public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r14-285] tree-optimization/109594 - wrong register promotion
@ 2023-04-27  8:38 Richard Biener
  0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2023-04-27  8:38 UTC (permalink / raw)
  To: gcc-cvs

https://gcc.gnu.org/g:7bcdcf86e8272eeb524cc1dcb0ada8c8cfe6f27e

commit r14-285-g7bcdcf86e8272eeb524cc1dcb0ada8c8cfe6f27e
Author: Richard Biener <rguenther@suse.de>
Date:   Mon Apr 24 13:20:25 2023 +0200

    tree-optimization/109594 - wrong register promotion
    
    We fail to verify the constraints under which we allow handled
    components to wrap registers.  The gcc.dg/pr70022.c testcase shows
    that we happily end up with
    
      _2 = VIEW_CONVERT_EXPR<int[4]>(v_1(D))
    
    as produced by SSA rewrite and update_address_taken.  But the intent
    was that we wrap registers with at most a single level of handled
    components and specifically only allow __real, __imag, BIT_FIELD_REF
    and VIEW_CONVERT_EXPR on them, but not ARRAY_REF or COMPONENT_REF.
    
    Together with the improved gimple_load predicate taking advantage
    of the above and ASAN this eventually ICEd.
    
    The following fixes update_address_taken as to this constraint.
    
            PR tree-optimization/109594
            * tree-ssa.cc (non_rewritable_mem_ref_base): Constrain
            what we rewrite to a register based on the above.

Diff:
---
 gcc/tree-ssa.cc | 35 ++++++++++++++++++++++++++++-------
 1 file changed, 28 insertions(+), 7 deletions(-)

diff --git a/gcc/tree-ssa.cc b/gcc/tree-ssa.cc
index 4ca1f5f3104..70828355c2b 100644
--- a/gcc/tree-ssa.cc
+++ b/gcc/tree-ssa.cc
@@ -1572,14 +1572,30 @@ non_rewritable_mem_ref_base (tree ref)
   if (DECL_P (ref))
     return NULL_TREE;
 
-  if (! (base = CONST_CAST_TREE (strip_invariant_refs (ref))))
+  switch (TREE_CODE (ref))
     {
-      base = get_base_address (ref);
-      if (DECL_P (base))
-	return base;
-      return NULL_TREE;
+    case REALPART_EXPR:
+    case IMAGPART_EXPR:
+    case BIT_FIELD_REF:
+      if (DECL_P (TREE_OPERAND (ref, 0)))
+	return NULL_TREE;
+      break;
+    case VIEW_CONVERT_EXPR:
+      if (DECL_P (TREE_OPERAND (ref, 0)))
+	{
+	  if (TYPE_SIZE (TREE_TYPE (ref))
+	      != TYPE_SIZE (TREE_TYPE (TREE_OPERAND  (ref, 0))))
+	    return TREE_OPERAND (ref, 0);
+	  return NULL_TREE;
+	}
+      break;
+    /* We would need to rewrite ARRAY_REFs or COMPONENT_REFs and even
+       more so multiple levels of handled components.  */
+    default:;
     }
 
+  base = ref;
+
   /* But watch out for MEM_REFs we cannot lower to a
      VIEW_CONVERT_EXPR or a BIT_FIELD_REF.  */
   if (TREE_CODE (base) == MEM_REF
@@ -1630,9 +1646,14 @@ non_rewritable_mem_ref_base (tree ref)
       return decl;
     }
 
+  /* We cannot rewrite a decl in the base.  */
+  base = get_base_address (ref);
+  if (DECL_P (base))
+    return base;
+
   /* We cannot rewrite TARGET_MEM_REFs.  */
-  if (TREE_CODE (base) == TARGET_MEM_REF
-      && TREE_CODE (TREE_OPERAND (base, 0)) == ADDR_EXPR)
+  else if (TREE_CODE (base) == TARGET_MEM_REF
+	   && TREE_CODE (TREE_OPERAND (base, 0)) == ADDR_EXPR)
     {
       tree decl = TREE_OPERAND (TREE_OPERAND (base, 0), 0);
       if (! DECL_P (decl))

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2023-04-27  8:38 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-04-27  8:38 [gcc r14-285] tree-optimization/109594 - wrong register promotion Richard Biener

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).