[Merged by Bors] - bevy_pbr: Use a special first depth slice for clustered forward #3545

superdump · 2022-01-04T11:34:30Z

Objective

Using plain exponential depth slicing for perspective projection cameras results in unnecessarily many slices very close together close to the camera. If the camera is then moved close to a collection of point lights, they will likely exhaust the available uniform buffer space for the lists of which lights affect which clusters.

Solution

A simple solution to this is to use a different near plane value for the depth slicing and set it to where the first slice's far plane should be. The default value is 5 and works well. This results in the configured number of depth slices, maintains the exponential slicing beyond the initial slice, and no slices are too small such that they cause problems that are sensitive to the view position.

superdump · 2022-01-04T11:49:44Z

Thanks to @aclysma for pointing out a simple way of doing this.

Using plain exponential depth slicing for perspective projection cameras results in unnecessarily many slices very close together close to the camera. If the camera is then moved close to a collection of point lights, they will likely exhaust the available uniform buffer space for the lists of which lights affect which clusters. A simple solution to this is to use a different near plane value for the depth slicing and set it to where the first slice's far plane should be. The default value is 5 and works well. This results in the configured number of depth slices, maintains the exponential slicing beyond the initial slice, and no slices are too small such that they cause problems that are sensitive to the view position.

…e pbr shader

cart · 2022-01-07T21:22:51Z

Works for me / generally makes sense. Thanks!

cart · 2022-01-07T21:22:56Z

bors r+

# Objective - Using plain exponential depth slicing for perspective projection cameras results in unnecessarily many slices very close together close to the camera. If the camera is then moved close to a collection of point lights, they will likely exhaust the available uniform buffer space for the lists of which lights affect which clusters. ## Solution - A simple solution to this is to use a different near plane value for the depth slicing and set it to where the first slice's far plane should be. The default value is 5 and works well. This results in the configured number of depth slices, maintains the exponential slicing beyond the initial slice, and no slices are too small such that they cause problems that are sensitive to the view position.

bors · 2022-01-07T21:44:24Z

Pull request successfully merged into main.

Build succeeded:

# Objective We keep getting issues where things break at small window sizes, e.g #3368 (caused by #3153), #3596 ('caused' by #3545) ## Solution - Add a test that we can make small windows. Currently, this fails on my machine with some quite scary vulkan errors: ``` 2022-01-08T22:55:13.770261Z ERROR wgpu_hal::vulkan::instance: VALIDATION [VUID-VkSwapchainCreateInfoKHR-imageExtent-01274 (0x7cd0911d)] Validation Error: [ VUID-VkSwapchainCreateInfoKHR-imageExtent-01274 ] Object 0: handle = 0x1adbd410a60, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0x7cd0911d | vkCreateSwapchainKHR() called with imageExtent = (225,60), which is outside the bounds returned by vkGetPhysicalDeviceSurfaceCapabilitiesKHR(): currentExtent = (225,56), minImageExtent = (225,56), maxImageExtent = (225,56). The Vulkan spec states: imageExtent must be between minImageExtent and maxImageExtent, inclusive, where minImageExtent and maxImageExtent are members of the VkSurfaceCapabilitiesKHR structure returned by vkGetPhysicalDeviceSurfaceCapabilitiesKHR for the surface (https://vulkan.lunarg.com/doc/view/1.2.198.1/windows/1.2-extensions/vkspec.html#VUID-VkSwapchainCreateInfoKHR-imageExtent-01274) 2022-01-08T22:55:13.770808Z ERROR wgpu_hal::vulkan::instance: objects: (type: DEVICE, hndl: 0x1adbd410a60, name: ?) 2022-01-08T22:55:13.787403Z ERROR wgpu_hal::vulkan::instance: VALIDATION [VUID-VkSwapchainCreateInfoKHR-imageExtent-01274 (0x7cd0911d)] Validation Error: [ VUID-VkSwapchainCreateInfoKHR-imageExtent-01274 ] Object 0: handle = 0x1adbd410a60, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0x7cd0911d | vkCreateSwapchainKHR() called with imageExtent = (225,56), which is outside the bounds returned by vkGetPhysicalDeviceSurfaceCapabilitiesKHR(): currentExtent = (225,52), minImageExtent = (225,52), maxImageExtent = (225,52). The Vulkan spec states: imageExtent must be between minImageExtent and maxImageExtent, inclusive, where minImageExtent and maxImageExtent are members of the VkSurfaceCapabilitiesKHR structure returned by vkGetPhysicalDeviceSurfaceCapabilitiesKHR for the surface (https://vulkan.lunarg.com/doc/view/1.2.198.1/windows/1.2-extensions/vkspec.html#VUID-VkSwapchainCreateInfoKHR-imageExtent-01274) ``` etc. This might be a new issue here, although I'm surprised it's vulkan giving this error; wgpu should stop it if this is illegal.

# Objective We keep getting issues where things break at small window sizes, e.g bevyengine#3368 (caused by bevyengine#3153), bevyengine#3596 ('caused' by bevyengine#3545) ## Solution - Add a test that we can make small windows. Currently, this fails on my machine with some quite scary vulkan errors: ``` 2022-01-08T22:55:13.770261Z ERROR wgpu_hal::vulkan::instance: VALIDATION [VUID-VkSwapchainCreateInfoKHR-imageExtent-01274 (0x7cd0911d)] Validation Error: [ VUID-VkSwapchainCreateInfoKHR-imageExtent-01274 ] Object 0: handle = 0x1adbd410a60, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0x7cd0911d | vkCreateSwapchainKHR() called with imageExtent = (225,60), which is outside the bounds returned by vkGetPhysicalDeviceSurfaceCapabilitiesKHR(): currentExtent = (225,56), minImageExtent = (225,56), maxImageExtent = (225,56). The Vulkan spec states: imageExtent must be between minImageExtent and maxImageExtent, inclusive, where minImageExtent and maxImageExtent are members of the VkSurfaceCapabilitiesKHR structure returned by vkGetPhysicalDeviceSurfaceCapabilitiesKHR for the surface (https://vulkan.lunarg.com/doc/view/1.2.198.1/windows/1.2-extensions/vkspec.html#VUID-VkSwapchainCreateInfoKHR-imageExtent-01274) 2022-01-08T22:55:13.770808Z ERROR wgpu_hal::vulkan::instance: objects: (type: DEVICE, hndl: 0x1adbd410a60, name: ?) 2022-01-08T22:55:13.787403Z ERROR wgpu_hal::vulkan::instance: VALIDATION [VUID-VkSwapchainCreateInfoKHR-imageExtent-01274 (0x7cd0911d)] Validation Error: [ VUID-VkSwapchainCreateInfoKHR-imageExtent-01274 ] Object 0: handle = 0x1adbd410a60, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0x7cd0911d | vkCreateSwapchainKHR() called with imageExtent = (225,56), which is outside the bounds returned by vkGetPhysicalDeviceSurfaceCapabilitiesKHR(): currentExtent = (225,52), minImageExtent = (225,52), maxImageExtent = (225,52). The Vulkan spec states: imageExtent must be between minImageExtent and maxImageExtent, inclusive, where minImageExtent and maxImageExtent are members of the VkSurfaceCapabilitiesKHR structure returned by vkGetPhysicalDeviceSurfaceCapabilitiesKHR for the surface (https://vulkan.lunarg.com/doc/view/1.2.198.1/windows/1.2-extensions/vkspec.html#VUID-VkSwapchainCreateInfoKHR-imageExtent-01274) ``` etc. This might be a new issue here, although I'm surprised it's vulkan giving this error; wgpu should stop it if this is illegal.

github-actions bot added the S-Needs-Triage This issue needs to be labelled label Jan 4, 2022

superdump requested a review from cart January 4, 2022 11:35

superdump added A-Rendering Drawing game state to the screen C-Enhancement A new feature and removed S-Needs-Triage This issue needs to be labelled labels Jan 4, 2022

cart added this to the Bevy 0.6 milestone Jan 4, 2022

superdump force-pushed the special-nearest-cluster branch from 1ca12f1 to 77877fb Compare January 5, 2022 13:04

bevy_pbr: Restrict cluster index to the valid range to avoid UB in th…

a1ac77c

…e pbr shader

bors bot changed the title ~~bevy_pbr: Use a special first depth slice for clustered forward~~ [Merged by Bors] - bevy_pbr: Use a special first depth slice for clustered forward Jan 7, 2022

bors bot closed this Jan 7, 2022

DJMcNab mentioned this pull request Jan 8, 2022

[Merged by Bors] - Add an example to test small window sizes #3597

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Merged by Bors] - bevy_pbr: Use a special first depth slice for clustered forward #3545

[Merged by Bors] - bevy_pbr: Use a special first depth slice for clustered forward #3545

superdump commented Jan 4, 2022

superdump commented Jan 4, 2022

cart commented Jan 7, 2022

cart commented Jan 7, 2022

bors bot commented Jan 7, 2022

[Merged by Bors] - bevy_pbr: Use a special first depth slice for clustered forward #3545

[Merged by Bors] - bevy_pbr: Use a special first depth slice for clustered forward #3545

Conversation

superdump commented Jan 4, 2022

Objective

Solution

superdump commented Jan 4, 2022

cart commented Jan 7, 2022

cart commented Jan 7, 2022

bors bot commented Jan 7, 2022