Alternative for i915 driver (Android-x86 4.4-r3)

[Florian Wiedemann]

Hi,

is there an alternative for the default driver that is used for i915 graphic chip?

With the latest official release Android-x86 4.4-r3 I have a lot of graphical glitches (text not readable, background disordered) on my old Dell Notebook with that driver.
As far as I remember I read there is practically no chance that the driver will be fixed in the future (missing OpenGL functions).

Are the only alternatives to use the vesafb or uvesafb drivers for i915?

Kind regards
Florian

[Florian Wiedemann]

Hi,

I searched the whole group for kernel settings but with all of them (and all kind of combinations) I see only a blinking cursor but no GUI

• nomodeset UVESA_MODE=1024x768
• xforcevesa nomodeset UVESA_MODE=1024x768
• video=-16 video=1024x600 i915.modeset=0
• nomodeset vga=788

When I use the hardware accelerated default driver for i915 text and sometimes the background or other GUI controls are scrambled.



Any help is highly appreciated.

Kind regards
Florian

[rbg]

search forums for Hazy Fonts....issue maybe with mesa...using 10.4.2 mesa on older GPU's seems to solve it....

temporary until developers finish all mesa development I suspect

[Florian Wiedemann]

I assume that I have to adapt the official build 4.4 R3 ISO to use Mesa 10.4.2?
Or is Mesa 10.4.2 in the git repository so that I have to do a repo sync and build my own ISO?

[Florian Wiedemann]

Hi,

I managed to setup everything needed to build Android from source.
What steps are needed to build an ISO that will work with my i915 hardware and Mesa 10.4.2?
Or will it also work if I create a current kitkat / lollipop build that maybe contain fixes for the old Intel hardware?

Thanks and kind regards
Florian

Am Donnerstag, 3. September 2015 17:24:07 UTC+2 schrieb Florian Wiedemann:

[Mauro Rossi]

Hi Florian,

Could you provide your Dell system specs and a usual dmesg, logcat and dumpsys SurfaceFlinger,

the problem i saw was affecting i965 family, useful to know if also i915/i945 are affected. 

With mesa 10.5.0 initial version two different font problems appeared, can you confirm that you just see the second one (randomic)?

1) systematic problem: is affecting the fonts in the apps menu and the blue hints screen systematically, and it's like dark stains on the desktop GUI fonts. 
This issue appeared with 10.5 branching commit d1e21325cf3c394f5fb28808a113826ea3b7f54b, while did not appear with parents.

Are you also seeing this problem, when I last checked on 10.6 and later this was solved.

2) another defect is more randomic issue, which is an unwanted transparency affecting some characters in the drop down menu fonts and the apps fonts (e.g. Play Store).

Very strange that this last problem appears also prior to 10.5 branch commit, but not with mesa 10.4.6

If you see the second problem randomly and not the stains in Android desktop the bug you're facing is this one: https://code.google.com/p/android-x86/issues/detail?id=1740

I have a proposal, if you want to, for you to check if the remaining issue is already present in 10.4.7 and then move to 10.4.2 which shoulnd be affected for sure.

If you see the problem with 10.4.7 we could continue try bisecting between 10.4.6 and 10.4.7,

if you don't see the random problem at all with 10.4.7, then also the random problem has been caused by the 10.5.0 branching commit (my notes in the attachment)

I tried to find the offending commit, but for me (i965) the problem was too randomic,

if with your chipset it is more frequent that info can help very much to continue investigation.

Mauro

You can build mesa 10.4.7 using this github: https://github.com/maurossi/mesa/tree/10.4.7

here follows instructions:

cd [android-x86]

cd external/mesa

git remote add maurossi https://github.com/maurossi/mesa

git fetch maurossi 10.4.7

git checkout FETCH_HEAD

if you get some warnings preventing the checkout operation:

git reset --hard

git fetch maurossi 10.4.7

git checkout FETCH_HEAD

It should build, if you get some building errors drop me a PM and we will sort it out.

PS: for building mesa 10.4.2 you can use the same instructions with tree 10.4 (which is in fact updated to 10.4.2 version)

just by replacing '10.4.7' with '10.4'

[Florian Wiedemann]

Hi Mauro,

thanks for your detailed answer. I will do my best to provide all needed information and to do the explained steps. Attached please find the requested logfiles.
I must admit that I've chosen the wrong name for the thread. I learned from lspci that I have a i965 graphics chip and not a i915. I mixed it up with the kernel driver i915. Sorry for the confusion.

The specification of the notebook is:

Product name:     Dell Inspiron 1525
CPU:                     Intel Celeron Dual-Core T1400 1730MHz
RAM:                    2048 So-Dimm DDR2
HDD:                    160 GB SATA
Display:               15,4" WXGA (1280x800, 16:10=
DVD-RW:              yes
WLAN:                 yes

The detailed hardware specification is attached as lspci_android.txt and lspci_linux.txt.

As next step I will try to build ISO files of kitkat with Mesa 10.4.7 and report back if the hazy font problem still exists.

Thanks for your great support!

Kind regards
Florian



On Thursday, September 3, 2015 at 5:24:07 PM UTC+2, Florian Wiedemann wrote:

[Mike R]

Hey, I didn't come into groups to read the thread, replying by email, But I have a Dell 1505 and a Dell 1750, Not sure what could be the diff. between the 2 but Both run off the GM45 Chip set.... Just putting it out, Because, RC2-r3, and Lollipop 5.1.1 Works on my system, and just seems odd that dell would change the chip set in the middle (but not Unheard of by no means!!!!)....

Maybe will help, Maybe not 8-)

Mike

--
You received this message because you are subscribed to the Google Groups "Android-x86" group.
To unsubscribe from this group and stop receiving emails from it, send an email to android-x86...@googlegroups.com.
To post to this group, send email to andro...@googlegroups.com.
Visit this group at http://groups.google.com/group/android-x86.
For more options, visit https://groups.google.com/d/optout.

[rbg]

Mauro...FWIW

I have also solved the issue on my lollipop build when testing by just using a different launcher....

using this:

https://play.google.com/store/apps/details?id=com.l.launcher&hl=en

with default (Trebuchet) I see the issue on my older machine, reboot and select the Llauncher

and don't see the issue....

[Florian Wiedemann]

Hi Mauro,

with the i915 driver and kitkat 4.4.-r3 I get both font errors:

black stains on app drawer fonts and the blue hints and also random font problems in settings and Play Store.

Kind regards

Florian


Am Donnerstag, 3. September 2015 17:24:07 UTC+2 schrieb Florian Wiedemann:

[Miker1029]

Makes sense been using L.L. for 4+ years. 

Sent with AquaMail for Android
http://www.aqua-mail.com

[Povilas Staniulis]

I also live in Europe and I use git.android-x86.org
IIRC, SF mirror is outdated.

On 2015.09.12 21:13, Florian Wiedemann wrote:

Hi,

I'd like to build an ISO but have no luck with repo sync.

As I live in Europe I read that I should you the sourceforge URL:

repo init -u http://git.code.sf.net/p/android-x86/manifest -b android-x86

But the sync aborts with the following error message:

Fetching project platform/external/koush/Widgets
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
  0     0    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0curl: (22) The requested URL returned error: 404 Not Found
Server does not provide clone.bundle; ignoring.
fatal: repository 'http://git.code.sf.net/p/platform/external/koush/Widgets/' not found

When I use the following URL the download is like a EDGE connection (10-30 KB/s):

repo init -u http://git.android-x86.org/manifest -b kitkat-x86

Are there any working alternatives to get the source from?

Kind regards
Florian



Am Donnerstag, 3. September 2015 17:24:07 UTC+2 schrieb Florian Wiedemann:

--

[Florian Wiedemann]

Hi,

I'd like to build an ISO but have no luck with repo sync.

As I live in Europe I read that I should you the sourceforge URL:

repo init -u http://git.code.sf.net/p/android-x86/manifest -b kitkat-x86

But the sync aborts with the following error message:

Fetching project platform/external/koush/Widgets
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
  0     0    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0curl: (22) The requested URL returned error: 404 Not Found
Server does not provide clone.bundle; ignoring.
fatal: repository 'http://git.code.sf.net/p/platform/external/koush/Widgets/' not found

When I use the following URL the download is like a EDGE connection (10-30 KB/s):

repo init -u http://git.android-x86.org/manifest -b kitkat-x86

Are there any working alternatives to get the source from?

Kind regards
Florian

Am Donnerstag, 3. September 2015 17:24:07 UTC+2 schrieb Florian Wiedemann:

[Florian Wiedemann]

Hi Mauro,

I'm currently not able to do a repo sync as the git server does not work correctly.
Coud you by chance create a kitkat or lollipop build with Mesa 10.4.7 and upload it?
I would be eager to test and give feedback.

Regards
Florian

[Mauro Rossi]

Hi,

here is the link: https://drive.google.com/file/d/0B_OFHiIqgpSFZXM4LWxnRkdudG8/view?usp=sharing

I checked with Acer 5620 and there's no font artifact in GUI overlays and Playstore with 10.4.7

10.5.9 is affected almost systematically

10.6.7 latest, I need to check

With lollipop-x86 during the whole 10.6.x... 10.7.0devel...11.0.0devel histoiry of builds, if I remember correctly, in my tests the problem was less frequent, but still present

Uploading a lollipop-x86 iso with mesa 11.0 soon.

Since the 10.4.x is unaffected then:

1) the problems started exactly with the branch of 10.5.0devel and were never corrected in all 10.5.x

2) we can check the status on then next release 10.6.x (I think I saw it also there)

3) check with mesa 11 on lollipop-x86

M.

[Mauro Rossi]

Hi,

back again 10.6.7 on kitkat-x86 has a problem in the overlay menus, but "just sometimes", while desktop seems completely unaffected, so I would apply "the lesser evil principle" here

The defect I see on Acer 5620 is triggered by doing: 

[ALT]+[F1]

stop

start

[ALT]+[F7]

and using the mouse at top-right notification area and opening the panel

the fonts have thin vertical black strips.

It's really a mystery to me why this happens at second lauch of the GUI.

The kitkat-x86 build with mesa 11.0 release is unfortunately  affected by "black GUI screen" on i965GM,

but works with i945GM  (i915 driver), very intriguing... :-)

I need to check what happened because the black screen on i965GM is also affecting in the same way lollipop-x86 with mesa 11.0.

M.

[Mauro Rossi]

I need to check what happened because the black screen on i965GM is also affecting in the same way lollipop-x86 with mesa 11.0.

M.

It is the already known problem that "i915: add mock implementation of GL_OES_EGL_image_external" is needed for android-x86 

I just had forgot that it is needed for i965GM to work and It was misleading for me that for i965 GL_OES_EGL_image_external.

My guess is that i965Gm chipset relies on i915 driver, anyway I'm happy it was a false alarm.

Two additional builds for you to try

Mesa 10.6.x from mesa git:                 https://drive.google.com/file/d/0B_OFHiIqgpSFX0NVNVI2WFVxS28/view?usp=sharing

Mesa 11.0 release:            https://drive.google.com/file/d/0B_OFHiIqgpSFdW83WHFmanNQN2s/view?usp=sharing

NOTE: mesa 11 was built with Revert "nv50/ir: make edge splitting fix up phi node sources", avoiding tr1/c++11,  in order to build nouveau too

In my tests with 10.6.x and 11.0, GUI/desktop stains did never appear at first boot/GUI restart,

but Playstore's and overlay panels's text may become unreadable after some interaction of the user.

M.

[Mauro Rossi]

| I just had forgot that it is needed for i965GM to work and It was misleading for me that for i965 GL_OES_EGL_image_external.[?]

rephrasing

GL_OES_EGL_image_external had been already implemented in i965 driver, so I made the wrong assumption that it would suffice for i965GM chipset. 

GL_OES_EGL_image_external enabled in i915 is needed.

My guess is that i965Gm chipset relies on i915 driver, anyway I'm happy it was a false alarm.

M.

[Florian Wiedemann]

Hi Mauro,

thanks a lot for uploading the test builds. My test results so far are:

Build Mesa 10.4.7:
Works like a charm without any problems: No font glitches, tested with Play Store, OpenGL demos, Google Maps, Camera, Gallery, Browser, Youtube

Build Mesa 10.6.x from mesa git:
Booted but stopped at the Android animation, dmesg showed a lot of I/O. So could it be that the ISO is corrupt?

Build Mesa 11.0:
Booted and started GUI but repeating message "Unforutnately, Setup Wizard has stopped."

If I should provide further information (dmesg, logcat) of any of the test builds please let me know.

Kind Regards
Florian

[Mauro Rossi]

Hi Florian,

The two images booted fine on Acer 5620

I have uploaded parity files to enable you to check the files you have downloaded.

https://drive.google.com/file/d/0B_OFHiIqgpSFcTY0a0hSaXdYRWM/view?usp=sharing

https://drive.google.com/file/d/0B_OFHiIqgpSFOG42SmxheFNSMkU/view?usp=sharing

If you don't find par2 package you get it here: http://security.ubuntu.com/ubuntu/pool/universe/p/par2cmdline/par2_0.6.11-1_amd64.deb

Mauro

[Florian Wiedemann]

Hi Mauro,

the downloaded ISO files were ok:

 There are 1 recoverable files and 0 other files.
 The block size used was 218632 bytes.
 There are a total of 2000 data blocks.
 The total size of the data files is 437256192 bytes.

 Verifying source files:

 Target: "android_x86_k408_mesa_1067_19-sep-2015.iso" - found.

 All files are correct, repair is not required.

I wrote the ISO with Mesa 10.6.7 again with the Rufus tool to my USB stick and now I was able to boot and to start the GUI.
At the setup wizard I got a error message "Unfortunately, the process android.process.acore has stopped". But then everything worked. I didn't get the hazy font error during my tests (Play Store, Google Maps, App Drawer etc.).

Tomorrow I will test again your Mesa 11 test build.

Attached please find my logfiles for the test with the Mesa 10.6.7 build.

Kind regards
Florian

[Florian Wiedemann]

Hi Mauro,

now I've tested your Mesa 11 build from 20.9. With this build the apps Camera and Google Maps do not work. But that is known as I read in one of your posts.
Also with this build I get the hazy fonts problem as you can see in the attached screenshot.

Conclusion of my tests:
At least for my good old Dell Inspiron 1525 notebook the builds with Mesa version 10.4.7 and 10.6.7 worked without problems.

Kind regards
Florian

[Mauro Rossi]

Hi Florian,

We may be on the verge of solving this.

may I ask your help?

Using kitkat-x86 with mesa 11 it is possible to trace gl api, using the following procedure (it requires a separate linux environment)

I have uploaded egltrace.so prebuilt here: https://drive.google.com/file/d/0B_OFHiIqgpSFUzVjcUF0NmdMX0U/view?usp=sharing

Upload egltrace.so library to kitkat-x86 device under test

---------------------------------------------------------------------------

adb root

adb connect [IP]

adb push ./egltrace.so data/

adb shell chmod 0777 /data/egltrace.so

Tracing with apitrace egltrace.so

-------------------------------------------

[example procedure for Settings - tested and working on kitkat-x86]

adb root

PROCNAME=com.android.settings

adb shell setprop wrap.$PROCNAME LD_PRELOAD=/data/egltrace.so

adb shell setprop debug.apitrace.procname $PROCNAME

adb shell am start -S com.android.settings/.Settings

In a similar way it should be possible to trace other apps like the Browser, Maps, PlayStore.

If we could upload apitrace showing 0x500 error on https://bugs.freedesktop.org/show_bug.cgi?id=92323

that would speedup resolving the bug.

While opening the bug I was asked to git bisect, while it may the usual standard procedure to find the commit that introduced the problem for a software that can be built without touching the sources,  that is absolutely is not easy for the unstable build that mesa-10.5.0devel was for android targets, so not the best strategy here.

I have also tried on lollipop-x86, but apitrace wrapping does not work there, I could never get an apitrace on lollipop-x86.

On lollipop-x86,  I am logging by modifying libhwui code and using MESA_DEBUG, MESA_LOG_FILE and INTEL_DEBUG environment variables, still to try with mesa error.c

but that being said the apitrace can really give mesa developers the information they need.

Let me know if you or other people are willing to pick-up the apitraces with kitkat-x86.

Thanks for helping to solve this

 

Mauro

[Florian Wiedemann]

Hi Mauro,

of course I will try to deliver the trace file to support this great project with my limited possibilities. :-)

I can (hopefully!) delivert the trace file on Saturday or latest on Sunday.

It would be great if that annoying font bug could be solved.

Best regards
Florian

[Florian Wiedemann]

Hi Mauro,

attached please find my logcat from the apitrace. You will find some 0x500 errors. :-)
Interestingly this time I only get the hazy fonts in the Youtube app. In the Settings and the App Drawer I couldn't reproduce the error.

Please let me know if I can provide additional logfiles or do other tests.

Kind regards
Florian

[Mauro Rossi]

Hi,

I've collected some apitraces from kitkat-86 Settings app, but I could not apitrace visible artifacts.

https://drive.google.com/open?id=0B_OFHiIqgpSFSFBfSHlCWndZUHM
https://drive.google.com/open?id=0B_OFHiIqgpSFbUJsVktES1BaUms
https://drive.google.com/open?id=0B_OFHiIqgpSFalRwcnNaVW11cUU


maybe re-iterating with PlayStore and Browser I may be able to catch the artifact.

"Apitracing" the OneTimeSetup app, should provide the wanted trace, because artifacts are systematic at OneTimeSetup.

What is very annoying is that apitracing is not working on lollipop-x86 for me.

Mauro

[Florian Wiedemann]

Hi Mauro,

you could also test with the Youtube app. At my last tests I only got the font glitches with the Youtube app.

Kind regards
Florian

[pstglia]

Hi Mauro,

I did some debugging with this "Hazy Fonts" issue.

As I could check, the error code related to this problem (0X500 - GL_INVALID_ENUM) is being raised by external/mesa/src/mesa/main/bufferobj.c ( function bind_buffer_object).

   bindTarget = get_buffer_target(ctx, target);

   if (!bindTarget) {

      _mesa_error(ctx, GL_INVALID_ENUM, "glBindBufferARB(target 0x%x)", target);

      return;

   }

One of the received parameters by this function is target (in this case it's value is 0x88ec - GL_PIXEL_UNPACK_BUFFER / GL_PIXEL_UNPACK_BUFFER_EXT ). This target is tested inside get_buffer_target. If not valid, NULL is returned, which returns the 0x500 error we are facing.

The first thing get_buffer_target checks is if your context is an "OpenGL Desktop" type or "OpenGL ES 3.0". If not, it's considered invalid and NULL is returned:

   /* Other targets are only supported in desktop OpenGL and OpenGL ES 3.0.

    */

   if (!_mesa_is_desktop_gl(ctx) && !_mesa_is_gles3(ctx)

       && target != GL_ARRAY_BUFFER && target != GL_ELEMENT_ARRAY_BUFFER)

      return NULL;

I suggest you checking which type of context is being used here and at which point it is being set (I'd say is a GLES 2.0 one...) 

I would test removing this check and see what happens also...

Hope this helps,

Pstglia

[Mauro Rossi]

Hi,

thanks because I was having hard time to find where the _mesa_error could be generated and propagated,

in the meantime I've investigated Intel i965 dri by enabling INTEL_DEBUG environment variable in init.x86.rc

and by adding a lot of checkpoints, here the result.

We have a workaround and a restricted scope of commits that introduced the problem.

Here follows the report posted on https://bugs.freedesktop.org/show_bug.cgi?id=92323

If I comment _mesa_meta_pbo_TexSubImage() the hazi fonts are GONE!!!

M.

Hi,

I have added a few checkpoints in src/mesa/drivers/dri/i965/intel_texsubimage.c

and src/mesa/drivers/dri/i965/intel_texsubimage.c to see the value of bool variable named ok

The result is:

11-01 16:49:04.645  4817  4945 W INTEL-MESA: intelTexSubImage mesa_format MESA_FORMAT_A_UNORM8 target GL_TEXTURE_2D format GL_ALPHA type GL_UNSIGNED_BYTE level 0 1024x512x1

11-01 16:49:04.645  4817  4945 W INTEL-MESA: intelNewTextureObject

11-01 16:49:04.645  4817  4945 W INTEL-MESA: intelNewTextureImage

11-01 16:49:04.652  4817  4945 W INTEL-MESA: MAUROSSI intelTexSubImage _mesa_meta_pbo_TexSubImage() returns: 1

11-01 16:49:04.652  4817  4945 D libEGL  : [glTexSubImage2D] 0x500

this means that _mesa_meta_pbo_TexSubImage() reports to have completed its job.

In order to check if the font artifact were related to meta implementation I've skipped _mesa_meta_pbo_TexSubImage() completely with /*  ... */ and logged 

intel_texsubimage_tiled_memcpy() and mesa_store_texsubimage() execution.  

11-02 02:20:44.825  5078  5096 W INTEL-MESA: intel_texsubimage_tiled_memcpy: MAUROSSI 1st check failed, packing?

11-02 02:20:44.825  5078  5096 W INTEL-MESA: MAUROSSI intelTexSubImage intel_texsubimage_tiled_memcpy() returns: 0

11-02 02:20:44.826  5078  5096 W INTEL-MESA: MAUROSSI intelTexSubImage _mesa_store_texsubimage() executed.

The result I get by skipping _mesa_meta_pbo_TexSubImage() is that intel_texsubimage_tiled_memcpy() bails out because of packing,

_mesa_store_texsubimage() is executed and there are no font artifacts at all (triple checked).

This means that the offending commit lies among the following ones:

i965/tex_subimage: use meta instead of the blitter for PBO TexSubImage

i965/tex_image: Use meta for instead of the blitter PBO TexImage and GetTexImage

i965/pixel_read: Use meta_pbo_GetTexSubImage for PBO ReadPixels

meta: Add an implementation of GetTexSubImage for PBOs

meta: Add a BlitFramebuffers-based implementation of TexSubImage

Mauro

[Mauro Rossi]

The only problem is that with all DBG() entries, misteriously the GUI restarts at first installation,

with SIGSEGV MAPERR related to libc.so and some (v)printf function

I had to test with a preinstalled Lollipop-x86 and push gallium_dri.so i965_dri.so in /system/lib/dri and restart the GUI (stop+start).

I still have to try with minimal workaround patch attached for mesa 11.0.3 in lollipop-x86,

to people with X3100, X4500MHD please provide confirmation about hazi font disappearence.

Thanks

M.

[pstglia]

I still have to try with minimal workaround patch attached for mesa 11.0.3 in lollipop-x86,

to people with X3100, X4500MHD please provide confirmation about hazi font disappearence.

Thanks

Thanks for your efforts on solving this. I tested your patch with GMA X4500 Integrated graphics (DG41WV motherboard - the same I used to collect the info I sent above) and it worked (no more hazing fonts)

I also executed Antutu and played some youtube videos on integrated browser. No problems reported. Nice job!

* * * 

 

I looked one of the functions you commented out, and it seems to create/bind a buffer using the "GL_PIXEL_UNPACK_BUFFER" I reported above:

# ./src/mesa/drivers/common/meta_tex_subimage.c

# function: _mesa_meta_pbo_TexSubImage

   pbo_tex_image = create_texture_for_pbo(ctx, create_pbo,

                                          GL_PIXEL_UNPACK_BUFFER,

                                          width, full_height,

                                          format, type, pixels, packing,

                                          &pbo, &pbo_tex);

Seems the type of ctx + target used are not supported by OpenGL ES contexts < 3.0

Regards,

Pstglia

[stavros daliakopoulos]

On a laptop with GM45 graphics and opengl 2.0 I get hazy fonts in the latest 5.1.1 

release.

I undestand it is a bug with opengl 2.0 and android.

[fgdn17]

interesting info here:

http://feedback.wildfiregames.com/report/opengl/

[Mauro Rossi]

Il giorno lunedì 2 novembre 2015 16:53:18 UTC+1, pstglia ha scritto:

I still have to try with minimal workaround patch attached for mesa 11.0.3 in lollipop-x86,

to people with X3100, X4500MHD please provide confirmation about hazi font disappearence.

Thanks

Thanks for your efforts on solving this. I tested your patch with GMA X4500 Integrated graphics (DG41WV motherboard - the same I used to collect the info I sent above) and it worked (no more hazing fonts)

I also executed Antutu and played some youtube videos on integrated browser. No problems reported. Nice job!

Hi, could finally test OneTime Initialization and not see the garbled fonts.

M.

 

I looked one of the functions you commented out, and it seems to create/bind a buffer using the "GL_PIXEL_UNPACK_BUFFER" I reported above:

# ./src/mesa/drivers/common/meta_tex_subimage.c

# function: _mesa_meta_pbo_TexSubImage

   pbo_tex_image = create_texture_for_pbo(ctx, create_pbo,

                                          GL_PIXEL_UNPACK_BUFFER,

                                          width, full_height,

                                          format, type, pixels, packing,

                                          &pbo, &pbo_tex);

Seems the type of ctx + target used are not supported by OpenGL ES contexts < 3.0

Regards,

Pstglia

dumpsys SurfaceFlinger reports OpenGL ES 2.0  for X3100,

I don't understand why OpenGL ES 3.0 features, like packed PBOs, are used "internally" when OpenGL ES 2.0 is advertised to 

In my understanding _mesa_meta_pbo_TexSubImage() would be suboptimal for i965GM (gen4-5) of compared to intel_blit_texsubimage() , even if the latter can be used just for  2D images without Y tiling.

would our font scenario (which may be 2D images w/o Y tiling, but I'm not sure) would still benefit by using intel_blit_texsubimage() or by the time being, with packing, we would end up at  _mesa_store_texsubimage() i.e CPU involved for memory storage?

I think the original commiters know more than me, by a long shot...

M. 

[Steve Trotman]

i have an old thinkpad r61 intel gma950 that gets the bad fonts almost immediately on 4.4r3 and 5.1.1. if someone posts an iso with the fixes discussed im happy to test and report results

[Chih-Wei Huang]

Nice work, Mauro.

Jason, you are the author of the commit
"meta: Add a BlitFramebuffers-based implementation of TexSubImage".
Could you please comment it?

https://bugs.freedesktop.org/show_bug.cgi?id=92323

--
Chih-Wei
Android-x86 project
http://www.android-x86.org

[Mauro Rossi]

Hi Florian,
Jason Ekstrand asked if it's possibile to collect an apitrace,
on lollipop-x86 I am having problema to collect those,
could you just confirm if with YouTube app launch you see the artifacts systematically?
If so I will collect the apitrace for Youtube app with ad hoc procedure, this evening.
M.

[Mauro Rossi]

Il giorno martedì 3 novembre 2015 02:21:40 UTC+1, Steve Trotman ha scritto:
> i have an old thinkpad r61 intel gma950 that gets the bad fonts almost immediately on 4.4r3 and 5.1.1. if someone posts an iso with the fixes discussed im happy to test and report results

Hi Steve,
Coulf you please provide info about app generating the artifacts, when you say almost immediately are you referring to OneTimeInitialization or to a specific activity?

Apitrace would be appreciated where is possibile to collect them i.e. on kitkat-x86

M.

[Florian Wiedemann]

Hi Mauro,

the problem is that I couldn't reproduce the error with the YouTube app. I tried just intensively all kind of apps (Maps, Camera, Browser, Settings) but didn't see the hazy fonts.

I will try the next days if I can reproduce the problem with any app.

Kind regards
Florian