2015-10-22 15:30 GMT+02:00 Martell Malone <[email protected]>:
> The reason I am asking is because there is a calling convention bug for
> mingw-w64 and clang that is only present on x86_64.
>
> Are you sure there are no differences from a Linux target?

Yes, I am.  At least at that time I made x64 target ready, it used the
80-bit floating-point too.  Nevertheless it is most naturally to use
same floating-point precision for x64 as used for x87, as targets need
to be remain compatible here.

So not sure what ABI difference to Clang exists here?  I don't see
such issue on gcc-side.

>
> On Thursday, October 22, 2015, Kai Tietz <[email protected]> wrote:
>>
>> Hi guys,
>>
>> 2015-10-22 12:54 GMT+02:00 Martell Malone <[email protected]>:
>> > Hi guys,
>> >
>> > Can someone confirm the format of long double for the x86_x64 target of
>> > mingw-w64.
>> >
>> > Under i686 it is x87 80 bit precision
>> > But under x86_64?
>>
>> It is also 80 bit precision.  We kept it the same as for linux x86_64
>> targets.In general this is already an extension of x64 ABI, as for the
>> MS' compiler 'long double' has just 64-bit precision.
>>
>>
>> > Is it sse2 128 or 80 bit or is it still x87
>>
>> We use for 80 bit still x87 opcodes.  Indeed it could be an
>> improvement to use a soft-emu using sse2 instructions and just
>> down-scale to 80-bit precision from it.  At least for x86_64 this
>> might be indeed a speed-improvement in some cases.
>>
>> > Kind regards
>> > Martell
>>
>> Regards,
>> Kai
>>

------------------------------------------------------------------------------
_______________________________________________
Mingw-w64-public mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mingw-w64-public

Reply via email to