IRC logs for #openrisc Saturday, 2014-09-13

--- Log opened Sat Sep 13 00:00:50 2014
-!- E11Fangirl is now known as kira00:36
-!- kira is now known as Guest4342300:37
-!- Guest43423 is now known as E11fangirl00:37
-!- Netsplit *.net <-> *.split quits: Amadiro00:38
-!- Netsplit over, joins: Amadiro00:45
-!- Netsplit *.net <-> *.split quits: rokka00:59
-!- Netsplit *.net <-> *.split quits: enghong, jeremy_bennett01:06
-!- Netsplit *.net <-> *.split quits: trevorman01:20
poke53281Hmm, gtk+ compiles.  *Formally* I meet all requirements to compile Firefox.01:45
mor1kx[mor1kx] skristiansson pushed 2 new commits to master:
mor1kxmor1kx/master 438e3fc Stefan Kristiansson: change module definitions to verilog 2001 style05:14
mor1kxmor1kx/master 973f039 Stefan Kristiansson: cappuccino/lsu: add option to disable store buffer...05:14
-!- Guest65615 is now known as exparrot05:26
stekernbah... I always forget mor1kx.v05:29
mor1kx[mor1kx] skristiansson pushed 2 new commits to master:
mor1kxmor1kx/master 88522fa Stefan Kristiansson: top: change module definition to verilog 2001 style05:37
mor1kxmor1kx/master d120827 Stefan Kristiansson: add missing FEATURE_STORE_BUFFER to top module...05:37
exparrotWhen compliling the toolchain, i've followed the instructions and built or1k-src with sim but without ork1sim, now I've compiled and installed or1ksim ( some unittest fails btw, is that a good sign?) should I recompile or1k-src (and therefore or1k-gcc) again with --enable-or1ksim? what will it get me?06:45
helmuthi. can someone update me on the state of or1k gcc upstreaming?06:59
-!- lauri__ is now known as lauri07:03
poke53281stekern: Is it possible to update gcc to 4.9.1? I have an compiler error which is maybe not related to openrisc07:05
poke53281this is the error I get
poke53281I can only compile -O0.07:57
poke53281with -O007:57
stekernpoke53281: we can give it a try ;)08:10
-!- knz_ is now known as knz08:18
stekernpoke53281: I've updated the or1k branch and created a musl-4.9.1 branch08:56
olofkhelmut: blueCmd probably knows the most about this, but I think that we are still missing the copyright assignment from one of the authors10:12
olofkexparrot: I *think* that compiling the toolchain with enable-or1ksim just allows you to use or1ksim as the simulator when you run the toolchain regression tests10:13
olofkWhat I normally do is run or1ksim as a stand-alone simulator, in which case there are no dependencies between or1ksim and the toolchain10:13
olofkwhich unittest fails btw? Is that when you run make check?10:14
olofkhmm.. I'm getting all kinds of fails when I do make check10:16
helmutolofk: the latest thing I heared of was
helmutolofk: that was an answer from blueCmd, but he went pretty much silent for the past two months10:23
olofkhelmut: He might show up if we call on him. blueCmd! blueCmd! blueCmd! blueCmd! blueCmd!10:26
helmutI guess he'll just be annoyed, because copyright assignment is tedious and boring. Still it is useful to have this code upstream.10:27
olofkAnyone familiar with the or1ksim regression tests? I'm getting several failed testcases, but I suspect they are false positives10:28
helmutI'm actually asking, because I ponder kicking or1k from CI unless it progresses. (
olofkThe log file seems to have problems with line breaks10:28
olofkhelmut: It would be great if you could wait with that for a while. We got the annual OpenRISC conference coming up in less than a month. Perhaps we can work something out then10:29
helmutwith "while" == "1 month"?10:32
olofkYeah, or if blueCmd says otherwise.10:34
olofkBut looking at the log I see this "*** Configuration or1k-unknown-linux-gnu not supported"10:35
helmutyes, that is due to the license clearing aspect10:53
helmutthe debian gcc maintainer refuses to take patches that are not submitted to gcc upstream.10:54
helmutsince the or1k gcc changes still haven't the copyright assignment done, they are considered unsubmitted10:54
helmutso the "not supported" bit should change fairly quickly when that copyright hassle is done10:54
blueCmdolofk: I have been summoned11:12
blueCmdI skimmed through the backlog, is this about gcc?11:12
blueCmdIf so, I still need to reach Yungsook for the floating point stuff, or we need to rewrite that11:14
helmutblueCmd: yes. I know it is annoying11:14
helmutblueCmd: so is that just one contributor left to do the assignment?11:15
blueCmdI sent some blind emails to what a person that used to work at the same company guessed the email would be, no response11:15
blueCmdhelmut: yep11:15
blueCmdhelmut: at least that I've been unable to reach11:15
helmutok... any how many still need to actually do the assignment?11:16
blueCmdhelmut: that's hard to know, I *think* everyone has sent their assignments, but that's only their word11:16
blueCmdI might be able to ask assign@gnu11:17
helmutI think that it is reasonable to trust contributors when they say they did it11:17
helmutthanks for the status update11:17
helmuthope you get the last issue sorted out quickly11:18
blueCmdthere is nothing quick about this :P11:18
helmutwell, rewriting may be possible.11:18
helmutthis copyright assignment thingy is annoying. nobody but the fsf get away with that. not even canonical.11:20
blueCmdMatjaz needs to confirm his assignment and then it's Yungsook11:26
blueCmdhelmut: I appriciate that you ping me, sometimes I need a kick in the but11:31
blueCmdhelmut: also, we need a code reviewer for the GCC commit11:32
blueCmdwhich apparently can be a pain to find, but I haven't tried11:32
helmutblueCmd: what qualifications are needed?11:37
blueCmdhelmut: that's what I know11:40
helmutbetween the lines, it says that you need a gcc developer11:41
helmutblueCmd: the good thing is that I guess having sent the formal mail is enough for doko to carry the patch in debian for a while.11:42
blueCmdhelmut: that's probably true yes11:42
helmutalso note that ppc64el and arm64 are progressing quite well, so that should make some space on d-ports "soon" (I estimate 4 months)11:43
helmutotherwise, kick powerpcspe to make space for or1k. :)11:43
blueCmdhelmut: cool! yes, they are doing very good work that helps openrisc a lot11:43
helmutso will you wait indefinitely for Yungsook, or is there some deadline at which point you discard his contributions?11:45
blueCmdI just now sent a linkedin message, maybe that will work11:46
blueCmdhelmut: I'll wait until someone steps up to rewrite the relevant code or that she responds11:47
blueCmdI don't think I can rewrite it since I know the code11:47
helmutsounds like you should use the openrisc conference to find someone willing to replace that code11:53
blueCmdI could, but I'm not going - I had a last minute thing come up that I need to attend :(13:01
helmutblueCmd: ok, so the point I was trying to make is that the conference seems like a good deadline. maybe someone else can bring that issue up13:29
blueCmdhelmut: sounds fair enough13:39
-!- Netsplit *.net <-> *.split quits: poke53281, trem13:55
-!- Netsplit over, joins: simoncook13:55
-!- Netsplit *.net <-> *.split quits: sb0, funfunctor13:55
-!- Netsplit over, joins: poke53281, trem, sb013:56
-!- Netsplit *.net <-> *.split quits: kiwichris13:56
-!- Netsplit over, joins: funfunctor13:57
-!- Netsplit *.net <-> *.split quits: poke53281, hesham14:26
-!- Netsplit over, joins: hesham14:26
-!- Netsplit over, joins: poke5328114:27
mor1kx[mor1kx] skristiansson pushed 3 new commits to master:
mor1kxmor1kx/master 5cfe894 Stefan Kristiansson: cappuccino/wb_mux: remove unused lsu_valid_i input15:17
mor1kxmor1kx/master 3f7597f Stefan Kristiansson: execute_alu: remove & {OPTION_OPERAND_WIDTH{1'b1}} from mul...15:17
mor1kxmor1kx/master d7a7d4f Stefan Kristiansson: execute_alu: Set result of disabled units to 0...15:17
stekernhmmm, I want to kill this:
-!- mboehnert is now known as martinb115:21
-!- martinb1 is now known as martinboehnert15:22
-!- martinboehnert is now known as hauslude15:22
-!- hauslude is now known as martinboehnert15:22
stekernI can't see why I need that, at least after I change the WRITE state to go back to IDLE instead of READ when it's done...15:22
-!- martinboehnert is now known as huaslude15:23
-!- huaslude is now known as hauslude15:23
heshamAre there performance counters that I can use (in or1ksim and/or mor1kx)? It's annoying (and buggy) to use tick timer for both  task scheduling and benchmark/profiling results.15:29
stekernhesham: what kind of performance do you want to measure?15:32
heshamCache functions15:32
stekernwhat kind of cache functions?15:33
heshamThere are some tests in RTEMS that tries to estimate some results about cache. First it tests accessing an array (1024 bytes) with cache invalidated, and then access it again when cache is worm. Both functions (with access of this 1024 bytes) need some timing results.15:35
stekernto answer your question, mor1kx doesn't implement the performance counters15:35
stekernaha, or1ksim will not give sensible results there15:35
stekernbut why is it error prone to use the tick timer?15:36
heshamSay I am getting the TTCR before the calling a cache function, and then read it again to calculate delta, TTCR may be overflowed during this cache function15:37
heshamI am using tick timer restart mode for scheduling15:37
stekernbut can't you just account for the overflow?15:38
stekernand isn't there any time-keeping 'frameworks' in RTEMS?15:38
stekernor1ksim is (naturally) slower with caches turned on15:39
heshamThere is time-keeping for each thread I think, and it's incremented every tick timer interrupt, I may use it.15:40
heshamI tried with or1ksim, it does not give reasonable results for cache functions15:40
heshamThe worm cache time is the same as if was not pre-fetched with data15:41
heshamSense the time-keeping variable in RTEMS is updated every tick timer interrupt, it won't give an accurate result too..15:42
heshamWill QEMU give sensible results ?15:49
heshamAnd how can I be sure that Cache functions are working properly on or1kim and/or QEMU?15:55
olofkblueCmd: Sorry to hear that you won't be able to come to orconf16:17
olofkstekern: Do you want to get rid of the parameter for simple_dpram_sclk, or just be able to set it to 0 for that instance16:18
poke53281Thanks stekern. Toolchain is built. Let's hope the best16:25
stekernolofk: I want to be able to set it to 0 for that instance (and the other instance in dcache)16:26
olofkah ok16:28
olofkI really should get a proper JTAG adapter. Considering getting a Bus Blaster from Dangerous prototypes16:29
olofkAll I really need is a breakout board for something like FTDI 223216:32
poke53281damn, no luck16:36
poke53281I think this is right18:15
poke53281firefox libraries now and then have headers in which they need such information.18:17
stekernpoke53281: can you back up a couple of steps ;)18:18
stekernbut, as far as I can see, those defines are right18:18
poke53281I try to compile the Netscape Portable Runtime (NSPR)18:22
poke53281Well, it's already done18:22
stekernoh, I know why I need the bypass now18:23
poke53281And now I try to compile firefox.18:23
stekernwhen there are several non-word writes to the same word coming in serie18:24
poke53281sicnce 20 minutes it compiles. So far, so good. I don't know how long firefox compiles. 1 hour?18:25
poke53281Hmm 54 standard build units18:26
stekernbut for that, I only need the bypass on the way_data_ram18:27
stekernpoke53281: I like that we are having two monologues going on at the same time ;)18:27
poke53281well, mybe someone is interested18:29
stekernI'm interested, don't get me wrong there, I just don't have anything to respond to what you are saying ;)18:31
poke53281the problem with you monologue is, that I don't know what you are saying.18:34
stekernI've sent mor1kx to fat-camp, but sometimes it just breaks down and cries for donuts... basically ;)18:39
blueCmdolofk: yeah, me too. it was a hard choice, I'll tell you that18:46
poke53281your FPGAs are too small?19:44
mor1kx[mor1kx] skristiansson pushed 2 new commits to master:
mor1kxmor1kx/master ed8f21c Stefan Kristiansson: dcache: minor whitespace fix19:45
mor1kxmor1kx/master 77a24c1 Stefan Kristiansson: dcache: disable bypass for tag ram...19:45
stekernmine aren't, but others might be ;)19:45
poke53281sooner or later I want to see an SMP system with 32 Cores on an FPGA19:47
stekernfor that, my FPGAs are too small19:48
stekernso in that sense, I'm actually motivated to do the area optimization for myself19:49
stekernI might fit 16 on my sockit board19:49
stekern...I think19:49
stekernbut I should try 4 before that ;)19:50
poke53281r is already pretty nice.19:50
poke53281the keys are next to each other19:51
stekernthey are19:51
poke53281and high have big fingers19:51
poke53281and I ....19:52
poke53281and my brain is much faster than my fingers. Therefore I always forget to write words in my sentences.19:53
poke53281Finally I have wrote a patch to alter the amount of memory before I boot. No recompiling of the kernel necessary.19:56
stekernI do that too, but I think that's just because I get distracted easily19:56
stekern"alter", in what way?19:56
poke53281set the amount of memory19:57
stekernyou mean what's in the .dts?19:57
poke53281I analyzed the compiled dts file19:57
poke53281would be nice to have something like u-boot.19:59
stekernyou could just keep around several dt blobs and use some small bootloader (or let jor1k do it) that loads it into mem and pass the pointer to it in r319:59
poke53281pointer in r3?20:00
stekernyes, to the memory where the compiled device tree is20:01
poke53281this is implemented?20:01
poke53281nice, didn't know this20:01
stekernthe kernel looks for a device tree in the memory pointed at by r3, if it doesn't find anything there, it use the built-in20:02
poke53281would be nice if this would work without a compiled device tree20:03
stekernyeah, well, the (higher levels) of the kernel expects a blob20:04
stekernu-boot and barebox can of course de-compile it and let you do changes20:05
stekernnot sure how bulky the fdt handling code is by itself, you could take a look if you can rip it out20:06
poke53281hmm, I might take a look. At the moment I am fine with my solution.20:13
stekernare you poking at the fdt in the kernel?20:16
poke53281yes, searching for "memory\0"20:16
poke53281I was afraid that they use a checksum.20:17
stekernyeah, I'm just trying to push you into a direction of doing something I'd like to have as well ;)20:17
stekernfor my sockit board, it'd be nice with a mini bootloader that could parse device-tree20:19
poke53281mhh, you are sure u-boot can do this?20:20
stekernI could of course just let the ARM core parse it, modify it and then dump it to a known memory area20:20
stekernyes, I'm sure, I've used it20:20
olofkAny pointers for where to start if I would want to write a kernel driver for wb_streamer?20:41
olofkI could always ask Alessandro at orconf :)20:41
olofkRubini, who wrote Linux Device Drivers. He said he'll come and do a presentation about the sdb (self-describing bus) stuff they're doing at CERN20:43
stekernah, cool!20:45
stekernin the meantime, I would sniff around drivers/dma to start with if I'd be you20:45
stekernI will need a driver too20:46
stekernthis nostalgia frenzy we are having really need some sound to it20:47
olofkSo we need an ISA core that we can hook up our old sound blasters to? :)20:49
stekernspeaking of which, someone has already done an OPL2 clone20:49
olofkNice. Is it any good?20:50
stekernI don't know, it's part of a x86 clone20:52
stekernbut *not* zet iirc20:52
stekernI should have made a note about it here when I saw it, now I can't find it...20:55
olofkNow it's saved in the cloud :)20:57
stekernthere it is, a sb 2.0 core ;)20:57
stekern"OPL2 (FM synthesis not fully working)."20:58
stekernmaybe answers your question20:58
olofkHere's my presentation from FPGAWorld if you're interested btw. There are some slides of the board I'm working on adding support for there
stekernhaha, I like that you add quotes, with yourself as the author of the quote ;)21:02
stekernI think qemu would have been worth noting too21:02
olofkah.. true :/21:03
olofkIt worked fine as a base presentation, so it's great with some input on how to improve it for other events21:03
poke53281you can also mention Debian.21:04
poke53281I am only waiting for blueCmd to put that online :)21:05
olofkI talked about Debian, but you're right, it's not written anywhere in the presentation21:07
olofkOh well. Time to sleep now21:07
stekernbut overall, good slides21:10
-!- Netsplit *.net <-> *.split quits: sb0, rjo21:45
-!- Netsplit *.net <-> *.split quits: funfunctor22:27
-!- Netsplit *.net <-> *.split quits: jeremypbennett, arokux222:27
--- Log closed Sun Sep 14 00:00:51 2014

Generated by 2.15.2 by Marius Gedminas - find it at!