cbuffer cbuffer_a : register(b0, space0) {
  uint4 a[4];
};

matrix<float16_t, 2, 4> a_load_1(uint offset) {
  const uint scalar_offset = ((offset + 0u)) / 4;
  uint4 ubo_load_1 = a[scalar_offset / 4];
  uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy);
  vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF));
  vector<float16_t, 2> ubo_load_yw = vector<float16_t, 2>(f16tof32(ubo_load >> 16));
  const uint scalar_offset_1 = ((offset + 8u)) / 4;
  uint4 ubo_load_3 = a[scalar_offset_1 / 4];
  uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy);
  vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF));
  vector<float16_t, 2> ubo_load_2_yw = vector<float16_t, 2>(f16tof32(ubo_load_2 >> 16));
  return matrix<float16_t, 2, 4>(vector<float16_t, 4>(ubo_load_xz[0], ubo_load_yw[0], ubo_load_xz[1], ubo_load_yw[1]), vector<float16_t, 4>(ubo_load_2_xz[0], ubo_load_2_yw[0], ubo_load_2_xz[1], ubo_load_2_yw[1]));
}

typedef matrix<float16_t, 2, 4> a_load_ret[4];
a_load_ret a_load(uint offset) {
  matrix<float16_t, 2, 4> arr[4] = (matrix<float16_t, 2, 4>[4])0;
  {
    for(uint i = 0u; (i < 4u); i = (i + 1u)) {
      arr[i] = a_load_1((offset + (i * 16u)));
    }
  }
  return arr;
}

[numthreads(1, 1, 1)]
void f() {
  const matrix<float16_t, 2, 4> l_a[4] = a_load(0u);
  const matrix<float16_t, 2, 4> l_a_i = a_load_1(32u);
  uint2 ubo_load_4 = a[2].zw;
  vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF));
  vector<float16_t, 2> ubo_load_4_yw = vector<float16_t, 2>(f16tof32(ubo_load_4 >> 16));
  const vector<float16_t, 4> l_a_i_i = vector<float16_t, 4>(ubo_load_4_xz[0], ubo_load_4_yw[0], ubo_load_4_xz[1], ubo_load_4_yw[1]);
  return;
}