| Age | Commit message (Collapse) | Author |
|
Change-Id: I1095edce0f26e2a4ced6fa301143622dc5cd0afd
|
|
Also straightened out the locking/unlocking for PushDescriptorSets.
Change-Id: Ic72bc693d16e7fa85df117d29abc1adc5a29ada6
|
|
Using PushDescriptorSets in a pipeline with normal Allocated descriptor
sets resulted in incorrect descriptor binding validation errors.
PushDescriptors are now run through the descriptor binding
state-update functions.
Change-Id: Ib702949af6de0b8638401c7eeb841c2590998e82
|
|
Change-Id: Ifabf7058b4ded33ab191a0043d9ef390b4c8cb06
|
|
Validation and state updates are now somewhat less tangled.
Change-Id: Ia666215a73ce02d2fbc326fb5b37b3361d83bbe3
|
|
Pulled from:
https://github.com/google/shaderc/blob/90a90d6f745704ac05afe70b7e8f759bec2ff493/known_good.json
|
|
The loader would return an unknown device extension handler when
specifying a function from a layer that was available, but not
enabled.
Change-Id: I0d77a7489d66e4d77aa8446018ec545c930f5bcf
|
|
Change-Id: Ic0777dda7b6958d2454dfec288630044e6f70db3
|
|
|
|
Complete support for cygwin environment by adding --no-sync support
to update_external_sources.bat. Also change command line parsing for
the the .bat script to be similar to the .sh script.
Change-Id: Ie813938654e7f4bb9575eef2587f36ca2933f30c
|
|
Partially support cygwin environment by adding --no-build to
update_external_sources.sh
Change-Id: I9fa35d6c720dd05a11c307d3da4d09415934899e
|
|
Change-Id: Icff4a71bbedc3ee95d2dc2b85224a3c2eee74cb6
|
|
Modify parameter validation codegen to handle multiple require blocks in
an extension, each with additional extension dependencies.
Change-Id: I336569ca19808f3dca342eb65fd076599c4fb329
|
|
|
|
Wrongly handled OOM error causes
dEQP-VK.api.object_management.alloc_callback_fail test to fail, because wrong
error code is returned.
|
|
|
|
This reverts commit 615dc4e3ea0495bc28130377b9e637e9979a560e.
|
|
This reverts commit 5d6aaa553b634747fb2761c6e5b665bdae54d206.
|
|
This reverts commit 7d35703bdc03d94cad272b5ea3ede8d586f57e88.
|
|
This reverts commit cfc1ed15584a12ef62e113c8daf673c6a68ce29c.
|
|
|
|
|
|
Parameter_validation's hammer for excluding some pdev-ish checks was
too large. This adds extension dependency checks for about 50 more
APIs.
Change-Id: Icb44cdb36c8c0384781e04e207ceb18d9f67ff0c
|
|
- updated vulkan.h
- updated vk.xml
- updated Win/Lin json files
- rebuilt and updated vulkan.hpp
- updated vk_validation_error_database.txt
- updated vk_validation_error_messages.h
- added new code-generated VUIDs (16) to error database
- modified object tracker for VUID corner-case
- updated vulkaninfo.c for STYPE name change
Change-Id: I0870e8bc4e84867e05bd466c13c7428b9b7365ac
|
|
|
|
Fixes #2020
Support from devs to remove this perf warning so just killing it.
Added a comment for future work where we can track which descriptors
are disturbed and reference that if a user attempts to use a
disturbed descriptor that isn't bound.
Also commented out associated test and filed #2022 to track enhancement
making note of disturbed descriptors if used unbound.
|
|
Change-Id: I8e25adabefde3a04be5536ce4c5c2d5920a76140
|
|
This commit adds vkjson and vkjson_info to the Android build.
It is only generated as an executable for devices with root
access, but it could be expanded to an APK with some more
boilerplate work.
See updates to BUILD.md for how to build and run on the device.
|
|
|
|
|
|
It's perfectly valid to bind the compute pipeline inside a renderpass.
An application only may not /dispatch/ compute work inside a
renderpass.
|
|
|
|
|
|
We won't actually check these at draw time since we have more precise
per-viewport and per-scissor support there, but we need these in order
to easily reject CmdSetViewport/CmdSetScissor when the pipeline doesn't
have the appropriate dynamic state enabled.
|
|
Removes app_dummy from apps that use native activity.
For more details refer to: https://github.com/android-ndk/ndk/issues/381
|
|
Previously the layout state in the primary command buffer was just
overwritten with the state in the secondary command buffer.
This can lead to spurious errors upon submission of the primary CB, if
an image used in the secondary CB is used prior to CmdExecuteCommands
in the primary CB with a different initial layout. This is because the
initialLayout field for the image on the primary CB is overwritten
with that of the secondary CB.
An example case that was incorrectly erroring (these commands refer to
the same image, newly created):
(primary buffer)
vkCmdPipelineBarrier, transition UNDEFINED -> TRANSFER_DST_OPTIMAL
vkCmdExecuteCommands
(secondary buffer)
...
vkCmdCopyBufferToImage, layout TRANSFER_DST_OPTIMAL
Just before the CmdExecuteCommands, the image had initialLayout =
UNDEFINED and layout = TRANSFER_DST_OPTIMAL on the primary CB, and
initialLayout = layout = TRANSFER_DST_OPTIMAL on the secondary.
After CmdExecuteCommands, the primary CB state is overwritten, so
it has initialLayout = layout = TRANSFER_DST_OPTIMAL. At submission,
this then errors because the global image layout is UNDEFINED.
Fix by only setting initialLayout on the primary CB from the secondary
if the image hasn't already been used on the primary.
Change-Id: Iae248ed09b1f87598fd192ba3d6f0bcf7ad38ada
|
|
When a secondary command buffer is executed, copy all functions in its
queue_submit_functions list to the primary command buffer.
This is needed e.g. to ensure that functions registered to validate and
update memory validity state for a command in the secondary CB get run
upon submission of the primary CB.
Fixes spurious memory invalid warnings on later commands after an
image's memory has been initially made valid inside a secondary CB.
Change-Id: I39d5a34f074942d5dedbf4c5a0295af957a92361
|
|
Couple of VUIDS were not added to the database.
Change-Id: I8ad8868182c8d129622d616241f5e0a14cb5054e
|
|
|
|
This gives us two advantages:
- We no longer need to recalculate the state bits to remove for the old
pipeline when binding a new pipeline.
- We now have the set of bits to test against when trying to bind
dynamic state for #614
|
|
Previously we'd only ever or in new bits, causing us to miss cases where
some state changes from static to dynamic within a CB.
Also only fiddle with the state masks when binding a graphics pipeline
-- previously we'd miss some other cases because binding a compute
pipeline would set all the state bits.
|
|
|
|
|
|
Simplifies steps necessary for new custom functions.
Change-Id: Ib2d80346ebc42011dce74fdd8cee62adb4b3432c
|
|
Changed the codegen to autogenerate validation functions for any new
APIs without user intervention. Manual updates are only necessary to
extend PV functionality.
- parameter_validation.h is now checked into the layers directory
- parameter_validation.cpp is now generated
- parameter_validation_utils.cpp is new and contains any custom
(non-generated) or housekeeping functions.
- parameter_validation_utils.h is deprecated
- updated database for new VUID coverage
Change-Id: Ib9261894386145573ba9e8906ba4ad4e9fecbdfd
|
|
Change-Id: Ibb2ac5b5f4b4176f5745007598c2fe64665237c2
|
|
When an object reaches the end of its "curve" path
it receives a new curve. To ease the transition from
the old curve to the new curve, the origin of the new
curve is the ending point of the old curve. However,
this can result in the starting point of the new curve
being out of the view. This results in objects slowing
leaving the view.
Add code to move the origin of the new curve back inside
the original view bounds only if the end of the old curve
is out of the view bounds. This does not add any additional
jarring because any object getting a new origin is already
outside the view anyway.
Change-Id: I50e88be453f31b9c37878f63c97d4c7c8154e718
|
|
- updated vulkan.h
- updated vk.xml
- updated cgenerator.py
- updated Win/Lin json files
- updated vulkan.hpp
- updated vk_validation_error_messages.h
- updated vk_validation_error_database.h
Change-Id: I8c525b6a2bd69914440c6d7984a67446b27e6c23
|
|
Change-Id: Id19d11e909ffc21c4aa6a419b071ba53594ae94f
|
|
Error in for loop condition
Change-Id: I8c775f961a6c05dd36e7acfdda6e94399365f90d
|