- Newest
- Most votes
- Most comments
By specifying --sp multiexp_kernel_1.point_p:DDR[0:1], you are connecting this kernel port (or argument) to both DDRs. This means that the addressable space, as seen by the kernel, will indeed by 32GB.
However, the XRT buffers declared in your host application must fit into a single DDR bank. Each buffer can therefore be 16GB at the most.
By connecting your kernel to DDR[0:1], you are allowing XRT to chose whether the buffer will be allocated in DDR[0] or DDR[1], but it cannot be allocated across both banks.
If your kernel needs to access 32GB worth of data, then you need to create at least 2 different buffers in your host code. Your kernel should then have as many arguments as you will declare buffers in the host code. Each of these kernel arguments can then be mapped to the desired DDR bank.
Ok makes sense, thanks.
(Seems you had the same formatting issues as me :) )
Relevant content
- asked a year ago
- asked 4 years ago
- asked a year ago
- asked 2 years ago
- AWS OFFICIALUpdated 5 months ago
- AWS OFFICIALUpdated 8 months ago
- AWS OFFICIALUpdated 4 months ago
- AWS OFFICIALUpdated 4 months ago