Purpose

JIL Execution in the VM is responsible for a small but important subset of node execution. While most nodes today are executed via the FFI (Foreign Function Interpreter) mechanisms which invoke C# code, some basic utility methods especially in Math category are written in native design script. This type of function is handled by the JILFunctionEndPoint. While many of these enhancements are targeted at this execution path, they also yield net improvements to the general execution of all EndPoint types. This is especially true in the POP_Handler, DEP_Handler, SetupExecutive, and RestoreFromCall methods within the VM's Executive. For a specific test graph with a mixture of geometry and mathematical operations, these enhancements specifically reduce overall UpdateGraph run time by 35% (17s to 13s). The net impact is also dramatically reduces temp memory allocation. For the case of this specific test graph the temporary memory allocation associated with UpdateGraph went from 11.3gb to 3.6gb. In summary this PR optimizes the execution of functions handled via the JIL Endpoint but has a net improvement to all node types.

Specifically this PR does the following

• Clean up dependency on CurrentStackFrame property of RuntimeMemory. This getter creates a new StackFrame object to reference a subset of items at a specific location in the VM's Stack. The CurentStackFrame property is utilized 99% of the time from the IsGlobalScope method. This issue is IsGlobalScope can be called tens of millions of times during a Graph Execution run which creates a new StackFrame object every time the CurrentStackFrame property is referenced. This optimization simply removes the need to allocate a temporary StackValue object when the data which is needed can be easily referenced directly from the Stack. This optimization represents the majority of the extra gigabytes of temporary allocation described above in the sample graph performance delta.

• Refactor RestoreFromCall to not allocate empty list until after the required check of runtimeCore.Options.RunMode == InterpreterMode.Expression. This allocation is done repeatedly with no items added to the collection. The call later to check the list via Any() checks can be refactored to a null check.

• Fast path for GetGraphNodesAtScope when asking for a Invalid ClassIndex and ProcessIndex (ie -1). This is another function that can be called millions of times during a UpdateGraph run. Many calls that are routed through this function are looking for the same item in the graphNodeMap dictionary. This optimization creates a shortcut when the lookup is asking for the specific case of the invalid ClassIndex and ProcessIndex that avoids accessing the object from the dictionary. Note, in the case of this function, caching the previous lookup would not speed up the lookup as the method usage typically alternates between values.

• Refactor UpdateGraph to not allocate a temporary list of GraphNodes.

Declarations

Check these if you believe they are true

• The codebase is in a better state after this PR
• Is documented according to the standards
• The level of testing this PR includes is appropriate
• User facing strings, if any, are extracted into *.resx files
• All tests pass using the self-service CI.
• Snapshot of UI changes, if any.
• Changes to the API follow Semantic Versioning and are documented in the API Changes document.
• This PR modifies some build requirements and the readme is updated

Reviewers

@sm6srw @aparajit-pratap

FYIs

@jasonstratton @mjkkirschner