Ensuring Update-After-Bind is supporting by the Device #492
Conversation
JeanPhilippeKernel
changed the title
JeanPhilippeKernel
added
vulkan-api
area-macOS
|
@MathewBensonCode It will be great if you can test the linux artificat of this PR :) |
|
Interesting you should ask this now as I was working on this yesterday. It works fine and the previous issue is not present. A new issue actually arises because of the change in Swapchain Image count affecting the commandbuffer manager and other related code. I Wasn't sure how to raise up this new issue when this pr wasn't yet merged. So I was going to wait. check out the |
Interesting ! I think you can open a PR against this branch, so we can iterate on your proposal - just checked and It seems correct but there are few adjustments. |
…ount (#495) * Re-Arrange Code flow after swapchainimage count change - The code that relies on the SwapchainImageCount such as the command buffer are re-arranged to ensure that they run after we establish the correct number of Swapchain Images * Pull Request #495 Requested Changes - Removed Redundant parameter to the CommandBufferManager.Initialize method - Moved the EnquedCommandBuffers.init method to be called right after the m_buffer_manager.Initialize function call --------- Co-authored-by: Mathew Benson <mathew@benson.co.ke>
|
@MathewBensonCode It'b be great if you can make a final testing on linux for this PR artificat - We should be pretty solid now on all platforms |
|
Linux is showing the UI as expected. Able to launch Obelisk through the Panzerfaust UI. |
|
On Windows, i'm observing a crash due to dangling pointers in Shader and RenderPass classes - mostly on how we manipulate/process string |
|
I tried this on windows yesterday and it worked ok. |
|
merging this since its confirmed working |
3 participants
In this PR, I revisited the process of Shader source reflexion to generate DescriptorSet and Layout.
I introduced the TextureArray as global binding across all shaders instead of per- shader.
I also revisited the logic around DescriptorPool size creation, which was consumed large resources than needed.
Then, I hardened the detection and support of Bindless resource of the Device so we can switch to a fallback option in case of no support.
Close #493