This is the mail archive of the
libc-hacker@cygnus.com
mailing list for the glibc project.
glibc optimizations and x86 variants
- To: libc-alpha@cygnus.com
- Subject: glibc optimizations and x86 variants
- From: "Joseph S. Myers" <jsm28@cam.ac.uk>
- Date: Tue, 19 Jan 1999 00:10:12 +0000 (GMT)
Some of the optimized assembler sources in glibc (2.0.110) for x86
variants seem to use non-386 instructions. Does glibc substitute 386
versions at runtime if needed? (The INSTALL file says that a library
compiled for a Pentium will still work on a 386.)
Some of the inline assembler in the headers may have the same problem.
egcs defines the relevant preprocessor variables based on the -mcpu value
rather than the -march value, so this could break the expectation that
unless you use -march the binary will work on any x86. Should egcs be
defining separate macros for the -mcpu and -march values (and glibc using
them in the headers to determine what assembler versions to use)?
--
Joseph S. Myers
jsm28@cam.ac.uk