Cross Reference: /external/lisa/libs/wlgen/wlgen/rta.py

History log of /external/lisa/libs/wlgen/wlgen/rta.py
Revision	Date	Author	Comments (<<< Hide modified files) (Show modified files >>>)
29a01e184f667f9e3b9d2a5a893856e0445e6e4f	04-Aug-2017	Brendan Jackman <brendan.jackman@arm.com>	rta: custom: Don't require duration if not needed If the provided JSON template doesn't include the __DURATION__ replacement string, then don't require the user to provide a duration value in the python. /external/lisa/libs/wlgen/wlgen/rta.py
89ece72cedda4cc7e3cf0d596afe08bc7676ffc6	21-Jul-2017	Patrick Bellasi <derkling@gmail.com>	Merge pull request #366 from bjackman/fix-rta-target-cpu rta: Fix setting up "calibration" field in rt-app global conf
89facb762f30211ab0b6915d6ef7d3f38a7f39cc	21-Jul-2017	Brendan Jackman <brendan.jackman@arm.com>	rta: Disallow invalid composition of delayed tasks This example: wl = RTA(te.target, 'mywl', calibration=calib) wl.conf('profile', params = {'t': Periodic(delay_s=3, duration_s=1) + Periodic(delay_s=2, duration_s=1)).get()}) Should semantically result in 1. 3s delay 2. 1s of periodic behaviour 3. 2s delay 4. 1s of periodic behaviour. But part 3 (the second delay) will be ignored due to the way we use the "delay" field. We could probably fix this using a "sleep" event, but to keep things simple let's just disallow this type of composition, unless we later find that we need that feature. /external/lisa/libs/wlgen/wlgen/rta.py
72c8f4f7afe58d8ca8141fcbb0c8b8fc0c96345f	21-Jul-2017	Zhifei Yang <zhifei.yang@arm.com>	BUG: RTA workload property delay_s will cause kernel userspace rt-app segmentation fault. Configure the correct RTA settings to meet the RT-APP requirement ( up to commit=a9a0c6c ) Signed-off-by: Zhifei Yang <zhifei.yang@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
366a69c9f3acad897bc71be36b84ca69b35debc4	06-Jul-2017	Brendan Jackman <brendan.jackman@arm.com>	rta: Fix & simplify calibration configuration This simplifies the behaviour for setting up rt-app's "calibration" field, aiming to ensure a sensible default is provided when no information is provided explicitly about the platform. - When we have calibrated rt-app, use the smallest value in self.pload (or the largest if loadref="LITTLE"). This means we don't need any big.LITTLE data at all in this case. - If we don't have a calibration, then we wan to pick a big CPU as the "target". If we have the platform data then determine the first big CPU from that, otherwise use the highest-numbered CPU (since it happens that most big.LITTLE systems number the LITTLE CPUs lowest) Also add some simple tests for this behaviour. /external/lisa/libs/wlgen/wlgen/rta.py
5c2aba5c77d29534b043002e7bd9f745f7679031	06-Jun-2017	Brendan Jackman <brendan.jackman@arm.com>	rta: Save and restore governor tunables in calibration Governor tunables are reset to default when switching governors. This is particularly relevant when using the Executor, which calls te.calibration() after _setup_cpufreq. If the calibration has not been done yet, that will result in setting the performance governor then restoring the previous governor. However if _setup_cpufreq set up any tunables they are not currently restored. This commit fixes that. /external/lisa/libs/wlgen/wlgen/rta.py
d9015a7db7b1009df2901a8f0f3a915e6f61e3bb	06-Jun-2017	Brendan Jackman <brendan.jackman@arm.com>	rta: Use cpufreq.iter_domains method This method was added recently to devlib to clarify this pattern. /external/lisa/libs/wlgen/wlgen/rta.py
c758d400e173106b7e7e046d7495991cb49a3156	24-May-2017	Brendan Jackman <brendan.jackman@arm.com>	rta: Fix check for task name length self.params does not have the task names as keys, self.params['profile'] does. /external/lisa/libs/wlgen/wlgen/rta.py
f108c0db83aa25982564176ac4b0e9015a2e44e2	10-Apr-2017	Brendan Jackman <brendan.jackman@arm.com>	wlgen/rta: Fix calibration value string replacement Currently, under the assumption that the calibration is an int, we don't quote it. However it may be a string like "CPU0" in which case we _do_ need to quote it. Also add a test for this case. /external/lisa/libs/wlgen/wlgen/rta.py
8684aed7f8be9f1f72ca532e392f4bc34074ad36	10-Apr-2017	Brendan Jackman <brendan.jackman@arm.com>	rta: Detect task names when using custom wload RTA.__postrun uses the `tasks` attribute to find the logfiles to pull from the target. So when using a custom rt-app configuration, set up this attribute in a similar way to "profile" workloads. /external/lisa/libs/wlgen/wlgen/rta.py
c9f6360e5e690af25fabb1fcfdd436737dff7b9c	24-Mar-2017	Valentin Schneider <valentin.schneider@arm.com>	libs/wlgen: Fix governor save/restore in rtapp calibration Governor was set to "performance" for all cpus during the rtapp calibration, but the previous governors weren't restored after that. /external/lisa/libs/wlgen/wlgen/rta.py
69c8070bb702c1b41a650d70f9f1903fb8045871	15-Feb-2017	Patrick Bellasi <derkling@gmail.com>	Merge pull request #297 from bjackman/wlgen-docs Sphinx documentation for Workload & RTA classes
b117b255584479470fe5e8f2c2f480ca3e1eb597	03-Jan-2017	Brendan Jackman <brendan.jackman@arm.com>	wlgen: Add some Sphinx-compatible docstrings /external/lisa/libs/wlgen/wlgen/rta.py
1ec0e9325c3eaab702a60565c9d7bb4d4172dd1f	14-Feb-2017	Patrick Bellasi <derkling@gmail.com>	Merge pull request #293 from bjackman/workload-cleanups Cleanups & Fixes for Workload class
fc51352881f5eb0430a782769428f5b976aa2fc1	08-Feb-2017	Brendan Jackman <brendan.jackman@arm.com>	rta: Use self.target.path instead of explicit '/'s /external/lisa/libs/wlgen/wlgen/rta.py
a1d17daaca43b8e6e8241eb75a7808dc2a773044	03-Feb-2017	Brendan Jackman <brendan.jackman@arm.com>	rta: Don't put single quotes around glob patterns This only works by accident over ADB and SSH. On a LocalConnection it results in trying to look for files containing quote characters. /external/lisa/libs/wlgen/wlgen/rta.py
2eb43b32feb25b1a9d475efa06af604bf82e1bae	03-Jan-2017	Brendan Jackman <brendan.jackman@arm.com>	rta: Rename _TaskBase -> RTATask This is just to make the class public so we can document it with sphinx /external/lisa/libs/wlgen/wlgen/rta.py
0e6969c856cf0829f1edfe9285c880ecb69ed516	03-Jan-2017	Brendan Jackman <brendan.jackman@arm.com>	workload: Remove rt-app specific `calibration` /external/lisa/libs/wlgen/wlgen/rta.py
f046fc8bda1d538623205c3b251538a06370c8ea	15-Dec-2016	Patrick Bellasi <patrick.bellasi@arm.com>	wlgen/rtapp: fix loading of custom JSON When 'kind=custom' the 'params' parameter must be used to specify the full path of a JSON file to be used for the rt-app confiuration. This feature is currently broken because we anticipate in the rtapp::conf() method some checks required only by the "profile" kind. This patch fixes this issue and improve some sanity checks on the values of the specified parameters. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
c278c2343c7f25083a80cb164b6bdc761d50050b	22-Nov-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs: use a more configurable logging strategy This is a big re-factoring which update all the logging statements to make use of a more sane and consistent configuration: - each module uses the same logger configuration but refers to a custom logger which has the same name of the module, this should allows a fine grained configuration of the loglevels for each module. - a new libs/utils/conf.py::LisaLogging class allows to load the configuration and customize the root logger level. - loggers are configured using a single logfile (logging.conf) which is provided by default in the LISA's root folder. Each logger can be configured to specify: - loglevel: INFO by default for all modules - output channels: console and logfile (lisa.log by default) for all modules - string format: the same by default for all modules If logging.conf does not specify a specific configuration for a logger, than the default logger configuration applies, which is the one defined by the "root" logger. Notebooks and LisaTest are required to call at least one time the LisaLogging::setup() method to properly intialize logging. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
9cb731a907d42210fa85c0532c2389e511953255	28-Oct-2016	Patrick Bellasi <patrick.bellasi@arm.com>	wlgen/rtapp: allow to specify task's CPU as single number or range-string Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
5d5ffca33ab8f720bb841e447dda844c7ea45c4a	03-Oct-2016	Brendan Jackman <brendan.jackman@arm.com>	libs/wlgen/rta: Limit task name to 15 characters rt-app uses pthread_setname_np(3) to set the names of tasks on the target [1]. This interface limits thread name length to 16 characters including the null '\0'. Although rt-app prints a message on the target, the workload execution goes ahead when the provided task names are too long, and the tasks are named "rt-app". This means they are not found when later examining the trace. Therefore, raise an exception on the host to prevent this situation arising. [1] https://github.com/scheduler-tools/rt-app/blob/dcaf5a04aa63b69d8b5e174d10a9e459f88772cf/src/rt-app.c#L152 /external/lisa/libs/wlgen/wlgen/rta.py
0685819e9106c35188afa360e35f383b22f98dd7	19-Jul-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: add a simple sanity check for calibration values On big.LITTLE platforms we always expect big CPUs to be more capable than LITTLE ones. If that does not happens, something wrong is going on and it's better either to repeat the calibration or just file a bug report for the specific platform. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
655e3f03767c663faf7824970f5e15ac2f206596	28-Jun-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rtapp: fix calibration for versions of Android >=N Apparently, since Android N, by default when running a command via ADB the STDERR is not forwarded along with the STDOUT and thus the RTApp calibration fails because we are not able to parse the pLoad value. This patch force every execution of RTApp to append the STDERR on STDOUT thus granting to collect from the host side all the output generated by the application. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
e057189af57fbaf4a30363a19d4d51613d9d1f72	23-Jun-2016	Juri Lelli <juri.lelli@arm.com>	rta: do not accumulate delay Successive conf() calls were making rta accumulate delay parameter (since the scaled values was saved in the internal dict). This was also causing the value to eventually overflow. Fix it by always using the dict parameter directly to initialize phases. /external/lisa/libs/wlgen/wlgen/rta.py
ce0cfe357b0b31425366ffd61d8bb857407de720	02-Mar-2016	Juri Lelli <juri.lelli@arm.com>	libs/wlgen/wlgen/rta: fix calibration after rtapp refactoring Periodic workloads were promoted to Class status after Patrick's rtapp refactoring. Change RTA calibration accordingly. Signed-off-by: Juri Lelli <juri.lelli@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
1adfcc7584e51391d9880f213898c47782656298	26-Feb-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: add support for modular RTA phases For some experiments it could be useful to generate workload which change over time and that can be easily defined as a composition of "basic shapes". The RTA workload generator allows already to defined some common shapes (e.g. Periodic, Pulse, Ramp) but its missing the possibility to compose these shapes to create more complex patterns. This patch add this functionality by allowing to concatenate tasks definitions to generate a single Task with variable behaviors along time. An example of this new composition feature is provided by the sched_freq smoke test notebook, which has been update to use the new API to express the phases required by the test. For example, the light-heavy-light task is described as: light1 = Periodic(duty_cycle_pct=10, duration_s=1.0, period_ms= 10, cpus=str(target.bl.bigs_online[0])) heavy1 = Periodic(duty_cycle_pct=90, duration_s=0.1, period_ms=100) light2 = Periodic(duty_cycle_pct=10, duration_s=1.0, period_ms= 10) lhl_task = light1 + heavy1 + light2 rtapp.conf(kind='profile', params={'task1': lhl_task.get()}) Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
870281b5d9e90abd45f66ee5d4fbf22e4bb8732d	25-Feb-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: use a namedtuple to represent RTApp phases This patch is just to improve code readability. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
818ba5cbc332f75b2c94cc1db22c3038006f59fa	25-Feb-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: refactor RTA methods to be class objects Currently we generate specific workloads patterns by using a member function of the RTA class which returns a list of phases to execute. This approach has two main disadvantages: 1. the embedded documentation is not exposed to Notebooks 2. phases generated by a class cannot be easily "composed" to define a more complex workload This patch refactor these RTA methods to be standalone classes. This allows to get embedded documentation available on Notebooks and prepare the ground to introduce workloads shapes composition. To this purpose, all the shape classes are now derived from a _TaskBase class. That class provides the new _TaskBase::get() method which can be used to get the dictionary used to configure RTApp. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
4848d50976577b0c90e106d03827ab775f687847	25-Feb-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: move conf definition to simplify following refactoring This patch move the definition of the conf method to make more self contained the following refactoring patches. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
ebffae6ae68b4edd071325ba1287ef4ef0105456	05-Feb-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen: update log statements formatting to match execution engine one During tests execution, the output of the execution engine is going to be intermixed with the output of the workload generator. Right now the two modules uses a different formatting template which makes a bit difficult to read the generated log. This patch updates the formatting string of the workload generator modules to match that of the execution engine. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
489fa75b7a5a9326db56a4009137ba12d86830a1	27-Jan-2016	Juri Lelli <juri.lelli@gmail.com>	libs/wlgen/wlgen/rta: delay phase has to be named delay Proper implementation of delayed task activation is requires "delay" label in rt-app json grammar (where we currently have "sleep"). Fix that. Signed-off-by: Juri Lelli <juri.lelli@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
1da7aa59785e84d517ffdbde5fd3e8ee5ecf77aa	19-Jan-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: fix typo on function call Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
bcc455129fc831d4282f40399926f4b083f8310f	19-Jan-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: fix bug in getTargetCpu If an RTA does not defined a loadref value, by default we calibrate on first big CPU. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
f307bd6af7fd55eac51ed6abde36ab1623c85661	14-Jan-2016	Patrick Bellasi <patrick.bellasi@arm.com>	rt-app: properly format reported calibration values The calibration values either computed or loaded from the configuration file are reported in a non copy-and-past friendly manner. This patch ensure that the key are always reported enclosed in quotes so that it's possible to copy-and-past directly the reported string over a JSON configuration file. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
4c73d59e3778f5413dd2a79704e99ca778ef7124	13-Jan-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: add support for calibration on SMP systems On SMP systems we do not have the "bl" module available while still we need to be able to calibrate RTApp. This patch adds the required support to return a proper CPU ID on SMP systems based on this convention: 1. the first big cluster is considered to be the last system cluster 2. the first LITTLE cluster is considered ot be the first system cluster Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
fcad294dc741105e23914816c13957425ff23332	11-Jan-2016	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: report selected calibration CPU The CPU used for the calibration of the workload, i.e. defining the RTApp pvalue, is reported in the logfile of the generated workload. This allows to better verify with respect to which CPU the workload will be executed. Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
d95e98d88a6cf4347853427dc4ef6fd0222a3881	04-Dec-2015	Patrick Bellasi <patrick.bellasi@arm.com>	Add proper copyright headers to all the contributed sources Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
df88a074c62788e0bce3d94933d031b92500729f	11-Nov-2015	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: reorder function parameters to mach base class Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py
ee2022ef3e69fd39325a47af40d377516fac1c19	11-Nov-2015	Patrick Bellasi <patrick.bellasi@arm.com>	libs/wlgen/rta: remove dead code Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com> /external/lisa/libs/wlgen/wlgen/rta.py