Showing results for 
Search instead for 
Did you mean: 
New Contributor I

floating point precision

Using Intel's i5 and i7 hardware, would a double precision ieee format be better utilized with a REAL( kind = 10 ) rather than a REAL ( kind = 8 ) format. The GNU gfortran uses kind=10 but our Intel v16 compiler barfs at kind=10. Don't you get segment faults with arrays of kind=8 but not with kind=10? What is the reason for not using kind=10 or kind=20?



Brooks V

0 Kudos
1 Reply

The GNU KIND=10 is using the

The GNU KIND=10 is using the ancient x87 80-bit floating registers. The modern (since Pentium 4) SSE floating point instructions don't support that, and neither does Intel Fortran.

There is no 10-byte memory format for floating. I thing gfortran uses 16 bytes to store these.

Retired 12/31/2016
0 Kudos