First of all, the code looks correct. Kudos.

Few notes though.

• No naked loops. Every loop implements an important algorithm, and therefore deserves a name. In your case, the loop

    for (int* i = p+1; i <= tmpStop; ++i)
  

  partitions the range. Lift it into a partition function. Not only it will make the code more readable, but also more testable. Now you can unit test partition.

• Most range algorithms are more naturally expressed in terms of a semi-open range (that is, the end does not belong to the range, but is one-beyond). With such contract you don't need to -1 anywhere, and you may use < instead of <=.

• I am not sure that std::vector is the best choice to represent the stack. STL conveniently offers a stack container.

• flip is likely inferior to std::swap.

• Stop using namespace std. It is a bad habit.

First of all, the code looks correct. Kudos.

Few notes though.

• No naked loops. Every loop implements an important algorithm, and therefore deserves a name. In your case, the loop

    for (int* i = p+1; i <= tmpStop; ++i)
  

  partitions the range. Lift it into a partition function. Not only it will make the code more readable, but also more testable. Now you can unit test partition. And more maintainable too: you may change the partitioning strategy and not touch the core of sorting.

• Most range algorithms are more naturally expressed in terms of a semi-open range (that is, the end does not belong to the range, but is one-beyond). With such contract you don't need to -1 anywhere, and you may use < instead of <=.

• I am not sure that std::vector is the best choice to represent the stack. STL conveniently offers a stack container.

• flip is likely inferior to std::swap.

• Stop using namespace std. It is a bad habit.