Extend stackmap format to allow for derived types. #40
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ltratt
reviewed
Aug 17, 2022
| while (ExpectArgs) { | ||
| SDNode *ItN = It->getNode(); | ||
| if ((ItN->getOpcode() == ISD::TargetConstant) && | ||
| (cast<ConstantSDNode>(ItN)->getZExtValue() == StackMaps::NextLive)) |
There was a problem hiding this comment.
I misread this one at first: it really feels like {...} brackets would help here?
vext01
commented
Aug 17, 2022
|
OK to squash? |
ltratt
reviewed
Aug 18, 2022
| std::stringstream Msg; | ||
| Msg << "stackmap ID " << CSI.ID | ||
| << ": too many locations in live variable #" << LiveIdx; | ||
| OS.getContext().reportError(SMLoc(), Msg.str()); |
There was a problem hiding this comment.
Should we return at this point or ... ?
There was a problem hiding this comment.
Yes, we should. It prevents code from crashing later and spamming the console with a backtrace! Good catch.
Fixed in 0896c8ff40bf
|
Please squash. |
Prior to this change, LLVM would write bogus stackmap entries for larger derived types: llvm#55957 To record the correct information in the stackmap section, we need to make two changes: 1) The ability for the stackmap format to express that any value could have been split among multiple register and/or memory locations. The stackmap format cannot, at-present, express this. Whereas before the entity relationship diagram was: (stackmap-callsite) -has-many-> (location) Now it is: (stackmap-callsite) -has-many-> (live-variable) -has-many-> (location) 2) The ability for the relevant DAG nodes to encode live variable groupings. This applies to stackmap, patchpoint and statepoint DAG nodes (which are all intertwined, otherwise I'd have left statepoint alone). Before, the relevant node operands were a flat list, and could not express that values could be made up of many other values. This change introduces a `NextLive` marker to encode how operands are grouped into live variables. For example, live variable operands in a DAG node like this: [1, 2, NextLive, 3, NextLive, 4, 5, 6, NextLive] expresses three live variables, made of: 1) [1, 2] 2) [3] 3) [4, 5, 6] Unfortunately changing the DAG nodes caused a lot of test failures. I've fixed most of them, but since we don't plan to upstream this right now, I've skipped some of the tests that are very annoying to update. Currently structs and vectors (aggregate types) are now correctly recorded in the stackmap section Vectors don't yet work and will require more work (they use a different DAG node structure which will require walking the graph to infer how to encode the stackmap operands). That work will build on this change.
|
splat. |
|
bors r+ |
|
Build succeeded: |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Prior to this change, LLVM would write bogus stackmap entries for
larger derived types:
llvm#55957
To record the correct information in the stackmap section, we need to
make two changes:
The ability for the stackmap format to express that any value could
have been split among multiple register and/or memory locations. The
stackmap format cannot, at-present, express this.
Whereas before the entity relationship diagram was:
(stackmap-callsite) -has-many-> (location)
Now it is:
(stackmap-callsite) -has-many-> (live-variable) -has-many-> (location)
The ability for the relevant DAG nodes to encode live variable
groupings.
This applies to stackmap, patchpoint and statepoint DAG nodes (which
are all intertwined, otherwise I'd have left statepoint alone).
Before, the relevant node operands were a flat list, and could not
express that values could be made up of many other values. This change
introduces a
NextLivemarker to encode how operands are grouped intolive variables.
For example, live variable operands in a DAG node like this:
[1, 2, NextLive, 3, NextLive, 4, 5, 6, NextLive]
expresses three live variables, made of:
Unfortunately changing the DAG nodes caused a lot of test failures. I've
fixed most of them, but since we don't plan to upstream this right now,
I've skipped some of the tests that are very annoying to update.
Currently structs and vectors (aggregate types) are now correctly
recorded in the stackmap section
Vectors don't yet work and will require more work (they use a different
DAG node structure which will require walking the graph to infer how to
encode the stackmap operands). That work will build on this change.
A companion PR
will follow: ykjit/yk#561