Friday, 2023-03-31

*** gnucode <gnucode!~gnucode@user/jab> has quit IRC00:03
sadoon[m]Good news, already have a script running that scans the full debian pool for .dsc files and uses the API call to instruct mini-buildd to build the package01:43
sadoon[m]Somewhat bad news: it kinda requires that you have a local mirror or an sshfs to the mirror to do the scan01:44
sadoon[m]I'm only concerned with it working for now though01:44
sadoon[m]The rest can be enhanced later01:44
lkclsadoon[m], nice. yes sensible. we can always get another grant and/or request a change to an MoU (not too often, though) to add an additional milestone02:16
sadoon[m]Technically you only need the repo structure with the .dsc files in it, nothing else03:09
sadoon[m]I can probably find a way to script a tar -cf into only taking in .dsc files03:10
sadoon[m]The reason is the script needs to know which packages are supposed to be built, and I do this by starting in a "pool", for looping over every folder (letters,etc), for looping again inside for every package, finding the .dsc file in there, and pushing it using the mini-buildd API like such03:11
sadoon[m]mini-buildd-api -l DEBUG --auto-confirm port_ext $PROTO://$USER@$SERVER:$PORT $SRCREPO/$dsc $REPO03:11
sadoon[m]Or more readable:03:12
sadoon[m]mini-buildd-api --auto-confirm port_ext http://admin@bookworm-mini-buildd-ppc64.sforza.local:8066 sid-bookworm-stable03:12
sadoon[m](Just an example)03:12
sadoon[m]Everything is checked with signatures against the debian keyring and built signed with a special key that mini-buildd assigns03:13
sadoon[m]There are only a few workable issues atm, the main one being the package naming and repo naming are not in sync with what debian requires03:14
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC06:26
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc06:26
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC06:43
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc06:44
*** Ritish <Ritish!~Ritish@2401:4900:1f2d:bbee:a5:6bc3:fcfc:d448> has joined #libre-soc07:14
*** Ritish <Ritish!~Ritish@2401:4900:1f2d:bbee:a5:6bc3:fcfc:d448> has quit IRC10:16
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC10:19
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc10:20
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC11:48
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc11:48
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC12:19
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc12:26
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC16:25
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc16:26
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC17:11
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc17:11
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC18:33
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc18:34
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC19:00
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc19:01
jni don't know if it has been mentioned here yet, but: - OpenRAM, a libre SRAM generator, has apparently just been published19:44
programmerjakeapparently it's been published since at least 2020:
*** gnucode <gnucode!~gnucode@user/jab> has joined #libre-soc19:57
jnhm, then i got confused by a tweet that seemed to claim it was published very recently. anyway, might be a useful tool still!20:33
programmerjakeyup! thx!20:37
lkclit's another one of those projects that on the surface seems great but when you investigate further has some technical limitations, i don't recall the precise details, i just recall that "there's some caveats"22:30
lkclmarkos, gnucode, perhaps the biggest "sleeper" in SVP64 is the "Data-Dependent Fail-First" mode22:31
lkclit's what allows a Vectorised implementation of a full and i do mean full implementation of strncpy to be implemented in 10 instructions22:31
lkclone of which is "move r3 to CTR"22:32
lkclthe other is "set r0 to 0"22:33
lkclit's required to use sv.bc which works directly in conjunction with the truncation of VL, to subtract *VL* from CTR rather than, as is done in RVV,22:35
lkcluse an intermediary GPR to *receive* a *copy* of VL that is then converted into a bitmask22:36
lkclthen incremented by one22:36
lkcloh dear.22:39
gnucode10 instructions vs. 36.  That's impressivr!23:14
lkclwell 11 vs 2523:40
lkclthe first trick is the "post-increment" mode added to LD/ST, which eliminates the usual subtraction needed23:42
lkclthat's one instruction gone23:43
lkclthe fail-first on sv.cmpi is set to hunt for the zero *and include it* (/vli - truncate VL *inclusive*)23:45
lkclstore that many VL bytes but again post-increment, avoiding yet another instruction (pre-subtraction)23:45
lkclsv.bc/all tests CTR "as usual" but subtracts VL, and also stops at the fail-point23:46
lkclit's all pretty obvious stuff, just a little bit more sophisticated23:47

Generated by 2.17.1 by Marius Gedminas - find it at!