From: Philipp K. K. <pk...@sp...> - 2013-12-20 17:46:38
|
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Am 19.12.2013 15:01, schrieb Maarten Brock: > Hi Philipp, > > I only payed some attention to absolute code size and execution > speed of the generated asm, not relative to function call overhead. > The main goal was generating correct code. > > But looking into this some more, I suggest to keep it in. The extra > size is not too big: +6 bytes. When the function appears only once > linking in the library will give larger code due to function > overhead (at least 9 bytes). I also expect most strncpy instances > use a small length where the overhead becomes significant in the > execution time. And the currently generated library implementation > is really not optimal. > > Btw. I was surprised that it passed > gcc-torture-execute-string-opt-5.c as it tests for this bug. But > apparently there it doesn't use the built-in version. > > Thanks for asking before ditching my contribution anyway. I'll > leave the final decision to you. > > Maarten Let's leave it built-in for now. But when we change the default calling convention to something more efficient, this issue should be reconsidered. Philipp -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.15 (GNU/Linux) Comment: Using GnuPG with Icedove - http://www.enigmail.net/ iEYEARECAAYFAlK0gnIACgkQbtUV+xsoLpol1QCfXrhp7PtMiS9E7XXdnsIFYmXR A9MAnik6rtJGnFTqWAFOIy4T8Kn/z6U3 =mr8e -----END PGP SIGNATURE----- |