iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Docs] Explain pointer-alignment JIT specialization in Triton skill (#57669)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
This commit is contained in:
Woosuk Kwon
2026-09-19 00:53:42 -07:00
committed by GitHub
parent 38537331d3
commit c3b4844634
1 changed files with 11 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.agents/skills/triton-kernel-writing/SKILL.md
+11 -5
View File
@@ -22,10 +22,14 @@ description: Write or review Triton kernels for vLLM, with practical guidance fo
a small, legible heuristic when workloads need different choices. Use a small, legible heuristic when workloads need different choices. Use
`triton.autotune` only when tuning is critical to performance, such as for a `triton.autotune` only when tuning is critical to performance, such as for a
matrix multiplication. Otherwise prioritize simple code and fast startup. matrix multiplication. Otherwise prioritize simple code and fast startup.
- Be careful to avoid unintended runtime JIT compilation. For example, put - Avoid unintended runtime JIT compilation. Put unimportant runtime integer
unimportant runtime integer scalars in `do_not_specialize`, especially those scalars in `do_not_specialize`, especially values that alternate between 0
that may alternate between values such as 0 and 1, which can produce and 1. Pointer alignment also affects specialization: an offset tensor view
different specialization keys. can trigger a new variant after warmup used an aligned allocation. For
scalar-only pointers whose alignment provides no useful optimization, use
`@triton.jit(do_not_specialize_on_alignment=["end_ptr"])`. Retain alignment
specialization when vectorized accesses benefit from it. See the
[Triton JIT options](https://triton-lang.org/main/python-api/generated/triton.jit.html).
- The Triton compiler does not guarantee safe ordering when a kernel writes to - The Triton compiler does not guarantee safe ordering when a kernel writes to
a pointer and subsequently reads from the same pointer. This pattern must a pointer and subsequently reads from the same pointer. This pattern must
have a `tl.debug_barrier()` between the write and read. The barrier have a `tl.debug_barrier()` between the write and read. The barrier
@@ -49,7 +53,9 @@ description: Write or review Triton kernels for vLLM, with practical guidance fo
## Validation ## Validation
- Check correctness at boundary shapes and at sizes that exercise masks and - Check correctness at boundary shapes and at sizes that exercise masks and
large offsets. large offsets. Include offset tensor views, not only fresh allocations.
Warm aligned buffers, then exercise offset views with vLLM's
`--jit-monitor-mode error --jit-monitor-verbose` to catch missing variants.
- Choose accumulation and intermediate dtypes explicitly. Test numerically - Choose accumulation and intermediate dtypes explicitly. Test numerically
difficult inputs, not only random, well-scaled tensors. difficult inputs, not only random, well-scaled tensors.
- Use `$kernel-microbenchmark` for benchmark construction, measurement, and - Use `$kernel-microbenchmark` for benchmark construction, measurement, and