The "precise" keyword was recently added to several shader inputs, but one was missed. This led to inconsistency between the depth and forward passes which resulted in artifacts when parallax pixel depth offset was in use. See https://github.com/o3de/o3de/pull/6536
Testing: I used AtomSampleViewer to make a local baseline of all screenshots in _fulltestsuite_.bv.lua. The only change was to the parallax test cases, which were a clear improvement.
Signed-off-by: santorac <55155825+santorac@users.noreply.github.com>
* Fix input/output params not being zero-based
* Expose missing Dump Database developer command
* Fix issue where developer gem name is not the same as the target
* Let the user specify folder to write to
Signed-off-by: Alex Peterson <26804013+AMZN-alexpete@users.noreply.github.com>
* moving material editor test from sandbox to main suite
Signed-off-by: Scott Murray <scottmur@amazon.com>
* define TEST_DIRECTORY and add logging
Signed-off-by: Scott Murray <scottmur@amazon.com>
Previously it was using `GetTickCount()` for the creation of the
directory which can collide if0 there are multiple processes
creating a temporary directory via teh ScopedAutoTempDirectory
constructor.
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* Fixed CMake use with Android NDK 23
This was done by only setting the `ANDROID_ARM_MODE` if the ANDROID_ABI
starts with `armeabi`.
The Android arm mode setting can't be used using non-`armeabi` ABIs.
In O3DE we default to `arm64-v8a` `ANDROID_ABI`
So `ANDROID_ARM_MODE` must not be set.
The CMake [Android-Determine.cmake](https://gitlab.kitware.com/cmake/cmake/-/blob/master/Modules/Platform/Android-Determine.cmake#L573-585) which is used to detect platform-wide information when the CMAKE_SYSTEM_NAME is set to android, enforces that if the `CMAKE_ANDROID_ARCH_ABI` doesn't start with `armeabi`, then it will fatal error if the CMAKE_ANDROID_ARM_MODE option is set.
In Android NDK 21 the `ANDROID_ARM_MODE` variable is used to set the CMAKE_ANDROID_ARM_MODE variable if the [ANDROID_ABI](https://android.googlesource.com/platform/ndk/+/refs/tags/ndk-r21e/build/cmake/android.toolchain.cmake#700) starts with `armeabi`.
This meant when using Android NDK 21, the CMake Android-Determine.cmake module would succeed, due
to the CMAKE_ANDROID_ARM_MODE not being set.
In Android NDK 23 the `ANDROID_ARM_MODE` now will set the
`CMAKE_ANDROID_ARM_MODE` variable to `TRUE` if it isn't defined.
Added an `--extra-cmake-configure-args` option to the `generate_android_project.py` script which can be used to append user specified CMake arguments to the cmake configure step (`cmake -B <build-dir> -S <source-dir> -DCMAKE_TOOLCHAIN_FILE=<android-toolchain-file> <extra-args>`)
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* Updated the O3DE Android toolchain wrapper to fatal error if the 64-bit
arm ABI isn't used.
Also removed the unneccessary setting of the ANDROID_ARM_MODE and
ANDROID_ARM_NEON option option now that the armeabi cannot be specified
as an ABI.
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
The CriticalAssetsCompiled event can be handled to detect when the
AssetProcessor has finished processing Critical Assets
Also with the new event, an audit has been performed over all the
locations where the AssetCatalogEventBus OnCatalogLoaded event was being
handle to make sure it was the proper event to use.
If the handler was actually examing the enumerating over the full
catalog or querying all assets within the catalog, then it was a proper
use.
For handlers that were interested in a particular asset it was not
Moreover added implementations of `OnCatalogAssetChanged` and
`OnCatalogAssetAdded` to the FileTagComponent and the MaterialViewportComponent.
Any applications which uses the AtomToolsApplication
class(MaterialEditor, AtomSampleViewerStandalone,
ShaderMangementConsole) now signals a "CriticalAssetsCompiled" lifecycle
event as well as loads the "assetcatalog.xml" if it exists.
The Launcher application signals the "CrticalAssetsCompiled" event and
reloads the "assetcatalog.xml" for the ${project}.GameLauncher and
${project}.ServerLauncher in Launcher.cpp
Finally the Editor signals the "CriticalAssetsCompiled" and reloads the
"assetcatalog.xml" in CryEdit.cpp
resolves#6093
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* Replaced TestDependenciesLevel's level.pak for TestDependenciesLevel.prefab to fix asset_bundler_batch_tests failure
Updated asset_bundler_batch_tests to reflect the update.
Signed-off-by: Rosario Cox <lesaelr@amazon.com>
* Missed one of the .spawnables changes
Signed-off-by: Rosario Cox <lesaelr@amazon.com>
* AZStd::basic_string improvements
The AZStd::basic_string class has a better implementation of the Short
String Optimization, which increases the amount of characters that can
be stored in a `basic_string<char>` from 15 characters to 22
characters(not-including null-terminating characters). For a
`basic_string<wchar_t>` on Windows the amount of characters that can be
stored increases from 7 to 10. Using `basic_string<wchar_t>` on Unix
platforms SSO character amount from 3 to 4 characters.
An additional benefit is that the size of the AZStd::basic_string class
has been reduced from 40 bytes to 32 bytes when using the
AZStd::allocator.
When using a stateless allocator with no non static data members such as
AZStd::stateless_allocator, the size of the AZStd::basic_string is 24
bytes.
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* Corrected comments and updated type alias to usings for AZStd::basic_string
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* Added Benchmarks for the basic_string and basic_fixed_string class
The benchmarks currently measure the speed of the `assign` overloads.
A benchmark has also been added to compare the speed swapping two
`basic_string` instances by 3 memcpy vs 3 pointer swap operations
Speed up string operation when in the iterator overload cases of the
`assign`, `append`, `insert` and `replace` function.
The code was always performing the logic to copy over a string that is
overlapping, without actually checking if the string was overlapping in
the first place.
Added an `az_builtin_is_constant_evaluated` macro that allows use of the
C++20 `std::is_constant_evaluated` feature to determine if an operation
is being performed at compile time vs run time.
That macro is being used to speed up the char_trait operations at run
time, by using the faster standard library functions.
For example char_traits::move now uses "memmove" at runtime, instead of
a for loop.
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
* Simplified string logic in AWSMetricsServiceApiTest.
Signed-off-by: lumberyard-employee-dm <56135373+lumberyard-employee-dm@users.noreply.github.com>
Script canvas component has the member m_sourceData (that points to the script canvas asset) that internally has data, id and path. Path is serialized as the absolute path of the pc that is saving the level.
So when another pc loads the same level it cannot find the script canvas asset.
The function CompleteDescription takes a look at the id and takes the path from the asset catalog, but at the moment it's doing an early return because id and path are not empty (but path is the value serialized from other user saving the level).
By removing the early return condition then it it will resolve by using id, looking into the catalog and getting the real path.
This fix makes several physics automated tests that relied on script canvas to work.
Signed-off-by: moraaar moraaar@amazon.com
jromnoa
santorac
Mike Balfour
tjmgd
sphrose
Vincent Liu
srikappa-amzn
Alex Peterson
Chris Galvan
SWMasterson
jckand-amzn
smurly
SergeyAMZN
lumberyard-employee-dm
Sergey Pereslavtsev
michabr
LesaelR
Esteban Papp
John Jones-Steele
Benjamin Jillich
moraaar
Junbo Liang