08-05-2019 08:08 PM
08-06-2019 02:34 AM
Multiple kernels can access to same buffer however single buffer size can not exceed 4GB.
In your case, the large dataset has to be split into several buffers.
08-06-2019 05:17 AM
Thanks for your reply! I can split the index into multiple buffers easily but I still wonder:
Is there any example of multiple kernels sharing same buffers?
Besides, in my case, I want the shared buffers stay in the global memory while other input buffers change in each iteration of my algorithm. Can this be done?
08-07-2019 11:08 PM
Here is an example of multiple kernels sharing one buffer: https://github.com/Xilinx/SDAccel_Examples/tree/master/getting_started/rtl_kernel/rtl_vadd_2kernels
Regarding your second question, the answer is yes. Say that you have buffer A and B on device ddr. At the begining, host writes data to both A and B. Then A may keep unchanged and host can re-write B as your wish.