Enabling L2+ Optimizations for EPs

[chilo-ms]

There are some requirements to modify the graph which are specific to the EP/hardware.
ORT has the hardcoded EP list for optimizations but that can't scale and it's hard be extended to enable EP custom optimizations.

This PR is to enable L2+ optimizations for EPs (The original prototype is provided by @skottmckay) as well as the TRT EP implementation for the ConstantFoldingDQ optimization.

The graph_optimizer_registry is designed for enabling L2+ graph optimizations tailored for EPs.
These optimizations are applied after the graph partitioner assigns ComputeCapability to the EP and before EP's "Compile" or fusion

Signatures for selection and optimization functions:

  - Selection: std::function<std::vector<std::unique_ptr<ComputeCapability>>(const GraphViewer&, const KeyValueConfig&, const GraphOptimizerRegistry&)>
  - Optimization: std::function<Status(const Graph&, const ComputeCapability& this_optimization, ComputeCapability& cc_to_update, const GraphOptimizerRegistry&)>

Optimizer's selection function:

• Selects a set of nodes from a given graph for optimization. Additional key/value strings can be provided to configure the optimizer. If needed, use graph_optimizer_registry to access the session options, the CPU EP and the logger.

Optimizer's optimization function:

• Gets the nodes in ComputeCapability from nodes_to_optimize. Use graph_optimizer_registry to access the session options, the CPU EP and the logger if needed to create the optimizer. Run optimization on the nodes/subgraph, and finally, update the ComputeCapability.

GetCapability

• call (new) provider bridge API to lookup pre-defined optimizer by name and get selection function

  • ComputeCapability.optimize_func, i.e. optimization function, would be set by the optimizer to the function that does the optimization

• EP has to update the returning ComputeCapability to include the optimization ComputeCapability in nodes_to_optimize. So that later ORT can perform optimization/transformation accordingly.

GraphPartitioner

• After assigning the ComputeCapability to the EP and prior to Compile, if the ComputeCapability has nodes_to_optimize, iterate that list
  • optimization function needs to be called with
    • a mutable Graph instance
    • the ComputeCapability for the individual optimization
    • the overall ComputeCapability so it can be updated

[github-actions]

You can commit the suggested changes from lintrunner.

[github-actions]

You can commit the suggested changes from lintrunner.