Lastest Froyo-x86 Mesa plateform mostly useless !!!

[Arobase40]

Hi developers,

I have spent a long time studying the mesa source code from lastest
Froyo-x86, and I have confirmation it is mostly useless with Intel
Laptops other than Intel Netbooks with i915 or closed drivers
(including GMA3150) and may be also Laptops with Radeon drivers. I
have no idea if laptops with Nvidea Nouveau drivers work or not, as I
don't have any...

I don't understand yet why it is supposed to work with Graphics Media
Accelerator integrated in Core i5 Processors while this does not work
with i915G, i945, GMA4500HD chipsets. I haven't tested with core i3
yet.

The code is sooo messy, that it's hard to read and understand : there
are some bugs in the nested define in the brw_reg.h file which
identify the graphical chipsets (and may be also intel_chipset.h), and
some files which are supposed to support i965 drivers are simple links
to the same file which support i915 drivers. ^^

What I did to verify a theory was to modify a test in /external/mesa/
src/egl/driver/android/droid.c file so as to force the load of the
i915 mesa driver with a GMA4500HD chipset, and at least I got a "nice"
display but the result is just as what Stefan S described in here
(capped, jerky and very slow display) :

http://groups.google.com/group/android-x86/browse_thread/thread/29df370b4f4b4f0a

So if you have laptops with i915G, i945, GMA4500HD and related
chipsets and unless you have time to debug the source code, you'd
better stay away from the lastest Froyo version identified as FRG83G,
unless you downgrade the mesa code.

I'll try to work some more time on this source code a bit, but not
sure I'd want to waste too long time on it. It's a very hard task
since I didn't created it and have little control on it... :(

Sorry to the authors of this, but you may need to simplify this source
code... ^^

[sknobl]

I can confirm that error - I switched back to android-x86-2.2 version
to get usable builds.

Best regards,
Stefan

On 2 Mai, 02:42, Arobase40 <arobas...@gmail.com> wrote:
> Hi developers,
>
> I have spent a long time studying the mesa source code from lastest
> Froyo-x86, and I have confirmation it is mostly useless with Intel
> Laptops other than Intel Netbooks with i915 or closed drivers
> (including GMA3150) and may be also Laptops with Radeon drivers. I
> have no idea if laptops with Nvidea Nouveau drivers work or not, as I
> don't have any...
>
> I don't understand yet why it is supposed to work with Graphics Media
> Accelerator integrated in Core i5 Processors  while this does not work
> with i915G, i945, GMA4500HD chipsets. I haven't tested with core i3
> yet.
>
> The code is sooo messy, that it's hard to read and understand : there
> are some bugs in the nested define in the brw_reg.h file which
> identify the graphical chipsets (and may be also intel_chipset.h), and
> some files which are supposed to support i965 drivers are simple links
> to the same file which support i915 drivers. ^^
>
> What I did to verify a theory was to modify a test in /external/mesa/
> src/egl/driver/android/droid.c file so as to force the load of the
> i915 mesa driver with a GMA4500HD chipset, and at least I got a "nice"
> display but the result is just as what Stefan S described in here
> (capped, jerky and very slow display) :
>

> http://groups.google.com/group/android-x86/browse_thread/thread/29df3...

[fuzzy7k]

Well, it's progress. Mostly useless is better than totally useless.

Status report, I periodically try to test out the R600G support. I've
yet to get it to compile unless I download the tree from the froyo-
next repo. Then, it will compile but I have noticed no speedup from
dissabling mesa entirely. I remember Chih-Wei saying something about
work moving toward honycomb so I've kept my mouth shut. Also, lately
I've been unable to compile an image that will boot in normal mode.
Same image will boot in debug mode but I've not looked into why yet.

Here is the tail of the log during R600G build failure.

target StaticLib: libmesa_gallium (out/target/product/tx2500/obj/
STATIC_LIBRARIES/libmesa_gallium_intermediates/libmesa_gallium.a)
target SharedLib: libGLES_mesa (out/target/product/tx2500/obj/
SHARED_LIBRARIES/libGLES_mesa_intermediates/LINKED/libGLES_mesa.so)
out/target/product/tx2500/obj/STATIC_LIBRARIES/
libmesa_st_mesa_intermediates/libmesa_st_mesa.a(vbo_exec_api.o): In
function `vbo_exec_FlushVertices':
external/mesa/src/mesa/vbo/vbo_exec_api.c:990: undefined reference to
`vbo_exec_FlushVertices_internal'
collect2: ld returned 1 exit status
make: *** [out/target/product/tx2500/obj/SHARED_LIBRARIES/
libGLES_mesa_intermediates/LINKED/libGLES_mesa.so] Error 1

On May 1, 10:42 pm, Arobase40 <arobas...@gmail.com> wrote:
> Hi developers,
>
> I have spent a long time studying the mesa source code from lastest
> Froyo-x86, and I have confirmation it is mostly useless with Intel
> Laptops other than Intel Netbooks with i915 or closed drivers
> (including GMA3150) and may be also Laptops with Radeon drivers. I
> have no idea if laptops with Nvidea Nouveau drivers work or not, as I
> don't have any...
>
> I don't understand yet why it is supposed to work with Graphics Media
> Accelerator integrated in Core i5 Processors  while this does not work
> with i915G, i945, GMA4500HD chipsets. I haven't tested with core i3
> yet.
>
> The code is sooo messy, that it's hard to read and understand : there
> are some bugs in the nested define in the brw_reg.h file which
> identify the graphical chipsets (and may be also intel_chipset.h), and
> some files which are supposed to support i965 drivers are simple links
> to the same file which support i915 drivers. ^^
>
> What I did to verify a theory was to modify a test in /external/mesa/
> src/egl/driver/android/droid.c file so as to force the load of the
> i915 mesa driver with a GMA4500HD chipset, and at least I got a "nice"
> display but the result is just as what Stefan S described in here
> (capped, jerky and very slow display) :
>

> http://groups.google.com/group/android-x86/browse_thread/thread/29df3...

[Arobase40]

Some more aspects about this next Froyo version :

- About Mesa again, I haven't found any parameters on how to
initialize the missing chipsets. These are calls to i915 drivers and
Readon drivers, with parameters to make them work but none about
Nouveau chipsets and i915G, i945, GMA4500HD chipsets.
For what I have seen most of the source codes comes from Tungsten
Graphics and from what I know about this company is that it has been
sold to VMWare and the development has just been abandoned and there
won't be any more support about Mesa from them... I hope I'm wrong !!!
So I don't understand what was the need to integrate these codes, if
we can't expect to have further support ???

- The second aspect of the FRG83G build is that sounds like the
physical keys for Back, Home, Menu are not working anymore ? Is this
something I did wrong and anyone can confirm this or invalidate my
words ?
In fact, even the physical keyboard isn't working anymore, as I can't
type the wep or wpa key for wifi connexion, nor can I type anything in
the Browser or in any applications. So the first thing I have to do is
to launch my own Softkeys app and to activate the SoftKeyboard in
order to use my tablet PC just as a Tablet... ^^ Not really my
purpose, or I would stay with my tablets.

- I replaced the existing Mesa drivers with the old one, and used the
hid-multitouch drivers with kernel 2.6.38 and even though it works,
looks like it is not as responsive as with the older one with kernel
2.6.35...

- Finally, setkeycodes isn't either working or may be I don't know how
to use it. So I guess I have to ask Al Sutton, as I have error
messages about outbound keys, or something like that. I have red the
documentation and I thought it was the same syntax as under linux.

I don't know if Chih Wei is planning to move toward Gingerbread or
Honeycomb, but I guess many people are still willing to stay with
Froyo, no ???

At least, untill these two newcomers are really ready to use.

Please, don't abandoned Froyo support untill it is totally stable ! :(

[Yi Sun]

It seems that the issue is because someone has move the
vbo_exec_FlushVertices_internal to a private function. And we may not
have right mesa configured for R6000G. Anyway, I did following change
and the build passed. But when I load the image, nothing change, it is
still slow on my AMD device.

diff --git a/src/mesa/vbo/vbo_exec_api.c b/src/mesa/vbo/vbo_exec_api.c
index fcd544d..e1bfb87 100644
--- a/src/mesa/vbo/vbo_exec_api.c
+++ b/src/mesa/vbo/vbo_exec_api.c
@@ -429,6 +429,23 @@ do {
\
#include "vbo_attrib_tmp.h"

+/**
+ * Flush (draw) vertices.
+ * \param unmap - leave VBO unmapped after flushing?
+ */
+static void
+vbo_exec_FlushVertices_internal(struct vbo_exec_context *exec, GLboolean unmap)
+{
+ if (exec->vtx.vert_count || unmap) {
+ vbo_exec_vtx_flush( exec, unmap );
+ }
+
+ if (exec->vtx.vertex_size) {
+ vbo_exec_copy_to_current( exec );
+ reset_attrfv( exec );
+ }
+}
+
#if FEATURE_beginend

@@ -531,7 +548,7 @@ static void GLAPIENTRY vbo_exec_EvalPoint2( GLint
i, GLint j )

#endif /* FEATURE_evaluators */

-
+#if 0
/**
* Flush (draw) vertices.
* \param unmap - leave VBO unmapped after flushing?
@@ -548,6 +565,7 @@ vbo_exec_FlushVertices_internal(struct
vbo_exec_context *exec, GLboolean unmap)
reset_attrfv( exec );
}
}
+#endif

> --
> You received this message because you are subscribed to the Google Groups "Android-x86" group.
> To post to this group, send email to andro...@googlegroups.com.
> To unsubscribe from this group, send email to android-x86...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/android-x86?hl=en.
>
>

--
Android-x86
http://www.android-x86.org

[Arobase40]

I think the best thing to do is to downgrade the Mesa platform to what
it was before, as there is already much work to do in the other parts
of Froyo and Android-x86 project is loosing a big part of its interest
and "some" developers, and specifically the new comers who won't
develop for Intel Atom based laptops...

I still have a copy of the old version, but it won't be easy for the
others to downgrade by themself.

[Yi Sun]

Let me to try something else, and see if it help or not.
Yi

[Yi Sun]

Since my device is only a prototype board, I do encourage people to try this patch out on other devices, to see if it is useful
Yi

[Arobase40]

Well, wish you good luck !!!

As I said above, even if Radeon Mesa support is buggy, at least you
have some initialization codes for it, while the support for i965C or
related seems to lead to a dead end. ^^

I just gave up for the moment as this is too much work specially when
you are not the initiator of the plateform.

If I feel brave enough and have enough time, I'll try to compare it
with what they've done on the linux side, if they share some common
codes. ^^

If not, I'll try to work on the PM codes I sent you some weeks ago...

[fuzzy7k]

Has anyone tried the latest froyo-next branch? I know Chih-Wei said
not to use it after the kernel and mesa was last bumped, but the
development work on that branch slowed down after that point and
hasn't seen any new work in about a month. Seems like Chia-I reached a
good stopping point? For me (R600G), it doesn't boost performance, but
it does compile without problems.

[Arobase40]

Nope...

I just thought that the FRG83G build was a result of the froyo-next
branch. Sounds like I also tried the "lastest froyo-next branch" after
the failed of the FRG83G and got EXACTLY the same result.
That why I just revert to the old version and get stuck on it since
now, trying in parallel to debug the FRG3G mesa version with the above
explained conclusions.

BTW, I don't have problem compiling the FRG83G version !!! It's just
that it won't work as I just have a black display. ^^ sigh

I also tried to force and fake the auto-detect mechanism to work with
i915G, i945, GMA4500HD chipsets letting him "think" it was a i915
chipset and if I got a working display, it was REALLYYY TOOO slowww to
be usable.

Another word, when you said : "I remember Chih-Wei saying something
about work moving toward honycomb so I've kept my mouth shut.", I
guess you were talking about cyanogenmod, isn't it ???

if so, don't forget cyanogenmod don't make use of Mesa, but OpenGL...
^^

At the moment, I'm creating a build for the Fujitsu P1610, P1510
Tablet PC and got a display with WiFi and Bluetooth working, even
though HW Acceleration is disabled yet.
I have to add serial touchscreen support, and after that (or in
between) I'll try to give FRG83G a last chance... ^^ ;-)

because, when I said it was a mess, IT IS REALLY a mess in THIS Mesa
source code ! :(

It's a long time I didn't dig my head so deeply in that kind of a
situation, as I'm not as patient anymore as I was "before"...

I got some Mesa source codes from Linux and I'll see if I can merge
some parts of the codes, and it may be some kind of a patchwork... ^^

[fuzzy7k]

> BTW, I don't have problem compiling the FRG83G version !!! It's just
> that it won't work as I just have a black display. ^^ sigh

I understand this, I was just thinking maybe if it fixed my build
problem, maybe it would fix your run problem.

> Another word, when you said : "I remember Chih-Wei saying something
> about work moving toward honycomb so I've kept my mouth shut.", I
> guess you were talking about cyanogenmod, isn't it ???
>
> if so, don't forget cyanogenmod don't make use of Mesa, but OpenGL...
> ^^

No, I was not talking about cyanogenmod. The following is the Chih-Wei
post I was referring to....

> From: Chih-Wei Huang <cwhu...@android-x86.org>
> Date: Wed, 13 Apr 2011 09:23:19 +0800
> Subject: Re: Major changes in froyo-x86 branch
> Oops! Did I mention any special action
> to get mesa in THIS post?
>
> Everything is now in our repository and
> what you need to do is just repo sync.
>
> The olv's mesa repository has entered
> the next stage which aimed to honeycomb.

Now, it doesn't make sense to me that honeycomb work would be taking
place in "froyo-next", so I'm thinking maybe he has just halted work
on this branch to start working on a honeycomb branch. And on that
note, work did not halt until one week after Chih-Wei's post. In that
time there were a lot of additions and code cleanups.

http://cgit.freedesktop.org/~olv/mesa/log/?h=android-froyo-next

So, I suggest trying out the latest froyo-next branch, commit
65823206d9254c78cba3c2443a7c92661ea6a8f9.
Also, if I'm not mistaken, the build ID is a google thing, which would
define things like API's and such. Since mesa is not a google thing,
it should not define mesa anything, but rather mesa should be
configured to work with this build ID.

> ...
>
> read more »

[Arobase40]

> No, I was not talking about cyanogenmod. The following is the Chih-Wei
> post I was referring to....
>
> > From: Chih-Wei Huang <cwhu...@android-x86.org>
> > Date: Wed, 13 Apr 2011 09:23:19 +0800
> > Subject: Re: Major changes in froyo-x86 branch
> > Oops! Did I mention any special action
> > to get mesa in THIS post?
>
> > Everything is now in our repository and
> > what you need to do is just repo sync.
>
> > The olv's mesa repository has entered
> > the next stage which aimed to honeycomb.

Oh yeah, I didn't even noticed as I thought he made a mistake when
talking about HC, as how could he start working on HC when google
didn't even release the source code since nowadays...

>
> Now, it doesn't make sense to me that honeycomb work would be taking
> place in "froyo-next", so I'm thinking maybe he has just halted work
> on this branch to start working on a honeycomb branch.

In fact, I thought he was working on the gingerbread mesa platform...
^^ :(

> And on that
> note, work did not halt until one week after Chih-Wei's post. In that
> time there were a lot of additions and code cleanups.
>
> http://cgit.freedesktop.org/~olv/mesa/log/?h=android-froyo-next

Okeyyy, I didn't try continuing to upgrade as I thought FRG83G build
just merged the result of froyo-next... ^^

>
> So, I suggest trying out the latest froyo-next branch, commit
> 65823206d9254c78cba3c2443a7c92661ea6a8f9.
> Also, if I'm not mistaken, the build ID is a google thing, which would
> define things like API's and such. Since mesa is not a google thing,
> it should not define mesa anything, but rather mesa should be
> configured to work with this build ID.

Yes, but android-x86 is not a google think either. :p

I have tried to repo sync many times over FRG83G build and did not
find any changes. That's why I thought Mesa development was over and
as I haven't seen any changes on the official web site for a long
time. ^_^

And I also thought android-x86 had its own way to upgrade and with its
own features.

Don't understand why nobody told me that before so I wouldn't waste my
time on this and I would have continue the development of my own build
and the merging of Cyanogenmod features with android-x86. blahh

Do you know if the auto-detect mechanism is working or not ?

Thanks you so much anyway for the information.

I'll try the update, but after a backup of my actual source code, of
course... ;-)

> ...
>
> plus de détails »

[Arobase40]

Well, I did the update from froyo-next and from I have tested I can't
tell there are much progresses... ^^ :(

At least, no more black screen,but "just" a garbled display as it was
before... some months ago.

So I revert back to the old version.

I'll see if I have time to study the changes they made and why there
are not working.

[Arobase40]

I haven't got time to analyze the modified mesa source code from froyo-
next yet, but looks like the latest changes are dated from April 20th
2011... ^^

So I better understand why there were no real changes with previous
codes.

> ...
>
> plus de détails »