Friday, 2022-08-26

*** ghostmansd <ghostmansd!> has quit IRC01:15
lkclah, it was there as a general comment for that section02:33
lkcli'll sort it02:33
*** ghostmansd <ghostmansd!> has joined #libre-soc08:57
*** markos <markos!~Konstanti@> has quit IRC09:15
*** markos <markos!~Konstanti@> has joined #libre-soc09:28
*** ghostmansd <ghostmansd!> has quit IRC09:49
*** ghostmansd <ghostmansd!> has joined #libre-soc10:17
*** octavius <octavius!> has joined #libre-soc11:31
octaviuslkcl, I did wonder why budget-sync didn't work. Sorry about that11:34
lkcloctavius, you can see other bugreports for examples. all sorted12:10
lkclbtw can you fix URLs here
lkcl404 not found on wiki12:10
octaviuswill do12:13
lkclcomment 0 is a main priority12:15
lkclit gets submitted to FundingBox12:15
octaviusupdated the links12:19
lkcli'll be putting in milestone 3 in the next 2-ish-or-so weeks12:21
lkclbut need everything in order first12:21
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC12:26
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc12:27
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC12:42
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc12:43
*** octavius <octavius!> has quit IRC12:52
*** lx0 <lx0!~lxo@gateway/tor-sasl/lxo> has quit IRC13:41
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC14:33
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc14:34
lkclprogrammerjake, on "setvl.", check - changing the behaviour of "svstep." is... dodgy14:41
lkcland the "vertical-first" stepping, it's making me nervous14:44
lkcl  33         if Rc = 1 then14:44
lkcl  34             if step = 0 then c <- 0b00114:44
lkclfurther down? no problem14:44
lkcl  61         if Rc = 1 then14:45
lkcl  62             if VL = 0 then c <- 0b00114:45
lkcl  63             else c <- 0b01014:45
lkclnope, thought about it: reverting14:52
lkclhandle_comparison should be the place - if at all - where Rc=1 is handled.14:53
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC15:15
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc15:16
lkclok i found the source of the problem - GPR(_RT) <- xxxx instead of RT <- xxxx15:37
lkclwhat i _did_ realise is, of course, an overflow flag is needed (to set CR0.SO)15:39
lkcli'll have to update the spec for that15:39
lkclsimplev.mdwn has to go back to what it was, and, sigh, ISACaller.handle_comparison has to be handed a copy of SVSTATE.vl16:02
lkcl*NOT* RT16:03
lkclthis to cope with RT=0 - CR0 *still* has to be update when Rc=1 and RT=016:03
lkclfrickin 'ell :)16:03
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC16:31
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc16:31
openpowerbot[mattermost] <lkcl> - call for papers on openpower 2022 summit - 10th november, 2 weeks until CFP deadline17:02
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC17:22
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc17:22
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc17:24
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC17:35
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc17:35
programmerjakelkcl: your latest openpower-isa commit is broken:17:41
programmerjake>       return (SVSTATE, CTR, overflow,)17:41
programmerjakeE       UnboundLocalError: local variable 'overflow' referenced before assignment17:41
programmerjakesrc/openpower/decoder/isa/ UnboundLocalError17:42
programmerjakeimho this is why you should subscribe to build failure emails, you can just delete them after you get them, but you need some kind of active notification, since you don't seem to check CI that often...17:44
programmerjakelemme know if you want that and I'll send you a link to change your subscription17:45
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC17:45
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc17:45
programmerjakei went to a lot of effort to get openpower-isa to pass CI specifically so it can catch stuff like this17:45
*** octavius <octavius!> has joined #libre-soc18:00
lkclprogrammerjake, i'm good19:31
lkclv. tired.19:32
lkcladded two more setvl. (Rc=1) unit tests, these are for CTR-mode19:39
programmerjakealso forgot declaring SO, OV and OV32 to be modified, assuming that's what overflow is for19:40
lkclprogrammerjake, i'll pm you the secret URL for formal-proofs RFPs19:41
programmerjakeimho setvl should *not* set SO, since that'll just make any following Rc=1 instruction block on the setvl unnecessarily19:42
lkclfortunately in SFFS compliancy level XER.SO can be ignored completely.19:44
programmerjakewell, not all cpus are only sffs, we need to not unnecessarily add stuff that makes it slower...knowing if setvl chose a value smaller than the input can easily be done with cmpldi for the very rare cases when that's necessary. the rest of the time not blocking on SO is imho more important19:48
programmerjakenote that imho that's *not* setvl overflowing, because setvl is defined to take the min(input, MVL) and the output of the min won't ever overflow19:49
lkclit's overflowing the limit of MAXVL.19:50
lkcli can see that being useful to know.19:50
lkclPower ISA scalar operations set XER.SO.19:51
lkclan exception to that requires justification and it will be the sole exclusive scalar instruction which does not set XER.SO19:51
programmerjakepower scalar operations only set SO if you use the *o instruction option, which they state is often quite slow.19:51
lkclIBM defined the Linux/AIX Levels as requiring XER.SO to be set at all times from CR0.SO19:51
lkclyes, because of hazard dependencies19:52
lkcland nobody uses them19:52
programmerjakethere are plenty of instructions that do not set SO, e.g. modulus19:52
* lkcl thinks...19:52
lkcli need rest, urgently. i can see the case for not setting XER.SO. leave it with me.19:52
* lkcl afk19:53
programmerjakelkcl: openpower-isa ci passed, thx for fixing it!20:07
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC22:23
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc22:30
*** octavius <octavius!> has quit IRC22:38

Generated by 2.17.1 by Marius Gedminas - find it at!