- Use ref types
- Support using rapidjson::Value in lieu of rapidjson::Document
- Use the existing JSON comparison util function in tests
Signed-off-by: Nicholas Van Sickle <nvsickle@amazon.com>
AZStd::fixed_vector had all its functions marked with constexpr, but this requires all member variables to be fully initialized. This meant that the internal array used to store elements always has to be fully initialized. This was done for trivial classes but not for non-trivial classes. As a result trivial classes always did a memset (or more optimized versions for smaller buffers) while the non-trivial version couldn't actually be stored in a constexpr variable. Since AZStd::fixed_vector is meant to be dynamic the choice was made to remove the constexpr from all non-static member functions in favor of avoiding the overhead of memset, which profiling showed was a considerable overhead depending on the reserved size. If a truly constexpr array is needed than AZStd::array is a better choice as that's designed to not by dynamic.
Signed-off-by: AMZN-koppersr <82230785+AMZN-koppersr@users.noreply.github.com>
* Renamed ArchiveLocationPriority enum to FileSearchPriority and made it a proper enum class
Added an ArchiveVars.cpp which checks the a new define: `LY_ARCHIVE_FILE_SEARCH_MODE_DEFAULT`
That define represents the default value to use for the Archive system search mode
Moved the FileSearchLocation enum to the ArchiveVars.h header
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* Updated the AssetBundleComponent to use AZ::IO::Path for level dirs
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* Added a LY_ARCHIVE_FILE_SEARCH_MODE cache variable
The Cache Variable default value is to Archive File Search Mode to PakOnly
in Release. This can be overridden using a value for all configurations
by specifying a number of 0, 1 or 2.
Alternatively a generator expression can be used to set the Archive File
Search Mode in specific configurations.
For example to set the FileSearchMode to 1 in profile and 2 in release
the following LY_ARCHIVE_FILE_SEARCH_MODE value can be used
`$<$<CONFIG:profile>:1>$<$<CONFIG:release>:2>`
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* Updated AssetBundler(Batch) VS Debugger arguments to populate the
project-path optoin if a single project is configured.
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* Add support for serializing older versions of the AssetBundleManifest
This is done by attaching the "ObjectStreamWriteElementOverride"
attribute to the AssetBundleManifest reflection.
That attribute contains a function which outputs an older serialized
version of the AssetBundleManifest based on the `m_bundleVersion` member
value.
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* AZStd::variant Serialization fix
The AttributeData<T> type is no longer suitable for storing the
ObjectStreamWriterOverrideCB function
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
This was due to incorrectly setting m_lastSimulationTickTimeUs when t_simulationTickDeltaOverride was enabled to the override.
m_lastSimulationTickTimeUs is now always set to the current time so the delta calculation will always be correct.
Signed-off-by: amzn-sean <75276488+amzn-sean@users.noreply.github.com>
- Replace AZ_TRAIT_MAX_JOB_MANAGER_WORKER_THREADS with AZ_TRAIT_THREAD_NUM_JOB_MANAGER_WORKER_THREADS that allows the number of threads created by the job manager to be set directly.
- Add AZ_TRAIT_THREAD_NUM_TASK_GRAPH_WORKER_THREADS that allows the number of threads created by the task graph to be set directly.
- Add a define that forces the AsyncUploadQueue to use the primary copy queue instead of creating a secondary copy queue.
Signed-off-by: bosnichd <bosnichd@amazon.com>
* Fix race condition where asset would finish loading and another request would start before the streamer request could be cleared
Signed-off-by: amzn-mike <80125227+amzn-mike@users.noreply.github.com>
* Re-enable disabled test which was failing for the same reason.
Fix test timeout which was way too long.
Reduce test iterations to keep test time safely under 5 seconds.
Signed-off-by: amzn-mike <80125227+amzn-mike@users.noreply.github.com>
* Instead of using a mutex, re-order the statements to remove the streamer request first
Signed-off-by: amzn-mike <80125227+amzn-mike@users.noreply.github.com>
Replaced and removed the CryTimer (gEnv->pTimer). The new TimeSystem is a merger of the current time functionality found in the engine.
* Rename TimeSystemComponent.h/.cpp to TimeSystem.h/.cpp
* Adding New TimeSystem
* remove old timer cvars
* small improvements to the time system.
- updated parts to use the time conversion functions.
- in AdvanceTickDeltaTimes applying t_simulationTickScale is now uses doubles instead of floats.
* Replace gEnv->pTimer / ITimer usages with TimeSystem
* Updating usages of AZ::TimeMs{ 0 } and AZ::TimeUs{ 0 } to AZ::Time::ZeroTimeMs and AZ::Time::ZeroTimeUs
* red code the CryTimer
* using TimeUs instead of TimeMs is some cases + updating usages of old cvars to new
Signed-off-by: amzn-sean <75276488+amzn-sean@users.noreply.github.com>
the owner has removed it!
This fixes the issue by forcing NameDictionary to not transfer
ownership. This means ComponentApplication()::Destroy will fully destroy
the NameDictionary before the OS::Allocator is destroyed.
In Windows the bug was not happening when running AssetProcessorBatch
because for Windows, _exit() is called before the application shutsdown
forcing all module to properly decrease the reference count of
EnvironmentVaqriableHolderBase::m_useCount for NameDictionary.
In MacOS, there's no _exit() so when the NameDictionary destructor was being called
before existing the application the reference count wouldn't be 0, and
would eventually try to destry the NameDictionary BUT the OS::Allocator
was already destroyed.
Signed-off-by: galibzon <66021303+galibzon@users.noreply.github.com>
This change adds a `DomBackend` interface and a `DomBackendRegistry` for backend discovery (currently not hooked up to anything) alongside a JSON backend implementation.
The JSON backend comes with a small suite of unit and performance tests. The unit tests validate generic DOM conversion to and from serialized JSON and rapidjson::Document objects (the support for which lives in `JsonSerializationUtils.h`).
The performance tests show a throughput decrease compared to directly using the rapidjson serializer when mirroring our current pattern of copying strings from the serialized JSON representation, but with the coming in-memory store we have the opportunity to keep the buffer in memory and deserialize in-situ using rapidjson's API, which is consistently at least 100MiB/s faster on my machine (Ryzen Threadripper 3970X). The first parameter is the nested object complexity (N*N, objects with N keys comprised of arrays with N values) and the second parameter is the base size of the strings within each entry of this object.
```
----------------------------------------------------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
----------------------------------------------------------------------------------------------------------------------
DomJsonBenchmark/DomDeserializeToDocumentInPlace/10/5 0.050 ms 0.050 ms 10000 bytes_per_second=355.816M/s
DomJsonBenchmark/DomDeserializeToDocumentInPlace/10/50 0.057 ms 0.057 ms 11200 bytes_per_second=386.064M/s
DomJsonBenchmark/DomDeserializeToDocumentInPlace/100/5 4.77 ms 4.88 ms 112 bytes_per_second=364.046M/s
DomJsonBenchmark/DomDeserializeToDocumentInPlace/100/500 11.6 ms 11.7 ms 64 bytes_per_second=554.518M/s
DomJsonBenchmark/DomDeserializeToDocumentWithCopies/10/5 0.084 ms 0.084 ms 7467 bytes_per_second=212.55M/s
DomJsonBenchmark/DomDeserializeToDocumentWithCopies/10/50 0.099 ms 0.100 ms 6400 bytes_per_second=220.608M/s
DomJsonBenchmark/DomDeserializeToDocumentWithCopies/100/5 8.22 ms 8.16 ms 90 bytes_per_second=217.847M/s
DomJsonBenchmark/DomDeserializeToDocumentWithCopies/100/500 23.2 ms 22.9 ms 30 bytes_per_second=283.56M/s
DomJsonBenchmark/JsonUtilsDeserializeToDocument/10/5 0.070 ms 0.070 ms 11200 bytes_per_second=255.049M/s
DomJsonBenchmark/JsonUtilsDeserializeToDocument/10/50 0.086 ms 0.087 ms 8960 bytes_per_second=253.258M/s
DomJsonBenchmark/JsonUtilsDeserializeToDocument/100/5 6.86 ms 6.84 ms 112 bytes_per_second=260.033M/s
DomJsonBenchmark/JsonUtilsDeserializeToDocument/100/500 22.8 ms 22.9 ms 32 bytes_per_second=283.158M/s
```
For `AZ::DOM::Document`, the current plan is to offer helper methods that can load from a file path or string using a given backend that can take advantage of in-place parsing by internally storing the serialized buffer.
Signed-off-by: Nicholas Van Sickle <nvsickle@amazon.com>
Nicholas Van Sickle
lumberyard-employee-dm
Tom Hulton-Harrop
AMZN-koppersr
Esteban Papp
amzn-mike
amzn-sean
Artur K
Tommy Walton
bosnichd
amzn-sj
rgba16f
galibzon