Re: Remove gimple_body after cfg build; do not copy cfun structure

[Diego Novillo <dnovillo at google dot com>]

Jan Hubicka wrote:

> 	PR tree-optimization/37315
> 	* cgraph.c (cgraph_create_edge): Use gimple_has_body_p.
> 	* cgraphunit.c (verify_cgraph_node): drop gimple_body check.
> 	(cgraph_analyze_functions): Use node->analyzed
> 	(cgraph_mark_functions_to_output): Likewise.
> 	(cgraph_expand_function): All functions can be released after
> 	expanding.
> 	(cgraph_optimize): Use gimple_has_body_p.
> 	* ipa-inline.c (cgraph_clone_inlined_nodes): Use analyzed flag.
> 	(cgraph_decide_inlining_incrementally): Likewise.
> 	(inline_transform): Inline transform.
> 	* tree-inline.c (initialize_cfun): Do now shallow copy structure;
> 	copy fields needed.
> 	(inlinable_function_p): Drop gimple_body check.
> 	(expand_call_inline): Use gimple_has_body_p.
> 	* gimple.c (gimple_has_body_p): New.
> 	* gimple.h (gimple_has_body_p): Add prototype.
> 	* tree-cfg.c (execute_build_cfg): Remove gimple_body.
> 	(dump_function_to_file): Use gimple_has_body_p check.

Nice cleanup, thanks.  This is OK.  If you have some free cycles, could 
you also apply this patch to the lto branch?

>     /* ??? Can happen with nested function of extern inline.  */
>     gcc_assert (TREE_ASM_WRITTEN (decl));
>     current_function_decl = NULL;
> !   gcc_assert (!cgraph_preserve_function_body_p (decl));
> !   cgraph_release_function_body (node);
> !   /* Eliminate all call edges.  This is important so the call_expr no longer
> !      points to the dead function body.  */

s/call_expr/GIMPLE_CALL/

> + /* Return true when FNDECL has Gimple body either in unlowered
> +    or CFG form.  */
> + bool
> + gimple_has_body_p (tree fndecl)
> + {
> +   struct function *fn = DECL_STRUCT_FUNCTION (fndecl);
> +   return (gimple_body (fndecl) || (fn && fn->cfg));
> + }

This could be a static inline in gimple.h, but this is OK too.


Diego.