Friday, 2022-08-19

ghostmansdlkcl, about 5bf3c705cc0736cf34456d173fc4c50da5aba03703:59
ghostmansdI know what the fix is :-)03:59
ghostmansdthe problem is, I don't understand what's going wrong if I pass FSI instead of SI04:00
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC04:36
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC06:51
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc06:52
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC08:57
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc08:58
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc09:51
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC10:38
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc10:38
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC10:54
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc10:58
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC11:11
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc11:20
programmerjakeinteresting reading:
markosnice article, I like the historical background given behind all those languages12:07
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC12:25
*** octavius <octavius!> has joined #libre-soc13:35
*** octavius <octavius!> has quit IRC13:39
*** josuah <josuah!~irc@> has quit IRC15:27
*** josuah <josuah!~irc@> has joined #libre-soc15:32
ghostmansdI'm now re-considering the way we handle field mappings. I posted some of the investigations here: Below you can find how the new fields look.16:10 works16:16
ghostmansdUnsurprizingly, though: I haven't yet switched the code to these new classes (except for insndb and binutils).16:17
*** tplaten <tplaten!> has joined #libre-soc16:23
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC16:37
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc16:37
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC16:46
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc16:50
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC17:00
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc17:01
lkclghostmansd, as long as PowerDecoderSubset can recognise the fields in their original form (by CSV column names), it's perfectly fine17:18
lkclmore to the point17:18
lkclif the fieldnames created as a PowerOp remain absolutely the same under all circumstances and under absolutely no circumstances change17:19
lkclfor all **TWELVE** uses across twelve pipelines17:19
lkcland in the ISACaller Simulator17:19
lkclit's "fine"17:19
lkclthe good news there, being: get it wrong it'll be bloody obvious17:20
ghostmansdWell I'm mostly centered on pysvp64asm/pysvp64dis/binutils for now17:21
ghostmansdBut, what was obvious too, many our tricks are really covered by such fields, let's call them mappings.17:22
ghostmansdMostly stuff from
ghostmansdPlus also some bits which were in the (RM and Prefix fields).17:23
ghostmansdWhat I really like about it is how it's remapped according to the actual layout.17:25
ghostmansdFor example, how RM adopts the layout to match it to Prefix bits.17:25
ghostmansdIn fact, inherit or incorporate these structures as many times you want, there will be exactly 1 selectable int throughout the whole course. And only 1 layer of access: the fields are remapped upon the time the class is created.17:28
ghostmansdThat is, svp64_insn.prefix.rm.elwidth keeps the same SelectableInt created by the time we created svp64_insn instance.17:29
ghostmansdAnd all offsets were already "statically" known upon module import. That is, when I access elwidth, I access SI immediately with the exact index.17:29
ghostmansdThis index already took into account all the changes related to layout.17:30
lkclyes - i very much wanted the spec to be directly reflected in the code, so preserving field names/layout was important19:45
lkclthere is a huge penalty associated with arithmetic done on FSIs though! it literally has to do a bit-level reconstruction of the value, perform the arithmetic operation, then push the result back again *one bit at a time* into the underlying SI19:51
lkclwhich is mental19:51
lkcland we don't care19:51
lkclISACaller is still 1,000 times faster than running against TestIssuer, so serves its purpose extremely well19:52
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc20:58
*** tplaten <tplaten!> has quit IRC20:59
*** ghostmansd <ghostmansd!> has quit IRC21:43

Generated by 2.17.1 by Marius Gedminas - find it at!