On Mon, Aug 14, 2017 at 1:25 PM, Richard Biener <richard.guent...@gmail.com> wrote: > On Mon, Aug 14, 2017 at 1:23 PM, Richard Biener > <richard.guent...@gmail.com> wrote: >> On Wed, Aug 9, 2017 at 1:39 PM, Martin Liška <mli...@suse.cz> wrote: >>> On 08/09/2017 11:43 AM, Richard Biener wrote: >>>> I only have the patch I sent you so I can't re-diff. >>>> >>>> Richard. >>> >>> Hi. >>> >>> I'm sending rebased version of the patch. However the patch >>> eats all my memory when e.g. building ../../../libgcc/libgcov-merge.c. >>> If you have time, please try to make it working for &STRING_CST. I can >>> continue >>> then. >> >> Note I didn't send the patch to make you use it -- it changes too much and I >> never fixed all the fall out. It was meant as an exercise on how to >> use a CONST_DECL. > > Btw. the > > @@ -732,7 +730,7 @@ is_gimple_reg (tree t) > if (virtual_operand_p (t)) > return false; > > - if (TREE_CODE (t) == SSA_NAME) > + if (TREE_CODE (t) == SSA_NAME || TREE_CODE (t) == CONST_DECL) > return true; > > if (!is_gimple_variable (t)) > > hunk looks wrong.
Oh, and the gimplifier.c hunk was misapplied - the gimplification to CONST_DECL is supposed to only happen for STRING_CSTs, not arbitrary constants. Updated patch attached. Richard. > > Richard. > >> Richard. >> >>> Thanks, >>> Martin
fix-pr50199
Description: Binary data