[Agents] Link Triton skill to JIT kernel warmup guide (#56499)
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
@@ -26,6 +26,10 @@ description: Write or review Triton kernels for vLLM, with practical guidance fo
|
||||
unimportant runtime integer scalars in `do_not_specialize`, especially those
|
||||
that may alternate between values such as 0 and 1, which can produce
|
||||
different specialization keys.
|
||||
- For new warmable kernels, follow the
|
||||
[JIT Kernel Warmup guide](../../../docs/contributing/jit_kernel_warmup.md)
|
||||
to define the warmup wrapper and call its `register_warmup(...)` where the
|
||||
runtime implementation is selected during runner setup.
|
||||
- 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
|
||||
have a `tl.debug_barrier()` between the write and read. The barrier
|
||||
|
||||
Reference in New Issue
Block a user