Torbjorn Granlund tg at
Tue Apr 27 23:54:33 CEST 2004

Kevin Ryde <user42 at> writes:

  We can detect the bad gas by grepping an object file, and can spot the
  bad gcc from the sample case.  But one gcc sample which comes out ok
  is not much of a test, we'd really want a good reason to think it'll
  never arise.
Not sure what you mean.

We don't want to detect and reject the bad gas, since that'd be
too broad, right?

There is no evil in gcc exposed by this, as far as I understand.
(I am not familiar with the ELF x86 ABI, but it would surprise me
if it mandates a register to use for a PIC function's local use.
Even if it does that, I would argue that gas should assemble what
the compiler generates.)

This leaves us to detect the situation that leads to triggering
the bad gas.  Even it is not gcc's fault, our only option is to
act as it it were; we need to change optimization level to avoid
triggering the gas bug.


More information about the gmp-devel mailing list