Re: [Gm2] Tarball of 14/10/14 (or 9/10) with Raspian Wheezy on ARM (RPi) - further improvement

[Gaius Mulley]

JD <address@hidden> writes:

> On 14/10/14 17:00, address@hidden wrote:
>
>     2. ARM - Raspberry Pi
>
>     === gm2 Summary ===
>     
>     # of expected passes 6744
>     # of unexpected failures 1513
>     # of unresolved testcases 1495
>     make: [check-gm2] Error 1 (ignored)
> But this one wasn't straightforward. I took up your hint to look at
> the ARM ABI. There are several cases on the web of the parameter
> passing error when calling the VFP unit. Unfortunately, there was no
> consistent solution. So, I tried a few changes to the configure
> options, and after trial and error this worked:
>
>     1. Force the architecture with the following extra options to
>     configure:
>     --build=arm-linux-gnueabihf --host=arm-linux-gnueabihf -
>     -target=arm-linux-gnueabihf 
>     This didn't itself solve the problem but it did force the use of
>     arm-linux-gnueabihf-gcc instead of plain gcc.
>     2. The option that made the difference was:
>     --with-float=hard
>     
>
> I tried compiling a few of my own modules and this showed that I didn't
> yet have libpth-dev installed. So I installed it and tried to run:
> make check-gm2
> but that failed eventually. So I rebuilt gm2 - but I'm not sure with
> which tarball - and the results are enormously improved:
>
>     === gm2 Summary ===
>     
>     # of expected passes 9584
>     # of unexpected failures 144
>     # of unresolved testcases 12
>     make: [check-gm2] Error 1 (ignored)
> BTW, four standard types on the ARM differ in size from on x86:
> On x86 REAL96 0; REAL128 16
> COMPLEX96 0; COMPLEX128 32
>
> On ARM REAL96 12; REAL128 0
> COMPLEX96 24; COMPLEX128 0
>
> Regards,
> John

Hi John,

wow much better!  On Wheezy x86_64 I get:

# of expected passes            9608
# of unexpected failures        120
# of unresolved testcases       12

so both architectures are converging.  So with respect to type sizes,
is it true that with hardfloat the arm has:

REAL96
COMPLEX96

but does not have:

REAL128 (and COMPLEX128)

Thanks for mentioning these types - it looks as if this is the failure
area.  I wonder if you could possibly post another gm2.sum ?

thanks,
Gaius