Re: Proposal for handling large numbers of registers

[Stan Shebs <shebs@cygnus.com>]

   Date: Sat, 07 Aug 1999 13:26:45 +1200 (NZST)
   From: Michael Hayes <m.hayes@elec.canterbury.ac.nz>

   J.T. Conklin writes:
    > I'm not sure that a compilers notion of register classes make sense
    > for debugging and vice versa.  For debugging, broad categories like
    > "integer registers", "floating point registers", "system registers"
    > seem to be most appropriate; while compilers think of classes like
    > "registers that can be used for indexing", "special register used
    > for integer divide instructions", etc.

   For most targets I have looked at, these other classes you mention are
   usually sub-classes of the "integer registers", "fp registers", etc.
   However, I can see from most user's perspectives that they will not be
   interested in these special sub-classes.

Plus which compilers don't usually have any information about special
or system registers, not least because that part of the register set
may vary wildly among members of a family.

Certainly it would be cool to have some sort of abstract architectural
description that could be crunched into compiler and debugger support
code, but we're a ways away from it (it's pretty hard to do too; I
tackled that one in grad school ten years ago, with limited success).

								Stan