Support for a scalable simd representation #118917
Draft
+795
−100
Commits on Jul 24, 2024
-
The representation of the element type has been changed to be a slice rather than a zero length array. Two feature gates are now required in core_arch unsized fn params and unsized locals. This still leaves unsized return types being an issue. For this we are currently bypassing some of the `Sized` trait checking to pass when the type is scalable simd. This still leaves the copy issue. For that we have marked scalable simd types as trivally pure clone copy. We have still had to remove some trait checks for the copy trait with this though as they are still performed in certain situations. The implementation of `transmute` is also an issue for us. For this a new SIMD intrinsic has been created simd_reinterpret which performs a transmute on SIMD vectors. A few intrinsics need to be able to produce an LLVM `undef` this intrinsic will also produce that when given a zero sized input.
-
-
-
Restrict the spilling of scalable types to the problematic ones.
Rather than not spilling any scalable SIMD types for debug info, we only avoid spilling the ones that are going to cause a problem. Currently the only ones known to cause a problem are the internal svbool types for AArch64.
-
Scalable SIMD tests for disallowing scalable types in types.
Tests to ensure that scalable SIMD types can't exist in struct, union, enum variants and compound types. This also changes the well formed checking of types to improve the error message when scalable SIMD types are included. The previous implementation would also allow a scalable SIMD type as the last element within a struct in some cases which we currently don't want to allow.
-
Allow SVE types to be taken by reference.
Ensures that an SVE type can have a reference taken to it. This currently emits a `dereferenceable` attribute for the ptr using the element size as the number of bytes. While not perfect this is correct as a vector will always have a least one primitive.
-
-
Don't require the user to enable the unsized features.
Rather than forcing the user to enable the unsized_fn_params and unsized_locals features, we condition those features tests with if the type is a scalable simd type.
-
Add some comments explaining some not so obvious things. Mark code as unreachable on architectures that don't currently have scalable vectors. Remove some unused (and incorrect) checks that were being performed. Refactor some code to improve it's readability.
-
Disallow scalable vectors in async fn.
Scalable vectors are unsized types so we can't hold them across an await point. Also clean up the tests to remove unsized features from tests as a user shouldn't have to enable them when using scalable vectors.
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.