Open jeffsetter opened 5 years ago
My intended semantics for coreir.mem was to have a 0-latency read, so this is a bug that should be fixed. I am surprised errors from this has not materialized till now.
@dillonhuff, could you take a look at this bug?
The memory abstraction that I would like for simulating my circuits is a memory where the read data appears at the same time as the read address comes in. Note that this might not be feasible in a normal memory, but retiming should fix this later.
Currently,
coreir.mem
has a cycle delay between the address being calculated and the data being provided. We should create another memory that has no latency during simulation.