How to deal alpha blending with cuda or opencl efficiently

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi Everyone,
Does any one know how to deal alpha blending with opencl efficiently? I had ported the host code to opencl and cuda, but the performance come out too poor. does anybody has any advice?

2

Hi Everyone,
Following is the code of opencl kernel for alpha blending. Can anyone give me some suggestion or guide me to optimize the opencl kernel.

uint applyAlphaToPixel(uint uPixel, uint uAlpha)
{
uint uColor = uPixel;
uPixel &= 0x00FF00FF;
uPixel = uAlpha;
uPixel += 0x00800080;
uPixel &= 0xFF00FF00;
uPixel >>= 8;
uColor &= 0xFF00FF00;
uColor >>= 8;
uColor = uAlpha;
uColor += 0x00800080;
uColor &= 0xFF00FF00;
uColor |= uPixel;
return uColor;
}
__kernel void alphaBlend_kernel(__global uint pTarget, __global uint pSource, const uint width, const uint height)
{
uint gx = get_global_id(0);
uint gy = get_global_id(1);
if (gx >= width || gy >= height)
return;
uint id = gy * width + gx;
uint tc = pTarget[id];
uint sc = pSource[id];
uint sa = sc >> 24;
if (sa == 0x0) {
pTarget[id] = tc;
} else if (sa == 0xFF) {
pTarget[id] = sc;
} else {
sc = applyAlphaToPixel(sc, sa);
tc = applyAlphaToPixel(tc, (0xFF ^ sa));
pTarget[id] = sc + tc;
}
}