This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

[2] egcs-{core|g77}-970901 on m68k-next-nextstep3


Hi All,

Well, I was able to build HIRLAM (URL:http://www.knmi.nl/hirlam for  
an explanation of what it is), but it didn't complete succesfully.

There are at least two routines that compile to code that leads to  
a Memory Fault at run time.  The first one was easy to spot, and I  
"repaired" the problem by compiling that specific routine with -O2  
instead of -O2 -fforce-addr.  At the second memory fault I gave up.

Timings so far:

egcs-970901:

0SUPOBS TOOK :       21.09100341797
0DATACH TOOK :     12391.4404296875
0ANAEVA TOOK :     12846.1289062500
0GRPEVA TOOK :     23506.7128906250
0HUMSUP TOOK :        1.68750000000
0DATACH TOOK :       698.2617187500
0HUMEVA TOOK :       317.0781250000
0GRPEVA TOOK :       399.8398437500
 PREPARATIONS TOOK         51.3353 SECONDS
 FORECAST TOOK        318.1753 SECONDS
 PREPARATIONS TOOK         57.4092 SECONDS

egcs-970821:

0SUPOBS TOOK :       21.24188232422
0DATACH TOOK :     13005.2167968750
0ANAEVA TOOK :     12855.5527343750
0GRPEVA TOOK :     24094.5722656250
0HUMSUP TOOK :        1.87500000000
0DATACH TOOK :       715.5117187500
0HUMEVA TOOK :       318.5078125000
0GRPEVA TOOK :       402.7421875000
 PREPARATIONS TOOK         52.4014 SECONDS
 FORECAST TOOK        296.5845 SECONDS
 PREPARATIONS TOOK         57.6717 SECONDS

i.e. slightly better, but not enough:

g77-0.5.21-970811 + gcc-2.7.2.2:

0SUPOBS TOOK :       20.02850341797
0DATACH TOOK :      9476.3925781250
0ANAEVA TOOK :     11479.2958984375
0GRPEVA TOOK :     22881.7519531250
0HUMSUP TOOK :        1.60937500000
0DATACH TOOK :       579.2109375000
0HUMEVA TOOK :       292.9531250000
0GRPEVA TOOK :       384.8281250000
 PREPARATIONS TOOK         49.1833 SECONDS
 FORECAST TOOK        286.4648 SECONDS
 PREPARATIONS TOOK         55.1645 SECONDS

BTW, Jeff, if you're trying to reproduce the compiler crash on  
compiling LOWPASS.F using options -O2 -fforce-addr, note that I  
changed the g77 front end thusly (file f/com.c):

/* Simple definitions and enumerations. */

#define FFECOM_sizeMAXSTACKITEM 0	<-- *** Note This ***
#ifndef FFECOM_sizeMAXSTACKITEM
#define FFECOM_sizeMAXSTACKITEM 32*1024 /* Keep user-declared things

Regards,
Toon.

PS: I'm concentrating for the time being on g77-0.5.21-970902 +  
gcc-2.7.2.3, because we want to get g77-0.5.21 out.

PS2: Craig, note that two recent alias patches by Jim Wilson are  
not in the backend patch for g77-0.5.21-970902.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]