AccessViolationException calling Fortran from C#

Question

I'm trying to wrap the glmnet library (http://cran.r-project.org/web/packages/glmnet/index.html) so I can solve ~~models~~ sparse general linear models in C#. However, the original function has a somewhat 20 parameters, so I started (completely new to Fortran) with a tiny subroutine for testing how to pass data. Unfortunately I always get an AccessViolationException.

Here's the code:

The Fortran subroutine. I compile it into a dll using the gfortran compiler that comes with Rtools (http://cran.r-project.org/bin/windows/Rtools/), using the -m64 option (yes, 64bit is neccessary since I handle quite big chunks of data). Yes, the use of i could lead to out-of-bounds... but this is just for testing.

subroutine testaaa  (f,i,fa,ia)
real fa(i)                                                      
integer ia(i)
ia(1) = 1337
ia(i) = 666
fa(1) = 8.15
fa(i) = 333
end subroutine testaaa

The C# PInvoke code:

[DllImport("ftest.dll", EntryPoint = "testaaa_", CallingConvention = CallingConvention.StdCall)]
public static extern void Test(
    [MarshalAs(UnmanagedType.R4)] float f,
    [MarshalAs(UnmanagedType.I4)] int i,
    IntPtr fa,
    IntPtr ia);

And here is how it's called:

var fa = new float[4];
var ia = new int[4];
IntPtr faPtr = Marshal.AllocHGlobal(fa.Length * sizeof(float));
Marshal.Copy(fa, 0, faPtr, fa.Length);

IntPtr iaPtr = Marshal.AllocHGlobal(ia.Length * sizeof(float));
Marshal.Copy(ia, 0, iaPtr, ia.Length);

GlmnetDllWrapper.Test(0.4f, 4,faPtr,iaPtr);

I also tried passing the arrays directly and giving them the [MarshalAs(UnmanagedType.LPArray)] attribute. Nothing worked for me.

Do you have any suggestions where to start or what to change?

Update 1: Even passing only the float and the int already causes the exception:

subroutine testbbb  (f,i)
i = 815
f = 8.15
return
end subroutine testbbb

C# Pinvoke and call are changed accordingly. What am I doing wrong?

Is is possible that your Fortan library receives the parameters by pointer? Also, why are you testing everything all at once. Why have you not simplified? Can you pass a single parameter of type `int`? Of type `float`? Then try an array. Don't test the most complicated thing first. How can you tell which part fails? — David Heffernan, Mar 02 '13 at 09:19
OK, your update and my comment crossed. Well done for simplifying. Try passing as pointers: `public static extern void Test(ref float f, ref int i)` — David Heffernan, Mar 02 '13 at 09:21
Once you get past the blockage on the simple types, we can show you how to do the arrays. And there's no need for `IntPtr`. We can get the pinvoke marshaller to pin the arrays and make the calling code very simple. — David Heffernan, Mar 02 '13 at 09:37
@DavidHeffernan `ref` does the job. Passing the arrays without any attribute works. Adding the `LPArray` attribute is just as fine. Manipulating arrays is also fine. Trying a more complex szenario now. Will comment when there's more to tell. — Ahue, Mar 02 '13 at 09:43

score 2 · Accepted Answer · answered Mar 02 '13 at 09:49

2

The main problem is that your Fortran library expects the scalar parameters to be passed by reference. So you need to declare your p/invoke to match.

The array parameters can be passed quite simply as arrays and the p/invoke marshaller will pin them for you.

So, your p/invoke declaration should be like this:

[DllImport("ftest.dll", EntryPoint = "testaaa_")]
public static extern void Test(
    ref float f,
    ref int i,
    [In, Out] float[] fa,
    [In, Out] int[] ia
);

You can adjust the [In, Out] attributes to meet your needs.

answered Mar 02 '13 at 09:49

David Heffernan

601,492
42
1,072
1,490

You'd better put CalingConvention back, Fortran is almost always stdcall. More about gfortran [here](http://gcc.gnu.org/onlinedocs/gcc-4.7.0/gfortran/Interoperability-with-C.html#Interoperability-with-C) – Hans Passant Mar 02 '13 at 12:12
@HansPassant `StdCall` is the default which is why I removed it. Or am I misinterpreting you? – David Heffernan Mar 02 '13 at 14:18
Yes, you're right, I fumbled that, sorry. The link I left should give the OP hints what calling convention and whether or not *ref* is required from the Fortran source code. The VALUE and USE statements matter. – Hans Passant Mar 02 '13 at 16:57
Yep, I think that's it! Thanks a lot for your help! The glmnet library is working... now I have to deal with its output. However, I left out the `In` and `Out` arguments. It works with and without them. Maybe adding them back makes the code more readable - or this there more to this? – Ahue Mar 03 '13 at 10:15
1

There is a litte more. Read @Hans's answer to this question: http://stackoverflow.com/questions/14366066/are-p-invoke-in-out-attributes-optional-for-marshaling-arrays – David Heffernan Mar 03 '13 at 11:35

Michael · Answer 2 · 2013-03-02T18:43:13.683

-1

Please have a look at http://msdn.microsoft.com/en-en/library/chfa2zb8%28v=VS.80%29.aspx (unsafe code) and play around with your project settings Project/Properties/Build: allow unsafe code. But be aware of the consequences. :)

Update: Do not "play around" - I meant: "Check out the "unsafe" features". "Unsafe" doesn't mean "dangerous".

edited Mar 02 '13 at 18:43

answered Mar 02 '13 at 09:31

Michael

621
5
17

1

No, please do not do this. There's no need for unsafe code. And "play around with your settings" does not an answer make. – David Heffernan Mar 02 '13 at 09:32
Yes, sorry. I meant: "Try wrapping the area with unsafe code and allow unsafe code once" to see if the issue is located there. It's a way to handle pointers especially while working with ext. non-.Net assemblies, if "ref" doesn't work. But you're right. It's no need, because a simple "ref" works. – Michael Mar 02 '13 at 09:50

AccessViolationException calling Fortran from C#

2 Answers2