2048 | A small clone of 1024 ( https

You can try like this in config.xml under android platform -

The original commit message sheds some light on this. In short, libc++ originally used just prime numbers. This commit introduces an optimization for power-of-2 numbers in case the user explicitly requests them.

Also note that the new function __constrain_hash() in that commit checks if it is a power-of-2 and then does not use the modulo operation. According to the commit message, the cost for the additional check if the input is a power-of-2 number is outweighed by not using a modulo operation. So even if you do not know the information at compile time, you can get a performance boost.

Quote of the commit message:

This commit establishes a new bucket_count policy in the unordered containers: The policy now allows a power-of-2 number of buckets to be requested (and that request honored) by the client. And if the number of buckets is set to a power of 2, then the constraint of the hash to the number of buckets uses & instead of %. If the client does not specify a number of buckets, then the policy remains unchanged: a prime number of buckets is selected. The growth policy is that the number of buckets is roughly doubled when needed. While growing, either the prime, or the power-of-2 strategy will be preserved. There is a small run time cost for putting in this switch. For very cheap hash functions, e.g. identity for int, the cost can be as high as 18%. However with more typical use cases, e.g. strings, the cost is in the noise level. I've measured cases with very cheap hash functions (int) that using a power-of-2 number of buckets can make look up about twice as fast. However I've also noted that a power-of-2 number of buckets is more susceptible to accidental catastrophic collisions. Though I've also noted that accidental catastrophic collisions are also possible when using a prime number of buckets (but seems far less likely). In short, this patch adds an extra tuning knob for those clients trying to get the last bit of performance squeezed out of their hash containers. Casual users of the hash containers will not notice the introduction of this tuning knob. Those clients who swear by power-of-2 hash containers can now opt-in to that strategy. Clients who prefer a prime number of buckets can continue as they have.

Unfortunately I can't provide you with a fix, but I've found a workaround for that exact same problem (company-hosted bitbucket resulting in exact same error). I also don't know exactly why the problem occurs, but my best guess would be that the libressl library shipped with Monterey has some sort of problem with specific (?TLSv1.3) certs. This guess is because the brew-installed openssl v1.1 and v3 don't throw that error when executed with /opt/homebrew/opt/openssl/bin/openssl s_client -connect ...:443

To get around that error, I've built git from source built against different openssl and curl implementations:

1. install autoconf, openssl and curl with brew (I think you can select the openssl lib you like, i.e. v1.1 or v3, I chose v3)
2. clone git version you like, i.e. git clone --branch v2.33.1 https://github.com/git/git.git
3. cd git
4. make configure (that is why autoconf is needed)
5. execute LDFLAGS="-L/opt/homebrew/opt/openssl@3/lib -L/opt/homebrew/opt/curl/lib" CPPFLAGS="-I/opt/homebrew/opt/openssl@3/include -I/opt/homebrew/opt/curl/include" ./configure --prefix=$HOME/git (here LDFLAGS and CPPFLAGS include the libs git will be built against, the right flags are emitted by brew on install success of curl and openssl; --prefix is the install directory of git, defaults to /usr/local but can be changed)
6. make install
7. ensure to add the install directory's subfolder /bin to the front of your $PATH to "override" the default git shipped by Monterey
8. restart terminal
9. check that git version shows the new version

This should help for now, but as I already said, this is only a workaround, hopefully Apple fixes their libressl fork ASAP.

In C++, the standard way to make an associative table is std::map:

Here's one portable way (untested):

The slowdown happens because of the way OFFSET works: it fetches all the data and only then drops the part before the offset. For your case it means the grouping will happen not only for the current page, but for all the previous pages too.

The standard trick to fix this kind of problem is to use Keyset Pagination. When fetching the page, you need to remember its last parent. Then in order to fetch the next page, you use your query with the

Imo, it's a bug in Paramiko. It does not handle correctly absence of server-sig-algs extension on the server side.

Try disabling rsa-sha2-* on Paramiko side altogether:

https://godbolt.org/z/PT6T1691W it seems that -O2 -funroll-loops does the trick, apparently that option needs to be on for the pragma to tell GCC how much to unroll. (Update: Or at least makes it have some effect. See comments, this doesn't seem to be a complete answer yet.)

(-funroll-loops is not on by default unless you use -fprofile-use, after doing a -fprofile-generate run and running the program with representative input. It used to be on by default at -O3 a long time ago, but code bloat I-cache pressure usually made that worse for loops that aren't hot. This leads to bass-ackwards situations where the loop where GCC spends most of its time is a few instructions long with SIMD, but the fully-unrolled scalar prologue / epilogue are 10x the number of instructions, especially with wider vectors. Even with AVX-512, GCC usually just uses scalar for odd numbers of elements, not creating a mask. :/)

Fully unrolling loops is something GCC will do even at -O2, at least for very small trip-counts. (e.g. up to 3 for an int array p[i] += 1;, with -O2 -fno-tree-vectorize). https://godbolt.org/z/P5rvjYj1b

Fully-unrolling larger loops or higher trip counts (when the static code size would increase from doing so, perhaps) is not on by default at -O2 it seems. (GCC calls this peeling a loop in their tuning options/parameters, i.e. peeling all the iterations out of the loop so it goes away. -fpeel-loops is on with -O3, but not -O2. Since GCC11, -fverbose-asm no longer prints a list of optimization options enabled as asm comments.)

And BTW, it seems auto-vectorization is on by default at -O2 now in GCC trunk. Previously it was only on at -O3, so that's interesting.

Reversing the bits achieves the expected effect in 1D, you could combine this shuffling technique with another one where you get the x and y coordinates be selecting the even, resp. odd, bits of the resulting number. Combining both methods in a single shuffle is highly desirable to avoid costly bit twiddling operations.

You could also use Gray Codes to shuffle values with n significant bits into a pseudo random order. Here is a trivial function to produce gray codes: