Differences

This shows you the differences between two versions of the page.

--- en:services:application_services:high_performance_computing:performance_engineering_and_analysis:likwid [2022/10/27 14:18] – jogaja
+++ en:services:application_services:high_performance_computing:performance_engineering_and_analysis:likwid [2022/10/27 15:23] (current) – jogaja
@@ Line 1: / Line 1: @@
+==== LIKWID ====
+=== LIKWID - "Like I Knew What I'm Doing" ===
+LIKWID is developed by FAU for Performance Optimization, Modeling, and Architecture Analysis
+ * Bringing the thinking back to the performance engineer.
+ * Open-source
+ * Command-line and Software Lib interface. No GUI
+ * x86 Architecture (Intel & AMD) with initial support for Nvidia GPUs
+=== Quick Start in SCC ===
+LIKIWID Toolset is available in SCC as a module, thus before using LIKWID a user need to load their preferred LIKWID version module to set the environment correctly.
+<code>
+(base) gwdu101:25 17:17:18 ~ > module show likwid
+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
+  /opt/sw/modules/21.12/cascadelake/Core/likwid/5.2.0.lua:
+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
+...
+(base) gwdu101:25 17:17:24 ~ >
+(base) gwdu101:25 17:18:33 ~ > module load likwid
+(base) gwdu101:25 17:18:43 ~ >
+</code>
+The following tasks can be performed by LIKWID:
+=== Node architecture information ===
+<code>
+$ likwid-topology
+$ likwid-powermeter
+</code>
+Examples of node-architecture for SCC's amp016:
+{{:en:services:application_services:high_performance_computing:likwid-topology-cache_g.png?400|}}
+{{:en:services:application_services:high_performance_computing:likwid-topology-cache_g2.png?400|}}
+=== Affinity control and data placement ===
+<code>
+$ likwid-pin
+$ likwid-mpirun
+</code>
+=== Query and alter system settings ===
+<code>
+$ likwid-features
+$ likwid-setFrequencies
+</code>
+=== Application performance profiling (perf-counter) ===
+ * Using the available hardware counters to measure events that characterises the interaction between software and hardware
+ * Uses a light-weight marker API for code instrumentation
+<code>
+$ likwid-perfctr
+</code>
+=== Micro-benchmarking - Application and framework to enable: ===
+  * Quantify sustainable performance limits
+  * Separate influences considering isolated instruction code snippets
+  * Reverse-egineer processor features
+  * Discover hardware bugs
+<code>
+$ likwid-bench
+$ likwid-memsweeper
+</code>
+=== likwid-topology ===
+  * Thread topology: How processor IDs map on physical compute resources
+  * Cache topology: How processors share the cache hierarchy
+  * Cache properties: Detailed information about all cache levels
+  * NUMA topology: NUMA domains and memory sizes
+  * GPU topology: GPU information
+=== likwid-pin ===
+  * Explicitly supports pthread and the OpenMP implementations of Intel and GNU gcc
+  * Only used with "pthread_create" API call which are dynamically linked with the static placement of threads.
+=== likwid-perfctr ===
+  * a lightweight command-line application to configure and read out hardware performance data
+  * Can be used as a wrapper (no modification in the application) or by adding "Marker_API" functions inside the code
+  * There are preconfigured performance groups with useful event sets and derived metrics
+  * Since ***likwid-perfctr*** measures all events on the specified CPUs, it is necessary for processes and threads to dedicated resources.
+  * This can be done by pinning the application manually or using the built-in functionality
+=== Performance Groups ===
+  * An outstanding feature of LIKWID
+  * Organizes and combines micro-architecture events and counters with e.g. run-time and clock speed
+  * Provides a set of derived metrics for efficient analysis
+  * They are read on the fly without compilation by command-line selection
+  * Are found in the path "${INSTALL_PREFIX}/share/likwid"
+Examples of using "likwid-perfctr" on SCC's amp016 node
+  * Use option "-a" to see available performance groups:
+{{:en:services:application_services:high_performance_computing:likwid-perfctr-perfgroup.png?400|}}
+  * Use "likwid-perfctr -g CLOCK" to measure the clock speed.
+{{:en:services:application_services:high_performance_computing:likwid-perfctr-g-clock.png?400|}}
+  * Use "likwid-perfctr -g FLOPS_DP" to measure the Arithmetic Intensity in double precision.
+{{:en:services:application_services:high_performance_computing:likwid-perfctr-g-flop_dp.png?400|}}
+  * Use "likwid-perfctr -g MEM" to measure the bandwidth of primary memory.
+{{:en:services:application_services:high_performance_computing:likwid-perfctr-g-mem.png?400|}}
+=== Marker API ===
+  * Enables measurements of user-defined code regions.
+  * The Marker API offers 6 functions (for C/C++) to measure named regions
+  * Activated by "-DLIKWID_PERFORM" to compiler calls
+    <code>
+    LIKWID_MARKER_INIT //global initialization
+    LIKWID_MARKER_THREADINIT //individual thread initialization
+    LIKWID_MARKER_START('compute') //Start a code region named 'compute'
+    LIKWID_MARKER_STOP('compute') //Stop the code region named 'compute'
+    LIWKID_MARKER_SWITCH //Switches perfromance group or event set in a round-robin fashion
+    LIKWID_MARKER_CLOSE //global finalization
+    </code>
+==== Additional information ====
+  * [LIKWID](https://github.com/RRZE-HPC/likwid/wiki)