[Bug] SYCL: Provided range and/or offset does not fit in int

[MrDrMcCoy]

Git commit

Using Docker image: ghcr.io/leejet/stable-diffusion.cpp:master-sycl@sha256:47a7b35f4219b7470f48951808b08dfb557c94a224b25af67f4debb72968f9b2

Operating System & Version

Debian 13 (with backports and Proxmox kernel 7.0.2-6)

GGML backends

SYCL

Command-line arguments used

/sd-cli --backend sycl0 --diffusion-model /data/ai/flux.2k/base/flux-2-klein-9b.safetensors --llm /data/ai/encoders/Qwen3-8B-Q8_0.gguf --vae /data/ai/flux.2k/vae/diffusion_pytorch_model.safetensors --cfg-scale 1.01 --lora-model-dir /data/ai/flux.2k/lora-active/real --prompt "photograph, natural lighting, a woman" --height 1248 --width 832 --output /tmp/0.png --sampling-method dpm2 --scheduler lcm --seed 1337 --steps 4 --verbose

Steps to reproduce

1. podman run -it --rm --privileged --device /dev/dri --device /dev/kfd -v /data/ai:/data/ai -v /tmp:/tmp --entrypoint bash ghcr.io/leejet/stable-diffusion.cpp:master-sycl@sha256:439ab4c578c1955830682b389f627d975fa5d8297b1c16ed5c9b2ee51ef43c5f
2. /sd-cli --backend sycl0 --diffusion-model /data/ai/flux.2k/base/flux-2-klein-9b.safetensors --llm /data/ai/encoders/Qwen3-8B-Q8_0.gguf --vae /data/ai/flux.2k/vae/diffusion_pytorch_model.safetensors --cfg-scale 1.01 --lora-model-dir /data/ai/flux.2k/lora-active/real --prompt "photograph, natural lighting, a woman" --height 1248 --width 832 --output /tmp/0.png --sampling-method dpm2 --scheduler lcm --seed 1337 --steps 4 --verbose

What you expected to happen

An image to be output.

What actually happened

Application exits with an error during VAE phase.

Logs / error messages / stack trace

[INFO ] stable-diffusion.cpp:4265 - sampling completed, taking 37.82s
[INFO ] stable-diffusion.cpp:4285 - generating 1 latent images completed, taking 37.82s
[INFO ] stable-diffusion.cpp:3975 - decoding 1 latents
[DEBUG] ggml_extend.hpp:1924 - vae compute buffer size: 6592.98 MB(VRAM)
Provided range and/or offset does not fit in int. Pass `-fno-sycl-id-queries-fit-in-int' to remove this limit.Exception caught at file:/sd.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp, line:4593
Error OP IM2COL

Additional context / environment details

Hardware info:
CPU: AMD EPYC 7532
GPUs: 4x Intel B70, 4x AMD Radeon Pro AI 9700 XT
RAM: 256GB DDR4
Storage: ZFS pool. 4x HDD mirror for data, 2x Optane mirror for special, 4x SSD stripe for cache.

As much as I would prefer to use Vulkan on the Intel cards, doing so causes them to vanish after several runs until I can cold boot the system. Hoping to get SYCL working for them, but not been having much luck with the official Docker container. My own build with the latest Intel compute framework just segfaults when I try to run it.