llama.cpp b11282：修复 MUSA 后端 __CUDA_ARCH__ 未定义问题
原标题：b11282
AI 摘要
llama.cpp 发布 b11282 版本，主要修复 MUSA 后端：此前 MUSA 厂商头文件未定义 __CUDA_ARCH__，导致共享 ggml-cuda 源码中所有架构判断为 0，部分内核（如 convert.cu 中 q8_0 到 f16 的反量化内核）被编译为空。该提交参照 HIP 后端报告最新架构并显式排除 NVIDIA 专属特性，仅在设备编译阶段定义该宏，同时移除架构比较中冗余的 defined(__CUDA_ARCH__) 检查。版本同时提供 macOS、Linux、Windows、Android 等多平台预编译包。
正文节选
<details open> musa : define __CUDA_ARCH__ for device passes (#29508) The MUSA vendor header never defined __CUDA_ARCH__, so every architecture test in the shared ggml-cuda sources evaluated to 0. Kernel bodies gated on the architecture therefore compiled to nothing, for example the q8_0 -> f16 dequantization kernel in convert.cu, whose NO_DEVICE_CODE fallback expands to an empty body in host code. Report the newest architecture like the HIP backend does and exclude the NVIDIA-only features