[Bug]: RX 7000 series encoding broken with CreateSurfaceFromDX11Native on driver 25.10.2 / AMF 1.5.0

[Esvandiary]

Describe the bug
H.264 / HEVC / AV1 encoding results in broken output after updating to 25.10.2, if input surfaces are created via CreateSurfaceFromDX11Native (possibly only with textures in BGRA formats?). This issue appears to be specific to RX 7000 series GPUs.

To Reproduce
Steps to reproduce the behavior:
(Sorry for the rubbish steps, I'll try to make a minimal example when we have a confirmed in-house repro to test against and we're not still firefighting this issue with users.)

1. Create DX11 textures in DXGI_FORMAT_B8G8R8A8_UNORM format
2. Create AMF surfaces from them via CreateSurfaceFromDX11Native
3. Pass those surfaces as inputs to the AMF encoder
4. Observe that (after a maximum of ~10s) the video output will be broken with odd gridlines (possibly every 16x16...?)
5. (Optional) Rollback to 25.10.1 or earlier and observe the same thing doesn't happen

Setup:

• OS: Windows 10/11
• Driver Version: 25.10.2
• GPU: Seemingly any RX7000 series GPU
  • Observed as broken so far: 7800 XT, 7900 XTX, 7900 GRE, 7700S
  • Working so far: 5500 XT, 5700 XT, 9070 XT
• Component: Encoder

Debug Log (please upload or paste):
(No errors observed from AMF, and no warnings which weren't present in the previous driver release.)

Expected behavior
Normal H.264 video output

Screenshots

Additional context
We are currently building against the AMF 1.4.30 headers in case that's relevant. Issue persists when built against AMF 1.5.0 headers.
I haven't yet confirmed whether any other codecs are affected; all the logs I've seen so far are for H.264. Confirmation for HEVC and AV1 also.

[lazye53]

Although I don't like the image quality, I don't have that problem. RX 5700XT

https://www.twitch.tv/videos/2605117278

1920x1080 60fps bitrate = 5000 kbps, keyframe interval 2s, preset = quality, PA checked, max B frames = 2

MaxNumRefFrames=4 MaxConsecutiveBPictures=1 BPicturesPattern=2 AdaptiveMiniGOP=true LowLatencyInternal=true RateControlPreanalysisEnable=1 EnableVBAQ=false HighMotionQualityBoostEnable=true DeBlockingFilter=true BReferenceEnable=true HalfPixel=true QuarterPixel=true PAFrameSadEnable=true PALookAheadBufferDepth=31 PAPerceptualAQMode=1 PACAQStrength=2 PATemporalAQMode=2 PASceneChangeDetectionEnable=true

[Esvandiary]

Although I don't like the image quality, I don't have that problem. RX 5700XT

Yeah this seems to be specific to RX 7000 (which we're trying to get a test machine for currently). I've got a 9070XT and it's fine for me, and we have another team member with a 5500XT who's the same. Every user who's reported this so far is on some sort of RX 7xxx GPU.

I'm just making a test environment which builds against the AMF 1.5.0 headers to see if that makes any difference, and once we can repro the issue internally I'll try that, and then make a minimal repro example.

e: From an initial in-house test it looks like the issue sometimes only kicks in after ~10 seconds of recording; I've updated the main post to match.

[Esvandiary]

Update: this only happens if we provide input surfaces via CreateSurfaceFromDX11Native, if we provide surfaces via AllocSurface + a copy the problem doesn't occur. We currently provide textures as DXGI_FORMAT_B8G8R8A8_UNORM. Going to try and make a minimal repro from samples when I get time.