-
Notifications
You must be signed in to change notification settings - Fork 628
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Arm Release v1.3.0 #24
Conversation
1729e29
to
76fc543
Compare
samples/performance/layout_transitions/layout_transitions_tutorial.md
Outdated
Show resolved
Hide resolved
samples/performance/pipeline_barriers/pipeline_barriers_tutorial.md
Outdated
Show resolved
Hide resolved
samples/performance/specialization_constants/specialization_constants_tutorial.md
Outdated
Show resolved
Hide resolved
f3ab3a5
to
7f2d51d
Compare
Getting a crash during shader compilation in the Specialization Constants sample due to lack of versioning information:
Adding version qualifier #version 320 es (as per other project shaders) to shaders\specialization_constants\specialization_constant.frag and shaders\specialization_constants\UBOs.frag |
Broadcom are seeing issues with a number of the samples with this PR in place. I'll need some time to investigate. These samples are just showing a black screen with the UI on top: I get a std::bad_alloc from pipeline_cache command_buffer_usage now somehow manages to provoke a GPU lock-up! I also see lots of warnings about "Render target extent is not an optimal size" |
Interesting. I have tested this branch on an S10, S9, S9 Plus, S8 Plus, and a Huawei Mate 20 Pro and all these samples are working. Are you receiving any errors with validation layers turned on? Does a clean reinstall fix the problem? |
No validation errors that I can see. Clean install has the same problem. |
Also getting the Bad alloc error in the Pipeline management sample on S10 Adreno:
Stacktrace: PiplineManagement_BadAlloc_S10_Adreno.txt EDIT: As per Gary's suggestions, this is resolved by deleting the cache.data and pipeline_cache.data files. |
This went away for me when I deleted the /tmp/cache.data and /tmp/pipeline_cache.data files that the sample reads. |
So it does, thanks! This folder is located in /data/data/com.khronos.vulkan_samples/cache on Android. Deleting the files does indeed help. Perhaps there is a versioning issue? Wondering if this will become a common problem when infrastructure changes are made, and if a check is needed to ensure the cache matches the current version of the APK? |
Indeed. I agree that some version checking is probably needed. |
74f0951
to
14d4a8f
Compare
This issue is also affecting the Samsung Galaxy S8 Adreno (G950U). I am currently getting a black screen with UI on top in the following samples:
Also getting a crash inside the core validation layer when switching options in samples which re-create the swapchain (Swapchain Images, Surface Rotation):
|
Getting another crash on the S8 Adreno (G950U) in the Render Subpasses,Layout Transitions and Pipeline Barriers samples relating to a VK_INCOMPLETE return flag. Not sure why this is happening currently, it is apparently an invalid return flag from vkCreateGraphicsPipelines (which should not normally return that flag). This is likely a known driver bug: (https://developer.qualcomm.com/forum/qdn-forums/software/adreno-gpu-sdk/34709) However, I would ideally like to understand why it is happening -- as these samples worked previously.
I'll need to investigate this further.
By coincidence, there is also a validation error relating to the same previously discussed issue with valid format selection. This problem exists in all the samples which crash as above, and only in those samples.
It would be good if the used depth format could have a fallback option, if the given format is not available, as is done with the API examples. |
@MichaelParkinWhite Myself and @gary-sweet have narrowed down the problem to being to do with how dynamic descriptor sets are handled now (when we force to using static UBOs the samples work). I am going to get my hands on an Adreno device to try and replicate this issue so I can fix it. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Getting
$ vulkan-samples/framework/stats.cpp:46:54: error: no member named 'TransactionEliminations' in 'hwcpipe::GpuCounter'
{StatIndex::killed_tiles, {hwcpipe::GpuCounter::TransactionEliminations}},
Haven't had time to look into why yet
Have you tried updating your submodules? |
I have created an issue (#41) so we can solve this in the future. |
All of Broadcom's black render issues have been fixed with the latest patches. Thanks |
28199fd
to
f81c668
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks good!
Hi, we're getting a new crash in the latest build in the Command Buffer Usage sample, relating to RenderFrameReset:
|
@ZandroFargnoli the previous format issue has now been resolved, cheers :)! I also believe the black screen issue has been resolved as well. Though only have access to a limited set of devices right now, so will update on that later. However there appears to be a new validation issue (Samsung Galaxy S7 Adreno - G930V), which i'm assuming is associated with setting the STENCIL aspect flag when a format which does not have a stencil component is used?
This manifests as flickering and incorrect rendering in the 'Render Subpasses' and 'Pipeline Barriers' samples. The validation error also triggers in 'Layout Transitions' sample. |
I'm finding that I can only run the pipeline_cache example once after deleting /tmp/cache.data. Subsequent runs crash on startup until I delete /tmp/cache.data again. Is anyone else seeing the same? |
Same for me on an Intel IGP. Sample runs fine on first try and crashes on second run. Works again if I manually delete the cache data. |
@gary-sweet @SaschaWillems When fixing the validation error and crash Michael was experiencing I seemed to have fixed this problem. I am still trying to work out exactly why we are seeing this crash and will push the fixed code once I have done so |
1eb7e6b
c606e46
Updated liceneses and sample CMakeLists Updated sample tutorials to be more clear about mali specific content Ordering samples now removes duplicates from the order list Render granularity only displays the first time, and it is now calculated correctly Added missing headers to CMake, variant definitions are now OO, removed superfluous dynamic code Shader program removed and reverted back inside pipeline layout Removed FAQ to be added as a seperate MR Added contents to memory limits document Updated depth format function and integrated it into all existing depth buffer images
ef8a3e0
c0f116e
to
ef8a3e0
Compare
Description
Read the full release article for a more complete description of the PR.
Please read the contribution guidelines
Checklist:
Do not submit your PR without all of the below being checked: