Please document the difference between the two 'fallthrough' options

[zapta]

They are not mention in the specification text. Please cover both the difference in logic and in timing (the fallthrough option seems to have a longer propogation delay of the data out from the clock edge which is problematic in our case where we use it to drive an external FTDI fifo).

[dpretet]

Hello,

fallthrough means the read is asynchronous, data written is "immediately" available on read side. This behaviour is originating from the original code from Clifford Cummings and the option to add an extra FFD stage on read has been added later by a pull request (don't remember which one).

Thus, this implementation chosen by the designer is a big constraint and you can enable the FFD stage on read if you want. But keep in mind the two sides are asynchronous and a false path should be added. The presence of FFD on read would avoid possible CDC glitches when read and write pointers are close together, when the FIFO is read and write at maximum bandwidth, making possible glitches when extracting the data from write clock domain to read clock domain.

Best,
Damien