Constant PGA_SCLERP_EVAL_WGSL
Source pub const PGA_SCLERP_EVAL_WGSL: &str = "// PGA ScLERP Batch Compute Kernel (3D Projective Geometric Algebra)\n// Evaluates parallel Screw Linear Interpolation for N instances across SE(3) geodesics.\n\nstruct Motor {\n r_w: f32,\n r_x: f32,\n r_y: f32,\n r_z: f32,\n d_w: f32,\n d_x: f32,\n d_y: f32,\n d_z: f32,\n};\n\nstruct ScLerpParams {\n count: u32,\n t: f32,\n _pad0: f32,\n _pad1: f32,\n};\n\n@group(0) @binding(0) var<uniform> params: ScLerpParams;\n@group(0) @binding(1) var<storage, read> motors_start: array<Motor>;\n@group(0) @binding(2) var<storage, read> motors_end: array<Motor>;\n@group(0) @binding(3) var<storage, read_write> motors_out: array<Motor>;\n\nfn motor_conjugate(m: Motor) -> Motor {\n return Motor(m.r_w, -m.r_x, -m.r_y, -m.r_z, m.d_w, -m.d_x, -m.d_y, -m.d_z);\n}\n\nfn motor_mul(a: Motor, b: Motor) -> Motor {\n let rw = a.r_w * b.r_w - a.r_x * b.r_x - a.r_y * b.r_y - a.r_z * b.r_z;\n let rx = a.r_w * b.r_x + a.r_x * b.r_w + a.r_y * b.r_z - a.r_z * b.r_y;\n let ry = a.r_w * b.r_y - a.r_x * b.r_z + a.r_y * b.r_w + a.r_z * b.r_x;\n let rz = a.r_w * b.r_z + a.r_x * b.r_y - a.r_y * b.r_x + a.r_z * b.r_w;\n\n let dw = a.r_w * b.d_w - a.r_x * b.d_x - a.r_y * b.d_y - a.r_z * b.d_z\n + a.d_w * b.r_w - a.d_x * b.r_x - a.d_y * b.r_y - a.d_z * b.r_z;\n let dx = a.r_w * b.d_x + a.r_x * b.d_w + a.r_y * b.d_z - a.r_z * b.d_y\n + a.d_w * b.r_x + a.d_x * b.r_w + a.d_y * b.r_z - a.d_z * b.r_y;\n let dy = a.r_w * b.d_y - a.r_x * b.d_z + a.r_y * b.d_w + a.r_z * b.d_x\n + a.d_w * b.r_y - a.d_x * b.r_z + a.d_y * b.r_w + a.d_z * b.r_x;\n let dz = a.r_w * b.d_z + a.r_x * b.d_y - a.r_y * b.d_x + a.r_z * b.d_w\n + a.d_w * b.r_z + a.d_x * b.r_y - a.d_y * b.r_x + a.d_z * b.r_w;\n\n return Motor(rw, rx, ry, rz, dw, dx, dy, dz);\n}\n\n@compute @workgroup_size(64)\nfn pga_sclerp_main(@builtin(global_invocation_id) global_id: vec3<u32>) {\n let idx = global_id.x;\n if (idx >= params.count) {\n return;\n }\n\n let m0 = motors_start[idx];\n let m1 = motors_end[idx];\n let t = params.t;\n\n let m0_inv = motor_conjugate(m0);\n var delta = motor_mul(m0_inv, m1);\n\n if (delta.r_w < 0.0) {\n delta = Motor(-delta.r_w, -delta.r_x, -delta.r_y, -delta.r_z,\n -delta.d_w, -delta.d_x, -delta.d_y, -delta.d_z);\n }\n\n // Logarithm -> Scale by t -> Exponential\n let sin_sq = delta.r_x * delta.r_x + delta.r_y * delta.r_y + delta.r_z * delta.r_z;\n var res_motor: Motor;\n\n if (sin_sq < 1e-8) {\n // Pure translation limit\n let scaled_t = Motor(1.0, 0.0, 0.0, 0.0,\n 0.0, delta.d_x * t, delta.d_y * t, delta.d_z * t);\n res_motor = motor_mul(m0, scaled_t);\n } else {\n let sin_theta = sqrt(sin_sq);\n let theta = atan2(sin_theta, delta.r_w);\n let inv_sin = 1.0 / sin_theta;\n\n let pitch = -delta.d_w * inv_sin;\n let axis = vec3<f32>(delta.r_x * inv_sin, delta.r_y * inv_sin, delta.r_z * inv_sin);\n\n let scaled_theta = theta * t;\n let sin_scaled = sin(scaled_theta);\n let cos_scaled = cos(scaled_theta);\n\n let exp_rw = cos_scaled;\n let exp_rx = axis.x * sin_scaled;\n let exp_ry = axis.y * sin_scaled;\n let exp_rz = axis.z * sin_scaled;\n\n let d_trans = vec3<f32>(delta.d_x - pitch * delta.r_x,\n delta.d_y - pitch * delta.r_y,\n delta.d_z - pitch * delta.r_z) * t;\n\n let exp_motor = Motor(exp_rw, exp_rx, exp_ry, exp_rz,\n 0.0, d_trans.x, d_trans.y, d_trans.z);\n res_motor = motor_mul(m0, exp_motor);\n }\n\n motors_out[idx] = res_motor;\n}\n";