Triton · Gluon · MLIR

See the compiler think.

Every pass, in completion order, with the full IR before and after. Browse pinned examples or bring your own trace manifest.

static catalog online

Compile and inspect

Use the same pass viewer for a kernel you just wrote.

LOCAL WASM READY
Pass-trace compiler

Bring a kernel.

Trace any decorated Triton or Gluon kernel locally—no GPU launch. The AMD + NVIDIA Pyodide compiler runs in a dedicated browser worker.

Kernel configuration

Exact decorated function name; leading underscores are valid.
One item per parameter. Numeric values supply tl.constexpr parameters.
BACKEND:ARCH:WARP_SIZE · e.g. hip:gfx950:64
Source
loading Python/Triton grammar + editor…
Paste any valid decorated Triton or Gluon kernel · 100 KB limit · import torch is supported for compile-only source and annotations · fixed compiler invocation · isolated temporary directory

Trace catalog

versioned pass-trace/v1 · full snapshots included

3 examples
triton / hip:gfx950:64 triton-softmax-gfx950

softmax

Triton softmax lowering with layout assignment, pipelining, and LLVM conversion.

68 passes 14 changed
triton / cuda:80:32 triton-softmax-cuda80

softmax

Triton softmax lowering with layout assignment, pipelining, and LLVM conversion.

74 passes 19 changed
gluon / hip:gfx950:64 gluon-attention-gfx950

fused_attention

Gluon attention lowering with scheduling, coalescing, and LLVM conversion.

29 passes 13 changed
Inspect a local PassTrace JSON never leaves your browser. It is validated and stored in IndexedDB for deep-linking.