Torbjorn Granlund tg at
Tue Apr 27 23:32:49 CEST 2004

Kevin Ryde <user42 at> writes:

  > Perhaps
  > one could hope for the best and reject the compiler when it + the
  > assembler generates the critical object code.
  Past gcc never used anything but ebx for the GOT, so never tickled
  this bug/misfeature of gas, as far as I know.  Perhaps it still uses
  ebx exclusively unless the optimization is high enough.  Do you know
  if that's the case?

Sorry, I don't know.

(I checked what gcc 3.4.0 does with the testcase, and it does not
use the shorter eax form, and thus does not trigger the bug for
that particular test case.  I have not checked the compiler
sources, so no real conclusons can be drawn.)


More information about the gmp-devel mailing list