BFieldCache Change the code so as to emit more clang like assembly for gcc

BFieldCache Change the code so as to emit more clang like assembly for gcc

see [https://godbolt.org/z/zK145fPbe]

ping @amete , this seems to improve things a bit.

But is the reason I asked if we had something with "micro-arch" for vtune.

This is usually called every time the propagators needs to access the magnetic field and is hard to accurately see if was an issue or not.

callgrind shows less branches and less mis-prediction (but callgrind emulates an old processor here). clang assumes the branch will be hard to predict and emit assembly for that assumption, gcc not ...

In some sense for this (no just this bit but the full chain) would be good to get some info on if there is something to improve ...

MagneticField/MagFieldElements/MagFieldElements/BFieldCache.icc

@@ -57,7 +57,11 @@ BFieldCache::inside(double z, double r, double phi) const
   if (phi < m_phimin) {
     phi += 2.0 * M_PI;
   }
-  return (z <= m_zmax && z >= m_zmin && r <= m_rmax && r >= m_rmin &&
-          phi <= m_phimax && phi >= m_phimin);
+  // clang emits code that is more similar to using the "bit AND"
+  // rather than "logical AND" for this kind of check.
+  // Basically get rid of jmp instructions with calculating
+  // a bit more
+  return ((z <= m_zmax) & (z >= m_zmin) & (r <= m_rmax) & (r >= m_rmin) &
+          (phi <= m_phimax) & (phi >= m_phimin));
 }