samplerFilterMinmax enabled by default
brannegan opened this issue · 6 comments
Short info header:
- GFX version: gfx-backend-vulkan 0.7.0
- OS: Arch Linux
- GPU: GeForce GT 730
Gfx and wgpu-rs examples fails with "Unknown device creation error: ERROR_FEATURE_NOT_PRESENT".
Short investigation showed me that samplerFilterMinmax feature enabled by default here
but for my GPU this feature is not supported.
vulkaninfo cut:
VkPhysicalDeviceVulkan12Properties:
driverID = DRIVER_ID_NVIDIA_PROPRIETARY
driverName = NVIDIA
driverInfo = 460.56
conformanceVersion = 1.2.4.0
VkPhysicalDeviceVulkan12Features:
...
runtimeDescriptorArray = true
samplerFilterMinmax = false
scalarBlockLayout = true
...
Oh snap, thank you for filing!
So Vulkan 1.2 includes the extension in the core, but there is still a flag to check, which we missed.
Would you want to make a PR to fix this? That would be wonderful!
What happens when sampler_filter_minmax is not supported? Since it (and imagless_framebuffer) aren't in hal::Features, would a user of gfx be able to tell if the support is available?
@tangmi I was thinking that we'd just not expose ImageFeature::SAMPLED_MINMAX in this case in any format.
@kvark I assume the image format feature flags we get from Vulkan just won't have SAMPLED_IMAGE_FILTER_MINMAX if the device feature is disabled.
This was definitely my mistake from the device creation refactor, I think the PR I opened addresses it.
@tangmi I'm afraid this may be a bit trickier than this. Vulkan clients are expected to ignore the flags they aren't supposed to know about. So if the client (which is gfx-backend-vulkan in our case) doesn't request any sampler reduction feature/extension, they can still get the format feature bits, but they are supposed to ignore them :/
This means even with the PR you made, gfx may return the SAMPLED_IMAGE_FILTER_MINMAX feature on some formats, erroneously.