proof of concept #90058
proof of concept #90058
Conversation
ghost
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsFYI @EgorBo @jakobbotsch This gets the Not sure this is the best approach, but it does some interesting things. It leverage the local threading so hopefully isn't' too costly.
|
|
Sample diff 
|
To handle these cases we need to do something before or during local morph, otherwise we will still end up with address exposure. |
|
This is like an anti-head-merge... if we were to take this along with head merge we'd need to figure out how not to undo what we just did. |
|
|
||
| bool Compiler::fgTailDuplicateBlock(BasicBlock* const block) | ||
| { | ||
| // We only consider blocks with muliple preds without |
There was a problem hiding this comment.
| // We only consider blocks with muliple preds without | |
| // We only consider blocks with multiple preds without |
Isn't this more like an anti-tail-merge? It duplicates the same statement at the end of multiple predecessors, so those statements are candidates for tail merging. |
Yes, you're right. Actually an anti-head-merge (head duplication? I guess) might be interesting to consider, if we think the successor blocks have more interesting context than the current block. |
ghost
closed this
|
Draft Pull Request was automatically closed for 30 days of inactivity. Please let us know if you'd like to reopen it. |
ghost
locked as resolved and limited conversation to collaborators
3 participants
FYI @EgorBo @jakobbotsch
This gets the
?:case Egor was discussing and maybe the local address one Jakob was looking at.Not sure this is the best approach, but it does some interesting things. It leverage the local threading so hopefully isn't' too costly.