From: "Samson Fu" <samsonfu@hotmail.com>
Newsgroups: borland.public.delphi.graphics
References: <3b0de118$1_1@dnews> <Xns90AC7A1978F4Bmatthijs@127.0.0.1> <3b0f28e6_2@dnews> <Xns90ADB8436752Dmatthijs@127.0.0.1>
Subject: Re: Xor 2 Bitmap.
Date: Mon, 28 May 2001 11:26:52 +0800
Lines: 171
X-Priority: 3
X-MSMail-Priority: Normal
X-Newsreader: Microsoft Outlook Express 5.50.4522.1200
X-MimeOLE: Produced By Microsoft MimeOLE V5.50.4522.1200
NNTP-Posting-Host: 202.69.84.215
Message-ID: <3b11c61e_2@dnews>
X-Trace: dnews 991020574 202.69.84.215 (27 May 2001 20:29:34 -0700)
Path: dnews
Xref: dnews borland.public.delphi.graphics:39239

Thank you! Thank you!!
It 8 times faster than Bitblt for Intel CPU.

Thank you again!

Samson Fu


"Matthijs Laan" <matthijsln@nospam-remove.xs4all.nl> wrote in message
news:Xns90ADB8436752Dmatthijs@127.0.0.1...
> According to Samson Fu,
>
> > "Matthijs Laan" <matthijsln@nospam-remove.xs4all.nl> wrote in message
> > news:Xns90AC7A1978F4Bmatthijs@127.0.0.1...
> >> According to Samson Fu,
> >>
> >> > I want to Xor 2 Tbitmaps (BMP1 and BMP2), and save the result to
> >> > BMP3. It Bitblt the only choice?
> >> >
> >> > If I use Bitblt, I have to copy BMP1 to BMP3 first and Bitblt BMP2
> >> > and BMP3. (As Bitblt 2 Bitmaps).
>
> > I want to use it for any pixelformat. Can you kindly give me example to
> > use MMX for this case?
>
> Please don't top-post, thank you.
>
> Here it is. I haven't done any speed testing, though.
>
> The formatting might look a little strange, but the general idea is 8
space
> TABs. Assembler, you know ;-)
>
> procedure mmx_xor_bytes(P1, P2, P3: Pointer; bytes: Integer); register;
assembler;
> { P1 eax
>   P2 edx
>   P3 ecx
>   bytes [ebp+8] }
> asm
> push ebx
> push esi
> push edi
> mov edi, ecx
> mov esi, eax
> mov ecx, [ebp+8]
> test ecx, ecx
> jz @Quit
> and ecx, 4294967288 // ecx := bytes - (bytes mod 8)
> jz @NoQWORDs
> add edx, ecx // Move pointers to the end minus (bytes mod 8)
> add esi, ecx
> add edi, ecx
> shr ecx, 3 // ecx := bytes div 8 (the amount of QWORDs to process)
> neg ecx // [Pointer + ecx * 8] now points to the start
> @LoopQWORDs:
> db $0F,$6F,$04,$CE       /// movq    mm0, qword ptr [esi + ecx * 8]
> db $0F,$EF,$04,$CA       /// pxor    mm0, qword ptr [edx + ecx * 8]
> db $0F,$7F,$04,$CF       /// movq    qword ptr [edi + ecx * 8], mm0
> inc ecx
> jnz @LoopQWORDs
> @NoQWORDs:
> mov ecx, [ebp+8]
> and ecx, $7 // ecx = bytes mod 8
> jz @Quit
> xor ebx, ebx
> @LoopBYTEs:
> mov al, [esi + ebx] // Process the remaining bytes
> xor al, [edx + ebx]
> mov [edi + ebx], al
> inc ebx
> cmp ebx, ecx
> jnz @LoopBYTEs
> @Quit:
> pop edi
> pop esi
> pop ebx
> end;
>
> procedure EMMS; assembler;
> asm
> db $0F,$77               /// emms
> end;
>
> procedure mmx_xor_bitmaps(Source1, Source2, Dest: TBitmap);
> { Probably garbles 16th bit (pf15bit) or 4th byte (pf32bit) }
> var bpp, bits, bytes, y: Integer;
> begin
> if (Source1.Width <> Source2.Width) or (Source1.Height <> Source2.Height)
then
>         raise Exception.Create('mmx_xor_bitmaps: Source bitmaps are not
the same size');
>
> if Source1.PixelFormat <> Source2.PixelFormat then
>         raise Exception.Create('mmx_xor_bitmaps: Source bitmaps
pixelformats different');
>
> case Source1.PixelFormat of
> // These behave differently than Windows,
> // don't know how exactly how/why
> pf1bit: bpp := 1;
> pf4bit: bpp := 4;
> pf8bit: bpp := 8;
>
> // These should be the same as Windows
> pf15bit: bpp := 16; // They're aligned on WORDs anyway
> pf16bit: bpp := 16;
> pf24bit: bpp := 24;
> pf32bit: bpp := 32;
> else
>         raise Exception.Create('mmx_xor_bitmaps: Pixelformat not
supported');
>         end;
>
>         bits := Source1.Width * bpp;
>         bytes := bits div 8;
> //        if bits mod 8 <> 0 then Inc(bytes); // Hmm...
>
>         Dest.Width := Source1.Width;
>         Dest.Height := Source1.Height;
>         Dest.PixelFormat := Source1.PixelFormat;
>
> Source1.HandleType := bmDIB; // This could be changed back later
> Source2.HandleType := bmDIB;
> Dest.HandleType := bmDIB;
>
>         for y := 0 to Source1.Height - 1 do
>         mmx_xor_bytes(Source1.ScanLine[y], Source2.ScanLine[y],
Dest.ScanLine[y], bytes);
>
>         EMMS;
> end;
>
> For testing, make a form with 4 TImages, and load two different bitmaps
> in Image1 and Image2. Drop another button on the form; and make this the
> OnClick handler:
>
> procedure TForm1.Button1Click(Sender: TObject);
> var R: TRect;
> var j, k, l: word;
> begin
>         Image1.Picture.Bitmap.PixelFormat := pf32bit;
>         Image2.Picture.Bitmap.PixelFormat := pf32bit;
>         R :=
Rect(0,0,Image1.Picture.Bitmap.Width,Image1.Picture.Bitmap.Height);
>         Image3.Picture.Bitmap.Canvas.FillRect(R);
>         Image4.Picture.Bitmap.Canvas.FillRect(R);
>         Image3.Repaint;
>         Image4.Repaint;
> mmx_xor_bitmaps(Image1.Picture.Bitmap, Image2.Picture.Bitmap,
Image3.Picture.Bitmap);
>         Image3.Repaint;
>         Image4.Repaint;
>         Image4.Picture.Bitmap.Width := Image1.Picture.Bitmap.Width;
>         Image4.Picture.Bitmap.Height := Image1.Picture.Bitmap.Height;
>         Image4.Picture.Bitmap.PixelFormat := pf32bit;
>         Image4.Picture.Bitmap.Canvas.Draw(0,0,Image1.Picture.Bitmap);
> Image4.Picture.Bitmap.Canvas.CopyMode := cmSrcInvert;
>         Image4.Picture.Bitmap.Canvas.CopyRect(R,
Image2.Picture.Bitmap.Canvas, R);
> Image4.Picture.Bitmap.Canvas.CopyMode := cmSrcInvert;
> Image4.Picture.Bitmap.Canvas.CopyMode := cmSrcCopy;
>         Image3.Repaint;
>         Image4.Repaint;
> end;
>
> Hope this helps,
>
> Matthijs