amaranth
amaranth copied to clipboard
amaranth-lang
Inverted DiffPairs in connectors
I have a board that has some inverted differential pairs broken out to a connector. The connector is used on several boards and there are daughterboards that can be plugged into those connectors.
Right now I am modeling this using the nmigen platform dsl in the following way: My board defines connectors. The add-on boards are functions which take the platform as an argument and then add the resources they hold to the connectors. For example:
def hdmi_plugin_connect(platform, plugin_number):
platform.add_resources([
Resource("hdmi", plungin_number,
# high speed serial lanes
Subsignal("clock", DiffPairs("lvds3_p", "lvds3_n", dir='o', conn=("plugin", plugin_number)), Attrs(IOSTANDARD="LVDS_25")),
Subsignal("data", DiffPairs("lvds2_p lvds1_p lvds0_p", "lvds2_n lvds1_n lvds0_n", dir='o', conn=("plugin", plugin_number)), Attrs(IOSTANDARD="LVDS_25"))
)
])
In my platform i do:
self.add_connectors([
Connector(
"plugin", "north",
{
"lvds0_p": 21, "lvds0_n": 23,
"lvds1_p": 3, "lvds1_n": 5,
"lvds2_p": 9, "lvds2_n": 11,
"lvds3_p": 13, "lvds3_n": 15,
},
conn=("expansion", 0),
)
])
Now one of the Lanes on the mainboard (not the daughterboard) is inverted (lvds3_p is connected to the negative side of the FPGAs differential buffer and vice versa to ease routing on the PCB). This raises the requirement, that I somehow need to mark pins as inverted diff pairs right in the platform and to mark single pairs of a DiffPairs Object as inverted.
How would you model this using the existing platform dsl or what would be the preferred way to extend the dsl?
No idea. I never considered this case and the existing DSL isn't built to support it.
@anuejn Use the vendor macro, so for Xilinx Series7 this would be IBUFGDS_DIFF_OUT, then connect the OB instead of the O output as documented in [1].
[1] https://www.xilinx.com/support/documentation/user_guides/ug471_7Series_SelectIO.pdf
The question is about the platform DSL. Of course you can always bypass the DSL, but it would be nice if we didn't have to.
I suspect the best way to handle this (hopefully very rare!) corner case is to allow passing a tuple to invert= rather than a single bool, but I haven't thought very deeply about this.
Should I formulate a separate Issue as an RFC or is this enough space for discussion on that specific topic?
Ideally an RFC would be a separate issue, though it's not that important here. (It mostly matters when there's a lot of discussion to follow, since a separate issue allows summarizing all that without forcing people to skip unnecessary context.)