Perf: SIMD first-byte scan, eliminate try/catch and cached span in hot search loop

[Copilot]

The byte pattern scanner's hot path had several avoidable costs that compound badly when scanning large (50–100 MB) executables.

Hot-path changes (WitchHunt.cs)

• Removed try/catch from Match: blocked JIT register allocation, loop hoisting, and inlining — bounds are already checked explicitly
• [AggressiveInlining] on Match: eliminates call frame overhead on every byte of the buffer
• Made Match static, pass ReadOnlySpan<byte> as parameter: removes repeated ReadOnlyMemory<byte>.Span property access (allocates a new span header each call); cached once in FindSingle as var data = Data.Span
• Cached start.ToInt32(): was re-evaluated on every iteration of the while condition despite start being loop-invariant

SIMD first-byte skip (WitchHunt.cs)

When no match is found and the first byte has no wildcard (mask == 0xFF), replaced the manual byte-by-byte next-occurrence scan with MemoryExtensions.IndexOf:

// Before: manual O(n) loop
for (; indexOf < dataBuffer.Length; indexOf++)
{
    if ((dataBuffer[indexOf] & mask) != bmo) continue;
    break;
}

// After: SIMD-accelerated (SSE2/AVX2), processes 16–32 bytes per cycle
var nextIdx = remaining.IndexOf((byte)bmo);
return nextIdx < 0 ? -dataBuffer.Length : -(nextIdx + 1);

Pattern parsing (Utilities.cs)

Replaced Hexchars.Contains(t) — an O(24) linear string scan per character — with direct char range comparisons (>= '0' && <= '9', etc.). Removed the now-unused Hexchars constant.

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.