All OpenGL programs that doesn’t try to calculate everything on it’s own will use the any T&L capabilities of the graphic adapter. However, i’m am not sure if the actual calculation of the matrixes are done in hardware of these accelerators.
Whether or not a given OpenGL transform call uses hardware T&L is up to the driver and the hardware. It’s transparent to the programmer.
AFAIK hardware T&L isn’t so much about calculating the transforms (e.g. glRotate, glMultMatrix) as about transforming the vertices you issue by the current modelview matrix. Which isn’t something you do explicitly anyway - it happens whenever you call glVertex etc.