04-06-2016 07:39 PM
Is it possible to pipeline entire body of the function using OpenCL pipeline attribute ? or is there any other alternative to achieve the same? (It works in HLS kernel by using #pragma HLS PIPELINE).
Sample HLS Code Snippet :
04-06-2016 09:46 PM
xcl_pipeline_workitems is the example, note that you need to create a region - the attribute apply to a region!
this is one of the templtes that we have in the tools.
kernel __attribute__ ((reqd_work_group_size(3,1,1)))
int tid = get_global_id(0);
04-06-2016 10:09 PM
04-07-2016 11:27 AM
SDAccel takes C / C ++ / OpenCL kernels so if you feel like you are more confortable with C / C++ , then use that and the VHLS pragmas - make sure to read the UG to respect the function signature needed.
The xcl pipeline workitems should have same effect as the pragma pipeline that you show.
With it, all subloops are unrolled and the datapath is pipelined so that it can process data at II=1
In VHLS if you have no loop that's the same as in OpenCL kernel with a single thread.
I hope this helps.