Branch: refs/heads/blead
Home: https://github.com/Perl/perl5
Commit: e2b6b05ee153b1f2dad24d3b736a6ea974780183
https://github.com/Perl/perl5/commit/e2b6b05ee153b1f2dad24d3b736a6ea974780183
Author: Karl Williamson <k...@cpan.org>
Date: 2023-12-20 (Wed, 20 Dec 2023)
Changed paths:
M inline.h
Log Message:
-----------
Fallback to use BitScan(Forward|Reverse) on MingW
Perl has functions that return the positions of the least or most
significant one bit in a word. gcc has builtins that do these specific
operations, and they get compiled to a single instruction on many
hardware platforms. Windows instead has various BitScan functions that
do the same thing.
MingW has both gcc and Windows, so it should have access to both the
builtins and to the BitScan functions, but prior to this commit, if we
didn't find gcc builtins on MingW, we wouldn't bother to look for the
BitScan functions.
Reading the code caused me to pause to try to figure out why not use
these BitScan functions, and asked on #irc. The answer from @xenu was
that on MingW (as opposed to MSVC) they're emulated, and not as
efficient as the gcc builtins.
Then I looked at the code again, and realized we don't try them unless
the gcc builtins aren't available, unlikely these days. But we should
do something so that people who read this code in the future don't also
end up spending time unnecessarily wondering why we treat MingW
specially.
We could add a comment, but it seems to me better to just not make MingW
a special case. It's likely that the MingW emulation of the BitScan
functions is better than our further fallback, so I see that as better
and cleaner than special casing MingW.
