IRC logs for #openrisc Tuesday, 2014-05-06

--- Log opened Tue May 06 00:00:36 2014
skip__stekern: thanks for the help! :), I was off by one somewhere in the MMU, sim essentially boots now
stekernskip__: congrats!04:44
stekernbut why are you using an ancient Linux version?04:44
stekernthere are probably loads of bugs that are fixed in that04:44
stekernloads of bugs in that, which are fixed in later versions04:52
stekernolofk: I updated the systemc pull-request, was there any kind of notification about that?06:04
stekernjust want to know if such a work-flow could work06:05
stekernstill not completely happy how that works in githubs pull-requests06:05
stekernbut if at least some notification is sent when you update it like I just did, I guess it's something I can live with06:06
stekernXilinx website is as solid as their tools...06:39
stekernI went to download ISE, and it started to download vivado06:40
stekerni.e., I went to this page:
stekernand pressed "Download ISE WebPACK software for Windows and Linux."06:41
stekerneven with sw tlb refill, we're not that slow07:14
stekernat least not in terms of clock cycles ;)07:15
olofkstekern: #46 fot FuseSoc, right? No notice for the changes07:21
stekernyes #4607:24
stekernI even changed the "comment" message, why isn't such changes notified?07:24
stekernobviously you need to add new comments... and it's annoying that the old comments disappears when you force-push the fixed branch07:25
olofkahh..the Mill CPU08:27
olofkI haven't gotten my head around that thing yet08:27
olofkIs it any good?08:27
olofkThat link looked like a very readable introduction though. I should probably start there08:28
LoneTechit's fairly interesting.. whether it's good will be very program dependent08:31
LoneTechthey've reduced the penalties of a very wide issue VLIW machine a bit, but not eliminated them08:31
stekernI haven't read up on it yet, a guy at work came up and spoke warmly about it.08:54
stekernmostly as a contrast to the "boring RISC cpus"08:54
LoneTechthe memory design with automatically blanked cache line allocation and deferred load makes sense08:54
stekernhe's right of course =)08:54
stekernyeah, he explained the deferred load, I didn't get it 100% though. Need to read it to get it I think08:55
LoneTechit has to do with the absence of a register file as such. you do content addressing of the pipelines instead, largely08:56
LoneTechwith the deferred loads, you ask for the result to appear at a specific time in the instruction stream08:56
LoneTechthere's also a bunch of metadata tracking allowing things like not causing an exception on broken values (for instance, bad loads) if they end up unused08:57
stekernah, ok. so, basically *if* it were a traditional risc machine with register file it'd be like 1: load r1,addr; 2: add r2, r3; 3: add r3, r4; 4: add r4,r109:01
stekernbut instead, you say 'load @4'09:01
stekernwell, 'load @4, addr', perhaps09:02
stekernkinda like load-delay-slots on steroids09:03
stekernor am I completely off track?09:03
LoneTechyou're not completely off track09:08
LoneTechlooks like a pretty good description actually09:09
LoneTechthe basic effect is that the compiler is allowed to hoist and speculate fetches, so the processor does not need to predict them09:11
stekernand the processor doesn't need to keep track of register usage if/when it deferres loads internally09:19
LoneTechwell, the load takes a pipeline slot in the load unit, effectively. the address it appears on is stored there, and it doesn't appear until the marked delay - if the load isn't finished yet it may stall09:20
LoneTechmuch of the mill design is removing large logic blocks that served to reduce the large processor design to an obsolete ISA model09:22
stekernmy point was just that, deferring loads as such isn't anything spectalurly innovative, getting rid of the downsides is09:22
LoneTechthe "belt" uses instruction position in the execution stream to decide on value names, which removes the destination operand at the expense of old values dropping off09:28
LoneTech(turns out there aren't that many you want kept a long time - mostly the context pointers, which might be special cases, I don't recall right now)09:30
stekernyeah, it's definitely an interesting approach09:39
LoneTechthe generational naming also allows it to hide values from other subroutines on calls09:42
LoneTecha bit sparcish, but somewhat more precise09:43
stekernback to more down-to-earth problems... how the h*ll do I obtain a Vivado WebPACK License?09:55
LoneTechoh, did they make it more broken again? I'll never understand why these companies keep throwing so much money at flexlm to break their software09:58
stekernI don't know if this is flexlm related, but when I click the "get Vivado webpack licenses" in the license manager I get to a webpage that gives me this message:
LoneTechoh yes, that's most definitely flexlm breaking it for you10:04
LoneTechit's also extra helpful in guessing instead of diagnosing the problem10:04
stekernwhat a POS10:13
LoneTechthere's a possibility you have a license that could be found in an account page somewhere10:20
stekernbut isn't that for ISE?10:22
LoneTechcould be. it's just one of the guesses they provide you with is "we're not giving you a license because we already gave you one", and if so, that one might be stored elsewhere also10:23
LoneTech(of course the sensible thing to do would be handing it out again in the first place, but there's a serious divide between LM/DRM and sensible)10:24
stekernI have a bunch of .lic licenses, but this is the new "fancy" activation license stuff...10:26
LoneTechit's just harder to use. same fundamental crap10:27
stekernhah, yeah... from the vivado manual:
stekernI guess my harddrives can't be trusted or something then...10:30
LoneTechyep, you're the customer, ergo the enemy. isn't it a wonderful attitude?10:31
LoneTechnever mind that this "trusted area" differs from the file only in obscurity and fragility10:32
stekernluckily enough, non-paying customer...10:32
stekernerr, non-paying enemy, I mean10:33
stekernI don't get why they need to have any licenses for the web-pack stuff anyway...10:33
stekerncan't it just work to a certain level without a license..?10:34
LoneTechthey don't. it's pure BS. also, earlier versions didn't10:34
stekernwell, of course they don't *need* to. I mean, why do they *insist* on it ;)10:35
LoneTechthey've been flip-flopping on it. they removed it once, because it was troublesome, now they reintroduced it because somebody sold them on it10:35
stekernmaybe it's just part of an assimilation process, so you'll be accustomed to it when you buy the product10:36
LoneTechthey don't want you buying the product; they want you to petition them for the boon of renting partial access.10:37
stekernlooks like I'm attending Assembly this year10:41
stekern...collecting empty bottles10:42
stekernto raise money for my sons football team10:45
LoneTechgood luck10:46
stekernfor future references, running the vivado license manager as root "fixes" the issue11:33
blueCmdstekern: Assembly is just the poor mans Dreamhack :)12:17
stekernAssembly was bigger in the golden era of demo parties though12:29
LoneTechstekern: good to know. I am guessing the "trusted" region requires a privilege escalation (which does not give me a reason to trust it)12:33
stekernLoneTech: yes, maybe... the funny thing is that I could "install the .xml into the trusted area" as a regular user12:34
stekerncould be something silly that it couldn't read some info about the machine as a regular user to create the request too12:35
LoneTechor even sillier, like not finding the tool to do so in your path12:35
LoneTech(for instance, /sbin/ifconfig is runnable as user, you just can't change things)12:36
stekernyes, I was thinking about something in line with that too12:36
LoneTechsometimes the people writing installers and similar toss in checks for root for no reason at all12:37
stekernbut since it doesn't emit any messages at all about what's wrong, I guess we'll never know12:37
LoneTechyou could start over and strace it, but it's probably not worth the effort12:37
stekernheh, yeah... no, I think I pass12:38
--- Log closed Wed May 07 00:00:37 2014

Generated by 2.15.2 by Marius Gedminas - find it at!