#milkymist IRC log for Tuesday, 2012-08-28

stekernthis paper came across my path, thought it could perhaps be of interest to someone here: http://www.ece.ubc.ca/~lemieux/publications/brant-fccm2012.pdf05:31
wpwrakquite radical -)05:56
larscI think I proposed something similar a few months ago. good to know that somebody is actually implementing it07:34
wolfspraul40 times penalty?07:49
wolfspraulI think an fpga emulation in a high-end icore could compete with that, no?07:49
wolfspraullm32 would then run at 2mhz, not 80mhz? :-)07:50
wolfsprauldon't tell that to werner07:50
wolfspraulwell ok, it wouldn't fit anyway07:50
wpwrakhey, still faster than an AVR with factory settings ;-)07:50
wpwraki just skimmed the paper and didn't see what exactly the 40 times penalty refers to. oddly enough, they seemed to be happy about that factor. so i wonder what exactly it means.07:52
wolfspraulthey are happy because it's better than the 100* in other attempts, they say07:52
wolfsprauleverything is relative, no?07:52
wolfspraulI'm happy without any reason, now how is that???07:53
wpwrakah, have to put wine on my shopping list, too :)07:55
larscmaybe I'm blind, but I'm unable to find their implementation. All references to "ZUMA FPGA" just point to the paper.09:37
larscwpwrak: btw. it is not necessarily slower, just smaller09:40
wolfspraulyeah who knows :-) I found too many dead ends to still care09:46
wolfspraula lot of papers and text that is written as if they know *some* stuff and then decided "we have to report our results"09:47
wolfspraulrecobus is the most advanced I think, but sadly they also just publish 5-10 page papers for conferences that are dead-ends for learning, in my experience09:47
mumptai_fpga on fpga?09:47
wolfspraulwhy not, you never know what becomes possible later, but the #1 thing imho is what lars said - where are their sources and does it actually run?09:50
wolfspraultoo much talk... back to coding :-)09:50
larscmumptai_: yes09:51
mumptai_if i was developing a new fpga+toolchain, i would probably like to do that, but otherwise?09:52
larscwell with the big vendor fpga you have the problem that you don't really know what's inside the fpga09:56
mumptai_larsc, and they dont' tell that without an nda, or not at all11:09
larscso an open fpga is a way better target for developing tools11:10
mumptai_but wrapping it again is only a first step11:10
lekernelaha! I got the bios to compile with gcc 4.8/svn and Nick's patches11:10
mumptai_i would actually like to have a model of the fpga for other reasons, but the vendors seem to dislike that11:11
lekernelld later prints mysterious/fucked error messages "main.c:(.text+0xc): warning: internal error: dangerous error" though11:11
lekernelbut still generates a binary...11:12
lekernelI see gcc copied clang's error/warning reporting :) hehehe11:14
larsclast time I tried to build openwrt with gcc 4.7 it just hogged all memory and cpu, hopefully it works again with those patches11:16
larscopenwrt for lm32 that is11:16
lekernelthere's one patch that fixes a bug that resulted in a infinite recursion loop11:17
lekernelis that what you are talking about?11:17
larscmost likely11:18
larscas i said gcc just started to eat all of my cpu and memory11:19
lekernelnow there's still this weird linker problem...11:20
wpwrak(virtual FPGA a "better target") of course, with a factor of 40, that gives you a lot of headroom for making a rather conservative model of the real thing. kinda what wolfgang is doing.13:15
wpwraka car analogy: since we can't be sure all curves are safe to drive in at 240 km/h, we a) post speed limits of 30 km/h (that could be wolfgang's result) or b) don't allow cars that can go faster than 6 km/h (the 40x margin)13:17
wpwrakwhile a) would suck, it would still be 5x better than b) ...13:17
--- Wed Aug 29 201200:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!