Hi Masih,
So you'll need to make three changes, one to the source file and one to the header file, and to the place where you call the function.
So if you change your source code to:
void sds_sobel(unsigned short *img_in, unsigned short *img_out, int rows, int cols, int stride, int &rowbarrier) //, unsigned short Dout[M*M])
{
int row;
int col;
int row_edge_cnt=0;
rowbarrier=0;
int i=0;
and your header to this:
#pragma SDS data mem_attribute(img_in:PHYSICAL_CONTIGUOUS, img_out:PHYSICAL_CONTIGUOUS)
#pragma SDS data access_pattern(img_in:SEQUENTIAL, img_out:SEQUENTIAL)
#pragma SDS data data_mover(img_in:AXIDMA_SG, img_out:AXIDMA_SG)
#pragma SDS data copy(img_in[0:stride*rows], img_out[0:stride*rows])
void sds_sobel(unsigned short *img_in, unsigned short *img_out, int rows, int cols, int stride, int &rowbarrier);
and your caller to this:
void sds_filter(unsigned char *data_in, unsigned char *data_out, int height, int width, int stride, filter_type type)
{
int rowbarrier;
switch (type) {
case FILTER_TYPE_SOBEL:
sds_sobel((unsigned short *) data_in, (unsigned short *) data_out, height, width, stride, rowbarrier);
break;
default:
printf("sds_filter :: Invalid Filter Type \n");
}
//now here when your function sds_sobel returns, the rowbarrier will be in the PS memory
}
SDSoC will return your single integer from PL to PS.
Sam
