Function neon_nf4_lut

csrc/cpu_ops.cpp:44–69

ARM NEON NF4 lookup table (16 float values indexed by 4-bit code)

// ARM NEON NF4 lookup table (16 float values indexed by 4-bit code)
static inline void neon_nf4_lut(float32x4_t lut[4]) {
    // Indices 0-15 map to NF4 dequantized values
    // Order: index 0 = -1.0, index 1 = -0.6962, ..., index 7 = 0.0, ..., index 15 = 1.0
    static const float nf4_values[16] = {
        -1.0f,
        -0.6961928009986877f,
        -0.5250730514526367f,
        -0.39491748809814453f,
        -0.28444138169288635f,
        -0.18477343022823334f,
        -0.09105003625154495f,
        0.0f,
        0.07958029955625534f,
        0.16093020141124725f,
        0.24611230194568634f,
        0.33791524171829224f,
        0.44070982933044434f,
        0.5626170039176941f,
        0.7229568362236023f,
        1.0f
    };
    lut[0] = vld1q_f32(nf4_values);
    lut[1] = vld1q_f32(nf4_values + 4);
    lut[2] = vld1q_f32(nf4_values + 8);
    lut[3] = vld1q_f32(nf4_values + 12);
}

// ARM NEON FP4 lookup table
static inline void neon_fp4_lut(float32x4_t lut[4]) {