This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Copy assignments for non scalar types


Hi,

On Wed, Apr 14, 2010 at 01:31:05PM +0200, Richard Guenther wrote:
> On Wed, 14 Apr 2010, Diego Novillo wrote:
> 
> > On Wed, Apr 14, 2010 at 04:40, Richard Guenther <rguenther@suse.de> wrote:
> > 
> > > No.  make_rename_temp should go away.  Please.
> > 
> > I don't disagree, in principle (less code is always good).  What is
> > wrong with it?
> 
> It asks the SSA renamer to put your new variables into SSA form.
> It's very simple to do that manually (at least if no PHIs are
> involved), so better do that.
> 

The problem of using create_tmp_var directly is that the following
pattern is now bound to creep up at quite many places:

      tmp = create_tmp_var (TREE_TYPE (adj->base), "blah");
      if (TREE_CODE (TREE_TYPE (tmp)) == COMPLEX_TYPE
	  || TREE_CODE (TREE_TYPE (tmp)) == VECTOR_TYPE)
	DECL_GIMPLE_REG_P (tmp) = 1;

Perhaps we should have something like create_gimple_reg_tmp_var that
would do this?  If so, I'll be happy to add it.

Martin


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]