Unit test wgpu::example/msaa-line fails with panic on hardware: Intel(R) HD Graphics 4600 (HSW GT2) with driver: Mesa 21.2.6 with backend: Vulkan

[hsandt]

Description
The unit test wgpu::example/msaa-line failed on my machine (specs at the bottom).

Repro steps
On a machine with similar specs, run all unit tests:
$ cargo nextest run --no-fail-fast

Expected vs observed behavior

Expected: all tests pass

Actual:

1 failed test (excerpt just below, full log more below)

1 failed test:
FAIL [ 3.726s] wgpu::example/msaa-line msaa_line
test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 3.70s
...
--- STDERR: wgpu::example/msaa-line msaa_line ---
MESA-INTEL: warning: Haswell Vulkan support is incomplete
thread 'msaa_line' panicked at 'Image data mismatch! Outlier count 77230 over limit 65536. Max difference 255', wgpu/examples/msaa-line/../../tests/common/image.rs:134:13
note: run with RUST_BACKTRACE=1 environment variable to display a backtrace
thread 'msaa_line' panicked at 'UNEXPECTED TEST FAILURE DUE TO PANIC', wgpu/examples/msaa-line/../../tests/common/mod.rs:307:9
...
Summary [ 6.416s] 56 tests run: 55 passed, 1 failed, 0 skipped
FAIL [ 3.726s] wgpu::example/msaa-line msaa_line

Extra materials

Full test log:
nextest result on Intel(R) HD Graphics 4600 (HSW GT2) - Mesa 21.2.6 - Vulkan.txt

I also tried:
$ cargo run --bin wgpu-info -- cargo nextest run --no-fail-fast

in order to also test my discrete GPU, but it failed on linking, so I don't have test results for this one, unfortunately (that may be worth another ticket, but I must first ensure I setup everything correctly for this more complex test).

If you know a way to enforce using a specific GPU for the tests, I can also do that to verify that my discrete card (Nvidia GTX 860M) doesn't have the issue.

Platform
Linux Ubuntu 20.04 with Unity desktop
wgpu 0.14.2

Graphical hardware/driver info (AdapterInfo output from bevy engine):

AdapterInfo { name: "Intel(R) HD Graphics 4600 (HSW GT2)", vendor: 32902, device: 1046, device_type: IntegratedGpu, driver: "Intel open-source Mesa driver", driver_info: "Mesa 21.2.6", backend: Vulkan }

I got an integrated Intel(R) HD Graphics 4600 (HSW GT2) + a discrete Nvidia GTX 860M on my laptop, but the latter doesn't seem to be recognized by this particular program.

[cwfitzgerald]

Can you upload the resulting image? It will be saved in wgpu/examples/msaa-line/screenshot-failure.png or something similar

[hsandt]

Sure:

wgpu/examples/msaa-line/screenshot.png

wgpu/examples/msaa-line/screenshot-actual.png

Click on the picture above to open it in a new tab on white background, because of transparency you may not see much on dark theme. You should see something like this in the top-left area:

wgpu/examples/msaa-line/screenshot-difference.png

This is great info, it shows that this really is the same issue (or as least there is a common factor) as I got on Bevy: bevyengine/bevy#7232
as you can see the black and white image shrinked to the top-left part of the window in both case.

[cwfitzgerald]

Oh boy. I'm not sure we're going to be able to fix this in a meaningful way. Jason Ekstrand (one of the lead devs of intel/vulkan on linux) has said before that supporting vulkan on haswell was a mistake as it just can't be made compliant - and as such that driver will have a ton of bugs. What GLES version do you get from running glxinfo? We might consider disabling the vulkan backend on Haswell/Broadwell so it forces users onto the GLES drivers.

[hsandt]

I'm not familiar with this info dump but I think the important numbers are:
GLX version: 1.4
OpenGL ES 3.1

I put more detailed info below just in case:
$ glxinfo

name of display: :0
display: :0 screen: 0
direct rendering: Yes
server glx vendor string: SGI
server glx version string: 1.4
...
client glx vendor string: Mesa Project and SGI
client glx version string: 1.4
...
GLX version: 1.4
...
Extended renderer info (GLX_MESA_query_renderer):
Vendor: Intel Open Source Technology Center (0x8086)
Device: Mesa DRI Intel(R) HD Graphics 4600 (HSW GT2) (0x416)
Version: 21.2.6
Accelerated: yes
Video memory: 1536MB
Unified memory: yes
Preferred profile: core (0x1)
Max core profile version: 4.5
Max compat profile version: 3.0
Max GLES1 profile version: 1.1
Max GLES[23] profile version: 3.1
OpenGL vendor string: Intel Open Source Technology Center
OpenGL renderer string: Mesa DRI Intel(R) HD Graphics 4600 (HSW GT2)
OpenGL core profile version string: 4.5 (Core Profile) Mesa 21.2.6
OpenGL core profile shading language version string: 4.50
...
OpenGL version string: 3.0 Mesa 21.2.6
OpenGL shading language version string: 1.30
...
OpenGL ES profile version string: OpenGL ES 3.1 Mesa 21.2.6
OpenGL ES profile shading language version string: OpenGL ES GLSL ES 3.10
...

Yes that's fine, I'll be happy to fall back to OpenGL, whether automatically or by adding an environment variable.

[hsandt]

So I've tested environment variable WGPU_BACKEND=gl but for some reason it does not work with cargo nextest run --no-fail-fast. Do you know how to pass an env var, or alternatively a direct option to set a rendering backend, to nextest?

However, setting WGPU_BACKEND=gl when running a Bevy game directly did work and fixed my issue. So at least I'll be able to keep working on my project without the rendering bug.

[cwfitzgerald]

That should work, what error are you getting?

[hsandt]

Oh, in fact the backend does change and msaa_line passes this time.
However, I got another error with skybox:

FAIL [ 2.259s] wgpu::example/skybox skybox

--- STDOUT: wgpu::example/skybox skybox ---

running 1 test
0 outliers over max difference 1
test skybox ... FAILED

failures:

failures:
skybox

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 3 filtered out; finished in 2.25s

--- STDERR: wgpu::example/skybox skybox ---
thread 'skybox' panicked at 'UNEXPECTED TEST PASS: BACKEND', wgpu/examples/skybox/../../tests/common/mod.rs:305:9
note: run with RUST_BACKTRACE=1 environment variable to display a backtrace

I searched for existing issues and found this one: #1727
which is also about skybox and UNEXPECTED TEST PASS: BACKEND

However, the latter has been resolved. I tried RUST_BACKTRACE=1 and full too but the stack is approximately the same as in the ticket.

Worth opening another issue, or would you rather have me try prime-run and properly run on my discrete Nvidia GPU first?

By the way, I tried the second test command, cargo run --bin wgpu-info -- cargo nextest run --no-fail-fast again, and this time, it didn't fail on linking, it ran just like the first test command and failed "normally" on msaa_line.

I guess deleting the target folder in the meantime helped.

[teoxoy]

Haswell iGPUs are not yet Vulkan conformant:

https://gitlab.freedesktop.org/mesa/mesa/-/blob/bafc27583cf7c2d136fa05764e5e2145fedbc9cc/src/intel/vulkan_hasvk/anv_device.c#L1004-1024

You should report this issue to the mesa repo.