| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
|
| |
Change-Id: I5a4243b194611003e2b0a25a160d557c7f6d954f
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
|
|
|
|
|
| |
When we were bucketing results to try get better stability, this used to
give information about that. This was removed in f2ef76ce444db733fad3f6806f0019419d97b62c,
but this text remained. Remove it to avoid confusion.
Change-Id: I6a175807d920c3fe750b08b592bc412200c9a002
Reviewed-by: Mitch Curtis <mitch.curtis@qt.io>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
|
|
|
|
|
| |
Change-Id: I1497464741bbeb4f8ca680f9aeac95166aa81eb4
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
|
|
|
| |
Amends ab770345f2882782e412a0e8244b27887e1b4f22.
Change-Id: I4975e74113a6b28d4f36fc5cb1545496dd225ac3
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
|
|
|
|
|
|
|
|
|
| |
<cmath> needs to declare the math functions in the std namespace. Optionally it
can also put them in the global namespace, but for example the standard library
for QNX 6.6 does not do so.
Change-Id: Ie4d15d419d7a93c1c04d2b6982dba740d522740a
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Some tests, where the static count was set to 20000 would collect
all items along a diagonal line. It seems as if the reason for this
was that QML would evaluate all Xs first and then all Ys, and since
the static count was an exact multiple of the size of the
pregenerated array, the multiplier for X and Y would be the same,
so they would gather along the diagonal.
By setting the size of the array we make this unlikely (tests
with static count 20002 will still encounter the issue, but we
have none of those.)
Change-Id: I2d1df3c17f270a185e73e53d4ec9e2b56ddc799d
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In Qt 5.10 we introduce a new QRandomGenerator which is now
used by Math.random(), and there are signs that this is
significantly slower than qrand() which was used before. So
much so, that it is visible in many of the qmlbench benchmarks.
Note that this is a theory based on local runs and analysis of
the graphs and still needs to be proven.
Regardless, the qmlbench benchmarks are not intended to be
benchmarking Math.random(), it is just used to ensure the
positions tested are random. To eliminate that as a
possible cause for regressions, we instead introduce a
simple internal random function which just cycles through
a pregenerated array.
Note that this will reduce performance for all affected
tests on all branches of Qt, but it will hopefully reduce
the gap between Qt 5.9.0 and Qt 5.10.
Task-number: QTBUG-63922
Change-Id: Ib19c11323f38b23f9705e0d77a52a8a0f1157ac3
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
|
|
|
|
|
|
|
|
|
| |
Support for kerning/shaping was recently added to Qt Quick. These comparison
tests shows the performance benefit from disabling them, respectively.
Task-number: QTBUG-56728
Change-Id: I5a0cdf643b3f4d16980a832ecf9f79ebcad41d9b
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
|
|
|
|
| |
Custom properties on ShaderEffect can carry greater than normal
cost. Explicitly test a ShaderEffect with such a property.
Change-Id: Ib1344646585937f15a1c6c8b05806a4cea723994
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
|
|
|
|
| |
This allows you to override the default timeout of 10 minutes for
cases where one of the benchmarks is exceedingly slow, but you still
would like to collect the data for the others.
Change-Id: Ifcfc304a21e643d938e53bfd36ddaf6df6a8d515
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
|
|
|
|
|
|
|
| |
We should not count the warmup iterations in the overall number of
produced items. This seems wrong, and causes some confusion when
troubleshooting a broken test.
Note that a side effect of this change is that *all* results will now
drop by a value of 5 after this change is introduced.
Change-Id: Ib5e8a55411b663180f9d1a1a2a1df651bf0e4dc5
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
|
|
|
|
|
|
|
|
| |
This doesn't really have much point now that the split process mode has
stabilized
Change-Id: Ia88935b427d56ede40448d76bda7dbb261016827
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
| |
Change-Id: I00620d034c4d640e6f4a23ac1c890eec159067ad
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
| |
This had only a cosmetic effect of printing subprocess output with extra
linebreaks, but since it's not hard to fix, let's do that.
Change-Id: I5a81be5c29df84a5632fe2599ac31f35ae7e2f0f
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
| |
Change-Id: I07e675ff8acf7741e8a299b28f170624fccbdf13
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
These still weren't directly comparable (as they have different count
figures), and the change has been causing significant confusion when
examining results.
At the same time, we introduce apparently regressed QObject array benchmark as
a new test so we can track improvements in that area.
This reverts commit 45da7843d6c7b70ecd37c952fd68dd84f79c23e0.
Change-Id: Id76a0a3ca1bfea2ec18a88a8ffe2cca985992142
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
|
|
|
|
| |
This approach is valid, but it's essentially a coin toss as to whether
onTChanged or the Timer fires first, so some tests may still tick a
little too long with the Timer based approach.
Change-Id: Ic0d53b60cea7cabf87cd9bd8ad13b8d8bc8727b1
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
|
| |
We call quit(), but some tests may be executing QML for a bit longer,
not respecting the quit flag. Tear down the engine so that the quit
happens sooner. This fixes e.g. some of the animation tests running a
bit longer after measurement is over.
Change-Id: Ie05aff3d962d537c7f5d05a53f2de12522e9035d
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
|
| |
It's too slow to be of much use as-is on TX1 (even on the cheapest possible
workload, EarleyBoyer takes 3 minutes, and gives a very low result). So
let's leave this for manual runs for the time being, and perhaps rethink how we
run them in the future.
Change-Id: I5f5eaf7443d7b55cebec670747a600ddf91c05d1
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
|
|
|
|
|
| |
Change-Id: Ie91c6f16258b48cc9034193ff8def202e781a3b3
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
|
|
| |
We make use of the newly added V8Benchmark type to act as a harness for
the tests (doing setup, running, and teardown each qmlbench iteration).
We don't report the results v8-bench does, but instead, let qmlbench's
shell report a result.
Change-Id: Ice4a60a0ff08466ffe41cb6b596d75b86269cdbb
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
|
|
|
|
|
|
|
|
|
| |
This is going to be used to run v8-bench benchmarks using qmlbench as a harness.
This reverts commit 156cdf3b2ec213843f13240c160bcb25144a12c2.
Change-Id: Ieae04fd9fd968ac7bc419767afbf235b67242779
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
|
|
|
|
|
|
|
|
| |
We need to prevent writes to global properties. We also stub out the
BenchmarkSuite running stuff, as we will be doing that ourselves.
Change-Id: I647556f3d7a4e044db68754ce1f0eb0e35874536
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
|
|
|
|
|
|
|
| |
Imported from qtdeclarative/aa19268705f1a4f81ac6dcb519e217d40cefa633.
Change-Id: Ieb96d2ae70dca043cd15e5a622bef3082cdd6ffd
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
|
|
|
|
|
| |
Change-Id: I9120b41266be6c44e55c6fc9ec6a48e159f11840
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
| |
Change-Id: I6df853ccac46b770de1b0a1fcafcfd764422f250
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
| |
This scales a lot better, and is what we're doing with other tests.
Change-Id: I30772db9d72dbaf1ec65e3598c334c6c40573568
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In delegates_qobject, we now use a JS array to store items in before
deletion. This should make it a bit more heavy/slower.
In delegates_item_empty_jscreation, we remove the redundant parenting
(via 'root'), and just use the JS array. This should make it a bit
lighter/faster.
With these changes together, one can meaningfully get a bit of an idea
of the raw impact of allocating QQuickItem vs QObject.
Change-Id: I503a205760ff767b6ca8d42a2989755facaacb12
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
|
|
|
| |
And at the same time, introduce benchmarks for reading properties from a
TestModel instance.
This doesn't give us full coverage of model/view -- I'd ideally like to
see row additions/removals/moves on an otherwise static ListView for instance
-- but that can be done separately.
Change-Id: I45e25d89eb76bbe046d7e0c2c1a23f13b6b92fab
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
| |
Otherwise, due to queued signal handling, we may start processing
another (useless) run when 't' ticks before the queued signal to quit
kicks in.
Change-Id: I52cd0311e82363cec2f60ca19a85ff653a9b68fb
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
| |
Change-Id: I1e6a46cc8b85f63e29b1e80de938b6c0feba48d2
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
| |
Useful to determine if there's something funky going on in caching
somewhere
Change-Id: I359eabec3b740eff602898f71a035af169800d38
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
| |
Change-Id: I8d69bb1c0f9076ff9ee84f26cbb1981d36805ddd
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
|
|
|
|
|
|
|
|
|
|
|
| |
This reverts commit 16c6b173d5ad7e7d946a7d8dbae8fc8eda673bb2.
This reverts commit 0ecfdc41ebfabe50c816c6877192281bbf426cc8.
These are complete bunk after all. We recreate the shell each iteration,
so this has no impact at all.
Change-Id: I1693a1d4acbc18cbe0427422b2196398c83556da
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
| |
* Improve formatting
* Fix/remove outdated bits and pieces, and move some information like
test names out of the README (since they grow stale easily).
Change-Id: I70460b785f3112b47e6a4cb6cbe096e88781c2af
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
| |
Change-Id: Iaa99be3ef1b82aa62a2a54aa9ae55d4a71c50b8b
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
|
|
|
| |
This may contribute to some test instability, although I see no real
difference locally.
Change-Id: I078029e4abec08aaaf1347a2e028d38d2c43ac70
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
| |
This way, we can enable it while still outputting json to stdout, which
means we have feedback on what is going on.
Change-Id: I1fd65464c410ed61077aa664789a86e84dc57d96
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
| |
Change-Id: I424e7fd0c2b93541c37f7857a2a00cfe098b4aef
Reviewed-by: Robin Burchell <robin.burchell@crimson.no>
|
|
|
|
|
|
|
|
| |
The childrenRect test is quite simplistic in that it doesn't try to have
a large nested hierarchy, but this is better than no coverage.
Change-Id: I53a2c213925f14249f2fd9a042fc993f354a3cb0
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
|
|
|
| |
This time around, we're going totally bare-bones. A global counter is
used to tick animations for everything else, no animations, no per-item
script handlers or anything.
Comparing the results of this with the others may help reveal where any
remaining problems are now.
Change-Id: Idefe7ac43d05e6b091cd0b6cb28f585c35f6138e
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Remove Grid and sizing stuff, use the bare minimum here. Create items
once, position them, and run animations repeatedly.
Also remove the PauseAnimation (which in itself is hard on predictable
work, which the benchmark kind of relies on) and Math.random (which
makes that even more hard to predict and not easily reproduced)
Change-Id: I3e74dca1ab8a302048575dcfb607cb84c2d2e822
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
| |
These are actually treated differently, so it makes sense to test.
Change-Id: Ibeea3e2003f7d425e1aa62bffd0837107cd47d9f
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
| |
Change-Id: I91238e768cb0ce84ac63db897c059873aa4af641
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
| |
Change-Id: I1ee45c17afc5eec86130c54188d70c04b57335dd
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
| |
Reasonable amount of these, so let's give them their own place. Also
rename delegates_blendedrect to delegates_rect_blended so it fits the
pattern.
Change-Id: Ib939641de723b592b38d771e2b08c8214faf1b11
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
| |
These are used reasonably often, and easy to test, so cover it.
Change-Id: I2a1580bf9666199da8bc957191b0aa6fa96c4ead
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
| |
This is a different codepath, often used, so it is a good idea to test it.
Change-Id: If30922dbd24df818cd4a62c6ec2076c1ad057884
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
| |
We have two variants where possible: the "complex" version, and the
"simple" version.
Change-Id: I57f023a4f24f355272f21c568203fc46d7f3a01a
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|
|
|
|
|
|
|
|
|
| |
The performance of using states is around half that of using a simple
binding (which isn't too surprising given all the extra creation and
allocation involved).
Change-Id: I4f12819a71ef88e8f68cc83ae00b1726ca74609e
Reviewed-by: Gunnar Sletta <gunnar@crimson.no>
|