Re: Machine/Compiler Independent Results

John Harper <harper@xxxxxxxxxxxxx> wrote:

Sparc. Sorry, I didn't know exactly which info you needed.
Of the 4 compilers I used, only Sun f95 supports quad precision.
No "Oops!" with that program when converted to quad. Modifying it for
even more extended precision is not a game I want to play on the last
working day before Christmas!

I believe the problem is mostly x86. Kahan has some articles on
the advantages of keeping intermediate values in extended precision.

I believe that he wants consistent use of extended precision, which is
not what happens in most cases. The problem with x87 is that to do it
right, intermediate values must be stored and reloaded, hopefully
through a cache. It still slows it down much more than anyone wants.

If only intermediate values within an expression were kept in
extended precision then at least one could explicitly store values
to remove extra precision. With most optimizers, that doesn't even
work. (Assuming optimizing between statements is on.)

On MIPS, SPARC, HP-PA, and Alpha you should not see Oops!.

-- glen
.

Relevant Pages

  • Re: double to int conversion yields strange results
    ... giving me two different values for the int. ... the question raised is not about the precision of the ... is done in extended precision. ... It is the conversion to 'double' before the conversion to 'int' that ...
    (comp.lang.c)
  • Re: Who uses clapack?
    ... Use of traditional BLAS interposes a mandatory memory save ... > the sqrtin extended precision, but it is not forced to truncate ... > do a simple dot product to compute the vector norm, ...
    (sci.math.num-analysis)
  • Re: Consistent floating-point computations and performance
    ... numbers that were or were not rounded down to zero. ... x/y that should have been zero (if stored in double precision) was not ... Do you mean using extended precision options with all results stored to ... Most CPUs have options to set abrupt underflow, often with a significant potential increase in performance. ...
    (comp.lang.fortran)
  • Re: Are f.p. manipulation functions only used in initialization?
    ... mathematical libraries assume that extended precision ... libraries could suffer significant loss of accuracy, ... About the only reason to change the FPU precision is if you are ... having the FPU in double precision, not double extended precision. ...
    (comp.lang.fortran)
  • Re: Kind of NOT integer constant
    ... > A processor that does NOTas a default precision operation ... > set bits right adjusted in the extended precision result). ... > the variable on the left of the assignment is default precision ... > consistent and standard conforming. ...
    (comp.lang.fortran)