Update to wgpu 0.18 #10266

Elabajaba · 2023-10-25T20:20:33Z

Objective

Keep up to date with wgpu.

Solution

Update the wgpu version.

Currently blocked on naga_oil updating to naga 0.14 and releasing a new version.

3d scenes (or maybe any scene with lighting?) currently don't render anything due to

error: naga_oil bug, please file a report: composer failed to build a valid header: Type [2] '' is invalid
 = Capability Capabilities(CUBE_ARRAY_TEXTURES) is required

I'm not sure what should be passed in for wgpu::InstanceFlags, or if we want to make the gles3minorversion configurable (might be useful for debugging?)

Currently blocked on bevyengine/naga_oil#63, and gfx-rs/wgpu#4569 to be fixed upstream in wgpu first.

Known issues

Amd+windows+vulkan has issues with texture_binding_arrays (see the image here), but that'll be fixed in the next wgpu/naga version, and you can just use dx12 as a workaround for now (Amd+linux mesa+vulkan texture_binding_arrays are fixed though).

Changelog

Updated wgpu to 0.18, naga to 0.14.2, and naga_oil to 0.11.

Windows desktop GL should now be less painful as it no longer requires Angle.
You can now toggle shader validation and debug information for debug and release builds using WgpuSettings.instance_flags and InstanceFlags

Migration Guide

RenderPassDescriptor color_attachments (as well as RenderPassColorAttachment, and RenderPassDepthStencilAttachment) now use StoreOp::Store or StoreOp::Discard instead of a boolean to declare whether or not they should be stored.
RenderPassDescriptor now have timestamp_writes and occlusion_query_set fields. These can safely be set to None.
ComputePassDescriptor now have a timestamp_writes field. This can be set to None for now.
See the wgpu changelog for additional details

crates/bevy_render/src/lib.rs

crates/bevy_render/Cargo.toml

superdump · 2023-10-25T21:49:28Z

Cool, basically LGTM. :)

robtfm · 2023-11-16T11:29:49Z

wgpu/naga have updated, if you get a chance please check/approve the naga_oil pr and i'll merge and release (unless i need to do something further for the capabilities?)

Elabajaba · 2023-11-16T21:17:41Z

Rebased, fixed transmission, and it seems to work now.

Elabajaba · 2023-11-16T21:25:24Z

crates/bevy_render/src/render_resource/pipeline_cache.rs

@@ -197,6 +197,10 @@ impl ShaderCache {
            }
        }

+        // TODO: Check if this is supported, though I'm not sure if bevy works without this feature?
+        // We can't compile for native at least without it.
+        capabilities |= Capabilities::CUBE_ARRAY_TEXTURES;


We need this for native to compile, and webgl2 still works even with it enabled.

Ideally we feature gate it but I'm not sure how since it's a part of DownLevelFlags and doesn't seem to be exposed as a wgpu feature or limit.

if we want to check, we could pass the RenderAdapter down through PipelineCache::new and use adapter.get_downlevel_capabilities().flags. but i think it's ok not to since we already assume support in point shadow bindings anyway.

Hmm. Ok. We can fix it if it’s a problem anywhere.

robtfm · 2023-11-17T00:39:04Z

looks like we'll need at least one shader update too:

   ┌─ crates\bevy_pbr\src\render/utils.wgsl:35:1
   │
35 │ ╭ fn octahedral_encode(v: vec3<f32>) -> vec2<f32> {
36 │ │     var n = v / (abs(v.x) + abs(v.y) + abs(v.z));
37 │ │     let octahedral_wrap = (1.0 - abs(n.yx)) * select(vec2(-1.0), vec2(1.0), n.xy > 0.0);
   │ │                                                                             ^^^^^^^^^^ naga::Expression [26]
38 │ │     let n_xy = select(octahedral_wrap, n.xy, n.z >= 0.0);
39 │ │     return n_xy * 0.5 + 0.5;
   │ ╰────────────────────────────^ naga::Function [62]
   │
   = Expression [26] is invalid
   = Operation Greater can't work with [24] and [25]

works replacing the 0.0 with vec2(0.0)

Elabajaba · 2023-11-20T23:04:41Z

looks like we'll need at least one shader update too:

   ┌─ crates\bevy_pbr\src\render/utils.wgsl:35:1
   │
35 │ ╭ fn octahedral_encode(v: vec3<f32>) -> vec2<f32> {
36 │ │     var n = v / (abs(v.x) + abs(v.y) + abs(v.z));
37 │ │     let octahedral_wrap = (1.0 - abs(n.yx)) * select(vec2(-1.0), vec2(1.0), n.xy > 0.0);
   │ │                                                                             ^^^^^^^^^^ naga::Expression [26]
38 │ │     let n_xy = select(octahedral_wrap, n.xy, n.z >= 0.0);
39 │ │     return n_xy * 0.5 + 0.5;
   │ ╰────────────────────────────^ naga::Function [62]
   │
   = Expression [26] is invalid
   = Operation Greater can't work with [24] and [25]

works replacing the 0.0 with vec2(0.0)

It works, but it spits out a vulkan validation error when I do that.

Also the texture_binding_array example is broken on vulkan. (works fine on dx12 though)

superdump · 2023-11-22T16:32:04Z

I've seen mentions of the texture binding array bug on 0.12 I think? If so, it shouldn't block this PR as this PR does not cause the regression.

What is the vulkan validation error with robtfm's proposal?

Elabajaba · 2023-11-23T20:21:39Z

I've seen mentions of the texture binding array bug on 0.12 I think? If so, it shouldn't block this PR as this PR does not cause the regression.

It was an issue on AMD+Linux RADV drivers on 0.12. With wgpu 0.18 (this PR) it's fixed on AMD+Linux RADV, but broken on AMD+Windows (and apparently also broken on AMD+Linux amdvlk).

(amdvlk is AMD's official FOSS linux vulkan driver, but people generally use Mesa/RADV because it's better)

On Windows at least there's somewhat of a workaround because the issue doesn't happen on dx12.

What is the vulkan validation error with robtfm's proposal?

2023-11-23T20:19:56.968250Z ERROR log: VALIDATION [VUID-VkShaderModuleCreateInfo-pCode-01379 (0x2a1bf17f)]
        Validation Error: [ VUID-VkShaderModuleCreateInfo-pCode-01379 ] | MessageID = 0x2a1bf17f | SPIR-V module not valid: OpConstantComposite Constituent <id> count does not match Result Type <id> '5[%5]'s vector component count.
  %2173 = OpConstantComposite %v4float %329 %float_1
 The Vulkan spec states: If pCode is a pointer to GLSL code, it must be valid GLSL code written to the GL_KHR_vulkan_glsl GLSL extension specification (https://vulkan.lunarg.com/doc/view/1.3.261.1/windows/1.3-extensions/vkspec.html#VUID-VkShaderModuleCreateInfo-pCode-01379)

2023-11-23T20:19:58.192546Z ERROR log: VALIDATION [VUID-VkShaderModuleCreateInfo-pCode-01379 (0x2a1bf17f)]
        Validation Error: [ VUID-VkShaderModuleCreateInfo-pCode-01379 ] | MessageID = 0x2a1bf17f | SPIR-V module not valid: OpConstantComposite Constituent <id> count does not match Result Type <id> '3[%3]'s vector component count.
  %2814 = OpConstantComposite %v4float %717 %float_1
 The Vulkan spec states: If pCode is a pointer to GLSL code, it must be valid GLSL code written to the GL_KHR_vulkan_glsl GLSL extension specification (https://vulkan.lunarg.com/doc/view/1.3.261.1/windows/1.3-extensions/vkspec.html#VUID-VkShaderModuleCreateInfo-pCode-01379)

mockersf · 2023-11-24T22:53:27Z

I think we should merge this now and try to investigate / fix after

Elabajaba · 2023-11-25T01:03:39Z

There's an upstream patch that fixes the texture_binding_array example on windows+amd+vulkan.

I guess since deferred works despite the validation error, it's probably fine?

robtfm · 2023-11-25T11:27:47Z

this fixes the validation error. seems like it shouldn't be needed but 🤷

diff --git a/crates/bevy_pbr/src/deferred/pbr_deferred_functions.wgsl b/crates/bevy_pbr/src/deferred/pbr_deferred_functions.wgsl
index d401805eb..f888c0f30 100644
--- a/crates/bevy_pbr/src/deferred/pbr_deferred_functions.wgsl
+++ b/crates/bevy_pbr/src/deferred/pbr_deferred_functions.wgsl
@@ -71,7 +71,7 @@ fn pbr_input_from_deferred_gbuffer(frag_coord: vec4<f32>, gbuffer: vec4<u32>) ->
     let emissive = rgb9e5::rgb9e5_to_vec3_(gbuffer.g);
     if ((pbr.material.flags & STANDARD_MATERIAL_FLAGS_UNLIT_BIT) != 0u) {
         pbr.material.base_color = vec4(emissive, 1.0);
-        pbr.material.emissive = vec4(vec3(0.0), 1.0);
+        pbr.material.emissive = vec4(0.0, 0.0, 0.0, 1.0);
     } else {
         pbr.material.base_color = vec4(pow(base_rough.rgb, vec3(2.2)), 1.0);
         pbr.material.emissive = vec4(emissive, 1.0);

superdump · 2023-11-29T08:16:43Z

@robtfm so one can no longer construct a vecN from components up to vecM, M <= N?

robtfm · 2023-11-29T08:24:28Z

@robtfm so one can no longer construct a vecN from components up to vecM, M <= N?

it seems like some kind of inference issue in the particular case. just below we have pbr.material.emissive = vec4(emissive, 1.0); which doesn't trigger the error.

superdump · 2023-11-29T08:25:51Z

I’m curious if vec4(vec3(), ) works…

robtfm · 2023-11-29T08:28:46Z

I’m curious if vec4(vec3(), ) works…
^^^^ type can't be inferred

also doesn't work:
vec4(vec3<f32>(), 1.0) -> same validation error
vec4<f32>(vec3<f32>(0.0), 1.0); -> same validation error
let zero: vec3<f32> = 0.0; ... vec4(zero, 1.0); -> same validation error

Elabajaba · 2023-11-29T21:31:14Z

gfx-rs/wgpu#4695 is the upstream fix. Do we want this backported to avoid breaking downstream user's shaders? Does this actually break anything or is it just a validation error?

Elabajaba · 2023-12-06T22:10:11Z

gfx-rs/wgpu#4695 is the upstream fix. Do we want this backported to avoid breaking downstream user's shaders? Does this actually break anything or is it just a validation error?

Backported and released in naga 0.14.2, I think this should be ready to be merged assuming CI passes?

robtfm

looks good

alice-i-cecile added A-Rendering Drawing game state to the screen C-Dependencies A change to the crates that Bevy depends on S-Blocked This cannot move forward until something else changes labels Oct 25, 2023

superdump reviewed Oct 25, 2023

View reviewed changes

crates/bevy_render/src/lib.rs Outdated Show resolved Hide resolved

superdump reviewed Oct 25, 2023

View reviewed changes

crates/bevy_render/Cargo.toml Outdated Show resolved Hide resolved

Elabajaba mentioned this pull request Nov 1, 2023

Update naga requirement from 0.13.0 to 0.14.0 #10315

Closed

Elabajaba force-pushed the wgpu018 branch from 25b5a7d to 2944108 Compare November 16, 2023 21:00

Elabajaba commented Nov 16, 2023

View reviewed changes

Elabajaba force-pushed the wgpu018 branch from 19ae00f to 01414c9 Compare November 20, 2023 21:24

JMS55 added this to the 0.13 milestone Nov 23, 2023

Elabajaba force-pushed the wgpu018 branch from 3c5c538 to ff5519d Compare November 24, 2023 21:05

mockersf approved these changes Nov 24, 2023

View reviewed changes

Elabajaba marked this pull request as ready for review November 25, 2023 01:03

JMS55 approved these changes Nov 25, 2023

View reviewed changes

wgpu update

4eb1f32

Elabajaba added 11 commits December 6, 2023 17:05

storeop, timestamp_writes, and occlusion_query_set

318bd63

max_non_sampler_bindings limits

f8c39ab

make InstanceFlags and Gles3MinorVersion configurable

f4b4fee

missed naga_oil in bevy_pbr

7068c30

fix transmission

f33f989

capabilities (TODO: needs to be checked)

cbcc028

fmt

8c656ef

remove fixed todos

2fef4b7

use naga_oil 0.11

b8d1309

make deferred rendering work

8cb058a

force naga_oil 0.14.2

44358da

Elabajaba force-pushed the wgpu018 branch from ff5519d to 44358da Compare December 6, 2023 22:08

Elabajaba added 2 commits December 6, 2023 17:54

pub use the new wgpu settings

2f24b63

fmt

548f97d

Elabajaba requested review from robtfm and superdump December 13, 2023 22:21

robtfm approved these changes Dec 13, 2023

View reviewed changes

mockersf added S-Ready-For-Final-Review This PR has been approved by the community. It's ready for a maintainer to consider merging it and removed S-Blocked This cannot move forward until something else changes labels Dec 14, 2023

superdump approved these changes Dec 14, 2023

View reviewed changes

superdump added this pull request to the merge queue Dec 14, 2023

Merged via the queue into bevyengine:main with commit 70a592f Dec 14, 2023
27 checks passed

eerii mentioned this pull request Dec 21, 2023

Bevy 0.13 mvlabat/bevy_egui#236

Merged

nicopap mentioned this pull request Jan 29, 2024

0.13 announcement post bevyengine/bevy-website#891

Merged

sampettersson mentioned this pull request Feb 19, 2024

iOS Simulator not rendering due to missing CUBE_ARRAY_TEXTURES capability #11978

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Update to wgpu 0.18 #10266

Update to wgpu 0.18 #10266

Elabajaba commented Oct 25, 2023 •

edited

Loading

superdump commented Oct 25, 2023

robtfm commented Nov 16, 2023 •

edited

Loading

Elabajaba commented Nov 16, 2023

Elabajaba Nov 16, 2023

robtfm Nov 16, 2023

superdump Dec 14, 2023

robtfm commented Nov 17, 2023

Elabajaba commented Nov 20, 2023

superdump commented Nov 22, 2023

Elabajaba commented Nov 23, 2023 •

edited

Loading

mockersf commented Nov 24, 2023

Elabajaba commented Nov 25, 2023

robtfm commented Nov 25, 2023

superdump commented Nov 29, 2023

robtfm commented Nov 29, 2023

superdump commented Nov 29, 2023

robtfm commented Nov 29, 2023

Elabajaba commented Nov 29, 2023

Elabajaba commented Dec 6, 2023

robtfm left a comment

Update to wgpu 0.18 #10266

Update to wgpu 0.18 #10266

Conversation

Elabajaba commented Oct 25, 2023 • edited Loading

Objective

Solution

Known issues

Changelog

Migration Guide

superdump commented Oct 25, 2023

robtfm commented Nov 16, 2023 • edited Loading

Elabajaba commented Nov 16, 2023

Elabajaba Nov 16, 2023

Choose a reason for hiding this comment

robtfm Nov 16, 2023

Choose a reason for hiding this comment

superdump Dec 14, 2023

Choose a reason for hiding this comment

robtfm commented Nov 17, 2023

Elabajaba commented Nov 20, 2023

superdump commented Nov 22, 2023

Elabajaba commented Nov 23, 2023 • edited Loading

mockersf commented Nov 24, 2023

Elabajaba commented Nov 25, 2023

robtfm commented Nov 25, 2023

superdump commented Nov 29, 2023

robtfm commented Nov 29, 2023

superdump commented Nov 29, 2023

robtfm commented Nov 29, 2023

Elabajaba commented Nov 29, 2023

Elabajaba commented Dec 6, 2023

robtfm left a comment

Choose a reason for hiding this comment

Elabajaba commented Oct 25, 2023 •

edited

Loading

robtfm commented Nov 16, 2023 •

edited

Loading

Elabajaba commented Nov 23, 2023 •

edited

Loading