int_select_assign with {other:?} update is not implemented

Error message

int_select_assign with {other:?} update is not implemented

What it means

int_select_assign on the CubeCl backend only supports IndexingUpdateOp::Add (kernel::select_assign) and IndexingUpdateOp::Mul (kernel::select_assign_mul). Any other update op variant falls into the catch-all arm and panics with this unimplemented! message. It is a backend feature gap, not a user data error.

Source

Thrown at crates/burn-cubecl/src/ops/int_tensor.rs:183

    fn int_select_assign(
        tensor: IntTensor<Self>,
        dim: usize,
        indices: IntTensor<Self>,
        value: IntTensor<Self>,
        update: burn_backend::tensor::IndexingUpdateOp,
    ) -> IntTensor<Self> {
        match update {
            burn_backend::tensor::IndexingUpdateOp::Assign => {
                kernel::select_assign_replace(tensor, dim, indices, value)
            }
            burn_backend::tensor::IndexingUpdateOp::Add => {
                kernel::select_assign(tensor, dim, indices, value, false)
            }
            burn_backend::tensor::IndexingUpdateOp::Mul => {
                kernel::select_assign_mul(tensor, dim, indices, value)
            }
            other => {
                unimplemented!("int_select_assign with {other:?} update is not implemented")
            }
        }
    }

    fn int_equal(
        lhs: IntTensor<Self>,
        rhs: IntTensor<Self>,
        out_dtype: BoolDType,
    ) -> BoolTensor<Self> {
        kernel::equal(lhs, rhs, out_dtype.into())
    }

    fn int_not_equal(
        lhs: IntTensor<Self>,
        rhs: IntTensor<Self>,
        out_dtype: BoolDType,
    ) -> BoolTensor<Self> {
        kernel::not_equal(lhs, rhs, out_dtype.into())

View on GitHub (pinned to d16f7ba2ed)

Solutions

Use a supported op: Add or Mul; emulate Sub by assigning the negated value with Add.
Replace with gather/compute/scatter: gather at indices, apply the update, then select_assign with Add/Mul or slice_assign.
Patch crates/burn-cubecl/src/ops/int_tensor.rs to dispatch the missing variant to a new int select_assign kernel.
Check backend compatibility before using advanced indexing-update ops on IntTensors.

Example fix

// before
let out = tensor.select_assign(dim, indices, updates, IndexingUpdateOp::Sub);
// after
let out = tensor.select_assign(dim, indices, -updates, IndexingUpdateOp::Add);

Defensive patterns

Strategy: validation

Validate before calling

if !matches!(op, IndexingUpdateOp::Add | IndexingUpdateOp::Mul) {
    panic!("int_select_assign on CubeCl supports only Add/Mul, got {op:?}");
}

Type guard

fn is_supported_select_op(op: &IndexingUpdateOp) -> bool {
    matches!(op, IndexingUpdateOp::Add | IndexingUpdateOp::Mul)
}

Prevention

Restrict select_assign update ops to Add/Mul for IntTensors on CubeCl
Convert subtraction updates to Add-of-negated before assigning
Keep a compatibility table of ops vs backends in CI tests
Prefer slice_assign for exotic update semantics

When it happens

Trigger: Calling int_select_assign (or Tensor::select_assign with an update op) on an IntTensor with the CubeCl backend while passing an IndexingUpdateOp other than Add or Mul (e.g. Sub).

Common situations: Using conditional assignment with subtraction updates on integer tensors; code that worked on a float backend or another backend being reused for int tensors; new op variants added upstream without int kernel support.

Related errors

AI-assisted analysis of tracel-ai/burn@d16f7ba2ed (2026-09-05). Data as JSON: /api/errors/1352a71a21a05d1a. Report an issue: GitHub.