AlphaBlend and 16-bit color mode - windows

Question

Win32 graphics is not my gig, but I have to do some alpha blending. The following code works fine in 32-bit color mode but displays nothing except the white background in 16-bit mode. Sorry for the length, but I don't know where it's going wrong. This is as compact as I could make it.
hbm is a 32-bit ARGB bitmap with varying per-pixel alpha, size 16x16 (so, cx = cy = 16).
// Create a memory DC to construct the bits
HDC hdc = GetDC(hWnd);
HDC hdcMem = CreateCompatibleDC(hdc);
HBITMAP hbmMem = CreateBitmap(cx, cy, 1, 32, NULL);
SelectObject(hdcMem, hbmMem);
// Fill the BG
RECT rc = { 0, 0, cx, cy };
FillRect(hdcMem, &rc, (HBRUSH)GetStockObject(WHITE_BRUSH));
// Get the bitmap bits
BITMAPINFO bmi;
ZeroMemory(&bmi, sizeof(bmi));
bmi.bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
bmi.bmiHeader.biWidth = cx;
bmi.bmiHeader.biHeight = cy;
bmi.bmiHeader.biBitCount = 32;
bmi.bmiHeader.biPlanes = 1;
bmi.bmiHeader.biCompression = BI_RGB;
std::unique_ptr<BYTE[]> pvBits(new BYTE[cx * cy * 4]);
GetDIBits(hdcMem, hbm, 0, cy, reinterpret_cast<void*>(pvBits.get()), &bmi, DIB_RGB_COLORS);
// Premultiply all color channel values by the per-pixel alpha.
int ctPixels = cx * cy;
BYTE *prgba = pvBits.get();
for (int i = 0; i < ctPixels; ++i)
{
int alpha = *(prgba + 3);
for (int j = 0; j <= 2; ++j)
{
int k = *prgba;
*prgba++ = k * alpha / 255;
}
++prgba;
}
// Put the new bits back
SetDIBits(hdcMem, hbm, 0, cy, reinterpret_cast<void*>(pvBits.get()), &bmi, DIB_RGB_COLORS);
// Alpha blend into memory DC
HDC hdcSrc = CreateCompatibleDC(hdcMem);
HBITMAP hbmOld = static_cast<HBITMAP>(SelectObject(hdcSrc, hbm));
BLENDFUNCTION bfn = { AC_SRC_OVER, 0, 255, AC_SRC_ALPHA };
AlphaBlend(hdcMem, 0, 0, cx, cy, hdcSrc, 0, 0, cx, cy, bfn);
SelectObject(hdcSrc, hbmOld);
DeleteDC(hdcSrc);
// Blit the memory DC to the screen
BitBlt(hdc, 0, 0, cx, cy, hdcMem, 0, 0, SRCCOPY);
I have some vague suspicions about CreateCompatibleDC but other than that I'm flying blind.
Any help appreciated. TIA.

Answer 1

According to Microsoft's documentation per pixel alpha values are only supported with 32-bit bitmaps. SetDIBits converts your 32-bit DIB into a 16-bit DDB if the display using 16-bit colour, and a 16-bit DDB has no space to store alpha values. You'll have to set bfn.SourceConstantAlpha to your alpha value instead to get this to work. You won't need to premultiply your bitmap in that case either.