Bruno Haible <[EMAIL PROTECTED]> wrote: >> I have pushed changes for md2, md4 and md5. > > Here's an additional proposed refinement. With it, gcc generates much better > code (on x86: a single instruction instead of a function call of a function > with 5 instructions). > > 2008-01-31 Bruno Haible <[EMAIL PROTECTED]> > > * lib/md4.c (set_uint32): Mark as inline. > * lib/md5.c (set_uint32): Likewise. > * lib/sha1.c (set_uint32): Likewise. > * m4/md4.m4 (gl_MD4): Require AC_C_INLINE. > * m4/md5.m4 (gl_MD5): Likewise. > * m4/sha1.m4 (gl_SHA1): Likewise.
Good! I had prepared a similar patch earlier today, but hadn't measured anything, so, after Simon's commit of the non-inline versions got in first, I didn't plan to pursue it. So at least for md5, that's fine with me.
