# Compiler Bug

## Convlusion Cl Codes
`vstore2` and `vload2` function

__kernel void default_function_kernel0(__global float* restrict A, __global float* restrict W, __global float* restrict B) {
  float B_local[64];
  __local float Apad_shared[8];
  __local float W_shared[128];
  float Apad_shared_local[8];
  float W_shared_local[8];
  for (int ff_c_init = 0; ff_c_init < 4; ++ff_c_init) {
    for (int nn_c_init = 0; nn_c_init < 4; ++nn_c_init) {
      B_local[(((ff_c_init * 4) + nn_c_init))] = 0.000000e+00f;
      B_local[((((ff_c_init * 4) + nn_c_init) + 32))] = 0.000000e+00f;
      B_local[((((ff_c_init * 4) + nn_c_init) + 16))] = 0.000000e+00f;
      B_local[((((ff_c_init * 4) + nn_c_init) + 48))] = 0.000000e+00f;
    }
  }
  for (int rc_outer = 0; rc_outer < 2; ++rc_outer) {
    for (int ry = 0; ry < 3; ++ry) {
      for (int rx = 0; rx < 3; ++rx) {
        barrier(CLK_LOCAL_MEM_FENCE);
        if (((int)get_local_id(0)) < 1) {
          Apad_shared[((((int)get_local_id(1)) + ((int)get_local_id(0))))] = (((((1 <= ((((int)get_group_id(2)) / 14) + ry)) && (((((int)get_group_id(2)) / 14) + ry) < 15)) && (1 <= (rx + (((int)get_group_id(2)) % 14)))) && ((rx + (((int)get_group_id(2)) % 14)) < 15)) ? A[((((((((ry * 224) + (((int)get_group_id(2)) * 16)) + (rx * 16)) + (rc_outer * 8)) + ((int)get_local_id(1))) + ((int)get_local_id(0))) - 240))] : 0.000000e+00f);
        }
        vstore2(vload2(0, W + (((((ry * 768) + (rx * 256)) + (rc_outer * 128)) + (((int)get_local_id(1)) * 16)) + (((int)get_local_id(0)) * 2))), 0, W_shared + ((((int)get_local_id(1)) * 16) + (((int)get_local_id(0)) * 2)));
        barrier(CLK_LOCAL_MEM_FENCE);
        for (int rc_inner = 0; rc_inner < 8; ++rc_inner) {
          for (int ax3 = 0; ax3 < 4; ++ax3) {
            if (((((int)get_local_id(0)) * 4) + ax3) < 1) {
              Apad_shared_local[(ax3)] = Apad_shared[((((((int)get_local_id(0)) * 4) + rc_inner) + ax3))];
            }
          }
          for (int ax31 = 0; ax31 < 4; ++ax31) {
            if (((((int)get_local_id(1)) * 4) + ax31) < 16) {
              W_shared_local[(ax31)] = W_shared[((((rc_inner * 16) + (((int)get_local_id(1)) * 4)) + ax31))];
            }
          }
          for (int ff_c = 0; ff_c < 4; ++ff_c) {
            for (int nn_c = 0; nn_c < 4; ++nn_c) {
              if (((((int)get_local_id(1)) * 4) + ff_c) < 16) {
                if (((((int)get_local_id(0)) * 4) + nn_c) < 1) {
                  B_local[(((ff_c * 4) + nn_c))] = (B_local[(((ff_c * 4) + nn_c))] + (Apad_shared_local[(nn_c)] * W_shared_local[(ff_c)]));
                }
              }
            }
          }
        }
      }
    }
  }
  for (int ff_inner_inner_inner = 0; ff_inner_inner_inner < 4; ++ff_inner_inner_inner) {
    for (int nn_inner_inner_inner = 0; nn_inner_inner_inner < 4; ++nn_inner_inner_inner) {
      if (((((int)get_local_id(1)) * 4) + ff_inner_inner_inner) < 16) {
        if (((((int)get_local_id(0)) * 4) + nn_inner_inner_inner) < 1) {
          B[((((((((int)get_group_id(2)) * 16) + (((int)get_local_id(1)) * 4)) + (((int)get_local_id(0)) * 4)) + ff_inner_inner_inner) + nn_inner_inner_inner))] = B_local[(((ff_inner_inner_inner * 4) + nn_inner_inner_inner))];
        }
      }
    }
  }
}