Re: floating point

Bruce Dawson <comments@cygnus-software.com> writes:


>Bruce Dawson said:
>> > It can introduce double rounding, and it doesn't clamp the
>> > exponent. The exponent clamping can be forced by writing to memory,
>> > but the double rounding is inevitable (but rare?)


>Joseph D. Darcy wrote:


[double rounding on underflow is unlikely to cause a problem for
practical programs]


>Just to clarify: Doug Priest's quote was talking about double rounding
>caused by writing doubles to memory and having the exponent range
>reduced by this process. I was intending to refer to the double
>rounding that happens in the x86 when the rounding mode is set to
>double and the results are first rounded to 64-bit precision, and then
>to 53 bit precision - even before being written to memory. On this
>topic Doug Priest says:


When the x86's rounding precision is set to double, the answer is
rounded to 53 significand bits (with extended exponent range).
Likewise, when the rounding precision is set to float, only 24 bit
significands are generated.


If a compiler sets the rounding precision to double extended (64
significand bits) and then writes these values out as double, double
rounding can occur. However, if the intention is to only operate on
double values and not use double extended, there is no reason not to
set the rounding precision. Adds and multiplies go just as fast under
double precision and divide is somewhat faster.


-Joe Darcy
darcy@cs.berkeley.edu

Return to the comp.compilers page.
Search the comp.compilers archives again.