This proposed refactor is part of TensorIR refactoring. https://discuss.tvm.apache.org/t/rfc-tensorir-a-schedulable-ir-for-tvm/7872
To enable the TensorIR scheduling, we will need to introduce support for more annotations to the loop node, specifically two fields will be introduced: - Optional<IterVar> thread_binding: Used for optionally bind the loop to a context thread such as GPU's blockIdx.x - Map<String, Object> annotations: Used for additional hint information used in high level transformations We will also remove the legacy `device_api` field as the field is no longer being used The refactor will change the IR construction API of ForNode, which requires migration of the code, but won't results in major refactor of the existing code as most cases CopyOnWrite are used. -- You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: https://github.com/apache/tvm/issues/7302
