| Author |
Message |
|
|
 Posted: Dec 15, 2010 - 04:53 PM |
|
Joined: Dec 13, 2010
Posts: 35
|
|
When I try to compile a trivial floating point routine with -mhard-float, gcc crashes with an "internal compiler error". The file works fine without -mhard-float. I have reproduced this problem on both Ubuntu 10.04 amd64 and Windows XP SP3 x86. Does avr32-gcc's floating point support work at all? I opened a trouble ticket at support.atmel.no, but it has received no attention.
somers@tryptophan:~/ctest/avr32$ avr32-gcc --version
avr32-gcc (-(Linux-i386)) 4.3.2
somers@tryptophan:~/ctest/avr32$ avr32-gcc -mhard-float -c -o myfadd.o myfadd.c
myfadd.c: In function 'myfadd':
myfadd.c:3: error: insn does not satisfy its constraints:
(insn 34 17 36 myfadd.c:2 (set (reg:SF 16 f15)
(reg:SF 5 r10 [48])) 11 {*movsf_internal} (nil))
myfadd.c:3: internal compiler error: in note_invalid_constants, at config/avr32/avr32.c:6411
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://www.atmel.com/avr32> for instructions.
$ cat ~/ctest/avr32/myfadd.c
float myfadd(float x, float y){
return x + y;
} |
|
|
| |
|
|
|
|
|
|
Posted: Dec 15, 2010 - 06:26 PM |
|
Joined: Jan 07, 2003
Posts: 4580
Location: Oslo, Norway
|
|
| The AVR32 port of GCC does not support hardware floats. There is even only one AVR32 cpu family that has a hardware FPU, and that is UC3C. |
|
|
| |
|
|
|
|
|
|
Posted: Dec 15, 2010 - 09:05 PM |
|
Joined: Dec 13, 2010
Posts: 35
|
|
If it does not support hardware floats, it should abort with a suitable message when you pass -mhard-float. It should never crash; that is a bug.
Google found this document for me: http://distribute.atmel.no/tools/avr32/ ... readme.pdf . It claims that the AVR32 gnu toolchain v3.1.0.166 supports hard floats. However, I can't figure out how to download it. Version 2.4.2 is the only version that I can find on your website. Could you tell me where to download it?
Hardware floating point is a seriously exciting feature in the UC3C: it's biggest selling point in my opinion. Renesas RX is the only other chip that can compete. I'm very excited to use it and I hope that this project works out. |
|
|
| |
|
|
|
|
|
|
Posted: Dec 15, 2010 - 09:13 PM |
|
Joined: Jan 07, 2003
Posts: 4580
Location: Oslo, Norway
|
|
|
Quote:
Version 2.4.2 is the only version that I can find on your website. Could you tell me where to download it?
Grab the latest AVR32 Studio from http://www.atmel.no/beta_ware/ it comes with the toolchain as a plugin.
Quote:
Hardware floating point is a seriously exciting feature in the UC3C: it's biggest selling point in my opinion. Renesas RX is the only other chip that can compete. I'm very excited to use it and I hope that this project works out.
If the toolchain in AVR32 Studio 2.7 doesn't work for you, it might be an option to test out IAR just to evaluate. |
|
|
| |
|
|
|
|
|
|
Posted: Dec 15, 2010 - 10:59 PM |
|
Joined: Dec 13, 2010
Posts: 35
|
|
SOLVED. As hce says, to use the FPU with avr32-gcc, you need the beta release of AVR32 Studio (sorry, no standalone toolchain right now) from http://www.atmel.no/beta_ware/ . Note also that avr32-gcc will silently revert to soft floats unless you specific an FPU-equipped part with --mpart. Example:
avr32-gcc -mhard-float -mpart=uc3c264c -c -o float.o float.c |
|
|
| |
|
|
|
|
|
|
Posted: Dec 16, 2010 - 06:20 AM |
|
Joined: Jan 07, 2003
Posts: 4580
Location: Oslo, Norway
|
|
| You can compile more generic as well, by setting -march=ucr3fp. |
|
|
| |
|
|
|
|
|
|
Posted: Feb 10, 2011 - 03:53 PM |
|
Joined: Aug 07, 2007
Posts: 30
|
|
Here is the toolchain that I have when installing AVR32Studio 2.7:
avr32-gcc (AVR_Toolchain_3.0_124) 4.3.3
Copyright (C) 2008 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
The option that you indicate -march=ucr3fp does not work. Only the previous one -mhard-float is working.
Any idea? |
|
|
| |
|
|
|
|
|
|
Posted: Feb 11, 2011 - 02:37 AM |
|
Joined: Dec 13, 2010
Posts: 35
|
|
It works for me using the below command line. What exactly are you trying to do?
avr32-gcc -march=ucr3fp -mhard-float -c -o float.o float.c |
|
|
| |
|
|
|
|
|
|
Posted: Feb 11, 2011 - 02:51 PM |
|
Joined: Aug 07, 2007
Posts: 30
|
|
Sorry I used the wrong GCC version.
With the AVR32Studio 2.7, here is the report of $avr32-gcc --version command
avr32-gcc (AVR_Toolchain_3.1.0_166) 4.3.3
And it works! |
|
|
| |
|
|
|
|
|
FAQ
Search
Register
Log in to check your private messages
Log in
Maximize
RSS
