Folly VS abseil-cpp

https://github.com/facebook/folly/commit/b1391e1c57be71c1e2a...

https://github.com/facebook/folly/pull/2153

To set a HP on Linux, Folly just does a relaxed load of the src pointer, release store of the HP, compiler-only barrier, and acquire load. (This prevents the compiler from reordering the 2nd load before the store, right? But to my understanding does not prevent a hypothetical CPU reordering of the 2nd load before the store, which seems potentially problematic!)

Then on the GC/reclaim side of things, after protected object pointers are stored, it does a more expensive barrier[0] before acquire-loading the HPs.

I'll admit, I am not confident I understand why this works. I mean, even on x86, loads can be reordered before earlier program-order stores. So it seems like the 2nd check on the protection side could be ineffective. (The non-Linux portable version just uses an atomic_thread_fence SeqCst on both sides, which seems more obviously correct.) And if they don't need the 2nd load on Linux, I'm unclear on why they do it.

[0]: https://github.com/facebook/folly/blob/main/folly/synchroniz...

(This uses either mprotect to force a TLB flush in process-relevant CPUs, or the newer Linux membarrier syscall if available.)

folly provides functions to resize std::string & std::vector without initialization [0].

[0] https://github.com/facebook/folly/blob/3c8829785e3ce86cb821c...

My implementation was quite similar to Boost's spsc_queue and Facebook's folly/ProducerConsumerQueue.h.

> How is that relevant?

Roaring bitmaps and similar data structures get their speed from decoding together consecutive groups of elements, so if you do sequential decoding or decode a large fraction of the list you get excellent performance.

EF instead excels at random skipping, so if you visit a small fraction of the list you generally get better performance. This is why it works so well for inverted indexes, as generally the queries are very selective (otherwise why do you need an index?) and if you have good intersection algorithms you can skip a large fraction of documents.

I didn't follow the rest of your comment, select is what EF is good at, every other data structure needs a lot more scanning once you land on the right chunk. With BMI2 you can also use the PDEP instruction to accelerate the final select on a 64-bit block: https://github.com/facebook/folly/blob/main/folly/experiment...

C++ with folly's SCOPE_EXIT {} construct:

https://github.com/facebook/folly/blob/main/folly/ScopeGuard...

Seems like github issues taking a long time to get any response: https://github.com/facebook/folly

Can't speak for abseil and tbb, but in folly there are a few solutions for the common problem of sharing state between a writer that updates it very infrequently and concurrent readers that read it very frequently (typical use case is configs).

The most performant solutions are RCU (https://github.com/facebook/folly/blob/main/folly/synchroniz...) and hazard pointers (https://github.com/facebook/folly/blob/main/folly/synchroniz...), but they're not quite as easy to use as a shared_ptr [1].

Then there is simil-shared_ptr implemented with thread-local counters (https://github.com/facebook/folly/blob/main/folly/experiment...).

If you absolutely need a std::shared_ptr (which can be the case if you're working with pre-existing interfaces) there is CoreCachedSharedPtr (https://github.com/facebook/folly/blob/main/folly/concurrenc...), which uses an aliasing trick to transparently maintain per-core reference counts, and scales linearly, but it works only when acquiring the shared_ptr, any subsequent copies of that would still cause contention if passed around in threads.

[1] Google has a proposal to make a smart pointer based on RCU/hazptr, but I'm not a fan of it because generally RCU/hazptr guards need to be released in the same thread that acquired them, and hiding them in a freely movable object looks like a recipe for disaster to me, especially if paired with coroutines https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2020/p05...

Not sure if it's still the case but about 6 years ago Facebook's folly C++ library was something I'd point to for my junior engineers to get a sense of "good" C++ https://github.com/facebook/folly

Yeah, it's nice! And Abseil does it, IFF you use LLVM libc++.

https://github.com/abseil/abseil-cpp/blob/master/absl/string...

The standard adopted it as resize_and_overwrite. Which I think is a little clunky.

This may be confusing to those familiar with Google's libraries. The baseline is the Go BTree, which I personally never heard of until just now, not the C++ absl::btree_set. The benchmarks aren't directly comparable, but the C++ version also comes with good microbenchmark coverage.

https://github.com/google/btree

https://github.com/abseil/abseil-cpp/blob/master/absl/contai...

An implementation of call_once that accommodates callbacks that throw: https://github.com/abseil/abseil-cpp/blob/master/absl/base/c...

I wouldn't say it's that cryptic. It's just a few bitwise rotations/shifts/xor operations.

You can see hashing optimizations as well https://www.deepmind.com/blog/alphadev-discovers-faster-sort..., https://github.com/abseil/abseil-cpp/commit/74eee2aff683cc7d...

I was one of the members who reviewed expertly what has been done both in sorting and hashing. Overall it's more about assembly, finding missed compiler optimizations and balancing between correctness and distribution (in hashing in particular).

It was not revolutionary in a sense it hasn't found completely new approaches but converged to something incomprehensible for humans but relatively good for performance which proves the point that optimal programs are very inhuman.

Note that for instructions in sorting, removing them does not always lead to better performance, for example, instructions can run in parallel and the effect can be less profound. Benchmarks can lie and compiler could do something differently when recompiling the sort3 function which was changed. There was some evidence that the effect can come from the other side.

For hashing it was even funnier, very small strings up to 64 bit already used 3 instructions like add some constant -> multiply 64x64 -> xor upper/lower. For bigger ones the question becomes more complicated, that's why 9-16 was a better spot and it simplified from 2 multiplications to just one and a rotation. Distribution on real workloads was good, it almost passed smhasher and we decided it was good enough to try out in prod. We did not rollback as you can see from abseil :)

But even given all that, it was fascinating to watch how this system was searching and was able to find particular programs can be further simplified. Kudos to everyone involved, it's a great incremental change that can bring more results in the future.

Check out https://abseil.io. It offers absl::optional, which is a backport of std::optional.