Closed Bug 2085131 Opened 24 days ago Closed 24 days ago

Hardware acceleration issue : green flashes over videos

Categories

(Core :: Audio/Video: Playback, defect)

Firefox 160
defect

Tracking

()

RESOLVED DUPLICATE of bug 2084846

People

(Reporter: ezaacl35t, Unassigned)

Details

Attachments

(2 files)

Attached video Green.mp4 —

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:160.0) Gecko/20100101 Firefox/160.0

Steps to reproduce

Background :
Noticed issue after update 160.0a1 (20261021205324)
Updated to current version 160.0a1 (20261022084508), still noticing issue
Graphics card : AMD RX 6800
Drivers : Adrenalin 26.10.1

Steps:
Received update
Opened youtube
Searched for video
Attempted to play

Actual results

Playback corrupted (See attachement)
Message captured in about:support under graphics failure log
GP+[GFX1-]: RenderDXGITextureHost descriptor (426x240 fmt 16) does not match resource (512x256 DXGI 103)

Troubleshooted :
Turned hardware acceleration to off and restart firefox, issue resolves

Expected results

For playback to be normal and my eyes to not hurt :(

Issue doesn't replicate in other browsers (Edge, Firefox stable release)
Issue doesn't occur in troubleshooting mode
Issue doesn't occur with hardware acceleration turned on

I messed up
Under Expected Results, Issue doesn't occur with hardware acceleration turned off**

The Bugbug bot thinks this bug should belong to the 'Core::Audio/Video: Playback' component, and is moving the bug to that component. Please correct in case you think the bot is wrong.

Component: Untriaged → Audio/Video: Playback
Product: Firefox → Core

Can confirm this is also happening to me, with the latest RTX 3060 drivers (621.01). Issue does not happen in stable release.
When I looked into a fix, I found bug #1933755, and the fix mentioned in that still works. It's likely these are the same issue.

Can confirm. Currently on 160.0a1 (2026-10-22). Found this RESOLVED INCOMPLETE Bug 1864307, and following Bug 1864307 comment 18, I disabled gfx.direct3d11.reuse-decoder-device in about:config, restarted Firefox, and the green flashes are gone.

Attached file about:support —

Ty guys!

With hardware acceleration on, I tested turning off media.wmf.zero-copy-nv12-textures in about:config and that resolved the issue for me.
Turning off just gfx.direct3d11.reuse-decoder-device did not on my side.

I was curious about this issue and why that worked for you, so I asked Claude Opus 5.5 (on Extra) to assist in finding a root cause in your bug log and help point to a fix. Here are its findings for the development team:

What the failure log means
RenderDXGITextureHost descriptor (426x240 fmt 16) does not match resource (512x256 DXGI 103)
fmt 16 is SurfaceFormat::NV12 and DXGI 103 is DXGI_FORMAT_NV12, so the formats match. Only the size differs: 426x240 is the picture size, and 512x256 is the decoder's padded output surface.

Likely cause
In the zero-copy path, D3D11DXVA2Manager::WrapTextureWithImage wraps the decoder's own output texture and labels it with IntSize(mWidth, mHeight), i.e. the picture size.
RenderDXGITextureHost::EnsureD3D11Texture2D now requires the descriptor size to exactly match the resource's desc.Width/Height. The padded decoder surface fails this, mTexture is dropped, and Lock() returns an invalid image.
The copy path (CopyTextureToImage) allocates a texture at the picture size and uses CopySubresourceRegion into it, so it passes. That would explain why media.wmf.zero-copy-nv12-textures = false fixes it.

Why the reuse-decoder-device pref worked for me (NVIDIA) but not the reporter (AMD)
Zero-copy requires the decoder to share the compositor device; DXVA2Manager disables it otherwise. On NVIDIA, setting the pref to false gives the decoder its own device, so zero-copy turns off and the copy path is used. On AMD, gfxWindowsPlatform::InitReuseDecoderDeviceConfig force-enables reuse when the pref is off ("Always reused on AMD"). So zero-copy stays on and the check still fails. @reporter: your about:support may show REUSE_DECODER_DEVICE as force-enabled for that reason.

Workaround (all vendors): media.wmf.zero-copy-nv12-textures = false

Possible fix: let the check accept a resource at least as large as the descriptor (desc.Width >= mSize.width && desc.Height >= mSize.height), since padded decoder surfaces are expected.

Not confirmed which changeset added the check. mozregression between the 2026-10-20 and 2026-10-21 nightlies should pin it. Since the check lives in gfx/webrender_bindings, this may also belong in Core :: Graphics: WebRender.

Fixed in the current Nightly build

Status: UNCONFIRMED → RESOLVED
Closed: 24 days ago
Duplicate of bug: 2084846
Resolution: --- → DUPLICATE
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: