Mupen textures on Pandora

[scorpio16v]

As I've shown in my last video, Mupen64plus - AE has some problems with textures on the Pandora's Android GB port with Rice Video-Plugin.
Config is :
Pandora 512 MB rebirth edition OC@800MHz
Notaz Cyanogenmod port of Gingerbread 2.3.7
Mupen64plus - Android Edition last test version, posted in beta thread.
Rice Video with enabled texture import and Risio's retro mario beta. But smeared, wrong textures and missing shadows are also on stock textures.

logcat

http://db.tt/V2uxQYdv

Latest official beta shows the same result, tested with testversion because of the better performance.

[Paul]

Thanks for the logcat, I added the Pandora to the profiler so the missing shadows and stars should be fixed now, assuming it takes the same offset values as other OMAP devices.  I also tried using highp for the shader texture vertex instead of mediump, so hopefully that will fix the texture smearing bug (let me know).  Internet has been out all day so FTP'd it from my phone (hopefully that went OK).

HighP Build

[scorpio16v]

No difference between this and the other build, sorry. But it might be that the error is in Gingerbread port, not in your code.

attached, the new logcat.

[Paul]

OK, I'll post a series of builds later to see if we can zero in on the correct offset values for the shadows.  Not sure how to fix the smearing bug though.

[Paul]

This build has highp on all the texture coordinate variables (instead of just the  vTexCoord0 ones). Hopefully it will fix the smearing:

High Precision Test

I'll work on those tests for the missing shadow bug next.

[scorpio16v]

The result is the same on Rice-Plugin and Mario, like shown in the video, I've posted some days before.
The graphic output on gles2n64 was OK on the former version, so no need to fix something.

[Paul]

Oh well.  Here are those offset tests with the various values that seem to cover most devices.  Ignore the fact that the profile names may refer to completely different hardware.. it's just a way to keep them separate.

Pandora in QALCOMM profile
Pandora in IMAP profile
Pandora in TEGRA2 profile

If one or more of these settings are close but not quite right (flickery shadows, etc) let me know and I can zero in closer on the correct values.  If none are correct, let me know that too (could be that the correct values for Pandora are completely different than the "normal" ones).

[Paul]

The graphic output on gles2n64 was OK on the former version, so no need to fix something.

Oh.. were the shadows still missing in gles2n64 too, or just in gles2rice?

[scorpio16v]

No, it's only a Rice problem. Your profile for gles2n64 is OK for Pandora.

Just to show the graphics on the regulare texture. In my video, it's a custom one, so maybe hard to compare the wrong textures for you.

https://dl.dropbox.com/u/17774073/20120708_211859.jpg

The castle's color is grey, the blur effect is only, because it's a foto, not a screenshoot.
You can see only 2 stars at the bridge, regulare texture are 5, i think.
In the castle are no stars at the doors. The big carpet is missing. Floor looks striped. No shadows.

[Paul]

Weird the shadow problem doesn't happen for gles2n64, only gles2rice (the code is virtually the same, just arranged a little differently).  I assume the OMAP settings (my first reply) are correct for gles2n64 to get the shadows visible, right?

The smearing problem always was unique to gles2rice, so I'm not surprised that problem doesn't affect gles2n64.

I suppose one possibility is that somehow a cached version of gles2rice is being packaged into the APK rather than the current version.  I recall that happened a while back for no apparent reason, then it somehow fixed itself.  I think I'll post a build with some log output to rule that out as the cause.

[scorpio16v]

One sidenote, nothing to do with the problem, but maybe helpful for people with continous resume to home on start.
On Pandora, on both plugins, the emulator resumed to home on start, if RGB_8888 Mode is enabled.
Happened on the last test and the official beta.