This sounds like he's worrying a lot about data caches and not at all about instruction caches which may look good in microbenchmarks but is a bad direction to go for a standard lib, especially when it gets instantiated for 100s of types. :)
-
-
-
Data cache-friendlier data structures would be nice (although the top 2 ones I see used, std::vector and std::unordered_map, are not bad on that axis anyway), but the STL already instantiates worrying amounts of code for its containers and algs...
- 3 more replies
New conversation -
-
-
I worked at HP Labs on a major C++ product at the time and used to buy our home line of PCs for the team from CompUSA as we couldn’t get our own Vectra line internally. Plenty of HP-UX hardware though.
-
I was also partially responsible for porting and we had money for decent enough Solaris and AIX kit.
- 2 more replies
New conversation -
-
-
Most cpp projects reimpl large parts of the stdlib - strings, sets, maps (for small), different bucketing for larger maps, different hashing for heavy reads, etc.... Given enough time and engs on it.
-
It seems like almost every big software development company (with C++ projects) implements it's own version of STL...
End of conversation
New conversation -
-
-
Do you have a link?
New conversation -
-
-
BTW google_dense_hashmap is super fast compared to libc++/stdc++ STL map and unordered_map. dense_hashmap is a bucket-less scanning hashtable with tombstones for deleted entries. It doesn’t pointer chase like a bucketed hashtable so is cache friendly.https://github.com/sparsehash/sparsehash …
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
IIRC HP at the time was also building its own CORBA ORB as it didn’t buy Orbix.
-
Everybody was building an Orb after not buying Orbix (raises hand)
- 2 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.