fix(hiptensor): reject batched contraction with a shared A/B/D mode as NOT_SUPPORTED - #10826
Merged
harkgill-amd merged 6 commits intoAug 20, 2026
Merged
Conversation
✅ All Checks Passed — Ready for Review
📖 Need help? See the Policy FAQ for details on every check and how to fix failures. |
|
🎉 All checks passed! This PR is ready for review. |
Ryker0627
reviewed
Aug 14, 2026
Ryker0627
reviewed
Aug 14, 2026
…rd trinary create, log mode as char
evedovelli
reviewed
Aug 19, 2026
evedovelli
left a comment
evedovelli
left a comment
Contributor
There was a problem hiding this comment.
@harkgill-amd Would it be possible adding a couple of unit tests to verify that the batched contraction properly returns NOT_SUPPORTED for binary and trinary contractions?
Contributor
Author
Added the following three test cases,
All are self-explanatory and pass various inputs to trigger either the Tested all of these on a gfx942/MI300A system w/ the latest TheRock nightlies + hipTensor build from source. All tests pass as expected. |
evedovelli
reviewed
Aug 19, 2026
evedovelli
left a comment
evedovelli
left a comment
Contributor
There was a problem hiding this comment.
Thank you so much for adding the tests! I've left one last ask, if you have the time to work on it. The changes look good for me but I'll leave to @Ryker0627 for approving.
Ryker0627
reviewed
Aug 19, 2026
Ryker0627
approved these changes
Aug 20, 2026
Ryker0627
left a comment
Ryker0627
left a comment
Contributor
There was a problem hiding this comment.
It looks good to me. Thank you for the time and effort you put into this!
harkgill-amd
deleted the
users/harkgill/hiptensor-reject-batched-contraction
branch
August 20, 2026 17:23
assistant-librarian Bot
pushed a commit
to ROCm/hipTensor
that referenced
this pull request
Aug 20, 2026
fix(hiptensor): reject batched contraction with a shared A/B/D mode as NOT_SUPPORTED (#10826) ## Motivation Resolves ROCm/legacy-rocm-build#6559. ## Technical Details On hipTensor 2.3.0, a conventional batched contraction ``` D[b,m,n] = sum_k A[b,m,k] * B[b,k,n] ``` accepts valid descriptors and returns `HIPTENSOR_STATUS_SUCCESS` from every hipTensor call, but produces numerically incorrect output. Check for batched contraction (see below) and return HIPTENSOR_STATUS_NOT_SUPPORTED until further support is added. ``` std::set<int32_t> setB(modeBV.cbegin(), modeBV.cend()); std::set<int32_t> setD(modeDV.cbegin(), modeDV.cend()); for(int32_t m : modeAV) if(setB.count(m) && setD.count(m)) ``` ## Test Plan Build patched libhiptensor from source (develop @ 2.4.0, against nightly CK) on an MI300A (gfx942) node, then run two cases on real hardware: 1. Batched contraction -> expect NOT_SUPPORTED after patch. 2. Non-batched GEMM -> expect unchanged SUCCESS. ## Test Result - With patch applied, batched contraction correctly returns as not supported meanwhile, non batched GEMMs see no change as expected. ## Submission Checklist - [X] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
bghimireamd
pushed a commit
that referenced
this pull request
Aug 20, 2026
…s NOT_SUPPORTED (#10826) ## Motivation Resolves ROCm/legacy-rocm-build#6559. ## Technical Details On hipTensor 2.3.0, a conventional batched contraction ``` D[b,m,n] = sum_k A[b,m,k] * B[b,k,n] ``` accepts valid descriptors and returns `HIPTENSOR_STATUS_SUCCESS` from every hipTensor call, but produces numerically incorrect output. Check for batched contraction (see below) and return HIPTENSOR_STATUS_NOT_SUPPORTED until further support is added. ``` std::set<int32_t> setB(modeBV.cbegin(), modeBV.cend()); std::set<int32_t> setD(modeDV.cbegin(), modeDV.cend()); for(int32_t m : modeAV) if(setB.count(m) && setD.count(m)) ``` ## Test Plan Build patched libhiptensor from source (develop @ 2.4.0, against nightly CK) on an MI300A (gfx942) node, then run two cases on real hardware: 1. Batched contraction -> expect NOT_SUPPORTED after patch. 2. Non-batched GEMM -> expect unchanged SUCCESS. ## Test Result - With patch applied, batched contraction correctly returns as not supported meanwhile, non batched GEMMs see no change as expected. ## Submission Checklist - [X] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
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.
Motivation
Resolves ROCm/legacy-rocm-build#6559.
Technical Details
On hipTensor 2.3.0, a conventional batched contraction
accepts valid descriptors and returns
HIPTENSOR_STATUS_SUCCESSfrom every hipTensor call, but produces numerically incorrect output.Check for batched contraction (see below) and return HIPTENSOR_STATUS_NOT_SUPPORTED until further support is added.
Test Plan
Build patched libhiptensor from source (develop @ 2.4.0, against nightly CK) on an MI300A (gfx942) node, then run two cases on real hardware:
Test Result
Submission Checklist