Re: [Bug-epsilon] 64-bit Epsilon Testing

[Luca Saiu]

Thomas A. Cort wrote:
> I compiled and tested it as documented. I didn't look at the compiler 
> flags when I was testing it the first time. After checking just now, it 
> doesn't appear that the compilation uses -m32 or -m64, and the default 
> is -m32. So no, it didn't get compiled for 64-bit mode.

  So your UltraSparc was not seen any differently than my PowerPC: 
RISC, big-endian, 32-bit. Sure it worked :-) It has been my fault, 
sorry; I should have thought before of -m32 and -m64.

  Compiling with a flag it is simply a matter of passing it to 
configure via the environment variable CFLAGS (I would also like you to 
try an high optimization level such as -O9 -funroll-loops 
-finline-functions):

  cd DIRECTORY_OF_EPSILON_TARBALL_UNCOMPRESSED
  CFLAGS="-m64 -O9 -funroll-loops -finline-functions" \
    ./configure --prefix=/tmp/usr
  make
  make install

  If everything works as it should then please run
  /tmp/usr/eam-config
, and send me a copy of its output.
  Also, please run again the test in epsilon-test.tar.gz (but do *not* 
recompile epsilon: just run cd /tmp/epsilon-test; ./test) and send me 
the results as written in the README. I am especially interested in the 
GC time percentage (the one you reported the last time was quite high, 
maybe because the default optimization level is low).

> > Can I send you an updated tarball?

  Ok, the updated epsilon tarball and the test tarball are included.
  Thanks again for your help and patience.

--
Luca Saiu, maintainer of GNU epsilon
http://www.gnu.org/software/epsilon