Saturday, 2022-12-10

*** gnucode <gnucode!~gnucode@user/jab> has joined #libre-soc01:16
*** gnucode <gnucode!~gnucode@user/jab> has quit IRC02:45
*** lxo <lxo!> has quit IRC05:33
*** lxo <lxo!> has joined #libre-soc05:33
*** lxo <lxo!> has quit IRC05:40
*** lxo <lxo!> has joined #libre-soc05:40
lkclno problem. just remember always always always run budget-sync after changing/adding budgets/milestones/TOML fields, the report ends up looking like this once converted from mdwn to html
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC08:47
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc08:47
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC08:50
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc08:50
*** lxo <lxo!> has quit IRC10:19
*** ghostmansd <ghostmansd!~ghostmans@> has joined #libre-soc10:20
lkclghostmansd, i pinged the mou team for adding you to 2021-02A-051.10:39
ghostmansdThanks Luke!10:40
ghostmansdI've updated the milestones, now budget-sync seems to be happy.10:40
ghostmansdIs pi specifier stable enough to be picked and ported to insndb?10:43
ghostmansdOr it's WiP and this should be postponed?10:43
lkclpi specifier?10:59
lkcloh. pi specifier. 1 sec11:00
lkclit's part of the unit tests, so yes.11:01
ghostmansdOK, will add it soon11:01
lkcli just realised that ffirst mode needs to go11:02
lkclsorry it's been several weeks since i looked at this - i normally look at these pages 5-10 times a day11:03
ghostmansdlol, exactly after when we supported it in 4 places: binutils, disasm, asm and new asm :-)11:03
lkclyyep, i know.11:03
lkclit was a face-palm moment when i realised a couple months back that ffirst mode in LD/ST is not actually the original LD/ST "fault-first" of ARM SVE.11:04
ghostmansdOK, if you could update the docs, I could switch all these places to new mode11:04
ghostmansdOr maybe we can defer it11:05
lkcli mean, it _could_ be left in - it's kinda "load this data, test it against the bit, and truncate VL based on that test"11:05
lkcli mean, is that actually useful?11:05
lkcldata-dependent fail-first is supposed to be for applying to e.g. cmpi11:06
ghostmansdI don't see that many scenarios which could benefit, to be honest...11:06
lkcl"compare this data, if one of the CRfield bits is 1, truncate VL"11:06
ghostmansdbut I'm by no means an expert in this area11:06
lkclit's like combining a LD/ST with a cmpi-with-fail-first-mode into one instruction11:06
lkclit's a bit much11:07
ghostmansddropping ffirst would leave us two free modes, right?11:07
ghostmansdso these could be reused for something else11:07
lkclthe entire "01" block becomes free.. yes.11:07
ghostmansdwell, practically one...11:07
ghostmansdbecause it depends on Rc IIRC11:07
lkclit's CISC not RISC11:08
ghostmansdOK, I suggest I'll add /pi and concentrate on testing the new assembly; changing the modes will kinda now intervene the latter activity11:12
ghostmansdOnce this is done, we can return to this topic and raise the relevant tasks: docs and code for Python and code for binutils11:12
ghostmansdOh, nice, I see you already integrated this into disasm11:17
ghostmansd(I mean /pi specifier again, sorry)11:17
ghostmansdHm. From what I see LF is specifier too. But there's no code for it in pysvp64asm.11:22
ghostmansd    def specifiers(self, record):11:23
ghostmansd        if self.pi:11:23
ghostmansd            yield "pi"11:23
ghostmansd        if self.lf:11:23
ghostmansd            yield "lf"11:23
*** ghostmansd <ghostmansd!~ghostmans@> has quit IRC11:28
lkclcorrect, there isn't. it was part of the realisation a couple months ago that it's missing (conflated with ffirst)11:29
lkcli'm currently mulling over:11:29
lkclbit 0 - pi11:29
lkclbit 1 - lf11:29
lkclbit 2 - els11:30
lkclbit 3 - sz11:30
lkclbit 4 - dz11:30
*** lxo <lxo!> has joined #libre-soc13:02
*** tplaten <tplaten!> has joined #libre-soc15:56
*** tplaten <tplaten!> has quit IRC18:31
*** tplaten <tplaten!> has joined #libre-soc18:45
*** tplaten <tplaten!> has quit IRC18:49
*** klys <klys!> has joined #libre-soc20:16

Generated by 2.17.1 by Marius Gedminas - find it at!