RFR: 8333393: PhaseCFG::insert_anti_dependences can fail to raise LCAs and to add necessary anti-dependence edges [v11]

[Daniel Lundén]

On Thu, 30 Jan 2025 13:57:18 GMT, Quan Anh Mai <qamai at openjdk.org> wrote:

> This is a really bold assumption and I assume you do not have any concrete reason for it.

The assumption holds in all (fuzzer-generated) failure cases we have looked at. It also makes sense from a scheduling perspective to have `initial_mem` as early as possible to allow putting `early` as early as possible allowing maximum scheduling freedom. In fact, the reason why we get these late `early` blocks in relation to `initial_mem_block` is that the loads that fail get pinned from above by explicit control inputs.

I could do some further experiments to investigate if you wish.

> "After" in the sense of CFG can only emerge after scheduling, while:
> 
> * We are scheduling.
> * The load and its memory input have existed before scheduling, which makes it very hard to imagine that this is a valid assumption.
> 
> And "after" in the sense of graph edges seems not to be the case, either. As your example has pointed out.

I mean "after" in the sense of the dominator graph, which is available at this point. Also, all memory-defining nodes have been globally scheduled (i.e., they have a block). These are pre-existing assumptions in `insert_anti_dependences`.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/22852#issuecomment-2624683099