[Bug] Vulkan Flash Attention not working

[daniandtheweb]

Git commit

a564fdf

Operating System & Version

Artix Linux - Linux 7.0.0-1-cachyos

GGML backends

Vulkan

Command-line arguments used

./sd-cli -M img_gen -p "a cat" --sampling-method euler --steps 20 -W 1024 -H 1024 -b 1 --cfg-scale 5 -s -1 --clip-skip -1 --embd-dir /home/daniandtheweb/Workspace/sd.cpp-webui/models/embeddings/ --lora-model-dir /home/daniandtheweb/Workspace/sd.cpp-webui/models/loras/ -t 0 --rng cuda --sampler-rng cuda --lora-apply-mode auto -o /home/daniandtheweb/Workspace/sd.cpp-webui/outputs/txt2img/1765151582.png --diffusion-model /home/daniandtheweb/Workspace/sd.cpp-webui/models/unet/anima-preview3-base.safetensors --vae /home/daniandtheweb/Workspace/sd.cpp-webui/models/vae/qwen_image_vae.safetensors --llm /home/daniandtheweb/Workspace/sd.cpp-webui/models/text_encoders/qwen_3_06b_base.safetensors --scheduler kl_optimal --vae-tile-overlap 0.5 --vae-tile-size 32x32 --cache-mode spectrum --preview proj --preview-path /home/daniandtheweb/Workspace/sd.cpp-webui/outputs/txt2img/1765151582_preview.png --preview-interval 1 --offload-to-cpu --vae-tiling --fa --vae-conv-direct --mmap --color

Steps to reproduce

Running Ernie or Anima models on Vulkan with Flash Attention enabled causes the generation to fail.
When Flash Attention is disabled the generation works correctly.

What you expected to happen

What actually happened

Logs / error messages / stack trace

No response

Additional context / environment details

This issue started with 8f2967c.

[daniandtheweb]

This is probably related to the issue mentioned here: #509.

[Naster17]

Same