Open snowleopard opened 8 years ago
I have translated all the .work
files present in https://github.com/tuura/graph-families/tree/master/real/processors, into a format compatible with scenco
. That is:
.scenario name_scenario
Vertex_A Vertex_B
:Boolean_condition Vertex_A
.end
All of them are present inside the test folder, where I am going to insert a file named 90nm.genlib
, useful for testing technology mapping related part.
@allegroCoder Excellent! That's a good first step.
As far as I understand we still can't reproduce all the reported results, because we are missing cross-platform ABC-based synthesis and technology mapping algorithms?
By the way, I think the current text format will eventually become too limiting for us. At some point we will probably switch to the Farfalle language for describing processor instructions.
Last comment: Sure, Farfalle language would be nice for the processor instruction description. What about partial order description? I think that the format listed above can be fine for the partial orders.
In my view, such a file might still be used, even though it can be hidden by the whole framework. It can be used for passing the partial orders to scenco
via a temporary file.
And at that point we will use Farfalle language for the description of the tests we run, as you mentioned! :+1:
@allegroCoder Good idea about an intermediate format for representing partial orders, and in general graphs. I think we can use an algebraic form, e.g. p = a + b -> (c + d)
, which is used in @jrbeaumont's plugin for Workcraft. I will implement a parser for it as part of https://github.com/tuura/scenco/issues/4.
At the moment processor benchmarks are hard to reproduce: they require Workcraft and there is a high chance that the reported results are outdated due to on-going development of the SCENCO plugin: https://github.com/tuura/graph-families/tree/master/real/processors.
As soon as https://github.com/tuura/scenco is released we should include precise instructions for reproducing the results directly with
scenco
(without relying on Workcraft) and make sure the results are up-to-date. We can keep old results for comparison, this will also allow us to see how our encoding tools improve over time.