gcc/0098-CHREC-multiplication-and-undefined-overflow.patch

From c4e4fef145c1e402f0558cc35f6c1ed0a08beffb Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=83=91=E6=99=A8=E5=8D=89?= <zhengchenhui1@huawei.com>
Date: Thu, 25 Jul 2024 20:16:52 +0800
Subject: [PATCH] CHREC multiplication and undefined overflow

This optimization is brought from https://gcc.gnu.org/pipermail/gcc-patches/2024-February/646531.html

When folding a multiply CHRECs are handled like {a, +, b} * c
is {a*c, +, b*c} but that isn't generally correct when overflow
invokes undefined behavior.  The following uses unsigned arithmetic
unless either a is zero or a and b have the same sign.

I've used simple early outs for INTEGER_CSTs and otherwise use
a range-query since we lack a tree_expr_nonpositive_p and
get_range_pos_neg isn't a good fit.
---
 gcc/common.opt                          |  4 ++
 gcc/testsuite/gcc.dg/pr68317.c          |  6 +-
 gcc/testsuite/gcc.dg/torture/pr114074.c | 31 ++++++++++
 gcc/tree-chrec.cc                       | 81 +++++++++++++++++++++----
 gcc/tree-chrec.h                        |  2 +-
 gcc/value-range.cc                      | 12 ++++
 gcc/value-range.h                       |  2 +
 7 files changed, 123 insertions(+), 15 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/torture/pr114074.c

diff --git a/gcc/common.opt b/gcc/common.opt
index b18f0b944..d3af3ba39 100644
--- a/gcc/common.opt
+++ b/gcc/common.opt
@@ -1771,6 +1771,10 @@ floop-interchange
 Common Var(flag_loop_interchange) Optimization
 Enable loop interchange on trees.
 
+fchrec-mul-fold-strict-overflow
+Common Var(flag_chrec_mul_fold_strict_overflow) Init(0)
+Enable strict overflow handling during constant folding of multiply CHRECs.
+
 floop-block
 Common Alias(floop-nest-optimize)
 Enable loop nest transforms.  Same as -floop-nest-optimize.
diff --git a/gcc/testsuite/gcc.dg/pr68317.c b/gcc/testsuite/gcc.dg/pr68317.c
index bd053a752..671a67d95 100644
--- a/gcc/testsuite/gcc.dg/pr68317.c
+++ b/gcc/testsuite/gcc.dg/pr68317.c
@@ -1,5 +1,5 @@
 /* { dg-do compile } */
-/* { dg-options "-O2 -fdisable-tree-ethread" } */
+/* { dg-options "-O2 -fdisable-tree-ethread -fchrec-mul-fold-strict-overflow" } */
 
 /* Note: Threader will collapse loop.  */
 
@@ -12,8 +12,8 @@ foo ()
 {
  int32_t index = 0;
 
- for (index; index <= 10; index--) // expected warning here
+ for (index; index <= 10; index--) /* { dg-warning "iteration \[0-9\]+ invokes undefined behavior" } */
    /* Result of the following multiply will overflow
       when converted to signed int32_t.  */
-   bar ((0xcafe + index) * 0xdead);  /* { dg-warning "iteration \[0-9\]+ invokes undefined behavior" } */
+   bar ((0xcafe + index) * 0xdead);
 }
diff --git a/gcc/testsuite/gcc.dg/torture/pr114074.c b/gcc/testsuite/gcc.dg/torture/pr114074.c
new file mode 100644
index 000000000..9a383d8fc
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/torture/pr114074.c
@@ -0,0 +1,31 @@
+/* { dg-do run } */
+<<<<<<< HEAD
+/* { dg-options "-fchrec-mul-fold-strict-overflow" } */
+=======
+/* { dg-options "-fchrec-mul-fold-strict-overflow"" } */
+>>>>>>> 47092575e7696f5a21cf75284fe3d4feb0c813ab
+int a, b, d;
+
+__attribute__((noipa)) void
+foo (void)
+{
+  ++d;
+}
+
+int
+main ()
+{
+  for (a = 0; a > -3; a -= 2)
+    {
+      int c = a;
+      b = __INT_MAX__ - 3000;
+      a = ~c * b;
+      foo ();
+      if (!a)
+	break;
+      a = c;
+    }
+  if (d != 2)
+    __builtin_abort ();
+  return 0;
+}
diff --git a/gcc/tree-chrec.cc b/gcc/tree-chrec.cc
index c44cea754..3323901bc 100644
--- a/gcc/tree-chrec.cc
+++ b/gcc/tree-chrec.cc
@@ -38,6 +38,8 @@ along with GCC; see the file COPYING3.  If not see
 #include "gimple.h"
 #include "tree-ssa-loop.h"
 #include "dumpfile.h"
+#include "value-range.h"
+#include "value-query.h"
 #include "tree-scalar-evolution.h"
 
 /* Extended folder for chrecs.  */
@@ -404,6 +406,13 @@ chrec_fold_multiply (tree type,
       || automatically_generated_chrec_p (op1))
     return chrec_fold_automatically_generated_operands (op0, op1);
 
+  if (flag_chrec_mul_fold_strict_overflow)
+    {
+      if (TREE_CODE (op0) != POLYNOMIAL_CHREC
+	  && TREE_CODE (op1) == POLYNOMIAL_CHREC)
+	std::swap (op0, op1);
+    }
+
   switch (TREE_CODE (op0))
     {
     case POLYNOMIAL_CHREC:
@@ -428,10 +437,53 @@ chrec_fold_multiply (tree type,
 	  if (integer_zerop (op1))
 	    return build_int_cst (type, 0);
 
-	  return build_polynomial_chrec
-	    (CHREC_VARIABLE (op0),
-	     chrec_fold_multiply (type, CHREC_LEFT (op0), op1),
-	     chrec_fold_multiply (type, CHREC_RIGHT (op0), op1));
+	  if (flag_chrec_mul_fold_strict_overflow)
+	    {
+	      /* When overflow is undefined and CHREC_LEFT/RIGHT do not have the
+		 same sign or CHREC_LEFT is zero then folding the multiply into
+		 the addition does not have the same behavior on overflow.  Use
+		 unsigned arithmetic in that case.  */
+	      value_range rl, rr;
+	      if (!ANY_INTEGRAL_TYPE_P (type)
+		  || TYPE_OVERFLOW_WRAPS (type)
+		  || integer_zerop (CHREC_LEFT (op0))
+		  || (TREE_CODE (CHREC_LEFT (op0)) == INTEGER_CST
+		  && TREE_CODE (CHREC_RIGHT (op0)) == INTEGER_CST
+		  && (tree_int_cst_sgn (CHREC_LEFT (op0))
+		      == tree_int_cst_sgn (CHREC_RIGHT (op0))))
+		  || (get_range_query (cfun)->range_of_expr (rl, CHREC_LEFT (op0))
+		  && !rl.undefined_p ()
+		  && (rl.nonpositive_p () || rl.nonnegative_p ())
+		  && get_range_query (cfun)->range_of_expr (rr,
+							CHREC_RIGHT (op0))
+		  && !rr.undefined_p ()
+		  && ((rl.nonpositive_p () && rr.nonpositive_p ())
+		  || (rl.nonnegative_p () && rr.nonnegative_p ()))))
+		{
+		  tree left = chrec_fold_multiply (type, CHREC_LEFT (op0), op1);
+		  tree right = chrec_fold_multiply (type, CHREC_RIGHT (op0), op1);
+		  return build_polynomial_chrec (CHREC_VARIABLE (op0), left, right);
+		}
+	      else
+		{
+		  tree utype = unsigned_type_for (type);
+		  tree uop1 = chrec_convert_rhs (utype, op1);
+		  tree uleft0 = chrec_convert_rhs (utype, CHREC_LEFT (op0));
+		  tree uright0 = chrec_convert_rhs (utype, CHREC_RIGHT (op0));
+		  tree left = chrec_fold_multiply (utype, uleft0, uop1);
+		  tree right = chrec_fold_multiply (utype, uright0, uop1);
+		  tree tem = build_polynomial_chrec (CHREC_VARIABLE (op0),
+							left, right);
+		  return chrec_convert_rhs (type, tem);
+		}
+	     }
+	   else
+	     {
+	       return build_polynomial_chrec
+		  (CHREC_VARIABLE (op0),
+		   chrec_fold_multiply (type, CHREC_LEFT (op0), op1),
+		   chrec_fold_multiply (type, CHREC_RIGHT (op0), op1));
+	     }
 	}
 
     CASE_CONVERT:
@@ -449,13 +501,20 @@ chrec_fold_multiply (tree type,
       switch (TREE_CODE (op1))
 	{
 	case POLYNOMIAL_CHREC:
-	  gcc_checking_assert
-	    (!chrec_contains_symbols_defined_in_loop (op1,
-						      CHREC_VARIABLE (op1)));
-	  return build_polynomial_chrec
-	    (CHREC_VARIABLE (op1),
-	     chrec_fold_multiply (type, CHREC_LEFT (op1), op0),
-	     chrec_fold_multiply (type, CHREC_RIGHT (op1), op0));
+	  if (flag_chrec_mul_fold_strict_overflow)
+	    {
+	      gcc_unreachable ();
+	    }
+	  else
+	   {
+	     gcc_checking_assert
+		(!chrec_contains_symbols_defined_in_loop (op1,
+				CHREC_VARIABLE (op1)));
+	     return build_polynomial_chrec
+		(CHREC_VARIABLE (op1),
+		 chrec_fold_multiply (type, CHREC_LEFT (op1), op0),
+		 chrec_fold_multiply (type, CHREC_RIGHT (op1), op0));
+	    }
 
 	CASE_CONVERT:
 	  if (tree_contains_chrecs (op1, NULL))
diff --git a/gcc/tree-chrec.h b/gcc/tree-chrec.h
index fcf41710d..cdc97d5d9 100644
--- a/gcc/tree-chrec.h
+++ b/gcc/tree-chrec.h
@@ -63,7 +63,7 @@ extern tree chrec_fold_plus (tree, tree, tree);
 extern tree chrec_fold_minus (tree, tree, tree);
 extern tree chrec_fold_multiply (tree, tree, tree);
 extern tree chrec_convert (tree, tree, gimple *, bool = true, tree = NULL);
-extern tree chrec_convert_rhs (tree, tree, gimple *);
+extern tree chrec_convert_rhs (tree, tree, gimple * = NULL);
 extern tree chrec_convert_aggressive (tree, tree, bool *);
 
 /* Operations.  */
diff --git a/gcc/value-range.cc b/gcc/value-range.cc
index 000bbcf89..a1dc10a24 100644
--- a/gcc/value-range.cc
+++ b/gcc/value-range.cc
@@ -656,6 +656,18 @@ irange::contains_p (tree cst) const
 
   return false;
 }
+bool
+irange::nonnegative_p () const
+{
+  return wi::ge_p (lower_bound (), 0, TYPE_SIGN (type ()));
+}
+
+bool
+irange::nonpositive_p () const
+{
+  return wi::le_p (upper_bound (), 0, TYPE_SIGN (type ()));
+}
+
 
 
 /* Normalize addresses into constants.  */
diff --git a/gcc/value-range.h b/gcc/value-range.h
index d4cba22d5..2dc0907de 100644
--- a/gcc/value-range.h
+++ b/gcc/value-range.h
@@ -69,6 +69,8 @@ public:
   bool varying_p () const;
   bool singleton_p (tree *result = NULL) const;
   bool contains_p (tree) const;
+  bool nonnegative_p () const;
+  bool nonpositive_p () const;
 
   // In-place operators.
   void union_ (const irange &);
-- 
2.33.0