Re: float depth with stencil?

Thanks for the response (feel free to wait until the weekdays!). Some followup questions:

1) does it prefer 32 over 24 if available?
2) can you detect which you have?

I'm specifically concerned about implementing a reversed-z depth buffer:

https://nlguillemot.wordpress.com/2016/12/07/reversed-z-in-opengl/

I can fall back to multiple viewing volumes and 24-bit integer depth, but reversed-z is greatly preferred.

Thanks,
Rob

________________________________
From: Kai Ninomiya
Sent: Sunday, March 8, 2020 4:53 PM
To: Rob Conde
Cc: public-gpu
Subject: Re: float depth with stencil?

Our "depth24plus" maps to EITHER D24 or D32F. We still haven't worked out how this works with copies etc. though.

On Sun, Mar 8, 2020, 1:52 PM Kai Ninomiya <kainino@google.com<mailto:kainino@google.com>> wrote:
(Quick response; weekend)

Vulkan spec section:
https://renderdoc.org/vkspec_chunked/chap37.html#features-required-format-support

Table 65.

VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT feature must be supported for at least one of VK_FORMAT_X8_D24_UNORM_PACK32 and VK_FORMAT_D32_SFLOAT, and must be supported for at least one of VK_FORMAT_D24_UNORM_S8_UINT and VK_FORMAT_D32_SFLOAT_S8_UINT.

On Sun, Mar 8, 2020, 9:21 AM Rob Conde <rob.conde@ai-solutions.com<mailto:rob.conde@ai-solutions.com>> wrote:
I noticed a lack of something like GL_DEPTH32F_STENCIL8 in the draft spec.

https://gpuweb.github.io/gpuweb//#dom-gputextureformat-depth32float

Is the lack of a floating point depth buffer + stencil going to be a limitation of WebGPU, or is this accomplished through another means?

Thanks,
Rob Conde