[Patch]: Fix pb in libcall_dead_p

Richard Henderson rth@cygnus.com
Tue Jun 27 14:55:00 GMT 2000


On Wed, Jun 28, 2000 at 01:51:49AM +0200, Stephane Carrez wrote:
> This is what is generated and I've observed the same for sparc.
> For sparc, the first REG_RETVAL is removed by the life analysis pass.

Got a test case?  This _really_ shouldn't be happening.

> Because I don't have 32-bit registers and I've implemented 32-bit moves
> by using a 16-bit scratch register.  The result is much more efficient than
> spliting the move into pieces.

Except that one can always split a double-word move after reload without
requiring a scratch register.  


r~


More information about the Gcc-patches mailing list