#include "psprecomp/runtime.hpp"
#include "generated_units.hpp"
#include <bit>
#include <cmath>
#include <cstdint>
#include <limits>

namespace psprecomp {
static const std::uint16_t kEntryIds_recomp_unit_0132[4096] = {
    1, 0, 0, 0, 2, 0, 0, 0, 3, 0, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 5, 0, 0, 6, 0, 0, 7, 0, 0, 0, 0, 0,
    0, 0, 0, 8, 0, 0, 0, 0, 9, 0, 0, 0, 10, 0, 11, 0, 12, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    13, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 14, 0, 15, 0, 16, 17, 0, 0, 0, 0, 0, 18, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 19, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 20, 0, 21, 0, 22, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 23, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 24, 0, 0, 25, 0, 0,
    0, 0, 0, 0, 26, 0, 0, 0, 27, 0, 0, 0, 28, 0, 0, 0, 0, 29, 0, 0, 0, 0, 30, 0, 0, 0, 0, 0, 31, 0, 0, 0,
    32, 0, 0, 0, 0, 0, 33, 0, 0, 0, 0, 0, 0, 0, 34, 0, 0, 0, 0, 35, 0, 0, 36, 0, 0, 0, 0, 0, 37, 0, 0, 0,
    0, 0, 38, 0, 0, 0, 0, 0, 39, 0, 0, 0, 0, 0, 0, 0, 0, 40, 0, 0, 0, 0, 0, 41, 0, 0, 42, 0, 0, 0, 0, 0,
    43, 0, 0, 0, 0, 0, 44, 0, 0, 0, 0, 0, 0, 0, 0, 45, 0, 0, 0, 0, 0, 0, 46, 0, 0, 47, 0, 0, 0, 0, 0, 48,
    0, 0, 0, 0, 0, 49, 0, 0, 0, 0, 0, 0, 0, 0, 50, 0, 0, 0, 0, 0, 0, 51, 0, 0, 52, 0, 0, 0, 0, 0, 53, 0,
    0, 0, 0, 0, 54, 0, 0, 0, 0, 0, 0, 0, 0, 55, 0, 0, 0, 0, 0, 0, 56, 0, 0, 0, 0, 0, 0, 57, 0, 0, 0, 58,
    0, 0, 0, 59, 0, 0, 0, 0, 0, 0, 60, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 61, 0, 0, 0, 0, 0, 0, 0, 62, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 63, 0, 0, 0, 0, 0, 64, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 65, 0, 0, 0, 0, 0, 0, 66, 0, 0, 0, 67, 0, 0, 68, 0, 69, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 70, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 71, 0, 0, 72, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 73, 0, 74, 0, 75, 0, 0, 0, 0, 0, 0, 0, 0, 76, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 77, 0, 0, 0, 0, 0, 78, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 79, 0, 0,
    0, 0, 0, 0, 0, 0, 80, 0, 0, 81, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 82, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 83, 84, 0, 0, 0, 0, 85, 86, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 87, 0, 0, 0, 0,
    88, 0, 89, 0, 90, 0, 0, 0, 0, 0, 91, 0, 0, 0, 0, 0, 0, 92, 0, 0, 0, 0, 0, 0, 93, 0, 94, 0, 95, 0, 0, 0,
    0, 0, 0, 0, 96, 0, 97, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 98, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 99, 0, 0, 0, 0,
    0, 0, 0, 0, 100, 0, 101, 0, 0, 102, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 103, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 104, 105, 0, 0, 0, 106, 107, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 108, 0, 0, 0, 0, 109, 0, 0, 0, 0, 0, 0, 0, 110, 0, 111, 0, 112, 0, 0,
    0, 0, 0, 113, 0, 0, 0, 0, 0, 0, 114, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 115, 0, 0, 0, 0, 0, 0, 0, 0, 116, 0, 117, 0, 0, 118, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 119, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 120, 121, 0, 0, 0,
    122, 123, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 124, 0, 0, 0, 0, 0, 0, 0, 125, 0, 0, 0, 0,
    126, 0, 127, 0, 128, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 129, 0, 0, 0, 0, 0, 0, 0, 0, 130, 0, 0, 131, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    132, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 133, 134, 0, 0, 0, 135, 136, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 137, 0, 0, 0, 0, 0, 0, 0, 138, 0, 0, 0, 0, 139, 0, 0, 0, 0, 0, 0, 0, 140, 0, 141, 0, 142, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 143, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 144, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 145, 0, 0, 0, 0, 0, 146, 0, 0, 0, 147, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 148, 0, 0, 0, 149, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 150, 0, 0, 0, 0, 151, 0, 152, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 153, 0, 154, 0, 0, 155, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 156, 0, 157, 0, 0, 0, 0, 0,
    0, 158, 0, 0, 0, 159, 0, 0, 0, 0, 0, 0, 160, 0, 0, 0, 0, 0, 0, 161, 162, 0, 0, 0, 163, 0, 0, 0, 0, 0, 164, 0,
    0, 0, 0, 0, 0, 165, 0, 0, 0, 166, 0, 0, 0, 0, 0, 0, 167, 0, 0, 0, 0, 0, 0, 168, 169, 0, 0, 0, 170, 0, 0, 0,
    0, 0, 171, 0, 0, 0, 0, 0, 0, 172, 0, 0, 0, 173, 0, 0, 0, 0, 0, 0, 174, 0, 0, 175, 0, 0, 0, 0, 176, 0, 0, 0,
    0, 0, 0, 177, 0, 0, 0, 178, 0, 0, 0, 0, 0, 0, 179, 0, 0, 0, 0, 180, 0, 0, 0, 0, 181, 0, 0, 0, 0, 182, 0, 0,
    0, 0, 183, 0, 0, 0, 0, 0, 184, 0, 0, 0, 0, 0, 0, 185, 0, 0, 0, 186, 0, 0, 0, 0, 0, 0, 187, 0, 0, 0, 0, 188,
    0, 0, 0, 0, 189, 0, 0, 0, 0, 190, 0, 0, 0, 0, 191, 0, 0, 0, 0, 0, 192, 0, 0, 0, 0, 0, 0, 193, 0, 0, 0, 0,
    0, 0, 194, 0, 0, 0, 195, 0, 196, 0, 0, 197, 0, 0, 0, 198, 0, 0, 0, 199, 0, 0, 0, 200, 0, 201, 0, 202, 0, 0, 203, 0,
    204, 0, 0, 0, 205, 0, 0, 0, 0, 0, 0, 0, 0, 206, 0, 0, 0, 207, 0, 0, 0, 0, 0, 0, 208, 0, 0, 0, 209, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 210, 0, 0, 211, 0, 0, 0, 212, 0, 0, 213,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 214, 0, 215, 0, 0, 0, 216, 0, 0, 217, 0, 218, 0, 219, 0, 0, 0, 0, 0,
    0, 220, 0, 0, 0, 0, 0, 0, 221, 0, 0, 0, 222, 0, 223, 0, 0, 224, 0, 0, 0, 225, 0, 0, 0, 226, 0, 0, 0, 227, 0, 228,
    0, 229, 0, 0, 230, 0, 231, 0, 0, 0, 232, 0, 0, 0, 0, 0, 0, 0, 0, 233, 0, 0, 0, 234, 0, 0, 0, 0, 0, 0, 235, 0,
    0, 0, 236, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 237, 0, 238, 0, 0, 0,
    239, 0, 0, 240, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 241, 242, 0, 0, 0, 243, 0, 0, 244, 0, 245, 0, 246, 0, 0,
    0, 247, 0, 0, 0, 0, 0, 0, 248, 0, 0, 0, 0, 0, 249, 0, 250, 0, 0, 251, 0, 0, 0, 0, 0, 0, 252, 0, 0, 0, 0, 0,
    253, 0, 0, 0, 0, 254, 0, 0, 0, 0, 0, 0, 255, 0, 0, 256, 0, 0, 257, 0, 0, 0, 0, 0, 258, 0, 0, 0, 259, 0, 0, 0,
    260, 0, 0, 0, 0, 0, 0, 261, 0, 0, 0, 0, 0, 0, 262, 0, 0, 0, 0, 0, 0, 263, 0, 264, 0, 0, 265, 0, 0, 0, 266, 0,
    0, 0, 267, 0, 0, 0, 0, 268, 0, 0, 0, 269, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 270, 0, 0, 0, 271, 272, 0, 0, 0, 0, 0, 273, 0, 0, 0, 0, 0, 274, 0, 0, 0, 275, 0, 0, 0, 276, 0, 0, 0, 277,
    0, 0, 0, 0, 0, 0, 0, 278, 0, 279, 0, 0, 0, 280, 0, 281, 0, 0, 0, 0, 0, 0, 282, 0, 0, 0, 0, 0, 0, 283, 0, 284,
    0, 0, 285, 0, 0, 0, 286, 0, 0, 0, 287, 0, 0, 0, 0, 288, 289, 0, 290, 0, 0, 0, 0, 0, 0, 291, 0, 0, 292, 0, 0, 293,
    0, 0, 0, 0, 294, 0, 0, 0, 0, 0, 0, 295, 0, 0, 0, 0, 0, 0, 296, 0, 0, 0, 0, 297, 298, 0, 0, 0, 0, 0, 0, 299,
    0, 0, 0, 0, 0, 0, 300, 0, 0, 301, 0, 0, 0, 0, 0, 302, 0, 0, 0, 0, 0, 0, 303, 0, 0, 0, 0, 0, 0, 304, 0, 0,
    305, 0, 0, 0, 0, 306, 0, 0, 0, 0, 0, 307, 0, 308, 0, 0, 0, 0, 0, 0, 0, 0, 0, 309, 0, 0, 0, 0, 310, 0, 311, 312,
    0, 0, 0, 313, 0, 314, 0, 315, 0, 316, 0, 317, 0, 0, 318, 0, 319, 0, 0, 0, 0, 0, 0, 320, 0, 0, 0, 0, 0, 321, 0, 0,
    0, 322, 0, 323, 324, 0, 0, 0, 0, 0, 0, 0, 0, 0, 325, 0, 0, 326, 0, 0, 0, 0, 0, 0, 327, 0, 0, 0, 0, 0, 0, 328,
    0, 0, 329, 0, 0, 0, 0, 0, 0, 330, 0, 331, 0, 332, 0, 333, 0, 0, 334, 0, 0, 0, 335, 0, 336, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 337, 0, 0, 0, 338, 0, 0, 339, 0, 0, 0, 0, 0, 0, 340, 0, 341, 0, 0, 0,
    0, 0, 0, 342, 343, 0, 0, 0, 0, 0, 0, 344, 0, 0, 0, 0, 0, 0, 345, 0, 0, 0, 0, 346, 0, 0, 0, 347, 0, 0, 0, 348,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 349, 0, 0, 0, 0, 0, 350, 0, 0, 0, 351, 0, 352, 0, 0, 353, 0, 354, 0, 0,
    0, 0, 0, 0, 0, 355, 0, 0, 0, 0, 0, 356, 0, 0, 0, 0, 0, 357, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 358, 0, 0, 0, 0, 0, 0, 359, 0, 360, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 361, 0, 0, 0,
    0, 0, 0, 362, 0, 363, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 364, 0, 0, 0, 0, 0, 365, 0, 0, 0, 366, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 367, 0, 0, 0, 0, 0, 0, 368, 0, 0, 0, 369, 0, 370, 0, 0, 0, 371, 0, 0, 0,
    0, 372, 0, 0, 0, 0, 0, 0, 373, 0, 374, 0, 0, 0, 0, 0, 0, 375, 376, 0, 377, 0, 0, 0, 0, 0, 0, 0, 378, 0, 0, 0,
    0, 0, 0, 379, 0, 0, 0, 0, 380, 0, 0, 0, 381, 0, 0, 0, 382, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 383, 0, 0,
    0, 0, 0, 384, 0, 0, 0, 385, 0, 386, 387, 0, 388, 0, 0, 389, 0, 390, 0, 0, 0, 0, 0, 0, 0, 391, 0, 0, 0, 0, 0, 392,
    0, 0, 0, 0, 0, 393, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 394, 0, 0, 0, 0, 0, 0, 395,
    0, 396, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 397, 0, 0, 0, 0, 0, 0, 398, 0, 0, 0, 399, 0, 0, 0, 400,
    0, 0, 0, 401, 0, 0, 0, 0, 0, 0, 0, 0, 402, 0, 403, 0, 0, 404, 0, 405, 0, 0, 406, 407, 0, 0, 408, 0, 409, 0, 0, 0,
    0, 410, 0, 0, 411, 0, 0, 0, 0, 0, 0, 0, 412, 0, 413, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 414, 415, 0, 0, 0, 416,
    0, 417, 0, 0, 0, 0, 418, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 419, 0, 0, 0, 0, 0, 420, 0, 0, 0, 421, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 422, 0, 0, 0, 0, 0, 0, 423, 0, 0, 0, 424, 0, 425, 0, 0, 0, 426, 0, 0,
    0, 427, 0, 428, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 429, 0, 0, 0, 0, 0, 0, 0, 0, 0, 430, 0, 0, 0, 431, 0,
    0, 0, 432, 0, 433, 0, 434, 0, 0, 0, 0, 0, 0, 435, 0, 0, 0, 0, 436, 0, 0, 437, 0, 0, 0, 0, 0, 0, 438, 0, 0, 0,
    0, 439, 0, 0, 440, 0, 0, 0, 441, 0, 0, 0, 442, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 443, 0, 0, 0, 0, 0, 0, 444, 0, 0, 445, 0, 0, 0, 0, 0, 446, 0, 0, 0,
    447, 448, 0, 0, 449, 0, 450, 0, 451, 0, 452, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 453, 0, 0, 454, 0, 0, 0, 0, 0, 0, 0,
    455, 0, 0, 0, 0, 0, 456, 0, 0, 0, 457, 0, 0, 0, 0, 0, 458, 0, 0, 0, 0, 0, 459, 0, 460, 0, 0, 0, 0, 461, 462, 0,
    0, 0, 0, 463, 464, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 465, 0, 0, 0, 0, 0, 466, 0, 0, 0, 467, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 468, 0, 0, 0, 0, 0, 469, 0, 0, 470, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 471, 0, 0, 0, 0, 0, 0, 472, 0, 0, 0, 473, 0, 474, 0, 0, 0, 0, 0, 475,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 476, 0, 0, 0, 0, 0,
    0, 0, 477, 0, 0, 0, 0, 0, 478, 0, 479, 0, 0, 0, 0, 480, 481, 0, 0, 0, 0, 482, 483, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 484, 0, 0, 0, 0, 0, 485, 0, 0, 0, 486, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 487, 0, 0, 0, 0, 0,
    488, 0, 0, 489, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 490, 0,
    0, 0, 0, 0, 0, 491, 0, 492, 0, 0, 0, 0, 493, 0, 494, 0, 0, 0, 0, 0, 0, 0, 495, 0, 0, 0, 496, 0, 0, 0, 0, 0,
    497, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 498, 0, 499, 0, 0, 0, 0, 0, 500, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 501, 0, 0, 0, 0, 0, 0, 0, 502, 0, 0, 0, 0, 0, 503,
    0, 504, 0, 0, 0, 0, 505, 506, 0, 0, 0, 0, 507, 508, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 509, 0, 0, 0, 0, 0,
    510, 0, 0, 0, 511, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 512, 0, 0, 0, 0, 0, 513, 0, 0, 514, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 515, 0, 0, 0, 0, 0, 0, 516, 0, 517, 0,
    0, 0, 0, 518, 0, 519, 0, 0, 0, 0, 0, 0, 0, 520, 0, 0, 0, 521, 0, 0, 0, 0, 0, 522, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 523, 0, 524, 0, 0, 0, 525, 0, 0, 0, 0, 0, 0, 0, 0, 0, 526, 0, 0, 0, 0, 0, 527, 0, 528, 0, 0, 0, 0,
    529, 530, 0, 0, 0, 0, 531, 532, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 533, 0, 0, 0, 0, 0, 534, 0, 0, 535, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 536, 0, 0, 0, 0, 0, 0, 537, 0, 538, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 539, 0, 540, 0, 0, 0, 0, 541, 0, 0, 0, 0, 0, 0, 0, 542, 0, 543, 0, 0, 0, 0, 544, 0, 545, 0,
    0, 0, 0, 0, 0, 0, 546, 547, 0, 0, 0, 548, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 549, 0, 0, 0, 0, 0, 0, 550, 0,
    551, 0, 552, 0, 0, 0, 553, 0, 554, 0, 0, 0, 0, 0, 0, 555, 0, 0, 0, 556, 0, 557, 0, 0, 0, 558, 0, 0, 0, 0, 0, 0,
    559, 0, 560, 0, 561, 0, 0, 0, 562, 563, 0, 0, 0, 0, 564, 0, 565, 0, 0, 0, 0, 0, 0, 566, 0, 0, 0, 0, 0, 567, 0, 0,
    0, 0, 0, 0, 568, 0, 0, 569, 0, 0, 0, 570, 0, 0, 0, 571, 0, 572, 0, 0, 0, 573, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    574, 0, 0, 0, 575, 0, 0, 0, 0, 0, 0, 576, 0, 0, 0, 577, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 578, 0, 0, 579, 0,
    580, 0, 0, 0, 0, 0, 581, 0, 0, 0, 582, 0, 583, 0, 0, 584, 0, 0, 585, 0, 0, 0, 586, 0, 0, 0, 0, 587, 0, 0, 0, 588,
};
void recomp_unit_0132_entry(Runtime &rt, AllegrexContext &ctx, std::uint16_t direct_entry_id, GuestMemory::AotFastView &aot_mem) {
    std::uint32_t jump_target = 0u;
    std::uint32_t local_transfers = 0u;
    std::uint32_t local_pc = ctx.pc;
    std::uint32_t entry_id = direct_entry_id;
LOCAL_DISPATCH:
    {
    if (entry_id == 0u) {
        const std::uint32_t entry_delta = local_pc - 0x08A14000u;
        entry_id = (entry_delta < 16384u && (entry_delta & 3u) == 0u) ? kEntryIds_recomp_unit_0132[entry_delta >> 2u] : 0u;
    }
    switch (entry_id) {
    case 1u: goto L_08A14000;
    case 2u: goto L_08A14010;
    case 3u: goto L_08A14020;
    case 4u: goto L_08A14038;
    case 5u: goto L_08A14050;
    case 6u: goto L_08A1405C;
    case 7u: goto L_08A14068;
    case 8u: goto L_08A1408C;
    case 9u: goto L_08A140A0;
    case 10u: goto L_08A140B0;
    case 11u: goto L_08A140B8;
    case 12u: goto L_08A140C0;
    case 13u: goto L_08A14100;
    case 14u: goto L_08A1414C;
    case 15u: goto L_08A14154;
    case 16u: goto L_08A1415C;
    case 17u: goto L_08A14160;
    case 18u: goto L_08A14178;
    case 19u: goto L_08A141C0;
    case 20u: goto L_08A1422C;
    case 21u: goto L_08A14234;
    case 22u: goto L_08A1423C;
    case 23u: goto L_08A14330;
    case 24u: goto L_08A143E8;
    case 25u: goto L_08A143F4;
    case 26u: goto L_08A14410;
    case 27u: goto L_08A14420;
    case 28u: goto L_08A14430;
    case 29u: goto L_08A14444;
    case 30u: goto L_08A14458;
    case 31u: goto L_08A14470;
    case 32u: goto L_08A14480;
    case 33u: goto L_08A14498;
    case 34u: goto L_08A144B8;
    case 35u: goto L_08A144CC;
    case 36u: goto L_08A144D8;
    case 37u: goto L_08A144F0;
    case 38u: goto L_08A14508;
    case 39u: goto L_08A14520;
    case 40u: goto L_08A14544;
    case 41u: goto L_08A1455C;
    case 42u: goto L_08A14568;
    case 43u: goto L_08A14580;
    case 44u: goto L_08A14598;
    case 45u: goto L_08A145BC;
    case 46u: goto L_08A145D8;
    case 47u: goto L_08A145E4;
    case 48u: goto L_08A145FC;
    case 49u: goto L_08A14614;
    case 50u: goto L_08A14638;
    case 51u: goto L_08A14654;
    case 52u: goto L_08A14660;
    case 53u: goto L_08A14678;
    case 54u: goto L_08A14690;
    case 55u: goto L_08A146B4;
    case 56u: goto L_08A146D0;
    case 57u: goto L_08A146EC;
    case 58u: goto L_08A146FC;
    case 59u: goto L_08A1470C;
    case 60u: goto L_08A14728;
    case 61u: goto L_08A14758;
    case 62u: goto L_08A14778;
    case 63u: goto L_08A147D4;
    case 64u: goto L_08A147EC;
    case 65u: goto L_08A1481C;
    case 66u: goto L_08A14838;
    case 67u: goto L_08A14848;
    case 68u: goto L_08A14854;
    case 69u: goto L_08A1485C;
    case 70u: goto L_08A1488C;
    case 71u: goto L_08A148B8;
    case 72u: goto L_08A148C4;
    case 73u: goto L_08A14920;
    case 74u: goto L_08A14928;
    case 75u: goto L_08A14930;
    case 76u: goto L_08A14954;
    case 77u: goto L_08A149A4;
    case 78u: goto L_08A149BC;
    case 79u: goto L_08A14AF4;
    case 80u: goto L_08A14B18;
    case 81u: goto L_08A14B24;
    case 82u: goto L_08A14B58;
    case 83u: goto L_08A14B88;
    case 84u: goto L_08A14B8C;
    case 85u: goto L_08A14BA0;
    case 86u: goto L_08A14BA4;
    case 87u: goto L_08A14BEC;
    case 88u: goto L_08A14C00;
    case 89u: goto L_08A14C08;
    case 90u: goto L_08A14C10;
    case 91u: goto L_08A14C28;
    case 92u: goto L_08A14C44;
    case 93u: goto L_08A14C60;
    case 94u: goto L_08A14C68;
    case 95u: goto L_08A14C70;
    case 96u: goto L_08A14C90;
    case 97u: goto L_08A14C98;
    case 98u: goto L_08A14E10;
    case 99u: goto L_08A14E6C;
    case 100u: goto L_08A14E90;
    case 101u: goto L_08A14E98;
    case 102u: goto L_08A14EA4;
    case 103u: goto L_08A14F20;
    case 104u: goto L_08A14F50;
    case 105u: goto L_08A14F54;
    case 106u: goto L_08A14F64;
    case 107u: goto L_08A14F68;
    case 108u: goto L_08A14FB0;
    case 109u: goto L_08A14FC4;
    case 110u: goto L_08A14FE4;
    case 111u: goto L_08A14FEC;
    case 112u: goto L_08A14FF4;
    case 113u: goto L_08A1500C;
    case 114u: goto L_08A15028;
    case 115u: goto L_08A15208;
    case 116u: goto L_08A1522C;
    case 117u: goto L_08A15234;
    case 118u: goto L_08A15240;
    case 119u: goto L_08A152BC;
    case 120u: goto L_08A152EC;
    case 121u: goto L_08A152F0;
    case 122u: goto L_08A15300;
    case 123u: goto L_08A15304;
    case 124u: goto L_08A1534C;
    case 125u: goto L_08A1536C;
    case 126u: goto L_08A15380;
    case 127u: goto L_08A15388;
    case 128u: goto L_08A15390;
    case 129u: goto L_08A1561C;
    case 130u: goto L_08A15640;
    case 131u: goto L_08A1564C;
    case 132u: goto L_08A15700;
    case 133u: goto L_08A15730;
    case 134u: goto L_08A15734;
    case 135u: goto L_08A15744;
    case 136u: goto L_08A15748;
    case 137u: goto L_08A15790;
    case 138u: goto L_08A157B0;
    case 139u: goto L_08A157C4;
    case 140u: goto L_08A157E4;
    case 141u: goto L_08A157EC;
    case 142u: goto L_08A157F4;
    case 143u: goto L_08A15824;
    case 144u: goto L_08A158D4;
    case 145u: goto L_08A1591C;
    case 146u: goto L_08A15934;
    case 147u: goto L_08A15944;
    case 148u: goto L_08A159AC;
    case 149u: goto L_08A159BC;
    case 150u: goto L_08A15A1C;
    case 151u: goto L_08A15A30;
    case 152u: goto L_08A15A38;
    case 153u: goto L_08A15A98;
    case 154u: goto L_08A15AA0;
    case 155u: goto L_08A15AAC;
    case 156u: goto L_08A15AE0;
    case 157u: goto L_08A15AE8;
    case 158u: goto L_08A15B04;
    case 159u: goto L_08A15B14;
    case 160u: goto L_08A15B30;
    case 161u: goto L_08A15B4C;
    case 162u: goto L_08A15B50;
    case 163u: goto L_08A15B60;
    case 164u: goto L_08A15B78;
    case 165u: goto L_08A15B94;
    case 166u: goto L_08A15BA4;
    case 167u: goto L_08A15BC0;
    case 168u: goto L_08A15BDC;
    case 169u: goto L_08A15BE0;
    case 170u: goto L_08A15BF0;
    case 171u: goto L_08A15C08;
    case 172u: goto L_08A15C24;
    case 173u: goto L_08A15C34;
    case 174u: goto L_08A15C50;
    case 175u: goto L_08A15C5C;
    case 176u: goto L_08A15C70;
    case 177u: goto L_08A15C8C;
    case 178u: goto L_08A15C9C;
    case 179u: goto L_08A15CB8;
    case 180u: goto L_08A15CCC;
    case 181u: goto L_08A15CE0;
    case 182u: goto L_08A15CF4;
    case 183u: goto L_08A15D08;
    case 184u: goto L_08A15D20;
    case 185u: goto L_08A15D3C;
    case 186u: goto L_08A15D4C;
    case 187u: goto L_08A15D68;
    case 188u: goto L_08A15D7C;
    case 189u: goto L_08A15D90;
    case 190u: goto L_08A15DA4;
    case 191u: goto L_08A15DB8;
    case 192u: goto L_08A15DD0;
    case 193u: goto L_08A15DEC;
    case 194u: goto L_08A15E08;
    case 195u: goto L_08A15E18;
    case 196u: goto L_08A15E20;
    case 197u: goto L_08A15E2C;
    case 198u: goto L_08A15E3C;
    case 199u: goto L_08A15E4C;
    case 200u: goto L_08A15E5C;
    case 201u: goto L_08A15E64;
    case 202u: goto L_08A15E6C;
    case 203u: goto L_08A15E78;
    case 204u: goto L_08A15E80;
    case 205u: goto L_08A15E90;
    case 206u: goto L_08A15EB4;
    case 207u: goto L_08A15EC4;
    case 208u: goto L_08A15EE0;
    case 209u: goto L_08A15EF0;
    case 210u: goto L_08A15F54;
    case 211u: goto L_08A15F60;
    case 212u: goto L_08A15F70;
    case 213u: goto L_08A15F7C;
    case 214u: goto L_08A15FB4;
    case 215u: goto L_08A15FBC;
    case 216u: goto L_08A15FCC;
    case 217u: goto L_08A15FD8;
    case 218u: goto L_08A15FE0;
    case 219u: goto L_08A15FE8;
    case 220u: goto L_08A16004;
    case 221u: goto L_08A16020;
    case 222u: goto L_08A16030;
    case 223u: goto L_08A16038;
    case 224u: goto L_08A16044;
    case 225u: goto L_08A16054;
    case 226u: goto L_08A16064;
    case 227u: goto L_08A16074;
    case 228u: goto L_08A1607C;
    case 229u: goto L_08A16084;
    case 230u: goto L_08A16090;
    case 231u: goto L_08A16098;
    case 232u: goto L_08A160A8;
    case 233u: goto L_08A160CC;
    case 234u: goto L_08A160DC;
    case 235u: goto L_08A160F8;
    case 236u: goto L_08A16108;
    case 237u: goto L_08A16168;
    case 238u: goto L_08A16170;
    case 239u: goto L_08A16180;
    case 240u: goto L_08A1618C;
    case 241u: goto L_08A161C4;
    case 242u: goto L_08A161C8;
    case 243u: goto L_08A161D8;
    case 244u: goto L_08A161E4;
    case 245u: goto L_08A161EC;
    case 246u: goto L_08A161F4;
    case 247u: goto L_08A16204;
    case 248u: goto L_08A16220;
    case 249u: goto L_08A16238;
    case 250u: goto L_08A16240;
    case 251u: goto L_08A1624C;
    case 252u: goto L_08A16268;
    case 253u: goto L_08A16280;
    case 254u: goto L_08A16294;
    case 255u: goto L_08A162B0;
    case 256u: goto L_08A162BC;
    case 257u: goto L_08A162C8;
    case 258u: goto L_08A162E0;
    case 259u: goto L_08A162F0;
    case 260u: goto L_08A16300;
    case 261u: goto L_08A1631C;
    case 262u: goto L_08A16338;
    case 263u: goto L_08A16354;
    case 264u: goto L_08A1635C;
    case 265u: goto L_08A16368;
    case 266u: goto L_08A16378;
    case 267u: goto L_08A16388;
    case 268u: goto L_08A1639C;
    case 269u: goto L_08A163AC;
    case 270u: goto L_08A16488;
    case 271u: goto L_08A16498;
    case 272u: goto L_08A1649C;
    case 273u: goto L_08A164B4;
    case 274u: goto L_08A164CC;
    case 275u: goto L_08A164DC;
    case 276u: goto L_08A164EC;
    case 277u: goto L_08A164FC;
    case 278u: goto L_08A1651C;
    case 279u: goto L_08A16524;
    case 280u: goto L_08A16534;
    case 281u: goto L_08A1653C;
    case 282u: goto L_08A16558;
    case 283u: goto L_08A16574;
    case 284u: goto L_08A1657C;
    case 285u: goto L_08A16588;
    case 286u: goto L_08A16598;
    case 287u: goto L_08A165A8;
    case 288u: goto L_08A165BC;
    case 289u: goto L_08A165C0;
    case 290u: goto L_08A165C8;
    case 291u: goto L_08A165E4;
    case 292u: goto L_08A165F0;
    case 293u: goto L_08A165FC;
    case 294u: goto L_08A16610;
    case 295u: goto L_08A1662C;
    case 296u: goto L_08A16648;
    case 297u: goto L_08A1665C;
    case 298u: goto L_08A16660;
    case 299u: goto L_08A1667C;
    case 300u: goto L_08A16698;
    case 301u: goto L_08A166A4;
    case 302u: goto L_08A166BC;
    case 303u: goto L_08A166D8;
    case 304u: goto L_08A166F4;
    case 305u: goto L_08A16700;
    case 306u: goto L_08A16714;
    case 307u: goto L_08A1672C;
    case 308u: goto L_08A16734;
    case 309u: goto L_08A1675C;
    case 310u: goto L_08A16770;
    case 311u: goto L_08A16778;
    case 312u: goto L_08A1677C;
    case 313u: goto L_08A1678C;
    case 314u: goto L_08A16794;
    case 315u: goto L_08A1679C;
    case 316u: goto L_08A167A4;
    case 317u: goto L_08A167AC;
    case 318u: goto L_08A167B8;
    case 319u: goto L_08A167C0;
    case 320u: goto L_08A167DC;
    case 321u: goto L_08A167F4;
    case 322u: goto L_08A16804;
    case 323u: goto L_08A1680C;
    case 324u: goto L_08A16810;
    case 325u: goto L_08A16838;
    case 326u: goto L_08A16844;
    case 327u: goto L_08A16860;
    case 328u: goto L_08A1687C;
    case 329u: goto L_08A16888;
    case 330u: goto L_08A168A4;
    case 331u: goto L_08A168AC;
    case 332u: goto L_08A168B4;
    case 333u: goto L_08A168BC;
    case 334u: goto L_08A168C8;
    case 335u: goto L_08A168D8;
    case 336u: goto L_08A168E0;
    case 337u: goto L_08A16930;
    case 338u: goto L_08A16940;
    case 339u: goto L_08A1694C;
    case 340u: goto L_08A16968;
    case 341u: goto L_08A16970;
    case 342u: goto L_08A1698C;
    case 343u: goto L_08A16990;
    case 344u: goto L_08A169AC;
    case 345u: goto L_08A169C8;
    case 346u: goto L_08A169DC;
    case 347u: goto L_08A169EC;
    case 348u: goto L_08A169FC;
    case 349u: goto L_08A16A30;
    case 350u: goto L_08A16A48;
    case 351u: goto L_08A16A58;
    case 352u: goto L_08A16A60;
    case 353u: goto L_08A16A6C;
    case 354u: goto L_08A16A74;
    case 355u: goto L_08A16A94;
    case 356u: goto L_08A16AAC;
    case 357u: goto L_08A16AC4;
    case 358u: goto L_08A16B10;
    case 359u: goto L_08A16B2C;
    case 360u: goto L_08A16B34;
    case 361u: goto L_08A16B70;
    case 362u: goto L_08A16B8C;
    case 363u: goto L_08A16B94;
    case 364u: goto L_08A16BC8;
    case 365u: goto L_08A16BE0;
    case 366u: goto L_08A16BF0;
    case 367u: goto L_08A16C2C;
    case 368u: goto L_08A16C48;
    case 369u: goto L_08A16C58;
    case 370u: goto L_08A16C60;
    case 371u: goto L_08A16C70;
    case 372u: goto L_08A16C84;
    case 373u: goto L_08A16CA0;
    case 374u: goto L_08A16CA8;
    case 375u: goto L_08A16CC4;
    case 376u: goto L_08A16CC8;
    case 377u: goto L_08A16CD0;
    case 378u: goto L_08A16CF0;
    case 379u: goto L_08A16D0C;
    case 380u: goto L_08A16D20;
    case 381u: goto L_08A16D30;
    case 382u: goto L_08A16D40;
    case 383u: goto L_08A16D74;
    case 384u: goto L_08A16D8C;
    case 385u: goto L_08A16D9C;
    case 386u: goto L_08A16DA4;
    case 387u: goto L_08A16DA8;
    case 388u: goto L_08A16DB0;
    case 389u: goto L_08A16DBC;
    case 390u: goto L_08A16DC4;
    case 391u: goto L_08A16DE4;
    case 392u: goto L_08A16DFC;
    case 393u: goto L_08A16E14;
    case 394u: goto L_08A16E60;
    case 395u: goto L_08A16E7C;
    case 396u: goto L_08A16E84;
    case 397u: goto L_08A16EC0;
    case 398u: goto L_08A16EDC;
    case 399u: goto L_08A16EEC;
    case 400u: goto L_08A16EFC;
    case 401u: goto L_08A16F0C;
    case 402u: goto L_08A16F30;
    case 403u: goto L_08A16F38;
    case 404u: goto L_08A16F44;
    case 405u: goto L_08A16F4C;
    case 406u: goto L_08A16F58;
    case 407u: goto L_08A16F5C;
    case 408u: goto L_08A16F68;
    case 409u: goto L_08A16F70;
    case 410u: goto L_08A16F84;
    case 411u: goto L_08A16F90;
    case 412u: goto L_08A16FB0;
    case 413u: goto L_08A16FB8;
    case 414u: goto L_08A16FE8;
    case 415u: goto L_08A16FEC;
    case 416u: goto L_08A16FFC;
    case 417u: goto L_08A17004;
    case 418u: goto L_08A17018;
    case 419u: goto L_08A1704C;
    case 420u: goto L_08A17064;
    case 421u: goto L_08A17074;
    case 422u: goto L_08A170B0;
    case 423u: goto L_08A170CC;
    case 424u: goto L_08A170DC;
    case 425u: goto L_08A170E4;
    case 426u: goto L_08A170F4;
    case 427u: goto L_08A17104;
    case 428u: goto L_08A1710C;
    case 429u: goto L_08A17140;
    case 430u: goto L_08A17168;
    case 431u: goto L_08A17178;
    case 432u: goto L_08A17188;
    case 433u: goto L_08A17190;
    case 434u: goto L_08A17198;
    case 435u: goto L_08A171B4;
    case 436u: goto L_08A171C8;
    case 437u: goto L_08A171D4;
    case 438u: goto L_08A171F0;
    case 439u: goto L_08A17204;
    case 440u: goto L_08A17210;
    case 441u: goto L_08A17220;
    case 442u: goto L_08A17230;
    case 443u: goto L_08A172B0;
    case 444u: goto L_08A172CC;
    case 445u: goto L_08A172D8;
    case 446u: goto L_08A172F0;
    case 447u: goto L_08A17300;
    case 448u: goto L_08A17304;
    case 449u: goto L_08A17310;
    case 450u: goto L_08A17318;
    case 451u: goto L_08A17320;
    case 452u: goto L_08A17328;
    case 453u: goto L_08A17354;
    case 454u: goto L_08A17360;
    case 455u: goto L_08A17380;
    case 456u: goto L_08A17398;
    case 457u: goto L_08A173A8;
    case 458u: goto L_08A173C0;
    case 459u: goto L_08A173D8;
    case 460u: goto L_08A173E0;
    case 461u: goto L_08A173F4;
    case 462u: goto L_08A173F8;
    case 463u: goto L_08A1740C;
    case 464u: goto L_08A17410;
    case 465u: goto L_08A17444;
    case 466u: goto L_08A1745C;
    case 467u: goto L_08A1746C;
    case 468u: goto L_08A174A0;
    case 469u: goto L_08A174B8;
    case 470u: goto L_08A174C4;
    case 471u: goto L_08A17530;
    case 472u: goto L_08A1754C;
    case 473u: goto L_08A1755C;
    case 474u: goto L_08A17564;
    case 475u: goto L_08A1757C;
    case 476u: goto L_08A175E8;
    case 477u: goto L_08A17608;
    case 478u: goto L_08A17620;
    case 479u: goto L_08A17628;
    case 480u: goto L_08A1763C;
    case 481u: goto L_08A17640;
    case 482u: goto L_08A17654;
    case 483u: goto L_08A17658;
    case 484u: goto L_08A1768C;
    case 485u: goto L_08A176A4;
    case 486u: goto L_08A176B4;
    case 487u: goto L_08A176E8;
    case 488u: goto L_08A17700;
    case 489u: goto L_08A1770C;
    case 490u: goto L_08A17778;
    case 491u: goto L_08A17794;
    case 492u: goto L_08A1779C;
    case 493u: goto L_08A177B0;
    case 494u: goto L_08A177B8;
    case 495u: goto L_08A177D8;
    case 496u: goto L_08A177E8;
    case 497u: goto L_08A17800;
    case 498u: goto L_08A17830;
    case 499u: goto L_08A17838;
    case 500u: goto L_08A17850;
    case 501u: goto L_08A178C4;
    case 502u: goto L_08A178E4;
    case 503u: goto L_08A178FC;
    case 504u: goto L_08A17904;
    case 505u: goto L_08A17918;
    case 506u: goto L_08A1791C;
    case 507u: goto L_08A17930;
    case 508u: goto L_08A17934;
    case 509u: goto L_08A17968;
    case 510u: goto L_08A17980;
    case 511u: goto L_08A17990;
    case 512u: goto L_08A179C4;
    case 513u: goto L_08A179DC;
    case 514u: goto L_08A179E8;
    case 515u: goto L_08A17A54;
    case 516u: goto L_08A17A70;
    case 517u: goto L_08A17A78;
    case 518u: goto L_08A17A8C;
    case 519u: goto L_08A17A94;
    case 520u: goto L_08A17AB4;
    case 521u: goto L_08A17AC4;
    case 522u: goto L_08A17ADC;
    case 523u: goto L_08A17B0C;
    case 524u: goto L_08A17B14;
    case 525u: goto L_08A17B24;
    case 526u: goto L_08A17B4C;
    case 527u: goto L_08A17B64;
    case 528u: goto L_08A17B6C;
    case 529u: goto L_08A17B80;
    case 530u: goto L_08A17B84;
    case 531u: goto L_08A17B98;
    case 532u: goto L_08A17B9C;
    case 533u: goto L_08A17BD0;
    case 534u: goto L_08A17BE8;
    case 535u: goto L_08A17BF4;
    case 536u: goto L_08A17C28;
    case 537u: goto L_08A17C44;
    case 538u: goto L_08A17C4C;
    case 539u: goto L_08A17C98;
    case 540u: goto L_08A17CA0;
    case 541u: goto L_08A17CB4;
    case 542u: goto L_08A17CD4;
    case 543u: goto L_08A17CDC;
    case 544u: goto L_08A17CF0;
    case 545u: goto L_08A17CF8;
    case 546u: goto L_08A17D18;
    case 547u: goto L_08A17D1C;
    case 548u: goto L_08A17D2C;
    case 549u: goto L_08A17D5C;
    case 550u: goto L_08A17D78;
    case 551u: goto L_08A17D80;
    case 552u: goto L_08A17D88;
    case 553u: goto L_08A17D98;
    case 554u: goto L_08A17DA0;
    case 555u: goto L_08A17DBC;
    case 556u: goto L_08A17DCC;
    case 557u: goto L_08A17DD4;
    case 558u: goto L_08A17DE4;
    case 559u: goto L_08A17E00;
    case 560u: goto L_08A17E08;
    case 561u: goto L_08A17E10;
    case 562u: goto L_08A17E20;
    case 563u: goto L_08A17E24;
    case 564u: goto L_08A17E38;
    case 565u: goto L_08A17E40;
    case 566u: goto L_08A17E5C;
    case 567u: goto L_08A17E74;
    case 568u: goto L_08A17E90;
    case 569u: goto L_08A17E9C;
    case 570u: goto L_08A17EAC;
    case 571u: goto L_08A17EBC;
    case 572u: goto L_08A17EC4;
    case 573u: goto L_08A17ED4;
    case 574u: goto L_08A17F00;
    case 575u: goto L_08A17F10;
    case 576u: goto L_08A17F2C;
    case 577u: goto L_08A17F3C;
    case 578u: goto L_08A17F6C;
    case 579u: goto L_08A17F78;
    case 580u: goto L_08A17F80;
    case 581u: goto L_08A17F98;
    case 582u: goto L_08A17FA8;
    case 583u: goto L_08A17FB0;
    case 584u: goto L_08A17FBC;
    case 585u: goto L_08A17FC8;
    case 586u: goto L_08A17FD8;
    case 587u: goto L_08A17FEC;
    case 588u: goto L_08A17FFC;
    default:
        if (local_transfers == 0u) rt.unsupported(ctx.pc, 0u, "invalid internal function entry");
        else ctx.pc = local_pc;
        return;
    }
    }
L_08A14000:
    ctx.fpr[20] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    ctx.gpr[31] = (0x08A14010u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 123u, 0x08A0D2A0u>(ctx, &aot_mem) && ctx.pc == 0x08A14010u) goto L_08A14010;
    return;
L_08A14010:
    ctx.gpr[18] = (ctx.gpr[17] + static_cast<std::uint32_t>(128));
    ctx.gpr[4] = (ctx.gpr[18] | 0u);
    ctx.gpr[31] = (0x08A14020u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 123u, 0x08A0D2A0u>(ctx, &aot_mem) && ctx.pc == 0x08A14020u) goto L_08A14020;
    return;
L_08A14020:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(92)));
    ctx.gpr[4] = (ctx.gpr[4] + static_cast<std::uint32_t>(168));
    ctx.gpr[5] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[4] + static_cast<std::uint32_t>(0))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(4)));
    jump_target = ctx.gpr[6];
    ctx.gpr[31] = (0x08A14038u);
    ctx.gpr[4] = (ctx.gpr[17] + ctx.gpr[5]);
    ctx.pc = jump_target;
    if (rt.invoke_chained_call(ctx, &aot_mem) && ctx.pc == 0x08A14038u) goto L_08A14038;
    return;
L_08A14038:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(92)));
    ctx.gpr[5] = (ctx.gpr[4] + static_cast<std::uint32_t>(160));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[5] + static_cast<std::uint32_t>(0))))));
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(4)));
    jump_target = ctx.gpr[5];
    ctx.gpr[31] = (0x08A14050u);
    ctx.gpr[4] = (ctx.gpr[17] + ctx.gpr[4]);
    ctx.pc = jump_target;
    if (rt.invoke_chained_call(ctx, &aot_mem) && ctx.pc == 0x08A14050u) goto L_08A14050;
    return;
L_08A14050:
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    ctx.gpr[31] = (0x08A1405Cu);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 123u, 0x08A0D2A0u>(ctx, &aot_mem) && ctx.pc == 0x08A1405Cu) goto L_08A1405C;
    return;
L_08A1405C:
    ctx.gpr[4] = (ctx.gpr[18] | 0u);
    ctx.gpr[31] = (0x08A14068u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 123u, 0x08A0D2A0u>(ctx, &aot_mem) && ctx.pc == 0x08A14068u) goto L_08A14068;
    return;
L_08A14068:
    ctx.gpr[4] = (0u | 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 1u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(68)));
    ctx.gpr[6] = (65535u << 16u);
    ctx.gpr[6] = (ctx.gpr[6] + static_cast<std::uint32_t>(32767));
    ctx.gpr[5] = (ctx.gpr[5] & ctx.gpr[6]);
    ctx.gpr[4] = (ctx.gpr[4] << 15u);
    ctx.gpr[4] = (ctx.gpr[5] | ctx.gpr[4]);
    aot_mem.aot_store32(ctx.gpr[17] + static_cast<std::uint32_t>(68), ctx.gpr[4]);
    goto L_08A1408C;
L_08A1408C:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(112)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(116)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(120)));
    ctx.gpr[31] = (0x08A140A0u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 278u, 0x08A0E1E4u>(ctx, &aot_mem) && ctx.pc == 0x08A140A0u) goto L_08A140A0;
    return;
L_08A140A0:
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(112));
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(48));
    ctx.gpr[31] = (0x08A140B0u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0189_entry, 189u, 442u, 0x08AF9F84u>(ctx, &aot_mem) && ctx.pc == 0x08A140B0u) goto L_08A140B0;
    return;
L_08A140B0:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 1u);
      if (branch_taken) {
          goto L_08A140C0;
      }
      goto L_08A140B8;
    }
L_08A140B8:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 0u);
      if (branch_taken) {
          goto L_08A140C0;
      }
      goto L_08A140C0;
    }
L_08A140C0:
    ctx.fpr[20] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2020)));
    ctx.fpr[22] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2024)));
    ctx.fpr[24] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2028)));
    ctx.fpr[26] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2032)));
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2036)));
    ctx.gpr[17] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2040)));
    ctx.gpr[18] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2044)));
    ctx.gpr[19] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2048)));
    ctx.gpr[20] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2052)));
    ctx.gpr[21] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2056)));
    ctx.gpr[22] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2060)));
    ctx.gpr[23] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2064)));
    ctx.gpr[30] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2068)));
    ctx.gpr[31] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(2072)));
    jump_target = ctx.gpr[31];
    ctx.gpr[29] = (ctx.gpr[29] + static_cast<std::uint32_t>(2080));
    local_pc = jump_target;
    if (++local_transfers < 2048u) { entry_id = 0u; goto LOCAL_DISPATCH; }
    ctx.pc = jump_target;
    return;
L_08A14100:
    ctx.gpr[29] = (ctx.gpr[29] + static_cast<std::uint32_t>(-608));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(576), ctx.gpr[16]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(580), ctx.gpr[17]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(584), ctx.gpr[18]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(588), ctx.gpr[19]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(592), ctx.gpr[20]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(596), ctx.gpr[21]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(600), ctx.gpr[31]);
    ctx.gpr[10] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[4] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[11] = (2228u << 16u);
    ctx.gpr[11] = (aot_mem.aot_load32(ctx.gpr[11] + static_cast<std::uint32_t>(-31116)));
    ctx.gpr[11] = (aot_mem.aot_load16(ctx.gpr[11] + static_cast<std::uint32_t>(440)));
    ctx.gpr[21] = (ctx.gpr[4] | 0u);
    ctx.gpr[20] = (ctx.gpr[5] | 0u);
    ctx.gpr[19] = (ctx.gpr[6] | 0u);
    ctx.gpr[18] = (ctx.gpr[7] | 0u);
    ctx.gpr[17] = (ctx.gpr[8] | 0u);
    { const bool branch_taken = ctx.gpr[10] != ctx.gpr[11];
    ctx.gpr[16] = (ctx.gpr[9] | 0u);
      if (branch_taken) {
          goto L_08A14160;
      }
      goto L_08A1414C;
    }
L_08A1414C:
    ctx.gpr[31] = (0x08A14154u);
    // nop
    if (rt.invoke_chained_direct<&recomp_unit_0116_entry, 116u, 198u, 0x089D596Cu>(ctx, &aot_mem) && ctx.pc == 0x08A14154u) goto L_08A14154;
    return;
L_08A14154:
    { const bool branch_taken = ctx.gpr[20] == ctx.gpr[2];
    // nop
      if (branch_taken) {
          goto L_08A14160;
      }
      goto L_08A1415C;
    }
L_08A1415C:
    aot_mem.aot_store8(ctx.gpr[21] + static_cast<std::uint32_t>(435), static_cast<std::uint8_t>(0u));
    goto L_08A14160;
L_08A14160:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 8192u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A1423C;
      }
      goto L_08A14178;
    }
L_08A14178:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(16), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(20), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(24), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[4] = (ctx.gpr[21] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(16));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[13] = std::bit_cast<float>(0u);
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[13]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14234;
      }
      goto L_08A141C0;
    }
L_08A141C0:
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(208)));
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]) ^ 0x80000000u);
    aot_mem.aot_store32(ctx.gpr[18] + static_cast<std::uint32_t>(0), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(48), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(52), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(56), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(48));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(32));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(208)));
    ctx.gpr[5] = (std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[5]);
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / vfpu_s[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(64));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[31] = (0x08A1422Cu);
    ctx.gpr[4] = (ctx.gpr[21] + static_cast<std::uint32_t>(112));
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 565u, 0x0892753Cu>(ctx, &aot_mem) && ctx.pc == 0x08A1422Cu) goto L_08A1422C;
    return;
L_08A1422C:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 1u);
      if (branch_taken) {
          goto L_08A14930;
      }
      goto L_08A14234;
    }
L_08A14234:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 0u);
      if (branch_taken) {
          goto L_08A14930;
      }
      goto L_08A1423C;
    }
L_08A1423C:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(0)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(80), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(4)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(84), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(8)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(88), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(80));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[21] + static_cast<std::uint32_t>(48));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(96));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(96)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(100)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(104)));
    ctx.gpr[4] = (ctx.gpr[21] + static_cast<std::uint32_t>(128));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[21] + static_cast<std::uint32_t>(160));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(480));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(496), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(500), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(504), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(496));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(464));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[21] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(448));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[21] + static_cast<std::uint32_t>(144));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(112));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(128), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(132), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(136), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(128));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[14] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[12] = std::bit_cast<float>(0u);
    ctx.set_fpu_condition((ctx.fpr[14] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14928;
      }
      goto L_08A14330;
    }
L_08A14330:
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<4u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(16);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<5u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(32);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<6u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[21] + static_cast<std::uint32_t>(240));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<7u, 4u>(vfpu_value); }
    { float vfpu_matrix[16]{}, vfpu_target_raw[4]{}, vfpu_target[4]{}, vfpu_result[4]{};
      ctx.read_vfpu_matrix(vfpu_matrix, 36u, 3u);
      ctx.read_vfpu_vector_ct<7u, 3u>(vfpu_target_raw);
      constexpr std::uint32_t vfpu_side = 3u;
      constexpr std::uint32_t vfpu_input_length = 3u;
      for (std::uint32_t i = 0; i < 4u; ++i) vfpu_target[i] = i < vfpu_input_length ? vfpu_target_raw[i] : 0.0f;
      if (vfpu_side - 1u >= vfpu_input_length) vfpu_target[vfpu_side - 1u] = 1.0f;
      for (std::uint32_t row = 0; row + 1u < vfpu_side; ++row) {
        float sum = 0.0f;
        for (std::uint32_t column = 0; column < vfpu_side; ++column) sum += vfpu_matrix[row * 4u + column] * vfpu_target[column];
        vfpu_result[row] = sum;
      }
      float vfpu_final_row[4]{vfpu_matrix[(vfpu_side - 1u) * 4u + 0u], vfpu_matrix[(vfpu_side - 1u) * 4u + 1u],
                              vfpu_matrix[(vfpu_side - 1u) * 4u + 2u], vfpu_matrix[(vfpu_side - 1u) * 4u + 3u]};
      ctx.apply_vfpu_source_prefix_ct<4u, 0u>(vfpu_final_row);
      ctx.apply_vfpu_source_prefix_ct<4u, 1u>(vfpu_target);
      for (std::uint32_t column = 0; column < 4u; ++column) vfpu_result[vfpu_side - 1u] += vfpu_final_row[column] * vfpu_target[column];
      const std::uint32_t vfpu_destination_prefix = ctx.vfpu_ctrl[2];
      const std::uint32_t vfpu_last_lane = vfpu_side - 1u;
      ctx.vfpu_ctrl[2] = ((vfpu_destination_prefix & (1u << 8u)) << vfpu_last_lane) |
                         ((vfpu_destination_prefix & 3u) << (vfpu_last_lane * 2u));
      ctx.write_vfpu_vector_with_destination_prefix(vfpu_result, 0u, vfpu_side); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(176));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(208)));
    ctx.gpr[5] = (16256u << 16u);
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[5]);
    ctx.fpr[12] = ctx.fpr[13] / ctx.fpr[12];
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(96));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(160));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(128));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(144));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[16] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(212)));
    ctx.fpr[15] = ctx.fpr[15] / ctx.fpr[16];
    ctx.fpr[12] = ctx.fpr[12] + ctx.fpr[15];
    ctx.fpr[13] = ctx.fpr[13] / ctx.fpr[12];
    ctx.gpr[4] = (2230u << 16u);
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(-8144)));
    ctx.gpr[4] = (15363u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 4719u);
    ctx.fpr[16] = std::bit_cast<float>(ctx.gpr[4]);
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[16]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.gpr[4] = (0u | 0u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & 14u);
    ctx.gpr[5] = (ctx.gpr[5] ^ 8u);
    ctx.gpr[5] = (ctx.gpr[5] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[5] = (ctx.gpr[5] & 255u);
    { const bool branch_taken = ctx.gpr[5] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A143F4;
      }
      goto L_08A143E8;
    }
L_08A143E8:
    ctx.gpr[4] = (0u | 1u);
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A144CC;
      }
      goto L_08A143F4;
    }
L_08A143F4:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & 14u);
    ctx.gpr[5] = (ctx.gpr[5] ^ 4u);
    ctx.gpr[5] = (ctx.gpr[5] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[5] = (ctx.gpr[5] & 255u);
    { const bool branch_taken = ctx.gpr[5] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A144CC;
      }
      goto L_08A14410;
    }
L_08A14410:
    ctx.gpr[5] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[21] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[5] = (ctx.gpr[5] & 16u);
    { const bool branch_taken = ctx.gpr[5] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A144CC;
      }
      goto L_08A14420;
    }
L_08A14420:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(836)));
    ctx.gpr[6] = (0u | 5u);
    { const bool branch_taken = ctx.gpr[5] != ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A14470;
      }
      goto L_08A14430;
    }
L_08A14430:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & 496u);
    ctx.gpr[6] = (0u | 64u);
    { const bool branch_taken = ctx.gpr[5] == ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A14458;
      }
      goto L_08A14444;
    }
L_08A14444:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & 496u);
    ctx.gpr[6] = (0u | 80u);
    { const bool branch_taken = ctx.gpr[5] != ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A14470;
      }
      goto L_08A14458;
    }
L_08A14458:
    ctx.gpr[4] = (0u | 3u);
    ctx.gpr[5] = (16294u << 16u);
    ctx.gpr[5] = (ctx.gpr[5] | 26214u);
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[5]);
    { const bool branch_taken = 0u == 0u;
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[15]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
      if (branch_taken) {
          goto L_08A144CC;
      }
      goto L_08A14470;
    }
L_08A14470:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(836)));
    ctx.gpr[6] = (0u | 1u);
    { const bool branch_taken = ctx.gpr[5] != ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A14498;
      }
      goto L_08A14480;
    }
L_08A14480:
    ctx.gpr[4] = (0u | 4u);
    ctx.gpr[5] = (16281u << 16u);
    ctx.gpr[5] = (ctx.gpr[5] | 39322u);
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[5]);
    { const bool branch_taken = 0u == 0u;
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[15]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
      if (branch_taken) {
          goto L_08A144CC;
      }
      goto L_08A14498;
    }
L_08A14498:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(40)));
    ctx.gpr[5] = (48793u << 16u);
    ctx.gpr[5] = (ctx.gpr[5] | 39322u);
    ctx.fpr[16] = std::bit_cast<float>(ctx.gpr[5]);
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[16]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A144CC;
      }
      goto L_08A144B8;
    }
L_08A144B8:
    ctx.gpr[4] = (0u | 2u);
    ctx.gpr[5] = (16268u << 16u);
    ctx.gpr[5] = (ctx.gpr[5] | 52429u);
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[5]);
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[15]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    goto L_08A144CC;
L_08A144CC:
    ctx.gpr[5] = (0u | 1u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A1455C;
      }
      goto L_08A144D8;
    }
L_08A144D8:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & 4096u);
    ctx.gpr[5] = (0u < ctx.gpr[5] ? 1u : 0u);
    ctx.gpr[5] = (ctx.gpr[5] & 255u);
    { const bool branch_taken = ctx.gpr[5] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A1455C;
      }
      goto L_08A144F0;
    }
L_08A144F0:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(112)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A1455C;
      }
      goto L_08A14508;
    }
L_08A14508:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(116)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A1455C;
      }
      goto L_08A14520;
    }
L_08A14520:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(120)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.gpr[5] = (16384u << 16u);
    ctx.fpr[16] = std::bit_cast<float>(ctx.gpr[5]);
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[16]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[16] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[16] = fs * ft; }
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[16]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A1455C;
      }
      goto L_08A14544;
    }
L_08A14544:
    ctx.gpr[5] = (49018u << 16u);
    ctx.gpr[5] = (ctx.gpr[5] | 57672u);
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[5]);
    { const float fs = ctx.fpr[14]; const float ft = ctx.fpr[15]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[15] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[15] = fs * ft; }
    { const float fs = ctx.fpr[15]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[15] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[15] = fs * ft; }
    aot_mem.aot_store32(ctx.gpr[18] + static_cast<std::uint32_t>(0), std::bit_cast<std::uint32_t>(ctx.fpr[15]));
    goto L_08A1455C;
L_08A1455C:
    ctx.gpr[5] = (0u | 3u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A145D8;
      }
      goto L_08A14568;
    }
L_08A14568:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(112)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A145D8;
      }
      goto L_08A14580;
    }
L_08A14580:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(116)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A145D8;
      }
      goto L_08A14598;
    }
L_08A14598:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(120)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.gpr[5] = (16384u << 16u);
    ctx.fpr[16] = std::bit_cast<float>(ctx.gpr[5]);
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[16]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[16] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[16] = fs * ft; }
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[16]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A145D8;
      }
      goto L_08A145BC;
    }
L_08A145BC:
    ctx.gpr[4] = (48972u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 52429u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    { const float fs = ctx.fpr[14]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[18] + static_cast<std::uint32_t>(0), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
      if (branch_taken) {
          goto L_08A14778;
      }
      goto L_08A145D8;
    }
L_08A145D8:
    ctx.gpr[5] = (0u | 2u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A14654;
      }
      goto L_08A145E4;
    }
L_08A145E4:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(112)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14654;
      }
      goto L_08A145FC;
    }
L_08A145FC:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(116)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14654;
      }
      goto L_08A14614;
    }
L_08A14614:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(120)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.gpr[5] = (16384u << 16u);
    ctx.fpr[16] = std::bit_cast<float>(ctx.gpr[5]);
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[16]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[16] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[16] = fs * ft; }
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[16]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14654;
      }
      goto L_08A14638;
    }
L_08A14638:
    ctx.gpr[4] = (49003u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 34079u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    { const float fs = ctx.fpr[14]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[18] + static_cast<std::uint32_t>(0), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
      if (branch_taken) {
          goto L_08A14778;
      }
      goto L_08A14654;
    }
L_08A14654:
    ctx.gpr[5] = (0u | 4u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A146D0;
      }
      goto L_08A14660;
    }
L_08A14660:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(112)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A146D0;
      }
      goto L_08A14678;
    }
L_08A14678:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(116)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A146D0;
      }
      goto L_08A14690;
    }
L_08A14690:
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(120)));
    ctx.fpr[15] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[15]) & 0x7FFFFFFFu);
    ctx.gpr[4] = (16384u << 16u);
    ctx.fpr[16] = std::bit_cast<float>(ctx.gpr[4]);
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[16]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.set_fpu_condition((ctx.fpr[15] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A146D0;
      }
      goto L_08A146B4;
    }
L_08A146B4:
    ctx.gpr[4] = (48972u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 52429u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    { const float fs = ctx.fpr[14]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[18] + static_cast<std::uint32_t>(0), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
      if (branch_taken) {
          goto L_08A14778;
      }
      goto L_08A146D0;
    }
L_08A146D0:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A14758;
      }
      goto L_08A146EC;
    }
L_08A146EC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(836)));
    ctx.gpr[5] = (0u | 1u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A14758;
      }
      goto L_08A146FC;
    }
L_08A146FC:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[19] + static_cast<std::uint32_t>(30)));
    ctx.gpr[5] = (0u | 22u);
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A14728;
      }
      goto L_08A1470C;
    }
L_08A1470C:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(136)));
    ctx.gpr[4] = (16128u << 16u);
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[15]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14758;
      }
      goto L_08A14728;
    }
L_08A14728:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(224)));
    ctx.gpr[4] = (16384u << 16u);
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[4]);
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[15]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.gpr[4] = (16256u << 16u);
    ctx.fpr[16] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[12] = ctx.fpr[12] + ctx.fpr[16];
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[14]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]) ^ 0x80000000u);
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[18] + static_cast<std::uint32_t>(0), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
      if (branch_taken) {
          goto L_08A14778;
      }
      goto L_08A14758;
    }
L_08A14758:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(224)));
    ctx.gpr[4] = (16256u << 16u);
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[12] = ctx.fpr[12] + ctx.fpr[15];
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[14]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]) ^ 0x80000000u);
    aot_mem.aot_store32(ctx.gpr[18] + static_cast<std::uint32_t>(0), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    goto L_08A14778;
L_08A14778:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(0)));
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(128));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(192));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(208)));
    ctx.gpr[5] = (std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[5]);
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / vfpu_s[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(208));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A148B8;
      }
      goto L_08A147D4;
    }
L_08A147D4:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(72)));
    ctx.gpr[4] = (ctx.gpr[4] & 2u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A1485C;
      }
      goto L_08A147EC;
    }
L_08A147EC:
    ctx.gpr[4] = (ctx.gpr[21] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.gpr[4] = (15820u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 52429u);
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14848;
      }
      goto L_08A1481C;
    }
L_08A1481C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 2u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A1485C;
      }
      goto L_08A14838;
    }
L_08A14838:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] & 4u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A1485C;
      }
      goto L_08A14848;
    }
L_08A14848:
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(208));
    ctx.gpr[31] = (0x08A14854u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 565u, 0x0892753Cu>(ctx, &aot_mem) && ctx.pc == 0x08A14854u) goto L_08A14854;
    return;
L_08A14854:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A1488C;
      }
      goto L_08A1485C;
    }
L_08A1485C:
    ctx.gpr[4] = (16281u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 39322u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(208));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(240));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[31] = (0x08A1488Cu);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 565u, 0x0892753Cu>(ctx, &aot_mem) && ctx.pc == 0x08A1488Cu) goto L_08A1488C;
    return;
L_08A1488C:
    ctx.gpr[4] = (16204u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 52429u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<1u>(ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(192));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 1u, 3u>();
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A148C4;
      }
      goto L_08A148B8;
    }
L_08A148B8:
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(208));
    ctx.gpr[31] = (0x08A148C4u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 565u, 0x0892753Cu>(ctx, &aot_mem) && ctx.pc == 0x08A148C4u) goto L_08A148C4;
    return;
L_08A148C4:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(96));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(176));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(144));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(192));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(160));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(212)));
    ctx.gpr[5] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[5]);
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / vfpu_s[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(224));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[31] = (0x08A14920u);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 565u, 0x0892753Cu>(ctx, &aot_mem) && ctx.pc == 0x08A14920u) goto L_08A14920;
    return;
L_08A14920:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 1u);
      if (branch_taken) {
          goto L_08A14930;
      }
      goto L_08A14928;
    }
L_08A14928:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 0u);
      if (branch_taken) {
          goto L_08A14930;
      }
      goto L_08A14930;
    }
L_08A14930:
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(576)));
    ctx.gpr[17] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(580)));
    ctx.gpr[18] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(584)));
    ctx.gpr[19] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(588)));
    ctx.gpr[20] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(592)));
    ctx.gpr[21] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(596)));
    ctx.gpr[31] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(600)));
    jump_target = ctx.gpr[31];
    ctx.gpr[29] = (ctx.gpr[29] + static_cast<std::uint32_t>(608));
    local_pc = jump_target;
    if (++local_transfers < 2048u) { entry_id = 0u; goto LOCAL_DISPATCH; }
    ctx.pc = jump_target;
    return;
L_08A14954:
    ctx.gpr[29] = (ctx.gpr[29] + static_cast<std::uint32_t>(-1808));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1760), std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1764), std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1768), std::bit_cast<std::uint32_t>(ctx.fpr[24]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1772), ctx.gpr[16]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1776), ctx.gpr[17]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1780), ctx.gpr[18]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1784), ctx.gpr[19]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1788), ctx.gpr[20]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1792), ctx.gpr[21]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1796), ctx.gpr[31]);
    ctx.gpr[7] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(68)));
    ctx.gpr[7] = (ctx.gpr[7] & 8192u);
    ctx.gpr[7] = (0u < ctx.gpr[7] ? 1u : 0u);
    ctx.gpr[7] = (ctx.gpr[7] & 255u);
    ctx.gpr[16] = (ctx.gpr[4] | 0u);
    ctx.gpr[17] = (ctx.gpr[5] | 0u);
    ctx.gpr[18] = (ctx.gpr[6] | 0u);
    { const bool branch_taken = ctx.gpr[7] == 0u;
    ctx.fpr[20] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]));
      if (branch_taken) {
          goto L_08A14C10;
      }
      goto L_08A149A4;
    }
L_08A149A4:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 8192u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A14C10;
      }
      goto L_08A149BC;
    }
L_08A149BC:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(32), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(36), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(40), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[4] = (ctx.gpr[16] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(32));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(32), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(36), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(40), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.gpr[6] = (ctx.gpr[17] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[7] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[7]);
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(32), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(36), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(40), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.gpr[7] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[7]);
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(48));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(96));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(80), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(84), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(88), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[5] = (std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[5]);
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(80));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(64));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(112));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = std::fabs(std::sqrt(vfpu_s[i]));
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[22] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = std::fabs(std::sqrt(vfpu_s[i]));
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[5] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[24] = std::bit_cast<float>(ctx.gpr[5]);
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(128));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(0u);
    ctx.set_fpu_condition((!(std::isnan(ctx.fpr[22]) || std::isnan(ctx.fpr[12])) && ctx.fpr[22] == ctx.fpr[12]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14B18;
      }
      goto L_08A14AF4;
    }
L_08A14AF4:
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    ctx.set_vfpu_scalar_bits_ct<1u>(ctx.gpr[4]);
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / vfpu_s[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<1u, 1u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(128));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 1u, 3u>();
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A14B24;
      }
      goto L_08A14B18;
    }
L_08A14B18:
    ctx.gpr[4] = (2226u << 16u);
    ctx.gpr[31] = (0x08A14B24u);
    ctx.gpr[4] = (ctx.gpr[4] + static_cast<std::uint32_t>(1520));
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 117u, 0x08A0D1F0u>(ctx, &aot_mem) && ctx.pc == 0x08A14B24u) goto L_08A14B24;
    return;
L_08A14B24:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(208)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(208)));
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[22]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[14] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[14] = fs * ft; }
    { const float fs = ctx.fpr[13]; const float ft = ctx.fpr[24]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[15] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[15] = fs * ft; }
    ctx.fpr[14] = ctx.fpr[14] + ctx.fpr[15];
    ctx.fpr[16] = ctx.fpr[12] + ctx.fpr[13];
    ctx.fpr[14] = ctx.fpr[14] / ctx.fpr[16];
    ctx.fpr[17] = ctx.fpr[22] - ctx.fpr[14];
    ctx.fpr[18] = std::bit_cast<float>(0u);
    ctx.set_fpu_condition((ctx.fpr[17] <= ctx.fpr[18]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14C08;
      }
      goto L_08A14B58;
    }
L_08A14B58:
    ctx.fpr[22] = ctx.fpr[14] - ctx.fpr[22];
    { const float fs = ctx.fpr[22]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[22] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[22] = fs * ft; }
    ctx.fpr[12] = ctx.fpr[14] - ctx.fpr[24];
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.gpr[4] = (2230u << 16u);
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(-8144)));
    { const float fs = ctx.fpr[15]; const float ft = ctx.fpr[20]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[20] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[20] = fs * ft; }
    ctx.fpr[16] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]) ^ 0x80000000u);
    ctx.set_fpu_condition((ctx.fpr[22] < ctx.fpr[16]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14B8C;
      }
      goto L_08A14B88;
    }
L_08A14B88:
    ctx.fpr[22] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]) ^ 0x80000000u);
    goto L_08A14B8C;
L_08A14B8C:
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]) ^ 0x80000000u);
    ctx.set_fpu_condition((ctx.fpr[22] < ctx.fpr[13]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14BA4;
      }
      goto L_08A14BA0;
    }
L_08A14BA0:
    ctx.fpr[22] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]) ^ 0x80000000u);
    goto L_08A14BA4;
L_08A14BA4:
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(128));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(144));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[5] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[5]);
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(160));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(144)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(148)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(152)));
    ctx.gpr[31] = (0x08A14BECu);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 459u, 0x08A0F7E0u>(ctx, &aot_mem) && ctx.pc == 0x08A14BECu) goto L_08A14BEC;
    return;
L_08A14BEC:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(160)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(164)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(168)));
    ctx.gpr[31] = (0x08A14C00u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 459u, 0x08A0F7E0u>(ctx, &aot_mem) && ctx.pc == 0x08A14C00u) goto L_08A14C00;
    return;
L_08A14C00:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 1u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A14C08;
    }
L_08A14C08:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 0u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A14C10;
    }
L_08A14C10:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 8192u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A14FF4;
      }
      goto L_08A14C28;
    }
L_08A14C28:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A14C68;
      }
      goto L_08A14C44;
    }
L_08A14C44:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 12u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    if (ctx.gpr[4] != 0u) {
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(336)));
        goto L_08A14C70;
    }
    goto L_08A14C60;
L_08A14C60:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A14C98;
      }
      goto L_08A14C68;
    }
L_08A14C68:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 0u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A14C70;
    }
L_08A14C70:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(108)));
    ctx.gpr[6] = (ctx.gpr[5] + static_cast<std::uint32_t>(16));
    ctx.gpr[5] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[6] + static_cast<std::uint32_t>(0))))));
    ctx.gpr[4] = (ctx.gpr[4] + ctx.gpr[5]);
    ctx.gpr[5] = (2224u << 16u);
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[6] + static_cast<std::uint32_t>(4)));
    jump_target = ctx.gpr[6];
    ctx.gpr[31] = (0x08A14C90u);
    ctx.gpr[5] = (ctx.gpr[5] + static_cast<std::uint32_t>(-23884));
    ctx.pc = jump_target;
    if (rt.invoke_chained_call(ctx, &aot_mem) && ctx.pc == 0x08A14C90u) goto L_08A14C90;
    return;
L_08A14C90:
    { const bool branch_taken = ctx.gpr[2] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A14E90;
      }
      goto L_08A14C98;
    }
L_08A14C98:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(0)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(176), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(4)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(180), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(8)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(184), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(176));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[17] + static_cast<std::uint32_t>(48));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(240));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(240)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(244)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(248)));
    ctx.gpr[4] = (ctx.gpr[17] + static_cast<std::uint32_t>(128));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[17] + static_cast<std::uint32_t>(160));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(1216));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1232), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1236), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1240), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(1232));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(1200));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[17] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(1184));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[17] + static_cast<std::uint32_t>(144));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[19] = (ctx.gpr[29] + static_cast<std::uint32_t>(256));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(192), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(196), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(200), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[4] = (ctx.gpr[16] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(192));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[7] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[7]);
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(192), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(196), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(200), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    { const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[7] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[7]);
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(192), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(196), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(200), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.gpr[7] = (std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[7]);
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(208));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[20] = (ctx.gpr[29] + static_cast<std::uint32_t>(272));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[20] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[21] = (ctx.gpr[29] + static_cast<std::uint32_t>(224));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(176), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(180), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(184), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.gpr[31] = (0x08A14E10u);
    ctx.gpr[4] = (ctx.gpr[21] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 120u, 0x08A0D25Cu>(ctx, &aot_mem) && ctx.pc == 0x08A14E10u) goto L_08A14E10;
    return;
L_08A14E10:
    { const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(288));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[20] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = std::fabs(std::sqrt(vfpu_s[i]));
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[5] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[22] = std::bit_cast<float>(ctx.gpr[5]);
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = std::fabs(std::sqrt(vfpu_s[i]));
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[24] = std::bit_cast<float>(ctx.gpr[4]);
    { const std::uint32_t vfpu_address = ctx.gpr[20] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(304));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(0u);
    ctx.set_fpu_condition((!(std::isnan(ctx.fpr[22]) || std::isnan(ctx.fpr[12])) && ctx.fpr[22] == ctx.fpr[12]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14E98;
      }
      goto L_08A14E6C;
    }
L_08A14E6C:
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    ctx.set_vfpu_scalar_bits_ct<1u>(ctx.gpr[4]);
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / vfpu_s[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<1u, 1u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(304));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 1u, 3u>();
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A14EA4;
      }
      goto L_08A14E90;
    }
L_08A14E90:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 0u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A14E98;
    }
L_08A14E98:
    ctx.gpr[4] = (2226u << 16u);
    ctx.gpr[31] = (0x08A14EA4u);
    ctx.gpr[4] = (ctx.gpr[4] + static_cast<std::uint32_t>(1564));
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 117u, 0x08A0D1F0u>(ctx, &aot_mem) && ctx.pc == 0x08A14EA4u) goto L_08A14EA4;
    return;
L_08A14EA4:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(208)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(208)));
    ctx.gpr[4] = (16256u << 16u);
    ctx.fpr[14] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[13] = ctx.fpr[14] / ctx.fpr[13];
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(240));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(304));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(224));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[16] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(212)));
    ctx.fpr[15] = ctx.fpr[15] / ctx.fpr[16];
    ctx.fpr[13] = ctx.fpr[13] + ctx.fpr[15];
    ctx.fpr[13] = ctx.fpr[14] / ctx.fpr[13];
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[22]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[17] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[17] = fs * ft; }
    { const float fs = ctx.fpr[13]; const float ft = ctx.fpr[24]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[18] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[18] = fs * ft; }
    ctx.fpr[14] = ctx.fpr[17] + ctx.fpr[18];
    ctx.fpr[16] = ctx.fpr[12] + ctx.fpr[13];
    ctx.fpr[14] = ctx.fpr[14] / ctx.fpr[16];
    ctx.fpr[15] = ctx.fpr[22] - ctx.fpr[14];
    ctx.fpr[19] = std::bit_cast<float>(0u);
    ctx.set_fpu_condition((ctx.fpr[15] <= ctx.fpr[19]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14FEC;
      }
      goto L_08A14F20;
    }
L_08A14F20:
    ctx.fpr[22] = ctx.fpr[14] - ctx.fpr[22];
    { const float fs = ctx.fpr[22]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[22] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[22] = fs * ft; }
    ctx.fpr[12] = ctx.fpr[14] - ctx.fpr[24];
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.gpr[4] = (2230u << 16u);
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(-8144)));
    { const float fs = ctx.fpr[15]; const float ft = ctx.fpr[20]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[20] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[20] = fs * ft; }
    ctx.fpr[16] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]) ^ 0x80000000u);
    ctx.set_fpu_condition((ctx.fpr[22] < ctx.fpr[16]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14F54;
      }
      goto L_08A14F50;
    }
L_08A14F50:
    ctx.fpr[22] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]) ^ 0x80000000u);
    goto L_08A14F54;
L_08A14F54:
    ctx.set_fpu_condition((ctx.fpr[20] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A14F68;
      }
      goto L_08A14F64;
    }
L_08A14F64:
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    goto L_08A14F68;
L_08A14F68:
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(304));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(320));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[5] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[5]);
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(336));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(320)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(324)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(328)));
    ctx.gpr[31] = (0x08A14FB0u);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 459u, 0x08A0F7E0u>(ctx, &aot_mem) && ctx.pc == 0x08A14FB0u) goto L_08A14FB0;
    return;
L_08A14FB0:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(336)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(340)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(344)));
    ctx.gpr[31] = (0x08A14FC4u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 459u, 0x08A0F7E0u>(ctx, &aot_mem) && ctx.pc == 0x08A14FC4u) goto L_08A14FC4;
    return;
L_08A14FC4:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(336)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(340)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(344)));
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(240)));
    ctx.fpr[16] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(244)));
    ctx.fpr[17] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(248)));
    ctx.gpr[31] = (0x08A14FE4u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 461u, 0x08A0F82Cu>(ctx, &aot_mem) && ctx.pc == 0x08A14FE4u) goto L_08A14FE4;
    return;
L_08A14FE4:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 1u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A14FEC;
    }
L_08A14FEC:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 0u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A14FF4;
    }
L_08A14FF4:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 8192u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15390;
      }
      goto L_08A1500C;
    }
L_08A1500C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A1522C;
      }
      goto L_08A15028;
    }
L_08A15028:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(0)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(352), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(4)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(356), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(8)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(360), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(352));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[16] + static_cast<std::uint32_t>(48));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(416));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(416)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(420)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(424)));
    ctx.gpr[5] = (ctx.gpr[16] + static_cast<std::uint32_t>(128));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[16] + static_cast<std::uint32_t>(160));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(1344));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1360), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1364), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1368), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(1360));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(1328));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[16] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(1312));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[16] + static_cast<std::uint32_t>(144));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(432));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(368), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(372), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(376), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(368));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[7] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[7]);
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(368), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(372), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(376), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.gpr[7] = (ctx.gpr[17] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[7] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[8] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[8]);
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(368), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(372), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(376), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.gpr[8] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[8]);
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(384));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(448));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(352), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(356), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(360), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[6] = (std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(400));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[7] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(464));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = std::fabs(std::sqrt(vfpu_s[i]));
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[22] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = std::fabs(std::sqrt(vfpu_s[i]));
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[24] = std::bit_cast<float>(ctx.gpr[4]);
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(480));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(0u);
    ctx.set_fpu_condition((!(std::isnan(ctx.fpr[22]) || std::isnan(ctx.fpr[12])) && ctx.fpr[22] == ctx.fpr[12]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A15234;
      }
      goto L_08A15208;
    }
L_08A15208:
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    ctx.set_vfpu_scalar_bits_ct<1u>(ctx.gpr[4]);
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / vfpu_s[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<1u, 1u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(480));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 1u, 3u>();
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15240;
      }
      goto L_08A1522C;
    }
L_08A1522C:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 0u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A15234;
    }
L_08A15234:
    ctx.gpr[4] = (2226u << 16u);
    ctx.gpr[31] = (0x08A15240u);
    ctx.gpr[4] = (ctx.gpr[4] + static_cast<std::uint32_t>(1608));
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 117u, 0x08A0D1F0u>(ctx, &aot_mem) && ctx.pc == 0x08A15240u) goto L_08A15240;
    return;
L_08A15240:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(208)));
    ctx.gpr[4] = (16256u << 16u);
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[12] = ctx.fpr[13] / ctx.fpr[12];
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(416));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(480));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(400));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[14] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(212)));
    ctx.fpr[14] = ctx.fpr[14] / ctx.fpr[15];
    ctx.fpr[12] = ctx.fpr[12] + ctx.fpr[14];
    ctx.fpr[12] = ctx.fpr[13] / ctx.fpr[12];
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(208)));
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[22]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[16] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[16] = fs * ft; }
    { const float fs = ctx.fpr[13]; const float ft = ctx.fpr[24]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[17] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[17] = fs * ft; }
    ctx.fpr[14] = ctx.fpr[16] + ctx.fpr[17];
    ctx.fpr[15] = ctx.fpr[12] + ctx.fpr[13];
    ctx.fpr[14] = ctx.fpr[14] / ctx.fpr[15];
    ctx.fpr[18] = ctx.fpr[22] - ctx.fpr[14];
    ctx.fpr[19] = std::bit_cast<float>(0u);
    ctx.set_fpu_condition((ctx.fpr[18] <= ctx.fpr[19]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A15388;
      }
      goto L_08A152BC;
    }
L_08A152BC:
    ctx.fpr[22] = ctx.fpr[14] - ctx.fpr[22];
    { const float fs = ctx.fpr[22]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[22] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[22] = fs * ft; }
    ctx.fpr[12] = ctx.fpr[14] - ctx.fpr[24];
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.gpr[4] = (2230u << 16u);
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(-8144)));
    { const float fs = ctx.fpr[15]; const float ft = ctx.fpr[20]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[20] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[20] = fs * ft; }
    ctx.fpr[16] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]) ^ 0x80000000u);
    ctx.set_fpu_condition((ctx.fpr[22] < ctx.fpr[16]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A152F0;
      }
      goto L_08A152EC;
    }
L_08A152EC:
    ctx.fpr[22] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]) ^ 0x80000000u);
    goto L_08A152F0;
L_08A152F0:
    ctx.set_fpu_condition((ctx.fpr[20] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A15304;
      }
      goto L_08A15300;
    }
L_08A15300:
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    goto L_08A15304;
L_08A15304:
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(480));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(496));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[5] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[5]);
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(512));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(496)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(500)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(504)));
    ctx.gpr[31] = (0x08A1534Cu);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 459u, 0x08A0F7E0u>(ctx, &aot_mem) && ctx.pc == 0x08A1534Cu) goto L_08A1534C;
    return;
L_08A1534C:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(496)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(500)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(504)));
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(416)));
    ctx.fpr[16] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(420)));
    ctx.fpr[17] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(424)));
    ctx.gpr[31] = (0x08A1536Cu);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 461u, 0x08A0F82Cu>(ctx, &aot_mem) && ctx.pc == 0x08A1536Cu) goto L_08A1536C;
    return;
L_08A1536C:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(512)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(516)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(520)));
    ctx.gpr[31] = (0x08A15380u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 459u, 0x08A0F7E0u>(ctx, &aot_mem) && ctx.pc == 0x08A15380u) goto L_08A15380;
    return;
L_08A15380:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 1u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A15388;
    }
L_08A15388:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 0u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A15390;
    }
L_08A15390:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(0)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(528), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(4)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(532), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(8)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(536), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(528));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[16] + static_cast<std::uint32_t>(48));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(608));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(0)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(544), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(4)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(548), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(8)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(552), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(544));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[17] + static_cast<std::uint32_t>(48));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(624));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(608)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(612)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(616)));
    ctx.gpr[5] = (ctx.gpr[16] + static_cast<std::uint32_t>(128));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[16] + static_cast<std::uint32_t>(160));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(1472));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1488), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1492), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1496), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(1488));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(1456));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[16] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(1440));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[16] + static_cast<std::uint32_t>(144));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(640));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(624)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(628)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(632)));
    ctx.gpr[6] = (ctx.gpr[17] + static_cast<std::uint32_t>(128));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (ctx.gpr[17] + static_cast<std::uint32_t>(160));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(1600));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1616), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1620), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1624), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(1616));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(1584));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (ctx.gpr[17] + static_cast<std::uint32_t>(112));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(1568));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (ctx.gpr[17] + static_cast<std::uint32_t>(144));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(656));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(560), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(564), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(568), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[7] = (ctx.gpr[29] + static_cast<std::uint32_t>(560));
    { const std::uint32_t vfpu_address = ctx.gpr[7] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[8] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[8]);
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(560), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(564), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(568), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[7] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[8] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[8]);
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(560), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(564), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(568), std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    ctx.gpr[8] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[8]);
    { const std::uint32_t vfpu_address = ctx.gpr[7] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[7] = (ctx.gpr[29] + static_cast<std::uint32_t>(576));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[7] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[7] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(672));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(544), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(548), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[18] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(552), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[7] = (std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[7]);
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(592));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(688));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = std::fabs(std::sqrt(vfpu_s[i]));
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[22] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = std::fabs(std::sqrt(vfpu_s[i]));
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[24] = std::bit_cast<float>(ctx.gpr[4]);
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(704));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(0u);
    ctx.set_fpu_condition((!(std::isnan(ctx.fpr[22]) || std::isnan(ctx.fpr[12])) && ctx.fpr[22] == ctx.fpr[12]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A15640;
      }
      goto L_08A1561C;
    }
L_08A1561C:
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    ctx.set_vfpu_scalar_bits_ct<1u>(ctx.gpr[4]);
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / vfpu_s[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<1u, 1u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(704));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 1u, 3u>();
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A1564C;
      }
      goto L_08A15640;
    }
L_08A15640:
    ctx.gpr[4] = (2226u << 16u);
    ctx.gpr[31] = (0x08A1564Cu);
    ctx.gpr[4] = (ctx.gpr[4] + static_cast<std::uint32_t>(1656));
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 117u, 0x08A0D1F0u>(ctx, &aot_mem) && ctx.pc == 0x08A1564Cu) goto L_08A1564C;
    return;
L_08A1564C:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(208)));
    ctx.gpr[4] = (16256u << 16u);
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[12] = ctx.fpr[13] / ctx.fpr[12];
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(608));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(704));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(592));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[14] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(212)));
    ctx.fpr[14] = ctx.fpr[14] / ctx.fpr[15];
    ctx.fpr[12] = ctx.fpr[12] + ctx.fpr[14];
    ctx.fpr[12] = ctx.fpr[13] / ctx.fpr[12];
    ctx.fpr[16] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(208)));
    ctx.fpr[16] = ctx.fpr[13] / ctx.fpr[16];
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(624));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    ctx.execute_vfpu_cross_quat(0u, 1u, 2u, 3u);
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[14] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(212)));
    ctx.fpr[14] = ctx.fpr[14] / ctx.fpr[15];
    ctx.fpr[14] = ctx.fpr[16] + ctx.fpr[14];
    ctx.fpr[13] = ctx.fpr[13] / ctx.fpr[14];
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[22]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[17] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[17] = fs * ft; }
    { const float fs = ctx.fpr[13]; const float ft = ctx.fpr[24]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[18] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[18] = fs * ft; }
    ctx.fpr[14] = ctx.fpr[17] + ctx.fpr[18];
    ctx.fpr[15] = ctx.fpr[12] + ctx.fpr[13];
    ctx.fpr[14] = ctx.fpr[14] / ctx.fpr[15];
    ctx.fpr[16] = ctx.fpr[22] - ctx.fpr[14];
    ctx.fpr[19] = std::bit_cast<float>(0u);
    ctx.set_fpu_condition((ctx.fpr[16] <= ctx.fpr[19]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A157EC;
      }
      goto L_08A15700;
    }
L_08A15700:
    ctx.fpr[22] = ctx.fpr[14] - ctx.fpr[22];
    { const float fs = ctx.fpr[22]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[22] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[22] = fs * ft; }
    ctx.fpr[12] = ctx.fpr[14] - ctx.fpr[24];
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[13]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.gpr[4] = (2230u << 16u);
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(-8144)));
    { const float fs = ctx.fpr[15]; const float ft = ctx.fpr[20]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[20] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[20] = fs * ft; }
    ctx.fpr[16] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]) ^ 0x80000000u);
    ctx.set_fpu_condition((ctx.fpr[22] < ctx.fpr[16]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A15734;
      }
      goto L_08A15730;
    }
L_08A15730:
    ctx.fpr[22] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]) ^ 0x80000000u);
    goto L_08A15734;
L_08A15734:
    ctx.set_fpu_condition((ctx.fpr[20] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A15748;
      }
      goto L_08A15744;
    }
L_08A15744:
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    goto L_08A15748;
L_08A15748:
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(704));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(720));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[5] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[5]);
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(736));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(720)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(724)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(728)));
    ctx.gpr[31] = (0x08A15790u);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 459u, 0x08A0F7E0u>(ctx, &aot_mem) && ctx.pc == 0x08A15790u) goto L_08A15790;
    return;
L_08A15790:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(720)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(724)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(728)));
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(608)));
    ctx.fpr[16] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(612)));
    ctx.fpr[17] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(616)));
    ctx.gpr[31] = (0x08A157B0u);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 461u, 0x08A0F82Cu>(ctx, &aot_mem) && ctx.pc == 0x08A157B0u) goto L_08A157B0;
    return;
L_08A157B0:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(736)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(740)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(744)));
    ctx.gpr[31] = (0x08A157C4u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 459u, 0x08A0F7E0u>(ctx, &aot_mem) && ctx.pc == 0x08A157C4u) goto L_08A157C4;
    return;
L_08A157C4:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(736)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(740)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(744)));
    ctx.fpr[15] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(624)));
    ctx.fpr[16] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(628)));
    ctx.fpr[17] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(632)));
    ctx.gpr[31] = (0x08A157E4u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 461u, 0x08A0F82Cu>(ctx, &aot_mem) && ctx.pc == 0x08A157E4u) goto L_08A157E4;
    return;
L_08A157E4:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 1u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A157EC;
    }
L_08A157EC:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 0u);
      if (branch_taken) {
          goto L_08A157F4;
      }
      goto L_08A157F4;
    }
L_08A157F4:
    ctx.fpr[20] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1760)));
    ctx.fpr[22] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1764)));
    ctx.fpr[24] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1768)));
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1772)));
    ctx.gpr[17] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1776)));
    ctx.gpr[18] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1780)));
    ctx.gpr[19] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1784)));
    ctx.gpr[20] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1788)));
    ctx.gpr[21] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1792)));
    ctx.gpr[31] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1796)));
    jump_target = ctx.gpr[31];
    ctx.gpr[29] = (ctx.gpr[29] + static_cast<std::uint32_t>(1808));
    local_pc = jump_target;
    if (++local_transfers < 2048u) { entry_id = 0u; goto LOCAL_DISPATCH; }
    ctx.pc = jump_target;
    return;
L_08A15824:
    ctx.gpr[29] = (ctx.gpr[29] + static_cast<std::uint32_t>(-1680));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1616), std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1620), std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1624), std::bit_cast<std::uint32_t>(ctx.fpr[24]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1628), std::bit_cast<std::uint32_t>(ctx.fpr[26]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1632), std::bit_cast<std::uint32_t>(ctx.fpr[28]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1636), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1640), ctx.gpr[16]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1644), ctx.gpr[17]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1648), ctx.gpr[18]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1652), ctx.gpr[19]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1656), ctx.gpr[20]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1660), ctx.gpr[21]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1664), ctx.gpr[22]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1668), ctx.gpr[23]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1672), ctx.gpr[30]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1676), ctx.gpr[31]);
    ctx.gpr[20] = (ctx.gpr[4] | 0u);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1604), ctx.gpr[5]);
    ctx.gpr[4] = (0u | 9u);
    ctx.gpr[5] = (49024u << 16u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[5]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1608), ctx.gpr[4]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(32), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(36), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[22] = (0u | 0u);
    ctx.gpr[16] = (ctx.gpr[29] + static_cast<std::uint32_t>(1136));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[20] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[4] = (ctx.gpr[4] << 2u);
    ctx.gpr[17] = (2229u << 16u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(-11604)));
    ctx.gpr[4] = (ctx.gpr[5] + ctx.gpr[4]);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(20)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1152), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(4)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1156), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(8)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1160), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(1152));
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    ctx.gpr[31] = (0x08A158D4u);
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 118u, 0x08A0D21Cu>(ctx, &aot_mem) && ctx.pc == 0x08A158D4u) goto L_08A158D4;
    return;
L_08A158D4:
    { const std::uint32_t vfpu_address = ctx.gpr[16] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(48));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[20] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[4] = (ctx.gpr[4] << 2u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[17] + static_cast<std::uint32_t>(-11604)));
    ctx.gpr[4] = (ctx.gpr[5] + ctx.gpr[4]);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(20)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(12)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(60), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 12u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    if (ctx.gpr[4] == 0u) {
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1608)));
        goto L_08A15934;
    }
    goto L_08A1591C;
L_08A1591C:
    ctx.gpr[4] = (0u | 8u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1604)));
    ctx.gpr[5] = (ctx.gpr[5] + static_cast<std::uint32_t>(4));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1608), ctx.gpr[4]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1604), ctx.gpr[5]);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1608)));
    goto L_08A15934;
L_08A15934:
    ctx.gpr[5] = (ctx.gpr[4] + static_cast<std::uint32_t>(-1));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1600), ctx.gpr[5]);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1608), ctx.gpr[5]);
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 68u, 0x08A182D4u>(ctx, &aot_mem); return;
      }
      goto L_08A15944;
    }
L_08A15944:
    ctx.gpr[4] = (2232u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] + static_cast<std::uint32_t>(-31744));
    ctx.gpr[5] = (ctx.gpr[20] + static_cast<std::uint32_t>(300));
    ctx.gpr[6] = (ctx.gpr[20] + static_cast<std::uint32_t>(304));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1572), ctx.gpr[4]);
    ctx.gpr[4] = (2230u << 16u);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1568), ctx.gpr[5]);
    ctx.gpr[4] = (ctx.gpr[4] + static_cast<std::uint32_t>(-8107));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1564), ctx.gpr[6]);
    ctx.gpr[5] = (ctx.gpr[20] + static_cast<std::uint32_t>(112));
    ctx.gpr[6] = (ctx.gpr[20] + static_cast<std::uint32_t>(128));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1552), ctx.gpr[4]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1544), ctx.gpr[5]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1548), ctx.gpr[6]);
    ctx.fpr[30] = std::bit_cast<float>(0u);
    ctx.gpr[4] = (ctx.gpr[20] + static_cast<std::uint32_t>(48));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1588), ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[20] + static_cast<std::uint32_t>(16));
    ctx.gpr[5] = (2224u << 16u);
    ctx.gpr[5] = (ctx.gpr[5] + static_cast<std::uint32_t>(-23884));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1596), ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[20] + static_cast<std::uint32_t>(160));
    ctx.gpr[6] = (ctx.gpr[20] + static_cast<std::uint32_t>(144));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1592), ctx.gpr[5]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1580), ctx.gpr[4]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1584), ctx.gpr[6]);
    goto L_08A159AC;
L_08A159AC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1604)));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    { const bool branch_taken = ctx.gpr[4] == 0u;
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1576), ctx.gpr[4]);
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 67u, 0x08A182B0u>(ctx, &aot_mem); return;
      }
      goto L_08A159BC;
    }
L_08A159BC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1576)));
    ctx.gpr[19] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(8)));
    ctx.gpr[30] = (0u | 1u);
    ctx.gpr[4] = (ctx.gpr[30] | 0u);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1576), ctx.gpr[5]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & 512u);
    ctx.gpr[5] = (0u < ctx.gpr[5] ? 1u : 0u);
    ctx.gpr[6] = (ctx.gpr[5] & 255u);
    ctx.gpr[5] = (0u | 3u);
    ctx.gpr[8] = (2228u << 16u);
    ctx.gpr[9] = (2229u << 16u);
    ctx.gpr[10] = (2230u << 16u);
    ctx.gpr[11] = (0u + static_cast<std::uint32_t>(-2));
    ctx.gpr[7] = (48972u << 16u);
    ctx.gpr[7] = (ctx.gpr[7] | 52429u);
    ctx.fpr[20] = std::bit_cast<float>(ctx.gpr[7]);
    ctx.gpr[21] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1588)));
    ctx.gpr[2] = (0u | 241u);
    ctx.gpr[7] = (16168u << 16u);
    ctx.gpr[7] = (ctx.gpr[7] | 62915u);
    { const bool branch_taken = ctx.gpr[6] == 0u;
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[7]);
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 49u, 0x08A18210u>(ctx, &aot_mem); return;
      }
      goto L_08A15A1C;
    }
L_08A15A1C:
    ctx.gpr[23] = (2227u << 16u);
    ctx.gpr[6] = (aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(84)));
    ctx.gpr[7] = (aot_mem.aot_load16(ctx.gpr[23] + static_cast<std::uint32_t>(20720)));
    { const bool branch_taken = ctx.gpr[6] == ctx.gpr[7];
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 49u, 0x08A18210u>(ctx, &aot_mem); return;
      }
      goto L_08A15A30;
    }
L_08A15A30:
    { const bool branch_taken = ctx.gpr[19] == ctx.gpr[20];
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 49u, 0x08A18210u>(ctx, &aot_mem); return;
      }
      goto L_08A15A38;
    }
L_08A15A38:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[4] = (ctx.gpr[4] << 2u);
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[9] + static_cast<std::uint32_t>(-11604)));
    ctx.gpr[4] = (ctx.gpr[6] + ctx.gpr[4]);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(20)));
    { const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(16);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(32);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<2u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(48);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<3u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<4u, 4u>(vfpu_value); }
    { float vfpu_value[4]{};
      ctx.write_vfpu_vector_with_destination_prefix_ct<35u, 3u>(vfpu_value); }
    { float vfpu_value[4]{1.0f, 1.0f, 1.0f, 1.0f};
      ctx.write_vfpu_vector_with_destination_prefix_ct<99u, 1u>(vfpu_value); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(48));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<8u, 4u>(vfpu_value); }
    { float vfpu_matrix[16]{}, vfpu_target_raw[4]{}, vfpu_target[4]{}, vfpu_result[4]{};
      ctx.read_vfpu_matrix(vfpu_matrix, 32u, 4u);
      ctx.read_vfpu_vector_ct<4u, 4u>(vfpu_target_raw);
      constexpr std::uint32_t vfpu_side = 4u;
      constexpr std::uint32_t vfpu_input_length = 3u;
      for (std::uint32_t i = 0; i < 4u; ++i) vfpu_target[i] = i < vfpu_input_length ? vfpu_target_raw[i] : 0.0f;
      if (vfpu_side - 1u >= vfpu_input_length) vfpu_target[vfpu_side - 1u] = 1.0f;
      for (std::uint32_t row = 0; row + 1u < vfpu_side; ++row) {
        float sum = 0.0f;
        for (std::uint32_t column = 0; column < vfpu_side; ++column) sum += vfpu_matrix[row * 4u + column] * vfpu_target[column];
        vfpu_result[row] = sum;
      }
      float vfpu_final_row[4]{vfpu_matrix[(vfpu_side - 1u) * 4u + 0u], vfpu_matrix[(vfpu_side - 1u) * 4u + 1u],
                              vfpu_matrix[(vfpu_side - 1u) * 4u + 2u], vfpu_matrix[(vfpu_side - 1u) * 4u + 3u]};
      ctx.apply_vfpu_source_prefix_ct<4u, 0u>(vfpu_final_row);
      ctx.apply_vfpu_source_prefix_ct<4u, 1u>(vfpu_target);
      for (std::uint32_t column = 0; column < 4u; ++column) vfpu_result[vfpu_side - 1u] += vfpu_final_row[column] * vfpu_target[column];
      const std::uint32_t vfpu_destination_prefix = ctx.vfpu_ctrl[2];
      const std::uint32_t vfpu_last_lane = vfpu_side - 1u;
      ctx.vfpu_ctrl[2] = ((vfpu_destination_prefix & (1u << 8u)) << vfpu_last_lane) |
                         ((vfpu_destination_prefix & 3u) << (vfpu_last_lane * 2u));
      ctx.write_vfpu_vector_with_destination_prefix(vfpu_result, 5u, vfpu_side); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<100u, 1u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<104u, 1u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<100u, 1u>(vfpu_d); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<100u, 1u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<100u, 1u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = vfpu_s[i] * vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<100u, 1u>(vfpu_d); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<8u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<5u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<5u, 3u>(vfpu_d); }
    ctx.execute_vfpu_vdot_ct<4u, 5u, 5u, 3u>();
    ctx.execute_vfpu_vcmp_ct<4u, 100u, 1u, 7u>();
    ctx.gpr[4] = (0u | 0u);
    { const bool branch_taken = ((ctx.vfpu_ctrl[3] >> 0u) & 1u) != 0u;
    // vflush: architectural no-op that retains VFPU prefixes
      if (branch_taken) {
          goto L_08A15AA0;
      }
      goto L_08A15A98;
    }
L_08A15A98:
    ctx.gpr[4] = (0u + static_cast<std::uint32_t>(1));
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    goto L_08A15AA0;
L_08A15AA0:
    ctx.gpr[6] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[6] == 0u;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 49u, 0x08A18210u>(ctx, &aot_mem); return;
      }
      goto L_08A15AAC;
    }
L_08A15AAC:
    ctx.gpr[4] = (0u + static_cast<std::uint32_t>(-129));
    ctx.gpr[6] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[6] & ctx.gpr[4]);
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
    ctx.gpr[17] = (0u | 0u);
    ctx.gpr[18] = (0u | 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 2u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15AE8;
      }
      goto L_08A15AE0;
    }
L_08A15AE0:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[17] = (0u | 0u);
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A15AE8;
    }
L_08A15AE8:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15B78;
      }
      goto L_08A15B04;
    }
L_08A15B04:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(422))))));
    ctx.gpr[4] = (ctx.gpr[4] & 2u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15B78;
      }
      goto L_08A15B14;
    }
L_08A15B14:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    if (ctx.gpr[4] != 0u) {
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(40)));
        goto L_08A15B50;
    }
    goto L_08A15B30;
L_08A15B30:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 6u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15B78;
      }
      goto L_08A15B4C;
    }
L_08A15B4C:
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(40)));
    goto L_08A15B50;
L_08A15B50:
    ctx.set_fpu_condition((ctx.fpr[13] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A15B78;
      }
      goto L_08A15B60;
    }
L_08A15B60:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(452), ctx.gpr[19]);
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A15B78;
    }
L_08A15B78:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15C08;
      }
      goto L_08A15B94;
    }
L_08A15B94:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[19] + static_cast<std::uint32_t>(422))))));
    ctx.gpr[4] = (ctx.gpr[4] & 2u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15C08;
      }
      goto L_08A15BA4;
    }
L_08A15BA4:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    if (ctx.gpr[4] != 0u) {
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(40)));
        goto L_08A15BE0;
    }
    goto L_08A15BC0;
L_08A15BC0:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 6u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15C08;
      }
      goto L_08A15BDC;
    }
L_08A15BDC:
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(40)));
    goto L_08A15BE0;
L_08A15BE0:
    ctx.set_fpu_condition((ctx.fpr[13] < ctx.fpr[12]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A15C08;
      }
      goto L_08A15BF0;
    }
L_08A15BF0:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(452), ctx.gpr[20]);
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A15C08;
    }
L_08A15C08:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15C70;
      }
      goto L_08A15C24;
    }
L_08A15C24:
    ctx.gpr[6] = (aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(433)));
    ctx.gpr[4] = (0u | 4u);
    { const bool branch_taken = ctx.gpr[6] != ctx.gpr[4];
    // nop
      if (branch_taken) {
          goto L_08A15C70;
      }
      goto L_08A15C34;
    }
L_08A15C34:
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[6] = (ctx.gpr[6] & 14u);
    ctx.gpr[6] = (ctx.gpr[6] ^ 8u);
    ctx.gpr[6] = (ctx.gpr[6] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[6] = (ctx.gpr[6] & 255u);
    { const bool branch_taken = ctx.gpr[6] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15C70;
      }
      goto L_08A15C50;
    }
L_08A15C50:
    ctx.gpr[6] = (aot_mem.aot_load8(ctx.gpr[19] + static_cast<std::uint32_t>(433)));
    { const bool branch_taken = ctx.gpr[6] != ctx.gpr[4];
    // nop
      if (branch_taken) {
          goto L_08A15C70;
      }
      goto L_08A15C5C;
    }
L_08A15C5C:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A15C70;
    }
L_08A15C70:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15D20;
      }
      goto L_08A15C8C;
    }
L_08A15C8C:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(422))))));
    ctx.gpr[4] = (ctx.gpr[4] & 2u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15D20;
      }
      goto L_08A15C9C;
    }
L_08A15C9C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 2u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A15D20;
      }
      goto L_08A15CB8;
    }
L_08A15CB8:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[20] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[8] + static_cast<std::uint32_t>(-31116)));
    ctx.gpr[6] = (aot_mem.aot_load16(ctx.gpr[6] + static_cast<std::uint32_t>(2)));
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A15D08;
      }
      goto L_08A15CCC;
    }
L_08A15CCC:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[20] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[8] + static_cast<std::uint32_t>(-31116)));
    ctx.gpr[6] = (aot_mem.aot_load16(ctx.gpr[6] + static_cast<std::uint32_t>(4)));
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A15D08;
      }
      goto L_08A15CE0;
    }
L_08A15CE0:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[20] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[8] + static_cast<std::uint32_t>(-31116)));
    ctx.gpr[6] = (aot_mem.aot_load16(ctx.gpr[6] + static_cast<std::uint32_t>(6)));
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A15D08;
      }
      goto L_08A15CF4;
    }
L_08A15CF4:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[20] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[8] + static_cast<std::uint32_t>(-31116)));
    ctx.gpr[6] = (aot_mem.aot_load16(ctx.gpr[6] + static_cast<std::uint32_t>(8)));
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A15D20;
      }
      goto L_08A15D08;
    }
L_08A15D08:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(452), ctx.gpr[19]);
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A15D20;
    }
L_08A15D20:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15DD0;
      }
      goto L_08A15D3C;
    }
L_08A15D3C:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[19] + static_cast<std::uint32_t>(422))))));
    ctx.gpr[4] = (ctx.gpr[4] & 2u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15DD0;
      }
      goto L_08A15D4C;
    }
L_08A15D4C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 2u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A15DD0;
      }
      goto L_08A15D68;
    }
L_08A15D68:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[8] + static_cast<std::uint32_t>(-31116)));
    ctx.gpr[6] = (aot_mem.aot_load16(ctx.gpr[6] + static_cast<std::uint32_t>(2)));
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A15DB8;
      }
      goto L_08A15D7C;
    }
L_08A15D7C:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[8] + static_cast<std::uint32_t>(-31116)));
    ctx.gpr[6] = (aot_mem.aot_load16(ctx.gpr[6] + static_cast<std::uint32_t>(4)));
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A15DB8;
      }
      goto L_08A15D90;
    }
L_08A15D90:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[8] + static_cast<std::uint32_t>(-31116)));
    ctx.gpr[6] = (aot_mem.aot_load16(ctx.gpr[6] + static_cast<std::uint32_t>(6)));
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A15DB8;
      }
      goto L_08A15DA4;
    }
L_08A15DA4:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[8] + static_cast<std::uint32_t>(-31116)));
    ctx.gpr[6] = (aot_mem.aot_load16(ctx.gpr[6] + static_cast<std::uint32_t>(8)));
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[6];
    // nop
      if (branch_taken) {
          goto L_08A15DD0;
      }
      goto L_08A15DB8;
    }
L_08A15DB8:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(452), ctx.gpr[20]);
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A15DD0;
    }
L_08A15DD0:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15FE8;
      }
      goto L_08A15DEC;
    }
L_08A15DEC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15FE8;
      }
      goto L_08A15E08;
    }
L_08A15E08:
    ctx.gpr[16] = (ctx.gpr[20] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[16] + static_cast<std::uint32_t>(88))))));
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[2];
    // nop
      if (branch_taken) {
          goto L_08A15E20;
      }
      goto L_08A15E18;
    }
L_08A15E18:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
      if (branch_taken) {
          goto L_08A15FE0;
      }
      goto L_08A15E20;
    }
L_08A15E20:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(420)));
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A15E6C;
      }
      goto L_08A15E2C;
    }
L_08A15E2C:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(421))))));
    ctx.gpr[4] = (ctx.gpr[4] & 64u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A15E6C;
      }
      goto L_08A15E3C;
    }
L_08A15E3C:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & 2048u);
    { const bool branch_taken = ctx.gpr[5] != 0u;
    ctx.gpr[4] = (0u | 0u);
      if (branch_taken) {
          goto L_08A15E5C;
      }
      goto L_08A15E4C;
    }
L_08A15E4C:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(72)));
    ctx.gpr[5] = (ctx.gpr[5] & 4096u);
    { const bool branch_taken = ctx.gpr[5] == 0u;
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
      if (branch_taken) {
          goto L_08A15E64;
      }
      goto L_08A15E5C;
    }
L_08A15E5C:
    ctx.gpr[4] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    goto L_08A15E64;
L_08A15E64:
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A15FE0;
      }
      goto L_08A15E6C;
    }
L_08A15E6C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(452)));
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[19];
    // nop
      if (branch_taken) {
          goto L_08A15E80;
      }
      goto L_08A15E78;
    }
L_08A15E78:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
      if (branch_taken) {
          goto L_08A15FE0;
      }
      goto L_08A15E80;
    }
L_08A15E80:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(432)));
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[4]) < 20 ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15FE0;
      }
      goto L_08A15E90;
    }
L_08A15E90:
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(144), 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(148)));
    ctx.gpr[4] = (ctx.gpr[4] & ctx.gpr[11]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(148), ctx.gpr[4]);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[20] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[10] + static_cast<std::uint32_t>(-8152)));
    ctx.gpr[6] = (static_cast<std::int32_t>(ctx.gpr[4]) < static_cast<std::int32_t>(ctx.gpr[6]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[6] == 0u;
    ctx.gpr[5] = (0u | 0u);
      if (branch_taken) {
          goto L_08A15EC4;
      }
      goto L_08A15EB4;
    }
L_08A15EB4:
    ctx.gpr[4] = (ctx.gpr[4] << 2u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[9] + static_cast<std::uint32_t>(-11604)));
    ctx.gpr[4] = (ctx.gpr[5] + ctx.gpr[4]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    goto L_08A15EC4;
L_08A15EC4:
    ctx.gpr[8] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(20)));
    ctx.gpr[8] = (ctx.gpr[8] + static_cast<std::uint32_t>(32));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[20] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[10] + static_cast<std::uint32_t>(-8152)));
    ctx.gpr[6] = (static_cast<std::int32_t>(ctx.gpr[4]) < static_cast<std::int32_t>(ctx.gpr[6]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[6] == 0u;
    ctx.gpr[5] = (0u | 0u);
      if (branch_taken) {
          goto L_08A15EF0;
      }
      goto L_08A15EE0;
    }
L_08A15EE0:
    ctx.gpr[4] = (ctx.gpr[4] << 2u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[9] + static_cast<std::uint32_t>(-11604)));
    ctx.gpr[4] = (ctx.gpr[5] + ctx.gpr[4]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    goto L_08A15EF0;
L_08A15EF0:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(20)));
    ctx.gpr[4] = (ctx.gpr[4] + static_cast<std::uint32_t>(16));
    { const std::uint32_t vfpu_address = ctx.gpr[8] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[21] = (ctx.gpr[29] + static_cast<std::uint32_t>(160));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[20] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<4u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[20] + static_cast<std::uint32_t>(16);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<5u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[20] + static_cast<std::uint32_t>(32);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<6u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[20] + static_cast<std::uint32_t>(48);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<7u, 4u>(vfpu_value); }
    { float vfpu_matrix[16]{}, vfpu_target_raw[4]{}, vfpu_target[4]{}, vfpu_result[4]{};
      ctx.read_vfpu_matrix(vfpu_matrix, 36u, 3u);
      ctx.read_vfpu_vector_ct<1u, 3u>(vfpu_target_raw);
      constexpr std::uint32_t vfpu_side = 3u;
      constexpr std::uint32_t vfpu_input_length = 3u;
      for (std::uint32_t i = 0; i < 4u; ++i) vfpu_target[i] = i < vfpu_input_length ? vfpu_target_raw[i] : 0.0f;
      if (vfpu_side - 1u >= vfpu_input_length) vfpu_target[vfpu_side - 1u] = 1.0f;
      for (std::uint32_t row = 0; row + 1u < vfpu_side; ++row) {
        float sum = 0.0f;
        for (std::uint32_t column = 0; column < vfpu_side; ++column) sum += vfpu_matrix[row * 4u + column] * vfpu_target[column];
        vfpu_result[row] = sum;
      }
      float vfpu_final_row[4]{vfpu_matrix[(vfpu_side - 1u) * 4u + 0u], vfpu_matrix[(vfpu_side - 1u) * 4u + 1u],
                              vfpu_matrix[(vfpu_side - 1u) * 4u + 2u], vfpu_matrix[(vfpu_side - 1u) * 4u + 3u]};
      ctx.apply_vfpu_source_prefix_ct<4u, 0u>(vfpu_final_row);
      ctx.apply_vfpu_source_prefix_ct<4u, 1u>(vfpu_target);
      for (std::uint32_t column = 0; column < 4u; ++column) vfpu_result[vfpu_side - 1u] += vfpu_final_row[column] * vfpu_target[column];
      const std::uint32_t vfpu_destination_prefix = ctx.vfpu_ctrl[2];
      const std::uint32_t vfpu_last_lane = vfpu_side - 1u;
      ctx.vfpu_ctrl[2] = ((vfpu_destination_prefix & (1u << 8u)) << vfpu_last_lane) |
                         ((vfpu_destination_prefix & 3u) << (vfpu_last_lane * 2u));
      ctx.write_vfpu_vector_with_destination_prefix(vfpu_result, 0u, vfpu_side); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<7u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(64));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(168)));
    ctx.gpr[4] = (ctx.gpr[19] + static_cast<std::uint32_t>(48));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(8)));
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[13]));
    // nop
    if (!ctx.fpu_condition()) {
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1612), ctx.gpr[30]);
        goto L_08A15F60;
    }
    goto L_08A15F54;
L_08A15F54:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[16] + static_cast<std::uint32_t>(452), ctx.gpr[19]);
      if (branch_taken) {
          goto L_08A15FBC;
      }
      goto L_08A15F60;
    }
L_08A15F60:
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    ctx.gpr[30] = (ctx.gpr[29] + static_cast<std::uint32_t>(80));
    ctx.gpr[31] = (0x08A15F70u);
    ctx.gpr[5] = (ctx.gpr[30] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0128_entry, 128u, 511u, 0x08A06548u>(ctx, &aot_mem) && ctx.pc == 0x08A15F70u) goto L_08A15F70;
    return;
L_08A15F70:
    ctx.gpr[4] = (ctx.gpr[30] | 0u);
    ctx.gpr[31] = (0x08A15F7Cu);
    ctx.gpr[5] = (ctx.gpr[2] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0128_entry, 128u, 527u, 0x08A06A44u>(ctx, &aot_mem) && ctx.pc == 0x08A15F7Cu) goto L_08A15F7C;
    return;
L_08A15F7C:
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[30] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<4u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[30] + static_cast<std::uint32_t>(16);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<5u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[30] + static_cast<std::uint32_t>(32);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<6u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[30] + static_cast<std::uint32_t>(48);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<7u, 4u>(vfpu_value); }
    { float vfpu_matrix[16]{}, vfpu_target_raw[4]{}, vfpu_target[4]{}, vfpu_result[4]{};
      ctx.read_vfpu_matrix(vfpu_matrix, 36u, 3u);
      ctx.read_vfpu_vector_ct<1u, 3u>(vfpu_target_raw);
      constexpr std::uint32_t vfpu_side = 3u;
      constexpr std::uint32_t vfpu_input_length = 3u;
      for (std::uint32_t i = 0; i < 4u; ++i) vfpu_target[i] = i < vfpu_input_length ? vfpu_target_raw[i] : 0.0f;
      if (vfpu_side - 1u >= vfpu_input_length) vfpu_target[vfpu_side - 1u] = 1.0f;
      for (std::uint32_t row = 0; row + 1u < vfpu_side; ++row) {
        float sum = 0.0f;
        for (std::uint32_t column = 0; column < vfpu_side; ++column) sum += vfpu_matrix[row * 4u + column] * vfpu_target[column];
        vfpu_result[row] = sum;
      }
      float vfpu_final_row[4]{vfpu_matrix[(vfpu_side - 1u) * 4u + 0u], vfpu_matrix[(vfpu_side - 1u) * 4u + 1u],
                              vfpu_matrix[(vfpu_side - 1u) * 4u + 2u], vfpu_matrix[(vfpu_side - 1u) * 4u + 3u]};
      ctx.apply_vfpu_source_prefix_ct<4u, 0u>(vfpu_final_row);
      ctx.apply_vfpu_source_prefix_ct<4u, 1u>(vfpu_target);
      for (std::uint32_t column = 0; column < 4u; ++column) vfpu_result[vfpu_side - 1u] += vfpu_final_row[column] * vfpu_target[column];
      const std::uint32_t vfpu_destination_prefix = ctx.vfpu_ctrl[2];
      const std::uint32_t vfpu_last_lane = vfpu_side - 1u;
      ctx.vfpu_ctrl[2] = ((vfpu_destination_prefix & (1u << 8u)) << vfpu_last_lane) |
                         ((vfpu_destination_prefix & 3u) << (vfpu_last_lane * 2u));
      ctx.write_vfpu_vector_with_destination_prefix(vfpu_result, 0u, vfpu_side); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<7u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(176));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(184)));
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[30]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    ctx.gpr[30] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1612)));
      if (branch_taken) {
          goto L_08A15FBC;
      }
      goto L_08A15FB4;
    }
L_08A15FB4:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    aot_mem.aot_store32(ctx.gpr[16] + static_cast<std::uint32_t>(452), ctx.gpr[19]);
    goto L_08A15FBC;
L_08A15FBC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(148)));
    ctx.gpr[4] = (ctx.gpr[4] & 1u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15FE0;
      }
      goto L_08A15FCC;
    }
L_08A15FCC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(144)));
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A15FE0;
      }
      goto L_08A15FD8;
    }
L_08A15FD8:
    ctx.gpr[31] = (0x08A15FE0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(144)));
    if (rt.invoke_chained_direct<&recomp_unit_0134_entry, 134u, 263u, 0x08A1D574u>(ctx, &aot_mem) && ctx.pc == 0x08A15FE0u) goto L_08A15FE0;
    return;
L_08A15FE0:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A15FE8;
    }
L_08A15FE8:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A161F4;
      }
      goto L_08A16004;
    }
L_08A16004:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A161F4;
      }
      goto L_08A16020;
    }
L_08A16020:
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    ctx.gpr[6] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[4] + static_cast<std::uint32_t>(88))))));
    { const bool branch_taken = ctx.gpr[6] != ctx.gpr[2];
    // nop
      if (branch_taken) {
          goto L_08A16038;
      }
      goto L_08A16030;
    }
L_08A16030:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
      if (branch_taken) {
          goto L_08A161EC;
      }
      goto L_08A16038;
    }
L_08A16038:
    ctx.gpr[6] = (aot_mem.aot_load8(ctx.gpr[4] + static_cast<std::uint32_t>(420)));
    { const bool branch_taken = ctx.gpr[6] == ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A16084;
      }
      goto L_08A16044;
    }
L_08A16044:
    ctx.gpr[5] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[4] + static_cast<std::uint32_t>(421))))));
    ctx.gpr[5] = (ctx.gpr[5] & 64u);
    { const bool branch_taken = ctx.gpr[5] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A16084;
      }
      goto L_08A16054;
    }
L_08A16054:
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(68)));
    ctx.gpr[6] = (ctx.gpr[6] & 2048u);
    { const bool branch_taken = ctx.gpr[6] != 0u;
    ctx.gpr[5] = (0u | 0u);
      if (branch_taken) {
          goto L_08A16074;
      }
      goto L_08A16064;
    }
L_08A16064:
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(72)));
    ctx.gpr[6] = (ctx.gpr[6] & 4096u);
    { const bool branch_taken = ctx.gpr[6] == 0u;
    ctx.gpr[5] = (ctx.gpr[5] & 255u);
      if (branch_taken) {
          goto L_08A1607C;
      }
      goto L_08A16074;
    }
L_08A16074:
    ctx.gpr[5] = (ctx.gpr[30] | 0u);
    ctx.gpr[5] = (ctx.gpr[5] & 255u);
    goto L_08A1607C;
L_08A1607C:
    { const bool branch_taken = ctx.gpr[5] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A161EC;
      }
      goto L_08A16084;
    }
L_08A16084:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(452)));
    { const bool branch_taken = ctx.gpr[5] != ctx.gpr[20];
    // nop
      if (branch_taken) {
          goto L_08A16098;
      }
      goto L_08A16090;
    }
L_08A16090:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
      if (branch_taken) {
          goto L_08A161EC;
      }
      goto L_08A16098;
    }
L_08A16098:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[4] + static_cast<std::uint32_t>(432)));
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[4]) < 20 ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A161EC;
      }
      goto L_08A160A8;
    }
L_08A160A8:
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(272), 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(276)));
    ctx.gpr[4] = (ctx.gpr[4] & ctx.gpr[11]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(276), ctx.gpr[4]);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[10] + static_cast<std::uint32_t>(-8152)));
    ctx.gpr[6] = (static_cast<std::int32_t>(ctx.gpr[4]) < static_cast<std::int32_t>(ctx.gpr[6]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[6] == 0u;
    ctx.gpr[5] = (0u | 0u);
      if (branch_taken) {
          goto L_08A160DC;
      }
      goto L_08A160CC;
    }
L_08A160CC:
    ctx.gpr[4] = (ctx.gpr[4] << 2u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[9] + static_cast<std::uint32_t>(-11604)));
    ctx.gpr[4] = (ctx.gpr[5] + ctx.gpr[4]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    goto L_08A160DC;
L_08A160DC:
    ctx.gpr[8] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(20)));
    ctx.gpr[8] = (ctx.gpr[8] + static_cast<std::uint32_t>(32));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[10] + static_cast<std::uint32_t>(-8152)));
    ctx.gpr[6] = (static_cast<std::int32_t>(ctx.gpr[4]) < static_cast<std::int32_t>(ctx.gpr[6]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[6] == 0u;
    ctx.gpr[5] = (0u | 0u);
      if (branch_taken) {
          goto L_08A16108;
      }
      goto L_08A160F8;
    }
L_08A160F8:
    ctx.gpr[4] = (ctx.gpr[4] << 2u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[9] + static_cast<std::uint32_t>(-11604)));
    ctx.gpr[4] = (ctx.gpr[5] + ctx.gpr[4]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    goto L_08A16108;
L_08A16108:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(20)));
    ctx.gpr[4] = (ctx.gpr[4] + static_cast<std::uint32_t>(16));
    { const std::uint32_t vfpu_address = ctx.gpr[8] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[16] = (ctx.gpr[29] + static_cast<std::uint32_t>(288));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[16] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[16] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<4u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(16);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<5u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(32);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<6u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[19] + static_cast<std::uint32_t>(48);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<7u, 4u>(vfpu_value); }
    { float vfpu_matrix[16]{}, vfpu_target_raw[4]{}, vfpu_target[4]{}, vfpu_result[4]{};
      ctx.read_vfpu_matrix(vfpu_matrix, 36u, 3u);
      ctx.read_vfpu_vector_ct<1u, 3u>(vfpu_target_raw);
      constexpr std::uint32_t vfpu_side = 3u;
      constexpr std::uint32_t vfpu_input_length = 3u;
      for (std::uint32_t i = 0; i < 4u; ++i) vfpu_target[i] = i < vfpu_input_length ? vfpu_target_raw[i] : 0.0f;
      if (vfpu_side - 1u >= vfpu_input_length) vfpu_target[vfpu_side - 1u] = 1.0f;
      for (std::uint32_t row = 0; row + 1u < vfpu_side; ++row) {
        float sum = 0.0f;
        for (std::uint32_t column = 0; column < vfpu_side; ++column) sum += vfpu_matrix[row * 4u + column] * vfpu_target[column];
        vfpu_result[row] = sum;
      }
      float vfpu_final_row[4]{vfpu_matrix[(vfpu_side - 1u) * 4u + 0u], vfpu_matrix[(vfpu_side - 1u) * 4u + 1u],
                              vfpu_matrix[(vfpu_side - 1u) * 4u + 2u], vfpu_matrix[(vfpu_side - 1u) * 4u + 3u]};
      ctx.apply_vfpu_source_prefix_ct<4u, 0u>(vfpu_final_row);
      ctx.apply_vfpu_source_prefix_ct<4u, 1u>(vfpu_target);
      for (std::uint32_t column = 0; column < 4u; ++column) vfpu_result[vfpu_side - 1u] += vfpu_final_row[column] * vfpu_target[column];
      const std::uint32_t vfpu_destination_prefix = ctx.vfpu_ctrl[2];
      const std::uint32_t vfpu_last_lane = vfpu_side - 1u;
      ctx.vfpu_ctrl[2] = ((vfpu_destination_prefix & (1u << 8u)) << vfpu_last_lane) |
                         ((vfpu_destination_prefix & 3u) << (vfpu_last_lane * 2u));
      ctx.write_vfpu_vector_with_destination_prefix(vfpu_result, 0u, vfpu_side); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<7u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(192));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[16] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[21] + static_cast<std::uint32_t>(8)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16170;
      }
      goto L_08A16168;
    }
L_08A16168:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
      if (branch_taken) {
          goto L_08A161C8;
      }
      goto L_08A16170;
    }
L_08A16170:
    ctx.gpr[21] = (ctx.gpr[29] + static_cast<std::uint32_t>(208));
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[31] = (0x08A16180u);
    ctx.gpr[5] = (ctx.gpr[21] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0128_entry, 128u, 511u, 0x08A06548u>(ctx, &aot_mem) && ctx.pc == 0x08A16180u) goto L_08A16180;
    return;
L_08A16180:
    ctx.gpr[4] = (ctx.gpr[21] | 0u);
    ctx.gpr[31] = (0x08A1618Cu);
    ctx.gpr[5] = (ctx.gpr[2] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0128_entry, 128u, 527u, 0x08A06A44u>(ctx, &aot_mem) && ctx.pc == 0x08A1618Cu) goto L_08A1618C;
    return;
L_08A1618C:
    { const std::uint32_t vfpu_address = ctx.gpr[16] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<4u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(16);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<5u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(32);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<6u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(48);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<7u, 4u>(vfpu_value); }
    { float vfpu_matrix[16]{}, vfpu_target_raw[4]{}, vfpu_target[4]{}, vfpu_result[4]{};
      ctx.read_vfpu_matrix(vfpu_matrix, 36u, 3u);
      ctx.read_vfpu_vector_ct<1u, 3u>(vfpu_target_raw);
      constexpr std::uint32_t vfpu_side = 3u;
      constexpr std::uint32_t vfpu_input_length = 3u;
      for (std::uint32_t i = 0; i < 4u; ++i) vfpu_target[i] = i < vfpu_input_length ? vfpu_target_raw[i] : 0.0f;
      if (vfpu_side - 1u >= vfpu_input_length) vfpu_target[vfpu_side - 1u] = 1.0f;
      for (std::uint32_t row = 0; row + 1u < vfpu_side; ++row) {
        float sum = 0.0f;
        for (std::uint32_t column = 0; column < vfpu_side; ++column) sum += vfpu_matrix[row * 4u + column] * vfpu_target[column];
        vfpu_result[row] = sum;
      }
      float vfpu_final_row[4]{vfpu_matrix[(vfpu_side - 1u) * 4u + 0u], vfpu_matrix[(vfpu_side - 1u) * 4u + 1u],
                              vfpu_matrix[(vfpu_side - 1u) * 4u + 2u], vfpu_matrix[(vfpu_side - 1u) * 4u + 3u]};
      ctx.apply_vfpu_source_prefix_ct<4u, 0u>(vfpu_final_row);
      ctx.apply_vfpu_source_prefix_ct<4u, 1u>(vfpu_target);
      for (std::uint32_t column = 0; column < 4u; ++column) vfpu_result[vfpu_side - 1u] += vfpu_final_row[column] * vfpu_target[column];
      const std::uint32_t vfpu_destination_prefix = ctx.vfpu_ctrl[2];
      const std::uint32_t vfpu_last_lane = vfpu_side - 1u;
      ctx.vfpu_ctrl[2] = ((vfpu_destination_prefix & (1u << 8u)) << vfpu_last_lane) |
                         ((vfpu_destination_prefix & 3u) << (vfpu_last_lane * 2u));
      ctx.write_vfpu_vector_with_destination_prefix(vfpu_result, 0u, vfpu_side); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<7u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] + vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(304));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(312)));
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[30]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A161C8;
      }
      goto L_08A161C4;
    }
L_08A161C4:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    goto L_08A161C8;
L_08A161C8:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(276)));
    ctx.gpr[4] = (ctx.gpr[4] & 1u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A161EC;
      }
      goto L_08A161D8;
    }
L_08A161D8:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(272)));
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A161EC;
      }
      goto L_08A161E4;
    }
L_08A161E4:
    ctx.gpr[31] = (0x08A161ECu);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(272)));
    if (rt.invoke_chained_direct<&recomp_unit_0134_entry, 134u, 263u, 0x08A1D574u>(ctx, &aot_mem) && ctx.pc == 0x08A161ECu) goto L_08A161EC;
    return;
L_08A161EC:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A161F4;
    }
L_08A161F4:
    ctx.gpr[5] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[20] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[4] = (0u | 270u);
    { const bool branch_taken = ctx.gpr[5] != ctx.gpr[4];
    // nop
      if (branch_taken) {
          goto L_08A16240;
      }
      goto L_08A16204;
    }
L_08A16204:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & 14u);
    ctx.gpr[5] = (ctx.gpr[5] ^ 6u);
    ctx.gpr[5] = (ctx.gpr[5] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[5] = (ctx.gpr[5] & 255u);
    { const bool branch_taken = ctx.gpr[5] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16240;
      }
      goto L_08A16220;
    }
L_08A16220:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(56)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(56)));
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[13]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16240;
      }
      goto L_08A16238;
    }
L_08A16238:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A16240;
    }
L_08A16240:
    ctx.gpr[5] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    { const bool branch_taken = ctx.gpr[5] != ctx.gpr[4];
    // nop
      if (branch_taken) {
          goto L_08A16294;
      }
      goto L_08A1624C;
    }
L_08A1624C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 6u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16294;
      }
      goto L_08A16268;
    }
L_08A16268:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(56)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(56)));
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[13]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16294;
      }
      goto L_08A16280;
    }
L_08A16280:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A16294;
    }
L_08A16294:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 6u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16300;
      }
      goto L_08A162B0;
    }
L_08A162B0:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(1412)));
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[19];
    // nop
      if (branch_taken) {
          goto L_08A162C8;
      }
      goto L_08A162BC;
    }
L_08A162BC:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(324)));
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A16300;
      }
      goto L_08A162C8;
    }
L_08A162C8:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (256u << 16u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(408)));
    ctx.gpr[4] = (ctx.gpr[5] & ctx.gpr[4]);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A162F0;
      }
      goto L_08A162E0;
    }
L_08A162E0:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(416)));
    ctx.gpr[4] = (ctx.gpr[4] & 8u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A162F0;
    }
L_08A162F0:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A16300;
    }
L_08A16300:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A1653C;
      }
      goto L_08A1631C;
    }
L_08A1631C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 12u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A1653C;
      }
      goto L_08A16338;
    }
L_08A16338:
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(336)));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(108)));
    ctx.gpr[5] = (ctx.gpr[4] + static_cast<std::uint32_t>(64));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[5] + static_cast<std::uint32_t>(0))))));
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(4)));
    jump_target = ctx.gpr[5];
    ctx.gpr[31] = (0x08A16354u);
    ctx.gpr[4] = (ctx.gpr[16] + ctx.gpr[4]);
    ctx.pc = jump_target;
    if (rt.invoke_chained_call(ctx, &aot_mem) && ctx.pc == 0x08A16354u) goto L_08A16354;
    return;
L_08A16354:
    { const bool branch_taken = ctx.gpr[2] != ctx.gpr[30];
    // nop
      if (branch_taken) {
          goto L_08A16534;
      }
      goto L_08A1635C;
    }
L_08A1635C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(96)));
    if (ctx.gpr[4] != 0u) {
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(96)));
        goto L_08A16388;
    }
    goto L_08A16368;
L_08A16368:
    ctx.gpr[5] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[16] + static_cast<std::uint32_t>(10))))));
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(1264));
    ctx.gpr[31] = (0x08A16378u);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0176_entry, 176u, 222u, 0x08AC5938u>(ctx, &aot_mem) && ctx.pc == 0x08A16378u) goto L_08A16378;
    return;
L_08A16378:
    aot_mem.aot_store32(ctx.gpr[16] + static_cast<std::uint32_t>(96), ctx.gpr[2]);
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[29] + static_cast<std::uint32_t>(1264)));
    aot_mem.aot_store8(ctx.gpr[16] + static_cast<std::uint32_t>(4), static_cast<std::uint8_t>(ctx.gpr[4]));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(96)));
    goto L_08A16388;
L_08A16388:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[4] + static_cast<std::uint32_t>(340)));
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16524;
      }
      goto L_08A1639C;
    }
L_08A1639C:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(675))))));
    ctx.gpr[5] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(3))))));
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A16524;
      }
      goto L_08A163AC;
    }
L_08A163AC:
    ctx.gpr[4] = (ctx.gpr[19] + static_cast<std::uint32_t>(48));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(320));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; vfpu_value[0u] = 1.0f;
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 1u, 1u, 3u>();
    ctx.execute_vfpu_vcmp_ct<28u, 32u, 1u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / std::sqrt(vfpu_s[i]);
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.execute_vfpu_vscl_ct<1u, 1u, 28u, 3u>();
    ctx.execute_vfpu_vcmov_ct<1u, 0u, 3u, 0u, false>();
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<1u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1544)));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(336));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; vfpu_value[0u] = 1.0f;
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 1u, 1u, 3u>();
    ctx.execute_vfpu_vcmp_ct<28u, 32u, 1u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / std::sqrt(vfpu_s[i]);
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.execute_vfpu_vscl_ct<1u, 1u, 28u, 3u>();
    ctx.execute_vfpu_vcmov_ct<1u, 0u, 3u, 0u, false>();
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<1u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1596)));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(352));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; vfpu_value[0u] = 1.0f;
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 1u, 1u, 3u>();
    ctx.execute_vfpu_vcmp_ct<28u, 32u, 1u, 3u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / std::sqrt(vfpu_s[i]);
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.execute_vfpu_vscl_ct<1u, 1u, 28u, 3u>();
    ctx.execute_vfpu_vcmov_ct<1u, 0u, 3u, 0u, false>();
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<1u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 1u, 3u>();
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(620)));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 2u>();
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = std::fabs(std::sqrt(vfpu_s[i]));
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[14] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.gpr[4] = (16128u << 16u);
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[15]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    ctx.gpr[4] = (16025u << 16u);
      if (branch_taken) {
          goto L_08A1649C;
      }
      goto L_08A16488;
    }
L_08A16488:
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[20]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A164CC;
      }
      goto L_08A16498;
    }
L_08A16498:
    ctx.gpr[4] = (16025u << 16u);
    goto L_08A1649C;
L_08A1649C:
    ctx.gpr[4] = (ctx.gpr[4] | 39322u);
    ctx.fpr[16] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[14] <= ctx.fpr[16]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    ctx.gpr[4] = (15651u << 16u);
      if (branch_taken) {
          goto L_08A164FC;
      }
      goto L_08A164B4;
    }
L_08A164B4:
    ctx.gpr[4] = (ctx.gpr[4] | 55050u);
    ctx.fpr[16] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[13] <= ctx.fpr[16]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A164FC;
      }
      goto L_08A164CC;
    }
L_08A164CC:
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[15]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A1651C;
      }
      goto L_08A164DC;
    }
L_08A164DC:
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[20]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A1651C;
      }
      goto L_08A164EC;
    }
L_08A164EC:
    ctx.set_fpu_condition((ctx.fpr[14] <= ctx.fpr[15]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A1651C;
      }
      goto L_08A164FC;
    }
L_08A164FC:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(3))))));
    ctx.gpr[5] = (0u | 20u);
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(675), static_cast<std::uint8_t>(ctx.gpr[4]));
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(676), static_cast<std::uint8_t>(ctx.gpr[5]));
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
    goto L_08A1651C;
L_08A1651C:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16534;
      }
      goto L_08A16524;
    }
L_08A16524:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
    goto L_08A16534;
L_08A16534:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A1653C;
    }
L_08A1653C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 12u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A165C8;
      }
      goto L_08A16558;
    }
L_08A16558:
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(336)));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(108)));
    ctx.gpr[5] = (ctx.gpr[4] + static_cast<std::uint32_t>(64));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[5] + static_cast<std::uint32_t>(0))))));
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(4)));
    jump_target = ctx.gpr[5];
    ctx.gpr[31] = (0x08A16574u);
    ctx.gpr[4] = (ctx.gpr[16] + ctx.gpr[4]);
    ctx.pc = jump_target;
    if (rt.invoke_chained_call(ctx, &aot_mem) && ctx.pc == 0x08A16574u) goto L_08A16574;
    return;
L_08A16574:
    { const bool branch_taken = ctx.gpr[2] != ctx.gpr[30];
    // nop
      if (branch_taken) {
          goto L_08A165C0;
      }
      goto L_08A1657C;
    }
L_08A1657C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(96)));
    if (ctx.gpr[4] != 0u) {
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(96)));
        goto L_08A165A8;
    }
    goto L_08A16588;
L_08A16588:
    ctx.gpr[5] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[16] + static_cast<std::uint32_t>(10))))));
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(1296));
    ctx.gpr[31] = (0x08A16598u);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0176_entry, 176u, 222u, 0x08AC5938u>(ctx, &aot_mem) && ctx.pc == 0x08A16598u) goto L_08A16598;
    return;
L_08A16598:
    aot_mem.aot_store32(ctx.gpr[16] + static_cast<std::uint32_t>(96), ctx.gpr[2]);
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[29] + static_cast<std::uint32_t>(1296)));
    aot_mem.aot_store8(ctx.gpr[16] + static_cast<std::uint32_t>(4), static_cast<std::uint8_t>(ctx.gpr[4]));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(96)));
    goto L_08A165A8;
L_08A165A8:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[4] + static_cast<std::uint32_t>(340)));
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A165C0;
      }
      goto L_08A165BC;
    }
L_08A165BC:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    goto L_08A165C0;
L_08A165C0:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A165C8;
    }
L_08A165C8:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 6u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16610;
      }
      goto L_08A165E4;
    }
L_08A165E4:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(1412)));
    { const bool branch_taken = ctx.gpr[20] == ctx.gpr[4];
    // nop
      if (branch_taken) {
          goto L_08A165FC;
      }
      goto L_08A165F0;
    }
L_08A165F0:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[19] + static_cast<std::uint32_t>(324)));
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A16610;
      }
      goto L_08A165FC;
    }
L_08A165FC:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A16610;
    }
L_08A16610:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 6u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A1662C;
    }
L_08A1662C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A16648;
    }
L_08A16648:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(416)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[30]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16660;
      }
      goto L_08A1665C;
    }
L_08A1665C:
    ctx.gpr[18] = (ctx.gpr[30] | 0u);
    goto L_08A16660;
L_08A16660:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A166BC;
      }
      goto L_08A1667C;
    }
L_08A1667C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 2u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A166BC;
      }
      goto L_08A16698;
    }
L_08A16698:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(324)));
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A16714;
      }
      goto L_08A166A4;
    }
L_08A166A4:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(452), ctx.gpr[20]);
      if (branch_taken) {
          goto L_08A16714;
      }
      goto L_08A166BC;
    }
L_08A166BC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16714;
      }
      goto L_08A166D8;
    }
L_08A166D8:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 2u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A16714;
      }
      goto L_08A166F4;
    }
L_08A166F4:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[19] + static_cast<std::uint32_t>(324)));
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A16714;
      }
      goto L_08A16700;
    }
L_08A16700:
    ctx.gpr[17] = (ctx.gpr[30] | 0u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] | 128u);
    aot_mem.aot_store8(ctx.gpr[20] + static_cast<std::uint32_t>(322), static_cast<std::uint8_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(452), ctx.gpr[20]);
    goto L_08A16714;
L_08A16714:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 512u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16734;
      }
      goto L_08A1672C;
    }
L_08A1672C:
    { const bool branch_taken = ctx.gpr[17] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A167C0;
      }
      goto L_08A16734;
    }
L_08A16734:
    ctx.gpr[4] = (aot_mem.aot_load16(ctx.gpr[23] + static_cast<std::uint32_t>(20720)));
    aot_mem.aot_store16(ctx.gpr[19] + static_cast<std::uint32_t>(84), static_cast<std::uint16_t>(ctx.gpr[4]));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(92)));
    ctx.gpr[5] = (ctx.gpr[4] + static_cast<std::uint32_t>(176));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[5] + static_cast<std::uint32_t>(0))))));
    ctx.gpr[4] = (ctx.gpr[20] + ctx.gpr[4]);
    ctx.gpr[7] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(4)));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1572)));
    jump_target = ctx.gpr[7];
    ctx.gpr[31] = (0x08A1675Cu);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.pc = jump_target;
    if (rt.invoke_chained_call(ctx, &aot_mem) && ctx.pc == 0x08A1675Cu) goto L_08A1675C;
    return;
L_08A1675C:
    ctx.gpr[4] = (ctx.gpr[2] | 0u);
    ctx.gpr[5] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(323))))));
    ctx.gpr[5] = (ctx.gpr[5] & 8u);
    { const bool branch_taken = ctx.gpr[5] == 0u;
    ctx.gpr[5] = (2229u << 16u);
      if (branch_taken) {
          goto L_08A1677C;
      }
      goto L_08A16770;
    }
L_08A16770:
    { const bool branch_taken = static_cast<std::int32_t>(ctx.gpr[4]) > 0;
    // nop
      if (branch_taken) {
          goto L_08A16794;
      }
      goto L_08A16778;
    }
L_08A16778:
    ctx.gpr[5] = (2229u << 16u);
    goto L_08A1677C;
L_08A1677C:
    ctx.gpr[5] = (aot_mem.aot_load8(ctx.gpr[5] + static_cast<std::uint32_t>(27452)));
    ctx.gpr[4] = (ctx.gpr[4] | ctx.gpr[5]);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A1679C;
      }
      goto L_08A1678C;
    }
L_08A1678C:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 66u, 0x08A182A4u>(ctx, &aot_mem); return;
      }
      goto L_08A16794;
    }
L_08A16794:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 1u);
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 69u, 0x08A182D8u>(ctx, &aot_mem); return;
      }
      goto L_08A1679C;
    }
L_08A1679C:
    ctx.gpr[31] = (0x08A167A4u);
    // nop
    if (rt.invoke_chained_direct<&recomp_unit_0116_entry, 116u, 198u, 0x089D596Cu>(ctx, &aot_mem) && ctx.pc == 0x08A167A4u) goto L_08A167A4;
    return;
L_08A167A4:
    { const bool branch_taken = ctx.gpr[20] != ctx.gpr[2];
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 66u, 0x08A182A4u>(ctx, &aot_mem); return;
      }
      goto L_08A167AC;
    }
L_08A167AC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(1412)));
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[19];
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 66u, 0x08A182A4u>(ctx, &aot_mem); return;
      }
      goto L_08A167B8;
    }
L_08A167B8:
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(1412), 0u);
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 66u, 0x08A182A4u>(ctx, &aot_mem); return;
      }
      goto L_08A167C0;
    }
L_08A167C0:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 2u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[4] = (2227u << 16u);
      if (branch_taken) {
          goto L_08A16810;
      }
      goto L_08A167DC;
    }
L_08A167DC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 16384u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[4] = (2227u << 16u);
      if (branch_taken) {
          goto L_08A16810;
      }
      goto L_08A167F4;
    }
L_08A167F4:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[19] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] & 8u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[4] = (2227u << 16u);
      if (branch_taken) {
          goto L_08A16810;
      }
      goto L_08A16804;
    }
L_08A16804:
    { const bool branch_taken = ctx.gpr[18] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17328;
      }
      goto L_08A1680C;
    }
L_08A1680C:
    ctx.gpr[4] = (2227u << 16u);
    goto L_08A16810;
L_08A16810:
    ctx.gpr[4] = (aot_mem.aot_load16(ctx.gpr[4] + static_cast<std::uint32_t>(20720)));
    aot_mem.aot_store16(ctx.gpr[19] + static_cast<std::uint32_t>(84), static_cast<std::uint16_t>(ctx.gpr[4]));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(92)));
    ctx.gpr[5] = (ctx.gpr[4] + static_cast<std::uint32_t>(176));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[5] + static_cast<std::uint32_t>(0))))));
    ctx.gpr[4] = (ctx.gpr[20] + ctx.gpr[4]);
    ctx.gpr[7] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(4)));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1572)));
    jump_target = ctx.gpr[7];
    ctx.gpr[31] = (0x08A16838u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.pc = jump_target;
    if (rt.invoke_chained_call(ctx, &aot_mem) && ctx.pc == 0x08A16838u) goto L_08A16838;
    return;
L_08A16838:
    ctx.gpr[21] = (ctx.gpr[2] | 0u);
    { const bool branch_taken = static_cast<std::int32_t>(ctx.gpr[21]) <= 0;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 66u, 0x08A182A4u>(ctx, &aot_mem); return;
      }
      goto L_08A16844;
    }
L_08A16844:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 6u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A168E0;
      }
      goto L_08A16860;
    }
L_08A16860:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 12u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A168E0;
      }
      goto L_08A1687C;
    }
L_08A1687C:
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(336)));
    { const bool branch_taken = ctx.gpr[16] == 0u;
    ctx.gpr[4] = (0u | 0u);
      if (branch_taken) {
          goto L_08A168B4;
      }
      goto L_08A16888;
    }
L_08A16888:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(108)));
    ctx.gpr[5] = (ctx.gpr[4] + static_cast<std::uint32_t>(16));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[5] + static_cast<std::uint32_t>(0))))));
    ctx.gpr[4] = (ctx.gpr[16] + ctx.gpr[4]);
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(4)));
    jump_target = ctx.gpr[6];
    ctx.gpr[31] = (0x08A168A4u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1592)));
    ctx.pc = jump_target;
    if (rt.invoke_chained_call(ctx, &aot_mem) && ctx.pc == 0x08A168A4u) goto L_08A168A4;
    return;
L_08A168A4:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    ctx.gpr[4] = (0u | 0u);
      if (branch_taken) {
          goto L_08A168B4;
      }
      goto L_08A168AC;
    }
L_08A168AC:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
      if (branch_taken) {
          goto L_08A168B4;
      }
      goto L_08A168B4;
    }
L_08A168B4:
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A168E0;
      }
      goto L_08A168BC;
    }
L_08A168BC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(120)));
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[20];
    // nop
      if (branch_taken) {
          goto L_08A168E0;
      }
      goto L_08A168C8;
    }
L_08A168C8:
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A168D8u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    if (rt.invoke_chained_direct<&recomp_unit_0109_entry, 109u, 206u, 0x089B8AD4u>(ctx, &aot_mem) && ctx.pc == 0x08A168D8u) goto L_08A168D8;
    return;
L_08A168D8:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 1u);
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 69u, 0x08A182D8u>(ctx, &aot_mem); return;
      }
      goto L_08A168E0;
    }
L_08A168E0:
    ctx.gpr[18] = (0u | 0u);
    ctx.fpr[22] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(368), std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(372), std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(376), std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(384), std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(388), std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(392), std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 4096u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    ctx.gpr[23] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1544)));
    ctx.gpr[30] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1548)));
    ctx.gpr[5] = (15692u << 16u);
    ctx.gpr[5] = (ctx.gpr[5] | 52429u);
    ctx.fpr[26] = std::bit_cast<float>(ctx.gpr[5]);
    ctx.gpr[5] = (16544u << 16u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    ctx.fpr[28] = std::bit_cast<float>(ctx.gpr[5]);
      if (branch_taken) {
          goto L_08A16C60;
      }
      goto L_08A16930;
    }
L_08A16930:
    ctx.gpr[17] = (0u | 0u);
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[17]) < static_cast<std::int32_t>(ctx.gpr[21]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17104;
      }
      goto L_08A16940;
    }
L_08A16940:
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1572)));
    ctx.gpr[4] = (0u | 0u);
    ctx.gpr[16] = (ctx.gpr[4] + ctx.gpr[16]);
    goto L_08A1694C;
L_08A1694C:
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    ctx.gpr[7] = (ctx.gpr[29] + static_cast<std::uint32_t>(32));
    ctx.gpr[8] = (ctx.gpr[29] + static_cast<std::uint32_t>(368));
    ctx.gpr[31] = (0x08A16968u);
    ctx.gpr[9] = (ctx.gpr[29] + static_cast<std::uint32_t>(384));
    goto L_08A14100;
L_08A16968:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16C48;
      }
      goto L_08A16970;
    }
L_08A16970:
    ctx.gpr[18] = (ctx.gpr[18] + static_cast<std::uint32_t>(1));
    ctx.gpr[18] = (ctx.gpr[18] & 255u);
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[22]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16990;
      }
      goto L_08A1698C;
    }
L_08A1698C:
    ctx.fpr[22] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    goto L_08A16990;
L_08A16990:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16B94;
      }
      goto L_08A169AC;
    }
L_08A169AC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 12u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A169DC;
      }
      goto L_08A169C8;
    }
L_08A169C8:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A169DCu);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 221u, 0x08A0DAB8u>(ctx, &aot_mem) && ctx.pc == 0x08A169DCu) goto L_08A169DC;
    return;
L_08A169DC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(836)));
    ctx.gpr[5] = (0u | 1u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A169FC;
      }
      goto L_08A169EC;
    }
L_08A169EC:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.gpr[5] = (0u | 22u);
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A16A58;
      }
      goto L_08A169FC;
    }
L_08A169FC:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(29)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(400), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(404), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(408), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16A58;
      }
      goto L_08A16A30;
    }
L_08A16A30:
    aot_mem.aot_store16(ctx.gpr[20] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(300), ctx.gpr[19]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1568)));
    ctx.gpr[31] = (0x08A16A48u);
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A16A48u) goto L_08A16A48;
    return;
L_08A16A48:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(400));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1564)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A16A58;
L_08A16A58:
    ctx.gpr[31] = (0x08A16A60u);
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 202u, 0x08925840u>(ctx, &aot_mem) && ctx.pc == 0x08A16A60u) goto L_08A16A60;
    return;
L_08A16A60:
    ctx.gpr[4] = (0u | 4u);
    { const bool branch_taken = ctx.gpr[2] != ctx.gpr[4];
    // nop
      if (branch_taken) {
          goto L_08A16A74;
      }
      goto L_08A16A6C;
    }
L_08A16A6C:
    ctx.gpr[4] = (0u | 18u);
    aot_mem.aot_store8(ctx.gpr[16] + static_cast<std::uint32_t>(28), static_cast<std::uint8_t>(ctx.gpr[4]));
    goto L_08A16A74;
L_08A16A74:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(40)));
    ctx.gpr[4] = (48921u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 39322u);
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[13]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16B34;
      }
      goto L_08A16A94;
    }
L_08A16A94:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(112)));
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[26]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16B34;
      }
      goto L_08A16AAC;
    }
L_08A16AAC:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(116)));
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[26]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16B34;
      }
      goto L_08A16AC4;
    }
L_08A16AC4:
    ctx.gpr[5] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(28)));
    ctx.gpr[4] = (15820u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 52429u);
    ctx.fpr[14] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    { const float fs = ctx.fpr[12]; const float ft = ctx.fpr[14]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[14] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[14] = fs * ft; }
    { const std::uint32_t vfpu_address = ctx.gpr[23] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[30] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]));
        goto L_08A16B10;
    }
    goto L_08A16B10;
L_08A16B10:
    ctx.gpr[7] = (ctx.gpr[5] | 0u);
    ctx.gpr[8] = (ctx.gpr[4] | 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1552)));
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[6] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A16B2Cu);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    if (rt.invoke_chained_direct<&recomp_unit_0024_entry, 24u, 137u, 0x088649D4u>(ctx, &aot_mem) && ctx.pc == 0x08A16B2Cu) goto L_08A16B2C;
    return;
L_08A16B2C:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16C48;
      }
      goto L_08A16B34;
    }
L_08A16B34:
    ctx.gpr[5] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(28)));
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    { const std::uint32_t vfpu_address = ctx.gpr[23] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[30] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]));
        goto L_08A16B70;
    }
    goto L_08A16B70;
L_08A16B70:
    ctx.gpr[7] = (ctx.gpr[5] | 0u);
    ctx.gpr[8] = (ctx.gpr[4] | 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1552)));
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[6] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A16B8Cu);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    if (rt.invoke_chained_direct<&recomp_unit_0024_entry, 24u, 137u, 0x088649D4u>(ctx, &aot_mem) && ctx.pc == 0x08A16B8Cu) goto L_08A16B8C;
    return;
L_08A16B8C:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16C48;
      }
      goto L_08A16B94;
    }
L_08A16B94:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(29)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(432), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(436), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(440), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16BF0;
      }
      goto L_08A16BC8;
    }
L_08A16BC8:
    aot_mem.aot_store16(ctx.gpr[20] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(300), ctx.gpr[19]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1568)));
    ctx.gpr[31] = (0x08A16BE0u);
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A16BE0u) goto L_08A16BE0;
    return;
L_08A16BE0:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(432));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1564)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A16BF0;
L_08A16BF0:
    ctx.gpr[5] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(28)));
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    { const std::uint32_t vfpu_address = ctx.gpr[23] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[30] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]));
        goto L_08A16C2C;
    }
    goto L_08A16C2C;
L_08A16C2C:
    ctx.gpr[7] = (ctx.gpr[5] | 0u);
    ctx.gpr[8] = (ctx.gpr[4] | 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1552)));
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[6] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A16C48u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    if (rt.invoke_chained_direct<&recomp_unit_0024_entry, 24u, 137u, 0x088649D4u>(ctx, &aot_mem) && ctx.pc == 0x08A16C48u) goto L_08A16C48;
    return;
L_08A16C48:
    ctx.gpr[17] = (ctx.gpr[17] + static_cast<std::uint32_t>(1));
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[17]) < static_cast<std::int32_t>(ctx.gpr[21]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[16] = (ctx.gpr[16] + static_cast<std::uint32_t>(32));
      if (branch_taken) {
          goto L_08A1694C;
      }
      goto L_08A16C58;
    }
L_08A16C58:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17104;
      }
      goto L_08A16C60;
    }
L_08A16C60:
    ctx.gpr[17] = (0u | 0u);
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[17]) < static_cast<std::int32_t>(ctx.gpr[21]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17104;
      }
      goto L_08A16C70;
    }
L_08A16C70:
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1572)));
    ctx.fpr[24] = std::bit_cast<float>(ctx.gpr[21]);
    ctx.fpr[24] = static_cast<float>(static_cast<std::int32_t>(std::bit_cast<std::uint32_t>(ctx.fpr[24])));
    ctx.gpr[4] = (0u | 0u);
    ctx.gpr[16] = (ctx.gpr[4] + ctx.gpr[16]);
    goto L_08A16C84;
L_08A16C84:
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[7] = (ctx.gpr[29] + static_cast<std::uint32_t>(32));
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    ctx.gpr[8] = (ctx.gpr[29] + static_cast<std::uint32_t>(368));
    ctx.gpr[31] = (0x08A16CA0u);
    ctx.gpr[9] = (ctx.gpr[29] + static_cast<std::uint32_t>(384));
    goto L_08A14100;
L_08A16CA0:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A170F4;
      }
      goto L_08A16CA8;
    }
L_08A16CA8:
    ctx.gpr[18] = (ctx.gpr[18] + static_cast<std::uint32_t>(1));
    ctx.gpr[18] = (ctx.gpr[18] & 255u);
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[22]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16CC8;
      }
      goto L_08A16CC4;
    }
L_08A16CC4:
    ctx.fpr[22] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    goto L_08A16CC8;
L_08A16CC8:
    ctx.gpr[31] = (0x08A16CD0u);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 199u, 0x089257E4u>(ctx, &aot_mem) && ctx.pc == 0x08A16CD0u) goto L_08A16CD0;
    return;
L_08A16CD0:
    ctx.fpr[20] = ctx.fpr[0] / ctx.fpr[24];
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17018;
      }
      goto L_08A16CF0;
    }
L_08A16CF0:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 12u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16D20;
      }
      goto L_08A16D0C;
    }
L_08A16D0C:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A16D20u);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 221u, 0x08A0DAB8u>(ctx, &aot_mem) && ctx.pc == 0x08A16D20u) goto L_08A16D20;
    return;
L_08A16D20:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(836)));
    ctx.gpr[5] = (0u | 1u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A16D40;
      }
      goto L_08A16D30;
    }
L_08A16D30:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.gpr[5] = (0u | 22u);
    { const bool branch_taken = ctx.gpr[4] == ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A16DA4;
      }
      goto L_08A16D40;
    }
L_08A16D40:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(29)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(464), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(468), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(472), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16D9C;
      }
      goto L_08A16D74;
    }
L_08A16D74:
    aot_mem.aot_store16(ctx.gpr[20] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(300), ctx.gpr[19]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1568)));
    ctx.gpr[31] = (0x08A16D8Cu);
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A16D8Cu) goto L_08A16D8C;
    return;
L_08A16D8C:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(464));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1564)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A16D9C;
L_08A16D9C:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16DA8;
      }
      goto L_08A16DA4;
    }
L_08A16DA4:
    ctx.fpr[20] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    goto L_08A16DA8;
L_08A16DA8:
    ctx.gpr[31] = (0x08A16DB0u);
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 202u, 0x08925840u>(ctx, &aot_mem) && ctx.pc == 0x08A16DB0u) goto L_08A16DB0;
    return;
L_08A16DB0:
    ctx.gpr[4] = (0u | 4u);
    { const bool branch_taken = ctx.gpr[2] != ctx.gpr[4];
    // nop
      if (branch_taken) {
          goto L_08A16DC4;
      }
      goto L_08A16DBC;
    }
L_08A16DBC:
    ctx.gpr[4] = (0u | 18u);
    aot_mem.aot_store8(ctx.gpr[16] + static_cast<std::uint32_t>(28), static_cast<std::uint8_t>(ctx.gpr[4]));
    goto L_08A16DC4;
L_08A16DC4:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(40)));
    ctx.gpr[4] = (48921u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 39322u);
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[13]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16E84;
      }
      goto L_08A16DE4;
    }
L_08A16DE4:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(112)));
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[26]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16E84;
      }
      goto L_08A16DFC;
    }
L_08A16DFC:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(116)));
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]) & 0x7FFFFFFFu);
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[26]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16E84;
      }
      goto L_08A16E14;
    }
L_08A16E14:
    ctx.gpr[5] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(28)));
    ctx.gpr[4] = (15820u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 52429u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    { const float fs = ctx.fpr[13]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    { const std::uint32_t vfpu_address = ctx.gpr[23] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[30] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[14] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.set_fpu_condition((ctx.fpr[13] <= ctx.fpr[14]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[14] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[13]));
        goto L_08A16E60;
    }
    goto L_08A16E60;
L_08A16E60:
    ctx.gpr[7] = (ctx.gpr[5] | 0u);
    ctx.gpr[8] = (ctx.gpr[4] | 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1552)));
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[6] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A16E7Cu);
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    if (rt.invoke_chained_direct<&recomp_unit_0024_entry, 24u, 137u, 0x088649D4u>(ctx, &aot_mem) && ctx.pc == 0x08A16E7Cu) goto L_08A16E7C;
    return;
L_08A16E7C:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A16EDC;
      }
      goto L_08A16E84;
    }
L_08A16E84:
    ctx.gpr[5] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(28)));
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    { const std::uint32_t vfpu_address = ctx.gpr[23] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[30] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]));
        goto L_08A16EC0;
    }
    goto L_08A16EC0;
L_08A16EC0:
    ctx.gpr[7] = (ctx.gpr[5] | 0u);
    ctx.gpr[8] = (ctx.gpr[4] | 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1552)));
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[6] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A16EDCu);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    if (rt.invoke_chained_direct<&recomp_unit_0024_entry, 24u, 137u, 0x088649D4u>(ctx, &aot_mem) && ctx.pc == 0x08A16EDCu) goto L_08A16EDC;
    return;
L_08A16EDC:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[20] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[5] = (0u | 169u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    ctx.gpr[4] = (15948u << 16u);
      if (branch_taken) {
          goto L_08A16EFC;
      }
      goto L_08A16EEC;
    }
L_08A16EEC:
    ctx.gpr[4] = (ctx.gpr[4] | 52429u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    { const bool branch_taken = 0u == 0u;
    { const float fs = ctx.fpr[20]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[20] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[20] = fs * ft; }
      if (branch_taken) {
          goto L_08A16FEC;
      }
      goto L_08A16EFC;
    }
L_08A16EFC:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(836)));
    ctx.gpr[5] = (0u | 1u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A16F70;
      }
      goto L_08A16F0C;
    }
L_08A16F0C:
    ctx.gpr[4] = (ctx.gpr[16] + static_cast<std::uint32_t>(16));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(8)));
    ctx.gpr[4] = (16153u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 39322u);
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16F68;
      }
      goto L_08A16F30;
    }
L_08A16F30:
    ctx.gpr[31] = (0x08A16F38u);
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 202u, 0x08925840u>(ctx, &aot_mem) && ctx.pc == 0x08A16F38u) goto L_08A16F38;
    return;
L_08A16F38:
    ctx.gpr[4] = (0u | 3u);
    { const bool branch_taken = ctx.gpr[2] == ctx.gpr[4];
    ctx.gpr[4] = (16448u << 16u);
      if (branch_taken) {
          goto L_08A16F5C;
      }
      goto L_08A16F44;
    }
L_08A16F44:
    ctx.gpr[31] = (0x08A16F4Cu);
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 202u, 0x08925840u>(ctx, &aot_mem) && ctx.pc == 0x08A16F4Cu) goto L_08A16F4C;
    return;
L_08A16F4C:
    ctx.gpr[4] = (0u | 4u);
    { const bool branch_taken = ctx.gpr[2] != ctx.gpr[4];
    // nop
      if (branch_taken) {
          goto L_08A16FEC;
      }
      goto L_08A16F58;
    }
L_08A16F58:
    ctx.gpr[4] = (16448u << 16u);
    goto L_08A16F5C;
L_08A16F5C:
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    { const bool branch_taken = 0u == 0u;
    { const float fs = ctx.fpr[20]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[20] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[20] = fs * ft; }
      if (branch_taken) {
          goto L_08A16FEC;
      }
      goto L_08A16F68;
    }
L_08A16F68:
    { const bool branch_taken = 0u == 0u;
    ctx.fpr[20] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[30]));
      if (branch_taken) {
          goto L_08A16FEC;
      }
      goto L_08A16F70;
    }
L_08A16F70:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 496u);
    ctx.gpr[5] = (0u | 80u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    ctx.gpr[4] = (16448u << 16u);
      if (branch_taken) {
          goto L_08A16F90;
      }
      goto L_08A16F84;
    }
L_08A16F84:
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    { const bool branch_taken = 0u == 0u;
    { const float fs = ctx.fpr[20]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[20] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[20] = fs * ft; }
      if (branch_taken) {
          goto L_08A16FEC;
      }
      goto L_08A16F90;
    }
L_08A16F90:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(40)));
    ctx.gpr[4] = (16025u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 39322u);
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A16FB8;
      }
      goto L_08A16FB0;
    }
L_08A16FB0:
    { const bool branch_taken = 0u == 0u;
    ctx.fpr[20] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[30]));
      if (branch_taken) {
          goto L_08A16FEC;
      }
      goto L_08A16FB8;
    }
L_08A16FB8:
    ctx.gpr[4] = (15605u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 49807u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.gpr[4] = (16256u << 16u);
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    { const float fs = ctx.fpr[14]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.fpr[12] = ctx.fpr[12] + ctx.fpr[13];
    ctx.set_fpu_condition((ctx.fpr[28] < ctx.fpr[12]));
    // nop
    if (ctx.fpu_condition()) {
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[28]));
        goto L_08A16FE8;
    }
    goto L_08A16FE8;
L_08A16FE8:
    { const float fs = ctx.fpr[20]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[20] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[20] = fs * ft; }
    goto L_08A16FEC;
L_08A16FEC:
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    ctx.gpr[31] = (0x08A16FFCu);
    ctx.gpr[5] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 463u, 0x08A0F8D0u>(ctx, &aot_mem) && ctx.pc == 0x08A16FFCu) goto L_08A16FFC;
    return;
L_08A16FFC:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    ctx.gpr[4] = (0u + static_cast<std::uint32_t>(-4097));
      if (branch_taken) {
          goto L_08A170F4;
      }
      goto L_08A17004;
    }
L_08A17004:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[5] & ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[4] | 4096u);
    { const bool branch_taken = 0u == 0u;
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(68), ctx.gpr[4]);
      if (branch_taken) {
          goto L_08A170F4;
      }
      goto L_08A17018;
    }
L_08A17018:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(29)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(512), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(516), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(520), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17074;
      }
      goto L_08A1704C;
    }
L_08A1704C:
    aot_mem.aot_store16(ctx.gpr[20] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(300), ctx.gpr[19]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1568)));
    ctx.gpr[31] = (0x08A17064u);
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A17064u) goto L_08A17064;
    return;
L_08A17064:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(512));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1564)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A17074;
L_08A17074:
    ctx.gpr[5] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(28)));
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    { const std::uint32_t vfpu_address = ctx.gpr[23] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[30] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]));
        goto L_08A170B0;
    }
    goto L_08A170B0;
L_08A170B0:
    ctx.gpr[7] = (ctx.gpr[5] | 0u);
    ctx.gpr[8] = (ctx.gpr[4] | 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1552)));
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[6] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A170CCu);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    if (rt.invoke_chained_direct<&recomp_unit_0024_entry, 24u, 137u, 0x088649D4u>(ctx, &aot_mem) && ctx.pc == 0x08A170CCu) goto L_08A170CC;
    return;
L_08A170CC:
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    ctx.gpr[31] = (0x08A170DCu);
    ctx.gpr[5] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 463u, 0x08A0F8D0u>(ctx, &aot_mem) && ctx.pc == 0x08A170DCu) goto L_08A170DC;
    return;
L_08A170DC:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    ctx.gpr[4] = (0u + static_cast<std::uint32_t>(-4097));
      if (branch_taken) {
          goto L_08A170F4;
      }
      goto L_08A170E4;
    }
L_08A170E4:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[5] & ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[4] | 4096u);
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(68), ctx.gpr[4]);
    goto L_08A170F4;
L_08A170F4:
    ctx.gpr[17] = (ctx.gpr[17] + static_cast<std::uint32_t>(1));
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[17]) < static_cast<std::int32_t>(ctx.gpr[21]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[16] = (ctx.gpr[16] + static_cast<std::uint32_t>(32));
      if (branch_taken) {
          goto L_08A16C84;
      }
      goto L_08A17104;
    }
L_08A17104:
    { const bool branch_taken = static_cast<std::int32_t>(ctx.gpr[18]) <= 0;
    // nop
      if (branch_taken) {
          goto L_08A17320;
      }
      goto L_08A1710C;
    }
L_08A1710C:
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[18]);
    ctx.fpr[20] = static_cast<float>(static_cast<std::int32_t>(std::bit_cast<std::uint32_t>(ctx.fpr[12])));
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[4]);
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / vfpu_s[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(368));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[17] = (ctx.gpr[29] + static_cast<std::uint32_t>(544));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[17] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[4] = (ctx.gpr[23] | 0u);
    ctx.gpr[31] = (0x08A17140u);
    ctx.gpr[5] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 565u, 0x0892753Cu>(ctx, &aot_mem) && ctx.pc == 0x08A17140u) goto L_08A17140;
    return;
L_08A17140:
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[4]);
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / vfpu_s[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    ctx.gpr[16] = (ctx.gpr[29] + static_cast<std::uint32_t>(384));
    { const std::uint32_t vfpu_address = ctx.gpr[16] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[17] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[4] = (ctx.gpr[30] | 0u);
    ctx.gpr[31] = (0x08A17168u);
    ctx.gpr[5] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 565u, 0x0892753Cu>(ctx, &aot_mem) && ctx.pc == 0x08A17168u) goto L_08A17168;
    return;
L_08A17168:
    ctx.gpr[4] = (2230u << 16u);
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[4] + static_cast<std::uint32_t>(-7146)));
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A172B0;
      }
      goto L_08A17178;
    }
L_08A17178:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 496u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[4] = (0u | 0u);
      if (branch_taken) {
          goto L_08A17190;
      }
      goto L_08A17188;
    }
L_08A17188:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[4] = (0u | 1u);
      if (branch_taken) {
          goto L_08A17190;
      }
      goto L_08A17190;
    }
L_08A17190:
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A172B0;
      }
      goto L_08A17198;
    }
L_08A17198:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A172B0;
      }
      goto L_08A171B4;
    }
L_08A171B4:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(112)));
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[30]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(112)));
        goto L_08A171D4;
    }
    goto L_08A171C8;
L_08A171C8:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(112)));
    { const bool branch_taken = 0u == 0u;
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]) ^ 0x80000000u);
      if (branch_taken) {
          goto L_08A171D4;
      }
      goto L_08A171D4;
    }
L_08A171D4:
    ctx.gpr[4] = (15948u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 52429u);
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A172B0;
      }
      goto L_08A171F0;
    }
L_08A171F0:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(116)));
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[30]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(116)));
        goto L_08A17210;
    }
    goto L_08A17204;
L_08A17204:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(116)));
    { const bool branch_taken = 0u == 0u;
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]) ^ 0x80000000u);
      if (branch_taken) {
          goto L_08A17210;
      }
      goto L_08A17210;
    }
L_08A17210:
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A172B0;
      }
      goto L_08A17220;
    }
L_08A17220:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int8_t>(aot_mem.aot_load8(ctx.gpr[20] + static_cast<std::uint32_t>(322))))));
    ctx.gpr[4] = (ctx.gpr[4] & 16u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          goto L_08A172B0;
      }
      goto L_08A17230;
    }
L_08A17230:
    ctx.gpr[4] = (48793u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 39322u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(144)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(368)));
    { const float fs = ctx.fpr[14]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[14] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[14] = fs * ft; }
    ctx.fpr[15] = std::bit_cast<float>(ctx.gpr[21]);
    ctx.fpr[15] = static_cast<float>(static_cast<std::int32_t>(std::bit_cast<std::uint32_t>(ctx.fpr[15])));
    ctx.fpr[14] = ctx.fpr[14] / ctx.fpr[15];
    ctx.fpr[13] = ctx.fpr[13] + ctx.fpr[14];
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(144), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(148)));
    ctx.fpr[16] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(372)));
    { const float fs = ctx.fpr[16]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[16] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[16] = fs * ft; }
    ctx.fpr[16] = ctx.fpr[16] / ctx.fpr[15];
    ctx.fpr[13] = ctx.fpr[13] + ctx.fpr[16];
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(148), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.gpr[4] = (std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[4]);
    { const std::uint32_t vfpu_address = ctx.gpr[16] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(576));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[5] = (std::bit_cast<std::uint32_t>(ctx.fpr[15]));
    ctx.set_vfpu_scalar_bits_ct<28u>(ctx.gpr[5]);
    { float vfpu_s[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<28u, 1u, 0u>(vfpu_s);
      for (std::uint32_t i = 0; i < 1u; ++i) vfpu_d[i] = 1.0f / vfpu_s[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<28u, 1u>(vfpu_d); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vscl_ct<0u, 0u, 28u, 3u>();
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(560));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[31] = (0x08A172B0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1580)));
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 565u, 0x0892753Cu>(ctx, &aot_mem) && ctx.pc == 0x08A172B0u) goto L_08A172B0;
    return;
L_08A172B0:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    ctx.gpr[4] = (2230u << 16u);
      if (branch_taken) {
          goto L_08A17304;
      }
      goto L_08A172CC;
    }
L_08A172CC:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[19] + static_cast<std::uint32_t>(432)));
    { const bool branch_taken = ctx.gpr[4] == 0u;
    ctx.gpr[4] = (2230u << 16u);
      if (branch_taken) {
          goto L_08A17304;
      }
      goto L_08A172D8;
    }
L_08A172D8:
    ctx.gpr[4] = (16800u << 16u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[22] <= ctx.fpr[12]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    ctx.gpr[4] = (2230u << 16u);
      if (branch_taken) {
          goto L_08A17304;
      }
      goto L_08A172F0;
    }
L_08A172F0:
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    ctx.gpr[31] = (0x08A17300u);
    ctx.gpr[5] = (0u | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0014_entry, 14u, 484u, 0x0883F1F4u>(ctx, &aot_mem) && ctx.pc == 0x08A17300u) goto L_08A17300;
    return;
L_08A17300:
    ctx.gpr[4] = (2230u << 16u);
    goto L_08A17304;
L_08A17304:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[4] + static_cast<std::uint32_t>(-7145)));
    { const bool branch_taken = ctx.gpr[4] == 0u;
    ctx.gpr[22] = (0u | 1u);
      if (branch_taken) {
          goto L_08A17318;
      }
      goto L_08A17310;
    }
L_08A17310:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17320;
      }
      goto L_08A17318;
    }
L_08A17318:
    { const bool branch_taken = 0u == 0u;
    ctx.gpr[2] = (0u | 1u);
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 69u, 0x08A182D8u>(ctx, &aot_mem); return;
      }
      goto L_08A17320;
    }
L_08A17320:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 66u, 0x08A182A4u>(ctx, &aot_mem); return;
      }
      goto L_08A17328;
    }
L_08A17328:
    ctx.gpr[4] = (2227u << 16u);
    ctx.gpr[4] = (aot_mem.aot_load16(ctx.gpr[4] + static_cast<std::uint32_t>(20720)));
    aot_mem.aot_store16(ctx.gpr[19] + static_cast<std::uint32_t>(84), static_cast<std::uint16_t>(ctx.gpr[4]));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(92)));
    ctx.gpr[5] = (ctx.gpr[4] + static_cast<std::uint32_t>(176));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[5] + static_cast<std::uint32_t>(0))))));
    ctx.gpr[4] = (ctx.gpr[20] + ctx.gpr[4]);
    ctx.gpr[7] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(4)));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1572)));
    jump_target = ctx.gpr[7];
    ctx.gpr[31] = (0x08A17354u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.pc = jump_target;
    if (rt.invoke_chained_call(ctx, &aot_mem) && ctx.pc == 0x08A17354u) goto L_08A17354;
    return;
L_08A17354:
    ctx.gpr[30] = (ctx.gpr[2] | 0u);
    { const bool branch_taken = static_cast<std::int32_t>(ctx.gpr[30]) <= 0;
    ctx.gpr[17] = (ctx.gpr[29] + static_cast<std::uint32_t>(832));
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 48u, 0x08A18208u>(ctx, &aot_mem); return;
      }
      goto L_08A17360;
    }
L_08A17360:
    ctx.fpr[24] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    ctx.fpr[26] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[24]));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 4096u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17564;
      }
      goto L_08A17380;
    }
L_08A17380:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 4096u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17564;
      }
      goto L_08A17398;
    }
L_08A17398:
    ctx.gpr[17] = (0u | 0u);
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[17]) < static_cast<std::int32_t>(ctx.gpr[30]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    ctx.gpr[23] = (ctx.gpr[19] + static_cast<std::uint32_t>(300));
      if (branch_taken) {
          goto L_08A17D2C;
      }
      goto L_08A173A8;
    }
L_08A173A8:
    ctx.gpr[22] = (ctx.gpr[19] + static_cast<std::uint32_t>(304));
    ctx.gpr[21] = (ctx.gpr[19] + static_cast<std::uint32_t>(112));
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1572)));
    ctx.gpr[18] = (ctx.gpr[19] + static_cast<std::uint32_t>(128));
    ctx.gpr[4] = (0u | 0u);
    ctx.gpr[16] = (ctx.gpr[4] + ctx.gpr[16]);
    goto L_08A173C0;
L_08A173C0:
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[7] = (ctx.gpr[29] + static_cast<std::uint32_t>(32));
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[8] = (ctx.gpr[29] + static_cast<std::uint32_t>(36));
    ctx.gpr[31] = (0x08A173D8u);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0131_entry, 131u, 369u, 0x08A11F78u>(ctx, &aot_mem) && ctx.pc == 0x08A173D8u) goto L_08A173D8;
    return;
L_08A173D8:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A1754C;
      }
      goto L_08A173E0;
    }
L_08A173E0:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[24]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A173F8;
      }
      goto L_08A173F4;
    }
L_08A173F4:
    ctx.fpr[24] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    goto L_08A173F8;
L_08A173F8:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[26]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17410;
      }
      goto L_08A1740C;
    }
L_08A1740C:
    ctx.fpr[26] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    goto L_08A17410;
L_08A17410:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(29)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(592), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(596), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(600), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A1746C;
      }
      goto L_08A17444;
    }
L_08A17444:
    aot_mem.aot_store16(ctx.gpr[20] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(300), ctx.gpr[19]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1568)));
    ctx.gpr[31] = (0x08A1745Cu);
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A1745Cu) goto L_08A1745C;
    return;
L_08A1745C:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(592));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1564)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A1746C;
L_08A1746C:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(31)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(592), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(596), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(600), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A174C4;
      }
      goto L_08A174A0;
    }
L_08A174A0:
    aot_mem.aot_store16(ctx.gpr[19] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(300), ctx.gpr[20]);
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[31] = (0x08A174B8u);
    ctx.gpr[5] = (ctx.gpr[23] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A174B8u) goto L_08A174B8;
    return;
L_08A174B8:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(592));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[22] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A174C4;
L_08A174C4:
    ctx.gpr[5] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(28)));
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1544)));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(592));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[18] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1548)));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(624));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]));
        goto L_08A17530;
    }
    goto L_08A17530;
L_08A17530:
    ctx.gpr[7] = (ctx.gpr[5] | 0u);
    ctx.gpr[8] = (ctx.gpr[4] | 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1552)));
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[6] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A1754Cu);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    if (rt.invoke_chained_direct<&recomp_unit_0024_entry, 24u, 137u, 0x088649D4u>(ctx, &aot_mem) && ctx.pc == 0x08A1754Cu) goto L_08A1754C;
    return;
L_08A1754C:
    ctx.gpr[17] = (ctx.gpr[17] + static_cast<std::uint32_t>(1));
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[17]) < static_cast<std::int32_t>(ctx.gpr[30]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[16] = (ctx.gpr[16] + static_cast<std::uint32_t>(32));
      if (branch_taken) {
          goto L_08A173C0;
      }
      goto L_08A1755C;
    }
L_08A1755C:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17D2C;
      }
      goto L_08A17564;
    }
L_08A17564:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 4096u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17838;
      }
      goto L_08A1757C;
    }
L_08A1757C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1584)));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(656));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1580)));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(672));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[6] = (0u + static_cast<std::uint32_t>(-4097));
    ctx.gpr[7] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[6] = (ctx.gpr[7] & ctx.gpr[6]);
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(68), ctx.gpr[6]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(640), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(644), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(648), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(640));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(640), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(644), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(648), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[17] = (0u | 0u);
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[17]) < static_cast<std::int32_t>(ctx.gpr[30]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    ctx.gpr[23] = (ctx.gpr[19] + static_cast<std::uint32_t>(300));
      if (branch_taken) {
          goto L_08A177E8;
      }
      goto L_08A175E8;
    }
L_08A175E8:
    ctx.gpr[22] = (ctx.gpr[19] + static_cast<std::uint32_t>(304));
    ctx.gpr[21] = (ctx.gpr[19] + static_cast<std::uint32_t>(112));
    ctx.gpr[18] = (ctx.gpr[19] + static_cast<std::uint32_t>(128));
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1572)));
    ctx.fpr[20] = std::bit_cast<float>(ctx.gpr[30]);
    ctx.fpr[20] = static_cast<float>(static_cast<std::int32_t>(std::bit_cast<std::uint32_t>(ctx.fpr[20])));
    ctx.gpr[4] = (0u | 0u);
    ctx.gpr[16] = (ctx.gpr[4] + ctx.gpr[16]);
    goto L_08A17608;
L_08A17608:
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[7] = (ctx.gpr[29] + static_cast<std::uint32_t>(32));
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[8] = (ctx.gpr[29] + static_cast<std::uint32_t>(36));
    ctx.gpr[31] = (0x08A17620u);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0131_entry, 131u, 369u, 0x08A11F78u>(ctx, &aot_mem) && ctx.pc == 0x08A17620u) goto L_08A17620;
    return;
L_08A17620:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A177D8;
      }
      goto L_08A17628;
    }
L_08A17628:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[24]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17640;
      }
      goto L_08A1763C;
    }
L_08A1763C:
    ctx.fpr[24] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    goto L_08A17640;
L_08A17640:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[26]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17658;
      }
      goto L_08A17654;
    }
L_08A17654:
    ctx.fpr[26] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    goto L_08A17658;
L_08A17658:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(29)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(688), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(692), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(696), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A176B4;
      }
      goto L_08A1768C;
    }
L_08A1768C:
    aot_mem.aot_store16(ctx.gpr[20] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(300), ctx.gpr[19]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1568)));
    ctx.gpr[31] = (0x08A176A4u);
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A176A4u) goto L_08A176A4;
    return;
L_08A176A4:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(688));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1564)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A176B4;
L_08A176B4:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(31)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(688), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(692), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(696), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A1770C;
      }
      goto L_08A176E8;
    }
L_08A176E8:
    aot_mem.aot_store16(ctx.gpr[19] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(300), ctx.gpr[20]);
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[31] = (0x08A17700u);
    ctx.gpr[5] = (ctx.gpr[23] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A17700u) goto L_08A17700;
    return;
L_08A17700:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(688));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[22] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A1770C;
L_08A1770C:
    ctx.gpr[5] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(28)));
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1544)));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(688));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[18] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1548)));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(720));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]));
        goto L_08A17778;
    }
    goto L_08A17778;
L_08A17778:
    ctx.gpr[7] = (ctx.gpr[5] | 0u);
    ctx.gpr[8] = (ctx.gpr[4] | 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1552)));
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[6] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A17794u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    if (rt.invoke_chained_direct<&recomp_unit_0024_entry, 24u, 137u, 0x088649D4u>(ctx, &aot_mem) && ctx.pc == 0x08A17794u) goto L_08A17794;
    return;
L_08A17794:
    ctx.gpr[31] = (0x08A1779Cu);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 199u, 0x089257E4u>(ctx, &aot_mem) && ctx.pc == 0x08A1779Cu) goto L_08A1779C;
    return;
L_08A1779C:
    ctx.fpr[12] = ctx.fpr[0] / ctx.fpr[20];
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A177B0u);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    goto L_08A14954;
L_08A177B0:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    ctx.gpr[4] = (0u + static_cast<std::uint32_t>(-4097));
      if (branch_taken) {
          goto L_08A177D8;
      }
      goto L_08A177B8;
    }
L_08A177B8:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & ctx.gpr[4]);
    ctx.gpr[5] = (ctx.gpr[5] | 4096u);
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(68), ctx.gpr[5]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[5] & ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[4] | 4096u);
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(68), ctx.gpr[4]);
    goto L_08A177D8;
L_08A177D8:
    ctx.gpr[17] = (ctx.gpr[17] + static_cast<std::uint32_t>(1));
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[17]) < static_cast<std::int32_t>(ctx.gpr[30]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[16] = (ctx.gpr[16] + static_cast<std::uint32_t>(32));
      if (branch_taken) {
          goto L_08A17608;
      }
      goto L_08A177E8;
    }
L_08A177E8:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 4096u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[4] = (0u + static_cast<std::uint32_t>(-4097));
      if (branch_taken) {
          goto L_08A17830;
      }
      goto L_08A17800;
    }
L_08A17800:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[5] & ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[4] | 4096u);
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(68), ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(656));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1584)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(672));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1580)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A17830;
L_08A17830:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17D2C;
      }
      goto L_08A17838;
    }
L_08A17838:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 4096u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17B14;
      }
      goto L_08A17850;
    }
L_08A17850:
    ctx.gpr[4] = (ctx.gpr[19] + static_cast<std::uint32_t>(144));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[5] = (ctx.gpr[29] + static_cast<std::uint32_t>(752));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1560), ctx.gpr[4]);
    ctx.gpr[5] = (ctx.gpr[19] + static_cast<std::uint32_t>(160));
    { const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(768));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1556), ctx.gpr[5]);
    ctx.gpr[6] = (0u + static_cast<std::uint32_t>(-4097));
    ctx.gpr[7] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[6] = (ctx.gpr[7] & ctx.gpr[6]);
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(68), ctx.gpr[6]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(736), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(740), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(744), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(736));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(736), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(740), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(744), std::bit_cast<std::uint32_t>(ctx.fpr[30]));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[5] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[17] = (0u | 0u);
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[17]) < static_cast<std::int32_t>(ctx.gpr[30]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    ctx.gpr[23] = (ctx.gpr[19] + static_cast<std::uint32_t>(300));
      if (branch_taken) {
          goto L_08A17AC4;
      }
      goto L_08A178C4;
    }
L_08A178C4:
    ctx.gpr[22] = (ctx.gpr[19] + static_cast<std::uint32_t>(304));
    ctx.gpr[21] = (ctx.gpr[19] + static_cast<std::uint32_t>(112));
    ctx.gpr[18] = (ctx.gpr[19] + static_cast<std::uint32_t>(128));
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1572)));
    ctx.fpr[20] = std::bit_cast<float>(ctx.gpr[30]);
    ctx.fpr[20] = static_cast<float>(static_cast<std::int32_t>(std::bit_cast<std::uint32_t>(ctx.fpr[20])));
    ctx.gpr[4] = (0u | 0u);
    ctx.gpr[16] = (ctx.gpr[4] + ctx.gpr[16]);
    goto L_08A178E4;
L_08A178E4:
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[7] = (ctx.gpr[29] + static_cast<std::uint32_t>(32));
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[8] = (ctx.gpr[29] + static_cast<std::uint32_t>(36));
    ctx.gpr[31] = (0x08A178FCu);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0131_entry, 131u, 369u, 0x08A11F78u>(ctx, &aot_mem) && ctx.pc == 0x08A178FCu) goto L_08A178FC;
    return;
L_08A178FC:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17AB4;
      }
      goto L_08A17904;
    }
L_08A17904:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[24]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A1791C;
      }
      goto L_08A17918;
    }
L_08A17918:
    ctx.fpr[24] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    goto L_08A1791C;
L_08A1791C:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[26]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17934;
      }
      goto L_08A17930;
    }
L_08A17930:
    ctx.fpr[26] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    goto L_08A17934;
L_08A17934:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(29)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(784), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(788), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(792), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17990;
      }
      goto L_08A17968;
    }
L_08A17968:
    aot_mem.aot_store16(ctx.gpr[20] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(300), ctx.gpr[19]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1568)));
    ctx.gpr[31] = (0x08A17980u);
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A17980u) goto L_08A17980;
    return;
L_08A17980:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(784));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1564)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A17990;
L_08A17990:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(31)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(784), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(788), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(792), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A179E8;
      }
      goto L_08A179C4;
    }
L_08A179C4:
    aot_mem.aot_store16(ctx.gpr[19] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(300), ctx.gpr[20]);
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[31] = (0x08A179DCu);
    ctx.gpr[5] = (ctx.gpr[23] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A179DCu) goto L_08A179DC;
    return;
L_08A179DC:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(784));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[22] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A179E8;
L_08A179E8:
    ctx.gpr[5] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(28)));
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.fpr[14] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    { const std::uint32_t vfpu_address = ctx.gpr[21] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1544)));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(784));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[6]);
    { const std::uint32_t vfpu_address = ctx.gpr[18] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1548)));
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[6] = (ctx.gpr[29] + static_cast<std::uint32_t>(816));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[6] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[6] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[13] = std::bit_cast<float>(ctx.gpr[6]);
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    if (!ctx.fpu_condition()) {
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]));
        goto L_08A17A54;
    }
    goto L_08A17A54;
L_08A17A54:
    ctx.gpr[7] = (ctx.gpr[5] | 0u);
    ctx.gpr[8] = (ctx.gpr[4] | 0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1552)));
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[6] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A17A70u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[14]));
    if (rt.invoke_chained_direct<&recomp_unit_0024_entry, 24u, 137u, 0x088649D4u>(ctx, &aot_mem) && ctx.pc == 0x08A17A70u) goto L_08A17A70;
    return;
L_08A17A70:
    ctx.gpr[31] = (0x08A17A78u);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 199u, 0x089257E4u>(ctx, &aot_mem) && ctx.pc == 0x08A17A78u) goto L_08A17A78;
    return;
L_08A17A78:
    ctx.fpr[12] = ctx.fpr[0] / ctx.fpr[20];
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A17A8Cu);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    goto L_08A14954;
L_08A17A8C:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    ctx.gpr[4] = (0u + static_cast<std::uint32_t>(-4097));
      if (branch_taken) {
          goto L_08A17AB4;
      }
      goto L_08A17A94;
    }
L_08A17A94:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & ctx.gpr[4]);
    ctx.gpr[5] = (ctx.gpr[5] | 4096u);
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(68), ctx.gpr[5]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[5] & ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[4] | 4096u);
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(68), ctx.gpr[4]);
    goto L_08A17AB4;
L_08A17AB4:
    ctx.gpr[17] = (ctx.gpr[17] + static_cast<std::uint32_t>(1));
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[17]) < static_cast<std::int32_t>(ctx.gpr[30]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[16] = (ctx.gpr[16] + static_cast<std::uint32_t>(32));
      if (branch_taken) {
          goto L_08A178E4;
      }
      goto L_08A17AC4;
    }
L_08A17AC4:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 4096u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[4] = (0u + static_cast<std::uint32_t>(-4097));
      if (branch_taken) {
          goto L_08A17B0C;
      }
      goto L_08A17ADC;
    }
L_08A17ADC:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[5] & ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[4] | 4096u);
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(68), ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(752));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1560)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(768));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1556)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A17B0C;
L_08A17B0C:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17D2C;
      }
      goto L_08A17B14;
    }
L_08A17B14:
    ctx.gpr[21] = (0u | 0u);
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[21]) < static_cast<std::int32_t>(ctx.gpr[30]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    ctx.gpr[4] = (ctx.gpr[19] + static_cast<std::uint32_t>(300));
      if (branch_taken) {
          goto L_08A17D2C;
      }
      goto L_08A17B24;
    }
L_08A17B24:
    ctx.gpr[5] = (ctx.gpr[19] + static_cast<std::uint32_t>(304));
    ctx.gpr[23] = (ctx.gpr[19] + static_cast<std::uint32_t>(112));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1540), ctx.gpr[4]);
    ctx.gpr[22] = (ctx.gpr[19] + static_cast<std::uint32_t>(128));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1536), ctx.gpr[5]);
    ctx.gpr[16] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1572)));
    ctx.fpr[28] = std::bit_cast<float>(ctx.gpr[30]);
    ctx.fpr[28] = static_cast<float>(static_cast<std::int32_t>(std::bit_cast<std::uint32_t>(ctx.fpr[28])));
    ctx.gpr[4] = (0u | 0u);
    ctx.gpr[16] = (ctx.gpr[4] + ctx.gpr[16]);
    goto L_08A17B4C;
L_08A17B4C:
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[7] = (ctx.gpr[29] + static_cast<std::uint32_t>(32));
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[8] = (ctx.gpr[29] + static_cast<std::uint32_t>(36));
    ctx.gpr[31] = (0x08A17B64u);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0131_entry, 131u, 369u, 0x08A11F78u>(ctx, &aot_mem) && ctx.pc == 0x08A17B64u) goto L_08A17B64;
    return;
L_08A17B64:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17D1C;
      }
      goto L_08A17B6C;
    }
L_08A17B6C:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[24]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17B84;
      }
      goto L_08A17B80;
    }
L_08A17B80:
    ctx.fpr[24] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    goto L_08A17B84;
L_08A17B84:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[26]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17B9C;
      }
      goto L_08A17B98;
    }
L_08A17B98:
    ctx.fpr[26] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    goto L_08A17B9C;
L_08A17B9C:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(29)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(832), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(836), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(840), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17BF4;
      }
      goto L_08A17BD0;
    }
L_08A17BD0:
    aot_mem.aot_store16(ctx.gpr[20] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(300), ctx.gpr[19]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1568)));
    ctx.gpr[31] = (0x08A17BE8u);
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A17BE8u) goto L_08A17BE8;
    return;
L_08A17BE8:
    { const std::uint32_t vfpu_address = ctx.gpr[17] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1564)));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A17BF4;
L_08A17BF4:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(31)));
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(36)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(16)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(832), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(20)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(836), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(24)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(840), std::bit_cast<std::uint32_t>(ctx.fpr[13]));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(296)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17C4C;
      }
      goto L_08A17C28;
    }
L_08A17C28:
    ctx.gpr[18] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1536)));
    aot_mem.aot_store16(ctx.gpr[19] + static_cast<std::uint32_t>(320), static_cast<std::uint16_t>(ctx.gpr[4]));
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(296), std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(300), ctx.gpr[20]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1540)));
    ctx.gpr[31] = (0x08A17C44u);
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0013_entry, 13u, 689u, 0x0883B8A4u>(ctx, &aot_mem) && ctx.pc == 0x08A17C44u) goto L_08A17C44;
    return;
L_08A17C44:
    { const std::uint32_t vfpu_address = ctx.gpr[17] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[18] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    goto L_08A17C4C;
L_08A17C4C:
    ctx.gpr[18] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(28)));
    ctx.gpr[17] = (aot_mem.aot_load8(ctx.gpr[16] + static_cast<std::uint32_t>(30)));
    ctx.fpr[22] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(32)));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(1612), ctx.gpr[17]);
    { const std::uint32_t vfpu_address = ctx.gpr[23] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1544)));
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(832));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    ctx.execute_vfpu_vdot_ct<28u, 0u, 0u, 3u>();
    ctx.gpr[4] = (ctx.vfpu_scalar_bits_ct<28u>());
    ctx.fpr[20] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1548)));
    ctx.gpr[17] = (ctx.gpr[29] + static_cast<std::uint32_t>(864));
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    ctx.gpr[31] = (0x08A17C98u);
    ctx.gpr[5] = (ctx.gpr[22] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 119u, 0x08A0D244u>(ctx, &aot_mem) && ctx.pc == 0x08A17C98u) goto L_08A17C98;
    return;
L_08A17C98:
    ctx.gpr[31] = (0x08A17CA0u);
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 121u, 0x08A0D278u>(ctx, &aot_mem) && ctx.pc == 0x08A17CA0u) goto L_08A17CA0;
    return;
L_08A17CA0:
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[0]));
    ctx.set_fpu_condition((ctx.fpr[20] <= ctx.fpr[12]));
    ctx.gpr[17] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1612)));
    if (!ctx.fpu_condition()) {
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
        goto L_08A17CB4;
    }
    goto L_08A17CB4;
L_08A17CB4:
    ctx.fpr[13] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[12]));
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1552)));
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[6] = (ctx.gpr[19] | 0u);
    ctx.gpr[7] = (ctx.gpr[18] | 0u);
    ctx.gpr[8] = (ctx.gpr[17] | 0u);
    ctx.gpr[31] = (0x08A17CD4u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[22]));
    if (rt.invoke_chained_direct<&recomp_unit_0024_entry, 24u, 137u, 0x088649D4u>(ctx, &aot_mem) && ctx.pc == 0x08A17CD4u) goto L_08A17CD4;
    return;
L_08A17CD4:
    ctx.gpr[31] = (0x08A17CDCu);
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0072_entry, 72u, 199u, 0x089257E4u>(ctx, &aot_mem) && ctx.pc == 0x08A17CDCu) goto L_08A17CDC;
    return;
L_08A17CDC:
    ctx.fpr[12] = ctx.fpr[0] / ctx.fpr[28];
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A17CF0u);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    goto L_08A14954;
L_08A17CF0:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    ctx.gpr[4] = (0u + static_cast<std::uint32_t>(-4097));
      if (branch_taken) {
          goto L_08A17D18;
      }
      goto L_08A17CF8;
    }
L_08A17CF8:
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[5] = (ctx.gpr[5] & ctx.gpr[4]);
    ctx.gpr[5] = (ctx.gpr[5] | 4096u);
    aot_mem.aot_store32(ctx.gpr[20] + static_cast<std::uint32_t>(68), ctx.gpr[5]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[5] & ctx.gpr[4]);
    ctx.gpr[4] = (ctx.gpr[4] | 4096u);
    aot_mem.aot_store32(ctx.gpr[19] + static_cast<std::uint32_t>(68), ctx.gpr[4]);
    goto L_08A17D18;
L_08A17D18:
    ctx.gpr[17] = (ctx.gpr[29] + static_cast<std::uint32_t>(832));
    goto L_08A17D1C;
L_08A17D1C:
    ctx.gpr[21] = (ctx.gpr[21] + static_cast<std::uint32_t>(1));
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[21]) < static_cast<std::int32_t>(ctx.gpr[30]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    ctx.gpr[16] = (ctx.gpr[16] + static_cast<std::uint32_t>(32));
      if (branch_taken) {
          goto L_08A17B4C;
      }
      goto L_08A17D2C;
    }
L_08A17D2C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 6u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    ctx.gpr[16] = (2229u << 16u);
    ctx.gpr[5] = (16968u << 16u);
    ctx.fpr[20] = std::bit_cast<float>(ctx.gpr[5]);
    ctx.gpr[17] = (ctx.gpr[29] + static_cast<std::uint32_t>(896));
    ctx.gpr[5] = (16800u << 16u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    ctx.fpr[22] = std::bit_cast<float>(ctx.gpr[5]);
      if (branch_taken) {
          goto L_08A17DD4;
      }
      goto L_08A17D5C;
    }
L_08A17D5C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17DD4;
      }
      goto L_08A17D78;
    }
L_08A17D78:
    ctx.gpr[31] = (0x08A17D80u);
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0103_entry, 103u, 841u, 0x089A38A0u>(ctx, &aot_mem) && ctx.pc == 0x08A17D80u) goto L_08A17D80;
    return;
L_08A17D80:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17DBC;
      }
      goto L_08A17D88;
    }
L_08A17D88:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(72)));
    ctx.gpr[4] = (ctx.gpr[4] & 2u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17DD4;
      }
      goto L_08A17D98;
    }
L_08A17D98:
    ctx.gpr[31] = (0x08A17DA0u);
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1544)));
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 121u, 0x08A0D278u>(ctx, &aot_mem) && ctx.pc == 0x08A17DA0u) goto L_08A17DA0;
    return;
L_08A17DA0:
    ctx.gpr[4] = (15139u << 16u);
    ctx.gpr[4] = (ctx.gpr[4] | 55051u);
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    ctx.set_fpu_condition((ctx.fpr[0] <= ctx.fpr[12]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17DD4;
      }
      goto L_08A17DBC;
    }
L_08A17DBC:
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    ctx.gpr[5] = (ctx.gpr[20] | 0u);
    ctx.gpr[31] = (0x08A17DCCu);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[26]));
    if (rt.invoke_chained_direct<&recomp_unit_0055_entry, 55u, 442u, 0x088E1D48u>(ctx, &aot_mem) && ctx.pc == 0x08A17DCCu) goto L_08A17DCC;
    return;
L_08A17DCC:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 32u, 0x08A18178u>(ctx, &aot_mem); return;
      }
      goto L_08A17DD4;
    }
L_08A17DD4:
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[5] = (0u | 197u);
    { const bool branch_taken = ctx.gpr[4] != ctx.gpr[5];
    // nop
      if (branch_taken) {
          goto L_08A17E40;
      }
      goto L_08A17DE4;
    }
L_08A17DE4:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 6u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17E40;
      }
      goto L_08A17E00;
    }
L_08A17E00:
    ctx.gpr[31] = (0x08A17E08u);
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0103_entry, 103u, 841u, 0x089A38A0u>(ctx, &aot_mem) && ctx.pc == 0x08A17E08u) goto L_08A17E08;
    return;
L_08A17E08:
    { const bool branch_taken = ctx.gpr[2] == 0u;
    ctx.gpr[4] = (16384u << 16u);
      if (branch_taken) {
          goto L_08A17E24;
      }
      goto L_08A17E10;
    }
L_08A17E10:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(72)));
    ctx.gpr[4] = (ctx.gpr[4] & 2u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17E40;
      }
      goto L_08A17E20;
    }
L_08A17E20:
    ctx.gpr[4] = (16384u << 16u);
    goto L_08A17E24;
L_08A17E24:
    ctx.fpr[12] = std::bit_cast<float>(ctx.gpr[4]);
    { const float fs = ctx.fpr[24]; const float ft = ctx.fpr[12]; if ((std::isinf(fs) && ft == 0.0f) || (std::isinf(ft) && fs == 0.0f)) ctx.fpr[12] = std::bit_cast<float>(0x7FC00000u); else ctx.fpr[12] = fs * ft; }
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[31] = (0x08A17E38u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0055_entry, 55u, 442u, 0x088E1D48u>(ctx, &aot_mem) && ctx.pc == 0x08A17E38u) goto L_08A17E38;
    return;
L_08A17E38:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 32u, 0x08A18178u>(ctx, &aot_mem); return;
      }
      goto L_08A17E40;
    }
L_08A17E40:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 8u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 3u, 0x08A18010u>(ctx, &aot_mem); return;
      }
      goto L_08A17E5C;
    }
L_08A17E5C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[19] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 512u);
    ctx.gpr[4] = (0u < ctx.gpr[4] ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 3u, 0x08A18010u>(ctx, &aot_mem); return;
      }
      goto L_08A17E74;
    }
L_08A17E74:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[20] + static_cast<std::uint32_t>(68)));
    ctx.gpr[4] = (ctx.gpr[4] & 14u);
    ctx.gpr[4] = (ctx.gpr[4] ^ 4u);
    ctx.gpr[4] = (ctx.gpr[4] < static_cast<std::uint32_t>(1) ? 1u : 0u);
    ctx.gpr[4] = (ctx.gpr[4] & 255u);
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 3u, 0x08A18010u>(ctx, &aot_mem); return;
      }
      goto L_08A17E90;
    }
L_08A17E90:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[19] + static_cast<std::uint32_t>(432)));
    { const bool branch_taken = ctx.gpr[4] == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17EC4;
      }
      goto L_08A17E9C;
    }
L_08A17E9C:
    ctx.set_fpu_condition((ctx.fpr[26] <= ctx.fpr[22]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17EC4;
      }
      goto L_08A17EAC;
    }
L_08A17EAC:
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[26]));
    ctx.gpr[31] = (0x08A17EBCu);
    ctx.gpr[5] = (0u | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0014_entry, 14u, 484u, 0x0883F1F4u>(ctx, &aot_mem) && ctx.pc == 0x08A17EBCu) goto L_08A17EBC;
    return;
L_08A17EBC:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 32u, 0x08A18178u>(ctx, &aot_mem); return;
      }
      goto L_08A17EC4;
    }
L_08A17EC4:
    ctx.gpr[4] = (aot_mem.aot_load8(ctx.gpr[19] + static_cast<std::uint32_t>(432)));
    ctx.gpr[4] = (static_cast<std::int32_t>(ctx.gpr[4]) < 20 ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[4] != 0u;
    // nop
      if (branch_taken) {
          (void)rt.invoke_chained_direct<&recomp_unit_0133_entry, 133u, 32u, 0x08A18178u>(ctx, &aot_mem); return;
      }
      goto L_08A17ED4;
    }
L_08A17ED4:
    ctx.gpr[4] = (0u + static_cast<std::uint32_t>(-2));
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(960), 0u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(964)));
    ctx.gpr[4] = (ctx.gpr[5] & ctx.gpr[4]);
    aot_mem.aot_store32(ctx.gpr[29] + static_cast<std::uint32_t>(964), ctx.gpr[4]);
    ctx.gpr[9] = (2230u << 16u);
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[9] + static_cast<std::uint32_t>(-8152)));
    ctx.gpr[6] = (static_cast<std::int32_t>(ctx.gpr[4]) < static_cast<std::int32_t>(ctx.gpr[6]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[6] == 0u;
    ctx.gpr[5] = (0u | 0u);
      if (branch_taken) {
          goto L_08A17F10;
      }
      goto L_08A17F00;
    }
L_08A17F00:
    ctx.gpr[4] = (ctx.gpr[4] << 2u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(-11604)));
    ctx.gpr[4] = (ctx.gpr[5] + ctx.gpr[4]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    goto L_08A17F10;
L_08A17F10:
    ctx.gpr[8] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(20)));
    ctx.gpr[8] = (ctx.gpr[8] + static_cast<std::uint32_t>(32));
    ctx.gpr[4] = (static_cast<std::uint32_t>(static_cast<std::int32_t>(static_cast<std::int16_t>(aot_mem.aot_load16(ctx.gpr[19] + static_cast<std::uint32_t>(88))))));
    ctx.gpr[6] = (aot_mem.aot_load32(ctx.gpr[9] + static_cast<std::uint32_t>(-8152)));
    ctx.gpr[6] = (static_cast<std::int32_t>(ctx.gpr[4]) < static_cast<std::int32_t>(ctx.gpr[6]) ? 1u : 0u);
    { const bool branch_taken = ctx.gpr[6] == 0u;
    ctx.gpr[5] = (0u | 0u);
      if (branch_taken) {
          goto L_08A17F3C;
      }
      goto L_08A17F2C;
    }
L_08A17F2C:
    ctx.gpr[4] = (ctx.gpr[4] << 2u);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[16] + static_cast<std::uint32_t>(-11604)));
    ctx.gpr[4] = (ctx.gpr[5] + ctx.gpr[4]);
    ctx.gpr[5] = (aot_mem.aot_load32(ctx.gpr[4] + static_cast<std::uint32_t>(0)));
    goto L_08A17F3C;
L_08A17F3C:
    ctx.gpr[4] = (aot_mem.aot_load32(ctx.gpr[5] + static_cast<std::uint32_t>(20)));
    ctx.gpr[4] = (ctx.gpr[4] + static_cast<std::uint32_t>(16));
    { const std::uint32_t vfpu_address = ctx.gpr[8] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<0u, 4u>(vfpu_value); }
    { const std::uint32_t vfpu_address = ctx.gpr[4] + static_cast<std::uint32_t>(0);
      float vfpu_value[4]{
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 0u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 4u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 8u)),
        std::bit_cast<float>(aot_mem.aot_load32(vfpu_address + 12u))};
      ctx.write_vfpu_vector_ct<1u, 4u>(vfpu_value); }
    { float vfpu_s[4]{}, vfpu_t[4]{}, vfpu_d[4]{};
      ctx.read_vfpu_vector_with_source_prefix_ct<0u, 3u, 0u>(vfpu_s);
      ctx.read_vfpu_vector_with_source_prefix_ct<1u, 3u, 1u>(vfpu_t);
      for (std::uint32_t i = 0; i < 3u; ++i) vfpu_d[i] = vfpu_s[i] - vfpu_t[i];
      ctx.write_vfpu_vector_with_destination_prefix_ct<0u, 3u>(vfpu_d); }
    ctx.gpr[16] = (ctx.gpr[29] + static_cast<std::uint32_t>(976));
    { float vfpu_value[4]{}; ctx.read_vfpu_vector_ct<0u, 4u>(vfpu_value);
      const std::uint32_t vfpu_address = ctx.gpr[16] + static_cast<std::uint32_t>(0);
      aot_mem.aot_store32(vfpu_address + 0u, std::bit_cast<std::uint32_t>(vfpu_value[0]));
      aot_mem.aot_store32(vfpu_address + 4u, std::bit_cast<std::uint32_t>(vfpu_value[1]));
      aot_mem.aot_store32(vfpu_address + 8u, std::bit_cast<std::uint32_t>(vfpu_value[2]));
      aot_mem.aot_store32(vfpu_address + 12u, std::bit_cast<std::uint32_t>(vfpu_value[3])); }
    ctx.gpr[18] = (ctx.gpr[29] + static_cast<std::uint32_t>(880));
    ctx.gpr[4] = (ctx.gpr[18] | 0u);
    ctx.gpr[5] = (ctx.gpr[19] | 0u);
    ctx.gpr[31] = (0x08A17F6Cu);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 118u, 0x08A0D21Cu>(ctx, &aot_mem) && ctx.pc == 0x08A17F6Cu) goto L_08A17F6C;
    return;
L_08A17F6C:
    ctx.gpr[4] = (ctx.gpr[16] | 0u);
    ctx.gpr[31] = (0x08A17F78u);
    ctx.gpr[5] = (ctx.gpr[18] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 122u, 0x08A0D290u>(ctx, &aot_mem) && ctx.pc == 0x08A17F78u) goto L_08A17F78;
    return;
L_08A17F78:
    ctx.gpr[31] = (0x08A17F80u);
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0189_entry, 189u, 355u, 0x08AF993Cu>(ctx, &aot_mem) && ctx.pc == 0x08A17F80u) goto L_08A17F80;
    return;
L_08A17F80:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[2] + static_cast<std::uint32_t>(8)));
    ctx.fpr[13] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(984)));
    ctx.set_fpu_condition((ctx.fpr[12] <= ctx.fpr[13]));
    // nop
    { const bool branch_taken = ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17FB0;
      }
      goto L_08A17F98;
    }
L_08A17F98:
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    ctx.gpr[31] = (0x08A17FA8u);
    ctx.gpr[5] = (0u | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0014_entry, 14u, 484u, 0x0883F1F4u>(ctx, &aot_mem) && ctx.pc == 0x08A17FA8u) goto L_08A17FA8;
    return;
L_08A17FA8:
    { const bool branch_taken = 0u == 0u;
    // nop
      if (branch_taken) {
          goto L_08A17FFC;
      }
      goto L_08A17FB0;
    }
L_08A17FB0:
    ctx.gpr[4] = (ctx.gpr[20] | 0u);
    ctx.gpr[31] = (0x08A17FBCu);
    ctx.gpr[5] = (ctx.gpr[17] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0128_entry, 128u, 511u, 0x08A06548u>(ctx, &aot_mem) && ctx.pc == 0x08A17FBCu) goto L_08A17FBC;
    return;
L_08A17FBC:
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    ctx.gpr[31] = (0x08A17FC8u);
    ctx.gpr[5] = (ctx.gpr[2] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0128_entry, 128u, 527u, 0x08A06A44u>(ctx, &aot_mem) && ctx.pc == 0x08A17FC8u) goto L_08A17FC8;
    return;
L_08A17FC8:
    ctx.gpr[4] = (ctx.gpr[29] + static_cast<std::uint32_t>(992));
    ctx.gpr[5] = (ctx.gpr[17] | 0u);
    ctx.gpr[31] = (0x08A17FD8u);
    ctx.gpr[6] = (ctx.gpr[16] | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0130_entry, 130u, 118u, 0x08A0D21Cu>(ctx, &aot_mem) && ctx.pc == 0x08A17FD8u) goto L_08A17FD8;
    return;
L_08A17FD8:
    ctx.fpr[12] = std::bit_cast<float>(aot_mem.aot_load32(ctx.gpr[29] + static_cast<std::uint32_t>(1000)));
    ctx.set_fpu_condition((ctx.fpr[12] < ctx.fpr[30]));
    // nop
    { const bool branch_taken = !ctx.fpu_condition();
    // nop
      if (branch_taken) {
          goto L_08A17FFC;
      }
      goto L_08A17FEC;
    }
L_08A17FEC:
    ctx.gpr[4] = (ctx.gpr[19] | 0u);
    ctx.fpr[12] = std::bit_cast<float>(std::bit_cast<std::uint32_t>(ctx.fpr[20]));
    ctx.gpr[31] = (0x08A17FFCu);
    ctx.gpr[5] = (0u | 0u);
    if (rt.invoke_chained_direct<&recomp_unit_0014_entry, 14u, 484u, 0x0883F1F4u>(ctx, &aot_mem) && ctx.pc == 0x08A17FFCu) goto L_08A17FFC;
    return;
L_08A17FFC:
    ctx.gpr[4] = (ctx.gpr[17] | 0u);
    ctx.pc = 0x08A18000u; return;
}

void recomp_unit_0132(Runtime &rt, AllegrexContext &ctx) {
    auto aot_mem = rt.memory().aot_fast_view();
    recomp_unit_0132_entry(rt, ctx, 0u, aot_mem);
}

void register_generated_unit_132(Runtime &runtime) {
    runtime.register_generated_unit(132u, 0x08A14000u, 16384u, &recomp_unit_0132, &recomp_unit_0132_entry);
    runtime.register_function(0x08A14000u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14010u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14020u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14038u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14050u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1405Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14068u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1408Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A140A0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A140B0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A140B8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A140C0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14100u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1414Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14154u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1415Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14160u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14178u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A141C0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1422Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14234u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1423Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14330u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A143E8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A143F4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14410u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14420u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14430u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14444u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14458u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14470u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14480u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14498u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A144B8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A144CCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A144D8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A144F0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14508u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14520u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14544u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1455Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14568u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14580u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14598u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A145BCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A145D8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A145E4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A145FCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14614u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14638u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14654u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14660u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14678u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14690u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A146B4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A146D0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A146ECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A146FCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1470Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14728u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14758u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14778u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A147D4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A147ECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1481Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14838u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14848u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14854u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1485Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1488Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A148B8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A148C4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14920u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14928u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14930u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14954u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A149A4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A149BCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14AF4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14B18u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14B24u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14B58u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14B88u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14B8Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14BA0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14BA4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14BECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14C00u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14C08u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14C10u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14C28u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14C44u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14C60u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14C68u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14C70u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14C90u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14C98u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14E10u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14E6Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14E90u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14E98u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14EA4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14F20u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14F50u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14F54u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14F64u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14F68u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14FB0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14FC4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14FE4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14FECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A14FF4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1500Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15028u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15208u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1522Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15234u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15240u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A152BCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A152ECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A152F0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15300u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15304u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1534Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1536Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15380u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15388u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15390u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1561Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15640u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1564Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15700u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15730u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15734u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15744u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15748u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15790u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A157B0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A157C4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A157E4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A157ECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A157F4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15824u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A158D4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1591Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15934u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15944u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A159ACu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A159BCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15A1Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15A30u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15A38u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15A98u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15AA0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15AACu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15AE0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15AE8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15B04u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15B14u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15B30u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15B4Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15B50u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15B60u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15B78u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15B94u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15BA4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15BC0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15BDCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15BE0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15BF0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15C08u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15C24u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15C34u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15C50u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15C5Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15C70u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15C8Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15C9Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15CB8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15CCCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15CE0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15CF4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15D08u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15D20u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15D3Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15D4Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15D68u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15D7Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15D90u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15DA4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15DB8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15DD0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15DECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E08u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E18u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E20u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E2Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E3Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E4Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E5Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E64u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E6Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E78u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E80u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15E90u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15EB4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15EC4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15EE0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15EF0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15F54u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15F60u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15F70u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15F7Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15FB4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15FBCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15FCCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15FD8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15FE0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A15FE8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16004u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16020u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16030u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16038u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16044u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16054u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16064u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16074u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1607Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16084u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16090u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16098u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A160A8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A160CCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A160DCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A160F8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16108u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16168u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16170u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16180u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1618Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A161C4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A161C8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A161D8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A161E4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A161ECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A161F4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16204u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16220u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16238u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16240u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1624Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16268u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16280u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16294u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A162B0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A162BCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A162C8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A162E0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A162F0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16300u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1631Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16338u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16354u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1635Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16368u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16378u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16388u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1639Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A163ACu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16488u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16498u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1649Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A164B4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A164CCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A164DCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A164ECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A164FCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1651Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16524u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16534u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1653Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16558u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16574u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1657Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16588u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16598u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A165A8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A165BCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A165C0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A165C8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A165E4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A165F0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A165FCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16610u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1662Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16648u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1665Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16660u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1667Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16698u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A166A4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A166BCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A166D8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A166F4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16700u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16714u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1672Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16734u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1675Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16770u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16778u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1677Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1678Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16794u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1679Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A167A4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A167ACu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A167B8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A167C0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A167DCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A167F4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16804u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1680Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16810u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16838u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16844u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16860u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1687Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16888u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A168A4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A168ACu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A168B4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A168BCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A168C8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A168D8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A168E0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16930u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16940u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1694Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16968u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16970u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1698Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16990u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A169ACu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A169C8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A169DCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A169ECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A169FCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16A30u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16A48u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16A58u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16A60u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16A6Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16A74u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16A94u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16AACu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16AC4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16B10u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16B2Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16B34u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16B70u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16B8Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16B94u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16BC8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16BE0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16BF0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16C2Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16C48u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16C58u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16C60u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16C70u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16C84u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16CA0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16CA8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16CC4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16CC8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16CD0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16CF0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16D0Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16D20u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16D30u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16D40u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16D74u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16D8Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16D9Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16DA4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16DA8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16DB0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16DBCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16DC4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16DE4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16DFCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16E14u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16E60u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16E7Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16E84u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16EC0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16EDCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16EECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16EFCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F0Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F30u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F38u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F44u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F4Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F58u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F5Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F68u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F70u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F84u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16F90u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16FB0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16FB8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16FE8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16FECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A16FFCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17004u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17018u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1704Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17064u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17074u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A170B0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A170CCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A170DCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A170E4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A170F4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17104u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1710Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17140u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17168u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17178u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17188u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17190u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17198u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A171B4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A171C8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A171D4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A171F0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17204u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17210u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17220u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17230u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A172B0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A172CCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A172D8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A172F0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17300u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17304u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17310u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17318u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17320u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17328u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17354u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17360u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17380u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17398u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A173A8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A173C0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A173D8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A173E0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A173F4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A173F8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1740Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17410u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17444u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1745Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1746Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A174A0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A174B8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A174C4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17530u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1754Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1755Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17564u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1757Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A175E8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17608u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17620u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17628u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1763Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17640u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17654u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17658u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1768Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A176A4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A176B4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A176E8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17700u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1770Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17778u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17794u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1779Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A177B0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A177B8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A177D8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A177E8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17800u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17830u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17838u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17850u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A178C4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A178E4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A178FCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17904u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17918u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A1791Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17930u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17934u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17968u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17980u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17990u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A179C4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A179DCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A179E8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17A54u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17A70u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17A78u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17A8Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17A94u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17AB4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17AC4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17ADCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17B0Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17B14u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17B24u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17B4Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17B64u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17B6Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17B80u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17B84u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17B98u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17B9Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17BD0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17BE8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17BF4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17C28u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17C44u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17C4Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17C98u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17CA0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17CB4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17CD4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17CDCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17CF0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17CF8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17D18u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17D1Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17D2Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17D5Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17D78u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17D80u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17D88u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17D98u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17DA0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17DBCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17DCCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17DD4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17DE4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E00u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E08u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E10u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E20u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E24u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E38u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E40u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E5Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E74u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E90u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17E9Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17EACu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17EBCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17EC4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17ED4u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17F00u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17F10u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17F2Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17F3Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17F6Cu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17F78u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17F80u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17F98u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17FA8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17FB0u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17FBCu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17FC8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17FD8u, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17FECu, &recomp_unit_0132, "recomp_unit_0132");
    runtime.register_function(0x08A17FFCu, &recomp_unit_0132, "recomp_unit_0132");
}
} // namespace psprecomp