CodeXL User GuideCodeXL Power Profiler provides a command line interface utility for users who prefer to use command interpreters like cmd.exe on Windows and bash on Linux. This CLI utility can be used to collect and analyze the profile data. It can also be used from a batch file or a test script.
Usage:
On Windows:
CodeXLPowerProfiler.exe <options>
On Linux:
CodeXLPowerProfiler <options>
The following options are supported:
|
-P <profile options> |
Specify what types of data will be collected during the profile session. The following data types are supported. power - collect all the available power counters temperature - collect all the available temperature counters frequency - collect all the available frequency counters cu_power - collect cpu compute-unit power counters cu_temperature - collect cpu compute-unit temperature counters gpu_power - collect gpu power counters gpu_temperature - collect gpu temperature counters core - collect core specific attributes. This includes core frequency, p-state and process id executing on the core. all - collect all the supported counters |
|
-l |
List all the counters supported by the local hardware, and the hardware devices that the profiler recognizes. The counter IDs can be used with the ‘-e’ option. The hardware device IDs can be used with the ‘-D’ option. |
|
-e <counter,...> |
Specify the comma separated list of counter names to be collected. Use option ‘-l’ to get the supported counter names. Note: use any one of the options -P or -e. |
|
-D <counter,...> |
Specify the comma separated list of device ids. All the counters of these devices will be profiled and collected. Use option (-l) to get the supported devices. |
|
-T <sampling interval> |
Sampling interval in milli-seconds. The minimum value is 20ms. |
|
-d <duration> |
Profile duration in seconds. |
|
-o <path> |
Specify the output file path. The default path will be %Temp%\Codexl-Power_<timestamp> on Windows and /tmp/ Codexl-Power_<timestamp> on Linux. |
|
-F <csv|txt> |
Define the output file format: · csv - Comma Separated Value text file · txt - plain text file The default file format is CSV file. |
|
-C <core mask> |
Specify core affinity mask for the application to be launched. Default affinity mask is all the available CPU cores. 0x2 = Core 1 0x4 = Core 2 0x8 = Core 3 |
|
Terminate the launched application after the specified profile duration. |
|
|
-w |
Specify the working directory. Default will be the path of the launched application. |
|
-h |
Displays this help information. |
|
-v |
Print version string. |
|
-z <db file output dir> |
Export results to a *.cxldb file which can be imported to CodeXL GUI application. |
|
-M process |
Process profiling is based on the IPC load. Collects power consumption of all running processes during their profile run. These power values may differ with the actual power consumption. These power values can be used to get a notion of power trend and relative power consumption among running processes. Note: Currently PMC counters are supported only on windows operating system. Supported AMD platforms are Kaveri, Mullins, Temash and Carrizo. Please refer limitation section for further details. |
Examples
· Collect power values for CPU Compute Units and GPU for the duration of 10 seconds, with sampling interval of 100 milliseconds:
CodeXLPowerProfiler.exe –P cpu_power -P gpu_power -o c:\Temp\powerprof-out.txt –T 100 –d 10
· Collect all the supported counter values for the duration of 10 seconds with sampling interval of 100 milliseconds:
CodeXLPowerProfiler.exe –P all -o c:\Temp\powerprof-out.txt –T 100 –d 10
· Collect all the supported counter values for the duration of 300 seconds with sampling interval of 100 milliseconds, and output the data to a binary file that can be imported into the graphic client:
CodeXLPowerProfiler.exe –P all –C 0x3 -z c:\Temp\PowerOutput –T 100 –d 300
· Collect all running process with their energy consumption shares as well as all counters during a particular profile run. Where profile duration is set to 300 seconds and profile sampling period is set to 10 milliseconds.
CodeXLPowerProfiler.exe –M process –e all c:\Temp\PowerOutput –T 10 –d 300
· Display help:
CodeXLPowerProfiler.exe -h
· Display version string:
CodeXLPowerProfiler.exe -v
Sample Text output files
Sample 1: Collecting non-cumulative counters
The following command collects all non-cumulative power counters for 10 seconds, sampling them every 500 milliseconds and dumping the results to a text file:
CodeXLPowerProfiler.exe -P power -d 10 -T 500 -o C:\temp\pwr_out.txt –F txt
When run on a Kaveri APU, the content of the result text file is as below:
CODEXL POWER PROFILE REPORT
PROFILE DETAILS
CPU Details: Family(0x15) Model(48)
CPU Core Mask: 0xf
Sampling Interval: 500 milli-seconds
Profile Start Time: Dec-09-2014_18-10-08
Profile Duration: 10 seconds
PROFILED COUNTERS
COUNTER ID NAME CATEGORY UNIT DESCRIPTION
0. pcie-ctrl-power Power Watt Average PCIe-Controller Power for the sampling period, …
1. mem-ctrl-power Power Watt Average DDR Memory-Controller Power for the sampling period,…
2. total-apu-power Power Watt Average APU Power for the sampling period, reported in Watts…
4. display-ctrl-power Power Watt Average Display-Controller Power for the sampling period,…
5. cpu-cu0-power Power Watt Average CPU Compute Unit Power for the sampling period, …
16. cpu-cu1-power Power Watt Average CPU Compute Unit Power for the sampling period,…
27. igpu-power Power Watt Average Integrated-GPU Power for the sampling period, …
PROFILE RECORDS
RecordId Timestamp pcie-ctrl-power mem-ctrl-power total-apu-power display-ctrl-powe cpu-cu0-power cpu-cu1-power igpu-power
0 18:10:9:002 0.56 4.88 56.28 0.00 15.49 17.36 17.98
1 18:10:9:501 0.56 4.79 39.92 0.00 13.29 13.19 8.08
2 18:10:10:003 0.56 4.83 38.00 0.00 12.86 11.58 8.17
3 18:10:10:503 0.56 4.80 38.05 0.00 12.70 12.02 7.96
4 18:10:11:003 0.56 4.79 38.80 0.00 12.84 12.53 8.07
5 18:10:11:503 0.56 4.79 38.95 0.00 13.05 12.63 7.92
6 18:10:12:003 0.56 4.82 38.16 0.00 12.15 12.57 8.06
7 18:10:12:503 0.56 4.81 37.93 0.00 12.23 12.42 7.91
8 18:10:13:003 0.56 4.80 37.46 0.00 11.97 12.06 8.05
9 18:10:13:503 0.56 4.80 37.64 0.00 12.32 11.97 7.99
10 18:10:14:003 0.56 4.83 38.91 0.00 12.49 12.99 8.04
11 18:10:14:501 0.56 4.79 38.56 0.00 12.54 12.71 7.96
12 18:10:15:003 0.56 4.81 38.89 0.00 12.38 13.08 8.06
13 18:10:15:501 0.56 4.80 39.38 0.00 12.71 13.30 8.00
14 18:10:16:003 0.56 4.80 39.37 0.00 13.03 12.91 8.07
15 18:10:16:503 0.56 4.80 39.25 0.00 12.99 12.91 7.99
16 18:10:17:003 0.56 4.80 39.72 0.00 12.50 13.80 8.05
17 18:10:17:503 0.56 4.80 38.97 0.00 12.93 12.67 8.01
18 18:10:18:003 0.56 4.81 38.95 0.00 12.97 12.56 8.05
19 18:10:18:503 0.56 4.80 39.21 0.00 12.75 13.11 7.99
20 18:10:19:003 0.56 4.80 38.66 0.00 12.69 12.56 8.05
Sample 2: Collecting cumulative counters
The following command collects the cumulative power counters for 10 seconds and dumping the results to a text file:
CodeXLPowerProfiler.exe -d 10 -o c:\temp\cumulative.txt -e 3,6,17,28 –F txt
When run on a Kaveri APU, the content of the result text file is as below:
CODEXL POWER PROFILE REPORT
PROFILE DETAILS
CPU Details: Family(0x15) Model(48)
CPU Core Mask: 0xf
Sampling Interval: 100 milli-seconds
Profile Start Time: Dec-11-2014_15-57-44
Profile Duration: 10 seconds
PROFILED COUNTERS
COUNTER ID NAME CATEGORY UNIT DESCRIPTION
3. total-apu-power-cuml Power Joule Cumulative APU Power, reported in Joules.
6. cpu-cu0-power-cuml Power Joule Cumulative CPU Compute Unit Power, reported in Joules.
17. cpu-cu1-power-cuml Power Joule Cumulative CPU Compute Unit Power, reported in Joules.
28. igpu-power-cuml Power Joule Cumulative Integrated-GPU Power, reported in Joules.
CUMULATIVE COUNTERS
COUNTER CUMULATIVE VALUE
total-apu-power-cuml 228.11
cpu-cu0-power-cuml 48.85
cpu-cu1-power-cuml 50.30
igpu-power-cuml 78.65
Sample 3: Collecting histogram counters
The following command collects the frequency histogram counters for 10 seconds and dumping the results to a text file:
CodeXLPowerProfiler.exe -d 10 -o c:\temp\histogram.txt -e 10,14,21,25,31 –F txt
When run on a Kaveri APU, the content of the result text file is as below:
CODEXL POWER PROFILE REPORT
PROFILE DETAILS
CPU Details: Family(0x15) Model(48)
CPU Core Mask: 0xf
Sampling Interval: 100 milli-seconds
Profile Start Time: Dec-11-2014_16-00-35
Profile Duration: 10 seconds
PROFILED COUNTERS
COUNTER ID NAME CATEGORY UNIT DESCRIPTION
10. cpu-core0-frequency-hist Frequency MHz Histogram of CPU Core Effective Frequency.
14. cpu-core1-frequency-hist Frequency MHz Histogram of CPU Core Effective Frequency.
21. cpu-core2-frequency-hist Frequency MHz Histogram of CPU Core Effective Frequency.
25. cpu-core3-frequency-hist Frequency MHz Histogram of CPU Core Effective Frequency.
31. igpu-frequency-hist Frequency MHz Histogram of Integrated-GPU Frequency.
HISTOGRAMS OF COUNTERS
COUNTER cpu-core0-frequency-hist
HISTOGRAM
low high count
0 200 0
200 400 0
400 600 0
600 800 16
800 1000 16
1000 1200 5
1200 1400 22
1400 1600 29
1600 1800 11
1800 2000 1
2000 2200 0
2200 2400 0
2400 2600 0
2600 2800 0
2800 3000 0
3000 3200 0
3200 3400 0
3400 3600 0
3600 3800 0
3800 4000 0
COUNTER cpu-core1-frequency-hist
HISTOGRAM
low high count
0 200 0
200 400 0
400 600 0
600 800 6
800 1000 79
1000 1200 13
1200 1400 1
1400 1600 1
1600 1800 0
1800 2000 0
2000 2200 0
2200 2400 0
2400 2600 0
2600 2800 0
2800 3000 0
3000 3200 0
3200 3400 0
3400 3600 0
3600 3800 0
3800 4000 0
COUNTER cpu-core2-frequency-hist
HISTOGRAM
low high count
0 200 0
200 400 0
400 600 0
600 800 0
800 1000 0
1000 1200 0
1200 1400 0
1400 1600 2
1600 1800 98
1800 2000 0
2000 2200 0
2200 2400 0
2400 2600 0
2600 2800 0
2800 3000 0
3000 3200 0
3200 3400 0
3400 3600 0
3600 3800 0
3800 4000 0
COUNTER cpu-core3-frequency-hist
HISTOGRAM
low high count
0 200 0
200 400 0
400 600 0
600 800 0
800 1000 28
1000 1200 45
1200 1400 25
1400 1600 1
1600 1800 1
1800 2000 0
2000 2200 0
2200 2400 0
2400 2600 0
2600 2800 0
2800 3000 0
3000 3200 0
3200 3400 0
3400 3600 0
3600 3800 0
3800 4000 0
COUNTER igpu-frequency-hist
HISTOGRAM
low high count
0 100 0
100 200 0
200 300 0
300 400 100
400 500 0
500 600 0
600 700 0
700 800 0
800 900 0
900 1000 0
Sample 3: Collecting process profiling data
The following command collects the process profiling data for 100 seconds and dumping the results to a text file:
CodeXLPowerProfiler.exe -d 10 –T 10 -o c:\temp\histogram.txt –M process –F txt
When run on a Carrizo APU, the content of the result text file is as below:
CODEXL POWER PROFILE REPORT
PROFILE DETAILS
CPU Details: Family(0x15) Model(0x60)
CPU Core Mask: 0xf
Sampling Interval: 10 milli-seconds
Profile Start Time: Mar-30-2016_00-20-44
Profile Duration: 100 seconds
PROCESS PROFILING DATA
SNo PID Samples IPC Power(Joules) Power(%) Name Path
0 4956 11 3.34 0.04 0.03 CodeXLPowerProfiler.exe \Program Files (x86)\CodeXL
1 0 383503 31755 109.58 95.28 [System Process] Unable to read path
2 648 1 0.38 0.01 0.01 lsass.exe C:\Windows\System32
3 900 54 13.72 0.06 0.05 dwm.exe C:\Windows\System32
4 4 224 96.76 0.24 0.21 System Unable to read path
5 2140 294 232.54 0.82 0.71 explorer.exe C:\Windows
6 760 4 1.04 0.00 0.00 svchost.exe C:\Windows\System32
7 1648 16 5.71 0.02 0.01 MsMpEng.exe Unable to read path
8 548 15 2.21 0.01 0.01 csrss.exe Unable to read path
9 4396 1066 710.18 3.86 3.36 iexplore.exe C:\Program Files (x86)\Internet Explorer
10 680 178 170.15 0.29 0.25 svchost.exe C:\Windows\System32
11 5296 1 0.44 0.00 0.00 conhost.exe C:\Windows\System32
12 2500 2 0.29 0.00 0.00 svchost.exe C:\Windows\System32
13 1320 6 2.57 0.01 0.01 svchost.exe C:\Windows\System32
14 1160 2 0.80 0.00 0.00 SearchIndexer.exe C:\Windows\System32
15 456 35 15.70 0.04 0.03 svchost.exe C:\Windows\System32
16 1108 11 7.65 0.02 0.01 svchost.exe C:\Windows\System32
17 640 3 0.84 0.00 0.00 services.exe Unable to read path
18 936 2 1.00 0.00 0.00 svchost.exe C:\Windows\System32
19 1448 3 1.22 0.00 0.00 taskhostex.exe C:\Windows\System32
20 720 2 0.46 0.00 0.00 svchost.exe C:\Windows\System32
21 1612 2 0.63 0.00 0.00 WUDFHost.exe C:\Windows\System32
22 1296 4 1.42 0.00 0.00 spoolsv.exe C:\Windows\System32
23 2736 1 0.25 0.00 0.00 sppsvc.exe Unable to read path
Profile Sesssion Power Consumption: 115.00
Total PID record collected 385440