This is the mail archive of the
gcc-help@gcc.gnu.org
mailing list for the GCC project.
Re: [Consult] tile: How to construct testsuite for tile architecture without hardware.
- From: Chen Gang S <gang dot chen at sunrus dot com dot cn>
- To: Mike Stump <mikestump at comcast dot net>
- Cc: Chris Metcalf <cmetcalf at ezchip dot com>, walt at tilera dot com, Jeff Law <law at redhat dot com>, gcc-help <gcc-help at gcc dot gnu dot org>, Michael Eager <eager at eagerm dot com>
- Date: Wed, 04 Feb 2015 07:00:06 +0800
- Subject: Re: [Consult] tile: How to construct testsuite for tile architecture without hardware.
- Authentication-results: sourceware.org; auth=none
- References: <54BA30F3 dot 4080202 at sunrus dot com dot cn> <54BD2C5E dot 1070804 at ezchip dot com> <54BDB71D dot 90703 at sunrus dot com dot cn> <54C39ECA dot 3000603 at sunrus dot com dot cn> <54CC0C79 dot 1070007 at sunrus dot com dot cn> <D90BD733-35A0-4ADD-9D1E-0913064A6EF9 at comcast dot net> <54CC6D9B dot 30106 at sunrus dot com dot cn> <54CCBB78 dot 6020408 at sunrus dot com dot cn> <9B72BD56-C4CF-4AC3-B9AB-1D71E76BF11D at comcast dot net> <54D0D934 dot 7060904 at sunrus dot com dot cn> <FAB87024-222B-49F2-838F-92C121C80AD0 at comcast dot net>
On 02/04/2015 01:24 AM, Mike Stump wrote:
> On Feb 3, 2015, at 6:20 AM, Chen Gang S <gang.chen@sunrus.com.cn> wrote:
>> If what I said above is really correct, for me, we have to try to find
>> another ways to solve this issue (it is not worthy enough to modify
>> quite a few code only for avoiding the suggesting warnings).
>
> So, ultimately, it is reasonable to ask, is it better to so fix the warnings or leave them as is…
>
OK, thanks. And excuse me, originally, I am not quite familiar with gcc,
so the related consult may be not quite precise.
> Also, one is free to research better techniques on how to either improve the warning or eliminate it entirely. For example, one could have an modeless operation be auto (a la C++) instead of always VOIDmode and have auto turn into VOIDmode unless the mode can be deduced. For example (set (reg:auto ) (and:DI )) deduces to DI. Maybe failure to deduce would be what we warn for. This could be another way to get rid of some of the warnings.
>
OK, thank you for your suggestions and ideas.
For me, I prefer to add additional mode in rtl template (e.g. IVOIDmode)
to let genrecog know about the rtl template intends to use VOIDmode, so
needn't warn again (only treat IVOIDmode as VOIDmode without warning).
- Just like most of languages (C, C++), can use type cast to avoid
warnings, so it is easy understanding to add the same features for
rtl template.
- This modification is the lowest cost: not only tile, but also I guess
another archs may also have this kind of issue.
- This modification is easy and clear enough to be documented to let
all other members know about it. And it also compitables with all
original rtl template features.
Welcome any additional ideas, suggestions and completions.
Thanks.
--
Open, share, and attitude like air, water, and life which God blessed.