~~NOTOC~~
====== Perfsuite ======
PerfSuite is a collection of tools, utilities, and libraries for software performance analysis. It is relatively easy to use and is targeted to users at all levels of expertise.
===== Installed version =====
* **1.1.4** built with GCC and **PAPI**
PerfSuite is available as a module. To load PerfSuite, use the module load command as follows:
$ module load perf/perfsuite/1.1.4
===== Basic usage =====
- run program with ''psrun'' command
- view statitistics with ''psprocess'' command
To measure code performance using PerfSuite, use the ''psrun'' command as follows:
For serial codes:
$ psrun [-c configuration.xml] a.out
For OpenMP codes:
$ export OMP_NUM_THREADS x
$ psrun -p [-c configuration.xml] a.out
For MPI codes:
$ module load mpi/openmpi
$ mpirun -np xx psrun -f [-c configuration.xml] a.out
An XML will be generated at the end of the execution
The ''psrun'' command uses an XML configuration file to determine the mode of operation (counting or profiling). If you do not specify an XML configuration file, the default file ''itimer.xml'' will be used.
For post-processing the output, use the ''psprocess'' command as follows:
$ psprocess a.out.pid.hostname.xml
==== Example ====
Execute Matmul (''omp_matmul_03.exe'') program with N=5000
**Execute:**
$ export OMP_NUM_THREADS=8
$ psrun ./omp_matmul_03.exe 5000
The output file is ''omp_matmul_03.exe.49333.mesoshared.xml''
**Postprocessing:**
$ psprocess omp_matmul_03.exe.49333.mesoshared.xml
**Statistic output:**
PerfSuite Hardware Performance Summary Report
Version : 1.0
Created : Tue Dec 06 16:39:13 CET 2016
Generator : psprocess 0.5
XML Source : omp_matmul_03.exe.49333.mesoshared.xml
Execution Information
============================================================================================
Collector : libpshwpc
Date : Tue Dec 6 16:37:47 2016
Host : mesoshared
Process ID : 49333
Thread : 0
User : kmazouzi
Command : omp_matmul_03.exe
Processor and System Information
============================================================================================
Node CPUs : 32
Vendor : Intel
Brand : Intel(R) Xeon(R) CPU X7550 @ 2.00GHz
CPUID : family: 6, model: 46, stepping: 6
CPU Revision : 6
Clock (MHz) : 2000.180
Memory (MB) : 64409.21
Pagesize (KB) : 4
Cache Information
============================================================================================
Cache levels : 3
--------------------------------
Level 1
Type : instruction
Size (KB) : 32
Linesize (B) : 64
Assoc : 4
Type : data
Size (KB) : 32
Linesize (B) : 64
Assoc : 8
--------------------------------
Level 2
Type : unified
Size (KB) : 256
Linesize (B) : 64
Assoc : 8
--------------------------------
Level 3
Type : unified
Size (KB) : 18432
Linesize (B) : 64
Assoc : 24
Index Description Counter Value
============================================================================================
1 Conditional branch instructions.................................. 1000412152
2 Branch instructions.............................................. 1058656734
3 Conditional branch instructions mispredicted..................... 1436660
4 Conditional branch instructions taken............................ 867951726
5 Floating point operations........................................ 4377146474
6 Level 1 data cache accesses...................................... 9466165484
7 Level 1 data cache misses........................................ 603760004
8 Level 1 instruction cache accesses............................... 4358373809
9 Level 1 instruction cache misses................................. 3199281
10 Level 2 instruction cache accesses............................... 3064801
11 Level 2 instruction cache misses................................. 2001209
12 Level 2 total cache accesses..................................... 956543561
13 Level 2 cache misses............................................. 372782206
14 Load instructions................................................ 4994753947
15 Cycles stalled on any resource................................... 5075909775
16 Store instructions............................................... 2278401353
17 Data translation lookaside buffer misses......................... 1037980
18 Instruction translation lookaside buffer misses.................. 4486
19 Total cycles..................................................... 10721046042
20 Instructions issued.............................................. 11870790930
21 Instructions completed........................................... 12961269121
Event Index
============================================================================================
1: PAPI_BR_CN 2: PAPI_BR_INS 3: PAPI_BR_MSP 4: PAPI_BR_TKN
5: PAPI_FP_OPS 6: PAPI_L1_DCA 7: PAPI_L1_DCM 8: PAPI_L1_ICA
9: PAPI_L1_ICM 10: PAPI_L2_ICA 11: PAPI_L2_ICM 12: PAPI_L2_TCA
13: PAPI_L2_TCM 14: PAPI_LD_INS 15: PAPI_RES_STL 16: PAPI_SR_INS
17: PAPI_TLB_DM 18: PAPI_TLB_IM 19: PAPI_TOT_CYC 20: PAPI_TOT_IIS
21: PAPI_TOT_INS
Statistics
============================================================================================
Counting domain........................................................ user
Multiplexed............................................................ yes
Floating point operations per cycle.................................... 0.408
Floating point operations per graduated instruction.................... 0.338
Graduated instructions per cycle....................................... 1.209
Issued instructions per cycle.......................................... 1.107
Graduated instructions per issued instruction.......................... 1.092
Issued instructions per level 1 instruction cache miss................. 3710.456
Graduated instructions per level 1 instruction cache miss.............. 4051.307
Level 1 data cache accesses per graduated instruction.................. 0.730
% cycles stalled on any resource....................................... 47.345
Graduated loads and stores per floating point operation................ 1.662
Level 1 instruction cache misses per issued instruction................ 0.000
Level 1 cache miss ratio (data)........................................ 0.064
Level 1 cache miss ratio (instruction)................................. 0.001
Level 2 cache miss ratio (data), data cache miss and access counts derived 0.389
Level 2 cache miss ratio (instruction)................................. 0.653
Bandwidth used to level 2 cache (MB/s)................................. 4451.097
MFLOPS (cycles)........................................................ 816.626
MFLOPS (wall clock).................................................... 659.210
MIPS (cycles).......................................................... 2418.129
MIPS (wall clock)...................................................... 1952.000
CPU time (seconds)..................................................... 5.360
Wall clock time (seconds).............................................. 6.640
% CPU utilization...................................................... 80.724
====== PerfSuite Command-Line Tools ======
Command-Line Utilities
* ''psinv'': a utility that provides access to information about the characteristics of a machine (e.g., processor type, processor features, cache information, available performance counters)
* ''psprocess'': a utility that assists with a number of common tasks related to pre- and post-processing of performance measurements
* ''psrun'': a utility for hardware performance event counting and profiling of single-threaded, POSIX threads-based, and MPI applications. Performance counter multiplexing is supported. Optionally, psrun can also report information about the resource usage of your application such as memory usage, page faults, etc. psrun requires no source code changes to or relinking of your application.
* ''psconfigs'': a tool for easy management of PerfSuite configuration files (needs X session)
No XML file is currently available to gather overall performance characteristics for Haswell or Broadwell processors.
====== Links ======
* http://perfsuite.ncsa.illinois.edu/
* https://www.nas.nasa.gov/hecc/support/kb/using-perfsuite-for-performance-analysis_435.html