docs/internals/SPEC-notes.txt

From Vince Weaver:

I've been running the SPEC CPU 2006 benchmarks under valgrind (doing some
work on my BBV generating plugin).

There are two benchmarks that have issues, and I thought I'd share them
here for future reference.

1). zeusmp - does not run

    It has a 1GB data segment, which valgrind cannot handle on a 32-bit
    CPU.

2). dealII - runs forever, never ending

    It took a while, but I tracked this down to a 64bit/80bit
    floating point issue.

    The code in the QGauss<1>::QGauss() function has some code like this:

    const long double tolerance = std::max (static_cast<long double>
       (std::numeric_limits<double>::epsilon() / 100),
       static_cast<long double>(std::numeric_limits<long
       double>::epsilon() *5));

     do {
       ....
       various fp operations
       ....
     } while (abs(p1/pp) > tolerance);


     The tolerance in this case is being set to ~2.22e-18, but the
     abs(p1/pp) value never gets below ~2.586e-17 under valgrind.

     [This is because Valgrind only uses 64-bit FP values on x86, not 80-bit
     values.]

     This is similar to an issue that happens with the "art"
     benchmark on SPEC CPU 2000, but in the "art" case it only
     makes the code take longer to finish; this "dealII" problem
     makes the benchmark loop forever.