Re:   problem while building arm vfp softfloat gcc 

[Dan Kegel <dank at kegel dot com>]

Add reports that the arm soft-float toolchain generated by
either crosstool-0.27 or the latest ptxdist can't do
  printf("%f", 1.0)
properly, whereas Montavista's arm toolchain handles it fine.
All three use nico's patch.  No idea where the problem lies;
I'm just passing on add's report, since he's having trouble posting.

Add also reports that the problem happens in a chroot environment
populated with crosstool.  I don't know whether he put it together
properly, but he did try.

Sure wish I had time to work on this...
- Dan

-------- Original Message --------
Subject: Re:   problem while building arm vfp softfloat gcc 
Date: 8 Feb 2004 10:37:10 +0800
From: add <addsub@eyou.com>
Reply-To: add <addsub@eyou.com>
To: dank@kegel.com

Because I have a built montavista toolchain, which is patched by nico.pitre 's
vfp patch, and I'm comparing the asm generated by mine toolchain and
montavista's. Then I found the only difference is the r2 and r3 sequence in
param passing. the binary generated by montavista can run well in my
environment, but mine should first be modified by these two line.
I think it has nothing to do with libgcc_s.so:) maybe I will test it for sure
on Monday.
ptxdist mailinglist seems down...
Thanks alot, hoping nico can send out a response.
> add wrote:
> > I'm not sure whether these code are right, because the foo testcase
doesn't
> > generate a output for me to judge.
>
> That's easy to fix.  Just do various operations that use float, and
> assert() that the result is as expected.
>
> > no, I don't think it's related to glibc compile, because my printf test
case
> > reveal there is something wrong in gcc float param passing.
> > In my testcase, I used glibc compiled by crosstool-0.27, and staticlly
linked
> > it.my kernel is linux-2.4.21-rmk1-pxa1.
>
> But did you install the libgcc_s.so generated by crosstool?
>
> IMHO you may have to create a chroot environment to actually test
> this; simply compiling a static executable might not be enough.
>
> > I will not only report his bug, I want to help you fix this, so I began
> > digging into this, but only this finding by far.I don't understand gcc
well.
> > Before anything I can do, I need somebody to tell me which line in gcc
source
> > tree generate the two asm lines. 
>
> That's a tough one, isn't it?
>
> > maybe nico,pitre knows this? but I think he is to busy to do this :)
>
> Indeed.
>
> > and I cannot subscribe to Robert's pxt mailing list, so I don't know how he
is
> > thinking about this...
>
> Why can't you subscribe?  Or email him directly, if it comes to that?
> - Dan





--http://www.eyou.com
--Îȶ¨¿É¿¿µÄµç×ÓÐÅÏä  ÓïÒôÓʼþ  Òƶ¯ÊéÇ©  ÈÕÀú·þÎñ  ÍøÂç´æ´¢...ÒÚÓÊδ¾¡

--http://vip.eyou.com
--¿ì¿ìµÇ¼ÒÚÓÊVIPÐÅÏä  ×¢²áÄúÖÐÒâµÄÓû§Ãû






------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sources.redhat.com