[Bug rtl-optimization/21485] [4.0/4.1/4.2 Regression] BYTEmark numsort: codegen regression with -O3

[steven at gcc dot gnu dot org]

------- Comment #13 from steven at gcc dot gnu dot org  2006-01-09 22:23 -------
AMD64 timings for today's GCC 4.1 (20060109) branch:

flags used                               score
for compilation                          (avg. of 3, higher is better)
-O3                                      685.2
-O3 -fprofile-generate                   636
-O3 -fprofile-use                        785.6
-O3 -fno-tree-pre                        958.4
-O3 -fno-tree-pre -fprofile-generate     673.8
-O3 -fno-tree-pre -fprofile-use          964.2
-O3 -fno-tree-pre -fmove-loop-invariants 967.2


The problem really is that we present the register allocator with a harder
problem.  Tough, there is really not much that can be done about this other
than crippling PRE for SMALL_REGISTER_CLASSES machines, but I don't know if we
should want to do that...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21485