Why does one single load/store consume much RAMs?

--- Quote Start ---

Other than disabling the cache using volatile, there is nothing else that can be done. Is there a specific reason you want to do this? Removing those FIFOs will have a very large negative impact on performance.

--- Quote End ---

I also saw that using volatile can disable the cache.

I want to apply coarse grained parallel on external memory access, e.g.

# pragma unroll

for (i = 0; i < 64; ++i)

for (j = 0; j < 1000; ++j)

... = a[i * 1000 + j];

It will consume a lot of RAMs (64 * 16 RAMs) while total of RAMs is about 2700 RAMs in arria10.

After that, we have little optimization space because of lack of RAMs.

Thanks.