https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91895
Bug ID: 91895
Summary: Compile the code with -O1 or -O2 is slower than with
-O3 and -Os
Product: gcc
Version: 4.6.4
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: hehaochen at hotmail dot com
Target Milestone: ---
Compiling the following code in gcc (GCC) 4.6.4:
# time gcc -O1 test.cpp
real 0m18.180s
user 0m16.368s
sys 0m1.812s
# time gcc -O2 test.cpp
real 0m40.856s
user 0m38.808s
sys 0m2.047s
# time gcc -O3 test.cpp
real 0m9.954s
user 0m9.115s
sys 0m0.840s
Interestingly, we get:
# time gcc -Os test.cpp
real 0m0.051s
user 0m0.034s
sys 0m0.017s
According to
https://gcc.gnu.org/onlinedocs/gcc-9.2.0/gcc/Optimize-Options.html#Optimize-Options
-O3 implies -O1 and -O2, so -O3 should be the slowest.
--------------------------------------------
int a, b, c, d, e;
int foo() {
int x = 0;
for (a = 0; a < 17; a++)
for (b = 0; b < 17; b++)
for (c = 0; c < 17; c++)
// change 17 to 18 or larger, the problem "solved"
for (d = 0; d < 17; d++)
x++;
}
int main ()
{
foo();
}
--------------------------------------------
Another interesting thing is, if you change any iteration count number larger,
the problem vanished. If you change the iteration count number smaller, the
slow down get relieved.
The compile time hog lies in "tree VRP", "tree reassociation"
--------------------------------------------
+++ slow +++
--------------------------------------------
time gcc -O1 -ftime-report test.cpp
Execution times (seconds)
name lookup : 0.02 ( 0%) usr 0.03 ( 2%) sys 0.04 ( 0%) wall
83 kB ( 0%) ggc
tree CFG cleanup : 0.03 ( 0%) usr 0.00 ( 0%) sys 0.03 ( 0%) wall
6 kB ( 0%) ggc
tree SSA rewrite : 0.05 ( 0%) usr 0.00 ( 0%) sys 0.05 ( 0%) wall
2 kB ( 0%) ggc
tree SSA other : 0.01 ( 0%) usr 0.00 ( 0%) sys 0.00 ( 0%) wall
0 kB ( 0%) ggc
tree SSA incremental : 0.50 ( 3%) usr 0.38 (21%) sys 0.85 ( 5%) wall
6 kB ( 0%) ggc
tree operand scan : 0.32 ( 2%) usr 0.53 (30%) sys 0.79 ( 4%) wall
9894 kB (19%) ggc
dominator optimization: 0.19 ( 1%) usr 0.19 (11%) sys 0.40 ( 2%) wall
5 kB ( 0%) ggc
tree CCP : 0.00 ( 0%) usr 0.00 ( 0%) sys 0.01 ( 0%) wall
1 kB ( 0%) ggc
tree reassociation : 14.78 (90%) usr 0.03 ( 2%) sys 14.82 (82%) wall
2 kB ( 0%) ggc
tree aggressive DCE : 0.02 ( 0%) usr 0.00 ( 0%) sys 0.01 ( 0%) wall
0 kB ( 0%) ggc
complete unrolling : 0.41 ( 3%) usr 0.60 (34%) sys 1.09 ( 6%) wall
41245 kB (78%) ggc
tree rename SSA copies: 0.00 ( 0%) usr 0.00 ( 0%) sys 0.01 ( 0%) wall
0 kB ( 0%) ggc
out of ssa : 0.00 ( 0%) usr 0.01 ( 1%) sys 0.01 ( 0%) wall
0 kB ( 0%) ggc
rest of compilation : 0.01 ( 0%) usr 0.01 ( 1%) sys 0.01 ( 0%) wall
5 kB ( 0%) ggc
repair loop structures: 0.01 ( 0%) usr 0.00 ( 0%) sys 0.00 ( 0%) wall
1 kB ( 0%) ggc
TOTAL : 16.35 1.78 18.13
52963 kB
real 0m18.180s
user 0m16.368s
sys 0m1.812s
--------------------------------------------
+++ VERY SLOW +++
--------------------------------------------
time gcc -O2 -ftime-report test.cpp
Execution times (seconds)
name lookup : 0.02 ( 0%) usr 0.02 ( 1%) sys 0.05 ( 0%) wall
83 kB ( 0%) ggc
tree CFG cleanup : 0.02 ( 0%) usr 0.00 ( 0%) sys 0.03 ( 0%) wall
6 kB ( 0%) ggc
tree VRP : 22.39 (58%) usr 0.60 (30%) sys 22.83 (56%) wall
4513 kB ( 8%) ggc
tree SSA rewrite : 0.04 ( 0%) usr 0.00 ( 0%) sys 0.04 ( 0%) wall
2 kB ( 0%) ggc
tree SSA incremental : 0.51 ( 1%) usr 0.32 (16%) sys 1.03 ( 3%) wall
6 kB ( 0%) ggc
tree operand scan : 0.58 ( 1%) usr 0.34 (17%) sys 0.94 ( 2%) wall
9894 kB (17%) ggc
dominator optimization: 0.01 ( 0%) usr 0.00 ( 0%) sys 0.01 ( 0%) wall
5 kB ( 0%) ggc
tree reassociation : 14.77 (38%) usr 0.02 ( 1%) sys 14.80 (36%) wall
2 kB ( 0%) ggc
tree aggressive DCE : 0.01 ( 0%) usr 0.00 ( 0%) sys 0.02 ( 0%) wall
7 kB ( 0%) ggc
complete unrolling : 0.40 ( 1%) usr 0.69 (34%) sys 1.03 ( 3%) wall
41248 kB (72%) ggc
tree rename SSA copies: 0.01 ( 0%) usr 0.00 ( 0%) sys 0.00 ( 0%) wall
0 kB ( 0%) ggc
out of ssa : 0.00 ( 0%) usr 0.01 ( 0%) sys 0.01 ( 0%) wall
0 kB ( 0%) ggc
integrated RA : 0.00 ( 0%) usr 0.01 ( 0%) sys 0.00 ( 0%) wall
3 kB ( 0%) ggc
rest of compilation : 0.01 ( 0%) usr 0.00 ( 0%) sys 0.01 ( 0%) wall
5 kB ( 0%) ggc
repair loop structures: 0.01 ( 0%) usr 0.00 ( 0%) sys 0.00 ( 0%) wall
1 kB ( 0%) ggc
TOTAL : 38.78 2.02 40.81
57496 kB
real 0m40.856s
user 0m38.808s
sys 0m2.047s
--------------------------------------------
+++ faster than O1/O2, but still slow +++
--------------------------------------------
time gcc -O3 -ftime-report test.cpp
Execution times (seconds)
ipa cp : 0.00 ( 0%) usr 0.01 ( 1%) sys 0.00 ( 0%) wall
0 kB ( 0%) ggc
name lookup : 0.02 ( 0%) usr 0.03 ( 4%) sys 0.01 ( 0%) wall
83 kB ( 0%) ggc
tree CFG cleanup : 0.02 ( 0%) usr 0.00 ( 0%) sys 0.01 ( 0%) wall
5 kB ( 0%) ggc
tree VRP : 1.34 (15%) usr 0.07 ( 9%) sys 1.41 (14%) wall
929 kB ( 3%) ggc
tree SSA rewrite : 0.02 ( 0%) usr 0.00 ( 0%) sys 0.02 ( 0%) wall
2 kB ( 0%) ggc
tree SSA incremental : 0.26 ( 3%) usr 0.22 (27%) sys 0.50 ( 5%) wall
6 kB ( 0%) ggc
tree operand scan : 0.19 ( 2%) usr 0.18 (22%) sys 0.31 ( 3%) wall
4966 kB (17%) ggc
dominator optimization: 0.00 ( 0%) usr 0.00 ( 0%) sys 0.01 ( 0%) wall
5 kB ( 0%) ggc
tree reassociation : 7.03 (77%) usr 0.02 ( 2%) sys 7.04 (71%) wall
2 kB ( 0%) ggc
complete unrolling : 0.21 ( 2%) usr 0.27 (33%) sys 0.52 ( 5%) wall
20656 kB (73%) ggc
out of ssa : 0.00 ( 0%) usr 0.01 ( 1%) sys 0.01 ( 0%) wall
0 kB ( 0%) ggc
rest of compilation : 0.00 ( 0%) usr 0.00 ( 0%) sys 0.01 ( 0%) wall
5 kB ( 0%) ggc
unaccounted todo : 0.01 ( 0%) usr 0.00 ( 0%) sys 0.01 ( 0%) wall
0 kB ( 0%) ggc
repair loop structures: 0.00 ( 0%) usr 0.00 ( 0%) sys 0.01 ( 0%) wall
1 kB ( 0%) ggc
TOTAL : 9.10 0.81 9.91
28392 kB
real 0m9.954s
user 0m9.115s
sys 0m0.840s
