Get float constant to be the same the runtime result (and VB.NET)

问题

I tried the equivalent of Michael Meadows EDIT 2, but in VB.NET and got a different result. (Specifically both the Double and Decimal results were 600000.0238418580.)

I have determined the difference is with the compile-time accuracy of a float (Single) division stored into a float in C#, (which seems to be more equivalent to VB.NET's accuracy when storing into a Double) and what happens (in both languages unsurprisingly) when you force the division to occur at runtime.

So, THREE_FIFTHS and vTHREE_FIFTHS provide different results for the asDouble summation:

const int ONE_MILLION = 1000000;

float THREEsng = 3f;
float FIVEsng = 5f;
float vTHREE_FIFTHS = THREEsng / FIVEsng;

const float THREE_FIFTHS = 3f / 5f;

Console.WriteLine("Three Fifths: {0}", THREE_FIFTHS.ToString("F10"));
float asSingle = 0f;
double asDouble = 0d;
decimal asDecimal = 0M;

for (int i = 0; i < ONE_MILLION; i++)
{
    asSingle += (float) THREE_FIFTHS;
    asDouble += (double) THREE_FIFTHS;
    asDecimal += (decimal) THREE_FIFTHS;
}
Console.WriteLine("Six Hundred Thousand: {0:F10}", THREE_FIFTHS * ONE_MILLION);
Console.WriteLine("Single: {0}", asSingle.ToString("F10"));
Console.WriteLine("Double: {0}", asDouble.ToString("F10"));
Console.WriteLine("Decimal: {0}", asDecimal.ToString("F10"));

Console.WriteLine("vThree Fifths: {0}", vTHREE_FIFTHS.ToString("F10"));
asSingle = 0f;
asDouble = 0d;
asDecimal = 0M;

for (int i = 0; i < ONE_MILLION; i++)
{
    asSingle += (float) vTHREE_FIFTHS;
    asDouble += (double) vTHREE_FIFTHS;
    asDecimal += (decimal) vTHREE_FIFTHS;
}
Console.WriteLine("Six Hundred Thousand: {0:F10}", vTHREE_FIFTHS * ONE_MILLION);
Console.WriteLine("Single: {0}", asSingle.ToString("F10"));
Console.WriteLine("Double: {0}", asDouble.ToString("F10"));
Console.WriteLine("Decimal: {0}", asDecimal.ToString("F10"));

The result with the difference hightlighted is:

Three Fifths: 0.6000000000
Six Hundred Thousand: 600000.0000000000
Single: 599093.4000000000
Double: 599999.9999886850
Decimal: 600000.0000000000
vThree Fifths: 0.6000000000
Six Hundred Thousand: 600000.0000000000
Single: 599093.4000000000
Double: 600000.0238418580
Decimal: 600000.0000000000

My question is, can you get C# to get a const float expression with the equivalent of the runtime (and VB.NET) result? (I.e. produce a THREE_FIFTHS with the same results as vTHREE_FIFTHS.)

回答1:

THREE_FIFTHS has the same value as vTHREE_FIFTHS in your example (you can see this with BitConverter.GetBytes). It only differs in the way it is added to a double in your code.

I think your difference is due to the way that the C# compiler handles consts as if they were literals, in some ways. E.g. this operation, though it's normally not allowed without a cast, is ok because the const lets the compiler see that the int is small enough to work out:

const int i = 5;
byte b = i;

In your case, this means that it doesn't add the single-precision value 3/5 to the double, but calculates the double value 3/5 and adds that. Unfortunately this extra "intelligence" has a side effect. You can get around it by storing the const float as a float first, e.g.:

float f = THREE_FIFTHS;
asDouble += (double) f;

With this, both ways calculate the double as 600000.0238418580.

You can see more detail on the weirdness with these outputs:

string GetByteString(double d)
{
    return string.Join("", BitConverter.GetBytes(d).Select(b=>b.ToString("X2")));
}
string GetByteString(float f)
{
    return string.Join("", BitConverter.GetBytes(f).Select(b=>b.ToString("X2")));
}
double vd = vTHREE_FIFTHS;
double d = THREE_FIFTHS;
const double cd = THREE_FIFTHS;
float f = THREE_FIFTHS;
const double cd2 = 3d / 5d;
double d2 = 3d / 5d;
double df = f;

// doubles
Console.WriteLine(GetByteString((double)THREE_FIFTHS));
Console.WriteLine(GetByteString(vd));
Console.WriteLine(GetByteString(df));
Console.WriteLine(GetByteString(d));
Console.WriteLine(GetByteString(cd));
Console.WriteLine(GetByteString(cd2));
Console.WriteLine(GetByteString(d2));

// floats
Console.WriteLine(GetByteString(f));
Console.WriteLine(GetByteString(vTHREE_FIFTHS));
Console.WriteLine(GetByteString(THREE_FIFTHS));

prints this:
333333333333E33F
000000403333E33F <-- these are the only ones that were actually
000000403333E33F <-- converted from 32-bit float values to doubles
333333333333E33F
333333333333E33F
333333333333E33F
333333333333E33F
9A99193F
9A99193F
9A99193F

回答2:

It does look like the answer is "it can't be done", because, as Tim S. pointed out, the compile-time constant is not really constant, but actually reinterpreted at each use. Specifically the (double) cast version of the constant produces a different result to the runtime result.

void Main()
{
        const int ONE_MILLION = 1000000;

        float THREEsng = 3f;
        float FIVEsng = 5f;
        float vTHREE_FIFTHS = THREEsng / FIVEsng;

        const float THREE_FIFTHS = 3f / 5f;

        Console.WriteLine("Three Fifths: {0}", THREE_FIFTHS.ToString("F10"));
        float asSingle = 0f;
        double asDouble = 0d;
        decimal asDecimal = 0M;

        for (int i = 0; i < ONE_MILLION; i++)
        {
            asSingle += (float) THREE_FIFTHS;
            asDouble += (double) THREE_FIFTHS;
            asDecimal += (decimal) THREE_FIFTHS;
        }
        Console.WriteLine("Six Hundred Thousand: {0:F10}", THREE_FIFTHS * ONE_MILLION);
        Console.WriteLine("Single: {0}", asSingle.ToString("F10"));
        Console.WriteLine("Double: {0}", asDouble.ToString("F10"));
        Console.WriteLine("Decimal: {0}", asDecimal.ToString("F10"));
        Console.WriteLine(GetByteString((float) THREE_FIFTHS));
        Console.WriteLine(GetByteString((double) THREE_FIFTHS));
        Console.WriteLine(GetByteString((decimal) THREE_FIFTHS));

        Console.WriteLine("vThree Fifths: {0}", vTHREE_FIFTHS.ToString("F10"));
        asSingle = 0f;
        asDouble = 0d;
        asDecimal = 0M;

        for (int i = 0; i < ONE_MILLION; i++)
        {
            asSingle += (float) vTHREE_FIFTHS;
            asDouble += (double) vTHREE_FIFTHS;
            asDecimal += (decimal) vTHREE_FIFTHS;
        }
        Console.WriteLine("Six Hundred Thousand: {0:F10}", vTHREE_FIFTHS * ONE_MILLION);
        Console.WriteLine("Single: {0}", asSingle.ToString("F10"));
        Console.WriteLine("Double: {0}", asDouble.ToString("F10"));
        Console.WriteLine("Decimal: {0}", asDecimal.ToString("F10"));
        Console.WriteLine(GetByteString((float) vTHREE_FIFTHS));
        Console.WriteLine(GetByteString((double) vTHREE_FIFTHS));
        Console.WriteLine(GetByteString((decimal) vTHREE_FIFTHS));

}

// Define other methods and classes here
string GetByteString(double d)
{
    return "#" + string.Join("", BitConverter.GetBytes(d).Select(b=>b.ToString("X2")));
}
string GetByteString(decimal d)
{
    return "D" + string.Join("", Decimal.GetBits(d).Select(b=>b.ToString("X8")));
}
string GetByteString(float f)
{
    return "S" + string.Join("", BitConverter.GetBytes(f).Select(b=>b.ToString("X2")));
}

Output:

Three Fifths: 0.6000000000
Six Hundred Thousand: 600000.0000000000
Single: 599093.4000000000
Double: 599999.9999886850
Decimal: 600000.0000000000
S9A99193F
#333333333333E33F
D00000006000000000000000000010000
vThree Fifths: 0.6000000000
Six Hundred Thousand: 600000.0000000000
Single: 599093.4000000000
Double: 600000.0238418580
Decimal: 600000.0000000000
S9A99193F
#000000403333E33F
D00000006000000000000000000010000

来源：https://stackoverflow.com/questions/19593234/get-float-constant-to-be-the-same-the-runtime-result-and-vb-net

标签

const

floating-point-conversion