expand: Fold x - y < 0 to x < y during expansion [PR94802]

My earlier patch to simplify x - y < 0 etc. for signed subtraction
with undefined overflow into x < y in match.pd regressed some tests,
even when it was guarded to be post-IPA, the following patch thus
attempts to optimize that during expansion instead (which is the last
time we can do it, afterwards we lose the information whether it was
x - y < 0 or (int) ((unsigned) x - y) < 0 for which we couldn't
optimize it.

2021-01-05  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/94802
	* expr.h (maybe_optimize_sub_cmp_0): Declare.
	* expr.c: Include tree-pretty-print.h and flags.h.
	(maybe_optimize_sub_cmp_0): New function.
	(do_store_flag): Use it.
	* cfgexpand.c (expand_gimple_cond): Likewise.

	* gcc.target/i386/pr94802.c: New test.
	* gcc.dg/Wstrict-overflow-25.c: Remove xfail.

5 files changed, 110 insertions, 1 deletions

diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c
index 1d3f96f..b73019b 100644
--- a/gcc/cfgexpand.c
+++ b/gcc/cfgexpand.c
@@ -2621,6 +2621,14 @@ expand_gimple_cond (basic_block bb, gcond *stmt)
       && TREE_CODE (op1) == INTEGER_CST)
     code = maybe_optimize_mod_cmp (code, &op0, &op1);
 
+  /* Optimize (x - y) < 0 into x < y if x - y has undefined overflow.  */
+  if (!TYPE_UNSIGNED (TREE_TYPE (op0))
+      && (code == LT_EXPR || code == LE_EXPR
+	  || code == GT_EXPR || code == GE_EXPR)
+      && integer_zerop (op1)
+      && TREE_CODE (op0) == SSA_NAME)
+    maybe_optimize_sub_cmp_0 (code, &op0, &op1);
+
   last2 = last = get_last_insn ();
 
   extract_true_false_edges_from_block (bb, &true_edge, &false_edge);
diff --git a/gcc/expr.c b/gcc/expr.c
index d472eea..04ef5ad 100644
--- a/gcc/expr.c
+++ b/gcc/expr.c
@@ -62,6 +62,8 @@ along with GCC; see the file COPYING3.  If not see
 #include "ccmp.h"
 #include "gimple-fold.h"
 #include "rtx-vector-builder.h"
+#include "tree-pretty-print.h"
+#include "flags.h"
 
 
 /* If this is nonzero, we do not bother generating VOLATILE
@@ -12275,6 +12277,37 @@ maybe_optimize_mod_cmp (enum tree_code code, tree *arg0, tree *arg1)
   *arg1 = c4;
   return code == EQ_EXPR ? LE_EXPR : GT_EXPR;
 }
+
+/* Optimize x - y < 0 into x < 0 if x - y has undefined overflow.  */
+
+void
+maybe_optimize_sub_cmp_0 (enum tree_code code, tree *arg0, tree *arg1)
+{
+  gcc_checking_assert (code == GT_EXPR || code == GE_EXPR
+		       || code == LT_EXPR || code == LE_EXPR);
+  gcc_checking_assert (integer_zerop (*arg1));
+
+  if (!optimize)
+    return;
+
+  gimple *stmt = get_def_for_expr (*arg0, MINUS_EXPR);
+  if (stmt == NULL)
+    return;
+
+  tree treeop0 = gimple_assign_rhs1 (stmt);
+  tree treeop1 = gimple_assign_rhs2 (stmt);
+  if (!TYPE_OVERFLOW_UNDEFINED (TREE_TYPE (treeop0)))
+    return;
+
+  if (issue_strict_overflow_warning (WARN_STRICT_OVERFLOW_COMPARISON))
+    warning_at (gimple_location (stmt), OPT_Wstrict_overflow,
+		"assuming signed overflow does not occur when "
+		"simplifying %<X - Y %s 0%> to %<X %s Y%>",
+		op_symbol_code (code), op_symbol_code (code));
+
+  *arg0 = treeop0;
+  *arg1 = treeop1;
+}
 
 /* Generate code to calculate OPS, and exploded expression
    using a store-flag instruction and return an rtx for the result.
@@ -12363,6 +12396,14 @@ do_store_flag (sepops ops, rtx target, machine_mode mode)
 	}
     }
 
+  /* Optimize (x - y) < 0 into x < y if x - y has undefined overflow.  */
+  if (!unsignedp
+      && (ops->code == LT_EXPR || ops->code == LE_EXPR
+	  || ops->code == GT_EXPR || ops->code == GE_EXPR)
+      && integer_zerop (arg1)
+      && TREE_CODE (arg0) == SSA_NAME)
+    maybe_optimize_sub_cmp_0 (ops->code, &arg0, &arg1);
+
   /* Get the rtx comparison code to use.  We know that EXP is a comparison
      operation of some type.  Some comparisons against 1 and -1 can be
      converted to comparisons with zero.  Do so here so that the tests
diff --git a/gcc/expr.h b/gcc/expr.h
index 6f99680..1f0177a 100644
--- a/gcc/expr.h
+++ b/gcc/expr.h
@@ -298,6 +298,7 @@ extern tree string_constant (tree, tree *, tree *, tree *);
 extern tree byte_representation (tree, tree *, tree *, tree *);
 
 extern enum tree_code maybe_optimize_mod_cmp (enum tree_code, tree *, tree *);
+extern void maybe_optimize_sub_cmp_0 (enum tree_code, tree *, tree *);
 
 /* Two different ways of generating switch statements.  */
 extern int try_casesi (tree, tree, tree, tree, rtx, rtx, rtx, profile_probability);
diff --git a/gcc/testsuite/gcc.dg/Wstrict-overflow-25.c b/gcc/testsuite/gcc.dg/Wstrict-overflow-25.c
index 774474d..0091644 100644
--- a/gcc/testsuite/gcc.dg/Wstrict-overflow-25.c
+++ b/gcc/testsuite/gcc.dg/Wstrict-overflow-25.c
@@ -7,5 +7,5 @@
 int
 foo (int x, int y)
 {
-  return x - y < 0; /* { dg-warning "assuming signed overflow does not occur" "correct warning" { xfail *-*-* } } */
+  return x - y < 0; /* { dg-warning "assuming signed overflow does not occur" "correct warning" } */
 }
diff --git a/gcc/testsuite/gcc.target/i386/pr94802.c b/gcc/testsuite/gcc.target/i386/pr94802.c
new file mode 100644
index 0000000..6c1bc66
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr94802.c
@@ -0,0 +1,59 @@
+/* PR tree-optimization/94802 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -masm=att" } */
+/* { dg-final { scan-assembler-not "\ttestl\t" } } */
+/* { dg-final { scan-assembler-times "\tcmpl\t" 8 } } */
+
+void foo (void);
+
+int
+f1 (int a, int b)
+{
+  return (a - b) >= 0;
+}
+
+int
+f2 (int a, int b)
+{
+  return (a - b) > 0;
+}
+
+int
+f3 (int a, int b)
+{
+  return (a - b) <= 0;
+}
+
+int
+f4 (int a, int b)
+{
+  return (a - b) < 0;
+}
+
+void
+f5 (int a, int b)
+{
+  if ((a - b) >= 0)
+    foo ();
+}
+
+void
+f6 (int a, int b)
+{
+  if ((a - b) > 0)
+    foo ();
+}
+
+void
+f7 (int a, int b)
+{
+  if ((a - b) <= 0)
+    foo ();
+}
+
+void
+f8 (int a, int b)
+{
+  if ((a - b) < 0)
+    foo ();
+}