Debian Codegraph @

docs/internal_team_notes/base_perf_job_gen/results

raw
1 RESULTS
2 ============================================================================
3 cd $WK ; cp /var/tmp/ma0/DEFS/metabuilder.def .
4 time Base/bin/gcc-7.3.0/release/perf_job_gen ./metabuilder.def
5 submitted 4553 out of 4559
6
7 real 0m2.22s
8 user 0m1.82s
9 sys 0m0.39s
10
11
12
13 strace -c Base/bin/gcc-7.3.0/release/perf_job_gen ./metabuilder.def
14 submitted 4553 out of 4559
15 % time seconds usecs/call calls errors syscall
16 ------ ----------- ----------- --------- --------- ----------------
17 32.97 0.098267 1 163674 read
18 15.05 0.044856 1 41404 write
19 13.59 0.040501 1 73919 lseek
20 11.46 0.034150 3 10760 17 open
21 10.51 0.031337 3 10742 close
22 9.67 0.028827 1 31948 12134 stat
23 2.40 0.007139 2 4553 chmod
24 1.88 0.005617 1 4555 munmap
25 1.41 0.004200 1 4578 mmap
26 0.89 0.002646 1 4563 fstat
27 0.11 0.000334 1 243 brk
28 0.06 0.000193 4 51 writev
29 0.00 0.000000 0 2 1 lstat
30 0.00 0.000000 0 14 mprotect
31 0.00 0.000000 0 3 rt_sigaction
32 0.00 0.000000 0 4 rt_sigprocmask
33 0.00 0.000000 0 1 1 access
34 0.00 0.000000 0 1 execve
35 0.00 0.000000 0 1 uname
36 0.00 0.000000 0 1 unlink
37 0.00 0.000000 0 1 getrlimit
38 0.00 0.000000 0 1 arch_prctl
39 0.00 0.000000 0 2 futex
40 0.00 0.000000 0 1 set_tid_address
41 0.00 0.000000 0 1 set_robust_list
42 ------ ----------- ----------- --------- --------- ----------------
43 100.00 0.298067 351023 12153 total
44
45
46 # baseline
47 perf stat -r 10 -d Base/bin/gcc-7.3.0/release/perf_job_gen ./metabuilder.def
48 Performance counter stats for 'Base/bin/gcc-7.3.0/release/perf_job_gen ./metabuilder.def' (10 runs):
49
50 2637.099995 task-clock (msec) # 0.996 CPUs utilized ( +- 2.31% )
51 218 context-switches # 0.083 K/sec ( +- 74.32% )
52 0 cpu-migrations # 0.000 K/sec ( +-100.00% )
53 9,658 page-faults # 0.004 M/sec ( +- 0.00% )
54 10,122,555,334 cycles # 3.839 GHz ( +- 2.30% ) (49.94%)
55 <not supported> stalled-cycles-frontend
56 <not supported> stalled-cycles-backend
57 12,804,414,325 instructions # 1.26 insns per cycle ( +- 0.10% ) (62.47%)
58 3,209,766,689 branches # 1217.158 M/sec ( +- 0.07% ) (62.49%)
59 121,179,840 branch-misses # 3.78% of all branches ( +- 0.47% ) (62.53%)
60 3,061,094,579 L1-dcache-loads # 1160.781 M/sec ( +- 0.10% ) (62.23%)
61 274,551,085 L1-dcache-load-misses # 8.97% of all L1-dcache hits ( +- 0.44% ) (25.03%)
62 87,217,818 LLC-loads # 33.073 M/sec ( +- 1.71% ) (25.04%)
63 1,055,192 LLC-load-misses # 1.21% of all LL-cache hits ( +- 5.56% ) (37.50%)
64
65 2.647587094 seconds time elapsed ( +- 2.27% )
66
67
68 # move check for holding day/date to where we check for time dependencies. Avoid redundant class, and reuse calendar
69 git develop/55f9b8c6fae
70 perf stat -r 10 -d Base/bin/gcc-7.3.0/release/perf_job_gen ./metabuilder.def
71 Performance counter stats for 'Base/bin/gcc-7.3.0/release/perf_job_gen ./metabuilder.def' (10 runs):
72
73 2569.686692 task-clock (msec) # 0.997 CPUs utilized ( +- 0.99% )
74 71 context-switches # 0.028 K/sec ( +- 21.34% )
75 0 cpu-migrations # 0.000 K/sec ( +- 66.67% )
76 9,658 page-faults # 0.004 M/sec ( +- 0.00% )
77 9,868,102,965 cycles # 3.840 GHz ( +- 1.04% ) (49.95%)
78 <not supported> stalled-cycles-frontend
79 <not supported> stalled-cycles-backend
80 12,801,757,236 instructions # 1.30 insns per cycle ( +- 0.08% ) (62.50%)
81 3,210,290,124 branches # 1249.292 M/sec ( +- 0.05% ) (62.52%)
82 122,763,538 branch-misses # 3.82% of all branches ( +- 0.23% ) (62.58%)
83 3,061,292,518 L1-dcache-loads # 1191.310 M/sec ( +- 0.09% ) (62.24%)
84 274,899,137 L1-dcache-load-misses # 8.98% of all L1-dcache hits ( +- 0.31% ) (24.99%)
85 88,426,942 LLC-loads # 34.412 M/sec ( +- 0.98% ) (25.02%)
86 998,813 LLC-load-misses # 1.13% of all LL-cache hits ( +- 4.56% ) (37.47%)
87
88 2.576901973 seconds time elapsed ( +- 0.97% )
89
90
91 # Only loop over job line once, when removing comment/manual and job lines
92 git develop/c1875ca6f5e
93 perf stat -r 10 -d Base/bin/gcc-7.3.0/release/perf_job_gen ./metabuilder.def
94 Performance counter stats for 'Base/bin/gcc-7.3.0/release/perf_job_gen ./metabuilder.def' (10 runs):
95
96 2439.205537 task-clock (msec) # 0.991 CPUs utilized ( +- 0.82% )
97 81 context-switches # 0.033 K/sec ( +- 23.79% )
98 0 cpu-migrations # 0.000 K/sec ( +- 50.92% )
99 9,658 page-faults # 0.004 M/sec ( +- 0.00% )
100 9,346,162,727 cycles # 3.832 GHz ( +- 0.92% ) (49.94%)
101 <not supported> stalled-cycles-frontend
102 <not supported> stalled-cycles-backend
103 12,409,439,249 instructions # 1.33 insns per cycle ( +- 0.09% ) (62.47%)
104 3,107,965,124 branches # 1274.171 M/sec ( +- 0.04% ) (62.53%)
105 116,968,249 branch-misses # 3.76% of all branches ( +- 0.19% ) (62.60%)
106 2,976,484,027 L1-dcache-loads # 1220.268 M/sec ( +- 0.08% ) (62.25%)
107 268,221,140 L1-dcache-load-misses # 9.01% of all L1-dcache hits ( +- 0.23% ) (24.99%)
108 88,439,925 LLC-loads # 36.258 M/sec ( +- 0.86% ) (25.00%)
109 1,011,364 LLC-load-misses # 1.14% of all LL-cache hits ( +- 5.26% ) (37.46%)
110
111 2.461374810 seconds time elapsed ( +- 0.82% )
112
113
114 # Use std::move() to avoid string copy, when processing job lines, remove dangling string reference
115 git develop/e0bab94495c,d77fa0e7424
116
117 perf stat -r 10 -d Base/bin/gcc-7.3.0/release/perf_job_gen ./metabuilder.def
118 Performance counter stats for 'Base/bin/gcc-7.3.0/release/perf_job_gen ./metabuilder.def' (10 runs):
119
120 2252.060052 task-clock (msec) # 0.997 CPUs utilized ( +- 0.21% )
121 59 context-switches # 0.026 K/sec ( +- 3.66% )
122 0 cpu-migrations # 0.000 K/sec
123 9,656 page-faults # 0.004 M/sec ( +- 0.00% )
124 8,655,849,508 cycles # 3.844 GHz ( +- 0.22% ) (50.03%)
125 <not supported> stalled-cycles-frontend
126 <not supported> stalled-cycles-backend
127 11,243,970,572 instructions # 1.30 insns per cycle ( +- 0.05% ) (62.57%)
128 2,815,604,290 branches # 1250.235 M/sec ( +- 0.03% ) (62.60%)
129 108,327,891 branch-misses # 3.85% of all branches ( +- 0.09% ) (62.59%)
130 2,760,671,220 L1-dcache-loads # 1225.843 M/sec ( +- 0.06% ) (62.16%)
131 258,914,707 L1-dcache-load-misses # 9.38% of all L1-dcache hits ( +- 0.16% ) (24.95%)
132 87,798,314 LLC-loads # 38.986 M/sec ( +- 0.93% ) (25.05%)
133 952,271 LLC-load-misses # 1.08% of all LL-cache hits ( +- 4.52% ) (37.53%)
134
135 2.259696222 seconds time elapsed ( +- 0.20% )
136
137