[Bug rtl-optimization/19097] [4.1/4.2 regression] Quadratic behavior with many sets for the same register in VRP

[steven at gcc dot gnu dot org]

------- Comment #36 from steven at gcc dot gnu dot org  2006-02-05 21:37 -------
I think the easiest way to fix this is to limit the length of the EQUIVALENCE
chains somehow.  I've collected some numbers about the number of elements in
the EQUIVALENCE chains of GCC 2.7.2 (which is the version of GCC in SPEC2000):


frequency   # of elements
  20342          0
  43484          1
  13230          2
   1701          3
    456          4
    207          5
     81          6
     44          7
     24          8
     18          9
      4          10
      4          11
      4          12
      4          13
      2          14
      2          15
      2          16

So if we cut off the element lists at, say, 20 elements, we shouldn't be
missing a significant number of optimizations.

I haven't looked at how such a cut-off should be implemented.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19097