#!/usr/bin/env python3
"""Insert perturb_stress_kernel into observer .cu before calculate_asymmetry_magnifying."""
path = '/mnt/d/resonance-engine-active/cuda/khra_gixx_1024_v5_observer.cu'
with open(path, 'r', encoding='utf-8') as f:
    lines = f.readlines()

# Find the line with 'float calculate_asymmetry_magnifying'
insert_idx = None
for i, line in enumerate(lines):
    if 'float calculate_asymmetry_magnifying' in line:
        # Find the closing brace of inject_density_kernel just before it
        for j in range(i - 1, -1, -1):
            if lines[j].strip() == '}':
                insert_idx = j + 1  # insert after this brace
                break
        break

if insert_idx is None:
    print('ERROR: could not find insertion point')
    exit(1)

kernel_lines = [
    '// === STRESS PERTURBATION KERNEL ===\n',
    '// Writes a perturbation into the NON-EQUILIBRIUM (shear-stress) moment.\n',
    '// f[i] += G * d_cx[i] * d_cy[i] -- the sxy basis has zero density moment,\n',
    '// zero momentum moment, and non-zero sxy shear-stress. Pure f_neq write.\n',
    '// Same Gaussian targeting as inject_density: (x, y, sigma, strength), periodic.\n',
    '__global__ void perturb_stress_kernel(float* f,\n',
    '                                      float cx, float cy,\n',
    '                                      float sigma, float strength) {\n',
    '    int x = blockIdx.x * blockDim.x + threadIdx.x;\n',
    '    int y = blockIdx.y * blockDim.y + threadIdx.y;\n',
    '    if (x >= NX || y >= NY) return;\n',
    '\n',
    '    float dx = (float)x - cx;\n',
    '    float dy = (float)y - cy;\n',
    '    if (dx >  NX * 0.5f) dx -= NX;\n',
    '    if (dx < -NX * 0.5f) dx += NX;\n',
    '    if (dy >  NY * 0.5f) dy -= NY;\n',
    '    if (dy < -NY * 0.5f) dy += NY;\n',
    '\n',
    '    float r2 = dx * dx + dy * dy;\n',
    '    float G = strength * expf(-r2 / (2.0f * sigma * sigma));\n',
    '\n',
    '    int f_idx = (y * NX + x) * Q;\n',
    '    for (int i = 0; i < Q; i++) {\n',
    '        f[f_idx + i] += G * d_cx[i] * d_cy[i];\n',
    '    }\n',
    '}\n',
    '\n',
]

# Insert kernel lines
for i, line in enumerate(kernel_lines):
    lines.insert(insert_idx + i, line)

with open(path, 'w', encoding='utf-8') as f:
    f.writelines(lines)

print(f'Inserted {len(kernel_lines)} lines at index {insert_idx + 1}')

# Verify
with open(path, 'r', encoding='utf-8') as f:
    content = f.read()
if 'perturb_stress_kernel' in content:
    print('VERIFIED: kernel present')
else:
    print('ERROR: kernel not found')