Patch for XFree86 to allow hardware acceleration on the newport video subsystem. This patch only effects the mips arch that uses the newport. Reproducible: Always Steps to Reproduce: 1.Apply patch 2.Compile 3.Behold XFree86 Actual Results: http://www.mindwerks.net/~psi/mips/screenshot2.jpeg CPU overhead on SGI Indy workstation is dramatically reduced and allows for more functionality while in X. There are known bugs, HWcursor "true" cursor isn't quite visable, but HWcursor "false" makes for strange artifacts. XFree86 configure doesn't generate a XF86Config file.
Created attachment 24009 [details, diff] XFree86 hardware accelerated newport driver patch
With this particular patch, XFree86 must be compiled statically as dlopen() hasn't been implemented yet, but is forth comming. This patch has yet to make it into XFree86's cvs.
Have you made this patch cleanly against gentoo's xfree as opposed to vanilla xfree?
Correct, this patch applied cleanly to the xfree86-4.3.99.902.ebuild after putting the patch into XFree86-4.3.99.902-patches-0.1.tar.bz2 and re-running the digest. I watched the auto patcher take the patch and apply it. The results can been seen in the screenshot. Still testing it out, GTK+2 text isn't displaying much of anything right now, but standard gtk+ apps are just fine. I would rather run X with this patch than without at the moment, at least for further testing in MIPS.
OK, I'll get it added in the next patchset.
Created attachment 24027 [details, diff] Updated patch. Did some checking to make sure everything went cleanly against both XFree86 CVS and then against Gentoo and noticed some inconsistancies with location of patched files, this will take of it. --- xc/programs/Xserver/hw/xfree86/drivers/newport.orig/newport_window.c 1970-01-01 01:00:00.000000000 +0100 +++ xc/programs/Xserver/hw/xfree86/drivers/newport/newport_window.c 2003-12-02 15:36:27.000000000 +0100
By the way, this patch will only be applied on USE=static.
Created attachment 43083 [details, diff] patch to accelerate 24bpp mode for the newport driver TCMAN's latest, greatest, and final XAA re-write of the newport acceleration patch is attached here now. It should apply against all xorg-x11 ebuilds with no problems (tested 6.8.0-r1 and 6.8.0-r2).
One more comment...this patch doesn't necessarily need USE="static" anymore. You just have to make sure to load the xaa module in your xorg.conf.
Created attachment 43120 [details, diff] newport-accel-4.patch here's a better version. this one knows how to load XAA for itself, and also has Render acceleration that will actually do something on 6.8 - no conformance guarantees, but then 6.7 Render accel was never conformant so this is no big deal. if this works out i'd like to see it applied upstream too, if the author is willing.
Created attachment 43121 [details, diff] newport-accel-5.patch woops, missed a spot, set up the legal destination formats too. since i mentioned conformance, the canonical Render conformance test is called rendercheck: http://freedesktop.org/~xapps/release/rendercheck-1.0.3.tar.bz2 which i'll post an ebuild for in the next few days. the Render setup hooks in 6.7 and before only passed the source picture format, so you _cannot_ write a conformant Render accel using the old hooks.
Fixed in 6.8.0-r2 in CVS. Please test.
Created attachment 43627 [details, diff] fixes a typo in newport-accel-5.patch
reopened so that the new patch with a typo fix is committed
Ok, don't commit the new patch just yet. TCMAN updated his driver to a revision #10. I am testing it right now, and will post a diff against the patch currently in the xorg patchset.
Created attachment 43830 [details, diff] this diff modifies the 5200_all_6.8.0-newport-accel.patch to the newest version of dominik's driver Dominik released a version 10 of his newport accel driver. This version was written specifically with xorg-x11 in mind. The attached patch is a diff against the current newport patch in the xorg-x11 patchset which will update it to the new driver. It will also fix the spelling mistake.
Just in case anybody wants to see it, the full diff of version 10 against a stock xorg-x11 source tree is located at http://dev.gentoo.org/~geoman/newportaccel.diff
Ajax, could you please vet the patch in comment #16?
looking at patches to patches makes my eyes go all crossy. the patch in comment #17 looks fine though. if there are no objections - and if Dominik doesn't beat me to it - i'll open a bug upstream to get this included. i'd still like to see rendercheck results though.
Rendercheck has an ebuild in portage now, so just emerge and test.
Added to 6.8.0-r2 in patchset 0.2.7.
Created attachment 44142 [details, diff] latest accel patch , with fix for swapped red and blue rel10 + fixes apps that use RENDER like gnome-term, gvim, xchat2
Please reopen when adding things that require action on our part.
Added to 6.8.0-r4 in patchset 0.2.8.
Dominik, you need to make sure you obsolete the proper stuff if you have a new patch. Donnie, the patch in comment 22 is actually newer than the one in xorg-6.8.0-r4 now. It fixes a small problem with font color swapping in gtk+2 apps that was only recently confirmed to be a problem with the driver. Hopefully we can get this stuff upstream so we can get rid of this bug! Dominik? ;)
Stephen, please diff it with the 5200 patch in the patchset 0.2.8. They should be the same. If they aren't, I screwed up.
Hmmm, they are the same. Dominik, are you *sure* you fixed the color problems? I'm still having problems with it.
Well it seems it fixes some of the color problems - namely gvim - that uses RENDER. I tried xchat2 and compared it to my other computer and red and blue is still reversed. I suspect it must be an application or gtk2 problem. The colorformat that the driver reports to X in xf86SetWeight at init time and then accepts internally from XAA is XBGR (in 32bit word). I suspect some widgets in gtk2 do not check the color format and pass XRGB or the app itself passes XRGB color to gtk2.
Ok, the newport patch in the xorg-x11 patchset needs to be reverted to the one in comment #17 (http://dev.gentoo.org/~geoman/newportaccel.diff), because the later versions incorrectly try to fix the xaa color swapping problem which is described and fixed in https://bugs.freedesktop.org/show_bug.cgi?id=1895
This is in 6.8.1.902 patchset 0.1.2 -- please test ASAP. I want to push it to 6.8.0-r4 and get that stabled for the release if possible.
Tested and working! Go ahead and add it to the 6.8.0-r4 patchset.
patch has been comitted to 6.8.0-r4, closing this bug now
Please post a bug at bugs.freedesktop.org with this patch, and stick the URL here.
https://bugs.freedesktop.org/show_bug.cgi?id=2356 freedesktop.org bug for inclusion of this patch upstream
