Function float_to_bf16

csrc/cpu_ops.h:139–144

void quantize_cpu_fp16(float* code, fp16_t* A, float* absmax, unsigned char* out, long long blocksize, long long n);

static inline bf16_t float_to_bf16(float x) {
    uint32_t bits;
    std::memcpy(&bits, &x, 4);
    uint32_t r = bits + 0x7FFF + ((bits >> 16) & 1);
    return bf16_t{static_cast<uint16_t>(r >> 16)};
}

static float bf16_to_float(uint16_t bf16) {
    uint32_t bits = (uint32_t)bf16 << 16;