Profiling

[ryanwilsond]

Description

Very simple profiler to compare the efficiency of the Evaluator, Interpreter, and Executor.

Findings so far:

• Interpreter is always the slowest
• Executor beats Evaluator unless the input is very short and linear

Next research:

• Wider variety of tests (arrays, higher heap stress, etc.)
• (Stretch) use each test and calculate node frequency scores compounded by test time to over time create a heuristic for expensive vs cheap nodes to potentially use in a deferring algorithm for using the Evaluator over Executor

The AutoRun build mode will now always default to Executor (assuming bug-free)
Can later add some logic to redirect to the Evaluator if the program is tiny and dynamic loop-free (could statically check loops and still defer to Evaluator if they are known to be small in scope)

Also fixed a lot of bugs with the CodeGenerator, especially around arrays.