Qt-embedded-interest Archive, October 2007
double and float in qtembedded linked apps on ARM
Message 1 in thread
Hi the list,
I use Qtembedded 3.3.4 compiled for arm to develop applications on a cirrus logic
EP9315 arm9 based system.
The tools chains I use are the ones from ELDK 31 and ELDK 41 from <www.denx.de>.
libqte is used as shared library.
libqte is configured with -DQT_QLOCALE_USES_FCVT switch.
when I compile and link my application with the linker flag "-lqte" I get a
perfectly working apps, BUT for the use of double and float types.
The same code (not calling any qt symbol) compiled and linked without "-lqte" works
fine, but if I add "-lqte" to the LIBS var in Makefile, I get strange behaviour when using
floating point types, especially if allocated with new operator.
EXAMPLE:
the code
________________________________
#include <stdio.h>
int main()
{
printf( "build: %s [%s]\r\n", __DATE__,__TIME__);
double d1 = 1.0 ;
double * pd = new double(2.);
printf( "%s:%d: (*pd)=%lf, d1=%lf, (*pd)+d1=%lf.\r\n",__FILE__,__LINE__, (*pd),
d1, (*pd)+d1 );
printf( "%s:%d: (*pd)=%lf, d1=%lf, (*pd)+d1=%lf.\r\n",__FILE__,__LINE__, (*pd),
d1, (*pd)+d1 );
delete pd;
return 0;
}
________________________________
linked with -lqte gives the following output:
________________________________
build: Oct 8 2007 [19:05:03]
..../src/main.cpp:10: (*pd)=2.000000, d1=1.000000, (*pd)+d1=3.000000.
..../src/main.cpp:11: (*pd)=2.000000, d1=1.000000, (*pd)+d1=inf.
________________________________
Does someone knows something about this?
Thank you.
--
[ signature omitted ]
Message 2 in thread
Hi Giovanni,
please try to compile it without compiler optimizations, this means replace "-O2" by "-O0" in the compiler switches.
This can be first manually done (call compiler commands directly) and if it helps, take a look at the "volatile" type qualifier.
Perhaps this helps.
Best regards,
Paul
-----Ursprüngliche Nachricht-----
Von: giovanniangeli@xxxxxxxxxxxxxxxxxxx [mailto:giovanniangeli@xxxxxxxxxxxxxxxxxxx]
Gesendet: Dienstag, 9. Oktober 2007 15:55
An: qt-embedded-interest@xxxxxxxxxxxxx
Betreff: double and float in qtembedded linked apps on ARM
Hi the list,
I use Qtembedded 3.3.4 compiled for arm to develop applications on a cirrus logic
EP9315 arm9 based system.
The tools chains I use are the ones from ELDK 31 and ELDK 41 from <www.denx.de>.
libqte is used as shared library.
libqte is configured with -DQT_QLOCALE_USES_FCVT switch.
when I compile and link my application with the linker flag "-lqte" I get a
perfectly working apps, BUT for the use of double and float types.
The same code (not calling any qt symbol) compiled and linked without "-lqte" works
fine, but if I add "-lqte" to the LIBS var in Makefile, I get strange behaviour when using
floating point types, especially if allocated with new operator.
EXAMPLE:
the code
________________________________
#include <stdio.h>
int main()
{
printf( "build: %s [%s]\r\n", __DATE__,__TIME__);
double d1 = 1.0 ;
double * pd = new double(2.);
printf( "%s:%d: (*pd)=%lf, d1=%lf, (*pd)+d1=%lf.\r\n",__FILE__,__LINE__, (*pd),
d1, (*pd)+d1 );
printf( "%s:%d: (*pd)=%lf, d1=%lf, (*pd)+d1=%lf.\r\n",__FILE__,__LINE__, (*pd),
d1, (*pd)+d1 );
delete pd;
return 0;
}
________________________________
linked with -lqte gives the following output:
________________________________
build: Oct 8 2007 [19:05:03]
..../src/main.cpp:10: (*pd)=2.000000, d1=1.000000, (*pd)+d1=3.000000.
..../src/main.cpp:11: (*pd)=2.000000, d1=1.000000, (*pd)+d1=inf.
________________________________
Does someone knows something about this?
Thank you.
--
[ signature omitted ]
Message 3 in thread
Hi Paul, thanks for your answer.
I tried to compile with 0 optimization (-O0) and to declare my floating point vars as volatile, but
this don't solve the problem.
I remark that it is stricly connected with the linking to the libqte: the same code not linked with
-lqte works fine. I think that some symbol defined in libqte (maybe in the qlocale) overwrite some
symbol in the system libs, but I don't know how to search in this direction.
Best regards, Giovanni.
Bussmann, Paul said:
> Hi Giovanni,
>
> please try to compile it without compiler optimizations, this means replace "-O2" by "-O0" in
> the compiler switches.
>
> This can be first manually done (call compiler commands directly) and if it helps, take a look
> at the "volatile" type qualifier. Perhaps this helps.
>
> Best regards,
> Paul
>
>
> -----Ursprüngliche Nachricht-----
> Von: giovanniangeli@xxxxxxxxxxxxxxxxxxx [mailto:giovanniangeli@xxxxxxxxxxxxxxxxxxx] Gesendet:
> Dienstag, 9. Oktober 2007 15:55
> An: qt-embedded-interest@xxxxxxxxxxxxx
> Betreff: double and float in qtembedded linked apps on ARM
>
>
> Hi the list,
>
> I use Qtembedded 3.3.4 compiled for arm to develop applications on a cirrus logic EP9315 arm9
> based system.
> The tools chains I use are the ones from ELDK 31 and ELDK 41 from <www.denx.de>. libqte is used
> as shared library.
> libqte is configured with -DQT_QLOCALE_USES_FCVT switch.
>
> when I compile and link my application with the linker flag "-lqte" I get a perfectly working
> apps, BUT for the use of double and float types. The same code (not calling any qt symbol)
> compiled and linked without "-lqte" works fine, but if I add "-lqte" to the LIBS var in
> Makefile, I get strange behaviour when using floating point types, especially if allocated with
> new operator.
>
> EXAMPLE:
> the code
> ________________________________
> #include <stdio.h>
>
> int main()
> {
> printf( "build: %s [%s]\r\n", __DATE__,__TIME__);
>
> double d1 = 1.0 ;
> double * pd = new double(2.);
>
> printf( "%s:%d: (*pd)=%lf, d1=%lf, (*pd)+d1=%lf.\r\n",__FILE__,__LINE__, (*pd), d1, (*pd)+d1
> );
> printf( "%s:%d: (*pd)=%lf, d1=%lf, (*pd)+d1=%lf.\r\n",__FILE__,__LINE__, (*pd), d1, (*pd)+d1
> );
>
> delete pd;
>
> return 0;
> }
> ________________________________
>
> linked with -lqte gives the following output:
>
> ________________________________
> build: Oct 8 2007 [19:05:03]
> ..../src/main.cpp:10: (*pd)=2.000000, d1=1.000000, (*pd)+d1=3.000000. ..../src/main.cpp:11:
> (*pd)=2.000000, d1=1.000000, (*pd)+d1=inf.
> ________________________________
>
>
> Does someone knows something about this?
>
> Thank you.
>
>
> --
> Giovanni Angeli
>
>
> To unsubscribe - send "unsubscribe" in the subject to
> qt-embedded-interest-request@xxxxxxxxxxxxx
--
[ signature omitted ]
