changeset 8422:fbb8f75498f4

8073670: TypeF::eq and TypeD::eq do not handle NaNs correctly Summary: Change TypeF:eq and TypeD:eq to compare NaN values using a bitwise comparison. Reviewed-by: kvn Contributed-by: Stefan Anzinger <stefan.anzinger@oracle.com>
author dbuck
date Fri, 11 Aug 2017 03:30:28 -0400
parents 1832e44a9889
children b55756ea22d8
files src/share/vm/opto/type.cpp test/compiler/c2/FloatingPointFoldingTest.java
diffstat 2 files changed, 171 insertions(+), 30 deletions(-) [+]
line wrap: on
line diff
--- a/src/share/vm/opto/type.cpp	Wed Jul 19 09:00:13 2017 +0200
+++ b/src/share/vm/opto/type.cpp	Fri Aug 11 03:30:28 2017 -0400
@@ -1001,21 +1001,10 @@
 
 //------------------------------eq---------------------------------------------
 // Structural equality check for Type representations
-bool TypeF::eq( const Type *t ) const {
-  if( g_isnan(_f) ||
-      g_isnan(t->getf()) ) {
-    // One or both are NANs.  If both are NANs return true, else false.
-    return (g_isnan(_f) && g_isnan(t->getf()));
-  }
-  if (_f == t->getf()) {
-    // (NaN is impossible at this point, since it is not equal even to itself)
-    if (_f == 0.0) {
-      // difference between positive and negative zero
-      if (jint_cast(_f) != jint_cast(t->getf()))  return false;
-    }
-    return true;
-  }
-  return false;
+bool TypeF::eq(const Type *t) const {
+  // Bitwise comparison to distinguish between +/-0. These values must be treated
+  // as different to be consistent with C1 and the interpreter.
+  return (jint_cast(_f) == jint_cast(t->getf()));
 }
 
 //------------------------------hash-------------------------------------------
@@ -1116,21 +1105,10 @@
 
 //------------------------------eq---------------------------------------------
 // Structural equality check for Type representations
-bool TypeD::eq( const Type *t ) const {
-  if( g_isnan(_d) ||
-      g_isnan(t->getd()) ) {
-    // One or both are NANs.  If both are NANs return true, else false.
-    return (g_isnan(_d) && g_isnan(t->getd()));
-  }
-  if (_d == t->getd()) {
-    // (NaN is impossible at this point, since it is not equal even to itself)
-    if (_d == 0.0) {
-      // difference between positive and negative zero
-      if (jlong_cast(_d) != jlong_cast(t->getd()))  return false;
-    }
-    return true;
-  }
-  return false;
+bool TypeD::eq(const Type *t) const {
+  // Bitwise comparison to distinguish between +/-0. These values must be treated
+  // as different to be consistent with C1 and the interpreter.
+  return (jlong_cast(_d) == jlong_cast(t->getd()));
 }
 
 //------------------------------hash-------------------------------------------
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/test/compiler/c2/FloatingPointFoldingTest.java	Fri Aug 11 03:30:28 2017 -0400
@@ -0,0 +1,163 @@
+/*
+ * Copyright (c) 2015, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+/**
+ * @test
+ * @bug 8073670
+ * @summary Test that causes C2 to fold two NaNs with different values into a single NaN.
+ * @run main/othervm -XX:-TieredCompilation -Xcomp -XX:CompileCommand=compileonly,FloatingPointFoldingTest.test_double_inf -XX:CompileCommand=compileonly,FloatingPointFoldingTest.test_double_zero -XX:CompileCommand=compileonly,FloatingPointFoldingTest.test_double_nan -XX:CompileCommand=compileonly,FloatingPointFoldingTest.test_float_inf -XX:CompileCommand=compileonly,FloatingPointFoldingTest.test_float_zero -XX:CompileCommand=compileonly,FloatingPointFoldingTest.test_float_nan FloatingPointFoldingTest
+ */
+
+public class FloatingPointFoldingTest {
+    // Double values.
+    public static final long MINUS_INF_LONGBITS = 0xfff0000000000000L;
+    public static final double DOUBLE_MINUS_INF = Double.longBitsToDouble(MINUS_INF_LONGBITS);
+
+    public static final long PLUS_INF_LONGBITS = 0x7ff0000000000000L;
+    public static final double DOUBLE_PLUS_INF = Double.longBitsToDouble(PLUS_INF_LONGBITS);
+
+    public static final long MINUS_ZERO_LONGBITS = 0x8000000000000000L;
+    public static final double DOUBLE_MINUS_ZERO = Double.longBitsToDouble(MINUS_ZERO_LONGBITS);
+
+    // We need two different NaN values. A floating point number is
+    // considered to be NaN is the sign bit is 0, all exponent bits
+    // are set to 1, and at least one bit of the exponent is not zero.
+    //
+    // As java.lang.Double.NaN is 0x7ff8000000000000L, we use
+    // 0x7ffc000000000000L as a second NaN double value.
+    public static final long NAN_LONGBITS = 0x7ffc000000000000L;
+    public static final double DOUBLE_NAN = Double.longBitsToDouble(NAN_LONGBITS);
+
+    // Float values.
+    public static final int MINUS_INF_INTBITS = 0xff800000;
+    public static final float FLOAT_MINUS_INF = Float.intBitsToFloat(MINUS_INF_INTBITS);
+
+    public static final int PLUS_INF_INTBITS = 0x7f800000;
+    public static final float FLOAT_PLUS_INF = Float.intBitsToFloat(PLUS_INF_INTBITS);
+
+    public static final int MINUS_ZERO_INTBITS = 0x80000000;
+    public static final float FLOAT_MINUS_ZERO = Float.intBitsToFloat(MINUS_ZERO_INTBITS);
+
+    // As java.lang.Float.NaN is 0x7fc00000, we use 0x7fe00000
+    // as a second NaN float value.
+    public static final int NAN_INTBITS = 0x7fe00000;
+    public static final float FLOAT_NAN = Float.intBitsToFloat(NAN_INTBITS);
+
+
+    // Double tests.
+    static void test_double_inf(long[] result) {
+        double d1 = DOUBLE_MINUS_INF;
+        double d2 = DOUBLE_PLUS_INF;
+        result[0] = Double.doubleToRawLongBits(d1);
+        result[1] = Double.doubleToRawLongBits(d2);
+    }
+
+    static void test_double_zero(long[] result) {
+        double d1 = DOUBLE_MINUS_ZERO;
+        double d2 = 0;
+        result[0] = Double.doubleToRawLongBits(d1);
+        result[1] = Double.doubleToRawLongBits(d2);
+    }
+
+    static void test_double_nan(long[] result) {
+        double d1 = DOUBLE_NAN;
+        double d2 = Double.NaN;
+        result[0] = Double.doubleToRawLongBits(d1);
+        result[1] = Double.doubleToRawLongBits(d2);
+    }
+
+    // Float tests.
+    static void test_float_inf(int[] result) {
+        float f1 = FLOAT_MINUS_INF;
+        float f2 = FLOAT_PLUS_INF;
+        result[0] = Float.floatToRawIntBits(f1);
+        result[1] = Float.floatToRawIntBits(f2);
+    }
+
+    static void test_float_zero(int[] result) {
+        float f1 = FLOAT_MINUS_ZERO;
+        float f2 = 0;
+        result[0] = Float.floatToRawIntBits(f1);
+        result[1] = Float.floatToRawIntBits(f2);
+    }
+
+    static void test_float_nan(int[] result) {
+        float f1 = FLOAT_NAN;
+        float f2 = Float.NaN;
+        result[0] = Float.floatToRawIntBits(f1);
+        result[1] = Float.floatToRawIntBits(f2);
+    }
+
+    // Check doubles.
+    static void check_double(long[] result, double d1, double d2) {
+        if (result[0] == result[1]) {
+            throw new RuntimeException("ERROR: Two different double values are considered equal. \n"
+                                       + String.format("\toriginal values: 0x%x 0x%x\n", Double.doubleToRawLongBits(d1), Double.doubleToRawLongBits(d2))
+                                       + String.format("\tvalues after execution of method test(): 0x%x 0x%x", result[0], result[1]));
+        }
+    }
+
+    // Check floats.
+    static void check_float(int[] result, float f1, float f2) {
+        if (result[0] == result[1]) {
+            throw new RuntimeException("ERROR: Two different float values are considered equal. \n"
+                                       + String.format("\toriginal values: 0x%x 0x%x\n", Float.floatToRawIntBits(f1), Float.floatToRawIntBits(f2))
+                                       + String.format("\tvalues after execution of method test(): 0x%x 0x%x", result[0], result[1]));
+        }
+    }
+
+    public static void main(String[] args) {
+        // Float tests.
+
+        int[] iresult = new int[2];
+
+        // -Inf and +Inf.
+        test_float_inf(iresult);
+        check_float(iresult, FLOAT_MINUS_INF, FLOAT_PLUS_INF);
+
+        // 0 and -0.
+        test_float_zero(iresult);
+        check_float(iresult, FLOAT_MINUS_ZERO, 0);
+
+        // Diferrent NaNs.
+        test_float_nan(iresult);
+        check_float(iresult, FLOAT_NAN, Float.NaN);
+
+        // Double tests.
+
+        long[] lresult = new long[2];
+
+        // -Inf and +Inf.
+        test_double_inf(lresult);
+        check_double(lresult, DOUBLE_MINUS_INF, DOUBLE_PLUS_INF);
+
+        // 0 and -0.
+        test_double_zero(lresult);
+        check_double(lresult, DOUBLE_MINUS_ZERO, 0);
+
+        // Diferrent NaNs.
+        test_double_nan(lresult);
+        check_double(lresult, DOUBLE_NAN, Double.NaN);
+    }
+}