note_02 concurrent executing pipe work

note_02 - concurrent executing - pipe - work

https://github.com/dsmv/sdaccel/commit/8f444f938c336b085e63d425eaad201735c369a4

I use code:

pipe ulong8 pipe_input __attribute__((xcl_reqd_pipe_depth(512)));

__kernel __attribute__ ((reqd_work_group_size(1,1,1)))

void check_read_input( __global ulong8 *src, const uint size ) {

   read_input(src, size);

}

__kernel __attribute__ ((reqd_work_group_size(1,1,1)))

void check_cnt_m2a( __global uint16 *pStatus, const ulong8 expect, const uint size ) {

   check_data( pStatus, expect, size );

}

pipe_input is FIFO that used for data transfer between kernels. We must to start both kernels at the same time. It is possible if attribute CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE is used for cl::CommandQueue;

Screen shot: https://github.com/dsmv/sdaccel/blob/develop/docs/check_transfer/screenshots/Screenshot%20from%202017-08-27%20application%20timeline.png

Attention!

Pipe may be used for communication between kernels only. Read and write pipe into one kernel is allowed in Emulation-CPU mode only.