cudaArray synchronized?


I’m editing texture which is in cudaArray.
Access patterns are random and I would like to know if texture in cudaArray is somehow synchronized?
I mean what happens if two blocks/threads try to access the same pixel (i.e. change color) at the same time using code from examples:

float* pixelPtr = reinterpret_cast<float*>((m_surfacePtr + y * m_pitch)) + sizeof(float) * x;
pixelPtr[0] = R;
pixelPtr[1] = G;
pixelPtr[2] = B;
pixelPtr[3] = A;