There is new feature in FPGA OpenCL v19.1 that one could force memory blocks to be implemented in MLAB or M20K as described in the guide:
But it only works on variables that are explicitly defined in the code. Some time the compiler will generate lots of memory in the clusters like follows:
When I use multiple kernels, it became very costly in M20K.
So is there any way that one could force such memory to be implemented in MLAB rather than M20K ?