Optimize VM performance with pre-compiled jump table

[Jim8y]

Description

This PR significantly improves Neo VM performance by replacing the reflection-based jump table initialization with a pre-compiled static approach.

Performance Improvements

Benchmark Results:

• Jump Table Construction: 91μs → 6.8μs (13.3x faster)
• VM Initialization: 88μs → 0.69μs (127x faster)
• Overall Performance Gain: 1,235% - 12,613%

Key Changes:

1. Replaced runtime reflection with compile-time method assignments
2. Eliminated expensive and calls
3. Direct opcode-to-method mapping using static initialization

Technical Details

Before (Reflection-based):

foreach (var mi in GetType().GetMethods())
{
    if (Enum.TryParse<OpCode>(mi.Name, true, out var opCode))
    {
        Table[(byte)opCode] = (DelAction)mi.CreateDelegate(typeof(DelAction), this);
    }
}

After (Pre-compiled):

Table[(byte)OpCode.PUSH1] = Push1;
Table[(byte)OpCode.ADD] = Add;
Table[(byte)OpCode.MUL] = Mul;
// ... direct assignments for all opcodes

Testing

• ✅ All existing unit tests pass (1,599+ tests)
• ✅ Added comprehensive benchmarks
• ✅ No breaking changes to public API
• ✅ Full backward compatibility maintained

Benchmarks

Two benchmark suites included:

1. • Full BenchmarkDotNet suite
2. • Quick demonstration

Run benchmarks:

NEO_VM_JUMPTABLE_BENCHMARK=1 dotnet run -c Release -f net9.0

Impact

This optimization provides dramatic performance improvements for:

• Node startup times
• Smart contract execution initialization
• High-throughput scenarios with multiple VM instances
• Development and testing cycles

The changes are particularly beneficial for blockchain nodes that need to create many VM instances for transaction processing.

[Wi1l-B0t]

Good job !

[Wi1l-B0t]

Where ar pre-compiled jump tables saved? Or need to compile every time?

[Jim8y]

no longer needed, close to clean up the pr list, such that we can focus on active prs.