Improving screen capture performance - c

Question

I am going to create some kind of "remote desktop" application that streams the content of the screen over a socket to a connected client.
In order to take a screenshot, I've come up with the following piece of code, which is a modified version of examples I've seen here and there.
#include <windows.h>
#include <tchar.h>
#include <stdio.h>
int _tmain( int argc, _TCHAR * argv[] )
{
int ScreenX = 0;
int ScreenY = 0;
BYTE* ScreenData = 0;
HDC hScreen = GetDC(GetDesktopWindow());
ScreenX = GetDeviceCaps(hScreen, HORZRES);
ScreenY = GetDeviceCaps(hScreen, VERTRES);
ScreenData = (BYTE*)calloc(4 * ScreenX * ScreenY, sizeof(BYTE) );
BITMAPINFOHEADER bmi = {0};
bmi.biSize = sizeof(BITMAPINFOHEADER);
bmi.biPlanes = 1;
bmi.biBitCount = 32;
bmi.biWidth = ScreenX;
bmi.biHeight = -ScreenY;
bmi.biCompression = BI_RGB;
bmi.biSizeImage = 0; // 3 * ScreenX * ScreenY;
int iBegTc = ::GetTickCount();
// Take 100 screen captures for a more accurante measurement of the duration.
for( int i = 0; i < 100; ++i )
{
HBITMAP hBitmap = CreateCompatibleBitmap(hScreen, ScreenX, ScreenY);
HDC hdcMem = CreateCompatibleDC (hScreen);
HGDIOBJ hOld = SelectObject(hdcMem, hBitmap);
BitBlt(hdcMem, 0, 0, ScreenX, ScreenY, hScreen, 0, 0, SRCCOPY);
SelectObject(hdcMem, hOld);
GetDIBits(hdcMem, hBitmap, 0, ScreenY, ScreenData, (BITMAPINFO*)&bmi, DIB_RGB_COLORS);
DeleteDC(hdcMem);
DeleteObject(hBitmap);
}
int iEndTc = ::GetTickCount();
printf( "%d ms", (iEndTc - iBegTc) / 100 );
system("PAUSE");
ReleaseDC(GetDesktopWindow(),hScreen);
return 0;
}
My problem is that the code within the loop takes too long too execute. In my case it's about 36 ms per iteration.
I am wondering if there are statements that could be done just once and thus put outside of the loop, likI did for the byte buffer. I don't know however which are the ones that I must do for each new image, and which are the ones I can only do one time.

Answer 1

Keep BitBlt and GetDIBits inside the loop, move the rest outside the loop as follows:
HBITMAP hBitmap = CreateCompatibleBitmap(hScreen, ScreenX, ScreenY);
HDC hdcMem = CreateCompatibleDC (hScreen);
HGDIOBJ hOld = SelectObject(hdcMem, hBitmap);
for( int i = 0; i < 100; ++i )
{
BitBlt(hdcMem, 0, 0, ScreenX, ScreenY, hScreen, 0, 0, SRCCOPY);
//hBitmap is updated now
GetDIBits(hdcMem, hBitmap, 0, ScreenY, ScreenData, (BITMAPINFO*)&bmi, DIB_RGB_COLORS);
//wait...
}
SelectObject(hdcMem, hOld);
DeleteDC(hdcMem);
DeleteObject(hBitmap);
In addition bmi.biSizeImage should be set to data size, in this case 4 * ScreenX * ScreenY
This won't make the code noticeably faster. The bottle neck is at BitBlt. It's still about 30 frames/sec, this should be okay unless there is a game or movie on the screen.
You might also try saving to a 24 bit bitmap. It won't make any difference in this code but data size would be smaller ((width * bitcount + 31) / 32) * 4 * height)

Answer 2

The Aero feature of Windows seems to affect the BitBlt speed.
If you iteratively BitBlt even one pixel from the display, it will run at about 30 frames per second, and the CPU usage will be near idle. But if you turn off the Aero feature of Windows, you'll get BitBlt speeds that are remarkably faster.