Message195571
> Instead of only a second lookup, could you try for example 4 lookup
> and align j to fit in a cache line?
Accessing 4 entries per probe is a tempting line of development, but will be subject to diminishing returns (second and third collisions aren't frequent).
I like the idea of aligning j to fit in a cache line, but the computation would need to be cheap and portable (standard C with no pointer tricks that rely on non-guaranteed behavior).
Have you had a chance to run the benchmarks on your machine? I'm curious how this works out on other processors and with other compilers. |
|
Date |
User |
Action |
Args |
2013-08-18 16:54:55 | rhettinger | set | recipients:
+ rhettinger, vstinner, christian.heimes, serhiy.storchaka |
2013-08-18 16:54:55 | rhettinger | set | messageid: <1376844895.13.0.85981629311.issue18771@psf.upfronthosting.co.za> |
2013-08-18 16:54:55 | rhettinger | link | issue18771 messages |
2013-08-18 16:54:54 | rhettinger | create | |
|