Re: Alignment question

From: Aslan Kral (spamtrap_at_crayne.org)
Date: 02/27/05

  • Next message: Aslan Kral: "Re: Alignment question" 
    Date: Sun, 27 Feb 2005 19:40:54 +0000 (UTC)

    "kingzog" <spamtrap@crayne.org>, haber iletisinde sunlari
    yazdi:1109454150.697734.259800@o13g2000cwo.googlegroups.com...
    > Aslan Kral wrote:
    > > The following C++ class is used to jump to the first pointer in the
    > > vtable.
    > > class KWndProc
    > > {
    > > uint32 jmp_buf[3];
    > > protected:
    > > virtual LRESULT WndProc(HWND hWnd,UINT nMessage,WPARAM
    > > wParam,LPARAM lParam)=0;
    > >
    > >
    > > public:
    > > #if 1
    > > KWndProc()
    > > {
    > > jmp_buf[0] = 0xB9909090; // mov ecx,
    > > jmp_buf[1] = (uint32)this;// this
    >
    >
    > In Visual C++, the default for non vararg functions is thiscall, so
    > ecx=this at all times, so you could skip these two lines.
    >
    >
    http://msdn.microsoft.com/library/default.asp?url=/library/en-us/vccore98/ht
    ml/_core_results_of_calling_example.asp
    >
    > I think you should have a pure C++ version too, so it will build for
    > Win64 - just put in a regular function call, e.g.
    >
    > #ifdef _M_X86
    > // asm version
    > #else
    > // c++ version
    > func();
    > #end
    >
    > Have you checked the ATL source code? It has a nice technique for
    > generating some thunk code for a WndProc that changes the first
    > parameter from a HWND to this.
    >
    No. But I have now. Basically it is the same logic (Jumping and registering
    classes, but they use different instructions). My class is defined a little
    bit different. I simply derive a class from it and implement the windows
    procedure (virtual function) in it.

    > (snip version where this is aligned)
    >
    > >
    > > Which one is better for alignment? The first pointer in the vtable
    > > is a window procedure so it will be called many times by windows
    > > through this thunk data.
    >
    > I don't know to be honest - I'd try testuing them, or counting cycles.
    > I'd guess that the CPU doesn't care if the immediate data after the mov
    > is aligned or not, so the version with the NOPs will be slower.
    >
    > Zog.
    >

  • Next message: Aslan Kral: "Re: Alignment question"