BUG: fixed regression in gdcmImageCodec.cxx

[issakomi]

S. post https://sourceforge.net/p/gdcm/bugs/554/

---

An unexpected behavior was introduced in commit 4ecd77a, causing pixel values being all greater or equal to zero on DICOM images where negative values are expected. When trying to solve the following cppcheck issues:

shiftNegativeLHS,GDCM/Source/MediaStorageAndFileFormat/gdcmImageCodec.cxx:465,portability,Shifting a negative value is technically undefined behaviour
shiftNegativeLHS,GDCM/Source/MediaStorageAndFileFormat/gdcmImageCodec.cxx:521,portability,Shifting a negative value is technically undefined behaviour

The changes in gdcmImageCodec.cxx:465 and gdcmImageCodec.cxx:521 implies a potential difference in the sign of nmask, depending on the specific values of PF.GetBitsAllocated() and PF.GetBitsStored(). If the right shift operation in the modified version produces a non-negative value, the result will be positive, whereas the result in the non modified version will always be negative due to the initial assignment of 0x8000 as a signed integer.

---

I could reproduce the issue:

#include <iostream>
#include <cstdint>
#include <vector>

int main()
{
  const std::vector<unsigned short> a{ 7, 6, 5, 4, 3, 2, 1, 0 };

  std::cout << "Old:" << '\n';
  for (size_t x = 0; x < a.size(); ++x)
  {
    //////////////////////////////////////////////
    //
    //
    int16_t nmask = (int16_t)0x8000;
    nmask = (int16_t)(nmask >> a.at(x));
    //
    //
    //////////////////////////////////////////////
    std::cout << nmask << '\n';
  }

  std::cout << "New:" << '\n';
  for (size_t x = 0; x < a.size(); ++x)
  {
    //////////////////////////////////////////////
    //
    //
    int16_t nmask = (int16_t)(0x8000U >> a.at(x));
    //
    //
    //////////////////////////////////////////////
    std::cout << nmask << '\n';
  }

  return 0;
}

Old:
-256
-512
-1024
-2048
-4096
-8192
-16384
-32768
New:
256
512
1024
2048
4096
8192
16384
-32768

[malaterre]

Thanks !