TDA2SX: how about the gpu performance ? such as BW and render capability

Hi,

May I know is there any updates?

Thanks and Best Regards,

Cherry

Cherry,

Sorry, the issue description is not very clear.

What exact do you mean by app would break down once OpenGL call glDrawArrays?

Is there a reference to a testcase that can be reproduced on the TI EVM? Or some logs to indicate the problem?

We need a reference testcase or logs or a clear description of the problem to proceed.

Regards

Karthik

Hi Karthik,

The updates from customer are as follows:

1, the usecase (partly) which app used:
    Capture -> Dup
    Dup -> Sync_disp -> Gate_camera_display -> Merge_display -> SgxFrmcpy (A15) -> Display
  
    in SgxFrmCpy link, the customer put the frameRender function in SgxFrmCpy processFrame function, i.e. SgxFrmCpy link would call frameRender function to render the screen, such as the video captured by cameras and ui widgets which rendered by openGL in this function when SgxFrmCpy link get new data from previous link.

2，the app could process user's touch operation on touch screen, and switch to responding ui, that means frameRender could render different ui widgets when sgxFrmCpy link receive new image data.

3，the app would break down when frameRender function try to render the setting ui widgets almost on every time.

4，when app breaks down, the customer found it's call openGL glDrawArrays function to render some icons.

5, the icon comes from jpg image file in tf card which mounted on board, the processdure as below:
    1, read image from tfcard;
    2, bind the image to tex (a data struct which is used to render iamge and word);
    3, call openGL api function to render the image as icon, resize the image and draw on the framebuffer;

6，there are 8 icons need to render, but app is always killed by linux os. so, the customer wanna know the details and performance of GPU.

btw:
this is the function to render the icons:

void className::fixed_paint_tex(TEX tex,float px,float py,float lx,float ly,float xx,float yy,float winWidth,float winHeight,float width,float height,bool stretch,float angle,float opacity)
{

    GLboolean  cull_enable=glIsEnabled(GL_CULL_FACE );
    glEnable(GL_CULL_FACE);
    glCullFace(GL_BACK);

    float x0,y0,x1,y1;
    if(stretch){
        x0=(xx-winWidth/2)/(winWidth/2);
        y0=(winHeight/2-yy)/(winHeight/2);
        x1=(xx+width-winWidth/2)/(winWidth/2);
        y1=(winHeight/2-(yy+height))/(winHeight/2);
    }else{
        x0=(xx-winWidth/2)/(winWidth/2);
        y0=(winHeight/2-yy)/(winHeight/2);
        x1=(xx+lx-winWidth/2)/(winWidth/2);
        y1=(winHeight/2-(yy+ly))/(winHeight/2);
    }

    //////////////////////////////////////////////////////////////////
    float vertices[30+1]={
        x0, y0, -0.1,  px/(float)tex.tex_width, py/(float)tex.tex_height,      //   0        2/3
        x0, y1, -0.1,  px/(float)tex.tex_width, (py+ly)/(float)tex.tex_height, //
        x1, y0, -0.1, (px+lx)/(float)tex.tex_width, py/(float)tex.tex_height,      //
        x1, y0, -0.1, (px+lx)/(float)tex.tex_width, py/(float)tex.tex_height,      //
        x0, y1, -0.1,  px/(float)tex.tex_width, (py+ly)/(float)tex.tex_height, //
        x1, y1, -0.1, (px+lx)/(float)tex.tex_width, (py+ly)/(float)tex.tex_height,  //   1/4       5
        0
    };

    float arc_angle=angle*M_PI/180.0f;

    float  mid_x=(x0+x1)/2.0;
    float  mid_y=(y0+y1)/2.0;

    if(angle > 0.000001 ||  angle < -0.000001)   /* omit calculation when angle == 0; */
    {
        for(int i=0;i<30;i+=5)
        {
            xx=(vertices[i]-mid_x)*cos(arc_angle)-(vertices[i+1]-mid_y)*sin(arc_angle)+mid_x;
            yy=(vertices[i]-mid_x)*sin(arc_angle)+(vertices[i+1]-mid_y)*cos(arc_angle)+mid_y;
            vertices[i]=xx;
            vertices[i+1]=yy;
        }
    }


    GLuint program = shaderObj->get_program();
    glUseProgram(program);
    glUniform1i(glGetUniformLocation(program, "sTexture"), 0);
    glUniform1f(glGetUniformLocation(program, "opacity"), opacity);

    glActiveTexture(GL_TEXTURE0);
    glBindTexture(GL_TEXTURE_2D, tex.texid);

    glDisable(GL_DEPTH_TEST);

    if(b_blend)
    {
        glEnable(GL_BLEND);
        glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
    }
    else
        glDisable(GL_BLEND);

    glBindBuffer(GL_ARRAY_BUFFER, 0);
    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
    glEnableVertexAttribArray(0);
    glEnableVertexAttribArray(1);

    glVertexAttribPointer(0, 3, GL_FLOAT, GL_FALSE, 20, vertices);
    glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, 20, vertices+3);

    glDrawArrays(GL_TRIANGLES, 0, 6);   /*  app breaks down when calls this function. */

    glDisableVertexAttribArray(0);
    glDisableVertexAttribArray(1);

    if(cull_enable)
        glEnable(GL_CULL_FACE);
    else
        glDisable(GL_CULL_FACE);
}

Thanks and Best Regards,

Cherry

Hi,

May I know is there any updates about the additional info above?

Thanks and Best Regards,

Cherry

Hi Cherry,

What does app breakdown mean?

Are you getting GPU crash or Linux Kernel panic?

Do you have some logs with the errors?

Regards,
Stanley

Hi Stanley,

Thanks for your support here!
Below is app break down log.

apps_jh6.out  is the app name.
when app calls the gldrawArrays(...) func, it breaks down and log printed as below.

linux kernel is unstable when app breaks down. it means that sometimes linux kernel breaks down either and resart, and sometimes, app halt, but linux still works well.

[ 1858.599018] apps_jh6.out invoked oom-killer: gfp_mask=0x24000c4, order=0, oom_score_adj=0

[ 1858.600066] apps_jh6.out cpuset=/ mems_allowed=0

Hi Stanley,

Could you check the info above?

Thanks and Best Regards,

Cherry

Hi Cherry,

This issue doesn't have anything to do with GPU performance.

From the trace, it looks like memory allocation issue.

Was the SGX framecopy use case working before the customer added their changes?

Regards,
Stanley