Marek Materzok

The cache could remember which part of the cache is flushed, and accept requests for the already flushed part.

If an unsupported instruction is decoded, we will soon get a trap, and the following instructions will never be executed, so why fetch them?

The error message doesn't show where the error occured, it should at least show which transactions/methods were involved. (Maybe catch the error and extend the message?)

Idea: instead of finding connected components of the relation graph, and then topologically sorting the components according to the scheduling order, do the following: - Find *strongly* connected components of...

Assembler-based tests have external dependencies and run slowly, an option to omit them would be useful.

Look at: from future import annotatons (PEP 563)

2

Could make sense in the implementation of transactions: https://peps.python.org/pep-0563/

In #187 there was a silent introduction of a class which simplifies writing typical tests, `SimpleTestCircuit`. It automatically creates `TestbenchIO`s for exposed methods, which allows to avoid creating a new...

There is some Java coding to be done probably.

More uses of av_comb, slight refactor of RF

1

Probably won't change much, but here we go.

To implement superscalarity on FPGAs, we need multi-ported FPGA memories. Such memories internally use two-ported FPGA memories and some glue logic to make them behave as a single multi-ported memory....