>> It happens only on the new GCC version on Ubuntu, and only with
optimization level 2.<<

Actually I think we concluded that it doesn't matter which Linux distro that
is used. It is only the fault of the GCC, the only thing we know at the
moment is that 4.2.3 is not working while 4.1.2 is working (both using O2).
Any versions in between have not been tested. I believe Alex tried with O1
on 4.2.3 and that worked too.

I don't have your simplest example anymore either, I didn't save that IRC
buffer where you posted it, however I think you just looped a simple new
statement and called gc each time or something, the loop would crash
somewhere between 100 and 200 if I remember correctly.

Thomas let me know if you for some reason can't compile with an older
version and I'll simply post the contents of bin compiled with 4.1.2 to this
list.

/Henrik

On Sat, Oct 4, 2008 at 12:50 AM, Alexander Burger <[EMAIL PROTECTED]>wrote:

> Hi Tomas,
>
> > Not sure what is going on exactly but some change in v2.3.3 must have
> > broken gc.  Looks like too much garbage is killing picolisp;-(
>
> I hope not ;-)
>
> At the moment, I suspect that it has another reason: The problems the
> latest versions of GCC have with code optimization.
>
> Henrik and I could produce a very simple example, basically consisting
> only of a single call 'new', repeated in a loop, to cause 'gc' to crash.
>
> @Henrik: Do you still have that example? I cannot find it any more :-(
>
> It happens only on the new GCC version on Ubuntu, and only with
> optimization level 2.
>
> Which version of GCC are you using? Could you try for a moment with -O1
> instead of -O2?
>
> BTW, I know since a few years that -O3 definitely produces wrong code
> (register allocation), so I switched back to -O2 then.
>
>
> If -O2 is really a problem, I don't know what to do. -O2 produces more
> than double as fast code as without optimization (-O0), and about 20%
> faster than -O1. And I don't want to give up that speed advantage.
>
> Cheers,
> - Alex
> --
> UNSUBSCRIBE: mailto:[EMAIL PROTECTED]
>

------=_Part_31798_26153809.1223099337428
Content-Type: text/html; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

<div dir="ltr">&gt;&gt;
It happens only on the new GCC version on Ubuntu, and only with<br>
optimization level 2.&lt;&lt;<br><br>Actually I think we concluded that it 
doesn&#39;t matter which Linux distro that is used. It is only the fault of the 
GCC, the only thing we know at the moment is that 4.2.3 is not working while 
4.1.2 is working (both using O2). Any versions in between have not been tested. 
I believe Alex tried with O1 on 4.2.3 and that worked too.<br>
<br>I don&#39;t have your simplest example anymore either, I didn&#39;t save 
that IRC buffer where you posted it, however I think you just looped a simple 
new statement and called gc each time or something, the loop would crash 
somewhere between 100 and 200 if I remember correctly.<br>
<br>Thomas let me know if you for some reason can&#39;t compile with an older 
version and I&#39;ll simply post the contents of bin compiled with 4.1.2 to 
this list.<br><br>/Henrik<br><br><div class="gmail_quote">On Sat, Oct 4, 2008 
at 12:50 AM, Alexander Burger <span dir="ltr">&lt;<a href="mailto:[EMAIL 
PROTECTED]">[EMAIL PROTECTED]</a>&gt;</span> wrote:<br>
<blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 
204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">Hi Tomas,<br>
<div class="Ih2E3d"><br>
&gt; Not sure what is going on exactly but some change in v2.3.3 must have<br>
&gt; broken gc. &nbsp;Looks like too much garbage is killing picolisp;-(<br>
<br>
</div>I hope not ;-)<br>
<br>
At the moment, I suspect that it has another reason: The problems the<br>
latest versions of GCC have with code optimization.<br>
<br>
Henrik and I could produce a very simple example, basically consisting<br>
only of a single call &#39;new&#39;, repeated in a loop, to cause &#39;gc&#39; 
to crash.<br>
<br>
@Henrik: Do you still have that example? I cannot find it any more :-(<br>
<br>
It happens only on the new GCC version on Ubuntu, and only with<br>
optimization level 2.<br>
<br>
Which version of GCC are you using? Could you try for a moment with -O1<br>
instead of -O2?<br>
<br>
BTW, I know since a few years that -O3 definitely produces wrong code<br>
(register allocation), so I switched back to -O2 then.<br>
<br>
<br>
If -O2 is really a problem, I don&#39;t know what to do. -O2 produces more<br>
than double as fast code as without optimization (-O0), and about 20%<br>
faster than -O1. And I don&#39;t want to give up that speed advantage.<br>
<br>
Cheers,<br>
- Alex<br>
<div><div></div><div class="Wj3C7c">--<br>
UNSUBSCRIBE: mailto:<a 
href="mailto:picolisp@software-lab.de";>picolisp@software-lab.de</a>?subject=Unsubscribe<br>
</div></div></blockquote></div><br></div>

------=_Part_31798_26153809.1223099337428--
-- 
UNSUBSCRIBE: mailto:[EMAIL PROTECTED]

Reply via email to