This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

SGX slow on the DM3730 with 2.6.37 kernel in Gingerbread based Rowboat

Other Parts Discussed in Thread: OMAP3530, DM3730, TSC2046

Hi,

The folks on #rowboat suggested I logged an entry here regarding some weird behavior I am seeing.

The current Gingerbread based Rowboat using a 2.6.37 kernel experiences a slow redraw frame rate on the DM3730. This behavior doesn't happen if I goto a OMAP3530 system or if I revert to the earlier release of Rowboat using the 2.6.32 kernel.

Setup:

Beagleboard xM (Rev C) for the DM3730 using my expansion board to provide a display (http://www.hy-research.com/beagle_lcd2.html). It is a basic 24 bit DPI interface with the TSC2046 as the touch screen controller on SPI4.Other then timing, the display should be equivalent to the on board DVI transmitter.

Beagleboard classic (Rev C, tried different ones, they all worked the same) for the OMAP3530 using the same expansion board.

For the 2.6.37 kernel, I am at commit 06ebbba35a2f2026b25240cea052d579736a697a paired with commit 3c2314a9fe811f7ac074608c5c0a214c49a22f8c for the SGX drivers under hardware/ti/sgx.

For the 2.6.32 kernel, I am at commit d9b04d8b50d19b7056ee051106057f703877dabc paired with commit 3674b877cb6b7c6e0267cb46c02c0f2ef9958a7e for the SGX drivers under hardware/ti/sgx.

Tools to reproduced

Droid2DTest-debug.apk from http://www.typhon4android.org/androidbugs/droid2dtest-21/

Symptoms:

Using the DM3730 and the 2.6.37 kernel drivers built with ES5.x, I see the hardware clock fps at 60 with some jitter and the other number at 10fps.

Changing things to a OMAP3530 but keeping the 2.6.37 kernel with drivers built with ES3.x, I see the hardware clock fps at 60 with jitter and the other number the same.

If I manually disable hardware acceleration on the DM3730 via egl.conf, I see the hardware clock fps remain at 60 and the software one jump between 20 to 40fps depending on system load. (as expected)

If I revert to the 2.6.32 kernel, the DM3730 setup works at 60 and 60 (as expected).

This slow fps manifests itself as a jerky mouse pointer if a mouse is used or an apparently laggy interface; the test app is used to provide a metric on this problem.Since there isn't any other versions of the SGX driver on the repo and it is closed source, I cannot debug this further.

  • Hello Hunyue Yau,

        I tried reproducing the problem on beagle-xm with 2.6.37 kernel with the application pointed in your post. I find no issues and observe ~60fps for both clock and display.

    Could you please try with prebuilt images from the following -

    http://software-dl.ti.com/dsps/dsps_public_sw/sdo_tii/TI_Android_DevKit/TI_Android_GingerBread_2_3_4_DevKit_2_1/index_FDS.html

    Also please refer manifest file (TI-Android-GingerBread-2.3.4-DevKit-2.1.xml) for more details.

    Regards,

    Aditya

  • Hi,

    Loading the binaries from the TI_Android_GingerBread_2_3_4_DevKit_2_1.tar.gz file on that page you referred to does indeed work.

    However, the XML file in that tarball is different from the one in Sources tarball in the same page which in turn is different from the one I got from GIT.

    What is the authoritative XML to use since even the 2 from the same page don't match?

    I haven't had time to go through it line by line and compare the different git repositories to see what the differences are.

    Thanks for looking into it.

  • The prebuilt images were built based on  TI-Android-GingerBread-2.3.4-DevKit-2.1.xml.

    Please sync your sources as per the manifest on release page and try.

    Please refer http://processors.wiki.ti.com/index.php/TI-Android-GingerBread-2.3.4-DevKit-2.1_DeveloperGuide#Source_locations for more details.

    Regards,

    Aditya

  • I went through the manifest line by line and checked each git repo and they are the same point. Apparently, some of the XML going around use tags instead of git hashes.

    (In case it matters -

    I went line by line through the XML file listed above.

    For each line, I looked at the package and path attributes to determine where that git repo is checked out to.

    I got that repo and do a git log -n 1 and compare that hash to the one shown in the revision attribute.

    )

    Anything else to check?

  • I would suggest you to sync your sources as per the manifest and try.

    Please refer developer guide which was pointed in the previous post for details.

    Regards,

    Aditya