N
Nobody
Armin Zingler said:
How about using ILGenerator.Emit Method?
How about using ILGenerator.Emit Method?
A
Armin Zingler
Am 11.01.2011 03:18, schrieb Nobody:
I don't know how I can make the VB compiler emit IL code inline
using this method.
I don't know how I can make the VB compiler emit IL code inline
using this method.
A
Armin Zingler
Am 11.01.2011 03:18, schrieb Nobody:
I think this is for compiler builders making use of the AssemblyBuilder class
and so on.
(in short:
AssemblyBuilder.DefineDynamicModule.DefineType.DefineMethod.GetILGenerator)
I think this is for compiler builders making use of the AssemblyBuilder class
and so on.
(in short:
AssemblyBuilder.DefineDynamicModule.DefineType.DefineMethod.GetILGenerator)
C
Cor
Armin,
Math is not my topic, but if I understand you correct, you want for xyz as
results by using the code below 7 7 7
Dim x As Integer = CInt(7.9)
Dim y As Integer = CInt(7.8)
Dim z As Integer = CInt(7.1)
Strange is that the default banking rounding is in my opinion not used
because i get 8 8 7, can you check if I understand something wrong?
Cor
"Armin Zingler" wrote in message
Hi,
is there a function in the Framework that cuts the places
after the decimal point of a Double and converts to an Integer
_without_ performing any rounding (as Convert.ToInt32 does)?
I mean, what the IL code "conv.i4" does? ATM I'm using a function
that exactly does what I need in a C++ helper library:
static int ToInt32(double value)
{
return int(value);
};
But I want to get rid of it because a whole library for this
function is oversized and because this simple function should be
in the Framework, shouldn't it?
Math is not my topic, but if I understand you correct, you want for xyz as
results by using the code below 7 7 7
Dim x As Integer = CInt(7.9)
Dim y As Integer = CInt(7.8)
Dim z As Integer = CInt(7.1)
Strange is that the default banking rounding is in my opinion not used
because i get 8 8 7, can you check if I understand something wrong?
Cor
"Armin Zingler" wrote in message
Hi,
is there a function in the Framework that cuts the places
after the decimal point of a Double and converts to an Integer
_without_ performing any rounding (as Convert.ToInt32 does)?
I mean, what the IL code "conv.i4" does? ATM I'm using a function
that exactly does what I need in a C++ helper library:
static int ToInt32(double value)
{
return int(value);
};
But I want to get rid of it because a whole library for this
function is oversized and because this simple function should be
in the Framework, shouldn't it?
A
Armin Zingler
Cor,
Am 11.01.2011 09:26, schrieb Cor:
You're right.
CInt only calls Convert.ToInt32. The latter is better described in the
documentation: http://msdn.microsoft.com/en-us/library/ffdk7eyz.aspx
You got it right. The only problem I have is that Convert.ToInt32 does
two steps:
1. Convert to Integer (simply by cutting the decimal places: 7 7 7 in your example)
2. Add/Subtract 1 to correct the result depending on the decimal places.
I need a function that only does step 1 (which is very quick) because step 2
is time-intensive and it is not required because the decimal places are always
*.0000000. The C++ function that I wrote does only this. I have to (and I can)
live with my one-function helper library, but I'm still a little surprised that
this is not in the Framework.
Armin
Am 11.01.2011 09:26, schrieb Cor:
You're right.
CInt only calls Convert.ToInt32. The latter is better described in the
documentation: http://msdn.microsoft.com/en-us/library/ffdk7eyz.aspx
You got it right. The only problem I have is that Convert.ToInt32 does
two steps:
1. Convert to Integer (simply by cutting the decimal places: 7 7 7 in your example)
2. Add/Subtract 1 to correct the result depending on the decimal places.
I need a function that only does step 1 (which is very quick) because step 2
is time-intensive and it is not required because the decimal places are always
*.0000000. The C++ function that I wrote does only this. I have to (and I can)
live with my one-function helper library, but I'm still a little surprised that
this is not in the Framework.
Armin
C
Cor
Armin,
You mean something like this but then with another name?
Dim x As Integer = CInt(Fix(7.9))
Dim y As Integer = CInt(Fix(7.8))
Dim z As Integer = CInt(Fix(7.1))
Or instead of the Fix the Int if the negative and positive should stay?
http://msdn.microsoft.com/en-us/library/xh29swte(v=VS.90).aspx
And then that in one function, I don't assume it is easy to get that.
Cor
"Armin Zingler" wrote in message
Cor,
Am 11.01.2011 09:26, schrieb Cor:
You're right.
CInt only calls Convert.ToInt32. The latter is better described in the
documentation: http://msdn.microsoft.com/en-us/library/ffdk7eyz.aspx
You got it right. The only problem I have is that Convert.ToInt32 does
two steps:
1. Convert to Integer (simply by cutting the decimal places: 7 7 7 in your
example)
2. Add/Subtract 1 to correct the result depending on the decimal places.
I need a function that only does step 1 (which is very quick) because step 2
is time-intensive and it is not required because the decimal places are
always
*.0000000. The C++ function that I wrote does only this. I have to (and I
can)
live with my one-function helper library, but I'm still a little surprised
that
this is not in the Framework.
Armin
You mean something like this but then with another name?
Dim x As Integer = CInt(Fix(7.9))
Dim y As Integer = CInt(Fix(7.8))
Dim z As Integer = CInt(Fix(7.1))
Or instead of the Fix the Int if the negative and positive should stay?
http://msdn.microsoft.com/en-us/library/xh29swte(v=VS.90).aspx
And then that in one function, I don't assume it is easy to get that.
Cor
"Armin Zingler" wrote in message
Cor,
Am 11.01.2011 09:26, schrieb Cor:
You're right.
CInt only calls Convert.ToInt32. The latter is better described in the
documentation: http://msdn.microsoft.com/en-us/library/ffdk7eyz.aspx
You got it right. The only problem I have is that Convert.ToInt32 does
two steps:
1. Convert to Integer (simply by cutting the decimal places: 7 7 7 in your
example)
2. Add/Subtract 1 to correct the result depending on the decimal places.
I need a function that only does step 1 (which is very quick) because step 2
is time-intensive and it is not required because the decimal places are
always
*.0000000. The C++ function that I wrote does only this. I have to (and I
can)
live with my one-function helper library, but I'm still a little surprised
that
this is not in the Framework.
Armin
N
Nobody
Armin Zingler said:
If only you could do this:
Imports System.Reflection.Emit
Public Class Form1
Private Sub Form1_Load(ByVal sender As System.Object, ByVal e As
System.EventArgs) Handles MyBase.Load
Dim d As Double = 7.9999999
Dim i As Integer
Dim o As ILGenerator
o = New ILGenerator
o.Emit(OpCodes.Conv_I4, d)
Console.WriteLine(i)
End Sub
End Class
But doesn't compile because of one line. You need to call GetILGenerator()
method instead of using New. I am seeing examples of creating methods in the
fly, and calling them, but this may introduce additional overhead that is
slower than the current alternative. I would search the web for "IL code
injection" to find if there is a solution.
N
Nobody
Cor said:
Int/Fix functions return the same data type as the input, so if you put 7.9
Double, you get 7 of type Double, and CInt would convert the 7 Double to 7
Integer, but it also runs the rounding routine, so it's still slow.
A
Armin Zingler
Cor,
Am 11.01.2011 18:16, schrieb Cor:
No, because CInt rounds. Once again, I don't want to round.
No matter if I use Fix or Int, it doesn't make sense to use CInt
afterwards.
It's easy. Look at my C++ function.
Armin
Am 11.01.2011 18:16, schrieb Cor:
No, because CInt rounds. Once again, I don't want to round.
No matter if I use Fix or Int, it doesn't make sense to use CInt
afterwards.
It's easy. Look at my C++ function.
Armin
A
Armin Zingler
Am 11.01.2011 18:24, schrieb Nobody:
Exactly!
Exactly!
C
Cor
But this means that you don't want to use CInt at all, because Cint(7D) is
the same as Cint(Fix(7.9)) for the second part.
Then there are in my idea less options.
Or do you mean that they should change the internal code for CInt(double),
few chance on that in my idea as that gives breaking chances.
Cor
"Armin Zingler" wrote in message
Am 11.01.2011 18:24, schrieb Nobody:
Exactly!
the same as Cint(Fix(7.9)) for the second part.
Then there are in my idea less options.
Or do you mean that they should change the internal code for CInt(double),
few chance on that in my idea as that gives breaking chances.
Cor
"Armin Zingler" wrote in message
Am 11.01.2011 18:24, schrieb Nobody:
Exactly!
C
Cor
And you really believe what you want is quicker than
Dim x = Cint(Fix(7.9999))
Every reflection method is slow in fact can Armin than simple use Option
Strict Of to obtain the same.
In my idea few chance on that.
Cor
"Nobody" wrote in message
If only you could do this:
Imports System.Reflection.Emit
Public Class Form1
Private Sub Form1_Load(ByVal sender As System.Object, ByVal e As
System.EventArgs) Handles MyBase.Load
Dim d As Double = 7.9999999
Dim i As Integer
Dim o As ILGenerator
o = New ILGenerator
o.Emit(OpCodes.Conv_I4, d)
Console.WriteLine(i)
End Sub
End Class
But doesn't compile because of one line. You need to call GetILGenerator()
method instead of using New. I am seeing examples of creating methods in the
fly, and calling them, but this may introduce additional overhead that is
slower than the current alternative. I would search the web for "IL code
injection" to find if there is a solution.
Dim x = Cint(Fix(7.9999))
Every reflection method is slow in fact can Armin than simple use Option
Strict Of to obtain the same.
In my idea few chance on that.
Cor
"Nobody" wrote in message
Armin Zingler said:
If only you could do this:
Imports System.Reflection.Emit
Public Class Form1
Private Sub Form1_Load(ByVal sender As System.Object, ByVal e As
System.EventArgs) Handles MyBase.Load
Dim d As Double = 7.9999999
Dim i As Integer
Dim o As ILGenerator
o = New ILGenerator
o.Emit(OpCodes.Conv_I4, d)
Console.WriteLine(i)
End Sub
End Class
But doesn't compile because of one line. You need to call GetILGenerator()
method instead of using New. I am seeing examples of creating methods in the
fly, and calling them, but this may introduce additional overhead that is
slower than the current alternative. I would search the web for "IL code
injection" to find if there is a solution.
A
Armin Zingler
Am 11.01.2011 19:34, schrieb Cor:
Correct, I do not want to use CInt (or Convert.ToInt32).
No, they shouldn't change CInt (or Convert.ToInt32), but it would be nice
to have an additional function the does not perform unnecessary rounding.
Just take the whole part before the "." as an Integer. That's all. No magic.
Very simple. It's not expected _more_, it's expected _less_.
Correct, I do not want to use CInt (or Convert.ToInt32).
No, they shouldn't change CInt (or Convert.ToInt32), but it would be nice
to have an additional function the does not perform unnecessary rounding.
Just take the whole part before the "." as an Integer. That's all. No magic.
Very simple. It's not expected _more_, it's expected _less_.
A
Armin Zingler
Am 11.01.2011 19:37, schrieb Cor:
Well, it's an option. I'm gonna try.....
And you really believe what you want is quicker than
Dim x = Cint(Fix(7.9999))
Every reflection method is slow in fact can Armin than simple use Option
Strict Of to obtain the same.
In my idea few chance on that.
Well, it's an option.
I'm gonna try.....
C
Cor
Armin,
this code
Dim x As Integer = CInt(Fix(7.9))
Dim y As Integer = Convert.ToInt32(Fix(7.9))
leads to this IL not so much in my idea.
..locals init (
[0] int32 x,
[1] int32 y)
L_0000: nop
L_0001: ldc.r8 7.9
L_000a: call float64
[Microsoft.VisualBasic]Microsoft.VisualBasic.Conversion::Fix(float64)
L_000f: call float64 [mscorlib]System.Math::Round(float64)
L_0014: conv.ovf.i4
L_0015: stloc.0
L_0016: ldc.r8 7.9
L_001f: call float64
[Microsoft.VisualBasic]Microsoft.VisualBasic.Conversion::Fix(float64)
L_0024: call int32 [mscorlib]System.Convert::ToInt32(float64)
L_0029: stloc.1
I assume also don't assume they have taken the Math round for nothing
instead of the convert.
Cor
"Armin Zingler" wrote in message
Am 11.01.2011 19:34, schrieb Cor:
Correct, I do not want to use CInt (or Convert.ToInt32).
No, they shouldn't change CInt (or Convert.ToInt32), but it would be nice
to have an additional function the does not perform unnecessary rounding.
Just take the whole part before the "." as an Integer. That's all. No magic.
Very simple. It's not expected _more_, it's expected _less_.
this code
Dim x As Integer = CInt(Fix(7.9))
Dim y As Integer = Convert.ToInt32(Fix(7.9))
leads to this IL not so much in my idea.
..locals init (
[0] int32 x,
[1] int32 y)
L_0000: nop
L_0001: ldc.r8 7.9
L_000a: call float64
[Microsoft.VisualBasic]Microsoft.VisualBasic.Conversion::Fix(float64)
L_000f: call float64 [mscorlib]System.Math::Round(float64)
L_0014: conv.ovf.i4
L_0015: stloc.0
L_0016: ldc.r8 7.9
L_001f: call float64
[Microsoft.VisualBasic]Microsoft.VisualBasic.Conversion::Fix(float64)
L_0024: call int32 [mscorlib]System.Convert::ToInt32(float64)
L_0029: stloc.1
I assume also don't assume they have taken the Math round for nothing
instead of the convert.
Cor
"Armin Zingler" wrote in message
Am 11.01.2011 19:34, schrieb Cor:
Correct, I do not want to use CInt (or Convert.ToInt32).
No, they shouldn't change CInt (or Convert.ToInt32), but it would be nice
to have an additional function the does not perform unnecessary rounding.
Just take the whole part before the "." as an Integer. That's all. No magic.
Very simple. It's not expected _more_, it's expected _less_.
A
Armin Zingler
Am 11.01.2011 20:21, schrieb Cor:
Also look inside Convert.ToInt32. I've posted it already. Why so much
rounding code for nothing? Again, the decimal places are zero after
calling Fix. There is nothing to round.
Also look inside Convert.ToInt32. I've posted it already. Why so much
rounding code for nothing? Again, the decimal places are zero after
calling Fix. There is nothing to round.
A
Armin Zingler
Am 11.01.2011 18:20, schrieb Nobody:
Works!
'Modified sample code from the documentation. (quick'n'dirty)
Imports System.Reflection
Imports System.Reflection.Emit
Public Class Form1
Private Delegate Function ToInt32Delegate(ByVal value As Double) As Integer
Private Shared Function CreateMyPersonalToInt32() As ToInt32Delegate
Dim aName As New AssemblyName("DynamicAssemblyExample")
Dim ab As AssemblyBuilder = _
AppDomain.CurrentDomain.DefineDynamicAssembly( _
aName, _
AssemblyBuilderAccess.Run)
Dim mb As ModuleBuilder = ab.DefineDynamicModule(aName.Name)
Dim tb As TypeBuilder = mb.DefineType("MyDynamicType", TypeAttributes.Public)
Dim parameterTypes As Type() = {GetType(Double)}
Dim methodBuilder = tb.DefineMethod("ToInt32", _
MethodAttributes.Static Or MethodAttributes.Public, GetType(Int32), parameterTypes)
Dim Generator As ILGenerator = methodBuilder.GetILGenerator()
Generator.Emit(OpCodes.Ldarg_0)
Generator.Emit(OpCodes.Conv_I4)
Generator.Emit(OpCodes.Ret)
Dim t As Type = tb.CreateType()
Dim mi As MethodInfo = t.GetMethod("ToInt32", BindingFlags.Public Or BindingFlags.Static)
Return DirectCast([Delegate].CreateDelegate(GetType(ToInt32Delegate), t, "ToInt32"), ToInt32Delegate)
End Function
Protected Overrides Sub OnLoad(ByVal e As System.EventArgs)
MyBase.OnLoad(e)
Dim MyConversionDelegate = CreateMyPersonalToInt32()
Dim i As Integer = MyConversionDelegate(7.9)
End Sub
End Class
Thank you for the hint! You know, I did this just out of interest.
Works!
'Modified sample code from the documentation. (quick'n'dirty)
Imports System.Reflection
Imports System.Reflection.Emit
Public Class Form1
Private Delegate Function ToInt32Delegate(ByVal value As Double) As Integer
Private Shared Function CreateMyPersonalToInt32() As ToInt32Delegate
Dim aName As New AssemblyName("DynamicAssemblyExample")
Dim ab As AssemblyBuilder = _
AppDomain.CurrentDomain.DefineDynamicAssembly( _
aName, _
AssemblyBuilderAccess.Run)
Dim mb As ModuleBuilder = ab.DefineDynamicModule(aName.Name)
Dim tb As TypeBuilder = mb.DefineType("MyDynamicType", TypeAttributes.Public)
Dim parameterTypes As Type() = {GetType(Double)}
Dim methodBuilder = tb.DefineMethod("ToInt32", _
MethodAttributes.Static Or MethodAttributes.Public, GetType(Int32), parameterTypes)
Dim Generator As ILGenerator = methodBuilder.GetILGenerator()
Generator.Emit(OpCodes.Ldarg_0)
Generator.Emit(OpCodes.Conv_I4)
Generator.Emit(OpCodes.Ret)
Dim t As Type = tb.CreateType()
Dim mi As MethodInfo = t.GetMethod("ToInt32", BindingFlags.Public Or BindingFlags.Static)
Return DirectCast([Delegate].CreateDelegate(GetType(ToInt32Delegate), t, "ToInt32"), ToInt32Delegate)
End Function
Protected Overrides Sub OnLoad(ByVal e As System.EventArgs)
MyBase.OnLoad(e)
Dim MyConversionDelegate = CreateMyPersonalToInt32()
Dim i As Integer = MyConversionDelegate(7.9)
End Sub
End Class
Thank you for the hint! You know, I did this just out of interest.
N
Nobody
Armin Zingler said:
It would be interesting to see if there is any overhead that makes it as
slow as the alternatives that you have tried, and compared to the overhead
in calling the DLL function.
A
Armin Zingler
Am 11.01.2011 21:28, schrieb Nobody:
I did some performance tests. Before posting the source and the results,
I wanted to make the code look neater. With the "redesigned" version, I
got different results even though it does exactly the same. Although I'm
aware of the side-effects of optimizations, there seem to be several
influences that don't allow me to post robust figures.
Despite, and only for the sake of completeness, I post the source, so you
can play with and modify it and make your own measure experiences. I don't
comment it now.
For my part, these files can be closed now.
Option Infer On
Option Strict On
Imports Microsoft.VisualBasic
Imports System
Imports System.Diagnostics
Imports System.Reflection
Imports System.Reflection.Emit
Friend Class Main
Private Delegate Function ToInt32Delegate(ByVal value As Double) As Integer
Private Shared ReadOnly ConversionDelegate As ToInt32Delegate = CreateConversionDelegate()
Private Shared D As Double = 7.9R
Private Shared i, j As Integer
Shared Sub Main()
Dim duration As TimeSpan
duration = RunTest()
MsgBox(duration.ToString)
duration = RunTest()
MsgBox(duration.ToString)
End Sub
Private Shared Function RunTest() As TimeSpan
Dim watch As Stopwatch
watch = Stopwatch.StartNew
For Project1.Main.i = 1 To 1000000000
' optimized not optimized
'Project1.Main.j = CInt(D) '23.3 16.3
'Project1.Main.j = ConversionDelegate(D) ' 6.9 14.5
'Project1.Main.j = DoubleToIntTableHelper.Main.ToInt32(D) ' 5.6 16.6
'Project1.Main.j = Convert.ToInt32(D) '11.9 22.0
Next
watch.Stop()
Return watch.Elapsed
End Function
Private Shared Function CreateConversionDelegate() As ToInt32Delegate
Dim AssyName As New AssemblyName("ConversionHelper")
Dim AssyBuilder = AppDomain.CurrentDomain.DefineDynamicAssembly(AssyName, AssemblyBuilderAccess.Run)
Dim ModuleBuilder = AssyBuilder.DefineDynamicModule(AssyName.Name)
Dim TypeBuilder = ModuleBuilder.DefineType("Main", TypeAttributes.Public)
Dim ParameterTypes As Type() = {GetType(Double)}
Dim MethodBuilder = TypeBuilder.DefineMethod( _
"ToInt32", MethodAttributes.Static Or MethodAttributes.Public, GetType(Int32), ParameterTypes)
Dim Generator As ILGenerator = MethodBuilder.GetILGenerator()
Dim t As Type
Generator.Emit(OpCodes.Ldarg_0)
Generator.Emit(OpCodes.Conv_I4)
Generator.Emit(OpCodes.Ret)
t = TypeBuilder.CreateType()
Return DirectCast( _
[Delegate].CreateDelegate(GetType(ToInt32Delegate), t, MethodBuilder.Name), _
ToInt32Delegate _
)
End Function
End Class
I did some performance tests. Before posting the source and the results,
I wanted to make the code look neater. With the "redesigned" version, I
got different results even though it does exactly the same. Although I'm
aware of the side-effects of optimizations, there seem to be several
influences that don't allow me to post robust figures.
Despite, and only for the sake of completeness, I post the source, so you
can play with and modify it and make your own measure experiences. I don't
comment it now.
For my part, these files can be closed now.
Option Infer On
Option Strict On
Imports Microsoft.VisualBasic
Imports System
Imports System.Diagnostics
Imports System.Reflection
Imports System.Reflection.Emit
Friend Class Main
Private Delegate Function ToInt32Delegate(ByVal value As Double) As Integer
Private Shared ReadOnly ConversionDelegate As ToInt32Delegate = CreateConversionDelegate()
Private Shared D As Double = 7.9R
Private Shared i, j As Integer
Shared Sub Main()
Dim duration As TimeSpan
duration = RunTest()
MsgBox(duration.ToString)
duration = RunTest()
MsgBox(duration.ToString)
End Sub
Private Shared Function RunTest() As TimeSpan
Dim watch As Stopwatch
watch = Stopwatch.StartNew
For Project1.Main.i = 1 To 1000000000
' optimized not optimized
'Project1.Main.j = CInt(D) '23.3 16.3
'Project1.Main.j = ConversionDelegate(D) ' 6.9 14.5
'Project1.Main.j = DoubleToIntTableHelper.Main.ToInt32(D) ' 5.6 16.6
'Project1.Main.j = Convert.ToInt32(D) '11.9 22.0
Next
watch.Stop()
Return watch.Elapsed
End Function
Private Shared Function CreateConversionDelegate() As ToInt32Delegate
Dim AssyName As New AssemblyName("ConversionHelper")
Dim AssyBuilder = AppDomain.CurrentDomain.DefineDynamicAssembly(AssyName, AssemblyBuilderAccess.Run)
Dim ModuleBuilder = AssyBuilder.DefineDynamicModule(AssyName.Name)
Dim TypeBuilder = ModuleBuilder.DefineType("Main", TypeAttributes.Public)
Dim ParameterTypes As Type() = {GetType(Double)}
Dim MethodBuilder = TypeBuilder.DefineMethod( _
"ToInt32", MethodAttributes.Static Or MethodAttributes.Public, GetType(Int32), ParameterTypes)
Dim Generator As ILGenerator = MethodBuilder.GetILGenerator()
Dim t As Type
Generator.Emit(OpCodes.Ldarg_0)
Generator.Emit(OpCodes.Conv_I4)
Generator.Emit(OpCodes.Ret)
t = TypeBuilder.CreateType()
Return DirectCast( _
[Delegate].CreateDelegate(GetType(ToInt32Delegate), t, MethodBuilder.Name), _
ToInt32Delegate _
)
End Function
End Class