long double format on I!64

long double format on I!64

[Gabriel Dos Reis]

Hi,

  I've been reviewing the fundamental data types format as
communicated to V3 sources. due to some mismatch in data, we're now
getting some information directly from the compiler.  See

      http://gcc.gnu.org/ml/gcc-patches/2002-08/msg01672.html

In the file

     libstdc++-v3/config/cpu/ia64/cpu_limits.h

I see the following 

    // While sizeof(long double) == 16, the format is the same as the x86.
    #define __glibcpp_long_double_bits 80

I don't know how exactly I should interpret that since 80 bits don't
make 16 bytes.  I guess that means some bits are ignored.  If so I need
to know exactly which ones. 

Thanks,

-- Gaby

Re: long double format on I!64

[Bo Thorsen]

On Thursday 29 August 2002 00:02, Gabriel Dos Reis wrote:
> Hi,
>
>   I've been reviewing the fundamental data types format as
> communicated to V3 sources. due to some mismatch in data, we're now
> getting some information directly from the compiler.  See
>
>       http://gcc.gnu.org/ml/gcc-patches/2002-08/msg01672.html
>
> In the file
>
>      libstdc++-v3/config/cpu/ia64/cpu_limits.h
>
> I see the following
>
>     // While sizeof(long double) == 16, the format is the same as the x86.
>     #define __glibcpp_long_double_bits 80
>
> I don't know how exactly I should interpret that since 80 bits don't
> make 16 bytes.  I guess that means some bits are ignored.  If so I need
> to know exactly which ones.

It's like x86, just with 6 bytes padding instead of 2. So the first 10 bytes 
are the double.

BTW, x86-64 does exactly the same.

Bo.

-- 

     Bo Thorsen                 |   Praestevejen 4
     Free software developer    |   5290 Marslev
     SuSE Labs                  |   Denmark

Re: long double format on I!64

[Gabriel Dos Reis]

Bo Thorsen <bo@sonofthor.dk> writes:

| > I don't know how exactly I should interpret that since 80 bits don't
| > make 16 bytes.  I guess that means some bits are ignored.  If so I need
| > to know exactly which ones.
| 
| It's like x86, just with 6 bytes padding instead of 2. So the first 10 bytes 
| are the double.

OK, thanks a lot.

Hmm, I guess I'm going to have another funny day with this stuf.

-- Gaby

Re: long double format on I!64

[Jim Wilson]

ia64-linux uses a long double with 80-bits of info stored in a 128-bit
slot.  The 80-bit format is used for compatibility with x86 systems.
The 128-bit slot is used to make alignment/padding simpler.  The first
ten bytes are the FP value, the last 6 bytes are the padding.  This is
true for both big and little endian.

ia64-hpux however uses an actual 128-bit long double.  This is for
compatibility with hppa-hpux systems.  I don't have access to ia64-hpux,
so I can't help you with this.

Jim

Re: long double format on I!64

[Gabriel Dos Reis]

Steve Ellcey <sje@cup.hp.com> writes:

| >     // While sizeof(long double) == 16, the format is the same as the x86.
| >     #define __glibcpp_long_double_bits 80
| 
| FYI:  While Linux on IA64 uses the 80 bit format, HP-UX does not.  It
| uses the full 128 bit format for long double.  In gcc/config/ia64 the
| floating point format is determined by the macro
| INTEL_EXTENDED_IEEE_FORMAT.  (1 for 80 bit format, 0 for 128 bit
| format).  This is because the HP-UX C and C++ compilers also use 128
| bits for long double in order to be compatible with the HP-UX PA
| compilers.

Thanks, that is helpful.

-- Gaby

Re: long double format on I!64

[Steve Ellcey]

>     // While sizeof(long double) == 16, the format is the same as the x86.
>     #define __glibcpp_long_double_bits 80

FYI:  While Linux on IA64 uses the 80 bit format, HP-UX does not.  It
uses the full 128 bit format for long double.  In gcc/config/ia64 the
floating point format is determined by the macro
INTEL_EXTENDED_IEEE_FORMAT.  (1 for 80 bit format, 0 for 128 bit
format).  This is because the HP-UX C and C++ compilers also use 128
bits for long double in order to be compatible with the HP-UX PA
compilers.

Steve Ellcey
sje@cup.hp.com