00:05:23 psilord [~psilord@c-69-180-173-249.hsd1.mn.comcast.net] has joined #sbcl 00:34:12 -!- slyrus [~chatzilla@173-228-44-92.dsl.static.sonic.net] has quit [Ping timeout: 265 seconds] 00:58:45 -!- Bike [~Glossina@wl-nat101.it.wsu.edu] has quit [Ping timeout: 248 seconds] 00:59:28 -!- samskulls [~user@S0106001111de1fc8.cg.shawcable.net] has quit [Remote host closed the connection] 00:59:55 -!- echo-area [~user@123.120.254.61] has quit [Remote host closed the connection] 01:06:10 Bike [~Glossina@gannon-wless-gw.resnet.wsu.edu] has joined #sbcl 01:50:49 -!- jaimef [jaimef@166.84.6.60] has quit [Excess Flood] 01:54:59 jaimef [jaimef@dns.mauthesis.com] has joined #sbcl 02:06:34 echo-area [~user@182.92.247.2] has joined #sbcl 02:12:54 ASau` [~user@p54AFFCA9.dip0.t-ipconnect.de] has joined #sbcl 02:16:27 -!- ASau [~user@p54AFF297.dip0.t-ipconnect.de] has quit [Ping timeout: 252 seconds] 02:22:28 slyrus [~chatzilla@137.164.119.50] has joined #sbcl 02:23:06 -!- scymtym_ [~user@2001:638:504:2093:baca:3aff:fe83:e736] has quit [Ping timeout: 245 seconds] 02:38:13 -!- jaimef [jaimef@dns.mauthesis.com] has quit [Excess Flood] 02:39:14 milosn_ [~milosn@user-5af505e0.broadband.tesco.net] has joined #sbcl 02:40:00 -!- milosn [~milosn@user-5af50c3b.broadband.tesco.net] has quit [Ping timeout: 245 seconds] 02:43:03 jaimef [jaimef@dns.mauthesis.com] has joined #sbcl 02:56:54 samskulls [~user@S0106001111de1fc8.cg.shawcable.net] has joined #sbcl 03:39:00 -!- christoph_debian [~christoph@ppp-88-217-35-39.dynamic.mnet-online.de] has quit [Ping timeout: 246 seconds] 17:28:20 ccl-logbot [~ccl-logbo@74.50.56.8] has joined #sbcl 17:28:20 17:28:20 -!- names: ccl-logbot yacks rpg oleo joshe Bike attila_lendvai Posterdati echo-area segv- kludge` slyrus scymtym_ prxq christoph_debian samskulls jaimef milosn ASau psilord _8hzp` reb |3b| pegu kanru fikusz xymox ferada easiere fe[nl]ix Blkt specbot jsnell foom @Krystof bege ivan`` flip214 danlentz nicdev pchrist antoszka asedeno luis ams redline6561 loke pkhuong daimrod Subfusc minion gko plathrop Tribal brucem 17:47:45 -!- attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has quit [Quit: Leaving.] 17:55:25 attila_lendvai [~attila_le@5.76.220.192] has joined #sbcl 17:55:25 -!- attila_lendvai [~attila_le@5.76.220.192] has quit [Changing host] 17:55:25 attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has joined #sbcl 18:03:31 -!- Bike [~Glossina@gannon-wless-gw.resnet.wsu.edu] has quit [Ping timeout: 272 seconds] 18:07:18 -!- attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has quit [Quit: Leaving.] 18:08:53 drmeiste_ [~drmeister@155.247.96.196] has joined #sbcl 18:09:59 Bike [~Glossina@wl-nat109.it.wsu.edu] has joined #sbcl 18:24:31 scymtym [~user@ip-5-147-115-29.unitymediagroup.de] has joined #sbcl 18:47:51 leuler [~user@p548FAC5D.dip0.t-ipconnect.de] has joined #sbcl 18:54:01 pkhuong: You were asking about committing the new register allocator. That made me want to try it out, so I cloned regalloc-cleanup-staging-rewrite-history. Unfortunately, it doesn't build. 18:55:06 When compiling src/code/early-alieneval.lisp, it breaks into the debugger with "The function SB!REGALLOC::PACK-ITERATIVE is undefined.". 18:55:49 leuler: I'm going over my history rewrites right now (: 18:56:15 It seems the commit added a new file but not the corresponding entry in build-order.lisp-expr. 18:57:07 ah yes. I wanted to conditionalize that and forgot to do it. 18:57:11 Let's make it unconditional 18:59:00 FEWER CONDITIONALS! 18:59:17 slyrus: otoh, more core ;) 18:59:55 leuler: should do it. 19:00:09 Thanks! Now trying. 19:00:39 -!- drmeiste_ [~drmeister@155.247.96.196] has quit [Remote host closed the connection] 19:01:39 -!- Posterdati [~kvirc@host111-230-dynamic.7-87-r.retail.telecomitalia.it] has quit [Excess Flood] 19:02:01 Posterdati [~kvirc@host111-230-dynamic.7-87-r.retail.telecomitalia.it] has joined #sbcl 19:05:06 -!- Bike [~Glossina@wl-nat109.it.wsu.edu] has quit [Ping timeout: 246 seconds] 19:11:42 This builds fine. 19:12:25 I'll make sure that the smaller commits to get there also build, and that I didn't break any voodoo in abarch's code. 19:13:05 Yes, it would be nice to remain able to bisect. 19:13:36 I'm pretty sure one day someone will either curse or thank me for all of these ftype declamations (or both?) 19:13:58 slyrus: so, one thing that worried me re ftype is that they might hinder development. 19:14:14 drmeiste_ [~drmeister@155.247.96.196] has joined #sbcl 19:15:29 Bike [~Glossina@wl-nat109.it.wsu.edu] has joined #sbcl 19:15:34 so maybe sb-fluid and/or very lax types (unless mandated by MOP/the standard) 19:16:10 hmm... interesting. I was thinking they might help flag potential problems if folks did change things. 19:16:36 take a look when i'm done and let me know what you think (it will still be some time) 19:24:23 Sagane [~Sagane@177.100-226-89.dsl.completel.net] has joined #sbcl 19:36:15 -!- scymtym [~user@ip-5-147-115-29.unitymediagroup.de] has quit [Ping timeout: 245 seconds] 19:39:00 -!- drmeiste_ [~drmeister@155.247.96.196] has quit [Remote host closed the connection] 19:40:09 pkhuong: I see something unexpected: I took the ...-again branch and compiled it as is (Linux, x86-64). The core size is 45,285,424 bytes. Then I changed the *register-allocation-method* in pack.lisp to :iterative so that SBCL builds itself using the new regalloc. The core size grows to 45,318,192 bytes. 19:42:28 leuler: yeah. Tweaking the parameters is something else. The iterative allocator forces TNs to the stack more aggressively, and is somewhat tuned for runtime performance more than speed (which could be changed as well). 19:42:57 * TNs are forcibly spilled when they're mostly live across calls. 19:43:40 *runtime performance more than space. 19:44:32 I see. I was expecting space savings; thinking about it, mostly because SBCL always showed unnecessary reg-reg moves. 19:56:46 leuler: removing the multiplication by loop depth in tn-spill-cost may save more space, for example. 20:02:02 -!- Sagane [~Sagane@177.100-226-89.dsl.completel.net] has quit [Quit: Leaving] 20:02:55 -!- Bike [~Glossina@wl-nat109.it.wsu.edu] has quit [Ping timeout: 245 seconds] 20:06:06 pkhuong: spot tests of disassembly outputs seem to show that the size increase comes from more unnecessary reg-reg moves :-(. But as SBCL builds itself with targeting disabled (you didn't change that? -- I didn't look) this may not be very interesting. 20:06:41 no, I didn't change anything to the logic that determines targetting desiderata. 20:09:36 the coalsecing logic could be worked on; e.g., TNs that should be in the same location aren't exactly treated as a single vertex during coloring. 20:23:11 Would that explain: For example in SB-PCL::%CHECK-CNM-ARGS, the old allocator moves NIL into EDX in three places immediately before jumping to L0 which contains the function return code. The new one moves NIL into EAX in all three places and needs an additional MOV RDX, RAX at L0. 20:23:26 Anyway, I don't have the time today but I will try the experiment to build SBCL with targeting enabled and look into this aspect again. 20:26:00 I'll see if I can come up with a simple tweak, but I want to make sure that branch does the same as abarch's branch first 20:27:09 -!- prxq [~mommer@x2f6d3c5.dyn.telefonica.de] has quit [Quit: Leaving] 20:27:24 prxq [~mommer@x2f6d3c5.dyn.telefonica.de] has joined #sbcl 20:30:59 Fine! Please don't let my obsession with space infect you ;-) 20:31:49 -!- leuler [~user@p548FAC5D.dip0.t-ipconnect.de] has quit [Quit: Good night.] 20:36:09 Yeah. I think I see how to improve targetting a lot. 20:36:10 helpful error message: Value of (VALUES) in ... is NIL, not a NIL. 20:37:00 slyrus: NIL is not of type NIL. 20:37:14 yeah, I know. it just reads funny. 20:45:59 Bike [~Glossina@wl-nat99.it.wsu.edu] has joined #sbcl 20:52:33 what does "note: type assertion too complex to check: (VALUES &OPTIONAL LIST &REST T)." suggest? 20:53:38 a return type of (values list &optional). 20:53:49 (the number of return values is unknown otherwise) 20:53:56 or sfunction 20:55:24 and why does that trigger a note? 20:56:37 because we don't check unknown-valued types. 20:58:03 -!- oleo [~oleo@xdsl-78-35-153-160.netcologne.de] has quit [Ping timeout: 245 seconds] 20:58:41 oleo [~oleo@xdsl-78-35-187-66.netcologne.de] has joined #sbcl 21:00:54 sdemarre [~serge@91.180.80.228] has joined #sbcl 21:08:13 Success: 45187120 output/sbcl.core 21:10:31 mm. wait. Forgot to force the new regalloc 21:18:48 minion: memo for leuler: it really seems to be something in the build script. recompiling %check-cnm-args doesn't show the same coalescing issue. 21:18:49 Remembered. I'll tell leuler when he/she/it next speaks. 21:29:13 drmeiste_ [~drmeister@wirelessNAT188.wireless.temple.edu] has joined #sbcl 21:29:48 well. and I have the same issue when recompiling with (speed 2). Not sure what's going on here. 21:43:20 -!- drmeiste_ [~drmeister@wirelessNAT188.wireless.temple.edu] has quit [Remote host closed the connection] 21:43:40 drmeiste_ [~drmeister@wirelessNAT188.wireless.temple.edu] has joined #sbcl 21:52:44 -!- prxq [~mommer@x2f6d3c5.dyn.telefonica.de] has quit [Quit: good night] 22:05:40 I think I see what's going on. At high speed, we try to pack registers where there's a lot of wired TNs... in our case, arg passing/return locations. 22:06:56 so some register in an inner loop gets RDX and interferes with the register we'd really like to get RDX because of targetting. 22:11:17 -!- sdemarre [~serge@91.180.80.228] has quit [Ping timeout: 248 seconds] 22:27:58 -!- _8hzp` [~user@87-95-84-107.bb.dnainternet.fi] has quit [Read error: Connection reset by peer] 22:28:04 -!- drmeiste_ [~drmeister@wirelessNAT188.wireless.temple.edu] has quit [Remote host closed the connection] 22:45:58 Vivitron [~Vivitron@c-50-172-44-193.hsd1.il.comcast.net] has joined #sbcl 22:55:13 I really don't know that that information is useful though. In fact, just preferring registers that are at low indices (e.g. rax, rcx) results in a smaller core, and avoids that specific targetting badness. 22:57:13 minion: memo for leuler: I pushed a tweak that should work better. 22:57:13 Remembered. I'll tell leuler when he/she/it next speaks. 22:57:24 Bike_ [~Glossina@wl-nat101.it.wsu.edu] has joined #sbcl 22:59:55 -!- Bike [~Glossina@wl-nat99.it.wsu.edu] has quit [Ping timeout: 272 seconds] 23:03:11 -!- Bike_ is now known as Bike 23:07:53 -!- Bike [~Glossina@wl-nat101.it.wsu.edu] has quit [Ping timeout: 272 seconds] 23:17:06 -!- rpg [~rpg@198-74-7-110.fttp.usinternet.com] has quit [Quit: rpg] 23:29:20 drmeiste_ [~drmeister@wirelessNAT188.wireless.temple.edu] has joined #sbcl 23:29:33 -!- segv- [~mb@95-91-240-228-dynip.superkabel.de] has quit [Remote host closed the connection] 23:29:40 and with that tweak, the new regalloc is 1 page tighter for the full core. 23:31:51 -!- scymtym_ [~user@2001:638:504:2093:baca:3aff:fe83:e736] has quit [Ping timeout: 245 seconds] 23:41:24 scymtym_ [~user@2001:638:504:2093:baca:3aff:fe83:e736] has joined #sbcl 23:56:40 -!- drmeiste_ [~drmeister@wirelessNAT188.wireless.temple.edu] has quit [Ping timeout: 245 seconds]