[Bug c/15239] New: suboptimal mult-by-const expansion cost limit

["l_belev at yahoo dot com" <gcc-bugzilla at gcc dot gnu dot org>]

Look in the file gcc/expmed.c, function expand_mult(). The code 
      mult_cost = rtx_cost (gen_rtx_MULT (mode, op0, op1), SET); 
      mult_cost = MIN (12 * add_cost, mult_cost); 
calculates the initial cost limit before the recursive search for 
a good expansion is invoked. 
 
The problem is that the limiting of the cost with the cost of 12 adds (the 
second line) is arbitrary and for some architectures (like pentium4) where 
the multiplication is far more expensive than 12 adds, the generated code is 
likely to be suboptimal as execution speed. 
It would be better if this limit is somehow platform and/or optimizing-level 
dependent (i.e. don't limit artificially the expansion for -O3 or higher). 
Intel's compiler (icc) doesn't limit the expansion in such a way even 
though for some constants it generates rather long sequences of 
adds/subs/leas/shls (instead of an imul).

-- 
           Summary: suboptimal mult-by-const expansion cost limit
           Product: gcc
           Version: 3.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: l_belev at yahoo dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: *
  GCC host triplet: *
GCC target triplet: *


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15239