OMAP3530: underrun!!! (at least 30.640 ms long) ... WHY????

   How often do you reload glViewport()?  Every frame?

   I have not seen any app attempt to reload glViewport() mid-scene in this way on the SGX and I suspect it is causing problems with the PowerVR's deferred rendering architecture.  I recommend you search the PowerVR documentation for any mention of this.  Have you seen this done anywhere in the SDK demos or the training course examples?  It would be a good experiment to modify one of them to see if you get the same problem.

   There are examples of reloading glViewport() to render to an FBO or pbuffer, and that approach would probably solve the problem. Switch your render target to an FBO to render your 2D to a texture, then use that texture in your 3D scene.  This way, the viewport for the framebuffer is constant

http://wiki.davincidsp.com/index.php/Render_to_Texture_with_OpenGL_ES

Regards, Clay

Clay,

We do reload the glVeiwport on every frame but I tried disabling that and it made no difference.  It definatly has something to do with a large texture being posted at about 2-3Hz.

Also, I am only seeing about 10Hz on my frame rate for a pretty busy VGA screen.  Does this seem right?

Thanks,

Ken

Ken,

    Loading and Binding large textures is expensive.  glTexImage2D must copy the texture and glBindTexture performs the twiddling.  You should probably reduce their size, load frequency or look at using texture streaming.

http://wiki.davincidsp.com/index.php/OpenGLES_Texture_Streaming_-_bc-cat_User_Guide

Regards, Clay

I checked the SGX speed by placing the following lines in GFX_Linux_KM/services4/system/omap3430/sysuils_linux.c:EnableSGXClocks()

    if (atomic_read(&psSysSpecData->sSGXClocksEnabled) != 0)
    {
        return PVRSRV_OK;
    }

    PVR_DPF((PVR_DBG_MESSAGE, "EnableSGXClocks: Enabling SGX Clocks"));
        /*
        rateMpu = clk_get_rate(psSysSpecData->psMPU_CK);
        printk(KERN_ERR "EnableSGXClocks: CPU Clock is %luMhz", HZ_TO_MHZ(rateMpu));
        */

#if defined(DEBUG)
    {
       
        IMG_UINT32 rate = clk_get_rate(psSysSpecData->psMPU_CK);
        PVR_DPF((PVR_DBG_MESSAGE, "EnableSGXClocks: CPU Clock is %dMhz", HZ_TO_MHZ(rate)));
    }
#endif

....

    res = clk_set_rate(psSysSpecData->psSGX_FCK, lNewRate);
    if (res < 0)
    {
        PVR_DPF((PVR_DBG_ERROR, "EnableSGXClocks: Couldn't set SGX function clock rate (%d)", res));
        return PVRSRV_ERROR_GENERIC;
    }
        /*
        rateSgx = clk_get_rate(psSysSpecData->psSGX_FCK);
        printk(KERN_ERR "EnableSGXClocks: SGX Functional Clock is %luMhz", HZ_TO_MHZ(rateSgx));
        */
#if defined(DEBUG)
    {
       
        IMG_UINT32 rate = clk_get_rate(psSysSpecData->psSGX_FCK);
        PVR_DPF((PVR_DBG_MESSAGE, "EnableSGXClocks: SGX Functional Clock is %dMhz", HZ_TO_MHZ(rate)));
    }
#endif

The output of these changes seem to happen at or near frame rate.

EnableSGXClocks: CPU Clock is 0Mhz
EnableSGXClocks: SGX Functional Clock is 110Mhz
EnableSGXClocks: CPU Clock is 0Mhz
EnableSGXClocks: SGX Functional Clock is 110Mhz
EnableSGXClocks: CPU Clock is 0Mhz
EnableSGXClocks: SGX Functional Clock is 110Mhz
EnableSGXClocks: CPU Clock is 0Mhz
EnableSGXClocks: SGX Functional Clock is 110Mhz
EnableSGXClocks: CPU Clock is 0Mhz
EnableSGXClocks: SGX Functional Clock is 110Mhz
EnableSGXClocks: CPU Clock is 0Mhz
EnableSGXClocks: SGX Functional Clock is 110Mhz
EnableSGXClocks: CPU Clock is 0Mhz
EnableSGXClocks: SGX Functional Clock is 110Mhz
EnableSGXClocks: CPU Clock is 0Mhz
EnableSGXClocks: SGX Functional Clock is 110Mhz

Why is the EnableSGXClocks called so often and is there something I can do to eliminate it.  Also, is it commond for the CPU clock to be ZERO?

Thanks,