This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
update_life_info stuff
- To: rth at cygnus dot com
- Subject: update_life_info stuff
- From: Jeffrey A Law <law at cygnus dot com>
- Date: Sun, 17 Oct 1999 14:36:26 -0600
- Cc: gcc at gcc dot gnu dot org
- Reply-To: law at cygnus dot com
We really need to fix the REG_BASIC_BLOCK, REG_LIVE_LENGTH, etc etc
stuff that broke when you revamped haifa's life stuff. That work has
also significantly slowed down compilation on the x86.
Building insn-attrtab.c -O2 with your change:
time in parse: 12.160000 (3%)
time in integration: 0.000000 (0%)
time in jump: 22.890000 (5%)
time in cse: 21.280000 (5%)
time in gcse: 19.370000 (5%)
time in loop: 1.940000 (0%)
time in cse2: 22.590000 (5%)
time in branch-prob: 0.000000 (0%)
time in flow: 7.580000 (2%)
time in combine: 11.460000 (3%)
time in regmove: 4.690000 (1%)
time in sched: 0.000000 (0%)
time in local-alloc: 8.400000 (2%)
time in global-alloc: 45.270000 (11%)
time in flow2: 6.170000 (1%)
time in peephole2: 5.990000 (1%)
time in sched2: 207.150000 (49%)
time in shorten-branch: 3.340000 (1%)
time in stack-reg: 0.490000 (0%)
time in final: 4.100000 (1%)
time in varconst: 0.030000 (0%)
time in symout: 0.000000 (0%)
time in dump: 0.000000 (0%)
time in gc: 3.410000 (1%)
Note the sched2 time.
compute_bb_for_insn accounts for half of the total compilation time via
the call to update_flow_info from sched.c
Backing out your sched.c changes the sched2 time drops to just 36 seconds.
jeff