program Ural1528;

var
  p:array[1..10000]of qword;
  q,u,e:array[1..5000]of qword;
  ans:array[1..5000]of qword;
  m:qword;
  a,n,max,x,y:longint;

procedure qsort(l,r:longint);

var
  i,j:longint;
  mid,swap:qword;

begin
  i:=l;j:=r;mid:=u[(i+j)div 2];
  repeat
    while u[i]<mid do inc(i);
    while u[j]>mid do dec(j);
    if i<=j then
    begin
      swap:=u[i];u[i]:=u[j];u[j]:=swap;
      swap:=q[i];q[i]:=q[j];q[j]:=swap;
      swap:=e[i];e[i]:=e[j];e[j]:=swap;
      inc(i);dec(j);
    end;
  until i>j;
  if i<r then qsort(i,r);
  if l<j then qsort(l,j);
end;

procedure readdata;

begin
  n:=0;max:=0;
  repeat
    readln(x,y);
    if (x=0)and(y=0) then exit;
    inc(n);inc(max);
    if x>=y then begin
      ans[n]:=0;
      dec(max);
    end else begin
      q[max]:=x;u[max]:=y;e[max]:=n;
    end;
  until (x=0)and(y=0);
end;

function gcd(o,z:qword):qword;

begin
  if z=0 then exit(o)
  else exit(gcd(z,o-o div z*z));
end;

procedure main;

var
  i,j,mu:longint;
  o:qword;

begin
  i:=1;
  j:=1;
  while i<=max do
    begin
      m:=1;
      mu:=0;
      o:=1;
      while (i<=max)and(trunc(ln(m)/ln(10))+trunc(ln(u[i])/ln(10))-trunc(ln(o)/ln(10))<=18)do
        begin
          m:=m div o;
          m:=m*u[i];
          if mu<q[i] then mu:=q[i];
          inc(i);
          if i<=max then
          if u[i]>m then o:=gcd(u[i],m)
          else o:=gcd(m,u[i]);
        end;
      p[1]:=1;
      for a:=2 to mu do
        p[a]:=p[a-1]*a mod m;
      for a:=j to i-1 do
        ans[e[a]]:=p[q[a]]mod u[a];
      j:=i;
    end;
end;

procedure writedata;

begin
  for a:=1 to n do
    writeln(ans[a]);
end;

begin
  readdata;
  qsort(1,max);
  main;
  writedata;
end.

Why WA5?
I already get AC.But I want to make it faster, so Where is wrong?