Wednesday, 2021-08-11

ghostmansd-pcfwiw, upon each push I see the following:08:39
ghostmansd-pcremote: warning: unrecognized negative pattern: '/*'08:39
ghostmansd-pcremote: warning: disabling cone pattern matching08:40
ghostmansd-pcdoes it come from some .gitignore file?08:40
jni'd suspect is comes from a gitolite configuration entry08:41
ghostmansd-pcok, this doesn't hurt, just FYI09:01
ghostmansd-pcthank you for ideas on how to improve the test, I'll take a look at it ASAP09:02
ghostmansd-pcOnce the performance is OK, I'll think about addg6s test. You mentioned I could compare it with qemu, right?09:03
lkcl-ghostmansd, yes ignore that. it's a post-update hook that rebuilds the wiki with a copy of the markdown files that are *NOT* part of the ikiwiki git repository09:56
lkcl-yes, addg6s could be run against qemu09:57
lkcl-to see what the heck it does09:57
lkcl-ghostmansd, i committed a first version (hacked together) of ISATestRunner, because extracting the unnecessary HDL-related code is quite complex10:23
lkcl-now you can see clearly there's virtually no difference between ISATestRunner.run_all() and run_tst()10:24
ghostmansdNice, thank you! I'd like to be able to always check the whole tables.10:25
lkcl-in _theory_ it's possible to also re-use an ISA() instance (re-initialise it with a new program) however we haven't ever tried it.10:25
ghostmansdSo I'll follow the approach you suggested when I have some time to do it.10:25
*** lkcl- is now known as lkcl10:38
lkcljn: it's a post-update hook for ikiwiki underlays10:40
jni see10:44
ghostmansd-pcHi Luke, I'm trying to make use of the suggested test approach, but it looks like I misunderstand how to use it.19:46
ghostmansd-pcI have a `class BCDTestCase(TestAccumulatorBase)`, with some nice `case_*` methods, which all add a Program and initial registers via self.add_case.19:48
ghostmansd-pcFor each and every entry from tables (BCD_TO_DPD and DPD_TO_BCD), we have self.add_case() invocation.19:49
ghostmansd-pcI also have `class BCDTestRunner(ISATestRunner)`, which currently inherits from your ISATestRunner (I had to change some bits there, though).19:50
ghostmansd-pcBCDTestRunner does nothing, it's literally `pass` right after inheritance.19:51
ghostmansd-pcThen, in my __main__, I do the following:19:51
ghostmansd-pc    unittest.main(exit=False)19:51
ghostmansd-pc    runner = BCDTestRunner(BCDTestCase().test_data)19:51
ghostmansd-pc    runner.run_all()19:51
ghostmansd-pcFirst, I don't understand how to do assertEqual with this approach.19:53
ghostmansd-pcSecond, if I attempt to run over all 2000 entries (1000 for each table), I'm smashed with `OSError: [Errno 24] Too many open files`.19:55
ghostmansd-pcI've pushed these changes into bug-657 branch, could you, please, take a look, when you have time? I likely miss something and use it incorrectly.19:56
lkclarg branches, blech :)20:25
lkcldon't delete run_tst, it's used by other test_caller_* code20:26
lkclah, you have a _copy_ of run_tst - it should have been imported from test_caller.py20:30
lkclurrr... DRAT20:34
lkclyep, those HDL test cases are post-analysis (post-comparison)20:35
lkclwhere the results of the HDL simulation are run against the results of the ISACaller simulation, one instruction at a time.20:35
lkclhave to do this differently20:35
lkclwe would have to define a new system of "expected results"20:44
lkclwhich would be added through each case_xxxx function20:44
lkclbut for now what i've done is set up one single PowerDecoder/2 that can be re-used20:45
lkclwhat i suggest is, add 32 instructions to each listing, as overwrites (ctdbcd 0,0 then ctdbcd 1,1 etc. etc.)20:46
lkcland batch them20:46
lkclthere's 109 people in the channel, bear in mind though it's primarily for people with IBM POWER9 Talos-II Workstations, by Raptor Engineering20:53
lkcldoh, wrong channel.20:57
ghostmansdlkcl: I think system of expected results is possible; after all, each TestCase can have instance variables (and it already has, e.g. GP regs, FP regs and so on).21:17
ghostmansdlkcl: So if I got that correctly you suggest to follow the approach I suggested in bug-657, right?21:18
ghostmansdAnd drop per-case system for now21:19
ghostmansdOr, rather, postpone commiting this to master and focus on making things work with the current approach21:21
lkclcesar, programmerjake, lxo: meeting 5mins21:54

Generated by 2.17.1 by Marius Gedminas - find it at!