| *** kylel1 is now known as kylel | 09:46 | |
| * cesar is getting familiar with the Pipeline API by playing on a sandbox. | 10:06 | |
| lkcl | cesar, i have an idea: to convert core.py to a pipeline stage | 10:50 |
|---|---|---|
| lkcl | this would have the side-effect of allowing you to see how it works | 10:50 |
| cesar | Sure. But there is plenty of examples already in nmutil. No need to touch core.py just to show me another example... I agree though that adding handshake ports to core.py is a good idea. | 11:00 |
| lkcl | the process of conversion begins to define the in and out data structures | 11:01 |
| lkcl | which is a necessary step | 11:07 |
| lkcl | done, but, hmm, just discovered that test_issuer_svp64.py is not terminating | 11:13 |
| lkcl | ah. that's going back to before oct 28 commit 7beae237 | 11:15 |
| lkcl | ahh part of the test runner api change | 11:20 |
| lkcl | commit f25faf58a3 | 11:23 |
| lkcl | the change-point from running side-by-side to running completely through simulator (and waiting for KeyError) then completely through HDL | 11:24 |
| lkcl | TestIssuer doesn't terminate in SVP64 mode | 11:26 |
| lkcl | that'll be a long-standing bug, previously undiscovered | 11:36 |
| lkcl | https://bugs.libre-soc.org/show_bug.cgi?id=743 | 11:36 |
| lkcl | ok that's mostly converted | 14:46 |
| lkcl | ready/valid signalling on the input. | 14:46 |
| lkcl | core.issue_i is removed, using core.p.i_valid instead | 14:46 |
| lkcl | core.busy_o is removed, using ~core.p.o_ready instead | 14:47 |
| lkcl | exc_happened (previously in TestIssuer) has moved to core.n.o_data | 14:47 |
| * lkcl re-running complete test issuer all unit tests | 14:49 | |
| lkcl | passes no problem | 16:37 |
| *** tplaten <tplaten!~isengaara@55d4c982.access.ecotel.net> has left #libre-soc | 20:04 | |
| lkcl | cesar, i just altered how rdmaskn needs to be set in MultiCompUnit | 23:49 |
| lkcl | it goes into the src latch _input_, now, rather than masks the _output_ of the src latch | 23:49 |
| lkcl | but, that meant that the src data would be left held on anything excluded from rdmaskn | 23:50 |
| lkcl | which meant that on the next ALU request, previous data went into the ALU. whoops :) | 23:50 |
| lkcl | to fix that i added a reset / zero-out mechanism: when the ALU says "not busy", all the src data latches are cleared. | 23:51 |
| lkcl | fixes it | 23:51 |
| lkcl | now, rdmaskn does not have to be held HI | 23:51 |
| lkcl | it can be asserted only one clock: when issue_i is set | 23:51 |
| lkcl | test_compunit_multi.py still works | 23:52 |
Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!