I'm trying to sample portions of the screen several times a second while hooked into a DX9 app. I've read that GetFrontBuffer is extremely slow, the code I have now uses GetRenderTargetData and it also causes large slow downs. Is it possible to read out (portions of) the front buffer without taking a huge hit in performance? The sampling rate dosen't have to be very great, 2 to 4 frames a second, and I do expect to take some hit in performance, but I also know there is a better way than this.
1. Lock the back buffer.
2. Get a pointer to it.
3. Use an assembly loop to copy dwords from back buffer or secondary surface to your buffer.
You could use a FOR loop but you would want to make sure it is using REP MOVSD and not REP MOVSB with a value of (buffer_size*4) in ECX.
I think most compilers should use REP MOVSD.
It should be mentioned that the C standard library memory functions like memcpy, memmove, etc., DO NOT use anything but STOSB and MOVSB. They copy bytes at a time which will actually decrease performance by a factor of 4.
Last edited by VirtualAce; 05-04-2005 at 06:44 AM.
[edit] silly comment, no need for it to be here, and cant find a way to remove the post.
Last edited by skorman00; 05-04-2005 at 10:48 AM.
