C6678 Jpeg Encoder DRI Marker

Can anyone help me?

thank you

 Hi, which image resolution are you using? could you share your configuration file? also could you please try latest JPEG enc JPEG_E 02_02_01_05 (http://software-dl.ti.com/dsps/dsps_public_sw/codecs/C6678/JPEG_E/latest/index_FDS.html) and let us know if the behaviour is the same?

Thank you,

Paula

hi,Paula,

   thanks for your reply. The image resolution is 3264*2448, the configuration is :

  IALG_Handle JpegEncAlgCreate (uint32_t uwImageWidth, uint32_t uwImageHeight, XDM_ChromaFormat eForceChromaFormat)
{
   IALG_Handle hJpegEnc;
   IDMJPGE_TIGEM_Params tJpegEncParams;

   memset(&tJpegEncParams, 0, sizeof(IDMJPGE_TIGEM_Params));
   tJpegEncParams.params.size              = sizeof(IDMJPGE_TIGEM_Params);
   tJpegEncParams.params.maxWidth          = uwImageWidth;       
   tJpegEncParams.params.maxHeight         = uwImageHeight;
   tJpegEncParams.params.maxScans          = XDM_DEFAULT;
   tJpegEncParams.params.dataEndianness    = XDM_BYTE;
   tJpegEncParams.params.forceChromaFormat = eForceChromaFormat;     /* XDM_YUV_420P/YUV422P*/
   
    hJpegEnc = (IALG_Handle)ALG_create((IALG_Fxns *)&DMJPGE_TIGEM_IDMJPGE,
                                 (IALG_Handle)NULL, (IALG_Params *)(&tJpegEncParams));
    return hJpegEnc;
}

And the eForceChromaFormat is XDM_YUV_420P.

The Dynamic Params is :

XDAS_Void jpegEncSetDynamicParams (IDMJPGE_TIGEM_DynamicParams *ptJpegEncDynamicParams,
              uint32_t uwInputWidth, uint32_t uwInputHeight,
              int32_t wQValue, XDM_ChromaFormat eInputChromaFormat)
{
 memset(ptJpegEncDynamicParams, 0, sizeof(IDMJPGE_TIGEM_DynamicParams));
    ptJpegEncDynamicParams->params.size              = sizeof(IDMJPGE_TIGEM_DynamicParams);
    ptJpegEncDynamicParams->params.numAU             = 0;//(uwInputWidth >> 4) * (uwInputHeight >> 4);
    ptJpegEncDynamicParams->params.generateHeader    = XDM_ENCODE_AU;

    ptJpegEncDynamicParams->params.inputWidth        = uwInputWidth;
    ptJpegEncDynamicParams->params.inputHeight       = uwInputHeight;
    ptJpegEncDynamicParams->params.inputChromaFormat = eInputChromaFormat; //XDM_YUV_420P;
    ptJpegEncDynamicParams->params.qValue            = wQValue;
    ptJpegEncDynamicParams->params.captureWidth      = uwInputWidth;

    ptJpegEncDynamicParams->captureHeight            = uwInputHeight;
    ptJpegEncDynamicParams->quantTable               = &gtStandardQuantTable;

    ptJpegEncDynamicParams->DRI_Interval = 0; //(uwInputWidth >> 4) * (uwInputHeight >> 4);
    return;
}

And the eInputChromaFormat is XDM_YUV_420P.

I will try the latest JPEG enc JPEG_E 02_02_01_05 , then i will give you a message.

Thanks again.

qbliu

hi,Paula,

   I tested the  JPEG enc JPEG_E 02_02_01_05 ,  the time is about 290ms,without DRI Marker, about 77ms!!

  The alg create params are: 

IALG_Handle JpegEncAlgCreate (uint32_t uwImageWidth, uint32_t uwImageHeight, XDM_ChromaFormat eForceChromaFormat)
{
 IALG_Handle hJpegEnc;
 IDMJPGE_TIGEM_Params tJpegEncParams;

 tJpegEncParams.params.size              = sizeof(IDMJPGE_TIGEM_Params);
 tJpegEncParams.params.maxWidth          = uwImageWidth;
 tJpegEncParams.params.maxHeight         = uwImageHeight;
 tJpegEncParams.params.maxScans          = 0;
 tJpegEncParams.params.dataEndianness    = XDM_BYTE;
 tJpegEncParams.params.forceChromaFormat = eForceChromaFormat;     /* XDM_YUV_420P/XDM_YUV_422P*/
   
 tJpegEncParams.MaxThumbnail_H_size_App0 = XDM_DEFAULT;
 tJpegEncParams.MaxThumbnail_V_size_App0 = XDM_DEFAULT;
 tJpegEncParams.MaxThumbnail_H_size_App1 = XDM_DEFAULT;
 tJpegEncParams.MaxThumbnail_V_size_App1 = XDM_DEFAULT;
 tJpegEncParams.MaxThumbnail_H_size_App13 = XDM_DEFAULT;
 tJpegEncParams.MaxThumbnail_V_size_App13 = XDM_DEFAULT;

    hJpegEnc = (IALG_Handle)ALG_create((IALG_Fxns *)&DMJPGE_TIGEM_IDMJPGE,
                                 (IALG_Handle)NULL, (IALG_Params *)(&tJpegEncParams));
    return hJpegEnc;
}

and the Dynamic Params:

XDAS_Void jpegEncSetDynamicParams (IDMJPGE_TIGEM_DynamicParams *ptJpegEncDynamicParams,
              uint32_t uwInputWidth, uint32_t uwInputHeight,
              int32_t wQValue, XDM_ChromaFormat eInputChromaFormat)
{
 //memset(ptJpegEncDynamicParams, 0, sizeof(IDMJPGE_TIGEM_DynamicParams));
    ptJpegEncDynamicParams->params.size              = sizeof(IDMJPGE_TIGEM_DynamicParams);
    ptJpegEncDynamicParams->params.numAU             = 0;//(uwInputWidth >> 4) * (uwInputHeight >> 4);
    ptJpegEncDynamicParams->params.generateHeader    = XDM_ENCODE_AU;

    ptJpegEncDynamicParams->params.inputWidth        = uwInputWidth;
    ptJpegEncDynamicParams->params.inputHeight       = uwInputHeight;
    ptJpegEncDynamicParams->params.inputChromaFormat = eInputChromaFormat; //XDM_YUV_420P;
    ptJpegEncDynamicParams->params.qValue            = wQValue;
    ptJpegEncDynamicParams->params.captureWidth      = uwInputWidth;

    ptJpegEncDynamicParams->captureHeight            = uwInputHeight;
    ptJpegEncDynamicParams->quantTable               = &gtStandardQuantTable;

    ptJpegEncDynamicParams->DRI_Interval = (uwInputWidth >> 4); //(uwInputWidth >> 4) * (uwInputHeight >> 4);

    ptJpegEncDynamicParams->captureHeight = 0;
 ptJpegEncDynamicParams->Thumbnail_Index_App0 = 0;
 ptJpegEncDynamicParams->Thumbnail_Index_App1 = 0;
 ptJpegEncDynamicParams->Thumbnail_Index_App13 = 0;
 ptJpegEncDynamicParams->Thumbnail_H_size_App1 = 0;
 ptJpegEncDynamicParams->Thumbnail_V_size_App1 = 0;
 ptJpegEncDynamicParams->Thumbnail_H_size_App0 = 0;
 ptJpegEncDynamicParams->Thumbnail_V_size_App0 = 0;
 ptJpegEncDynamicParams->Thumbnail_H_size_App13 = 0;
 ptJpegEncDynamicParams->Thumbnail_V_size_App13 = 0;
 ptJpegEncDynamicParams->APPN0_numBufs = 0;
 ptJpegEncDynamicParams->APPN1_numBufs = 0;
 ptJpegEncDynamicParams->APPN13_numBufs = 0;
 ptJpegEncDynamicParams->APPN0_startBuf = 0;
 ptJpegEncDynamicParams->APPN1_startBuf = 0;
 ptJpegEncDynamicParams->APPN13_startBuf = 0;
 ptJpegEncDynamicParams->COMMENT_insert = 0;
 ptJpegEncDynamicParams->COMMENT_startBuf = 0;

    return;
}

The encoded jpeg is right. Can you check it.

thanks, qbliu

Hi qblui,

How are you measuring time? are you using JPEG Enc unitest framework? If so, can you please share with me Testparams.cfg? (located at C:\Program Files\Texas Instruments\Codecs\C66x_jpegenc_02_02_01_05_ELF\packages\ti\sdo\codecs\jpegenc\App\Client\Test\TestVecs\Config\ ), which DRI_Interval is giving you ~300ms?

Thank you,

Paula

hi,paula,

1、How are you measuring time?

     I use the module 'ti.sysbios.family.c64p.TimestampProvider' to measure time like that:

   uwStartClock = TimestampProvider_get32();

   jpegEncode();

    uwEndClock = TimestampProvider_get32();

   System_printf("jpeg encode time = %.2fms, jpeg size = %d.\n",
          (double)(uwEndClock - uwStartClock) / 1000.0 / 1000.0, uwJpgSize);  // the cpu clock is 1GHz

   2、can you please share with me Testparams.cfg?

     I create a new project, so i do not use the Testparams.cfg. And  I will share the project.

   In this project, i use the xds560v2 to load the yuv data(for example,Input_420p.yuv at D:\ti\Codecs\C66x_jpegenc_02_02_01_05_ELF\packages\ti\sdo\codecs\jpegenc\App\Client\Test\TestVecs\Input) to 0x0c000000(MSMCSRAM), and use the function:

  JpegEncAlgCreate() to create the encode alg. and then use the function jpegEnc() to encode the yuv data at 0x0c000000 .

3、Paula,you can use the project at D:\ti\Codecs\C66x_jpegenc_02_02_01_05_ELF\packages\ti\sdo\codecs\jpegenc\App\Client\Build\TestAppEncoderand change the params->DRI_Interval in the function TestApp_SetDynamicParams() in TestAppEncoder.c to imageWidth/16 or other value. Then you can measure the jpeg encode time with params->DRI_Interval = 0 and other value to check the time with DRI Marker and without DRI Marker. I will also do it.

Look forward to your reply!

qbliu

Hi qblui, As a test I encode same image (720x480) with and without DRI. In the case in which DRI was enabled I used DRI_Interval = 20, the rest of the parameter were the same in both configurations. I used unitest framework (CCS and JTAG), please see below results:

[C66xx_0] *******************************************

[C66xx_0] Read Configuration Set 1

[C66xx_0] *******************************************

[C66xx_0] Running in Output Dump Mode

[C66xx_0] Input file... ..\..\..\Test\TestVecs\Input\Input_420p.yuv

[C66xx_0] Test File... ..\..\..\Test\TestVecs\Output\Output_420p_DRI.jpg

[C66xx_0] Creating Algorithm Instance...

[C66xx_0] Algorithm Instance Creation Done...

[C66xx_0]

[C66xx_0] Encoder output dump completed

[C66xx_0] Encoder output dump completed

[C66xx_0] Image Encoded = 1

[C66xx_0] Width and Height = 720, 480

[C66xx_0] QP = 80

[C66xx_0] Ratio = 10.121837

[C66xx_0] MegaPixel/sec 500 : 50.154207 Cycles Taken : 3445374 Cycles per pixel : 9.969253 Total_BytesRead : 518400 TotalBytes: 0 Ratio : 10.121837

[C66xx_0]

[C66xx_0] *******************************************

[C66xx_0] Read Configuration Set 2

[C66xx_0] *******************************************

[C66xx_0] Running in Output Dump Mode

[C66xx_0] Input file... ..\..\..\Test\TestVecs\Input\Input_420p.yuv

[C66xx_0] Test File... ..\..\..\Test\TestVecs\Output\Output_420p_NoDRI.jpg

[C66xx_0] Creating Algorithm Instance...

[C66xx_0] Algorithm Instance Creation Done...

[C66xx_0]

[C66xx_0] Encoder output dump completed

[C66xx_0] Encoder output dump completed

[C66xx_0] Image Encoded = 2

[C66xx_0] Width and Height = 720, 480

[C66xx_0] QP = 80

[C66xx_0] Ratio = 10.121837

[C66xx_0] MegaPixel/sec 500 : 50.152227 Cycles Taken : 3445510 Cycles per pixel : 9.969647 Total_BytesRead : 518400 TotalBytes: 0 Ratio : 10.121837

[C66xx_0]

[C66xx_0] End of execution

At least for these resolution # cycles is ~= for both cases. Could you please share with me your input file?

Thank you,

Paula

hi, Paula ,

If you use the project at c:\ti\Codecs\C66x_jpegenc_02_02_01_05_ELF\packages\ti\sdo\codecs\jpegenc\App\Client\Build\TestAppEncoder. You must change the params->DRI_Interval = 0 in the function TestApp_SetDynamicParams or it will be useless.

qbliu

hi,paula,

Could you please give your email address, i will send my project to you.

Because I do not know how to upload my project here.

qbliu 

hi,Paula,

  with my project, I test the picture 720x480, Input_420p.yuv at c:\ti\Codecs\C66x_jpegenc_02_02_01_05_ELF\packages\ti\sdo\codecs\jpegenc\App\Client\Test\TestVecs\Input,

My encode time is about 3.64ms, about 3640000 cycles, and the encode time of your test is 3445374 cycles without DRI Marker or 3445510 cycles with DRI marker 20,

It looks like the same. But in my project if I use the DRI marker = 720/16(An integer multiple of the number of the MCU in one MCU row), its endode time is about 12.86ms=12860000cycles!!

qbliu

Hi qbliu,

I was able to reproduce you times (please see below console outputs). I will check with the codec team if this is a expected behaviour. By the way thanks for reporting the bug about no being able to set DRI_Interval using Testparams.cfg. In order to being able to get DRI_Interval from .cfg I commented //params->DRI_Interval = 0; inside TestApp_SetDynamicParams(IIMGENC1_DynamicParams *dynamicParams) function. I will open and IR for this so it will be fixed by next release.

>>>> DRI / console output >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

[C66xx_0] Encoder output dump completed

[C66xx_0] Image Encoded = 1

[C66xx_0] Width and Height = 720, 480

[C66xx_0] QP = 80

[C66xx_0] Ratio = 9.891053

[C66xx_0] MegaPixel/sec 500 : 13.826290 Cycles Taken : 12497930 Cycles per pixel : 36.162992 Total_BytesRead : 518400 TotalBytes: 0 Ratio : 9.891053

[C66xx_0] DRI_interval = 40

[C66xx_0]

>>>> No DRI / console output >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

[C66xx_0] Encoder output dump completed

[C66xx_0] Image Encoded = 2

[C66xx_0] Width and Height = 720, 480

[C66xx_0] QP = 80

[C66xx_0] Ratio = 10.121837

[C66xx_0] MegaPixel/sec 500 : 50.099677 Cycles Taken : 3449124 Cycles per pixel : 9.980104 Total_BytesRead : 518400 TotalBytes: 0 Ratio : 10.121837

[C66xx_0] DRI_interval = 0

thank you,

Paula

Hi, Paula,

   Ok and thanks for your reply.

   When you find out if this is a expected behaviour or not, please give me the answer as soon as possible.

   qbliu

hi, Paula,

   Do you have any results?

  Thanks

  qbliu

Hi qbliu, an SR was opened to the codec team, they will come back with a fix schedule. I will let you know when this is done.

Thank you,

Paula

hi, Paula,

   Do you have any results? Our project team need the result to do the next step.

  Thanks very much.

  qbliu

Hi qliu,

Some of the entropy coding functions (e.g Huffman entropy coding of DC DCT coefficients, Quantization and Run lenght encoding) are optimazed for encoding 4 blocks of 8x8 at once. However, when DRI marker is enabled, RST interval is also enabled and forces the encode to process each 8x8 block independently. Since RST interval makes the encoder to process one block of 8x8 at the time, no matters which interval is selected, the total processing time of previously mentioned entropy encoding functions increases ~x4times. So, the behivour that you are facing is expected for current implementation.

Thank you,

Paula

Hi,Paula,

Thanks for your reply.

About your answer to the question, I have some problems:

1、if i can set some interval to satisfy the condition of 4 blocks of 8x8, so the encode time with or without DRI marker is about the same? I think it is feasible. If not, could the codec team release a new version of jpeg encode library to fix it?

2、When i finish encoding  the image, the field  'currentAU' of 'IIMGENC1_OutArgs' is 56200, I do not know why. The image is 3264*2448, YUV420P, I think the 'currentAU' is 3264*2448/8/8 or 3264*2448/16/16.

thank you, qbliu

Hi qbliu,

I was checking outArgs->currentAU after IIMGENC1fxns->process() (TestAppEncoder.c) and results looks ok. I got outArgs->currentAU=8100 for a 720x480 image (720x480*3/2 *1/(8x8)=8100) which is the total number of 8x8 in a 420 input. Please help me to double check in your end, if still you get a wrong AU number for a especific input and configuration (Testparams) please attach then so I can test them in my end.

thank you,

Paula

hi,paula,

  Thanks.

Question1. Ok, And thanks.

Question2. I know the answer, but i think there is a bug,  outArgs->currentAU is a XDAS_Int32 in the <JPEG_BL_Encoder_C6678_UserGuide.pdf>, but i think it is a Uint16 Datatype. 3264*2448*1.5/8/8 = 187272 = 0x2DB88, if the DataType is Uint16, the value 56200 (0xDB88) is true.

qbliu

Hi qbliu, could you please try attached Jpeg lib and .out in your end and let me know if currentAU now show you the right value? I built them in debug so it might be a little slow.. Please let me know.

1373.jpgealg_ti.rar

1067.TestAppEncoder.rar

thanks,

Paula

Qbliu, forgot to mention I modify one u16 to u32 accordingly with your observation. If new lib gives you the right AU, I will submit an IR with the codec team for an official fix in the next release.

Thanks for your help,

Paula

hi,paula,

  With your 'jpgealg_ti.rar', there is a problem.

[C66xx_0] A0=0xf8920d18 A1=0xf8920d10
[C66xx_0] A2=0x0 A3=0x82b3d8
[C66xx_0] A4=0xf8920d18 A5=0x0
[C66xx_0] A6=0xf8920d18 A7=0x2
[C66xx_0] A8=0x0 A9=0x0
[C66xx_0] A10=0x0 A11=0x0
[C66xx_0] A12=0x0 A13=0x0
[C66xx_0] A14=0x0 A15=0x0
[C66xx_0] A16=0xf89202b4 A17=0x0
[C66xx_0] A18=0xf8920284 A19=0x20
[C66xx_0] A20=0x6c A21=0x4c
[C66xx_0] A22=0x0 A23=0x1
[C66xx_0] A24=0x0 A25=0x13
[C66xx_0] A26=0x0 A27=0x0
[C66xx_0] A28=0x0 A29=0x0
[C66xx_0] A30=0x0 A31=0x0
[C66xx_0] B0=0xf89202ec B1=0x0
[C66xx_0] B2=0x1 B3=0x81847a
[C66xx_0] B4=0xf89202ec B5=0x818440
[C66xx_0] B6=0x0 B7=0xc
[C66xx_0] B8=0xf89200f0 B9=0x0
[C66xx_0] B10=0x0 B11=0x0
[C66xx_0] B12=0x0 B13=0x0
[C66xx_0] B14=0x82c774 B15=0xf89202a0
[C66xx_0] B16=0xf8920b7c B17=0x0
[C66xx_0] B18=0xa B19=0x82a04c
[C66xx_0] B20=0x69 B21=0x1
[C66xx_0] B22=0x0 B23=0x40000000
[C66xx_0] B24=0x82a050 B25=0x83100000
[C66xx_0] B26=0x24a64 B27=0x81000000
[C66xx_0] B28=0x28fef B29=0x83000000
[C66xx_0] B30=0xffffffff B31=0xffffffff
[C66xx_0] NTSR=0x1000f
[C66xx_0] ITSR=0xf
[C66xx_0] IRP=0x81eada
[C66xx_0] SSR=0x0
[C66xx_0] AMR=0x0
[C66xx_0] RILC=0x0
[C66xx_0] ILC=0x0
[C66xx_0] Exception at 0x7f99c0
[C66xx_0] EFR=0x2 NRP=0x7f99c0
[C66xx_0] Internal exception: IERR=0x1
[C66xx_0] Instruction fetch exception
[C66xx_0] ti.sysbios.family.c64p.Exception: line 248: E_exceptionMin: pc = 0x0081eada, sp = 0xf89202a0.
[C66xx_0] To see more exception detail, use ROV or set 'ti.sysbios.family.c64p.Exception.enablePrint = true;'
[C66xx_0] xdc.runtime.Error.raise: terminating execution

qbliu

Hi qbliu, could be due to different tools versions between shared lib and your application. In order to align, I used compiler version v7.4.0 also please see snapshot with used versions for bios, xdc, xdais and FC. If you are using these tool versions and still experience errors please attach the input file and I will proceed to verify correctness of currentAU in my setup.

thank you,

Paula

hi, Paula,

  I change the compiler version to v7.4.0, The used version of ccs is 5.0.1.201105110900.

But there is also a error : ti.sysbios.heaps.HeapMem: line 294: out of memory: handle=0x82b710, size=2937323008.

I give you the input file and you can check it.

qbliu

4276.yuv_3264x2448.rar

Paula,

 The input image is 3264*2448, YUV420P.

The snapshot is :

qbliu

Hi, Paula,

  After I encoded the image with DRI marker, I tested the jpeg decode algorithm. There is a problem:

 The input image 720x480 YUV420P, 'Input_420p.yuv'.

 With the DRI marker = 720 * 480 / 4 / 8 / 8 (480 / 4 rows of 8*8 block),  I can decode the encoded picture.

But with the DRI marker = 720 * 8 / 4 / 8 / 8 (8 rows of 8*8 block),  I can not decode the encoded picture. The 'imgdecOutArgs.extendedError' is 4103.

Could you please try the attached  jpeg picture with your jpeg decode demo? Please let me know.

The version of jpeg decode is C66x_jpegdec_03_01_01_05_ELF.

Thanks, qbliu

7416.jpegPictures.rar

Hi Qliu,could you please help me to open a new e2e thread for this? so we can keep issues separate it. in the meantime I will try it.

thanks,

Paula

Hi Qblui, Please see below link with latest Jpeg encoder release. This release addressed DRI cycles increase and fixed outargs->currentAU for  images >2k

http://software-dl.ti.com/dsps/dsps_public_sw/codecs/C6678/JPEG_E/latest/index_FDS.html

thank you,

Paula