public void ShowDoubleRoundingPriestExplicit()
{
double a = DoubleConverter.FromFloatingPointBinaryString("1" + '0'.Repeat(51) + "1"); // 100....00001 (53-bits wide)
double b = DoubleConverter.FromFloatingPointBinaryString("0.0" + '1'.Repeat(53)); // 0.0111...111 (53 1's)
double expected53 = a;
double expected64 = a + 1.0; // The point is that this is different to expceted53.
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(a));
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(b));
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(expected53));
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(expected64));
// Set Fpu to 53-bit precision (the default)
FpuControl.SetState((uint)FpuControl.PrecisionControl.Double53Bits, FpuControl.Mask.PrecisionControl);
double result53 = a + b;
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(result53));
Assert.AreEqual(expected53, result53);
// Explicit rounding makes no difference here (since we're in Double53bits precision FPU mode)
result53 = (double)(a + b);
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(result53));
Assert.AreEqual(expected53, result53);
// Set Fpu to 64-bit precision (extended precision)
FpuControl.SetState((uint)FpuControl.PrecisionControl.Extended64Bits, FpuControl.Mask.PrecisionControl);
double result64 = (double)(a + b);
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(result64));
Assert.AreEqual(expected64, result64);
}
public void ShowDoubleRoundingPriest()
{
double a = Math.Pow(2.0, 52.0) + 1.0;
double b = 0.5 - Math.Pow(2.0, -54.0);
double expected53 = a;
double expected64 = a + 1.0; // Math.Pow(2.0, 52.0) + 2.0;
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(a));
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(b));
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(expected53));
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(expected64));
// Set Fpu to 53-bit precision (the default)
FpuControl.SetState((uint)FpuControl.PrecisionControl.Double53Bits, FpuControl.Mask.PrecisionControl);
double result53 = a + b;
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(result53));
Assert.AreEqual(expected53, result53);
// Set Fpu to 64-bit precision (extended precision)
FpuControl.SetState((uint)FpuControl.PrecisionControl.Extended64Bits, FpuControl.Mask.PrecisionControl);
double result64 = (double)(a + b);
Debug.Print(DoubleConverter.ToFloatingPointBinaryString(result64));
Assert.AreEqual(expected64, result64);
}
public void ShowDoubleRounding()
{
double a = DoubleConverter.FromFloatingPointBinaryString('1'.Repeat(52) + "0"); // 111....11110
double b = DoubleConverter.FromFloatingPointBinaryString("0.100000000001"); // 000....00000.100000000001
double expected53 = DoubleConverter.FromFloatingPointBinaryString('1'.Repeat(52) + "1"); // 111....11111
double expected64 = a;
// Set Fpu to 53-bit precision (the default)
FpuControl.SetState((uint)FpuControl.PrecisionControl.Double53Bits, FpuControl.Mask.PrecisionControl);
double result53 = a + b;
Assert.AreEqual(expected53, result53);
// Set Fpu to 64-bit precision (extended precision)
FpuControl.SetState((uint)FpuControl.PrecisionControl.Extended64Bits, FpuControl.Mask.PrecisionControl);
double result64 = (double)(a + b);
Assert.AreEqual(expected64, result64);
double result64_0 = (a + b) - a;
Assert.AreNotEqual(0.0, result64_0);
Assert.AreNotEqual(b, result64_0);
Assert.AreEqual(0.5, result64_0); // 000....00000.1
}
public void TestMethod1()
{
double before = TestCalc();
Assert.AreEqual(0.0, before);
var oldState = new FpuControl.State(FpuControl.GetState());
var oldPc = oldState.PrecisionControl;
uint err = FpuControl.SetState((uint)FpuControl.PrecisionControl.Extended64Bits, FpuControl.Mask.PrecisionControl);
var newState = new FpuControl.State(FpuControl.GetState());
var newPc = newState.PrecisionControl;
double after = TestCalc();
Assert.AreEqual(0.5, after);
double afterSafe = TestCalcSafe();
Assert.AreEqual(0.0, afterSafe);
FpuControl.SetState((uint)oldState.PrecisionControl, FpuControl.Mask.PrecisionControl);
double reset = TestCalc();
Assert.AreEqual(0.0, reset);
}
public void TestPrecision53()
{
double a = DoubleConverter.FromFloatingPointBinaryString('1'.Repeat(52) + "0"); // 111....11110
double b = DoubleConverter.FromFloatingPointBinaryString("0.1"); // 000....00000.1
double expected53 = a;
FpuControl.SetState((uint)FpuControl.PrecisionControl.Double53Bits, FpuControl.Mask.PrecisionControl);
double result53 = (double)(a + b);
Assert.AreEqual(expected53, result53);
double result53_0 = (a + b) - a;
Assert.AreEqual(0.0, result53_0);
double result53_X = ((double)(a + b)) - a;
Assert.AreEqual(0.0, result53_X);
}
public void TestPrecision64()
{
double a = DoubleConverter.FromFloatingPointBinaryString('1'.Repeat(52) + "0"); // 111....11110
double b = DoubleConverter.FromFloatingPointBinaryString("0.1"); // 000....00000.1
double expected64 = a;
FpuControl.SetState((uint)FpuControl.PrecisionControl.Extended64Bits, FpuControl.Mask.PrecisionControl);
double result64 = (double)(a + b);
Assert.AreEqual(expected64, result64);
double result64_b = (a + b) - a;
Assert.AreEqual(b, result64_b);
double result64_0 = ((double)(a + b)) - a;
Assert.AreNotEqual(b, result64_0);
Assert.AreEqual(0.0, result64_0);
}