WIP: feat(tensilelite): Use StinkyTofu with Subtile.#9303
Conversation
…riter Move rocIsaPass and StinkyTofu conversion logic from kernelBody() into reusable helpers with no behavior change for classic or subtile kernels. Co-authored-by: Cursor <cursoragent@cursor.com>
…ubtile Set _StinkySubtile on gfx1250 subtile solutions and run rocIsaPass plus a minimal StinkyTofu pipeline in kernelBodySubtile. Add unit tests. Co-authored-by: Cursor <cursoragent@cursor.com>
Emit PRELOOP/MAINLOOP/NGLL as loopBody and NLL as noLoadLoopBody so gfx1250 waitcnt region detection works without StinkyTofu converter changes. Co-authored-by: Cursor <cursoragent@cursor.com>
…itcnt Attach LDS buffer tokens to GR/LR loads and barriers so gfx1250 subtile kernels can pass MemTokenConsistencyCheck in the waitcnt-only pipeline. Co-authored-by: Cursor <cursoragent@cursor.com>
Expose loopBody as a top-level kernel-body module so ST can detect loopWithPrefetch, enable loop-carried token deps, and skip hand-rolled vmcnt adjustment when the waitcnt-only pipeline owns synchronization. Co-authored-by: Cursor <cursoragent@cursor.com>
Expose the tail loop as top-level noLoadLoopBody modules so Gfx1250 waitcnt insertion covers TAILLOOP and replaces hand-rolled tail waits. Co-authored-by: Cursor <cursoragent@cursor.com>
…n passes Centralize waitcnt-only StinkyTofu overrides and add asm checks that ST does not double-insert cluster barriers or wait-alu while OptLevel 0 skips DAG rescheduling. Co-authored-by: Cursor <cursoragent@cursor.com>
…is on Let StinkyTofu own GR/LR waitcnt insertion for gfx1250 subtile kernels by no-oping WaitGROp/WaitLROp emission while keeping scheduler deps. Co-authored-by: Cursor <cursoragent@cursor.com>
❌ PR Check — Action Required
📖 Need help? See the Policy FAQ for details on every check and how to fix failures. |
|
🚫 Please fix the failed policies before requesting reviews. The following policy checks failed:
The |
Codecov Report✅ All modified and coverable lines are covered by tests. ❌ Your project status has failed because the head coverage (76.92%) is below the target coverage (80.00%). You can increase the head coverage or adjust the target coverage. Additional details and impacted files@@ Coverage Diff @@
## develop #9303 +/- ##
===========================================
+ Coverage 64.99% 65.00% +0.01%
===========================================
Files 2685 2685
Lines 422183 422070 -113
Branches 62726 62719 -7
===========================================
- Hits 274365 274339 -26
+ Misses 127113 127026 -87
Partials 20705 20705
*This pull request uses carry forward flags. Click here to find out more.
🚀 New features to boost your workflow:
|
No description provided.