MXNet will integrate with many different kinds of acceleration libraries, including TVM, MKLDNN, TensorRT, Intel nGraph and more. These accelerators in general support a limited number of operators, and running computation in a model usually involves in interaction between accelerator operators and MXNet operators.
These accelerators share some common requirements:
- TVM , MKLDNN and nGraph uses customized data formats. Interaction between these accelerators with MXNet requires data format conversion.
- TVM, MKLDNN, TensorRT and nGraph provide fused operators.
Integration with these accelerators should happen in the granularity of subgraphs instead of in the granularity of operators. For fused operators, it's obvious that we need to divide a graph into subgraphs so that the operators in a subgraph can be replaced by a fused operator. To handle customized data formats, we should partition a computation graph into subgraphs as well. Each subgraph contains only the operators supported by the accelerators. In this way, MXNet converts data formats only when entering such a subgraph and the operators inside a subgraph handle format conversion themselves if necessary. This makes interaction of TVM and MKLDNN with MXNet much easier. Neither the MXNet executor nor the MXNet operators need to deal with customized data formats.
As such, integration with these accelerators may result in two levels of graph partitioning. In the first level, a subgraph only contains the operators supported by the accelerator. In the second level, a subgraph only contains the operators that can be fused.
- TVM requires two levels of partitioning because TVM searches for global scheduling among fused operators in order to achieve the best performance.
- MKLDNN requires two levels of partitioning. We want to isolate MKLDNN operators from MXNet operators so that we know where to insert MKLDNN format conversion operators. MKLDNN also wants to fuse operators to achieve the optimal performance.
- TensorRT requires only one level of partitioning. It only needs to fuse operators.
- nGraph probably has the same requirement as MKLDNN.
Step 1: graph partition
Graph partitioning is to traverse a computation graph and group operators into subgraphs based on certain rules. There already exists an TVM fuse pass in NNVM, which groups operators into subgraphs based on certain general rules (e.g., convolution followed by element-wise operations). This graph partitioner is TVM-specific. It doesn't work for other accelerators. We need more graph partitioners. For example, TensorRT and MKLDNN requires a partitioner that finds subgraphs with specific patterns (e.g., convolution followed by batchnorm, followed by activation, etc).
After generating a subgraph, we have two choices: replace a subgraph with a specific fused operator pre-registered in MXNet, or pass the subgraph to a special operator (e.g., TVM operator, MKLDNN operator) to invoke the computation in the subgraph. In the case of TVM, after we generate the first level of subgraphs, we need to write them out so that TVM can compile and find the optimal scheduling offline.
Step 2: subgraph operator (function call)
Although there are two levels of graph partitioning, we only need to handle one level of subgraphs in the executor because the subgraphs in the second level are fused into operators. We can execute these subgraphs inside special operators, which is specific to the accelerator.
- TVM execution operator: loads a subgraph from a TVM compiled binary, a graph JSON file and weight arrays, and executes the subgraph composed of fused operators. We can first use the TVM executor to execute the subgraph, but in the future we should use the MXNet executor because MXNet executes operators in multiple threads, which is useful for task parallelism. The operator needs to convert all output NDArrays of the subgraph to the default format.
- MKLDNN execution operator: gets a subgraph from the first step and runs operators in the MXNet executors. Like TVM operator, this operator also needs to convert all output NDArrays of the subgraph to the default format.
- TensorRT doesn't need a specific operator because it only has operator fusion.
- nGraph execution operators: should be similar to the MKLDNN execution operator.
For fast inference in TVM and MKLDNN, the subgraph execution operators need to maintain a copy of weight arrays (similar to the closure of a function). In this way, we can convert the data format of the weight arrays and cache the array inside the subgraph operator to avoid any redundant format conversion. The original weight arrays will still be part of the inputs of the subgraph operator. Even though the weight arrays are normally not modified, we still need to handle this case correctly. One solution is to maintain a version number for the var of an NDArray, which is increased by one whenever the NDArray is modified in the execution engine. We can use the version number to determine the weight arrays have been modified whenever the subgraph operator is invoked.
The benefit of invoking a subgraph inside an operator
Introducing a subgraph execution operator for TVM and MKLDNN may sound like unnecessary complexity. It actually significantly reduces the complexity of the integration. By using the subgraph operator, we can completely isolate TVM operators and MKLDNN operators from MXNet operators as well as the default MXNet memory planning. Inside the subgraph operators, we don't need to deal with data format conversion and can use a completely different memory plan for the subgraph.