Commit Graph

8 Commits (28049f63210bf1ff427fe6907b2002e7159a03ca)

Author SHA1 Message Date
Chris Burel 377da0ed0c Ignore unused benchmark state variables
Signed-off-by: Chris Burel <burelc@amazon.com>
4 years ago
Mike Balfour 8b82041361
Optimize SurfaceData bulk queries (#7593)
* Add comparison operators to SurfaceTagWeight.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Changed AddSurfaceTagWeight to always combine weights.
This simplifies the API a bit and defines the behavior if someone ever tries to add a duplicate tag.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Added benchmarks for measuring the performance-critical APIs.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Changed SurfaceTagWeights to a fixed_vector.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Add inPosition to AddSurfacePoint.
This will be used to detect which input the surface point is associated with.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Add inPositionIndex to the appropriate APIs.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Switched Gradient Surface benchmarks to use actual surface components.
The gradient unit tests and benchmarks were previously using a mock surface data system, which led to misleading benchmark results. Now, the actual SurfaceData system gets constructed, and the tests use a mock provider, but the benchmarks use actual shape providers for more realistic benchmarking.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Fixed unit tests to have better query ranges.
Half of each previous range was querying outside the surface provider's data.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* First attempt at removing SurfacePointLists.
This currently runs significantly slower than the previous code but passes the unit tests.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Another attempt at optimization.
This one runs faster than the previous, but still slow.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Fix the cmake dependency so that the gradient tests rebuild SurfaceData.dll when run.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Switch SurfaceAltitudeGradient over to the new bulk API.
Also, optimized the non-bulk API by having it reuse the SurfacePointList to avoid the repeated allocation / deallocation cost.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Switched to using an indirect index so that all allocations are consecutive in our reserved buffer.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Switched back to SurfaceTagWeight again.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Added runtime dependency to LmbrCentral for unit tests.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Switched code over to use the full EnumeratePoints in most cases.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Added knowledge of max surface point creation into the system.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Add generic GetSurfacePointsFromList API implementation for surface providers.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Fixed implementation to use the correct maximum number of input points based on the surface providers being queried.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Fix out-of-bounds references on empty lists.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Fix memory allocation that caused benchmark runs to crash.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Starting to clean up the API.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Move SurfacePointList into separate files for easier maintainability.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Fixed bug where too many points were filtered out due to using the position Z as a part of the AABB check.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Made FilterPoints an internal part of SurfacePointList so we can choose when and how to perform the filtering.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Final cleanup / comments.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Added includes for non-unity builds.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Changed how unit tests initialize the mock lists to try and fix the linux errors.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Fixed compile error.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>
4 years ago
Chris Burel 664403c5de Mark benchmark state variables in for loops as unused in benchmarks
Signed-off-by: Chris Burel <burelc@amazon.com>
4 years ago
amzn-sj 396ec8a247
[Terrain] Optimize bulk queries to the Terrain System to retrieve height, surface weights, and normals (#7357) 4 years ago
Mike Balfour c46c558038
Switched Gradient Surface benchmarks to use actual surface components. (#7468)
* Switched Gradient Surface benchmarks to use actual surface components.
The gradient unit tests and benchmarks were previously using a mock surface data system, which led to misleading benchmark results. Now, the actual SurfaceData system gets constructed, and the tests use a mock provider, but the benchmarks use actual shape providers for more realistic benchmarking.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Fixed unit tests to have better query ranges.
Half of each previous range was querying outside the surface provider's data.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>
4 years ago
Mike Balfour a6ddf4164f
SurfaceTagWeights optimization (#7436)
* Add comparison operators to SurfaceTagWeight.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Changed AddSurfaceTagWeight to always combine weights.
This simplifies the API a bit and defines the behavior if someone ever tries to add a duplicate tag.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Added benchmarks for measuring the performance-critical APIs.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Changed SurfaceTagWeights to a fixed_vector.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>
4 years ago
Mike Balfour d9ba0af645
SurfacePoint data structure encapsulations (#7413)
* First pass at encapsulating SurfacePointList.
The biggest challenge in optimizing SurfacePointList(s) usage is the overall memory management associated with it. There are M surface points with N surface mask entries created for every input point, which leads to a lot of container reallocation and memory shuffling when processing multiple input points. By encapsulating the list, it should become easier to preallocate the entries, as well as keep "helper data" around for managing the bookkeeping to associate the input points with the output points.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Small fixes and TODO reminders.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Encapsulate surface point creation and separate EnumeratePoints out from modifications.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Start removing SurfacePoint from the exposed API.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Changed SurfacePointList to split out the surface point storage to allow for span<> usage over time.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Removed entity id

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Removed SurfacePoint from SurfaceData, changed all remaining uses to AzFramework::SurfaceData::SurfacePoint.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Encapsulated SurfaceTagWeightMap and renamed to SurfaceTagWeights.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Fixed make file.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Better commenting and parameter naming.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Renamed methods to be more descriptive.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>
4 years ago
Mike Balfour b975111a93
Add benchmarks and unit tests for GetSurfacePoints*. (#7216)
* Add benchmarks and unit tests for GetSurfacePoints*.
The benchmarks are very enlightening - the existing implementation of GetSurfacePointsFromRegion (and GetSurfacePointsFromList) is currently measurably *slower* than just calling GetSurfacePoints() many times in a loop.  This is due to all of the extra allocation overhead that's currently happening with the way these data structures are built.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Small syntax improvement

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Small update to the benchmark to use filtered results.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>

* Removed accidental extra include.

Signed-off-by: Mike Balfour <82224783+mbalfour-amzn@users.noreply.github.com>
4 years ago