llama.cpp b11933：Vulkan 修复 MUL_MAT dim 3 广播

AI 摘要

llama.cpp 发布 b11933 版本，主要修复 Vulkan 后端在 MUL_MAT 算子中对 dim 3 广播的支持：此前 supports_op 会拒绝 src0 与 src1 在 ne3 维度不一致的 MUL_MAT，导致这类矩阵乘回退到 CPU 并复制权重，且切分布局随 ubatch 流数变化。该改动仅对 MUL_MAT_ID 保留严格检查，并将广播场景从要求 ne3 相等的 mul_mat_vec_nc 路径分流，同时新增针对 dim 3 广播的测试用例。

正文节选

<details open> vulkan: fix CI llvmpipe by allowing dim 3 broadcast in MUL_MAT (#31239) * vulkan: allow dim 3 broadcast in MUL_MAT supports_op rejected every MUL_MAT whose src0 and src1 differ in ne3, although mul_mm, mul_mmq and mul_mat_vec already broadcast src0 over dim 3 through broadcast3. Such a matmul then ran on the CPU with a copy of the weight, and the split layout changed with the number of streams in the ubatch. Keep the strict check for MUL_MAT_ID only, and route the broadcast ca

发布时间：2026-10-27 23:27

抓取时间：2026-10-27 23:48

来源机构：ggml-org