because it's much tighter on timings, basically it would flicker but it would be too fast for you to see.

[chumley]

Thanks, I'll play with that. It may be hard to determine the impact, since I can't seem to reproduce the issue on demand. PM15 avoids the issue most of the time already.

Thanks for your time Moonchild.

[chumley]

It seems, also judging by a related recent support request, that this is a flaw in the DirectX 10 implementation of the Intel drivers, and that forcing it to use DirectX 9 can fix all of the issues.

In about:config, try setting layers.prefer-d3d9 to true

I used this setting for a while and I think it may helped with the original issue, but I'm not certain since the issue was not strictly reproducible. However, this setting introduced another weird issue. I have font smoothing disabled on my machine, but that setting caused smoothed fonts to appear intermittently in different places. One repeatable example was the Yahoo mail delete confirmation dialog. Disabling the above setting or disabling hardware acceleration eliminated the font smoothing issue. So I'm going back to layers.prefer-d3d9 = false.

[Moonchild]

Usually font smoothing issues are related to forcing antialiasing in the graphics driver control panel. You want to disable that - games tend to have a built-in setting for this anyway so there's no need to enable it overall.