PROCESSOR-SDK-J784S4: tiovx

Hello Teknik,

Due to India holiday, responses will be delayed by a day or two.

Josue

So, basically can we say that one needs to work on A72 to use dsp cores with tiovx.

Then, what about vxlib? Or directly working on c7x dsp cores?

Hi,

teknik flesetah said:

So, basically can we say that one needs to work on A72 to use dsp cores with tiovx.

Yes, correct.

teknik flesetah said:

Then, what about vxlib? Or directly working on c7x dsp cores?

vxlibs are utilized in a c7x kernel running on the C7x target. But to provide the buffers required for this libs, you would require A72 in a TIOVX framework

Regards,

Nikhil

this code is provided in the vxlib_09_02_00_04 folder. Can I use this code in my code composer project which i believe directly works on c7x target?

#include "VXLIB_bufParams.h"
#include "VXLIB_types.h"
#include "vxlib.h"
#include <cstdio>
#include <stdint.h>

#define height (3)
#define width (3)

/******************************************************************************/
/*                                                                            */
/* main                                                                       */
/*                                                                            */
/******************************************************************************/

int main(void)
{

   // clang-format off

  // Setup input and output buffers for single- and double-precision datatypes
   uint8_t in0[] = {100, 50, 0,
		            255, 25, 40,
                    65, 80, 120};

   uint8_t in1[] = {100, 75, 5,
		            55, 15, 60,
		            40, 100, 10};
   
   uint8_t out[] = {0, 0, 0,
		            0, 0, 0,
		            0, 0, 0};

   // clang-format on

   // handles and struct for call to kernel
   VXLIB_STATUS           status;
   VXLIB_absDiff_InitArgs kerInitArgs;
   int32_t                handleSize = VXLIB_absDiff_getHandleSize(&kerInitArgs);
   VXLIB_kernelHandle     handle     = malloc(handleSize);

   VXLIB_bufParams2D_t bufParamsIn0, bufParamsIn1, bufParamsOut;

   // fill in input and output buffer parameters
   bufParamsIn0.data_type = VXLIB_UINT8;
   bufParamsIn0.dim_x     = width;
   bufParamsIn0.dim_y     = height;
   bufParamsIn0.stride_y  = width * sizeof(uint8_t);

   bufParamsIn1.data_type = VXLIB_UINT8;
   bufParamsIn1.dim_x     = width;
   bufParamsIn1.dim_y     = height;
   bufParamsIn1.stride_y  = width * sizeof(uint8_t);

   bufParamsOut.data_type = VXLIB_UINT8;
   bufParamsOut.dim_x     = width;
   bufParamsOut.dim_y     = height;
   bufParamsOut.stride_y  = width * sizeof(uint8_t);

   kerInitArgs.funcStyle = VXLIB_FUNCTION_OPTIMIZED;

   status = VXLIB_SUCCESS;

   // init checkparams
   if (status == VXLIB_SUCCESS)
      status = VXLIB_absDiff_init_checkParams(handle, &bufParamsIn0, &bufParamsIn1, &bufParamsOut, &kerInitArgs);

   // init
   if (status == VXLIB_SUCCESS)
      status = VXLIB_absDiff_init(handle, &bufParamsIn0, &bufParamsIn1, &bufParamsOut, &kerInitArgs);

   // exec checkparams
   if (status == VXLIB_SUCCESS)
      VXLIB_absDiff_exec_checkParams(handle, in0, in1, out);

   // exec
   if (status == VXLIB_SUCCESS)
      status = VXLIB_absDiff_exec(handle, in0, in1, out);

   printf("UINT8 DATA");
   // print results
   for (size_t i = 0; i < height; i++) {
      printf("\n\n");
      for (size_t j = 0; j < width; j++) {
         printf("%d, ", out[i * width + j]);
      }
   }
   printf("\n\n");

   // clang-format off

  // Setup input and output buffers for single- and double-precision datatypes
   int16_t in0_16[] = {32767, 10000, 5000,
		            20000, 25000, 400,
                    6500, 800, 1200};

   int16_t in1_16[] = {-32767, -7500, 500,
		            -5500, 150, 600,
		            400, 100, -10};
   
   int16_t out_16[] = {0, 0, 0,
		            0, 0, 0,
		            0, 0, 0};

   // clang-format on

   // fill in input and output buffer parameters
   bufParamsIn0.data_type = VXLIB_INT16;
   bufParamsIn0.dim_x     = width;
   bufParamsIn0.dim_y     = height;
   bufParamsIn0.stride_y  = width * sizeof(int16_t);

   bufParamsIn1.data_type = VXLIB_INT16;
   bufParamsIn1.dim_x     = width;
   bufParamsIn1.dim_y     = height;
   bufParamsIn1.stride_y  = width * sizeof(int16_t);

   bufParamsOut.data_type = VXLIB_INT16;
   bufParamsOut.dim_x     = width;
   bufParamsOut.dim_y     = height;
   bufParamsOut.stride_y  = width * sizeof(int16_t);

   kerInitArgs.funcStyle = VXLIB_FUNCTION_OPTIMIZED;

   status = VXLIB_SUCCESS;

   // init checkparams
   if (status == VXLIB_SUCCESS)
      status = VXLIB_absDiff_init_checkParams(handle, &bufParamsIn0, &bufParamsIn1, &bufParamsOut, &kerInitArgs);

   // init
   if (status == VXLIB_SUCCESS)
      status = VXLIB_absDiff_init(handle, &bufParamsIn0, &bufParamsIn1, &bufParamsOut, &kerInitArgs);

   // exec checkparams
   if (status == VXLIB_SUCCESS)
      VXLIB_absDiff_exec_checkParams(handle, in0_16, in1_16, out_16);

   // exec
   if (status == VXLIB_SUCCESS)
      status = VXLIB_absDiff_exec(handle, in0_16, in1_16, out_16);

   printf("INT16 DATA");
   // print results
   for (size_t i = 0; i < height; i++) {
      printf("\n\n");
      for (size_t j = 0; j < width; j++) {
         printf("%d, ", out_16[i * width + j]);
      }
   }
   printf("\n\n");

   free(handle);

   return 0;
}

Nikhil Dasan said:

Please find the build instructions here 

VXLIB User Guide: Ubuntu 22.04 Build Instructions for VXLIB

What do I need this for?

Thanks,

Teknik

To run VXLIB kernels examples for C7x target on CCS, you would require the .out file of the program which can be generated using the above build steps.

I assume this is your usecase?

Regards,

Nikhil

vxlib_09_02_00_04 has dilate kernel. What about erosion? I could not see it. How possibly it is not included?

Hi Nikhil,

Thank you for your quick response.

I am using vxlib_09_02_00_04 for dilation. And, I am going to use vxlib for erosion purposes. However, there are lots of file in the sdk source. Such as, VXLIB_dilate_3x3_i8u_o8u.h, VXLIB_dilate_3x3_i8u_o8u_cn.c, VXLIB_dilate_3x3_i8u_o8u_idat.c ...
what is _cn for, or idat for? which one I should use? Or there is a c66 folder, it is a bit confusing. Can you guide me, or is there any tutorial or example code? 

Best,
Tenik