[SCEV] Increase FoldID bits size cover common cases.

[fhahn]

Summary

This should fix the regressions introduced by a53d940cee6f, while
retaining support for arbitrary SCEV expressions.

Alternative to D144335.

Compile-time impact:

• NewPM-O3: -0.11%
• NewPM-ReleaseThinLTO: -0.10%

*NewPM-ReleaseLTO-g: -0.08%

https://llvm-compile-time-tracker.com/compare.php?from=df016a9525e5722dfd1e1e1632cec3ed7b33bc8a&to=c1c64de4a973bcecaddbc4038a539234eb39413b&stat=instructions:u

[nikic]

LGTM, this looks like the better backport candidate.

[vitalybuka]

llvm/include/llvm/Analysis/ScalarEvolution.h

1300

it's precisely 5 for 64 and 3 for 32

https://github.com/llvm/llvm-project/commit/f1124364336b8bfa341e8a39caba1170205cb6c3

its evaluates to the same improvement.
Either way prevent SmallVector from using heap,

1300

https://llvm-compile-time-tracker.com/compare.php?from=a53d940cee6f281ef1a20d4f0fb39b23b4e98614&to=f1124364336b8bfa341e8a39caba1170205cb6c3&stat=instructions:u

[vitalybuka]

Would you like to submit this one and back-port? it's just missed rc3

[fhahn]

Just landed with adjusted size to 5. I filed https://github.com/llvm/llvm-project/issues/60978 for the back-port.

llvm/include/llvm/Analysis/ScalarEvolution.h

bool loopHasNoAbnormalExits(const Loop *L) {
     return getLoopProperties(L).HasNoAbnormalExits;
   }
 
   /// Return true if this loop is finite by assumption.  That is,
   /// to be infinite, it must also be undefined.
   bool loopIsFiniteByAssumption(const Loop *L);
 
   class FoldID {
-    SmallVector<unsigned, 4> Bits;
+    SmallVector<unsigned, 5> Bits;
 
   public:
     void addInteger(unsigned long I) {
       if (sizeof(long) == sizeof(int))
         addInteger(unsigned(I));
       else if (sizeof(long) == sizeof(long long))
         addInteger((unsigned long long)I);
       else