[Release SM 6.9] Cherry-Pick Fix rawBufferVectorLoad/Store to widen min precision types to 32-bit#8369
Merged
alsepkow merged 2 commits intomicrosoft:release-1.9.2602from Apr 15, 2026
Conversation
…icrosoft#8274) ## Summary Fixes `RawBufferVectorLoad`/`Store` to use 32-bit element types (`i32`/`f32`) for min precision types (`min16int`, `min16uint`, `min16float`) instead of 16-bit (`i16`/`f16`). This matches how pre-SM6.9 `RawBufferLoad` handles min precision. Resolves microsoft#8273 ## Root Cause `TranslateBufLoad` in `HLOperationLower.cpp` creates the vector type directly from the min precision element type (`i16`/`f16`) without widening to `i32`/`f32`. This causes WARP (and potentially other drivers) to load/store 2 bytes per element instead of 4, mismatching the buffer layout. ## Fix Apply the same widening pattern used for bool types: - **Load**: Load as `v_i32`/`v_f32`, then trunc/fptrunc back to `i16`/`half` - **Store**: `sext`/`fpext` to `i32`/`f32`, then store as `v_i32`/`v_f32` ## Testing Added FileCheck test verifying all 3 min precision types produce `i32`/`f32` vector load/store ops. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Tex Riddell <texr@microsoft.com>
…icrosoft#8321) ## Summary Reverts the `DxilGenerationPass` ByteAddressBuffer scalar store changes and removes scalar store tests that were included in PR microsoft#8274. These changes were out of scope for the vector load/store fix and, on further discussion, were concluded to be incomplete. Follow-up issue for the proper fix: microsoft#8322 ## What changed - **DxilGenerationPass.cpp**: Reverted the ByteAddressBuffer fallback path added in `ReplaceMinPrecisionRawBufferStoreByType`. The `sext` fallback was wrong for `min16uint` (loses signedness), and the fix belongs in CodeGen where signedness info is still available. - **min_precision_raw_load_store.hlsl**: Removed scalar load/store tests. Scalar `ByteAddressBuffer::Store<min16int>()` hits a pre-existing crash in `TranslateMinPrecisionRawBuffer` (`cast<StructType>` on ByteAddressBuffer's `i32` inner element). Test now covers vector loads/stores only, which is the scope of the original fix. ## Context The original PR microsoft#8274 correctly fixed `RawBufferVectorLoad/Store` to widen min precision types to 32-bit. However, it also added a ByteAddressBuffer scalar store fix in `DxilGenerationPass` that: 1. Crept outside the scope of the vector load/store fix 2. Was incomplete — the `sext` fallback is wrong for unsigned types (`min16uint`) 3. Should instead be handled during Clang CodeGen, where signedness information is available Scalar ByteAddressBuffer template store widening for min precision types is a separate pre-existing issue that needs a proper fix in CodeGen. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This was referenced Apr 14, 2026
|
Azure Pipelines: Successfully started running 1 pipeline(s). |
tex3d
approved these changes
Apr 15, 2026
Icohedron
approved these changes
Apr 15, 2026
Collaborator
Icohedron
left a comment
Icohedron
left a comment
There was a problem hiding this comment.
Confirmed to be cherry-picks; the diff hashes match
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
3 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.
Cherry-pick PR (#8274) and revert of out-of-scope changes PR (#8321)
Assisted by gh copilot.
SHA dc4354b
SHA 71aa195