Re: GCC tickling obscure hardware bug or...?

culverk_at_yumyumyum.org wrote:
> Quoting Scott Reese <sreese_at_codysbooks.com>:
> 
> 
>>Hello all, I'm rephrasing my previous question to reflect new findings
>>in my situation in the hopes that someone may have an idea of what's
>>going on here (see thread "Internal compiler error in
>>reload_cse_simplify_operands" from earlier this week).
>>
>>In a nutshell, I have upgraded my machine to a PIV 2.4 GHz processor on
>>a VIA P4B 400 motherboard with 512 MB Samsung 2700 DDRAM.  I have a 300
>>Watt power supply with a ball-bearing fan and a Seagate 40 GB HD.  The
>>problem I'm having that I did not see with my PIII is that I'm getting
>>intermittent internal compiler errors when attempting to compile
>>anything (again, see previous thread for an example).  I even did a
>>fresh install of the system yesterday in order to rule out something
>>random getting hosed causing problems with my compiler.  Through
>>experimentation, I've found that toggling the CFLAGS from -O to -O2 or
>>vice-versa works around the problem (actually, I've had greater success
>>leaving my CFLAGS set to -O2 -pipe and occasionally bringing them back
>>to -O -pipe when I run into an ICE and then setting them back again).
>>My hunch is that some hardware bug is being tickled by gcc somehow.  I
>>don't think it's the standard "broken hardware" thing because I've not
>>received any signal 11/7/4 errors at all and the system runs
>>wonderfully.  So far it's been up for over 24 hours, compiling ports and
>>running without a hitch.  The only strangeness is the compiler's
>>behavior.  Also, I should mention that I added
>>
>>options DISABLE_PSE
>>options DISABLE_PG_G
>>
>>to my kernel config but the ICE's still persist, though they seem to be
>>less frequent now.
>>
>>Attached is my dmesg, please let me know if anyone would like any more
>>information about this or if you have any idea what might be going on
>>here.
>>
>>Thank you,
>>Scott
>>
> 
> 
> Whenever I've had problems like this, it's been either bad memory, or bad memory
> settings in bios. I'd check both of those things.
> 
> Ken

According to several bug reports on gcc.gnu.org related to cse and p4 I
strongly recommend to all whose using -On (with n>1) to add -fno-gcse
etc. for all cse options.

Of course, any interested can add themself as watcher for each
interested reported gcc bug.

Best,
Jens