This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [COMMITTED] tilegx: enable wordsize-64 support for ieee745 dbl-64.
- From: "H.J. Lu" <hjl dot tools at gmail dot com>
- To: Chris Metcalf <cmetcalf at ezchip dot com>
- Cc: Andrew Pinski <pinskia at gmail dot com>, Joseph Myers <joseph at codesourcery dot com>, GNU C Library <libc-alpha at sourceware dot org>
- Date: Fri, 26 Dec 2014 07:44:48 -0800
- Subject: Re: [COMMITTED] tilegx: enable wordsize-64 support for ieee745 dbl-64.
- Authentication-results: sourceware.org; auth=none
- References: <201412231905 dot sBNJ5c3C004622 at farm-0002 dot internal dot tilera dot com> <alpine dot DEB dot 2 dot 10 dot 1412231958370 dot 31777 at digraph dot polyomino dot org dot uk> <5499D4FE dot 8070405 at ezchip dot com> <5499DA84 dot 3010205 at ezchip dot com> <CAMe9rOr2DU5qYDQriF9RvAJCJF_wZ3gzqqHGDZXf+-OZFuJX0g at mail dot gmail dot com> <5499E115 dot 9000502 at ezchip dot com> <CAMe9rOrFEVEpPik-qMR=SPM-Eqgetwg9C3oztjT4yTf74S+EWw at mail dot gmail dot com> <549AF3A3 dot 8070007 at ezchip dot com> <CA+=Sn1kumpH2iCOYykmJb8LjNA1jufoKj8hhqiA4XR=KtCJA-g at mail dot gmail dot com> <549D7976 dot 8020403 at ezchip dot com>
On Fri, Dec 26, 2014 at 7:06 AM, Chris Metcalf <cmetcalf@ezchip.com> wrote:
> On 12/24/2014 6:45 PM, Andrew Pinski wrote:
>>
>> On Wed, Dec 24, 2014 at 9:10 AM, Chris Metcalf <cmetcalf@ezchip.com>
>> wrote:
>>>
>>> On 12/23/2014 5:39 PM, H.J. Lu wrote:
>>>>
>>>> You can provide a safe default and let x32 override it.
>>>
>>>
>>> I think it's conventional now to require that platforms are explicit
>>> about #defines, and use #if with -Wundef. It seems to make good sense
>>> in this case, since it's really not clear what the right default for
>>> any given platform is.
>>>
>>> And since there are currently just two platforms that use wordsize-64
>>> and are !_LP64 (x32 and tilegx32), it feels pretty reasonable to be
>>> explicit for both. The AArch64 guys will need to set the right default
>>> for their ILP32 mode; I'm not even sure what the right answer is for
>>> them.
>>
>> AARCH64:ILP32 is fine with having the upper bits being undefined when
>> dealing with passing and return values.
>
>
> So sounds like two in each "camp", x32 and AArch64:ILP32 not wanting a cast,
> and tilegx32 and MIPS needing the explicit cast, so requiring a
> preprocessor symbol to be defined for that code seems like the right thing.
>
> Any objections to the proposed code to do the extension in llround.c?
>
Can we find a better name that USE_32BIT_NORMALIZE? How
about REGISTER_CAST_INT32_TO_INT64? Also it should be an
error if the macro is undefined.
--
H.J.