This should fix a hypothetical issue such as:
address = load_global()
value = load_global(address, access=uses-smem)
where divergence analysis can't prove that 'address' is uniform, but can
prove that 'value' is uniform.
We might then add both load_global to the load_worklist, but only disable
helpers for the first because the second is uniform, making 'address'
divergent for real and potentially incorrect when used with
v_readfirstlane_b32.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36850>
