Hi,

this is an undefined behavior spotted by the sanitizer that has managed to go 
unnoticed until now.  Tested on x86-64/Linux, OK for the mainline?


2021-03-04  Eric Botcazou  <ebotca...@adacore.com>

        PR rtl-optimization/99376
        * rtlanal.c (nonzero_bits1) <arithmetic operators>: If the number
        of low-order zero bits is too large, set the result to 0 directly.

-- 
Eric Botcazou
diff --git a/gcc/rtlanal.c b/gcc/rtlanal.c
index d1240b0b7c5..a8ea1d72636 100644
--- a/gcc/rtlanal.c
+++ b/gcc/rtlanal.c
@@ -5053,11 +5053,17 @@ nonzero_bits1 (const_rtx x, scalar_int_mode mode, const_rtx known_x,
 	    gcc_unreachable ();
 	  }
 
+	/* Note that mode_width <= HOST_BITS_PER_WIDE_INT, see above.  */
 	if (result_width < mode_width)
 	  nonzero &= (HOST_WIDE_INT_1U << result_width) - 1;
 
 	if (result_low > 0)
-	  nonzero &= ~((HOST_WIDE_INT_1U << result_low) - 1);
+	  {
+	    if (result_low < HOST_BITS_PER_WIDE_INT)
+	      nonzero &= ~((HOST_WIDE_INT_1U << result_low) - 1);
+	    else
+	      nonzero = 0;
+	  }
       }
       break;
 

Reply via email to