Mario Rugiero
Mario Rugiero
There are performance benefits on using aligned data AFAIR, tho. Reads and writes of aligned data should hit the bus less often. Same way, they should invalidate the cache less...
But if you do `memcpy` you lose part of the benefits of using a frozen format in the first place, as a big one is the ability to simply `mmap`...
Off-topic: there's a case where it really is more important to align to cache, when atomics are involved. Atomic operations invalidate a cache line on all processors, so you want...
Oh, I thought `memcpy`ing the whole contents at once.
Coming back to this. I think an option would be to: 1. Convert from portable to frozen in a buffer, without creating an intermediate bitmap; 2. Create a view with...
A `Validate` function would be best. If the user trusts the input (they generated it, for example) they can avoid paying the price, but if something comes down the wire...
I think this is either a duplicate of #38 or it suffices to use several bitmaps and each thread process a few of them. You can also use a single...
Sorry, I thought this was CRoaring 🤦 I know some ops are parallel here. I think some more could be parallelized tho.
Is there a way to do this in two steps? It would be interesting to first hash only the keys, which should be faster, and only if that hash matches...
Of course. What I meant is if there's a way to use a several-checksums approach for keys in maps and the like. It wouldn't help a lot now that I...