[libcxx-commits] [PATCH] D59178: [libc++] Speedup to_string and to_wstring for integers using stack buffer and SSO

Ed Schouten via Phabricator via libcxx-commits libcxx-commits at lists.llvm.org
Thu Mar 14 04:45:30 PDT 2019

ed added inline comments.

Comment at: libcxx/src/string.cpp:422
-    return as_string(snprintf, initial_string<string, int>()(), "%d", val);
+    return i_to_string<string>(snprintf, "%d", val);
ivafanas wrote:
> ivafanas wrote:
> > ed wrote:
> > > Considering that performance is of the essence and that we only do simple conversions (without leading whitespace, zeroes, etc.), would it make sense to handroll this for the integer cases? This is likely a lot faster than using `snprintf()`.
> > I will try to implement that and estimate performance gain.
> Ed, I have locally implemented simple conversion and it gives significant performance boost (about 20x times when SSO plays).
> There is one more question I'm looking for answer for days.
> According to `std::to_string` documentation result for integer types must be equal to `sprintf` call with `%i` parameter (`%l` for long etc etc)
> https://en.cppreference.com/w/cpp/string/basic_string/to_string
> According to `to_chars` documentation `to_chars` is the only locale-independent formatter
> https://en.cppreference.com/w/cpp/utility/to_chars
> According to this note clang aims to work with the system C library
> http://clang-developers.42468.n3.nabble.com/C-Standard-Library-td2127533.html
> I've checked `glibc`, `ditelibc` and `musl` standard C lib implementations, and they provides the expected result for `sprintf(s, "%i", x)` calls without locale checks for thousands separator: `sprintf(s, "%i", 100500) --> 100500`
> I've tested msvc `sprintf(s, "%i", x)` calls with the different locales, it also prints integers without thousands separator.
> So, on the one side, performance boost is really impressive, on the other side there is no guarantee that `sprintf(s, "%i", x)` will always ignore thousands separator.
> I want to ask your opinion as much more experienced compiler developer.
> Should we use fast simple conversion or leave slow `sprintf` calls?
> So, on the one side, performance boost is really impressive, on the other side there is no guarantee that `sprintf(s, "%i", x)` will always ignore thousands separator.

Even though this is POSIX and not ISO C, it seems that thousands separators are only added to integers when `%'d` is used (notice the apostrophe). See this page:

http://pubs.opengroup.org/onlinepubs/9699919799/functions/fprintf.html <- search for 'apostrophe'

ISO C also says the following about `printf()`'s 'd' conversion specifier:

> The int argument is converted to signed decimal in the style [−]dddd. The precision specifies the minimum number of digits to appear; if the value being converted can be represented in fewer digits, it is expanded with leading zeros. The default precision is 1. The result of converting a zero value with a precision of zero is no characters.

Notice that the description doesn't mention anything about thousand separators.

In other words, I think it's fine to bypass sprintf() for integer literals. For floating point literals it is a different story, because formatting those is locale dependent.

  rCXX libc++



More information about the libcxx-commits mailing list