NSCCN/LIKWID：硬件性能监控与基准测试工具，用于量化分析CPU微架构级性能指标。

Introduction

Likwid is a simple to install and use toolsuite of command line applications and a library for performance oriented programmers. It works for Intel, AMD, ARMv8 and POWER9 processors on the Linux operating system. There is additional support for Nvidia and AMD GPUs. There is support for ARMv7 and POWER8/9 but there is currently no test machine in our hands to test them properly.

LIKWID Playlist (YouTube)

It consists of:

likwid-topology: print thread, cache and NUMA topology
likwid-perfctr: configure and read out hardware performance counters on Intel, AMD, ARM and POWER processors and Nvidia GPUs
likwid-powermeter: read out RAPL Energy information and get info about Turbo mode steps
likwid-pin: pin your threaded application (pthread, Intel and gcc OpenMP to dedicated processors)
likwid-bench: Micro benchmarking platform for CPU architectures
likwid-features: Print and manipulate cpu features like hardware prefetchers (x86 only)
likwid-genTopoCfg: Dumps topology information to a file
likwid-mpirun: Wrapper to start MPI and Hybrid MPI/OpenMP applications (Supports Intel MPI, OpenMPI, MPICH and SLURM)
likwid-perfscope: Frontend to the timeline mode of likwid-perfctr, plots live graphs of performance metrics using gnuplot
likwid-memsweeper: Sweep memory of NUMA domains and evict cachelines from the last level cache
likwid-setFrequencies: Tool to control the CPU and Uncore frequencies (x86 only)
likwid-sysFeatures: Tool to system settings like frequencies, powercaps and prefetchers (experimental)

For further information please take a look at the Wiki or contact us via Matrix chat LIKWID General.

Supported architectures

Intel

Intel Atom
Intel Pentium M
Intel Core2
Intel Nehalem
Intel NehalemEX
Intel Westmere
Intel WestmereEX
Intel Xeon Phi (KNC)
Intel Silvermont & Airmont
Intel Goldmont
Intel SandyBridge
Intel SandyBridge EP/EN
Intel IvyBridge
Intel IvyBridge EP/EN/EX
Intel Xeon Phi (KNL, KNM)
Intel Haswell
Intel Haswell EP/EN/EX
Intel Broadwell
Intel Broadwell D
Intel Broadwell EP
Intel Skylake
Intel Kabylake
Intel Coffeelake
Intel Skylake SP
Intel Cascadelake SP
Intel Icelake
Intel Icelake SP
Intel Tigerlake (experimental)
Intel SapphireRapids
Intel EmeraldRapids

AMD

AMD K8
AMD K10
AMD Interlagos
AMD Kabini
AMD Zen
AMD Zen2
AMD Zen3
AMD Zen4

ARM

ARMv7
ARMv8
Special support for Marvell Thunder X2
Fujitsu A64FX
ARM Neoverse N1 (AWS Graviton 2)
ARM Neoverse V1
HiSilicon TSV110
Apple M1 (only with Linux)

POWER (experimental)

IBM POWER8
IBM POWER9

Nvidia GPUs

AMD GPUs

Download, Build and Install

You can get the releases of LIKWID at: http://ftp.fau.de/pub/likwid/

For build and installation hints see INSTALL file or check the build instructions page in the wiki https://github.com/RRZE-HPC/likwid/wiki/Build

For quick install:

VERSION=stable
wget http://ftp.fau.de/pub/likwid/likwid-$VERSION.tar.gz
tar -xaf likwid-$VERSION.tar.gz
cd likwid-*
vi config.mk # configure build, e.g. change installation prefix and architecture flags
make
sudo make install # sudo required to install the access daemon with proper permissions

For ARM builds, the COMPILER flag in config.mk needs to changed to GCCARMv8 or ARMCLANG (experimental). For POWER builds, the COMPILER flag in config.mk needs to changed to GCCPOWER or XLC (experimental). For Nvidia GPU support, set NVIDIA_INTERFACE in config.mk to true and adjust build-time variables if needed For AMD GPU support, set ROCM_INTERFACE in config.mk to true and adjust build-time variables if needed

Usage examples

likwid-topology

--------------------------------------------------------------------------------
CPU name:    Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz
CPU type:    Intel Skylake processor
CPU stepping:    3
********************************************************************************
Hardware Thread Topology
********************************************************************************
Sockets:        1
Cores per socket:    4
Threads per core:    2
--------------------------------------------------------------------------------
HWThread        Thread        Core        Die        Socket        Available
0               0             0           0          0             *                
1               0             1           0          0             *                
2               0             2           0          0             *                
3               0             3           0          0             *                
4               1             0           0          0             *                
5               1             1           0          0             *                
6               1             2           0          0             *                
7               1             3           0          0             *                
--------------------------------------------------------------------------------
Socket 0:        ( 0 4 1 5 2 6 3 7 )
--------------------------------------------------------------------------------
********************************************************************************
Cache Topology
********************************************************************************
Level:            1
Size:            32 kB
Cache groups:        ( 0 4 ) ( 1 5 ) ( 2 6 ) ( 3 7 )
--------------------------------------------------------------------------------
Level:            2
Size:            256 kB
Cache groups:        ( 0 4 ) ( 1 5 ) ( 2 6 ) ( 3 7 )
--------------------------------------------------------------------------------
Level:            3
Size:            8 MB
Cache groups:        ( 0 4 1 5 2 6 3 7 )
--------------------------------------------------------------------------------
********************************************************************************
NUMA Topology
********************************************************************************
NUMA domains:        1
--------------------------------------------------------------------------------
Domain:            0
Processors:        ( 0 4 1 5 2 6 3 7 )
Distances:        10
Free memory:        318.203 MB
Total memory:        7626.23 MB
--------------------------------------------------------------------------------

likwid-perfctr

$ likwid-perfctr -C 0 -g L2 hostname
--------------------------------------------------------------------------------
CPU name:    Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz
CPU type:    Intel Skylake processor
CPU clock:    4.01 GHz
--------------------------------------------------------------------------------
mytesthost
--------------------------------------------------------------------------------
Group 1: L2
+-----------------------+---------+------------+
|         Event         | Counter | HWThread 0 |
+-----------------------+---------+------------+
|   INSTR_RETIRED_ANY   |  FIXC0  |     321342 |
| CPU_CLK_UNHALTED_CORE |  FIXC1  |     450498 |
|  CPU_CLK_UNHALTED_REF |  FIXC2  |    1118900 |
|    L1D_REPLACEMENT    |   PMC0  |       6670 |
|      L1D_M_EVICT      |   PMC1  |       1840 |
| ICACHE_64B_IFTAG_MISS |   PMC2  |       9293 |
+-----------------------+---------+------------+

+——————————–+————+
|             Metric             | HWThread 0 |
+——————————–+————+
|       Runtime (RDTSC) [s]      |     0.0022 |
|      Runtime unhalted [s]      |     0.0001 |
|           Clock [MHz]          |  1613.6392 |
|               CPI              |     1.4019 |
|  L2D load bandwidth [MBytes/s] |   197.8326 |
|  L2D load data volume [GBytes] |     0.0004 |
| L2D evict bandwidth [MBytes/s] |    54.5745 |
| L2D evict data volume [GBytes] |     0.0001 |
|     L2 bandwidth [MBytes/s]    |   528.0381 |
|     L2 data volume [GBytes]    |     0.0011 |
+——————————–+————+

likwid-pin

$ likwid-pin -c 0,1,2 ./a.out
[pthread wrapper] 
[pthread wrapper] MAIN -> 0
[pthread wrapper] PIN_MASK: 0->1  1->2  
[pthread wrapper] SKIP MASK: 0x0
    threadid 140566548539136 -> hwthread 1 - OK
    threadid 140566540146432 -> hwthread 2 - OK
Number of Threads requested = 3
Thread 0 running on processor 0 ....
Thread 1 running on processor 1 ....
Thread 2 running on processor 2 ....
[...]

likwid-bench

$ likwid-bench -t triad_avx -W N:2GB:3
Warning: Sanitizing vector length to a multiple of the loop stride 16 and thread count 3 from 62500000 elements (500000000 bytes) to 62499984 elements (499999872 bytes)
Allocate: Process running on hwthread 0 (Domain N) - Vector length 62499984/499999872 Offset 0 Alignment 512
Allocate: Process running on hwthread 0 (Domain N) - Vector length 62499984/499999872 Offset 0 Alignment 512
Allocate: Process running on hwthread 0 (Domain N) - Vector length 62499984/499999872 Offset 0 Alignment 512
Allocate: Process running on hwthread 0 (Domain N) - Vector length 62499984/499999872 Offset 0 Alignment 512
Initialization: Each thread in domain initializes its own stream chunks
--------------------------------------------------------------------------------
LIKWID MICRO BENCHMARK
Test: triad_avx
--------------------------------------------------------------------------------
Using 1 work groups
Using 3 threads
--------------------------------------------------------------------------------
Running without Marker API. Activate Marker API with -m on commandline.
--------------------------------------------------------------------------------
Group: 0 Thread 1 Global Thread 1 running on hwthread 4 - Vector length 20833328 Offset 20833328
Group: 0 Thread 0 Global Thread 0 running on hwthread 0 - Vector length 20833328 Offset 0
Group: 0 Thread 2 Global Thread 2 running on hwthread 1 - Vector length 20833328 Offset 41666656
--------------------------------------------------------------------------------
Cycles:            22977763263
CPU Clock:        4007946861
Cycle Clock:        4007946861
Time:            5.733051e+00 sec
Iterations:        96
Iterations per thread:    32
Inner loop executions:    1302083
Size (Byte):        1999999488
Size per thread:    666666496
Number of Flops:    3999998976
MFlops/s:        697.71
Data volume (Byte):    63999983616
MByte/s:        11163.34
Cycles per update:    11.488885
Cycles per cacheline:    91.911077
Loads per update:    3
Stores per update:    1
Load bytes per element:    24
Store bytes per elem.:    8
Load/store ratio:    3.00
Instructions:        2374999408
UOPs:            3749999040
--------------------------------------------------------------------------------

likwid-mpirun

$ likwid-mpirun -mpi slurm -np 4 -t 2 ./a.out
MPI started
Process with rank 0 running on Node f0846.nhr.fau.de core 0
Process with rank 2 running on Node f0859.nhr.fau.de core 0
Process with rank 3 running on Node f0859.nhr.fau.de core 36
Process with rank 1 running on Node f0846.nhr.fau.de core 36
Enter OpenMP parallel region
Start OpenMP threads
Rank 0 Thread 0 running on Node f0846.nhr.fau.de core 0
Rank 0 Thread 1 running on Node f0846.nhr.fau.de core 1
Rank 1 Thread 0 running on Node f0846.nhr.fau.de core 36
Rank 1 Thread 1 running on Node f0846.nhr.fau.de core 37
Rank 2 Thread 0 running on Node f0859.nhr.fau.de core 0
Rank 2 Thread 1 running on Node f0859.nhr.fau.de core 1
Rank 3 Thread 0 running on Node f0859.nhr.fau.de core 36
Rank 3 Thread 1 running on Node f0859.nhr.fau.de core 37

likwid-powermeter

$ likwid-powermeter 
--------------------------------------------------------------------------------
CPU name:    Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz
CPU type:    Intel Skylake processor
CPU clock:    4.01 GHz
--------------------------------------------------------------------------------
--------------------------------------------------------------------------------
Runtime: 2.00019 s
Measure for socket 0 on CPU 0
Domain PKG:
Energy consumed: 7.47705 Joules
Power consumed: 3.73817 Watt
Domain PP0:
Energy consumed: 5.42047 Joules
Power consumed: 2.70998 Watt
Domain PP1:
Energy consumed: 0.0872803 Joules
Power consumed: 0.043636 Watt
Domain DRAM:
Energy consumed: 1.02612 Joules
Power consumed: 0.513013 Watt
Domain PLATFORM:
Energy consumed: 0 Joules
Power consumed: 0 Watt
--------------------------------------------------------------------------------

likwid-features

$ likwid-features -c 0 -l
Feature               HWThread 0    
HW_PREFETCHER         on    
CL_PREFETCHER         on    
DCU_PREFETCHER        on    
IP_PREFETCHER         on    
FAST_STRINGS          on    
THERMAL_CONTROL       on    
PERF_MON              on    
FERR_MULTIPLEX        off    
BRANCH_TRACE_STORAGE  on    
XTPR_MESSAGE          off    
PEBS                  on    
SPEEDSTEP             on    
MONITOR               on    
SPEEDSTEP_LOCK        off    
CPUID_MAX_VAL         off    
XD_BIT                on    
DYN_ACCEL             off    
TURBO_MODE            on    
TM2                   off

Documentation

For a detailed documentation on the usage of the tools have a look at the html documentation build with doxygen. Call

make docs

or after installation, look at the man pages.

There is also a wiki at the github page: https://github.com/rrze-likwid/likwid/wiki

If you have problems or suggestions please let me know on the likwid mailing list: http://groups.google.com/group/likwid-users

or if it is bug, add an issue at: https://github.com/rrze-likwid/likwid/issues

You can also chat with us through Matrix:

General chat: https://matrix.to/#/#likwid:matrix.org?via=matrix.org
Development chat: https://matrix.to/#/#likwid-dev:matrix.org?via=matrix.org

Extras

If you want to use the Marker API with Java, you can find the Java module here: https://github.com/jacek-lewandowski/likwid-java-api
For Python you can find an interface to the LIKWID API here: https://github.com/RRZE-HPC/pylikwid or pip install pylikwid
A Julia interface to LIKWID is provided by the Paderborn Center for Parallel Computing (PC²) and the MIT JuliaLab: https://github.com/JuliaPerf/LIKWID.jl or ] add LIKWID

Survey

We opened a survey at the user mailing list to get a feeling who uses LIKWID and how. Moreover we would be interested if you are missing a feature or what annoys you when using LIKWID. Link to the survey: https://groups.google.com/forum/#!topic/likwid-users/F7TDho3k7ps

Funding

LIKWID development was funded by BMBF Germany under the FEPA project, grant 01IH13009. Since 2017 the development is further funded by BMBF Germany under the SeASiTe project, grant 01IH16012A. In 2022, the EE-HPC project is funded by BMBF Germany in the GreenHPC grant.