Page 2 of 4

Re: Official Release 20120903 Open Source Compiler

Posted: Tue Sep 25, 2012 5:38 am
by dangeljs
Brian,

I won't be able to try it out until Wednesday, but I'll definitely let you know what I find out then.

Thanks,

Jason

Re: Official Release 20120903 Open Source Compiler

Posted: Tue Sep 25, 2012 1:26 pm
by EmbeddedMan
Note that if the board you are using does NOT have those lines in the boards.txt file, then it is already using ::-fno-short-double and my suggestion won't work for you.

*Brian

Re: Official Release 20120903 Open Source Compiler

Posted: Wed Sep 26, 2012 6:35 pm
by dangeljs
Brian,
My file did not have the command appended at the end of each line, but as it turns out I am not having an issue after all. I had imported an Arduino library (PID) that required it's inputs to be doubles, so the variable I was working on was declared as a double not a float. I changed the PID library to use floats and also converted my variable to a float and I no longer have an issue with the angle returned from the ATAN function.
A double in Chipkit is still 32bits correct? If so is there ever a need to declare anything as a double?

Thanks,

Jason

Re: Official Release 20120903 Open Source Compiler

Posted: Sat Oct 06, 2012 5:22 am
by artbecker
Sorry, I posted this under the wrong topic yesterday.

20120903 fails on Ubuntu Linux 12.04. It works ok under Windows 7 and Mac OSX 10.8.2

A driver loads OK in linux (ttyACM0), but on starting MPIDE, I immediately get asked which directory to store my sketches in, and then, after either making a selection, or creating a new folder, the program just disappears.

Arduino IDE 1.0.1 works OK on Ubuntu 12.04

When running MPIDE in terminal mode, in Ubuntu, I see some java errors, such as "no rxtxSerial in java.library.path"

I am current with all linux updates for Ubuntu 12.04, though sometimes things like Java have to be downloaded externally, to get the latest version, rather than through software update.

=====

In Windows 7, when starting the 20120903 MPIDE, there is a temporary java file created, such as jna1515996498641544886.tmp, that triggers an alert from Microsoft Security Essentials (MSE), as being an unknown file, and thus potentially dangerous. Hoping for the best, I told MSE to always ignore files of the type jna*.tmp. I've never seen this happen before with MSE.

.

Re: Official Release 20120903 Open Source Compiler

Posted: Sat Oct 06, 2012 4:34 pm
by mikes
I don't have linux but on windows if C:\Users\YourName\AppData\Roaming\mpide\preferences.txt contains bad information it causes a silent failure. The cure is simply to delete the file and mpide recreates it with the default values. You might try finding the linux equivalent of that File.

Re: Official Release 20120903 Open Source Compiler

Posted: Wed Oct 10, 2012 10:18 pm
by ricklon
The upcoming test build has a working configuration for 64 bit. No 64 bit compiler, but will build properly.

Re: Official Release 20120903 Open Source Compiler

Posted: Thu Oct 11, 2012 12:07 pm
by JTRelectronics
This on the face of it looks like it could be a really good option for a PIC32 C compiler. Congrats to all those who have worked very hard on putting this together.

(I wonder if I can port my USB stack to it...)

Now I have a quick question. Does this support compiling in MIPS16 (and without time restriction.)

I do not need library support so that does not matter but rather I need to see the C calling conventions (I.E. reduced register set) to MIPS16 code and I am looking for a C compiler to reverse engineer this given I cannot find any microchip documentation on this.

Any heads-up appreciated.

Re: Official Release 20120903 Open Source Compiler

Posted: Thu Oct 11, 2012 1:00 pm
by EmbeddedMan
Yup. It does support MIPS16. It's really just C32, so any documentation you find for C32 will apply to the MPIDE compiler as well.

*Brian

Re: Official Release 20120903 Open Source Compiler

Posted: Thu Oct 11, 2012 1:55 pm
by JTRelectronics
Thanks for that Brain. Good to know for my future plans.

Re: Official Release 20120903 Open Source Compiler

Posted: Thu Oct 11, 2012 11:08 pm
by cmspider
I've noticed floating-point function problems with this release too. With an uno32, the floating point functions don't seem to get replaced with the single precision versions of them properly, unless the -f-no-short-double option is added as suggested above. I saw this with fabs, floor, and log. If I explicitly make them fabsf, floorf and logf then they are ok, but if I leave them without the final f (and compile without no-short-double), then seem to return 0.