Start Tcl design inspection methods #4678
Conversation
There was a problem hiding this comment.
Overall this lgtm, I'd suggest making a decision on flag / option parsing before merging a first version, and also fixing the unsigned MSB set int handling, both to avoid backwards incompatible changes. The other points I commented on can be addressed in a follow up if the goal is to get something usable merged quickly.
kernel/tclapi.cc
Outdated
| (!mod_flag && i != argc - 3) || | ||
| (string_flag + int_flag + bool_flag > 1)) | ||
| ERROR("bad usage: expected \"get_attr -mod [-string|-int|-bool] <module> <attrname>\"" | ||
| " or \"get_attr [-string|-int|-bool] <module> <identifier> <attrname>\"") |
There was a problem hiding this comment.
It might be nice to support an optional -default <default_value> (which will return the default value unchanged, not reinterpret it as the selected type).
One potential issue with adding more flags (whether it's this -default or something else) is that before doing so get_attr -mod -default some_attr would look up some_attr in a module -default but after adding that option it would be missing arguments. Maybe we should require using {\-default} (which already works due to escape_id) and/or -- and interpret all - strings as flags/options (producing an error if unknown) until -- is found. This is also an issue just for the existing flags, i.e. you have to use [rtlil::get_attr -mod -string {\-mod} some_attr] and can't use [rtlil::get_attr -mod -string -mod some_attr] to access module \-mod (... . I'm not sure what's the convention for tcl commands wrt to disambiguating options and strings starting with -.
kernel/tclapi.cc
Outdated
| Tcl_CreateCommand(interp, "rtlil::has_attr", tcl_has_attr, NULL, NULL); | ||
| Tcl_CreateCommand(interp, "rtlil::set_attr", tcl_set_attr, NULL, NULL); | ||
| Tcl_CreateCommand(interp, "rtlil::get_param", tcl_get_param, NULL, NULL); | ||
| Tcl_CreateCommand(interp, "rtlil::set_param", tcl_set_param, NULL, NULL); |
There was a problem hiding this comment.
I'd also expect rtlil::unset_attr as well as rtlil::attr_list to get a complete set of operations (same for param which is also missing rtlil::has_param).
Maybe also useful would be something like rtlil::attr_array which would produce all attributes as a key-value array such that you can do array set object_attrs [rtlil::attr_array some_mod some_object]; puts "the value of keep is $object_attrs(\\keep)" and use other tcl array functions. I don't actually use tcl enough to know how useful this would be, though.
|
Pulled fixes from @jix. Thanks! |
What are the reasons/motivation for this change?
Allow inspection and altering of the design from Tcl scripts for use-cases which are too specific to write a Yosys command for.
This is in addition to the existing ways of obtaining information about a design by
and variations thereof.
See sample usage here: https://github.com/povik/OpenROAD-flow-scripts/blob/72b084f68275aaefec903fbc8e4f8613679e52e0/flow/scripts/synth_preamble.tcl#L122
Another use case of the new API is replacing this Python script called from Tcl to process a JSON export with a direct Tcl implementation: https://github.com/The-OpenROAD-Project/OpenROAD-flow-scripts/blob/master/flow/scripts/mem_dump.py (Cc @oharboe, feel free to comment)
Explain how this is achieved.
We introduce a set of Tcl functions prefixed with
rtlil::to query attributes and parameters.Tcl bindings in other projects sometimes use text-formatted pointers as the Tcl-side handles, but for a RTLIL binding we can consistently use IDs of objects to refer to them from the Tcl side.