IRC logs for #openrisc Saturday, 2015-04-25

--- Log opened Sat Apr 25 00:00:21 2015
-!- Netsplit *.net <-> *.split quits: hno`, Amadiro06:58
-!- Netsplit over, joins: Amadiro07:00
bandvigGeneralStupid: the prebuild newlib-based toolchains could be found here:
GeneralStupidbandvig: ok, thanks but still i cant build14:04
GeneralStupidthe problem is, he cannot locate the libs14:07
GeneralStupidor1k-elf-gcc: /lib/x86_64-linux-gnu/ version `GLIBC_2.14' not found (required by or1k-elf-gcc)14:09
bandvighmm, I'm working under cygwin, I'm afraid I couldn't help you14:10
GeneralStupidok , no problem :)14:16
GeneralStupidmaybe its easier to run linux  than baremetal?14:17
bandvigmy experience answer: not. I spent much more time to build UClibc toolchain and compiling and running linux.14:19
GeneralStupidits hard - but i had some colleagues who tried an openrisc core earlier and they stopped it. So there must be a lot of work already done (fusesoc and stuf...)14:22
bandvigIt is true. Personally I spent a lot of time (~about 1-2 monthes at all) with learning tutorials and doing experiments to get working SoC (on Atlys board) and two operable toolchains. Plus, as I mentioned, I'm a Windows guy. So, for me there were additional difficults to get experience with free source development flow.14:34
GeneralStupidim more that software guy... and i work in a properitary world :-D14:43
bandvigMy experience tell me that in free source world anybody sould be ready to find solutions himself using actual source code, obsolete tutorials (mans) and deep diving in Inernet forums (mailing lists and so on).14:53
GeneralStupidits hard that way, that would mean you should actually have some knowledge about the system15:12
bandvigyes. For example, yesterday I asked how to configure NewLIB variants for building or1k-elf-* toolchain. Actually I asked the question second time. So, it looks like nobody remember how to do it. Well. I performing not deep diving into gcc source tree and found answer in several hours.15:33
bandvigToday morning I re-built NewLIB with support floating point arithmetic for trigonometric and other functions. Just now I'm going to recompile Whetstone benchmark to see improvement in functioncomputation.15:34
GeneralStupidthere are so many tools which are really PITA to work with15:40
GeneralStupidBut in my experience its more pita with closed source software15:44
bandvigIn fact, I don't know "gcc system" even briefly, I just try to find example and do what I need in the same way. Perhaps the such approach is only siutable (espetially for initial steps) as you are right: "there are too many tools..."15:45
GeneralStupidif you dont want to port it for your cpu i think thats all you need to know.15:52
GeneralStupidIMHO thats all just tools for solving real world problems15:53
bandvigI agree. If I want to port gcc/llvm I have to learn them in much more details "using actual source code, obsolete tutorials (mans) and deep diving in Inernet forums (mailing lists and so on)" ;)16:01
-!- Netsplit *.net <-> *.split quits: antgreen, stekern, NCommander16:23
bandvigwell, I've recompiled Whetstone benchmark with a NewLIB variant which were built with usage fp-arithmetic (+,-,/,*) for compute various functions (sinf,cosf,etc).17:00
bandvigAdditionally, NewLIB were built with hardware divivision support and cappuccino pipe configured with pipelined multiplier.  Let me 12 lines to post result table.17:01
bandvigLoop content                soft-float   OR1200       mor1kx      mor1kx17:01
bandvig                                           FPU        PFPU32       "A"17:02
bandvigN1 floating point (MFLOPS)    0.409       3.200       9.600       7.85517:02
bandvigN2 floating point (MFLOPS)    0.336       3.360       6.720       6.72017:02
bandvigN3 if then else     (MOPS)    0.000       0.000       0.000      46.57517:02
bandvigN4 fixed point      (MOPS)    2.250      31.500      31.500      28.35017:02
bandvigN5 sin,cos etc.     (MOPS)    0.019       0.020       0.020       0.32817:02
bandvigN6 floating point (MFLOPS)    0.409       2.075       7.706       7.19217:02
bandvigN7 assignments      (MOPS)    0.000       0.000       0.000      25.58817:03
bandvigN8 exp,sqrt etc.    (MOPS)    0.009       0.009       0.009       0.20217:03
bandvigMWIPS                         0.954       1.128       1.156      17.99317:03
bandvigLast column reflect the latest results17:04
bandvigExpress analysis: negatives: degradation in N117:04
bandvigExpress analysis: stranges: N3 and N7 are non-zeros (was someting wrong with time measurement in previous tests?)17:06
bandvigExpress analysis: positives: N5 and N8 (function computation) are ~15...20 times faster than for soft-float.17:07
--- Log closed Sun Apr 26 00:00:23 2015

Generated by 2.15.2 by Marius Gedminas - find it at!