Both the nvidia and ATI drivers compile this code without warning, but both exhibit apparently undefined behavior..
In D3D you can index a vector component with []. eg:
uint4 foo; foo[2] = 3; // now foo.z is 3.
The OpenCL compilers compile that, but they don't work!
eg
// Load in a and b terms
uint4 a, b;
/*
for(int i = 0; i < 4; ++i) {
a = pass1_values[leftIndex + i];
b = pass1_values[rightIndex + i];
}
*/
a.x = pass1_values[leftIndex + 0];
a.y = pass1_values[leftIndex + 1];
a.z = pass1_values[leftIndex + 2];
a.w = pass1_values[leftIndex + 3];
b.x = pass1_values[rightIndex + 0];
b.y = pass1_values[rightIndex + 1];
b.z = pass1_values[rightIndex + 2];
b.w = pass1_values[rightIndex + 3];
If I uncomment the commented part, the behavior is undefined. Is this a legal kernel operation? If not, the compiler should error.
.sean
