[0035] Adding conversion primitives - #819
Merged
Chris B (llvm-beanz) merged 8 commits intoMar 30, 2026
Merged
Conversation
This provides a different approach to addressing some of the issues around types and conversions. Instead of allowing the arithmetic operations to convert data during the operations it implements a conversion operation that converts data before the operation. Fixes microsoft#786, Fixes microsoft#568
Collaborator
Author
|
Tagging Anupama Chandrasekhar (@anupamachandra), Gregory Roth (@pow2clk), Jeff Bolz (@jeffbolznv), Michael Bedy (@mbedy-amd), & Nicolai Hähnle (@nhaehnle) for visibility. |
Nicolai Hähnle (nhaehnle)
left a comment
Nicolai Hähnle (nhaehnle)
left a comment
Contributor
There was a problem hiding this comment.
Having an explicit Convert like this makes sense to me, but I'm struggling to understand how a conversion to, say, F8_E4M3 is actually supposed to work given that I don't believe we can have a plain vector of that type.
Gregory Roth (pow2clk)
left a comment
Gregory Roth (pow2clk)
left a comment
Contributor
There was a problem hiding this comment.
I think it would be less confusing to use uint for packed 8-bit floats than the existing uint8_t4_packed native types as the example code uses.
I'd be interested to know what is expected if these native packed types are used for native vectors used in matvec multiplies.
Co-authored-by: Nicolai Hähnle <nhaehnle@gmail.com>
Contributor
|
Will there be a runtime feature check couterpart to check what conversions an implementation supports? |
Member
I don't see a change to how the PSV0 data structure as part of this, but yes we can make a corresponding change in the caps reporting. |
Mostly changes the inlined header to be consistent with itself, other parts of the spec and other corrections for compilability Finish update to SIZE_TYPE in header. Remove unused IsNativeScalar boolean. These finish the cleanups that microsoft#712 started to fix microsoft#570 Finish update to align mat/vec multiplies with coopvec spec. Finishes Update mat/vec multiply operations to not take a scope parameter, limiting them to thread scope Fixes microsoft#797 Remove unused template type for accumulate() methods fixes microsoft#795 Renumbered componenttype enum to be consistent with internal numbers fixes microsoft#798 Removed latent `p` prefix to overload names of groupshared dxil ops for consistency. Corrects omission in microsoft#769 fixing microsoft#756 Corrected latent mention of linAlgMatrixMulOp
This responds to API feedback provided offline. I've also fixed some formatting and other errors introduced in `main`.
Collaborator
Author
Ashley Coleman (V-FEXrt)
approved these changes
Mar 20, 2026
Ashley Coleman (V-FEXrt)
approved these changes
Mar 27, 2026
Co-authored-by: Ashley Coleman <ascoleman@microsoft.com>
Tex Riddell (tex3d)
approved these changes
Mar 30, 2026
Ashley Coleman (V-FEXrt)
added a commit
to microsoft/DirectXShaderCompiler
that referenced
this pull request
Mar 31, 2026
microsoft/hlsl-specs#819 Adds a signed output bool to the VecMul and VecMulAdd ops. This PR reflects that change to the spec.
Helena Kotas (hekota)
added a commit
to microsoft/DirectXShaderCompiler
that referenced
this pull request
Apr 1, 2026
Adds `Convert` function to `linalg.h`. This function converts a vector to an `InterpretedVector` of a different component type. It has been added to the spec in microsoft/hlsl-specs#819. It is implemented by calling a built-in function that was added in #8308.
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
8 participants
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 provides a different approach to addressing some of the issues around types and conversions. Instead of allowing the arithmetic operations to convert data during the operations it implements a conversion operation that converts data before the operation.
Fixes #786, Fixes #568