Consume Crossbar inputs during prologue

[guosran]

Summary

Addresses #292 with a minimized Crossbar prologue-consume fix. This PR only updates CrossbarRTL.py and its focused test.

Changes

• Consume inputs that are skipped by prologue so the input stream stays aligned with the control schedule.
• Do not forward the skipped token during the prologue cycle.
• Add a concrete example as an inline test comment.

No new ports or interfaces are introduced.

[guosran]

This PR is the isolated Crossbar part from the earlier prologue discussion. Concrete example: if ctrl addr 0 has prologue_count=1, the first input token belongs to the skipped prologue cycle. The Crossbar should consume that token to keep the stream aligned, but should not forward it. The next non-prologue token is then the first value observed by the sink.

[guosran]

Concrete example:
During prologue, a skipped input token should be consumed but not forwarded.

input token A arrives
prologue_count says this input is still in prologue

Before:
crossbar could either block on downstream rdy or incorrectly forward stale data.

After:
recv side can consume the prologue token,
send side does not require/forward that prologued output.

[yyan7223]

Concrete example: During prologue, a skipped input token should be consumed but not forwarded.

input token A arrives prologue_count says this input is still in prologue

Before: crossbar could either block on downstream rdy or incorrectly forward stale data.

After: recv side can consume the prologue token, send side does not require/forward that prologued output.

During prologue, a skipped input token should be consumed but not forwarded.
Sorry, why is that?
We have prologue for both fu (consumed you mentioned) and fu_crossbar/routing_crossbar (forwarded you mentioned).

[yyan7223]

Sorry I cannot understand the changes.

[yyan7223]

Why crossbar is allowed to perform routing during prologue?

[tancheng]

Concrete example: During prologue, a skipped input token should be consumed but not forwarded.
input token A arrives prologue_count says this input is still in prologue
Before: crossbar could either block on downstream rdy or incorrectly forward stale data.
After: recv side can consume the prologue token, send side does not require/forward that prologued output.

During prologue, a skipped input token should be consumed but not forwarded. Sorry, why is that? We have prologue for both fu (consumed you mentioned) and fu_crossbar/routing_crossbar (forwarded you mentioned).

+1, I thought prologue shouldn't consume input token. I thought prologue (iirc about my own design/motivation) is used to proceed even without caring about input, i.e.:

• Even input is not there, it can proceed
• Even input is there, it can proceed without consuming the input, as the input should be consumed in the future (after prologue cycles)