Improve compilation performance
Opened this issue · 0 comments
gbeaty commented
I explored a number of different implementations for scunits, and the current one is the fastest. However it seems like it could still get faster. In the ~40s it takes to compile the benchmark code, time is spend in the following tasks:
- EList multiplication: 13s
- Integer addition: 4s (included in the above)
- Figuring out if the right-hand side of a
Scalar
operation hasDims
which are<: DimsOf[quants]
: 15s
3 seems like the lowest hanging fruit.