| --- Log opened Thu Aug 27 00:00:17 2015 | ||
| stekern | andrzejr: the cache in the wb_sdram ctrl just caches bursts | 03:09 |
|---|---|---|
| stekern | but, I agree, if you are coupling it more closely with mor1kx cache, you will not need that | 03:11 |
| olofk | andrzejr: Hmm... how do you expect a burst termination signal to be used? | 05:16 |
| olofk | For example, will the receiver only latch a new address on the data after a burst termination? | 05:17 |
| olofk | And have you considered burst counters instead (like sending 8 on a separate line to indicate that there will now be 8 words in this burst | 05:18 |
| andrzejr | olofk, yes, counters would be even better. I've been thinking about it and I agree we could probably do without it but it would be useful for the slave to know that the write burst was finished so it can flush its buffers. | 08:20 |
| andrzejr | another extension is sending and returning an identifier, which would allow reordering of transactions. Both CAMD and UI assume that the order of reads matches the order of requests. But that may be an overkill. | 08:21 |
| olofk | I would say that both burst size/end indications and reordering are nice to have, but not essential | 09:09 |
| andrzejr_ | olofk, IMHO we should have at least burst termination for writes. word count is nice but sometimes we may not have this information available. reordering is too much in most cases, even sdram Ctrl returns data in order. | 17:29 |
| andrzejr_ | how do you request a read burst? by sending multiple commands like UI bus? | 17:30 |
| olofk | I made it simple. There is no specific burst mode. Both reads and writes are supposed to require one command/address per word | 21:29 |
| --- Log closed Fri Aug 28 00:00:19 2015 | ||
Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!