See the TODO attached to this bug (after a pending PR is submitted). Note that part of the downside of this is that for a large IR, we have to run the sparse constant propagation analysis twice to get the dependent analysis to even run, and this shows up as a slower compile time.
See the TODO attached to this bug (after a pending PR is submitted). Note that part of the downside of this is that for a large IR, we have to run the sparse constant propagation analysis twice to get the dependent analysis to even run, and this shows up as a slower compile time.