Hi,
On 03/15/2013 04:59 PM, Keith Lofstrom wrote:
... Are double and float synonyms for the same
double precision representation?
From http://en.wikipedia.org/wiki/C_data_types:
"The actual size of floating point types also varies by implementation.
The only guarantee is that the long double is not smaller than double,
which is not smaller than float. Usually, 32-bit and 64-bit IEEE 754
floating point formats are used, if supported by hardware."
The other issue could be that most calculations might be done using
register values with precision far exceeding the precision of the
variable in memory. That would complete 'hide' the expected effect of
the float or double on the calculations. Use -ffloat-store to avoid this
for your tests.
HTH,
Matthias
