You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Because all BLAS1 routines are bandwidth limited, there is essentially no value to using them over straightforward implementations by qualified programmers.
ZDOT_RETURN is a workaround for flaws in BLAS1 implementations. I think it would be prudent to just implement dot products in BAGEL directly.
The text was updated successfully, but these errors were encountered:
Sounds like a great idea (as long as the code is competitive with MKL; "qualified" is the key). The code should be used only when, e.g., MKL is not present.
It's much easier than you think :-)
Intel compiler with naive C or Fortran loops should beat MKL for BLAS1 because autovectorization will be sufficient and user code is easier to inline and/or fuse.
I'll perform some experiments and share full details.
Note also that C++ loops trivially avoids any incompatibility between std::complex and Fortran complex double, should any exist (I can't remember what ISO says about this).
Because all BLAS1 routines are bandwidth limited, there is essentially no value to using them over straightforward implementations by qualified programmers.
ZDOT_RETURN
is a workaround for flaws in BLAS1 implementations. I think it would be prudent to just implement dot products in BAGEL directly.The text was updated successfully, but these errors were encountered: