------- Additional Comments From rakdver at gcc dot gnu dot org 2005-01-30
11:59 -------
Patch #7 of http://gcc.gnu.org/ml/gcc-patches/2004-12/msg01381.html more or
less solves the problem (there is one small bit missing that does not affect
this testcase, but might be significant elsewhere).
The selection of ivs on x86 seems ok to me -- the cost of computing w - ivtmp.4
is the same as the cost of decrementing w, so this way is equivalent to the one
used in the original code (and would need less registers in case the original
value of w would be live across the loop).
--
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |patch
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19701
