A test of building the llvm compiler stack. The entire stack is built twice, once with Ninja and then with Unix Makefiles. Overall time is slightly faster for Ninja. The overall profile shows except for periods towards end of each build, almost all cores are kept busy.
Overall topdown profile is a workload dominated by frontend stalls with a lower retirement rate. This is similar to other build-* workloads, though backend/memory stalls are slightly higher with LLVM and frontend stalls are slightly lower.
AMD topdown metrics show little floating point, a high amount of branches and a moderate amount of L2 access/miss
elapsed 5163.931
on_cpu 0.945 # 15.11 / 16 cores
utime 72481.356
stime 5561.869
nvcsw 690828 # 26.70%
nivcsw 1896731 # 73.30%
inblock 752 # 0.15/sec
onblock 72693768 # 14077.21/sec
cpu-clock 78056877516054 # 78056.878 seconds
task-clock 78058299825553 # 78058.300 seconds
page faults 1723272985 # 22076.742/sec
context switches 2501754 # 32.050/sec
cpu migrations 84464 # 1.082/sec
major page faults 7702 # 0.099/sec
minor page faults 1723265283 # 22076.644/sec
alignment faults 0 # 0.000/sec
emulation faults 0 # 0.000/sec
branches 53968965816670 # 209.027 branches per 1000 inst
branch misses 1624267586008 # 3.01% branch miss
conditional 41377495312890 # 160.259 conditional branches per 1000 inst
indirect 1239736919294 # 4.802 indirect branches per 1000 inst
cpu-cycles 318616249965597 # 3.88 GHz
instructions 257239532203885 # 0.81 IPC
slots 638462591892552 #
retiring 83608634229281 # 13.1% (16.4%)
-- ucode 98629507536 # 0.0%
-- fastpath 83510004721745 # 13.1%
frontend 200104012133847 # 31.3% (39.2%)
-- latency 150156168729132 # 23.5%
-- bandwidth 49947843404715 # 7.8%
backend 209812204890246 # 32.9% (41.1%)
-- cpu 17481000115336 # 2.7%
-- memory 192331204774910 # 30.1%
speculation 16989232947648 # 2.7% ( 3.3%)
-- branch mispredict 16820144024682 # 2.6%
-- pipeline restart 169088922966 # 0.0%
smt-contention 127947932556371 # 20.0% ( 0.0%)
cpu-cycles 318995838054107 # 3.88 GHz
instructions 257262438648603 # 0.81 IPC
instructions 85844563756236 # 57.868 l2 access per 1000 inst
l2 hit from l1 4340327891636 # 21.96% l2 miss
l2 miss from l1 756553839542 #
l2 hit from l2 pf 293116256893 #
l3 hit from l2 pf 124042424042 #
l3 miss from l2 pf 210191796922 #
instructions 85821948979048 # 18.776 float per 1000 inst
float 512 38882 # 0.000 AVX-512 per 1000 inst
float 256 18532849 # 0.000 AVX-256 per 1000 inst
float 128 1611376882605 # 18.776 AVX-128 per 1000 inst
float MMX 0 # 0.000 MMX per 1000 inst
float scalar 5 # 0.000 scalar per 1000 inst
Intel metrics. Perhaps counted slightly differently but this workload as others shows higher IPC, lower GHz and then higher retirement rate than AMD.
elapsed 6082.871
on_cpu 0.952 # 15.23 / 16 cores
utime 87827.392
stime 4786.786
nvcsw 920851 # 30.45%
nivcsw 2103588 # 69.55%
inblock 423848 # 69.68/sec
onblock 72678008 # 11947.98/sec
cpu-clock 92626124571056 # 92626.125 seconds
task-clock 92627957812326 # 92627.958 seconds
page faults 1715040925 # 18515.370/sec
context switches 2950846 # 31.857/sec
cpu migrations 96215 # 1.039/sec
major page faults 2083 # 0.022/sec
minor page faults 1715038842 # 18515.348/sec
alignment faults 0 # 0.000/sec
emulation faults 0 # 0.000/sec
branches 53412577346061 # 207.493 branches per 1000 inst
branch misses 1337965435641 # 2.50% branch miss
conditional 53412581027341 # 207.493 conditional branches per 1000 inst
indirect 10008132965242 # 38.879 indirect branches per 1000 inst
slots 437772678130796 #
retiring 133128924746572 # 30.4% (30.4%)
-- ucode 10347229360575 # 2.4%
-- fastpath 122781695385997 # 28.0%
frontend 154233918698731 # 35.2% (35.2%)
-- latency 89782197058371 # 20.5%
-- bandwidth 64451721640360 # 14.7%
backend 97914185694265 # 22.4% (22.4%)
-- cpu 24835061651941 # 5.7%
-- memory 73079124042324 # 16.7%
speculation 53004289903871 # 12.1% (12.1%)
-- branch mispredict 51316773822425 # 11.7%
-- pipeline restart 1687516081446 # 0.4%
smt-contention 0 # 0.0% ( 0.0%)
cpu-cycles 194627005685821 # 2.00 GHz
instructions 194392697442484 # 1.00 IPC
l2 access 8477188822805 # 59.773 l2 access per 1000 inst
l2 miss 2579404345601 # 30.43% l2 miss
Process summary is incomplete. Ninja completes, Unix makefiles dies part way through, but at least half is sufficient to get a good profile. Overall cpu time is dominated by the C++ front end, with a lesser amount in tblgen program. Many invocations of the assembler with almost no time.
78068 processes
12328 cc1plus 48003.58 2933.97
1342 llvm-tblgen 2043.24 144.27
901 ld 423.71 113.10
12793 as 84.78 7.24
548 cc1 14.60 1.46
5 xz 12.57 0.90
34 clinfo 9.59 3.66
7885 cmake 9.49 10.41
1443 ninja 3.77 2.85
19 vulkaninfo 0.73 0.73
850 ranlib 0.69 5.26
1573 gmake 0.60 0.61
849 ar 0.59 5.16
10 tar 0.54 9.47
80 python3.10 0.50 0.04
13 rm 0.09 6.98
2 vulkani:disk$0 0.07 0.07
3 glxinfo:gdrv0 0.07 0.04
6 clang 0.06 0.06
1 llvmpipe-0 0.04 0.04
1 llvmpipe-1 0.04 0.04
1 llvmpipe-10 0.04 0.04
1 llvmpipe-11 0.04 0.04
1 llvmpipe-12 0.04 0.04
1 llvmpipe-13 0.04 0.04
1 llvmpipe-14 0.04 0.04
1 llvmpipe-15 0.04 0.04
1 llvmpipe-2 0.04 0.04
1 llvmpipe-3 0.04 0.04
1 llvmpipe-4 0.04 0.04
1 llvmpipe-5 0.04 0.04
1 llvmpipe-6 0.04 0.04
1 llvmpipe-7 0.04 0.04
1 llvmpipe-8 0.04 0.04
1 llvmpipe-9 0.04 0.04
5 py3versions 0.04 0.01
1 glxinfo 0.03 0.02
1 glxinfo:cs0 0.03 0.02
1 glxinfo:disk$0 0.03 0.02
1 glxinfo:sh0 0.03 0.02
1 glxinfo:shlo0 0.03 0.02
1 ps 0.00 0.01
22348 sh 0.00 0.00
12872 c++ 0.00 0.00
949 cc 0.00 0.00
901 collect2 0.00 0.00
30 uname 0.00 0.00
25 git 0.00 0.00
20 pkg-config 0.00 0.00
13 gcc 0.00 0.00
10 gsettings 0.00 0.00
8 systemd-detect- 0.00 0.00
7 stat 0.00 0.00
6 bash 0.00 0.00
6 llvm-link 0.00 0.00
6 sed 0.00 0.00
5 mkdir 0.00 0.00
5 mv 0.00 0.00
4 build-llvm 0.00 0.00
4 phoronix-test-s 0.00 0.00
3 gmain 0.00 0.00
2 dconf worker 0.00 0.00
2 which 0.00 0.00
1 date 0.00 0.00
1 dirname 0.00 0.00
1 grep 0.00 0.00
1 ifconfig 0.00 0.00
1 ip 0.00 0.00
1 lscpu 0.00 0.00
1 mktemp 0.00 0.00
1 python 0.00 0.00
1 python3 0.00 0.00
1 qdbus 0.00 0.00
1 readlink 0.00 0.00
1 realpath 0.00 0.00
1 sort 0.00 0.00
1 stty 0.00 0.00
1 systemctl 0.00 0.00
1 template.sh 0.00 0.00
1 wc 0.00 0.00
1 xrandr 0.00 0.00
1 xset 0.00 0.00
83 processes running
99 maximum processesThe following pattern is used for compilation
58628) ninja cpu=3 start=9.98 finish=9.98
58629) ninja cpu=5 start=9.99 finish=9.99
58630) ninja cpu=11 start=9.99 finish=9.99
58631) ninja cpu=4 start=9.99 finish=10.03
58632) sh cpu=15 start=10.00 finish=10.01
58633) cc cpu=0 start=10.00 finish=10.01
58634) cc1 cpu=1 start=10.00 finish=10.01
58635) as cpu=14 start=10.01 finish=10.01
58636) sh cpu=5 start=10.01 finish=10.03
58637) cc cpu=14 start=10.02 finish=10.03
58638) collect2 cpu=0 start=10.02 finish=10.03
58639) ld cpu=15 start=10.02 finish=10.03