0
$\begingroup$

I've been reading books which refers to vector update operations of the form: y := y + ax, where y and x are vector variables and a is a scalar as SAXPY. I understand ax plus y part, but why "single precision"?

I'm seeing it in the context of complexity estimates of iterative linear solvers, where I assume that most variables (the vectors and matrices) are double precision, not single. But vector updates are still described as SAXPY.

Thanks.

$\endgroup$
0
$\begingroup$

This is before my time, but the old Cray Single Precision was 64-bit, and so the name stuck for supercomputer users who got used to it. "Sax pee" is just more natural to say than "Dax pee", besides.

$\endgroup$
  • $\begingroup$ Wow, really? To me, SAXPY really just means 32-bit float or REAL*4 arithmetic, and DAXPY means 64-bit double or REAL*8. In fact, I hear people use the term DAXPY far more often than SAXPY these days, and that makes sense because nobody still uses single precision. $\endgroup$ – Wolfgang Bangerth Apr 11 '16 at 2:35
  • 1
    $\begingroup$ Me, too, until I remembered that Cray had it's own "float" type. We're too young, @WolfgangBangerth! $\endgroup$ – Bill Barth Apr 11 '16 at 3:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.