early exit in TypeLowering for address-only type #1253
Conversation
| for (auto eltType : type.getElementTypes()) { | ||
| switch (classifyType(eltType, M, Sig, Expansion)) { | ||
| case LoweredTypeKind::Trivial: | ||
| continue; | ||
| case LoweredTypeKind::AddressOnly: | ||
| isAddressOnly = true; | ||
| return asImpl().handleAddressOnly(type); | ||
| continue; |
There was a problem hiding this comment.
The 'continue' is unreachable now
|
Awesome, thank you so much! |
|
|
The logic that detects type circularity in TypeLowering has been replaced in Sema. By exiting as soon as an address-only type is detected, we can avoid walking the rest and gain some speedup.
dduan
force-pushed
the
delete_typelowering_dead_code
branch
from
bf269de to
9203f20
Compare
|
@slavapestov updated. |
| RecursiveErrorTypeLowering(SILType type) | ||
| : AddressOnlyTypeLowering(type) {} | ||
|
|
||
| bool isValid() const override { |
There was a problem hiding this comment.
One more nit-pick if you don't mind: It looks like TypeLowering::isValid() is never called anywhere and always returns true now, so it can be removed?
There was a problem hiding this comment.
It is used in lib/SILOptimizer/Mandatory/PredictableMemOpt.cpp#44, but since it's alway true, we can remove that check now. Right?
There was a problem hiding this comment.
I believe that check was added since we still run mandatory optimizations even if SILGen generated diagnostics.
So yeah, go ahead and remove it!
|
@swift-ci Please test |
|
Alright, looks good. I'll merge once CI is done. |
|
@swift-ci Please test |
|
Thanks! |
early exit in TypeLowering for address-only type
Test SwiftPM included in the distribution on CI
The logic that detects type circularity in TypeLowering has been replaced in
Sema. By exiting as soon as an address-only type is detected, we can avoid
walking the rest and gain some speedup.