This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
i[3456]86 test results
- To: egcs at cygnus dot com
- Subject: i[3456]86 test results
- From: Robert Lipe <robertl at dgii dot com>
- Date: Tue, 9 Jun 1998 12:03:01 -0500
Someone (Jeff?) suggested it might be fun to run the testsuite on a P6
with each set of {-mcpu=i386 -march=i386, -mcpu=486 -march=i486, ...}
That person would be wrong. :-) I was unable to convince dejagnu
in any reasonable way to loop through something like
--target_board unix{-mcpu=i386 -march=i386, -mcpu=i486 -march=i486}
I tried trainloads of different quoting mechanisms, command lines,
environmental variables, passing --tool_opts, dorking with config files,
and so on. I just couldn't make it take two options in each pass
instead of one. Then I remembered that not all things worth doing are
worth doing well. I just ran the test 4 times on my Linux (redhat 5.0)
system.
make -k check-gcc check-g++ RUNTESTFLAGS='--outdir /home/robertl/tmp/i386 --tool_opts "-mcpu=i386 -march=i386" '
make -k check-gcc check-g++ RUNTESTFLAGS='--outdir /home/robertl/tmp/i486 --tool_opts "-mcpu=i486 -march=i486" '
make -k check-gcc check-g++ RUNTESTFLAGS='--outdir /home/robertl/tmp/i586 --tool_opts "-mcpu=i586 -march=i586" '
make -k check-gcc check-g++ RUNTESTFLAGS='--outdir /home/robertl/tmp/i686 --tool_opts "-mcpu=i686 -march=i686" '
I boiled down the test results, and as of the last time I updated
(Monday of this week) there is only one different result, and I believe
it was in fact specifically added to catch a recent P6 problem.
The P6 compiler has only one additional failure:
FAIL: gcc.c-torture/execute/980604-1.c execution, -O1
FAIL: gcc.c-torture/execute/980604-1.c execution, -O2
FAIL: gcc.c-torture/execute/980604-1.c execution, -O2 -fomit-frame-pointer -finline-functions
FAIL: gcc.c-torture/execute/980604-1.c execution, -O2 -g
FAIL: gcc.c-torture/execute/980604-1.c execution, -Os
It took my 333Mhz system about five hours to grind through all these.
RJL