Skip to content

docs: add TileIR backend (ENABLE_TILE) usage guide#44

Merged
msaroufim merged 8 commits intogpu-mode:mainfrom
yf225:docs/tileir-usage-guide
Mar 14, 2026
Merged

docs: add TileIR backend (ENABLE_TILE) usage guide#44
msaroufim merged 8 commits intogpu-mode:mainfrom
yf225:docs/tileir-usage-guide

Conversation

@yf225
Copy link
Contributor

@yf225 yf225 commented Mar 13, 2026

Summary

  • Adds a new "Using the TileIR Backend" section to docs/helion-hackathon.md
  • Documents ENABLE_TILE=0 vs ENABLE_TILE=1 and the TileIR compilation pipeline (Triton IR → TileIR → CUBIN via tileiras)
  • Explains how to enable TileIR with Helion (ENABLE_TILE=1 + HELION_BACKEND=tileir)
  • Compares tunables: TileIR uses num_ctas/occupancy instead of num_warps/maxnreg
  • Shows how to hardcode TileIR configs in submissions (including the env var setup)
  • Recommends trying both backends and submitting whichever is faster

Test plan

  • Verified markdown renders correctly
  • Verified against helion source (_compat.py, config_spec.py, backend.py, settings.py) that described behavior is accurate
  • Confirmed TileIR tunables match helion's TileIRBackend.tunable_fragments(): num_ctas (1-2), occupancy (1-8)

🤖 Generated with Claude Code

@yf225 yf225 changed the title docs: add TileIR backend (ENABLE_TILE) usage guide [WIP] docs: add TileIR backend (ENABLE_TILE) usage guide Mar 13, 2026
Documents ENABLE_TILE=0 vs ENABLE_TILE=1 and the TileIR compilation
pipeline available via nvtriton on B200 instances. Covers how to enable
TileIR with Helion (ENABLE_TILE=1 + HELION_BACKEND=tileir), the
different tunables (num_ctas/occupancy vs num_warps/maxnreg), and how
to hardcode TileIR configs in submissions.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@yf225 yf225 force-pushed the docs/tileir-usage-guide branch from 9e72871 to a4ac5c4 Compare March 13, 2026 22:42
yf225 and others added 4 commits March 13, 2026 15:46
Group both sections under a single "Optional: Extra Performance Knobs"
heading to emphasize neither is required. Streamline both into
step 1 (autotune) / step 2 (hardcode) format. Add a "Which combination"
section showing all 4 options to try.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove duplicate bash export block — the Python os.environ in the
code example is sufficient for both local autotuning and submissions.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@codecov
Copy link

codecov bot commented Mar 13, 2026

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

yf225 and others added 2 commits March 13, 2026 15:56
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@yf225 yf225 changed the title [WIP] docs: add TileIR backend (ENABLE_TILE) usage guide docs: add TileIR backend (ENABLE_TILE) usage guide Mar 13, 2026
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@yf225 yf225 closed this Mar 14, 2026
@yf225 yf225 reopened this Mar 14, 2026
@msaroufim msaroufim merged commit 44a8b35 into gpu-mode:main Mar 14, 2026
14 of 16 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants