IRC logs for #openrisc Saturday, 2015-01-10

--- Log opened Sat Jan 10 00:00:49 2015
olofkTrying to understand these gcc/libc interdependencies. Do I need to build a stage 2 gcc for every libc?12:31
olofkSomething like this:12:31
olofk1. binutils12:31
olofk2. gcc stage 112:31
olofkfor c in libc:12:31
olofk  libc && gcc stage 212:31
olofkWorst attempt at pseudo code ever12:32
olofkOr is it:12:36
olofk1. binutils12:36
olofk2. for c in libc:12:36
olofk  gcc stage 1 && libc && gcc stage 212:36
poke53282binutils, linux-headers, gcc stage 1, libc, gcc stage 213:01
poke53282stage 2 is least needed by uclibc, musl and glibc.13:03
ysangkokpoke53282: when to use throw, and when to use message.Debug? some fatal errors use message.Debug, i don't understand that15:29
poke53282message.Debug is used for everything.15:38
poke53282And if his problem is critical also abort is executed.15:38
poke53282One plan is have a new function message.Error, which sends the abort signal itself.15:39
poke53282So, everytime, the abort function is called, change message.Debug to message.Error15:39
ysangkokok15:45
poke53282This are just a few lines of code.15:51
poke53282I can implement it15:51
ysangkokpoke53282: it would be useful if there was a way to invoke console.warn too... for example, if the file sizes don't match15:57
mor1kx[mor1kx] bandvig pushed 1 new commit to withfpu: https://github.com/openrisc/mor1kx/commit/ada2df5dabd711b43f8878f511001cca4b039c0916:01
mor1kxmor1kx/withfpu ada2df5 Andrey Bacherov: Size optimization: share post-operation (add/sub, mul/div, i2f) align shifter .16:01
poke53282ysangkok: You mean message.Warn16:02
ysangkokah soo16:02
ysangkokall right :)16:02
stekernpoke53282: the stdatomic-compare-exchange-1 fail seems to be related to ltoi16:03
stekern-i16:03
stekernit doesn't fail without -flto but does with16:04
poke53282try with and without inlining16:04
stekernwas that directed to me?16:07
poke53282Yes16:08
poke53282ysangkok: Added a commit.16:09
stekernok, I don't understand16:09
poke53282You are free to use those functions.16:09
ysangkokgreat :D16:09
poke53282stekern: well I think, inlining and -flto ist partly related.16:11
stekernmaybe, but there's nothing related to inlining in that test16:11
poke53282Ahh, Ok#16:11
stekernhttp://pastie.org/982405916:12
stekernthat's the part of the testcase that fails16:12
poke53282Did you try it on your hardware?16:13
stekernand that atomic_compare_xxx call boils down to the atomic stuff blueCmd added16:13
stekernI'm running in or1ksim16:13
stekernI'm diffing the disasm of the failing and passing now16:14
stekerncan't see any difference in the actual atomic related code that's generated though16:16
stekernso might be a general lto problem for us16:16
poke53282diff with objdump -d ?16:22
stekernyes16:22
ysangkokpoke53282: since the debug messages in the worker are relayed to the master using Send, it is hard to get backtraces for worker problems18:07
ysangkokpoke53282: using console.warn, Chrome would show the correct line in the console in the right side... but this would require no indirection18:08
ysangkokpoke53282: currently i just set a breakpoint on the worker debug function18:08
poke53282Yes, the lines are lost.22:37
poke53282But I had never problems with such things. The error message string is more than enough.22:38
ysangkokpoke53282: new PR https://github.com/s-macke/jor1k/pull/5123:21
poke53282great, how did you change the base.xml files?23:31
poke53282Ahh, manually I guess23:32
poke53282Another hack gone.23:41
--- Log closed Sun Jan 11 00:00:50 2015

Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!