• State
  • Locked Locked
  • Replies 2 replies
  • Subscribers 100 subscribers
  • Views 262 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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.

Why codec engine access ARM memory so slow?

feng xiao ming
Expert 1445 points

HW:DM8168

SW:ezsdk5.5.1.4+codec engine

I only run the following alg:

for(i=0;i<270;i++)
{
     for(j=0;j<480;j++)
    {
          dstbuffer[itmp+j]=0;

     }

}

 

dstbuffer is allocated from arm.

this will take 3millisecond,but when the alg run at the ARM,it only take 0.5millisecond.

Why the performance os dsp is slower  than ARM?

  • 0 in reply to Margarita Gashova
    Expert 1445 points

    Dear Margarita:

         I change the codec of video_copy examples.

    int MY_algorithm(char*orgbuffer,char*dstbuffer,int iwidth,int iheight)
    {
    unsigned int i=0;
    unsigned int j=0;

    for(i=0;i<270;i++)
    {
         for(j=0;j<480;j++)
        {
              dstbuffer[itmp+j]=0;

         }

    }


    }

     

    XDAS_Int32 VIDEOCOPY_TI_process(IVIDDEC_Handle h, XDM_BufDesc *inBufs,
    XDM_BufDesc *outBufs, IVIDDEC_InArgs *inArgs, IVIDDEC_OutArgs *outArgs)
    {
    XDAS_Int32 curBuf;
    XDAS_Int32 minSamples;

    Log_print5(Diags_ENTRY, "[+E] COLORCONVERT_TI_process(0x%lx, 0x%lx, 0x%lx, "
    "0x%lx, 0x%lx)",
    (IArg)h, (IArg)inBufs, (IArg)outBufs, (IArg)inArgs, (IArg)outArgs);

    /* validate arguments - this codec only supports "base" xDM. */
    if ((inArgs->size != sizeof(*inArgs)) ||
    (outArgs->size != sizeof(*outArgs))) {

    Log_print2(Diags_ENTRY, "[+E] COLORCONVERT_TI_process, unsupported size "
    "(0x%lx, 0x%lx)", (IArg)(inArgs->size), (IArg)(outArgs->size));

    return (IVIDDEC_EFAIL);
    }

    /* outArgs->bytesConsumed reports the total number of bytes consumed */
    outArgs->bytesConsumed = 0;

    /*
    * A couple constraints for this simple "copy" codec:
    * - Given a different number of input and output buffers, only
    * decode (i.e., copy) the lesser number of buffers.
    * - Given a different size of an input and output buffers, only
    * decode (i.e., copy) the lesser of the sizes.
    */

    for (curBuf = 0; (curBuf < inBufs->numBufs) &&
    (curBuf < outBufs->numBufs); curBuf++) {

    /* there's an available in and out buffer, how many samples? */
    minSamples = inBufs->bufSizes[curBuf] < outBufs->bufSizes[curBuf] ?
    inBufs->bufSizes[curBuf] : outBufs->bufSizes[curBuf];

    /* process the data: read input, produce output */
    //memcpy(outBufs->bufs[curBuf], inBufs->bufs[curBuf], minSamples);
    MY_algorithm(inBufs->bufs[curBuf],outBufs->bufs[curBuf],1920,1080);

    Log_print1(Diags_USER2, "[+2] COLORCONVERT_TI_process> "
    "Processed %d bytes.", (IArg)(minSamples ));
    outArgs->bytesConsumed += minSamples;
    }

    /* Fill out the rest of the outArgs struct */
    outArgs->extendedError = 0;
    outArgs->decodedFrameType = 0; /* TODO */
    outArgs->outputID = inArgs->inputID;
    outArgs->displayBufs.numBufs = 0; /* important: indicate no displayBufs */

    return (IVIDDEC_EOK);
    }

    And I run the algorithm video_copy.

    then I write an app use the MY_algorithm function at ARM side,not run  at DSP side.

    I found that the DSP run so slow than ARM.

    How to accelerate the memory access time ata DSP side?