qp.backline.triton_decoder

triton_decoder(decoder_fns, **build_options)[source]

Compile Triton quantum error correction decoder functions into a coprocessor function for use with backline.

This function accepts a tuple of un-jitted Triton decoder functions, and compiles them into a shared library that can be used as a CoprocessorFunction.

Warning

Backline is experimental and only usable through the Catalyst compiler.

Parameters:

decoder_fns (tuple[object, ...]) – Un-jitted Triton decoder functions. Each entry is jit compiled internally, and decoder_id selects the tuple index at runtime.

Keyword Arguments:
  • platform (str) – Required Triton platform string of the form "backend:arch:warp_size". For example, "hip:gfx942:64" or "cuda:80:32".

  • grid (tuple[int, int, int]) – Triton kernel launch grid dimensions. Defaults to (1, 1, 1).

  • num_warps (int) – Triton kernel launch warp count. Defaults to 1.

  • num_stages (int) – Triton pipeline stage count. Defaults to 1.

  • compiler (str) – Optional compiler executable override. Defaults to "".

  • cflags (tuple[str, ...]) – Extra compiler flags. Defaults to ().

Returns:

The compiled decode function, ready to pass as

coprocessor_fn.

Return type:

CoprocessorFunction

Raises:
  • ImportError – If Triton decoder support is unavailable.

  • TypeError – If decoder_fns contains already jit compiled Triton functions.

  • ValueError – If the decoder build options are invalid.

Example

>>> import pennylane as qp
>>> import triton.language as tl
>>> def steane_lookup(syndrome):
...     return tl.where(syndrome != 0, 1 << (syndrome - 1), 0)
>>> decoder = qp.backline.triton_decoder(
...     (steane_lookup, steane_lookup),
...     platform="hip:gfx942:64",
... )