import std.cpuid;
import std.stdio;

//-----------------------------------------------------------------------------
//  This code goes into the library
const uint  CPU_NO_EXT  = 0,
            CPU_MMX     = 1,
            CPU_SSE     = 2,
            CPU_SSE2    = 4,
            CPU_SSE3    = 8;

/******************************************************************************
    Set initializer to first instance that matches the available hardware
******************************************************************************/
void BindMTV(alias T, E...)(ClassInfo ci, size_t offset)
{
    uint extensions;
    if ( mmx )  extensions |= CPU_MMX;
    if ( sse )  extensions |= CPU_SSE;
    if ( sse2 ) extensions |= CPU_SSE2;
    if ( sse3 ) extensions |= CPU_SSE3;

    foreach ( i, ext; E )
    {
        if ( (E[i] & extensions) == E[i] ) {
            *cast(void**)&ci.init[offset] = &T!(cast(uint)E[i]);
            return;
        }
    }
    throw new Exception("unsupported architecture "~std.string.toString(extensions));
}

//-----------------------------------------------------------------------------
//  This is application code

/******************************************************************************
    A class with a multi-target function
******************************************************************************/
class MyMultiTargetClass
{
    char[] function() multi_target; /// recieves the pointer to the preferred version
    // a template that generates multiple version for different targets
    char[] multi_target_mtv(uint arch)()
    {
        //pragma(target, arch)
        {
            // target independent code, e.g. vector operations
            return "using "~std.string.toString(arch);
        }
    }

    static this()
    {
        // genereate 4 different versions and activate first match
        BindMTV!(multi_target_mtv, CPU_SSE3, CPU_SSE2, CPU_SSE|CPU_MMX, CPU_NO_EXT)(this.classinfo, multi_target.offsetof);
    }
}


/******************************************************************************
    Finally, the usage is completely opaque and there is no runtime overhead
    besides the detection at startup.
******************************************************************************/
void main()
{
    MyMultiTargetClass t = new MyMultiTargetClass;
    writefln("%s", t.multi_target());
}