new file containing very clever code for floating point "equality" comparisons

[ardour.git] / libs / pbd / pbd / floating.h

diff --git a/libs/pbd/pbd/floating.h b/libs/pbd/pbd/floating.h
new file mode 100644 (file)
index 0000000..f707dc9
--- /dev/null
+++ b/libs/pbd/pbd/floating.h
@@ -0,0 +1,53 @@
+/* Taken from
+ * http://www.cygnus-software.com/papers/comparingfloats/comparingfloats.htm
+ *
+ * Code assumed to be in the public domain.
+ */
+
+#ifndef __libpbd__floating_h__
+#define __libpbd__floating_h__
+
+namespace PBD {
+
+union Float_t
+{
+    Float_t (float num = 0.0f) : f(num) {}
+
+    // Portable extraction of components.
+    bool    negative() const { return (i >> 31) != 0; }
+    int32_t raw_mantissa() const { return i & ((1 << 23) - 1); }
+    int32_t raw_exponent() const { return (i >> 23) & 0xFF; }
+ 
+    int32_t i;
+    float f;
+};
+ 
+/* Note: ULPS = Units in the Last Place */
+
+static inline bool floateq (float a, float b, int max_ulps_diff)
+{
+    Float_t ua(a);
+    Float_t ub(b);
+ 
+    // Different signs means they do not match.
+    if (ua.negative() != ub.negative()) {
+        // Check for equality to make sure +0==-0
+           if (a == b) {
+                   return true;
+           }
+           return false;
+    }
+ 
+    // Find the difference in ULPs.
+    int ulps_diff = abs (ua.i - ub.i);
+
+    if (ulps_diff <= max_ulps_diff) {
+        return true;
+    }
+ 
+    return false;
+}
+
+} /* namespace */
+
+#endif /* __libpbd__floating_h__ */