Performance Speed Up (~5x): Implement adaptive step raytracing & stabilize N-body physics

[Quercius]

Overview

This PR significantly improves the overall performance of the simulation. By replacing the fixed-step integration in the compute shader with an adaptive approach and fixing the $O(N^2)$ gravity hot loop on the CPU, the rendering time has been cut by ~79%, yielding a ~4.7x speedup.

Key Changes

1. GPU Compute Shader (geodesic.comp)

• Adaptive Step Size: Replaced the brute-force, fixed D_LAMBDA step with a dynamically scaled step (current_dL). Rays far from the black hole take massive leaps, clamping to high-precision microscopic values (MIN_STEP) only when approaching extreme curvature.
• Smart Early Exit: Reduced ESCAPE_R to a realistic bound (1e12) and implemented an early exit for escaping rays (ray.r > ESCAPE_R && ray.dr > 0.0). This instantly kills rays heading into deep space, massively reducing Warp/Thread divergence.

2. CPU Physics Engine (black_hole.cpp)

• Fixed N-body Integration: The previous logic updated object positions during the nested distance calculation, leading to compounding mathematical errors. Forces are now properly accumulated into a totalAccel vector before applying velocities/positions.
• Delta Time (dt): Physics updates are now scaled by delta time, making the simulation framerate-independent.
• Hot Loop Cleanup: Removed expensive heap allocations (std::vector) and blocking I/O (std::cout) from the gravity hot loop, replacing them with stack-allocated glm::dvec3.

Performance & Benchmarks

Based on a 60-second weighted average sampling:

• Before: ~335 ms/frame (~3 FPS)
• After: ~71 ms/frame (~14 FPS)