Compiling with -msoft-float

Compiling with -msoft-float

[Athanasios Anastasiou]

Hello All

I am developing a quasi realtime DSP application in C++ that is to be
ported later to an ARM processor. As far as i know this processor does
not have an FPU.
To get an estimate on the performance penalty, i tried to g++ -o MyTest
MyTest.cpp -msoft-float on the development machine (Suse 9.1
proffesional) but i get a load of linker errors reporting that specific
low level math instructions "can not be found"

After a search through the internet and the gcc mailing lists i found
out a lot of fragmeneted information about this subject.

Could you please provide a simple walk through procedure where a "Hello
Soft Floating Point World" is compiled with -msoft-float succesfuly? (If 
this is possible...Or any other helpful comment on the subject?)

In case i have not looked thoroughly, could you please provide any
links, books, physical places, that compiling with soft floating point
is dealt with in detail?

All the best.
thanOS

Re: Compiling with -msoft-float

[Eljay Love-Jensen]

Hi thanOS,

>i tried to g++ -o MyTest MyTest.cpp -msoft-float 

Try:  g++ -o MyTest MyTest.cpp -msoft-float -lm

HTH,
--Eljay

Re: Compiling with -msoft-float

[Athanasios Anastasiou]

Hi LJ

I have done that but it was no good...

Here is a list of those "low level" math functions, ld doesnt seem to 
locate:

/tmp/ccKhuObA.o(.text+0x94): In function `main':
: undefined reference to `__fixdfsi'
/tmp/ccKhuObA.o(.text+0xc8): In function `main':
: undefined reference to `__fixdfsi'
/tmp/ccKhuObA.o(.text+0xfc): In function `main':
: undefined reference to `__fixdfsi'
/tmp/ccKhuObA.o(.text+0x144): In function `main':
: undefined reference to `__fixdfsi'
/tmp/ccKhuObA.o(.text+0x178): In function `main':
: undefined reference to `__fixdfsi'
/tmp/ccKhuObA.o(.text+0x1ac): more undefined references to `__fixdfsi' 
follow
/tmp/ccKhuObA.o(.text+0x204): In function `main':
: undefined reference to `__floatsidf'
/tmp/ccKhuObA.o(.text+0x221): In function `main':
: undefined reference to `__divdf3'
/tmp/ccKhuObA.o(.text+0x298): In function `main':
: undefined reference to `__floatsidf'
/tmp/ccKhuObA.o(.text+0x2b5): In function `main':
: undefined reference to `__divdf3'
/tmp/ccKhuObA.o(.gnu.linkonce.t._Z14benchmarkFloatv+0x4b): In function 
`benchmarkFloat()':
: undefined reference to `__muldf3'


All the best.
thanOS

Re: Compiling with -msoft-float

[lamikr]

Hi!

I suggest you to download the crosstool 0.35. With it it is easy to 
build own crosscompiler with soft-fp support.
It has lot of examples for different target cpu, target gcc and target 
glibc.
Just extract it and check demo-*.sh scripts...

But I have also a questiong for you? Could you send the test program you 
are using for me?
I am working with HP iPaq H6340 that has TI's arm925 cpu and I need 
somehow find out whether this CPU supports
fp or do I need to use soft-fp instead.
I have gcc 4.0 for this CPU without soft-fp support and at least some 
simple apps which make divisions and multibles with doubles
seems to work just fine. But before being sure I need more comprehensive 
test apps...

Mika

Athanasios Anastasiou wrote:

> Hello All
>
> I am developing a quasi realtime DSP application in C++ that is to be
> ported later to an ARM processor. As far as i know this processor does
> not have an FPU.
> To get an estimate on the performance penalty, i tried to g++ -o MyTest
> MyTest.cpp -msoft-float on the development machine (Suse 9.1
> proffesional) but i get a load of linker errors reporting that specific
> low level math instructions "can not be found"
>
> After a search through the internet and the gcc mailing lists i found
> out a lot of fragmeneted information about this subject.
>
> Could you please provide a simple walk through procedure where a "Hello
> Soft Floating Point World" is compiled with -msoft-float succesfuly? 
> (If this is possible...Or any other helpful comment on the subject?)
>
> In case i have not looked thoroughly, could you please provide any
> links, books, physical places, that compiling with soft floating point
> is dealt with in detail?
>
> All the best.
> thanOS
>
>