Andrei Alexandrescu wrote:
Don wrote:
Close, but that's technically not true in the case where abs(a/b) >
long.max. (The integer doesn't have to fit into a 'long').
But if real is 79-bit long (as on Intel), the largest integer that could
fit without loss in 1 << 63, and that would fit in a long. Are you
saying r could spill into large integers that cannot be represented
without loss?
The definition is without regard to the size of integral types. It only
means "integer".
In IEEE754, r= a % b is defined by the mathematical relation r = a –
b * n , where n is the integer nearest the exact number a/b ;
whenever abs( n – a/b) = 0.5 , then n is even. If r == 0 , its sign
is the same as a.
I take it D does not define a % b the IEEE 754 way (that's why I
eliminated that mention). Is that correct?
No, it is defined as fmod, which is IEEE754 %. In fact, it is quite
literally implemented by the FPREM instruction which is the same used
for fmod().
Hmm, I just noticed that the code generator should use FPREM1 instead to
get IEEE conformance. Darn.
http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc108.htm
http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc109.htm
