[pmo] Teach PMO that in ossa, load [take] invalidates an available va… #25354
Conversation
|
@swift-ci smoke test |
gottesmm
force-pushed
the
pr-3e14153374d6cc2731dac887138c185dcc5de96f
branch
from
d7aa40f to
657dc5b
Compare
|
@swift-ci smoke test |
gottesmm
force-pushed
the
pr-3e14153374d6cc2731dac887138c185dcc5de96f
branch
from
657dc5b to
292b745
Compare
|
@swift-ci smoke test |
| @@ -233,7 +234,9 @@ struct AvailableValue { | |||
| InsertionPoints.set_union(Other.InsertionPoints); | |||
| } | |||
|
|
|||
| void addInsertionPoint(StoreInst *I) & { InsertionPoints.insert(I); } | |||
| void addInsertionPoint(SILInstruction *I) & { | |||
There was a problem hiding this comment.
Did you do this because you have a follow up PR that adds different types of insertion points? By itself this change doesn't make sense. It's better to raise the assumed type up to the interface.
There was a problem hiding this comment.
Sure. I realized I did this for a stupid reason.
|
Andy and I talked a bit about this. There are larger changes he wants to make. I am going to add the comments/etc and then merge. This pass is the first poster child for how load/store promotion is done on ossa so it is the place where I am making tools. It is important that it is easy to read as a result (and this also will get rid of some technical debt as well). |
…lue. Example would be a case where we dynamically control if the load [take] occurs and the load [take] happens with conditional control flow. This was exposed by the throwing_initializer and failable_initializer interpreter tests in the test of PolarBear(n:before:during).
gottesmm
force-pushed
the
pr-3e14153374d6cc2731dac887138c185dcc5de96f
branch
from
292b745 to
ced5f4a
Compare
|
@swift-ci smoke test and merge |
…lue.
Example would be a case where we dynamically control if the load [take] occurs
and the load [take] happens with conditional control flow.
This was exposed by the throwing_initializer and failable_initializer
interpreter tests in the test of PolarBear(n:before:during).