Mail Archives: pgcc/1998/09/08/15:39:54
It has been recently reported that pgcc v1.1a has problems (can generate
incorrect code) with the fast-math and runtime-lift-stores switches.
Under v1.03a I saw problems with the risc-const switch when building the
Linux 2.1.x kernel, but all other -O4 switches worked fine on the same
code. My understanding is that the problem is due to the small register
set on the i386 CPUs, so it will not be seen on all the platforms
supported by the egcs that pgcc is based on. Because pgcc is oriented to
the Pentium+ CPUs, though, this is a problem that will be seen more
frequently by pgcc users.
I suggest that known problems with pgcc be noted on the FAQ posted on the
pgcc Web page. As things are now, users who do not subscribe to this list
will believe that all these options are safe because they are described
that way under the Optimizations section of the FAQ.
Thank you.
- Raw text -