-
Couldn't load subscription status.
- Fork 15k
[SPIRV] Add legalization for long vectors #165444
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Draft
s-perron
wants to merge
13
commits into
llvm:main
Choose a base branch
from
s-perron:legalize-long-vectors
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Draft
+1,027
−193
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 change sets the hasSideEffects flag to false on type and constant opcodes so that they can be considered trivially dead if their result is unused. This means that instruction selection will now be able to remove them.
The spv_bitcast intrinsic is currently replaced by an OpBitcast during prelegalization. This will cause a problem when we need to legalize the OpBitcast. The legalizer assumes that instruction already lowered to a target specific opcode is legal. We cannot lower it to a G_BITCAST because the bitcasts sometimes the LLT type will be the same, causing an error in the verifier, even if the SPIR-V types will be different. This commit keeps the intrinsic around until instructoin selection. We can create rules to legalize a G_INTRINISIC* instruction, and it does not create problem for the verifier.
The previous check for vector bitcasts in `loadVectorFromVector` only compared the number of elements, which is insufficient when the element types differ. This can lead to incorrect assumptions about the validity of the cast. This commit replaces the element count check with a comparison of the total size of the vectors in bits. This ensures that the bitcast is only performed between vectors of the same size, preventing potential miscompilations.
This commit refactors the SPIRV post-legalizer to use a worklist to process new instructions. Previously, the post-legalizer would iterate through all instructions and try to assign types. This could fail if a new instruction depended on another new instruction that had not been processed yet. The new implementation adds all new instructions that require a SPIR-V type to a worklist. It then iteratively processes the worklist until it is empty. This ensures that all dependencies are met before an instruction is processed. This change makes the post-legalizer more robust and fixes potential ordering issues with newly generated instructions. Existing tests cover existing functionality. More tests will be added as the legalizer is modified. Part of llvm#153091
…tcast-intrinsic' and 'legalize_ptr_cast' into legalize-long-vectors-final
This commit refactors the SPIRV post-legalizer to use a worklist to process new instructions. Previously, the post-legalizer would iterate through all instructions and try to assign types. This could fail if a new instruction depended on another new instruction that had not been processed yet. The new implementation adds all new instructions that require a SPIR-V type to a worklist. It then iteratively processes the worklist until it is empty. This ensures that all dependencies are met before an instruction is processed. This change makes the post-legalizer more robust and fixes potential ordering issues with newly generated instructions. Existing tests cover existing functionality. More tests will be added as the legalizer is modified. Part of llvm#153091
s-perron
added a commit
to s-perron/llvm-project
that referenced
this pull request
Oct 28, 2025
This patch introduces the necessary infrastructure to legalize vector operations on vectors that are longer than what the SPIR-V target supports. For instance, shaders only support vectors up to 4 elements. The legalization is done by splitting the long vectors into smaller vectors of a legal size. Specifically, this patch does the following: - Introduces `vectorElementCountIsGreaterThan` and `vectorElementCountIsLessThanOrEqualTo` legality predicates. - Adds legalization rules for `G_SHUFFLE_VECTOR`, `G_EXTRACT_VECTOR_ELT`, `G_BUILD_VECTOR`, `G_CONCAT_VECTORS`, `G_SPLAT_VECTOR`, and `G_UNMERGE_VALUES`. - Handles `G_BITCAST` of long vectors by converting them to `@llvm.spv.bitcast` intrinsics which are then legalized. - Updates `selectUnmergeValues` to handle extraction of both scalars and vectors from a larger vector, using `OpCompositeExtract` and `OpVectorShuffle` respectively. - Adds a test case to verify the legalization of a bitcast between a `<8 x i32>` and `<4 x f64>`, which is a pattern generated by HLSL's `asuint` and `asdouble` intrinsics. Fixes: llvm#165444
f50b420 to
ed13aeb
Compare
s-perron
commented
Oct 28, 2025
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I need to rename the file.
This patch introduces the necessary infrastructure to legalize vector operations on vectors that are longer than what the SPIR-V target supports. For instance, shaders only support vectors up to 4 elements. The legalization is done by splitting the long vectors into smaller vectors of a legal size. Specifically, this patch does the following: - Introduces `vectorElementCountIsGreaterThan` and `vectorElementCountIsLessThanOrEqualTo` legality predicates. - Adds legalization rules for `G_SHUFFLE_VECTOR`, `G_EXTRACT_VECTOR_ELT`, `G_BUILD_VECTOR`, `G_CONCAT_VECTORS`, `G_SPLAT_VECTOR`, and `G_UNMERGE_VALUES`. - Handles `G_BITCAST` of long vectors by converting them to `@llvm.spv.bitcast` intrinsics which are then legalized. - Updates `selectUnmergeValues` to handle extraction of both scalars and vectors from a larger vector, using `OpCompositeExtract` and `OpVectorShuffle` respectively. - Adds a test case to verify the legalization of a bitcast between a `<8 x i32>` and `<4 x f64>`, which is a pattern generated by HLSL's `asuint` and `asdouble` intrinsics. Fixes: llvm#165444
ed13aeb to
0e00c3a
Compare
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.
This patch introduces the necessary infrastructure to legalize vector
operations on vectors that are longer than what the SPIR-V target
supports. For instance, shaders only support vectors up to 4 elements.
The legalization is done by splitting the long vectors into smaller
vectors of a legal size.
Specifically, this patch does the following:
vectorElementCountIsGreaterThanandvectorElementCountIsLessThanOrEqualTolegality predicates.G_SHUFFLE_VECTOR,G_EXTRACT_VECTOR_ELT,G_BUILD_VECTOR,G_CONCAT_VECTORS,G_SPLAT_VECTOR, andG_UNMERGE_VALUES.G_BITCASTof long vectors by converting them to@llvm.spv.bitcastintrinsics which are then legalized.selectUnmergeValuesto handle extraction of both scalarsand vectors from a larger vector, using
OpCompositeExtractandOpVectorShufflerespectively.a
<8 x i32>and<4 x f64>, which is a pattern generated byHLSL's
asuintandasdoubleintrinsics.Fixes #165444