This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Problem in split_basic_block
- From: kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)
- To: wilson at tuliptree dot org
- Cc: gcc at gcc dot gnu dot org
- Date: Sun, 10 Aug 03 09:45:22 EDT
- Subject: Re: Problem in split_basic_block
addp4 does a 32-bit add and then copies the 2 most significant bits of
the 32-bit result (31 and 30) into the 2 most significant bits of the
target register (63 and 62). This gives you access to 4 segments
(regions).
Right, I know that.
You really should have some understanding of the IA-64 ISA if you are
trying to do a port.
I'm not the person most involved in the port, just the one that's
working on this particular bug. The people doing the port certainly have
that documentation.
See Vol 2 Section 4.1.9 32-bit Virtual Addressing. This documents how
to do zero-extend, sign-extend, or pointer-swizzling addressing via
addp4. All 3 are supported by the architecture.
OK, then there's indeed no efficiency issue after all. Good. HP obviously
gets to choose which one they want to use for VMS.