<addr_expr <realpart_expr <x>>> ?

[Diego Novillo]

On Sun, 2004-01-11 at 20:05, Richard Henderson wrote:

> Expanding complex operations just before SRA would probably do
> what we want.  It would eliminate enough dead code that we 
> wouldn't waste time.
> 
How about removing COMPLEX_TYPE, COMPLEX_CST and COMPLEX_EXPR from
tree.def?  Have the parser turn __complex__ into 'struct complex { <t>
real; <t> imag }'.

But in the end, anything that stops COMPLEX_* from getting into the
optimizers is fine with me.


Diego.