TypeLayout: Use AlignedGroups instead of ScalarLayouts #39053
Merged
Conversation
This file contains hidden or 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
This comment has been minimized.
This comment has been minimized.
gmittert
changed the title
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
gmittert
force-pushed
the
TypeLayoutEnhancementsUpstream
branch
from
5f63ca4 to
9cffbfb
Compare
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
gmittert
force-pushed
the
TypeLayoutEnhancementsUpstream
branch
from
9cffbfb to
290c6dd
Compare
|
@swift-ci please test |
|
Build failed |
|
Build failed |
gmittert
force-pushed
the
TypeLayoutEnhancementsUpstream
branch
from
290c6dd to
1e9127a
Compare
|
@swift-ci please smoke test |
gmittert
force-pushed
the
TypeLayoutEnhancementsUpstream
branch
from
1e9127a to
332b4f7
Compare
|
Rebasing and cleaning up a bit. |
|
@swift-ci please smoke test |
added 4 commits
September 21, 2021 22:21
Added ForceStructTypeLayouts. When enabled, IRGen will lower structs using the aligned group of TypeLayout rather than using TypeInfos. This potentially leads to a size increase as TypeInfos currently produce better code than the TypeLayout route.
If the Aligned Group is fixed sized and alignment, rather than doing an alignment computation at runtime, we compute the offset for each field and emit a GEP.
When generating TypeLayouts, rather than using a Scalar Layout, use an aligned struct so that we use the typelayout path rather than the typeinfo path when doing irgen.
Teach TypeLayout about when it can omit checking the enum tag and instead directly either ignore retaining/releasing the payload or retain/release it regardless of the enum tag. This ports to PO/NullableRefcounted/ForwardToPayload distinctions to TypeLayout.
gmittert
force-pushed
the
TypeLayoutEnhancementsUpstream
branch
from
332b4f7 to
b3c9316
Compare
|
@swift-ci please smoke test |
|
@swift-ci please test windows platform |
|
Windows build seems to be broken. Same test is failing in other PRs. |
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.
The goal of these changes is to reduce the uses of getOrCreateScalarEntry which is a passthrough to the typeinfo system rather than using the typelayout system. Removing these uses should increase the visibility that the typelayout system has in the types that it is lowering and allow it to generate better code.
However, simply changing the included getOrCreateScalarEntries to getOrCreateAligned groups causes a significant size increase. The two main contributors were enums not implementing pod/nullablerefcount/forwardtopayload optimizations and structs doing alignment computations for each field.
This PR teach typelayout how to do the above and then swaps to using aligned group entries instead of scalar group entries. In my testing it brought the size increase down to within 1%, but it might be good to test it against more samples.