Re: PING: PATCH - PPC-darwin Correct initialization of 'long long' with constant

[Fariborz Jahanian <fjahanian at apple dot com>]

On Jan 23, 2004, at 2:20 PM, Geoff Keating wrote:

> Fariborz Jahanian <fjahanian@apple.com> writes:
>
> > This is also: PR13779
> >
> > http://gcc.gnu.org/ml/gcc-patches/2004-01/msg01497.html
>
> - This should have a proper dejagnu testcase

Yes, there is. I sent it earlier. Here it is:

Attachment: darwin-longlong.c
Description: Text document

> - Why doesn't this cause a problem if the input is a REG?
>   It looks like every other part of this if/else chain takes
>   'bytepos' into account, except this one:
>
>       else if (CONSTANT_P (src)
>                || (GET_CODE (src) == REG && GET_MODE (src) == mode))
>         tmps[i] = src;
>
>   I think you are overriding the CONSTANT_P case, but what about the 
> REG
>   case?

Above is not my code. My code checks for parallel pattern generated for 
'long long' arguments and
generates the constants. Maybe some one else can respond how above 
pattern is used.

- Fariborz


> --
> - Geoffrey Keating <geoffk@geoffk.org>