Conversation
marleo02
force-pushed
the
disable-external-compression
branch
2 times, most recently
from
fba9eda to
abd053c
Compare
marleo02
force-pushed
the
disable-external-compression
branch
from
abd053c to
0af559d
Compare
| VKAPI_ATTR VkResult VKAPI_CALL layer_vkCreateSwapchainKHR<user_tag>(VkDevice device, | ||
| const VkSwapchainCreateInfoKHR* pCreateInfo, | ||
| const VkAllocationCallbacks* pAllocator, | ||
| VkSwapchainKHR* pSwapchain |
There was a problem hiding this comment.
Closing brace on same line as last parameter to keep clang=tidy happy. Ditto below.
There was a problem hiding this comment.
Reformatted everywhere, I hope. Do we have something to check that all the formatting is compliant?
layer_gpu_support/source/layer_device_functions_allocate_memory.cpp
Outdated
Show resolved
Hide resolved
| if (ci->arrayLayers != 1) return false; // Might introduce false negative, stereo images are excluded | ||
|
|
||
| // Prefer common present-ish color formats. | ||
| switch (ci->format) { |
There was a problem hiding this comment.
Brace on new line for coding style. (Ditto in all the files, not tracking per instance).
| // -------------------------------------------- | ||
| // Heuristic + external export intent detection | ||
| // -------------------------------------------- | ||
| const bool present_like = heuristic_is_present_like_image(pCreateInfo); |
There was a problem hiding this comment.
Don't we know which images came from the swapchain via vkGetSwapchainImagesKHR()? Can't we just track those, and avoid the heuristic based on format alone?
There was a problem hiding this comment.
The need for this heuristic arose from a Vulkan application (specifically, gfxreconstruct) that was replaying a trace originally captured on an Android device using an AHB for presentation. When replayed on a Linux platform, no swapchain was created, and the AHB was ignored, resulting in no presentation at all.
We’ve since developed a patch in gfxreconstruct to address this issue: a valid swapchain is now created when this situation occurs. However, this change has not yet been merged into either the downstream or upstream official releases.
I suspect there may be other scenarios where such a heuristic could also prove useful. Please share your thoughts on whether we should keep it or remove it.
| VkImageCompressionControlEXT* compressionControl = &newCompressionControl; | ||
| auto append = [&](VkBaseOutStructure* n){ | ||
| n->pNext = nullptr; | ||
| if (!rebuilt_head) rebuilt_head = rebuilt_tail = n; |
There was a problem hiding this comment.
Coding style follows MISRA for bracing (if/else should always use braces, and body should be on different line to the condition). It's a lot more readable than long one-liners.
layer_gpu_support/source/layer_instance_functions_get_physical_properties.cpp
Outdated
Show resolved
Hide resolved
marleo02
force-pushed
the
disable-external-compression
branch
from
520cf16 to
cdcf94c
Compare
marleo02
force-pushed
the
disable-external-compression
branch
from
cdcf94c to
93221b9
Compare
2 participants
This PR allows to disable only external compression, while leaving internal compression untouched, it also include an heuristic in case there is no swapchain creation / presentation that acts on vkCreateImage. Presentations case handled: WSI Swapchain, DMA-buf import/export, AHB import/export. For DMA-buf and AHB import, an error is returned as it is not possible to control external compression with a Vulkan layer.