[Bug tree-optimization/61572] [4.10 Regression] ICE: in assign_by_spills, at lra-assigns.c:1335

["rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61572

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
          Component|rtl-optimization            |tree-optimization
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
That's because of

       register __typeof__(0) c asm("rdx");

which AFAIK makes 'rdx' "fixed" in the whole function.

What changed recently is that we sunk the b = c; assignment right before
its use.  Not doing that fixes the issue (but I wonder if the use of
that asm isn't undefined from the start).

Note that we still happily sink stores to hard registers (that probably
always reduces hardreg lifetime).

Index: tree-ssa-sink.c
===================================================================
--- tree-ssa-sink.c     (revision 211928)
+++ tree-ssa-sink.c     (working copy)
@@ -374,6 +374,12 @@ statement_sink_location (gimple stmt, ba
         nearest to commondom.  */
       if (gimple_vuse (stmt))
        {
+         /* Do not sink loads from hard registers.  */
+         if (gimple_assign_single_p (stmt)
+             && TREE_CODE (gimple_assign_rhs1 (stmt)) == VAR_DECL
+             && DECL_HARD_REGISTER (gimple_assign_rhs1 (stmt)))
+           return false;
+
          imm_use_iterator imm_iter;
          use_operand_p use_p;
          basic_block found = NULL;