------- Additional Comments From law at redhat dot com 2004-12-28 23:31 ------- Yea, the code I've got lying around here to undo unprofitable const/copy propagations definitely helps this code. And better yet, we don't need the full-blown version (which is queued for 4.1). We can get by with the very simplistic version. For the testcase in comment #8 we generate the following .optimized dump:
# BLOCK 0 # PRED: ENTRY [100.0%] (fallthru,exec) <L4>:; # SUCC: 1 [100.0%] (fallthru,exec) # BLOCK 1 # PRED: 0 [100.0%] (fallthru,exec) 1 [89.0%] (dfs_back,false,exec) <L0>:; temp1 = temp2 * xlvj_ - temp3 * zlvj_; temp3 = temp2 * zlvj_ + temp3 * xlvj_; temp2 = temp1; if (temp3 != 0) goto <L2>; else goto <L0>; # SUCC: 2 [11.0%] (loop_exit,true,exec) 1 [89.0%] (dfs_back,false,exec) # BLOCK 2 # PRED: 1 [11.0%] (loop_exit,true,exec) <L2>:; return yv1j_ * yv2j_; # SUCC: EXIT [100.0%] } -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19038
