Ling

@AlexeyAB Is this project no longer maintained?

2

@AlexeyAB You seem to have not been active for a long time, do you not maintain this project anymore？

我有3个GPU，但get_gpu_count()=1

8

windows11，用的最新版ncnn. 但是用darknet训练或者识别都可以正常用到两个p40 

好兄弟，啥时候有yolov8n/s/m.cfg

13

darknet用起来顺手

下次发布可以提供带有mt版本的静态库吗？

2

现在是md的，有些用户的系统环境会导致加载ncnn main时直接在vs运行库dll内崩溃

mi_heap_new() Is multithreading unsafe, as expected?

16

windows - x86 heap = mi_heap_new() Error prone code in multithreading: address = mi_heap_malloc(heap, 666) .................................. mi_free(address) Code for safe operation without errors: EnterCriticalSection() address = mi_heap_malloc(heap, 666) LeaveCriticalSection() .....................................

老哥，vcpkg里面搜不到coost

1

.\vcpkg search coost

ncnnoptimize.exe 经常崩溃

1

错误应用程序名称: ncnnoptimize.exe，版本: 0.0.0.0，时间戳: 0x66c44b14 错误模块名称: VCRUNTIME140.dll，版本: 14.40.33810.0，时间戳: 0x92923569 异常代码: 0xc0000005 错误偏移量: 0x0000e62e 控制台输出：shape_inference 之后崩溃，有时候又正常 我的调用过程： darknet2ncnn.exe "yolo.cfg" "yolo.weights" "temp.param" "temp.bin" 1 ncnnoptimize.exe "temp.param" "temp.bin" "opt.param" "opt.bin" 65536 经常崩，还有就是NCNN本身也会在某些系统上崩溃，也是这个VCRUNTIME140.dll报错，运行库都安装了最新的

Error message: signal handler invoked for signal #11

4

[yolov4-tiny_3l_ncnn.cfg.zip](https://github.com/user-attachments/files/19253745/yolov4-tiny_3l_ncnn.cfg.zip) Darknet V4 "Slate" v4.0-33-ga48c04d0 CUDA runtime version 12050 (v12.5), driver version 12080 (v12.8) cuDNN version 12050 (v9.2.0), use of half-size floats is ENABLED => 0: NVIDIA GeForce RTX 3060...

hp-socket 使用vs2022 + VC-LTL + mimalloc 编译后 在DLLmain会引发内存堆损坏的问题

3

https://github.com/Chuyu-Team/VC-LTL5/issues/71

如题，是否可以考虑增加共享栈的方式