[QDP] Pr1 phase kernel opt

[aloha1357]

Related Issues

related #1385

Changes

• Bug fix
• New feature
• Refactoring
• Documentation
• Test
• CI/CD pipeline
• Other

Why

The original phase encoding and IQP encoding kernels suffered from GPU thread divergence due to conditional branching (if (val != 0.0) or if ((x >> i) & 1U)). Furthermore, the normalization factor (norm_factor) was being redundantly calculated inside the GPU kernel, consuming extra cycles. Eliminating these inefficiencies significantly improves the kernel's execution speed on the GPU.

How

• Replaced Conditional Branching: In both phase.cu and iqp.cu, the if conditions checking bit states were replaced with boolean arithmetic casting and multiplication (e.g., phases[bit] * (double)((idx >> bit) & 1U)). This ensures that all threads in a warp follow the exact same instruction path, eliminating warp divergence.
• Host-side Pre-calculation: Moved the norm_factor calculation to the host (CPU) before launching the kernel in phase.cu, passing the result as an immutable parameter.
• Added Explanatory Comments: Included inline documentation near the bitwise arithmetic lines to aid code reviewers in understanding the optimizations.

Benchmark Results

Environment: Dev Machine (NVIDIA RTX 4060)
Configuration: Qubits (N): 14, Batch Size: 128, Iterations: 5

Implementation	Execution Time (ms)	Notes
GPU phase (Before PR1)	1.26 ms (per sample) / 161.91 ms (total)	Strict checkout of unoptimized phase.cu
GPU phase (This PR)	1.16 ms (per sample) / 147.94 ms (total)	~9.4% Performance Gain with zero divergence.

Checklist

• Added or updated unit tests for all changes (Verified passing against existing CI test suite)
• Added or updated documentation for all changes (Added explanatory inline comments for PR)

[ryankert01]

I need you to do four things:

1. show the benchmark of this changes (before & after)
2. enhance our unit tests for this function
3. cleanup the code & comments from coding agents
4. read CONTRIBUTING.md

Thanks!