Skip to content

fix(tensilelite): correct gfx1250 gate residual instruction size - #12235

Merged
alex391a merged 1 commit into
developfrom
users/awang105/fix-gfx1250-gate-inline-literal
Sep 18, 2026
Merged

alex391a merged 1 commit into
developfrom
users/awang105/fix-gfx1250-gate-inline-literal

Conversation

@alex391a

@alex391a alex391a commented Sep 17, 2026 •

Copy link
Copy Markdown
Contributor

JIRA ID : AIHPSPAR-277

Motivation

The gate residual path selects 1.0 when the gate pointer is null. It represented this value as the integer literal 0x3f800000, which StinkyTofu counted as an additional 4-byte literal while LLVM encoded it as an inline floating-point constant. This caused the gfx1250 assembly-size check to report a 4-byte mismatch before client execution.

Technical Details

Change the gate residual select operand from hex(0x3f800000) to the floating-point literal 1.0. The generated machine instruction remains unchanged, while StinkyTofu and LLVM now agree on its 4-byte size.

Enable CheckASMCodeSize in spmm_gate.yaml so the gfx1250 FFM test checks the generated instruction size and detects future mismatches in this path.

Test Plan

Test Result

Submission Checklist

@therock-pr-bot

therock-pr-bot Bot commented Sep 17, 2026 •

Copy link
Copy Markdown

✅ All Checks Passed — Ready for Review

Check Status Details
📝 PR Description ✅ Pass —
⛔ Forbidden Files ✅ Pass —
🧪 Unit Test ⚠️ Warning Error: Source/code files changed without an accompanying unit test.
Expected: add at least one test file named like test_<name>.py / test_<name>.cpp (or <name>_test.*).
Current: code file(s) changed: projects/hipblaslt/tensilelite/Tensile/KernelWriterAssembly.py; no test file found
🔎 pre-commit ✅ Pass —
🚫 Draft PR 🔜 To Be Enabled —
🚩 Feature Flag 🔜 To Be Enabled —
📊 Code Coverage 🔜 To Be Enabled —
🤖 therock-pr-bot ✅ Pass —

🎉 All checks passed! This PR is ready for review.

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

🙋 Wish to Override Policy?

@therock-pr-bot

therock-pr-bot Bot commented Sep 17, 2026 •

Copy link
Copy Markdown

🎉 All checks passed! This PR is ready for review.

@alex391a
alex391a force-pushed the users/awang105/fix-gfx1250-gate-inline-literal branch from 098c6f7 to 9d3e434 Compare September 17, 2026 23:46
@alex391a
alex391a force-pushed the users/awang105/fix-gfx1250-gate-inline-literal branch from 9d3e434 to a40b130 Compare September 18, 2026 02:16
@codecov

codecov Bot commented Sep 18, 2026 •

Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 0% with 1 line in your changes missing coverage. Please review.

Files with missing lines Patch % Lines
...blaslt/tensilelite/Tensile/KernelWriterAssembly.py 0.00% 1 Missing ⚠️
Additional details and impacted files
@@             Coverage Diff             @@
##           develop   #12235      +/-   ##
===========================================
- Coverage    70.80%   70.80%   -0.00%     
===========================================
  Files         2845     2845              
  Lines       465630   465630              
  Branches     68575    68575              
===========================================
- Hits        329660   329659       -1     
  Misses      112251   112251              
- Partials     23719    23720       +1     
Flag Coverage Δ *Carryforward flag
TensileLite-CPP 46.40% <ø> (-0.01%) ⬇️
TensileLite-Unit 76.10% <0.00%> (ø)
hipBLAS 90.62% <ø> (ø) Carriedforward from 2c764a8
hipBLASLt 35.24% <ø> (ø) Carriedforward from 2c764a8
hipCUB 82.68% <ø> (ø) Carriedforward from 2c764a8
hipDNN 86.98% <ø> (ø) Carriedforward from 2c764a8
hipFFT 43.25% <ø> (ø) Carriedforward from 2c764a8
hipRAND 76.12% <ø> (ø) Carriedforward from 2c764a8
hipSOLVER 68.96% <ø> (ø) Carriedforward from 2c764a8
hipSPARSE 86.99% <ø> (ø) Carriedforward from 2c764a8
rocBLAS 48.31% <ø> (ø) Carriedforward from 2c764a8
rocFFT 51.59% <ø> (ø) Carriedforward from 2c764a8
rocRAND 57.42% <ø> (ø) Carriedforward from 2c764a8
rocSOLVER 76.83% <ø> (ø) Carriedforward from 2c764a8
rocSPARSE 74.61% <ø> (ø) Carriedforward from 2c764a8
rocThrust 93.15% <ø> (ø) Carriedforward from 2c764a8

*This pull request uses carry forward flags. Click here to find out more.

Files with missing lines Coverage Δ
...blaslt/tensilelite/Tensile/KernelWriterAssembly.py 70.27% <0.00%> (ø)

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
  • 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

@alex391a
alex391a merged commit 25e20fa into develop Sep 18, 2026
181 of 188 checks passed
@alex391a
alex391a deleted the users/awang105/fix-gfx1250-gate-inline-literal branch September 18, 2026 16:59
@assistant-librarian

Copy link
Copy Markdown
Contributor

TheRock Submodule Bump Activity

Newest first

tony-davis pushed a commit that referenced this pull request Oct 23, 2026
JIRA ID : AIHPBLAS-4764

## Motivation
`sparse/gfx1250/spmm_gate.yaml` doesn't run on the gfx1250 hardware
lane. #12235 fixed its failure, and #12355 removed its xfail after it
passed. Then #11777 marked the config `skip-gfx1250-strict` ("not
supported yet").

## Technical Details
- Add `sparse/gfx1250-strict/spmm_gate.yaml`, a copy of the gfx1250
config following the pattern of the other 37 sparse strict copies:
  - `skip-gfx1250` instead of `skip-gfx1250-strict`.
- The gfx1250 config is unchanged and keeps running where it runs today.

## Test Plan
- This PR's `gfx1250 HW hipBLASLt + TensileLite Tests` run collects and
passes `sparse/gfx1250-strict/spmm_gate.yaml`.

## Test Result
- New strict test
[passes](https://github.com/ROCm/rocm-libraries/actions/runs/39722965803/job/119640530321?pr=14175)
on gfx1250 HW run.

## Submission Checklist
- [x] Look over the contributing guidelines at
https://github.com/ROCm/TheRock/blob/main/GOVERNANCE.md#pull-requests.

## Risk level
Low. One new test config that runs only on gfx1250-strict.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants