<wpwrak> contemplated sending the M1 to the balcony. then the NOR corruption can freeze to death while i have it nice and warm ...
<wpwrak> s/contemplated/contemplates/
<wpwrak> hehe ... gfpus.c Efficiency: 68%, my scheduler 71%. now i just wonder if the code is correct :)
<wpwrak> NOR corruption definitely likes the night.
<wpwrak> yes !! code is functionally identical ;-)
<wpwrak> now let's benchmark the critter ...
<wpwrak> very satisfying :) the original scheduler takes about 31 seconds to compile "Godhead" 10k times. mine does it in 3.63 s (unoptimized)
<wpwrak> (all on the PC ... haven't tried building FN yet)
<kristianpaul> night -> night clubs -> ;-)
<wolfspraul> ok I think for the most part my shop landing page is done https://sharism.cc/shop/product_info.php?products_id=13
<wolfspraul> the one thing still missing is tests with a few more browsers (only chromium so far0
<wolfspraul> and maybe better hover texts for some images
<wolfspraul> and then of course feedback, if I get any :-)
<wolfspraul> next: write a few launch stories, and contact some journalists
<wpwrak> the four buttons would be even cuter if you could click them ;-))
<wpwrak> minor nit: the contrast of the links isn't great. yellow on grey ...
<wpwrak> "Here is how it works:" is a bit isolated. should it refer to the scenery image or also to the technical specs ?
<wpwrak> hmm, "regular updates" sounds like a weekly patchday. it also downplays the possibility of sharing patches without involving the manufactureer
<wpwrak> no "click to enlarge" for any of the images ? e.g., to get a closer look at the M1 or the accessories ?
<wolfspraul> all good feedback, thanks!
<wolfspraul> most likely I'll do it once creloaded is out ;-)
<wpwrak> i see some sparks on the eastern horizon. is someone grinding an axe ? ;-)
<mwalle> wpwrak: what soc release are you using for your corruption tests?
<wpwrak> you mean bitstream ? the july stuff
<mwalle> and the sw version
<wpwrak> it's all the july stuff: http://www.milkymist.org/updates/2011-07-13/
<wpwrak> whee ! another corruption ! today they're fast
<mwalle> wpwrak: diff?
<wpwrak> towards the end, right before the summary
<mwalle> wpwrak: +* ?
<mwalle> strange diff :)
<wpwrak> * = previous line
<wpwrak> hexdump did it :)
<wpwrak> so the 0x90 became a 0x10
<mwalle> so 90->10
<mwalle> yes
<wpwrak> this it with all safeties off - no clean shutdown, no locking
<wpwrak> s/it/is/
<mwalle> wpwrak: the ff corruptions are interesting
<wpwrak> the ff -> 00 ? yes. they're pretty frequent
<mwalle> and 81 4c
<mwalle> at the same offset
<wpwrak> oh yes, this is an unusual one
<wpwrak> ahh, here he is ! :)
<lekernel> hi
<wpwrak> lekernel: question: are code size and register allocations likely to be an issue with the PFPU ?
<wpwrak> or are there still plenty of reserves ?
<lekernel> there are reserves; you still run into register allocation problems at times atm only because the register allocation software is stupid
<lekernel> it preallocates registers for all variables, even unused ones
<wpwrak> also, what happens if one instruction issues from a register that's also in the destination slot of that instruction ? undefined behaviour ?
<lekernel> and if you define more variables, you easily run into problems
<wpwrak> (preallocate) yeah, and then you go and fish out the ones you don't need :)
<lekernel> yeah, I should do that
<wpwrak> you're doing that already
<lekernel> ah
<lekernel> yeah, but on the scheduling stage
<wpwrak> you check if any instruction names one of these registers
<wpwrak> yes
<wpwrak> before you have infinite registers, no ?
<lekernel> yes
<wpwrak> ah, perhaps not for the pre-initialized ones
<lekernel> but the problem is: the first N registers are mapped directly to PFPU registers
<lekernel> the scheduler will always use those exact PFPU registers when it sees numbers <= N
<lekernel> and the current algorithm preallocates all potential per-vertex and per-frame variables into that pool
<lekernel> if user code defines more variables, they will receive the next register numbers
<wpwrak> yup. i see what you mean
<lekernel> if this goes above N, you run into the "register allocation failed" error message
<wpwrak> you ought rename the user-defined ones (or do something similar)
<lekernel> the problem is that they should keep their values between PFPU runs
<wpwrak> ah, and i found a bug in your thesis: table 6.1, destination should be 6..0, not 7..0
<lekernel> if I define them "renamable", the scheduler might reuse the same physical register twice and destroy the first value
<wpwrak> naw, rename before scheduling
<wpwrak> move the "does anyone use this ?" check outside the scheduler
<lekernel> just strip out the unused per-frame and per-vertex variables from the preallocation...
<lekernel> scan the patch once for used variables, and only allocate those
<wpwrak> yup
<wpwrak> next question: which variables are outputs ? not only the ones names in VECTOUT, right ?
<wpwrak> s/names/named/
<lekernel> VECTOUT is only useful for the per-vertex variables
<wolfspraul> lekernel: different question, trying to write an email to Wim I couldn't find the sources for http://milkymist.org/leaflets/brochure_v4.pdf
<wolfspraul> is this already in git?
<lekernel> in fact, the per-frame code is run on the PFPU merely because it can share the same infrastructure, not for performance reasons like the per-vertex one
<wpwrak> how can i tell what the outputs of the per-frame code are ?
<lekernel> VECTOUT only writes two registers to memory with DMA, goes to the next vector, and runs again the PFPU code
<lekernel> wpwrak, using the preallocation
<wolfspraul> I am sort of happy with my shop landing page finally, feedback welcome https://sharism.cc/shop/product_info.php?products_id=13
<lekernel> wolfspraul, yes, saw your email. moment...
<wolfspraul> no rush
<wpwrak> okay. so if a fixed register is written, this always means its an output. not just an intermediate value ?
<wolfspraul> it's just that I want to avoid accidentally forking something with a commit when you have something locally...
<lekernel> wpwrak, the compiler does that yes
<lekernel> and it also uses fixed register for user variables (even if the generic code doesn't read them)
<lekernel> just so that values are kept between PFPU runs (having a fixed register also prevents the scheduler from reusing it multiple times)
<wpwrak> great. i'm trying to find a "tight" definition for what it means for two code sequences to be functionally equivalent. for now, i just look at vectout
<wpwrak> context: i'm writing a faster scheduler. the scheduler seems to be where compilation spends most of its time. i'm not completely finished yet, but it looks promising so far: http://projects.qi-hardware.com/index.php/p/wernermisc/source/tree/master/m1/perf/sched.c
<lekernel> in per-vertex code, the fixed registers also prevent the scheduler from damaging the values of the parameters (midi inputs, ...) that have been written by the renderer software
<wpwrak> with gcc and unoptimized, it's around 10x as fast as the original (tested with "Godhead")
<lekernel> otherwise only the first vertex would use the correct parameter values
<lekernel> wpwrak, cool!
<wpwrak> can data propagate from vertex to vertex ? i saw that push/pop stuff, but haven't looked at it in detail yet
<lekernel> very nice :)
<lekernel> no, data isn't supposed to propagate from vertex to vertex
<lekernel> so you think we can simply make the compiler so fast that no caching of the binary is needed?
<lekernel> that would be very cool :)
<wpwrak> the original scheduler achieves 68%. mine gets 71% witha peak allocation of 35 registers. if i enable a small optimization that increases total run time by about 15%, i get 79% with 26 registers
<wpwrak> that's exactly the idea ;-)
<lekernel> the PFPU is very fast already (not the limiting factor at all), so trading off code quality for runtime makes sense
<wpwrak> there's no gain in making code shorter, right ? (at long as it fits)
<lekernel> 79% efficienty is ~63 MFLOPS
<lekernel> we don't need that much
<lekernel> shorter code is faster
<wolfspraul> alright, I setup some redirects, sharism.cc/milkymist is the landing page
<lekernel> 1 instruction = 1 clock cycle at 80MHz
<lekernel> all the time, except for VECTOUT
<wpwrak> what my optimizer does is that, when given a set of instructions that can be scheduled, it picks the one that has the longest run time to the end of all the things that depend on it
<lekernel> wolfspraul, here it is http://milkymist.org/Brochure2/
<lekernel> made with Photoshop by Openpattern graphics designer Michael Belhatte
<wpwrak> without the optimizer, i simply pick the first that comes along (i.e., in input order) and that has no destination conflict
<lekernel> haven't opened the source files yet
<wolfspraul> ahh :-)
<wolfspraul> lekernel: totally new source files? so not related to the Scribus stuff we have in github?
<lekernel> wpwrak, I was wondering if that "longest latency first" heuristic would have any positive effect :) apparently it does.
<lekernel> we should try scheduling more patches though
<lekernel> and compare
<lekernel> (just for curiosity)
<lekernel> wolfspraul, yes
<wpwrak> lekernel: yup. 8% more efficiency, 25% fewer dynamically allocated registers
<wolfspraul> ok I will not move this over to github/scribus now, it's good that it's published there though
<lekernel> yes and the files are very big and binary
<wpwrak> the massive drop in allocated registers was a welcome surprise. i think my unoptimized version uses more than yours.
<lekernel> :) I haven't tried to optimize that scheduler a lot
<wpwrak> (more patches) i'm just running the whole set
<lekernel> just put together something that works
<wpwrak> correction, past tense. just finished :)
<wpwrak> grmbl. and i messed something up ... let's see why it thinks they're all the same ...
<wpwrak> ah, clever. that was basically just the empty program. it does a lot of stuff, so i didn't even notice ;-)
<wpwrak> "ignoring unknown parameter bMotionVectorsOn" ? hmm ...
<lekernel> ah, yes, MilkDrop legacy
<lekernel> this parameter is redundant with mv_a and unsupported by FN
<lekernel> if it's still in some patches, it's an oversight ...
<wpwrak> that came from  Custom/Philpraxis - Eight bit starfield.fnp
<lekernel> it's quite in a few others...
<lekernel> let me fix that
<GitHub73> [flickernoise] sbourdeauducq pushed 1 new commit to master: http://git.io/aNDtrQ
<GitHub73> [flickernoise/master] patches: remove bMotionVectorsOn - Sebastien Bourdeauducq
<wpwrak> thanks !
<lekernel> thank you for reporting :)
<GitHub8> [flickernoise] sbourdeauducq pushed 1 new commit to stable_1.0: http://git.io/qwUK_g
<GitHub8> [flickernoise/stable_1.0] patches: remove bMotionVectorsOn - Sebastien Bourdeauducq
<GitHub57> [flickernoise] sbourdeauducq pushed 2 new commits to master: http://git.io/TjsIrQ
<GitHub57> [flickernoise/master] ignore video patch when there is no camera connectted - Xiangfu Liu
<GitHub57> [flickernoise/master] performance: skip advanced patches and patches using unavailable video on patch switch - Sebastien Bourdeauducq
<wpwrak> the "computer-based" sounds weird
<lekernel> -> "with a laptop"?
<wpwrak> it sounds more like "we use live-action spray painting, to avoid those evil computer-based solutions"
<lekernel> and replace other "computer" with "laptop" to avoid confusion among nerds
<wpwrak> yeah, laptop. or just "PC". i guess the iFans will still realize that this applies to them as well :)
<wpwrak> yeah. the 2nd "computer" isn't as bad, because it's clear from the context that you mean something additional
<wolfspraul> ok let me share my wisdom about PR
<wolfspraul> our first practical problem is that none of us are English native speakers
<wpwrak> now it comes :)
<wolfspraul> that means our texts must be edited
<wolfspraul> nah
<wolfspraul> just quick
<wolfspraul> my key points
<wolfspraul> #1 - don't forget that we are not native speakers
<wolfspraul> #2 - the key thing is to work with journalists, email them directly
<wolfspraul> a story is not published if the text or 'news' is so good that they have to be published. this is not how it works, at all.
<wolfspraul> facts don't cease to be facts, but news ceases to be news
<wolfspraul> what's interesting in news is often the story behind, the motivation of people, get the secrets out
<wolfspraul> I will do this - I will write several intros/stories/paragraphs
<wolfspraul> and then email them to some journalists, and see what I get back
<wolfspraul> sebastien asked about PR agents in a mail - yes, they do work, definitely
<wolfspraul> but it's very expensive
<lekernel> ereleases is $399
<wolfspraul> you don't want a PR agent to write a short snippet for you, that's a waste of time
<wolfspraul> a PR agent acts like a 'pull station' for journalists
<lekernel> now, maybe you get what you pay for ...
<wolfspraul> so for example, a journalist of a DJ magazine has a "vj special" coming out in the December 2011 edition
<wolfspraul> but where will he get his articles from?
<wolfspraul> he calls/emails some PR agents he knows (!)
<wolfspraul> "hey, I have this VJ special in a few months, do you have something?"
<wolfspraul> the PR agent looks among his clients, and bang
<wolfspraul> that's how stories get into magazines
<wolfspraul> you can try ereleases and similar stuff for a few hundred USD, but I think it's a total waste of money
<wolfspraul> you can achieve the same by taking a few hours to copy/paste your news to the 'submit story' buttons of publications
<wolfspraul> which is what I will do as soon as I have some feedback from some journalists
<wolfspraul> as to which story they think might work
<wpwrak> aha ! i reached my first "if". and indeed, it causes trouble :)
<wolfspraul> if you really want to be represented well, you need to work with ongoing PR agents, that's an arrangement where you pay X thousand USD / month, and then you become one of his/her clients, and whenever journalists 'pull' a story from them your story will be pitched
<wpwrak> well, a little. very little :)
<wolfspraul> for example I can definitely recommend the services of Pat Meier-Johnson :-)
<wolfspraul> but too expensive for us now
<wolfspraul> www.patmeier.com
<lekernel> that's only for the tech press, no?
<wolfspraul> I have little experience outside tech, so until I have more, I assume it works the same there
<wolfspraul> of course there may be other PR agents specializing in music/dj/etc.
<wolfspraul> but a PR agent generally acts like a pull station for future articles
<wolfspraul> you think about it (you being a journalist) - you have these 'themes' coming up, a schedule for the next x months
<wolfspraul> but where do you get the articles from?
<wolfspraul> in the future!
<wolfspraul> it can only work if there is some sort of aggregator that can schedule stuff in the future
<wolfspraul> a PR agent
<wolfspraul> let's say some magazine always has a 'gadget focus' in the January edition
<wolfspraul> or in November, christmas gift ideas
<wolfspraul> etc. etc.
<wolfspraul> how can a journalist know in June which articles he should feature in the November edition?
<wolfspraul> so he calls some PR agencies he knows
<wolfspraul> "do you have anything interesting coming up for the November timeframe?"
<wolfspraul> I need to feature some products that are hot and new *at that time*
<wpwrak> how about viral ? try to place it at an online spot where it has high visibility, then let the news spread itself ?
<wolfspraul> sure, of course
<wolfspraul> we have been extremely bad at getting our rc2 users to open their mouths
<wolfspraul> viral is viral, either it is viral or not :-)
<wolfspraul> seems it is not, so far
<wolfspraul> so we have
<wolfspraul> 1) push news out, 'submit story' buttons, email known journalists to see which story/angle might work
<wolfspraul> 2) register with good PR agent to get into magazines, be 'on the radar' for future editions/features, etc.
<wolfspraul> 3) viral, have your customers be so happy that they take your news further
<wolfspraul> #1 we will be doing, now
<wolfspraul> #2 is too expensive now
<lekernel> let's just try everything until something works
<wolfspraul> #3, well, don't know but so far it didn't happen
<wolfspraul> if you are not sure about a good PR agent, you can also ask journalists
<wpwrak> viral can also come from sites like /., lwn, etc.
<wolfspraul> say in music, I am not sure
<wolfspraul> but I have an intro at this german magazine, so I may ask there
<lekernel> I don't think the journalist situation sucks as bad as you describe it, I already managed to get my articles in a few publications (with limited impact unfortunately)
<wpwrak> and yes, there must be similar places in the music area
<wolfspraul> "what is a good PR agent for the music press"
<wolfspraul> sucks?
<lekernel> slashdot worked great
<wolfspraul> I didn't mean that at all
<lekernel> yes sucks
<wpwrak> what do the vjs say where they go for their daily news fix ?
<lekernel> this PR agent system sucks
<wolfspraul> I'm just describing the mechanics as I've seen them working
<wolfspraul> no the agents are needed for quality control
<lekernel> it's unfair and journalists are supposed to do their job investigating stuff, not just drinking the kool aid from people who tell them whatever story from anyone that gives them money
<wpwrak> /. seems to create a lot of activity but i'm not sure much is left after the noise dies down. but it seems that still a lot of people are reading it. we saw that with the pickup at friendlier places.
<lekernel> so "sucks" is totally appropriate
<wolfspraul> journalists have no time for that
<wolfspraul> they just write write write
<wolfspraul> 10 stories or more per day
<wolfspraul> you would probably advise them all to quit their jobs :-)
<wolfspraul> let them tell you how they are paid :-)
<wolfspraul> by word, by minute (in radio), etc.
<wpwrak> they may make more on wellfare :)
<lekernel> oh, well, maybe the problem goes deeper
<lekernel> but it still sucks
<wolfspraul> you are probably right, without agents the journalists' bosses would have to give them a lot more time :-)
<wolfspraul> but the agents act like quality filters
<wolfspraul> of course it makes it very hard for newcomers
<lekernel> yes, only people with a lot of money can get their voice heard
<lekernel> what a great quality filter :-)
<wolfspraul> yes but once you take money out you really do have a lot of junk
<wolfspraul> it may be an unfair filter, but that's how it works now
<wolfspraul> if you can spend 5000 USD / month (or more) on being represented by a PR agent, you probably are a serious business :-)
<wolfspraul> so anyway, a good agent will get you into magazines, for sure
<wolfspraul> otherwise he/she would loose your business, obviously
<wolfspraul> so yes, it works, and it is very broadly used
<wolfspraul> the agents can also write texts, collect or make high quality pictures or videos, etc. etc.
<wolfspraul> will all cost, of course
<wpwrak> just imagines being served the daily unfiltered rubbish of all the great new products they have just received at dealextreme etc.
<wolfspraul> ha
<wolfspraul> just google for 'pitch story' and you will find more
<wolfspraul> there's a lot of nice links at the bottom of that url
<wolfspraul> need to read a bit more :-)
<wolfspraul> anyway, I dont' want to become a good PR agent, for sure
<wolfspraul> so we do this for a while ourselves, hopefully to get things moving
<wolfspraul> without being able to afford a good PR agent, we will alwyas have a very hard time to get our story out
<wolfspraul> because we compete with the professional PR guys who often act as pull stations
<wolfspraul> as soon as we can afford it, we should look for a suitable PR agent and enlist their help
<wolfspraul> they do hard work, it's not corruption money or wasted money
<wolfspraul> unless you want to criticize the entire work situation of journalists who are just the magazine equivalent of code monkeys, but that's naive and things won't change anyway because of you
<wolfspraul> writing stories, pitching stories, getting them out is an art form in and of itself
<wolfspraul> some people love it
<wolfspraul> it's their life
<wolfspraul> they wake up and have all these ideas for stories
<wolfspraul> every day
<wolfspraul> so either we need to accept that too for a while, and play the game right, and enjoy it, or we just accept we are horribly bad at it and find someone like that to help us right away
<wpwrak> i notice a bit of a psychopathological theme in those patches ... Lonely_Signal_Gone_Mad, Altars_Of_Madness, Interwoven_Nightmare, Dark_Subconscious, DMX_madness, ...
<wolfspraul> dismissing it as 'unfair', or 'we are not being heard' or 'we have no publicity' will achieve nothing
<wolfspraul> that's about all my wisdoms on the subject :-)
<wolfspraul> I will try my luck this coming week, now that my landing page is ok (I think)
<wolfspraul> yeah that's a good text [how to pitch a story], at least I agree with it all
<wolfspraul> helpareporter.com - interesting. didn't know about this before.
<lekernel> I didn't say we shouldn't do it, I just said it sucks and it's unfair
<lekernel> but unpopularity sucks more
<wolfspraul> did you read that how to pitch url?
<wolfspraul> google for 'pitch story' and you find more
<wolfspraul> maybe I will try this helpareporter.com thing
<wolfspraul> looks interesting
<wolfspraul> I guess we can register as experts for video, synthesizers, fpga, free culture, etc? :-) don't know how it works...
<wolfspraul> I guess it tries to automate the PR agent work a little, or at least pull out what can be automated into a site
<wolfspraul> lekernel: when you meet the edirol founder, ask him which PR agent they worked with at Roland, or before they were acquired by Roland
<wolfspraul> get a specific name and contact info. the agents I know are all small businesses (1-2 persons), there shouldn't be much secrecy in this information, if he knows.
<wolfspraul> getting an intro at a good/the right PR agent may be the next big thing we need for Milkymist's success
<wolfspraul> we just need to know which PR agent can do wonders for us :-)
<wolfspraul> who has the ear of music journalists? I don't know...
<wolfspraul> but he may have some leads for you, so please ask
<wolfspraul> also ask him which magazines/publications/journalists where the most effective in getting the Edirol story out
<wpwrak> lekernel: by the way, what does the hardware do if an instruction uses the same register as implicit/explicit operand and as destination ? does something useful happen, or just chaos ?
<wpwrak> i guess he just tried it :)
<mwalle> lol :b
<wpwrak> performance comparison of the schedulers: http://downloads.qi-hardware.com/people/werner/m1/perf/chart-20110918
<wpwrak> the two that don't have Y/Y/Y in the "Equiv" column are a bit worrisome. N/N/N means that original, new and new with optimization all produced different code. Y/N/N means that original and new produced the same but new+optimization didn't.
<wpwrak> the run time is for a loop that compiles the patch 10000 times. running on my Q6600, compiled with -pg without any -O
<roh> wpwrak: nice.. so your optimizations work besides for 2 patches?
<Fallenou> hum hum MacPorts "crossgcc" PortGroup (sort of template for Portfiles) seems not to be compatible with naming binaries "arch-systemVERSION-tool"
<Fallenou> so instead of lm32-rtems4.11-gcc we would get lm32-rtems-gcc
<Fallenou> if we use PortGroup crossgcc instead of a complety written from scratch Portfile
<Fallenou> They don't like my Portfile (the one I submitted) cause they want me to use their new feature "PortGroup" in order to factorize Portfile code
<Fallenou> =(
<Fallenou> I think it would really be better if the Portfile would generate a lm32-rtems4.11-gcc instead of a lm32-rtems-gcc, because every page of the wiki and the automated scripts Xiangfu has written all use "lm32-rtems4.11-gcc" reference
<Fallenou> what do you think ?
<roh> fix the scripts to have "lm32-rtems4.11-" in CROSS
<roh> neither lm32-rtems-gcc or lm32-rtems4.11-gcc should be hardcoded there anyhow
<qi-bot> The Firmware build was successfull, see images here: http://fidelio.qi-hardware.com/~xiangfu/build-milkymist/milkymist-firmware-09182011-1746/
<Fallenou> roh: Sorry I did not understand your workaround
<roh> its not a workaround. the compiler should never be specified explicitely by its complete string. same goes for the other utils (ar, strip, etc)
<Fallenou> How can make know which tool to use then ?
<Fallenou> instead of the usual native gcc
<Fallenou> I mean, in xiangfu script you have CC= AR= etc environment variables set, but they are set to "lm32-rtems4.11-TOOLNAME"
<Fallenou> so I guess it won't work if the system has a binary named "lm32-rtems-gcc"
<Fallenou> and so on
<roh> there should be a CROSS variable which has lm32-rtems4.11- in it
<roh> and the CC then becomes CROSS+"gcc"
<roh> etc
<roh> so you only need to change one variable, depending on what prefix your toolchain has
<Fallenou> oh ok
<Fallenou> But you think we should definitely stick with the "lm32-rtems4.11-TOOL" naming, right ?
<roh> i dont think that one should fix on one version. its irrealistic to make sure everybody uses the same compiler in th end
<Fallenou> Maybe, but we should try to stay coherent, when it's possible
<Fallenou> so if we have scripts with 4.11 in the binary names
<Fallenou> and wiki with 4.11
<roh> then thats stupid.
<Fallenou> it will be weird to provide a lm32-rtems-gcc Portfile to Macports which generates "lm32-rtems-gcc"
<roh> who knows how long 4.11 is recent?
<roh> so it needs to change anyhow
<Fallenou> and to advertise this MacPort Portfile in our wiki
<Fallenou> Ok I think I will drop an e-mail on the mailing list about this
<Fallenou> So that people can give their opinion on this
<Fallenou> but we may end up removing "4.11", to stay generic and prepare for next rtems release
<Fallenou> Thanks
<roh> yes. and fix some scripts which dont use CROSS or something similar  at all but hardcode stuff like LD=lm32-rtems4.11-ld \
<wpwrak> roh: (optimization) yup, it almost always produces better code as well,.. and it's always ~10 times faster than the original. (at least on the PC and without -O)
<wpwrak> ((hard-coded toolchain) e.g., milkymist/software/include.mak
<wpwrak> but its good to see that it's using gcc. i thought there was llvm somewhere in there
<kristianpaul> (llvm) just because the llvm prot for lm32 is not totally ready yet
<wpwrak> hmm, i don't mind at all not having to keep yet another kind of toolchain around. something like five flavour of gcc is already plenty :)
<wpwrak> hmm. valgrind doesn't see anything suspicious in the code the produces diverging results. complains a lot about fvm_dump, though.
<mwalle> imho all scripts should use a CROSS_COMPILE prefix which may default to lm32-rtems4.11- but you should be able to override it via an environment variable
<wpwrak> that sounds reasonable, yes
<wpwrak> of CROSS_COMPILE or CROSS
<Fallenou> but you need to choose wether you want to have a binary named lm32-rtems4.11-gcc or lm32-rtems-gcc
<wpwrak> or both :)
<wpwrak> version numbers in tool names are relatively uncommon
<kristianpaul> at least rtems seriouslly want to fork gcc .. ?
<wpwrak> hmm yes, i'm sure sebastien has wet dreams about maintaining his own gcc fork. maybe with autotools too, just for the sheer pleasure of it ;-)
<mwalle> Fallenou: so? if you are required to use one without version, do it, run the script with CROSS_COMPILE=lm32-rtems-gcc- ./script
<Fallenou> wpwrak: ahah he surely wants to get rid of autotools
<mwalle> Fallenou: btw does your macport receipt (dunno whats the name for it :)) support plain lm32-elf- too?
<Fallenou> the macports Portfile I've written is just compiling gcc 4.5.2 with newlib and applying rtems patches
<Fallenou> it's doing the same thing as written on the wiki
<Fallenou> but I could do a lm32-elf- Portfile too :) and generate .dmg files for it
<mwalle> Fallenou: would be cool, other users might write bare metal applications for lm32
<kristianpaul> likes mwalle idea
<Fallenou> mwalle: well yes good idea, I just wanted to allow compiling of flickernoise
<Fallenou> but yes, why not :)
<Fallenou> but first things first
<Fallenou> I still need to make this Portfile accepted
<Fallenou> either using PortGroup or not
<Fallenou> in the mean time, there is still the .dmg files hosted on my website to give instant toolchain to mac users :)
<Fallenou> please if you have a Mac, try the .dmg files and give me feed backs if it works for you
<Fallenou> but it should !
<mwalle> is it still ppc+x86?
<Fallenou> since I only compiled it for x86
<Fallenou> I think the .dmg is only x86
<Fallenou> it's not "universal"
<Fallenou> but ppc macs are really declining
<mwalle> hey :b
<mwalle> still have one
<Fallenou> I wonder how hard it would be for me, with my Intel mac, to generate universal binary (ppc+x86) :)
<Fallenou> I will ask on #macports ;)
<mwalle> well my father inherited it, now
<mwalle> Fallenou: (declining userbase) yes but its sad that not even mozilla compiles their software for ppc anymore..
<Fallenou> oh :(
<Fallenou> I guess they have limited resources
<Fallenou> and limited people committed for these kind of "now exotic" platforms
<Fallenou> Maybe they would just be really happy with someone saying "I will compile mozilla for ppc and release it"
<Fallenou> dunno if it's their policy or just a lack of resources
<Fallenou> it's true that nowadays people are compiling for x86, amd_64, arm and that's it
<roh> Fallenou: i think ppc isnt needed anymore. people who use osx have x86 based hw nowadays (osx is pain and unsupported on ppc)
<Fallenou> roh: well mac OS X can works on PPC, up to 10.5
<Fallenou> work*
<Fallenou> but yes it's old computers
<Fallenou> no more mac books are built with ppc
<Fallenou> it's all x86
<Fallenou> but there are still out there people with old mac books  :)
<Fallenou> good night !
<kristianpaul> n8
<wpwrak> Fallenou: can't they run linux ?
<roh> wpwrak: sure. and its much more fun than a slow and fat osx on hw too slow for fancy ui
<Fallenou> ahah
<Fallenou> I won't enter this troll :)
<Fallenou> I think those who buy a Mac don't wanna install a Linux on it, they want the OS that goes with their hardware
<Fallenou> and I can't blame them for this
<Fallenou> If they wanted Linux, they would have bought a PC, it's less expensive and much more supported
<Fallenou> time to go to bed :p
<Fallenou> see you :)
<wpwrak> pity. so they'll remain at the stage of cavemen and never learn what a good text-based user interface is like :)
<Fallenou> I guess they will pretty soon (if not already done) buy a new PC or a new Mac book ;)
<wpwrak> yeah, i would think so. it you follow that line of systems, you pretty much have to upgrade relatively often
<wpwrak> linux is a little friendlier when it comes to that. but of course, there's pressure too. e.g., my first linux pc (~20 MHz 386DX) compiled the kernel pretty quickly with gcc 1.something. things have changed a bit since then, and while that box - if it still existed - would probably still be able to boot contemporary linux, it would have a hard time doing much else :)
<Fallenou> I guess you had something like 8 MB of RAM
<Fallenou> hard to make linux boot with such a small RAM
<Fallenou> I added some RAM to my 486 DX2, it had 24 MB ;)
<Fallenou> not here anymore *zZzZ*
<wpwrak> at the beginning, i had 4 MB. then i decided to go for luxury, have X, and got myself another 4 MB :)
<kristianpaul> ram, i guess that a mainly wall for selfhosting :/
<wpwrak> hmm ?
<kristianpaul> i mean, you cant compile linux/gcc in your 12Mb router, but of course it can be cross-compiled
<kristianpaul> and still run :)
<kristianpaul> wpwrak: is my comment valid/understandable?
<kristianpaul> for example imagine you want to partially reconfigure a mm soc for a specific task, the at least you need to make run llhdl on it, and i wonder if not re-compile some apps also