cpuuse, rtmonuse and gcc

[Joel Sherrill]

Kitchen, George A. (JSC-ES5)[ESCG] wrote:
> Perhaps I misunderstand this email, so I will ask.  Are you saying 
> that one cannot use the FPU under the latest version of RTEMS with a 
> Sparc?
>  
No. The kernel itself is compiled with soft-float so it doesn't know the
difference until you need to context switch but user application code
can use the FPU -- you just have to specify tasks are FP and compile
without -msoft-float.

--joel
> Allen Kitchen
>
> ------------------------------------------------------------------------
> *From:* rtems-users-bounces at rtems.org on behalf of Joel Sherrill
> *Sent:* Fri 3/28/2008 8:58 AM
> *To:* Jiri Gaisler
> *Cc:* rtems-users at rtems.org
> *Subject:* Re: cpuuse, rtmonuse and gcc (was: erc32 wrong compilation 
> flagsfor initial task with fpu?)
>
> Jiri Gaisler wrote:
> > This is pretty bad. If the kernel uses floats, then compilation
> > with -msoft-float will not work if the application later is linked
> > without -msoft-float. Are the CPU_usage_Dump() and Period_usage_Dump()
> > new functions or have they always been around?
> I think our emails crossed.  Older versions had floating
> point math if the FPU defined was set to TRUE.
>
> Newer versions have been rewritten to completely
> use integer math.
>
> So the question for Aleix is what version of RTEMS
> do you have?  If it is the older FP using version,
> can you backport the integer version of the functions
> and post a patch.
> > Are there other
> > functions in the kernel that uses floats as well?
> >  
> There are not supposed to be.  But gcc has surprised
> us in the past by generating FP using code when you
> least expected it.
>
> So AFAIK "float/double" C types should not be found
> anywhere in the code except to declare the FP context
> structures anymore.
>
> --joel
> > Jiri.
> >
> > Aleix Conchillo Flaqué wrote:
> >  
> >> Hi again,
> >>
> >> I forced the SPARC_HAS_FPU to 1, so CPU_HARDWARE_FP is set to TRUE.
> >> Now (compiling RTEMS with -msoft-float), the CPU_usage_Dump and
> >> Period_usage_Dump print 0.000 for percentage values.
> >>
> >> I have disassembled the code and I have seen that the compiler forces
> >> those functions (by chance, only these two) to use __floatsidf and
> >> __divdf3 functions, which I think are for software floating point
> >> emulation. This is fine, the problem is that they use floating point
> >> registers internally.
> >>
> >> So, even RTEMS is compiled with -msoft-float, gcc floating point
> >> software emulation function use floating point registers, which is
> >> kind of scary.
> >>
> >> Luckily, my task had the RTEMS_FLOATING_POINT attribute set, so I
> >> could see the statistics. Removing the attribute, which I think it
> >> should, crashed the application because of the gcc problem cited
> >> above.
> >>
> >> The problems I am facing now are:
> >>
> >> - gcc uses floating point registers for some floating point software
> >> emulation routines.
> >> - cpu_usage_dump and period_usage_dump do not print correct percentage
> >> values. I am not sure if this is caused by the point above.
> >>
> >> Does this makes any sense at all?
> >>
> >>
> >> Aleix
> >>
> >>
> >>    
>
>
> --
> Joel Sherrill, Ph.D.             Director of Research & Development
> joel.sherrill at OARcorp.com        On-Line Applications Research
> Ask me about RTEMS: a free RTOS  Huntsville AL 35805
>    Support Available             (256) 722-9985
>
>
> _______________________________________________
> rtems-users mailing list
> rtems-users at rtems.com
> http://rtems.rtems.org/mailman/listinfo/rtems-users
>


-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel.sherrill at OARcorp.com        On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
   Support Available             (256) 722-9985