inliner: Don't emit copy stmts for empty type parameters [PR103989]

Hi!

The following patch avoids emitting a parameter copy statement when inlining
if the parameter has empty type.  E.g. the gimplifier does something similar
(except that it needs to evaluate side-effects if any, which isn't the case
here):
  /* For empty types only gimplify the left hand side and right hand
     side as statements and throw away the assignment.  Do this after
     gimplify_modify_expr_rhs so we handle TARGET_EXPRs of addressable
     types properly.  */
  if (is_empty_type (TREE_TYPE (*from_p))
      && !want_value
      /* Don't do this for calls that return addressable types, expand_call
         relies on those having a lhs.  */
      && !(TREE_ADDRESSABLE (TREE_TYPE (*from_p))
           && TREE_CODE (*from_p) == CALL_EXPR))
    {
      gimplify_stmt (from_p, pre_p);
      gimplify_stmt (to_p, pre_p);
      *expr_p = NULL_TREE;
      return GS_ALL_DONE;
    }
Unfortunately, this patch doesn't cure the uninit warnings in that PR,
but I think is desirable anyway.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2022-01-13  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/103989
	* tree-inline.c (setup_one_parameter): Don't copy parms with
	empty type.


	Jakub

On Thu, 13 Jan 2022, Jakub Jelinek wrote:

> Hi!
> 
> The following patch avoids emitting a parameter copy statement when inlining
> if the parameter has empty type.  E.g. the gimplifier does something similar
> (except that it needs to evaluate side-effects if any, which isn't the case
> here):
>   /* For empty types only gimplify the left hand side and right hand
>      side as statements and throw away the assignment.  Do this after
>      gimplify_modify_expr_rhs so we handle TARGET_EXPRs of addressable
>      types properly.  */
>   if (is_empty_type (TREE_TYPE (*from_p))
>       && !want_value
>       /* Don't do this for calls that return addressable types, expand_call
>          relies on those having a lhs.  */
>       && !(TREE_ADDRESSABLE (TREE_TYPE (*from_p))
>            && TREE_CODE (*from_p) == CALL_EXPR))
>     {
>       gimplify_stmt (from_p, pre_p);
>       gimplify_stmt (to_p, pre_p);
>       *expr_p = NULL_TREE;
>       return GS_ALL_DONE;
>     }
> Unfortunately, this patch doesn't cure the uninit warnings in that PR,
> but I think is desirable anyway.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

Hmm, but not emitting the initialization might cause even more such
warnings for the case where the passed in argument _is_ initialized
(or not visible as not, like when being a function parameter itself)?

Otherwise sure, it's the same what the gimplifier does.

I wonder if instead uninit warning should simply ignore uses of
"empty" typed variables?

OK.

Thanks,
Richard.

> 2022-01-13  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR tree-optimization/103989
> 	* tree-inline.c (setup_one_parameter): Don't copy parms with
> 	empty type.
> 
> --- gcc/tree-inline.c.jj	2022-01-11 23:11:23.422275652 +0100
> +++ gcc/tree-inline.c	2022-01-12 18:37:44.119950128 +0100
> @@ -3608,7 +3608,7 @@ setup_one_parameter (copy_body_data *id,
>  	      init_stmt = gimple_build_assign (def, rhs);
>  	    }
>  	}
> -      else
> +      else if (!is_empty_type (TREE_TYPE (var)))
>          init_stmt = gimple_build_assign (var, rhs);
>  
>        if (bb && init_stmt)
> 
> 	Jakub
> 
>

On Thu, Jan 13, 2022 at 10:54:15AM +0100, Richard Biener wrote:
> > The following patch avoids emitting a parameter copy statement when inlining
> > if the parameter has empty type.  E.g. the gimplifier does something similar
> > (except that it needs to evaluate side-effects if any, which isn't the case
> > here):
> >   /* For empty types only gimplify the left hand side and right hand
> >      side as statements and throw away the assignment.  Do this after
> >      gimplify_modify_expr_rhs so we handle TARGET_EXPRs of addressable
> >      types properly.  */
> >   if (is_empty_type (TREE_TYPE (*from_p))
> >       && !want_value
> >       /* Don't do this for calls that return addressable types, expand_call
> >          relies on those having a lhs.  */
> >       && !(TREE_ADDRESSABLE (TREE_TYPE (*from_p))
> >            && TREE_CODE (*from_p) == CALL_EXPR))
> >     {
> >       gimplify_stmt (from_p, pre_p);
> >       gimplify_stmt (to_p, pre_p);
> >       *expr_p = NULL_TREE;
> >       return GS_ALL_DONE;
> >     }
> > Unfortunately, this patch doesn't cure the uninit warnings in that PR,
> > but I think is desirable anyway.
> > 
> > Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
> 
> Hmm, but not emitting the initialization might cause even more such
> warnings for the case where the passed in argument _is_ initialized
> (or not visible as not, like when being a function parameter itself)?

Most of the time it won't be initialized either, but sure, there
can be some cases like when a larger struct is initialized with memset
and then we pass a field from that as an argument.

> Otherwise sure, it's the same what the gimplifier does.
> 
> I wonder if instead uninit warning should simply ignore uses of
> "empty" typed variables?

Apparently it does already:
  /* Avoid warning about empty types such as structs with no members.
     The first_field() test is important for C++ where the predicate
     alone isn't always sufficient.  */
  tree rhstype = TREE_TYPE (rhs);
  if (POINTER_TYPE_P (rhstype))
    rhstype = TREE_TYPE (rhstype);
  if (is_empty_type (rhstype))
    return NULL_TREE;
Though, the above
  if (POINTER_TYPE_P (rhstype))
    rhstype = TREE_TYPE (rhstype);
is just extremely suspicious, either we care about what type rhs has,
or it is dereferenced and it must be a pointer type and we care about
what it points to, but the simple fact whether rhs has a pointer type
or some other type shouldn't change what we test is_empty_type on.

When I was briefly looking at the assignment on which it actually warned,
it actually looked not empty type related.

	Jakub