#avr Logs

Oct 16 2017

12:02 AM day__ is now known as daey
05:13 AM thardin: what serial terminal do you all use on linux?
05:13 AM thardin: I've been giving minicom a go, but it craps out when it sees binary data
05:18 AM polprog: picocom
05:18 AM polprog: minicom for me craps out in menus
05:26 AM thardin: will give it a try
05:27 AM thardin: best would be if I could have avrdude and the serial terminal program play nice
05:27 AM thardin: currently I have a raspberry-pi entirely dedicated to serial
05:31 AM gnom: cutecom
05:31 AM gnom: http://cutecom.sourceforge.net/
05:36 AM LeoNerd: I run socat
05:37 AM thardin: being able to dump raw data to file would be nice. preferably unmolested
05:37 AM thardin: minicom changes line endings
05:37 AM LeoNerd: So, socat then
05:41 AM thardin: mayhaps. picocom plays nice enough with serial programming. cutecom is graphical so nearly useless
08:00 AM Pi3rrot: hi -all
08:02 AM Pi3rrot: is anyone have an issue with GGC7 to compile any AVR software?
08:03 AM Pi3rrot: like this : "/usr/include/bits/floatn.h:73:9: error: unknown type name '__float128'"
08:03 AM Pi3rrot: I see __float128 is only specified in new versions of GCC
08:04 AM Pi3rrot: should I continue to use an older version like GCC5.4 or is it possible to disable with a compiler option __float128?
08:54 AM thardin: maybe you need -gnu99 ?
08:54 AM thardin: err -std=gnu99
08:54 AM thardin: also did you check the manual? :]
10:57 AM Pi3rrot: thardin: sure, -std=gnu99 is by default in the compiler option
11:43 AM Cracki_: avrgcc has no float128
11:43 AM Cracki_: what is that header file?
11:44 AM Cracki_: (as far as I know)
11:45 AM aczid: Pi3rrot: maybe you can try using a: typedef __attribute__((vector_size(64)) unsigned int __float128; ?
11:45 AM aczid: err wait, not an int :D
11:46 AM aczid: this would work for integer types, I'm not sure if can work on floats too actually
11:46 AM aczid: doing long floating point math on AVR doesn't seem ideal anyway
11:49 AM polprog: so you coud in theory define a 16 byte integer?
11:50 AM aczid: yes
11:50 AM polprog: cool
11:50 AM aczid: actually, __float128 sounds like it should come from intrinsics.h Pi3rrot - maybe that's what you're missing?
11:51 AM aczid: polprog: the argument to vector_size is actually in bytes, so this is 512-bits
11:51 AM aczid: as you can see, for integers this all makes sense because all the vectorized cases are the same. it's easy for the compiler
11:51 AM aczid: floats however, I think would be different for each size right?
11:53 AM Cracki_: right.
11:53 AM Cracki_: an 128 bit float is very unusual
11:53 AM Cracki_: I'm not even sure avr has 64 bit floats
11:55 AM Cracki_: "having" is a matter of compiler support
11:55 AM aczid: does ISO even have them? :D
11:55 AM Cracki_: I'm sure iso defines them
11:55 AM Cracki_: https://en.wikipedia.org/wiki/Quadruple-precision_floating-point_format
11:55 AM aczid: aight
11:56 AM Cracki_: that header makes no sense for avr
11:56 AM aczid: wait, that makes sense, because that's also inside my intel cpu :)
11:56 AM Cracki_: avr has its own headers
11:56 AM aczid: ISO has defined up to octuple sizes according to wikipedia
11:56 AM Cracki_: your build setup must be broken
11:56 AM aczid: so up to 256 bit floats
11:57 AM aczid: wanting a 512-bit float might be a sign you're on the wrong track design-wise :P
11:57 AM Cracki_: perhaps not
11:57 AM aczid: might tho
11:57 AM aczid: on avr, more likely than on x86
11:58 AM Cracki_: what I want on a desktop CPU is half floats and very much wider vector instructions
11:58 AM aczid: it also doesn't exist on GCC?
11:58 AM polprog: that's very interesting
11:59 AM Cracki_: avr has no hardware floats, so avrgcc emulates those.
11:59 AM polprog: wonder how many cycles does it take to add two 64-byte ints :D
11:59 AM Cracki_: gcc for amd64 and such uses hardware capabilities, does no emulation
11:59 AM polprog: considering each one is twice the capacity of all registers
11:59 AM Cracki_: add, add with carry, ... done
11:59 AM Cracki_: oh right
12:00 PM Cracki_: lots of loads n stores
04:43 PM grog_l is now known as grog
10:27 PM Tom_itx is now known as Tom_L