JIT: build pred lists early, persist them throughout compilation #80193
Assignees
Labels
area-CodeGen-coreclr
CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
Milestone
Comments
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch, @kunalspathak Issue DetailsCurrently there's a lag between incorporation of profile data (which happens quite early and uses its own flow edge representation) and building pred lists, and pred lists are sometimes dropped and recomputed from scratch (#49030). We should be building pred lists early on and using those for profile incorporation, and then keep the lists valid and persistent throughout compilation.
|
|
Note this will likely happen incrementally -- moving the point where pred lists are built earlier and earlier and gradually fixing up the impacted areas of the jit. Related work:
|
AndyAyersMS
added a commit
that referenced
this issue
Jan 6, 2023
Add finer-grained notions of what post-phase checks are enabled. Start checking some IR and FG invariants earlier than we did before. This is largely done in anticipation of moving pred list building earlier (#80193). We should also consider moving some of the IR checking earlier as well.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Jan 19, 2023
Fix the EH opts to properly maintain pred lists, and move building of the pred lists to just before the EH opts. Contributes to dotnet#80193.
AndyAyersMS
added a commit
that referenced
this issue
Jan 20, 2023
Fix the EH opts to properly maintain pred lists, and move building of the pred lists to just before the EH opts. Contributes to #80193.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Jan 20, 2023
Move the pred list building up a bit further. Contributes to dotnet#80193.
AndyAyersMS
added a commit
that referenced
this issue
Jan 20, 2023
Move the pred list building up a bit further. Contributes to #80193.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Jan 20, 2023
In anticipation of moving pred list building before inlining, use the common block split utility instead of custom code. This does a better job of propagating IL offsets, and automatically handles pred list maintenance. Contributes to dotnet#80193.
AndyAyersMS
added a commit
that referenced
this issue
Jan 20, 2023
In anticipation of moving pred list building before inlining, use the common block split utility instead of custom code. This does a better job of propagating IL offsets, and automatically handles pred list maintenance. Contributes to #80193.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Jan 22, 2023
Move pred list building up a bit further. Note that this impacts both the root method and all inlinees, since we run a number of the initial phases on both. Since inlinees build their basic blocks and flow edges in the root compiler's memory pool, the only work required to unify the pred lists for a successful inline is to get things right at the boundaries. And for failed inlines there is no cross-referencing so we can just let the new pred lists leak away (like we alredy do for the inlinee blocks). Contributes towards dotnet#80193.
mdh1418
pushed a commit
to mdh1418/runtime
that referenced
this issue
Jan 24, 2023
Fix the EH opts to properly maintain pred lists, and move building of the pred lists to just before the EH opts. Contributes to dotnet#80193.
mdh1418
pushed a commit
to mdh1418/runtime
that referenced
this issue
Jan 24, 2023
Move the pred list building up a bit further. Contributes to dotnet#80193.
mdh1418
pushed a commit
to mdh1418/runtime
that referenced
this issue
Jan 24, 2023
In anticipation of moving pred list building before inlining, use the common block split utility instead of custom code. This does a better job of propagating IL offsets, and automatically handles pred list maintenance. Contributes to dotnet#80193.
AndyAyersMS
added a commit
that referenced
this issue
Jan 24, 2023
Move pred list building up a bit further. Note that this impacts both the root method and all inlinees, since we run a number of the initial phases on both. Since inlinees build their basic blocks and flow edges in the root compiler's memory pool, the only work required to unify the pred lists for a successful inline is to get things right at the boundaries. And for failed inlines there is no cross-referencing so we can just let the new pred lists leak away (like we already do for the inlinee blocks). Contributes towards #80193.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Jan 25, 2023
Move pred list building 3 phases earlier. It now happens just after instrumentation (or importation if we're not instrumenting) and just before expanding patchpoints. Revise the patchpoint, indirect call, and post importer cleanup phases to do proper pred list maintenance. Update the flow checker to handle cases we see when we haven't yet run the post importer cleanup. Contributes to dotnet#80193.
AndyAyersMS
added a commit
that referenced
this issue
Jan 26, 2023
Move pred list building 3 phases earlier. It now happens just after instrumentation (or importation if we're not instrumenting) and just before expanding patchpoints. Revise the patchpoint, indirect call, and post importer cleanup phases to do proper pred list maintenance. Update the flow checker to handle cases we see when we haven't yet run the post importer cleanup. Contributes to #80193.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Jan 26, 2023
This is used some early phases; make it pred list aware. Contributes to dotnet#80193.
AndyAyersMS
added a commit
that referenced
this issue
Jan 27, 2023
This is used some early phases; make it pred list aware. Contributes to #80193.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Jan 27, 2023
Move pred list building to just before instrumentation (and just after importation -- we are getting very close to the front of the phase list now). The block and edge count instrumenters were both using cheap preds to keep track of some relocated count probes. Revise this so they can use the regular pred lists. Also rework both approaches so their `RelocateProbes` methods are fairly similar and perhaps could be unified one day. Contributes to dotnet#80193.
AndyAyersMS
added a commit
that referenced
this issue
Jan 30, 2023
Move pred list building to just before instrumentation (and just after importation -- we are getting very close to the front of the phase list now). The block and edge count instrumenters were both using cheap preds to keep track of some relocated count probes. Revise this so they can use the regular pred lists. Also rework both approaches so their `RelocateProbes` methods are fairly similar and perhaps could be unified one day. Contributes to #80193.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Feb 1, 2023
Move pred list building before importation. It now runs as the first phase in the phase list. * Split up some unions in block.h as some things can't share storage anymore (may revisit this later). * Modify importer not to rely on cheap preds. Most of the work here is in `impImportLeave`. * Adjust OSR protection strategy for the method entry. In particular, watch for the degenerate case where the OSR entry is the method entry. * Ensure we don't double-decrement some ref counts when blocks with degenerate or folded flow are re-imported. * Update spill clique discovery to use the actual pred lists. * Add new method `impFixPredLists` for the importer to use at the end of importation. This adds pred list entries finally returns, which can't be done until all `BBJ_LEAVE` blocks are processed. Contributes to dotnet#80193.
AndyAyersMS
added a commit
that referenced
this issue
Feb 2, 2023
Move pred list building before importation. It now runs as the first phase in the phase list. * Split up some unions in block.h as some things can't share storage anymore (may revisit this later). * Modify importer not to rely on cheap preds. Most of the work here is in `impImportLeave`. * Adjust OSR protection strategy for the method entry. In particular, watch for the degenerate case where the OSR entry is the method entry. * Ensure we don't double-decrement some ref counts when blocks with degenerate or folded flow are re-imported. * Update spill clique discovery to use the actual pred lists. * Add new method `impFixPredLists` for the importer to use at the end of importation. This adds pred list entries finally returns, which can't be done until all `BBJ_LEAVE` blocks are processed. * trigger badcode inside `fgComputePreds` * fix issue with `STRESS_CATCH_ARG` Contributes to #80193.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Feb 2, 2023
This is the last in a (long) series. We now build the pred lists at the same time we are initially connecting up the flow graph. Pred lists are now always valid and need to be maintained by all phases. There are some changes needed in EH normalization, and one special case we need to handle in debug codegen where we create the scratch BB very early on. This was the last client for the cheap pred lists. Note some of the pred list info can't be added right away, in particular the "return" edges from finallies do not appear until we've made it through the importer. I have deferred cleaning up dead code; will do it in follow-up changes. Contributes to dotnet#80193.
AndyAyersMS
added a commit
that referenced
this issue
Feb 2, 2023
This is the last in a (long) series. We now build the pred lists at the same time we are initially connecting up the flow graph. Pred lists are now always valid and need to be maintained by all phases. There are some changes needed in EH normalization, and one special case we need to handle in debug codegen where we create the scratch BB very early on. This was the last client for the cheap pred lists. Note some of the pred list info can't be added right away, in particular the "return" edges from finallies do not appear until we've made it through the importer. I have deferred cleaning up dead code; will do it in follow-up changes. Contributes to #80193.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Feb 2, 2023
We no longer need cheap preds. We always have full preds available. Also, remove the ability to drop and recompute full preds, as we now compute full preds incrementally while building the flow graph, and never drop or need to rebuild them. Contributes to dotnet#80193.
AndyAyersMS
added a commit
that referenced
this issue
Feb 3, 2023
We no longer need cheap preds. We always have full preds available. Also, remove the ability to drop and recompute full preds, as we now compute full preds incrementally while building the flow graph, and never drop or need to rebuild them. Contributes to #80193.
AndyAyersMS
added a commit
to AndyAyersMS/runtime
that referenced
this issue
Feb 3, 2023
Since pred lists now exist pervasively, change most methods that had conditional pred list updates to assert preds exist and always update. To make sure I got all uses, I renamed `fgComputePredsDone` to `fgPredsComputed`. Remove the ability to drop EH, as it doesn't update pred lists properly and so has been broken for quite a while now. Remove some of the logic for fixing up finally targets, since we now always have pred lists around and so don't need the blanket invaliation. Closes dotnet#80193.
ghost
added
the
in-pr
AndyAyersMS
added a commit
that referenced
this issue
Feb 3, 2023
Since pred lists now exist pervasively, change most methods that had conditional pred list updates to assert preds exist and always update. To make sure I got all uses, I renamed `fgComputePredsDone` to `fgPredsComputed`. Remove the ability to drop EH, as it doesn't update pred lists properly and so has been broken for quite a while now. Remove some of the logic for fixing up finally targets, since we now always have pred lists around and so don't need the blanket invaliation. Closes #80193.
ghost
removed
the
in-pr
ghost
locked as resolved and limited conversation to collaborators
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Currently there's a lag between incorporation of profile data (which happens quite early and uses its own flow edge representation) and building pred lists, and pred lists are sometimes dropped and recomputed from scratch (#49030).
We should be building pred lists early on and using those for profile incorporation, and then keep the lists valid and persistent throughout compilation.
The text was updated successfully, but these errors were encountered: