SizeT – .NET and native code


In this post I wanted to introduce you with a new construct we added to the latest release of CUDA.NET (2.3.6) and will be available with the published OpenCL.NET library.

The problem

.NET is a very fixed environment, defining well known types, such that an int is always 4 bytes long (32 bit) and a long is always 8 bytes long (64 bit).

This is not the case with native code, for developers of C/C++. Writing a program in 32 bit environment, will always yield 32 bit types, unless using specific directives to get 64 bit variables. When writing 64 bit programs, they do get access to 64 bit wide variables as primitives supported by the compiler.

This clearly creates a portability problem for code and applications written in 32 and 64 bit environments.

Another example, is pointer size, where in C/C++ environments, under 32 bit the pointer is 4 bytes wide (int) and under 64 bit systems it is 8 bytes wide (long). The .NET environment (through different languages) provides a simple construct to overcome this problem, namely the IntPtr object, which some of you may be familiar with.

Now, coming back to our domain, the runtime API (also the driver in a new CUDA 2.3 function) and OpenCL makes extensive use of the C/C++ size_t data type. This data type ensures for developers that under different environments they will get the maximum width of the supported data type, unsigned int for 32 bit systems and unsigned long for 64 bit systems.

Possible options

By means of the interoprating library (wrapper), such as CUDA.NET, it creates a problem, since the API should provide several versions of the function, one given an uint (to map to 32 bit with unsigned int C/C++ type), and ulong (to map against unsigned long in 64 bit C/C++ systems). Supplying such an interface to the user will have to force him a specific behavior and system, since in .NET, the uint is always 32 bit wide, and ulong is 64 bit wide, no matter what.

Another option can be to provide a unique, standalone interface, using the IntPtr object, since .NET takes care to make it 32 bit wide in 32 bit systems and 64 bit wide for 64 bit systems, dynamically, without user intervention.

But using the IntPtr and a very serious downside, it is not dynamic, once it’s value is set, it cannot be changed through simple arithmetic operators, like +,-,*,/ or else.

The solution

Exactly for this purpose we created the SizeT object (structure). First, it maps to the same name as it’s native counterpart (size_t) and second it provides the dynamic mechanisms we want for working with 32 or 64 bit systems transparently.

SizeT can serve just like any other basic primitive in .NET.
For example:

SizeT temp = 15;
uint value = (uint)temp;
ulong value2 = (ulong)temp;
temp = value;

Internally, the SizeT wraps the IntPtr object to provide the same dynamic capabilities under 32 and 64 bit platforms.
It can host the required .NET primitives (int, uint, long, ulong), so when programming, one will make a good habit for using the SizeT instead of other data types (working with the runtime CUDA API).

For OpenCL the interface was built from the first place to use SizeT in mind, as the OpenCL API uses only size_t data types for cross platform functions.

Advanced data types with CUDA

Following with CUDA.NET 2.3.6 release, this article is meant to show you so of the more advanced constructs .NET can offer developers willing to get advanced interoperability with native code.
As most of you ar familiar, CUDA.NET offers to copy many types of arrays and data types to the GPU memory (through the different memcpy functions). These are based on well defined data types, mostly for numerical purposes.

Consider a basic data type of float, the corresponding array is declared as: float[], in C# or otherwise in different languages, but the principle is the same. In addition to these primitives (byte, short, int, long, float, double) there is also support for vector data types that CUDA support, such as Float2, where it is composed of 2 consequtive float elements.

What happens when you want to pass more complex data types that are not supported by CUDA.NET?

In this case, there are several techniques to achieve this goal, some maybe more complex to empploy than others, and it mostly depend on your expected usage.

1. Declaring a new copy function

Well, that’s always an option if you wish to extend the API of functions. In such case, the developer declares a new copy function to use, with expected parameters and consumes it.

The following example can show a little more:

// This is a dummy, complex data type
struct Test
public int value1;
public float value2;

// Define a new copy function to use with CUDA, assuming running under Linux
public static extern CUResult cuMemcpyHtoD(CUdeviceptr dst, Test[] src, uint bytes);

The definition above is for a function, to use, capable of copying data from an array of Test objects to device memory.
But, it may not always be convenient.

2. The dynamic, simpler way

Well, .NET offers one more possibility to convert .NET objects into native representation, without using “unsafe” mechanisms.

For this purpose, there is an object called “GCHandle” to use. This object provides an advanced control over the garbage collector of .NET to lock objects in memory and get their native pointer (IntPtr in .NET).

Since all copy functions in CUDA.NET support the IntPtr data type, one can use this mechanism as a generic way to copy data to the GPU. In practice, when a user calls one of the existing copy functions, the exact process is performed.

Again, consider the Test structure we created before.

// Getting native handle from an array
Test[] data = new Test[100];
// Fill in the array values...
GCHandle ptr = GCHandle.Alloc(data, GCHandleType.Pinned);
IntPtr src = ptr.AddrOfPinnedObject();
// Now copy to the GPU memory from this pointer...
// When finished, don't forget to free the GCHandle!

This is a simple process for exposing complex .NET data types to CUDA and CUDA.NET to be processed by the GPU.

In the next article we will present the new SizeT object we added for portability between 32 and 64 bit systems.