How much parallel speedup is there on multicore CPUs?

Twinside / Rasterific

A drawing engine in Haskell

BSD 3-Clause "New" or "Revised" License

140 stars 11 forks source link

How much parallel speedup is there on multicore CPUs? #32

Closed stephano33 closed 7 years ago

stephano33 commented 7 years ago

How much parallel speedup is there? None? Or a bit? Or full support?

Also how much effort would it be to rewrite the library in with parallel support?

PS: very impressed with the Rasterific library! Fantastic! Also because there is not so much documentation on gezira that I found. Even more impressive.

Twinside commented 7 years ago

Right now, absolutely none. The biggest bottleneck is the edge sample: sorting

What could be parallelized easily:

Composition, after the sorting, all CoverageSpan can be applied to the canvas in parallel
Primitive decomposition, but the cost to recombine them before sorting could be high

Maybe sorting in parallel could be done, but tuning performance can be tricky. A word of warning though, lists are used lazily across the library, and changing allocation/storage can also be tricky.

stephano33 commented 7 years ago

I am closing this issue; it is more a feature request than a bug.