Re: [PATCH] register CALL_INSN_FUNCTION_USAGE in find_all_hard_reg_sets

[Tom de Vries <Tom_deVries at mentor dot com>]

On 15-01-14 19:40, Richard Sandiford wrote:
> Tom de Vries <Tom_deVries@mentor.com> writes:
> > This patch adds scanning of clobbers in CALL_INSN_FUNCTION_USAGE to
> > find_all_hard_reg_sets.
> >
> > For MIPS, calls are split at some point. After the split, one of the
> > resulting insns may clobber $6. But before the split, that's not
> > explicit in the rtl representation of the unsplit call.
> >
> > For -fuse-caller-save, that's a problem, and Richard S. suggested to
> > add the clobber of $6 to the CALL_INSN_FUNCTION_USAGE of the unsplit
> > call.
>
> Maybe here's a good place to raise this, but I was hoping all targets
> could use CALL_INSN_FUNCTION_USAGE  and we'd be able to drop the new
> hook for -fuse-caller-save.  That seems better than adding another
> place that needs to be checked when figuring out what a call clobbers.

Hi Richard,

There are the following sets of registers involved:
 * CALL_REALLY_USED_REGS, which is the set normally clobbered by called
   functions (or PLTs, resolvers, or any other piece of code that runs during a
   function call)
* The set of registers discovered by the use-caller-save optimization, which
  takes into account only the registers clobbered by the called function itself
* The set of registers which are clobbered during a call by things like the plt
  - these are not picked up by the use-caller-save optimization. We need the
  hook to inform the compiler about these registers
* And finally, registers clobbered in the caller itself during a sequence of
  instructions implementing a function call. On mips, that's R6, which may be
  clobbered by the call. Normally that doesn't need mentioning in the RTL since
  it's a call_used_reg, but since use-caller-save might discover a set of
  registers for the called function that does not include R6, it becomes
  important to record this clobber explicitly. It could be represented in the
  RTL by a clobber on the insn, or a clobber in C_I_F_U. Or it could just be
  part of the registers returned by the hook - but that was previously deemed
  not acceptable (and it doesn't match the description of the hook).

> In other words, I didn't want MIPS to be different from the other targets.

MIPS just needs some special handling for the unsplit call. There might be other 
targets which need similar tweaks.

> If we decide to keep the hook then MIPS should use it too.

I think the current proposal is good:
- use the hook to model registers clobbered during a call
- use CALL_INSN_FUNCTION_USAGE to model the r6 clobber for the unsplit MIPS
  call_insn

Thanks,
- Tom

> Thanks,
> Richard