This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: powerpc & unaligned block moves with fp registers
- To: David Edelsohn <dje at watson dot ibm dot com>
- Subject: Re: powerpc & unaligned block moves with fp registers
- From: Mark Mitchell <mark at codesourcery dot com>
- Date: Fri, 09 Nov 2001 12:19:23 -0800
- cc: Richard Kenner <kenner at vlsi1 dot ultra dot nyu dot edu>, "gcc at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>
--On Friday, November 09, 2001 03:01:13 PM -0500 David Edelsohn
<dje@watson.ibm.com> wrote:
>>>>>> Mark Mitchell writes:
>
> Mark> This would certainly work, but I think it would be nicer to just
> always Mark> use BLKmode and have the back ends call back to a common
> routine if they Mark> want command handling. I would like the middle
> end/back end interface Mark> to be small, and then provide a library of
> common routines that back Mark> ends could use.
>
> As Richard Kenner pointed out, we want the structure to be an
> integer mode so that the structure is operated on in a register. That
> requires that some of the load/store operations be in that mode and any
> heuristic to use BLKmode moves needs to choose when to use emit_block_move
> and when to use emit_move_insn.
Right.
> If we don't want to use SLOW_UNALIGNED_ACCESS macro and "always"
> want to punt to BLKmode, it sounds like the criteria is memory-to-memory
> moves, as opposed to memory-to-register moves. In other words, when
Right.
> expand_assignment() sees that the LHS and RHS are memory, it should
> convert the mode to BLKmode, or something like that. Comments?
That is my fantasy of how things should work. Note, however, that
I don't really know how hard it would be to make that work, changing
all back ends, etc.
--
Mark Mitchell mark@codesourcery.com
CodeSourcery, LLC http://www.codesourcery.com