Could it be the lack of hardware vertex processing causing the shader version to break.
I tested the latest demo in my wifes work laptop(has been very problematic). Radeon Mobility X300. It's very slow with shadows in 32 bit, 2x as fast in 16 bit. And the shader works and doesn't have much impact on performance. the X300 does have a single vertex unit onboard though.
The shaders should look like this.
I tested the latest demo in my wifes work laptop(has been very problematic). Radeon Mobility X300. It's very slow with shadows in 32 bit, 2x as fast in 16 bit. And the shader works and doesn't have much impact on performance. the X300 does have a single vertex unit onboard though.
The shaders should look like this.