jax-ml/jax · error · NotImplementedError

Unsupported reduction kind: {op.kind}

Error message

Unsupported reduction kind: {op.kind}

What it means

vector.reduction lowering supports only add/max/min combining kinds; anything else (e.g. AND, OR, XOR) falls through to NotImplementedError.

Source

Thrown at jax/experimental/mosaic/gpu/dialect_lowering.py:871

  element_type = op.vector.type.element_type
  scratch = _slice_smem(
      ir.MemRefType.get([4], element_type, memory_space=utils.smem()),
      ir.IntegerAttr(op.attributes["offset"]).value,
      ctx.smem_requested_bytes,
  )
  axes = range(op.vector.type.rank)
  op_kind = _combining_kind(op.kind)
  is_signed = _is_reduction_signed(op_kind)
  a = _fragmented_array_from_ir(op.vector, layout, is_signed)
  match op_kind:
    case vector.CombiningKind.ADD:
      result = a.reduce("add", axes, scratch)
    case vector.CombiningKind.MAXSI | vector.CombiningKind.MAXUI | vector.CombiningKind.MAXIMUMF:
      result = a.reduce("max", axes, scratch)
    case vector.CombiningKind.MINUI | vector.CombiningKind.MINSI | vector.CombiningKind.MINIMUMF:
      result = a.reduce("min", axes, scratch)
    case _:
      raise NotImplementedError(f"Unsupported reduction kind: {op.kind}")
  assert isinstance(result.layout, fa.WGSplatFragLayout)
  return [result.registers.item()]


@_register_lowering(vector.MultiDimReductionOp)
def _vector_multi_dim_reduction_op_lowering_rule(
    ctx: LoweringContext, op: vector.MultiDimReductionOp
) -> Sequence[ir.Value]:
  [in_layout, acc_layout] = inference_utils.in_layouts(op)
  [out_layout] = inference_utils.out_layouts(op)
  if out_layout != acc_layout:
    raise ValueError(
        f"Output layout {out_layout} must match the accumulator layout"
        f" {acc_layout}"
    )

  if len(op.reduction_dims) != 1:
    raise NotImplementedError("Only 1 reduction dimension is supported.")

View on GitHub (pinned to 1e1c6a8fc0)

Solutions

1. Implement bitwise reduction manually via elementwise ops + add reduction on integer representation
2. Restrict to add/max/min reductions

SUPPORTED = {'add', 'max', 'min'}
assert kind in SUPPORTED, f'reduction kind {kind} unsupported'

When it happens

Trigger: Emitting vector.reduction with kind like AND/OR/XOR in a Mosaic kernel.

Common situations: Porting code that uses bitwise vector reductions, which Mosaic GPU does not lower.

Understand the failure class

Background: UnsupportedOperationException and "is not supported" errors: when a library deliberately refuses a call — this error's family across 30 libraries.

Related errors

• `strides` must contain only 1s.
• Only slicing with static indices allowed.
• Output layout {out_layout} must match the accumulator layout
• Only 1 reduction dimension is supported.
• Unsupported layout: {src.layout}

AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/957450bd7146ddd6. Report an issue: GitHub.