Get scene statistic from renderStats extended data
and use them instead of the artifically calculated ones.
This feature is supported from Qt 6.5 onwards.

Give some time for the newly loaded test to stabilize before
starting the performance measurement.

Note: works from Qt 6.3.0 onward

Note: It has to be compiled with Qt 5.15.x to be usable with
Qt 5.15.x version of the benchmarkdemo

Removed the global rotation and added local rotation to models instead.
This prevents models coming so close to the camera that it affects
the amount of fragment operations done drastically. Benchmark results
are more consistent between runs now.

Andy Nichols authored 3 years ago and Andy Nichols committed 3 years ago

Previously there was a lot of variablity in performance between runs on
certain hardware because models were spawned at random positions which
were differn't between runs. This variablity is visible when hardware is
bound by fragment operations.  When this is the case the more space a
model takes up on the screen as it rotates by the more fragments need to
be rasterized by the GPU.  If every run puts items in a differnt
position, it can be hard to get stable peformance results between runs,
and between differnt variations of the product.

This changes the MeshSpawner to a method for generating random numbers
that can be seeded to generate the same random numbers each run.
Math.random can not be seeded as there is no API to do so.

Split test scripts into smaller pieces. Originals were left
untouched, new ones are named _part#

This only builds the simple assets currently.

The performance will not be excellent, but it should
be runnable. Use "--mode demo --target embedded --preset entrylevel"
command line parameters to get all of the optimizations into use.