IRC logs for #openrisc Sunday, 2014-09-28

--- Log opened Sun Sep 28 00:00:13 2014
poke53282Hmm, atgreen, are you the maintainer of libffi?03:34
atgreenhi, poke53282 . yes.03:34
atgreenI just read the pull request for openrisc support.03:35
poke53282when I read your name, I was wondering. I remembered something :)03:35
poke53282Yes, I wrote this and pulled today.03:35
atgreenI want to fix a test suite issue before I merge it.03:35
atgreentrying right now.03:35
poke53282Just that you know, I neglected the complex number testcases. But they also failed on my x86 machine.03:36
atgreenthat's what I need to deal with.  I want to mark those as expected fail on unsupported hosts.03:37
poke53282I can tell you , that every speciality of the openrisc ABI was included in your testcases.03:37
poke53282So I can test it, if you want.03:38
poke53282In principle what I think you should include in your testsuite is a calculation with twenty integers or so before and after the ffi calls, but dependent on each other.  You could easily accidently overwrite a register or stack entry without noticing. Maybe this is already included in you testsuite, but I missed it.03:42
poke53282Oh, I read your blog. moxie-sim uses part of my code. :)04:02
atgreenpoke53282: oh? what part?04:54
atgreenpoke53282: just fixed the testsuite and merged your port.04:54
atgreenoh.. that part!  you're the jor1k guy!04:56
atgreenand a fellow Canadian04:57
poke53282the terminal and terminal-input04:57
poke53282<- jor1k guy04:58
poke53282but I am not Canadian. I live currently in Vancouver, yes.04:58
poke53282But actually I am German.04:58
atgreenthat's OK.. I'm only half Canadian. :)04:58
atgreen1am.. time top go. l8r04:58
poke53282thanks for merging04:59
poke53282good night04:59
poke53282blueCmd: Next official version of libffi will contain OpenRISC support.  Time to ask you, if the glibc problem is solved ;)05:01
-!- lauri__ is now known as lauri08:24
blueCmdpoke53282: wooo!13:38
blueCmdpoke53282: nah, I haven't done anything on that13:38
sb0when is it going to be the same for gcc and llvm?15:13
sb0it's been way faster for ARM...15:14
sb0...risc-v might get that right, but they take FPGA resource bloat to new heights :/15:16
sb0bottom line: open source CPUs suck15:16
stekernI agree, ARM serves you everything, nothing left to do. Boring like hell.16:01
olofkDoes anyone have upsizers and downsizers available for a stream interface?18:03
poke53282this 20030125-1.c testcase is really strange. I searched for half an hour. It is about optimization of floor and sin when converting floats and doubles. According to C99, the optimization for both are different. But I can't see a reason, why the or1k part of gcc should be responsible for that. We use soft float.18:12
stekernpoke53282: can't it be that optimizations are done before the result is handed to the actual insertion of the float 'instructions' (hard or float)?19:40
* olofk is awesome!19:44
poke53282stekern: I don't really understand what you mean. At which position?20:06
stekernolofk: can you be even more awesome and fix fusesoc's inabilitity to work with soft links? ;)20:06
poke53282For me this looks like a frontend problem. C99 is definitely defined for us.20:07
stekernobviously it has worked before, because I've used it in the travis setup for the i2s core20:07
stekernpoke53282: it was just a comment to the last sentence of what you said (why the or1k part of gcc should be responsible for that)20:08
poke53282But I am testing several things right now. Just looking at other code and change ours and compile and start the testsuite.20:08
poke53282we are the only ones who use uclibc-stdint.h for example.20:08
stekernpoke53282: btw, iirc, blueCmd didn't get that failure against glibc20:08
poke53282Ok, that's interesting.20:09
stekerndo we use that for musl (and glibc) too?20:09
poke53282for uclibc this testcase is omitted20:10
poke53282Yes, for all20:10
poke53282Look at gcc/config.gcc20:10
poke53282       tm_file="${tm_file} dbxelf.h elfos.h or1k/elf.h gnu-user.h linux.h or1k/linux-gas.h or1k/linux-elf.h uclibc-stdint.h"20:10
poke53282       tmake_file="or1k/t-or1k or1k/t-linux ${tmake_file}"20:10
poke53282       tm_defines="${tm_defines} OR1K_DELAY_DEFAULT=OR1K_DELAY_ON"20:10
poke53282       ;;20:10
poke53282I remove this part, other option would be to use glibc-stdint.h20:11
poke53282the other problem is, that if I link with -p I get an error "cannot find gcrt1.o: No such file or directory"20:19
olofkstekern: Yes, I just was that awesome :)20:27
poke53282I get the "-p" in sabotage linux too. So, I wonder how many of those testcases are related to musl.20:29
poke53282I ask in the other chatroom.20:30
stekernok, once again my i2s core passes the travis tests20:33
olofkstekern: That was quick20:33
stekernolofk: well, it was just a matter of fixing all the orpsoc/fusesoc breakage ;)20:34
stekernolofk: if you still don't get notifications:
olofkstekern: I actually got the first one. It was just that gmail's "Smart Inbox" or whatever they called it filed it automatically under "Social" or "Promotions" for some reason21:15
olofkNever really understood that. I have lots of folders already in gmail with filters to move stuff to where it should belong, so why divide the inbox in another way21:16
stekernI have that "smart inbox" feature turned off21:18
stekernbecause, I, as you, didn't get the point of it21:18
stekernI rather sort my mail according to my own rules21:19
stekern(except for perhaps spam)21:19
poke53282stekern: I slowly get the impression, that a lot of the failed tests are related to musl.21:19
stekernpoke53282: you should compare against the results that blueCmd got (search the irc logs for them, or ping him)21:19
stekernI think we actually had very few failures that were intersecting21:20
poke53282e.g. test autogen: Command not found fails in sabotage linux too21:21
poke53282e. g. test 20030125-1.c fails in sabotage linux too.21:22
stekernfor x86?21:23
poke53282in the sabotage linux chroot21:23
stekernok, that's comforting in the sense that then it's at least not "our" fault ;)21:23
poke53282maybe I played a little bit too much around with this file.21:24
poke53282arghh, I forgot the -O2 option.21:27
poke53282No, it doesn't fail in sabotage linux21:27
stekernback to blaming ourselfs then ;)21:28
stekernolofk: as I note in that commit message, only the writer is supported. next step is to start playing with the reader and get some capture working21:29
poke53282I run the gcc testsuite in sabotage linux now.21:30
olofkstekern: Nice work! The reader probably doesn't work though. I abandoned that in favour of the writer, but I am soon at the point where I need the reader too21:35
olofkBut first I need to implement some infrastructure components for streaming21:36
olofkMy downsizer seems to work. Wrote it down on paper, implemented it blindly in 41 lines of verilog and it actually seems to work straight away21:37
stekerndownsizer, sounds like a core for management22:15
poke53282                === gcc Summary ===22:16
poke53282# of expected passes            8851922:16
poke53282# of unexpected failures        5622:16
poke53282# of unexpected successes       4222:16
poke53282# of expected failures          26822:16
poke53282# of unresolved testcases       822:16
poke53282# of unsupported tests          125722:16
poke53282but the overlap is not that high22:22
poke53282only 7 testcases seem to overlap22:24
--- Log closed Mon Sep 29 00:00:15 2014

Generated by 2.15.2 by Marius Gedminas - find it at!