Fix floating point issues on Windows.

runtime/platform/floating_point_win.cc

+// Copyright (c) 2012, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+#include <math.h>
+#include <limits>
+
+#include "platform/globals.h"
+
+// Taken from third_party/v8/src/platform-win32.cc
+double fmod_ieee(double x, double y) {
+  // x is dividend, y is divisor.
+  // Work around MS fmod bugs. ISO Standard says:
+  // If dividend is finite and divisor is an infinity or
+  //    dividend is a zero and divisor is nonzero finite,
+  // then dividend is returned.
+  if (!(_finite(x) && (!_finite(y) && !isnan(y))) &&
+      !(x == 0 && (y != 0 && _finite(y)))) {
+    x = fmod(x, y);
+  }
+  return x;
+}
+
+// Bring MSVC atan2 behavior in line with ISO standard.
+// MSVC atan2 returns NaN when x and y are either +infinity or -infinity.
+// Per ISO standard:
+//  - If y is +/-infinity and x is -infinity, +/-3*pi/4 is returned.
+//  - If y is +/-infinity and x is +infinity, +/-pi/4 is returned.
+double atan2_ieee(double x, double y) {
+  int cls_x = _fpclass(x);
+  int cls_y = _fpclass(y);
+  if (((cls_x & (_FPCLASS_PINF | _FPCLASS_NINF)) != 0) &&
+      ((cls_y & (_FPCLASS_PINF | _FPCLASS_NINF)) != 0)) {
+    // atan2 values at infinities listed above are the same as values
+    // at (+/-1, +/-1). ndx_x is 0, when x is +infinty, 1 when x is -infinty.

[Mads Ager (google), 2012/11/29 13:25:14]

ndx_x and ndx_y -> index_x and index_y.

[aam-me, 2012/11/30 03:35:24]

Done.

+    // Same is with ndx_y.
+    int index_x = (cls_x & _FPCLASS_PINF) != 0 ? 0 : 1;
+    int index_y = (cls_y & _FPCLASS_PINF) != 0 ? 0 : 1;
+    static double atans_at_infinities[2][2] =
+      { { atan2(1.,  1.), atan2(1.,  -1.) },
+        { atan2(-1., 1.), atan2(-1., -1.) } };
+    return atans_at_infinities[index_x][index_y];
+  } else {
+    return atan2(x, y);
+  }
+}