返回全部动态
llama.cpp b12179：Hexagon 后端支持 conv_3d 并优化 conv_2d
原标题：b12179
AI 摘要
llama.cpp 发布 b12179 版本，主要更新 Hexagon 后端：将 CONV_2D 实现泛化以支持 CONV_3D，复用现有 HMX 分块（隐式 GEMM）卷积路径，并修复除零风险、在主机端预计算展开除数、支持任意正卷积步长、改进 tile 尺寸选择，同时解决 DMA 队列溢出问题。该更新提升了骁龙 Hexagon NPU 上卷积算子的覆盖范围与稳定性。
以上摘要由 AI 生成，可能存在误差。事实请以原文为准。
正文节选
<details open> hexagon: conv_2d optimizations and support for conv_3d (#31814) * hexagon: Generalization of CONV_2D implementation for CONV_3D - New op HTP_OP_CONV_3D is wired through htp-ops.h, main.c (shares op_conv_2d) and the op remap in htp-kparams.cpp. - conv_3d runs on the existing HMX tiled (implicit GEMM) conv2d path. KD and KH are folded into one virtual kernel row (KD*KH), and output rows are counted as N*OD*OH. - htp_kparams_solve_conv3d checks the conv_3d layout and calls the til
发布时间：2026-11-07 04:43
抓取时间：2026-11-07 05:35
来源机构：ggml-org