Re: question about registers that are set on entry to a function.

[Jim Wilson <wilson at specifix dot com>]

Kenneth Zadeck wrote:
> I have one question about EH_USES on the ia-64 (the only place it is 
> used). 

I know this is a bit late, but I didn't see any answer to this on this 
list, and figured you might still need one.

The IA-64 port uses EH_USES for registers that hold frame state.  We can 
optionally store some frame state in local registers in the current 
register window.  These are obviously live at the epilogue because the 
epilogue code uses them.  These are also live at EH edges, but this 
isn't obvious because there are no explicit uses in the rtl.  The uses 
are hidden inside the unwind library.  So EH_USES is used to make sure 
these regs are live at entry to an EH edge.  EH_USES serves the same 
purpose as EPILOGUE_USES, except one is for EH edges and one is for 
epilogues.

Why only IA-64 needs this, offhand, I don't know.  Maybe it has to do 
with the IA-64 unwinder that only IA-64 uses.  Or maybe it has something 
to do with how IA-64 register windows work.  I don't care enough to take 
a closer look.

These registers do not exist before the prologue/epilogue code is 
emitted, and hence they do not exist before reload.

These are normal allocatable registers.  If there are free registers, 
then we use some of them for storing frame state.  If there aren't any 
free registers, then the frame state goes on the stack which is the 
normal place for it.  So there is no latent register allocation problem 
here.

We only use 5 of 128 integer registers this way, so relatively speaking, 
this isn't a lot of registers.  Since these are local register windowed 
regs, they are pretty cheap.
--
Jim Wilson, GNU Tools Support, http://www.specifix.com