IRC logs for #openrisc Monday, 2014-10-06

--- Log opened Mon Oct 06 00:00:25 2014
poke53282Slides for the talk finished.03:24
poke53282Looks like the Hotel-Pension Theresia-Regina is the closest hotel to the conferene. 30 Euro per night, but no own shower.05:35
olofkah yes. Time to get a hotel now I guess06:58
olofkAnd book the flights06:58
stekernI've booked a room at hotel europe07:14
olofkOk. Flights booked now09:13
wallentoand what is it now?09:18
olofkdrakhogekko: Hello10:16
olofkwallento: what is what?10:17
wallentoolofk: your flight? munich or frankfurt?10:57
olofkwallento: Munich friday 18:1010:57
stekernhmm, I activated wb_stream_reader and now the writer doesn't work anymore15:11
stekernmaybe I've screwed up the intercon...15:11
poke53282stekern: Did you make progress on the gcc internal compiler error?16:29
poke53282It's strange, but for me the error happens only with the optimization -O1 and not with -O2 and so on.16:30
poke53282If i don't use -O1, but instead tons of -f... options it doesn't crash.16:39
stekernpoke53282: hmm, I started to look into it, yes. But I don't think I made any big break-throughs17:31
poke53282Yes, this is a hard problem. I guess we have to use -fdump-final-insns and find the exact command where it fails.17:36
stekernI dumped the gimple trees for O1 and O2:
stekernI think this is the problem maybe:
poke53282never heard of gimple trees17:39
poke53282At the moment I am reading LISP code:
stekernyou get them with -fdump-tree-all17:40
poke53282at the moment the code fails after he gernerated this rtl code.17:42
poke53282And I guess, it is just a special instruction there which fails. On a much lower level than the gimple trees.17:43
poke53282Because -O2 seems to work for me.17:43
stekernyes, 02 works17:45
stekernbut that's probably just because the problem got optimized away17:46
poke53282Ok, let me put a printf in the expr function to see at which rtl expression it fails.17:46
poke53282Is there an lxr project for gcc?17:47
poke53282online gcc cross reference?17:47
stekernhaven't seen one at least17:50
poke53282ICE happens in in expand_expr_addr_expr_1 because the result is (reg/v:TI 86 [17:56
poke53282func ]), which is obviously no memory.17:56
poke53282Issue seems here to be that x64 ABI has pre-allocated stack-area on call, so17:56
poke53282for the first 4 arguments no stack-allocation is necessary.  So we might need17:56
poke53282to specialize allocate_stack_slots_for_args-hook for this target.17:56
stekernhmm, yeah. that seems like it might be related17:58
poke53282It happens directly between the translation of the tree view and the rtl view.18:08
stekernolofk: shouldn't this be DEPTH_WIDTH:0?
olofkstekern: I'm planning to remove that fifo_fwft19:20
olofkstream_reader is the only thing that still uses it, so if that's what you're looking at, please replace that with stream_fifo instead19:21
olofkOr if you need a counter, do something like what is done in wb_stream_writer_fifo19:22
stekernI'm just looking at my old local fifo that works and one that doesn't19:26
stekern(which happened to come from .cache)19:26
olofkCool! Just read a tweet that some guy has got Parallella working under FuseSoC20:32
olofkLooks like only four small patches were needed20:38
olofkMaybe this means it's time to finally take my parallella out of the garage and power it up20:39
olofkstekern: Did you say that symlinks works in FuseSoC now? I'm wondering because it looks like I never pushed that commit20:41
--- Log closed Tue Oct 07 00:00:26 2014

Generated by 2.15.2 by Marius Gedminas - find it at!