pragma options "--bnd-unroll-amnt 6 --bnd-inbits 3 ";

int[16] transpose(int[16] m){
  int x, y;
  for(x = 0; x < 4; x++)
    for(y = 0; y ≤ x; y++)
      m[4*x+y] = m[4*y+x];
  return m;
}

generator int[4] shufps(int[4] xmm1, int[4] xmm2, bit[8] imm8){ /* automatically rewritten */
  int[4] ret;

  ret[0] = xmm1[(int)imm8[0::2]];
  ret[1] = xmm1[(int)imm8[2::2]];
  ret[2] = xmm2[(int)imm8[4::2]];
  ret[3] = xmm2[(int)imm8[6::2]];

  return ret;
}

int[16] sse_transpose(int[16] m) implements transpose {
  int[16] s = 0;
  int[16] t = 0;

  s[??::4] = shufps(m[??::4], m[??::4], ??);  
  s[??::4] = shufps(m[??::4], m[??::4], ??); 
  s[??::4] = shufps(m[??::4], m[??::4], ??); 
  s[??::4] = shufps(m[??::4], m[??::4], ??); 
	
  t[??::4] = shufps(s[??::4], s[??::4], ??);
  t[??::4] = shufps(s[??::4], s[??::4], ??);
  t[??::4] = shufps(s[??::4], s[??::4], ??);
  t[??::4] = shufps(s[??::4], s[??::4], ??);

  return t;
}