match.pd: Optimize MIN_EXPR <addr1, addr2> etc. addr1 < addr2 would be simplified [PR102951]

This patch outlines the decision whether address comparison can be folded
or not from the match.pd simple comparison simplification and uses it
both there and in a new minmax simplification, such that we fold e.g.
MAX (&a[2], &a[1]) etc.
Some of the Wstringop-overflow-62.c changes might look weird, but that
seems to be mainly due to gimple_fold_builtin_memset not bothering to
copy over location, will fix that incrementally.

2021-10-28  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/102951
	* fold-const.h (address_compare): Declare.
	* fold-const.c (address_compare): New function.
	* match.pd (cmp (convert1?@2 addr@0) (convert2? addr@1)): Use
	address_compare helper.
	(minmax cmp (convert1?@2 addr@0) (convert2?@3 addr@1)): New
	simplification.

	* gcc.dg/tree-ssa/pr102951.c: New test.
	* gcc.dg/Wstringop-overflow-62.c: Adjust expected diagnostics.

1 files changed, 126 insertions, 0 deletions

diff --git a/gcc/fold-const.c b/gcc/fold-const.c
index c7daf871..54f91f0 100644
--- a/gcc/fold-const.c
+++ b/gcc/fold-const.c
@@ -16492,6 +16492,132 @@ tree_nonzero_bits (const_tree t)
   return wi::shwi (-1, TYPE_PRECISION (TREE_TYPE (t)));
 }
 
+/* Helper function for address compare simplifications in match.pd.
+   OP0 and OP1 are ADDR_EXPR operands being compared by CODE.
+   BASE0, BASE1, OFF0 and OFF1 are set by the function.
+   GENERIC is true if GENERIC folding and false for GIMPLE folding.
+   Returns 0 if OP0 is known to be unequal to OP1 regardless of OFF{0,1},
+   1 if bases are known to be equal and OP0 cmp OP1 depends on OFF0 cmp OFF1,
+   and 2 if unknown.  */
+
+int
+address_compare (tree_code code, tree type, tree op0, tree op1,
+		 tree &base0, tree &base1, poly_int64 &off0, poly_int64 &off1,
+		 bool generic)
+{
+  gcc_checking_assert (TREE_CODE (op0) == ADDR_EXPR);
+  gcc_checking_assert (TREE_CODE (op1) == ADDR_EXPR);
+  base0 = get_addr_base_and_unit_offset (TREE_OPERAND (op0, 0), &off0);
+  base1 = get_addr_base_and_unit_offset (TREE_OPERAND (op1, 0), &off1);
+  if (base0 && TREE_CODE (base0) == MEM_REF)
+    {
+      off0 += mem_ref_offset (base0).force_shwi ();
+      base0 = TREE_OPERAND (base0, 0);
+    }
+  if (base1 && TREE_CODE (base1) == MEM_REF)
+    {
+      off1 += mem_ref_offset (base1).force_shwi ();
+      base1 = TREE_OPERAND (base1, 0);
+    }
+  if (base0 == NULL_TREE || base1 == NULL_TREE)
+    return 2;
+
+  int equal = 2;
+  /* Punt in GENERIC on variables with value expressions;
+     the value expressions might point to fields/elements
+     of other vars etc.  */
+  if (generic
+      && ((VAR_P (base0) && DECL_HAS_VALUE_EXPR_P (base0))
+	  || (VAR_P (base1) && DECL_HAS_VALUE_EXPR_P (base1))))
+    return 2;
+  else if (decl_in_symtab_p (base0) && decl_in_symtab_p (base1))
+    {
+      symtab_node *node0 = symtab_node::get_create (base0);
+      symtab_node *node1 = symtab_node::get_create (base1);
+      equal = node0->equal_address_to (node1);
+    }
+  else if ((DECL_P (base0)
+	    || TREE_CODE (base0) == SSA_NAME
+	    || TREE_CODE (base0) == STRING_CST)
+	   && (DECL_P (base1)
+	       || TREE_CODE (base1) == SSA_NAME
+	       || TREE_CODE (base1) == STRING_CST))
+    equal = (base0 == base1);
+  if (equal == 1)
+    {
+      if (code == EQ_EXPR
+	  || code == NE_EXPR
+	  /* If the offsets are equal we can ignore overflow.  */
+	  || known_eq (off0, off1)
+	  || TYPE_OVERFLOW_UNDEFINED (TREE_TYPE (op0))
+	  /* Or if we compare using pointers to decls or strings.  */
+	  || (POINTER_TYPE_P (type)
+	      && (DECL_P (base0) || TREE_CODE (base0) == STRING_CST)))
+	return 1;
+      return 2;
+    }
+  if (equal != 0)
+    return equal;
+  if (code != EQ_EXPR && code != NE_EXPR)
+    return 2;
+
+  HOST_WIDE_INT ioff0 = -1, ioff1 = -1;
+  off0.is_constant (&ioff0);
+  off1.is_constant (&ioff1);
+  if ((DECL_P (base0) && TREE_CODE (base1) == STRING_CST)
+       || (TREE_CODE (base0) == STRING_CST && DECL_P (base1))
+       || (TREE_CODE (base0) == STRING_CST
+	   && TREE_CODE (base1) == STRING_CST
+	   && ioff0 >= 0 && ioff1 >= 0
+	   && ioff0 < TREE_STRING_LENGTH (base0)
+	   && ioff1 < TREE_STRING_LENGTH (base1)
+	  /* This is a too conservative test that the STRING_CSTs
+	     will not end up being string-merged.  */
+	   && strncmp (TREE_STRING_POINTER (base0) + ioff0,
+		       TREE_STRING_POINTER (base1) + ioff1,
+		       MIN (TREE_STRING_LENGTH (base0) - ioff0,
+			    TREE_STRING_LENGTH (base1) - ioff1)) != 0))
+    ;
+  else if (!DECL_P (base0) || !DECL_P (base1))
+    return 2;
+  /* If this is a pointer comparison, ignore for now even
+     valid equalities where one pointer is the offset zero
+     of one object and the other to one past end of another one.  */
+  else if (!INTEGRAL_TYPE_P (type))
+    ;
+  /* Assume that automatic variables can't be adjacent to global
+     variables.  */
+  else if (is_global_var (base0) != is_global_var (base1))
+    ;
+  else
+    {
+      tree sz0 = DECL_SIZE_UNIT (base0);
+      tree sz1 = DECL_SIZE_UNIT (base1);
+      /* If sizes are unknown, e.g. VLA or not representable, punt.  */
+      if (!tree_fits_poly_int64_p (sz0) || !tree_fits_poly_int64_p (sz1))
+	return 2;
+
+      poly_int64 size0 = tree_to_poly_int64 (sz0);
+      poly_int64 size1 = tree_to_poly_int64 (sz1);
+      /* If one offset is pointing (or could be) to the beginning of one
+	 object and the other is pointing to one past the last byte of the
+	 other object, punt.  */
+      if (maybe_eq (off0, 0) && maybe_eq (off1, size1))
+	equal = 2;
+      else if (maybe_eq (off1, 0) && maybe_eq (off0, size0))
+	equal = 2;
+      /* If both offsets are the same, there are some cases we know that are
+	 ok.  Either if we know they aren't zero, or if we know both sizes
+	 are no zero.  */
+      if (equal == 2
+	  && known_eq (off0, off1)
+	  && (known_ne (off0, 0)
+	      || (known_ne (size0, 0) && known_ne (size1, 0))))
+	equal = 0;
+     }
+  return equal;
+}
+
 #if CHECKING_P
 
 namespace selftest {