ARM Patch: allow for better CSE of HImode and QImode loads

[Richard Earnshaw]

The following patch allows for better CSE of HImode and QImode memory load 
operations by converting these into zero-extend operations that load a 
full word and then copying a subreg of the result to the target.  For 
HImode we only do this if targeting a v4 processor, since otherwise we 
don't have a zero_extend operation.  In either case we only do this when 
optimizing and when we can create new pseudos.

2002-07-22  Richard Earnshaw  <rearnsha@arm.com>

        * arm.md (movqi): If optimizing and we can create pseudos, use
        a ZERO_EXTEND to load from memory, then copy the result into the
        target.
        (movhi): Likewise, but only for ARMv4.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: gcc-ze.patch
Type: text/x-patch
Size: 1658 bytes
Desc: gcc-ze.patch
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20020722/3bfae63d/attachment.bin>