In D54516#1301072, @tpr wrote:

EarlyCSE does seem to common up in this situation. And, if I disable that, I get GVN commoning it up.

Sorry about that! I am reverting the change for now. Waiting for the tests
to pass locally with the change reverted.

figured out how to fix the revision using "arc diff --edit"

I changed the first line in the change description to make it explicit that this is a workaround for DA, but it is not showing up here. I just used "arc diff master" ... how do I get the web interface to show the updated description?

addressed comments from @arsenm

In D52221#1238579, @sameerds wrote:

In D52221#1238116, @nhaehnle wrote:

Good to hear the new DA can handle this. We need lowered switches for the control flow lowering anyway, though, so we may as well do it this way.

The test case is extremely mysterious to me, though. What is it that's actually being tested here? From the description I understand it's the divergence of the phi in sw.epilog? I don't see how that relates to the CHECK lines.

Updated the test to better convey the intention. Does that help? Note that since this change is a backend-specific workaround, the test focuses on the symptom seen in the results of si-annotate-control-flow rather than checking the results of divergence analysis.

In D52221#1238116, @nhaehnle wrote:

Good to hear the new DA can handle this. We need lowered switches for the control flow lowering anyway, though, so we may as well do it this way.

The test case is extremely mysterious to me, though. What is it that's actually being tested here? From the description I understand it's the divergence of the phi in sw.epilog? I don't see how that relates to the CHECK lines.

In D52221#1238073, @simoll wrote:

This workaround will not longer be required with the new Divergence Analysis (https://reviews.llvm.org/D50433). I will add this test case as a unit test for the new implementation.

improve the lit test