peyer

How about providing c++/python version of wrapper of amx?

2

Thanks for your great work, but I am newer to rust, do you have idea to provide c++/python version of wrapper of amx?

**I'm submitting a ...** (check one with "x“) 我使用使用vs2013+win7编译EasyPR.sln，想避免手动配置头文件和依赖库，请问使用nuget怎么自动下载opencv，并安装好对应的库文件和头文件 ``` [ ] bug report [x ] help wanted [ ] feature request ``` **Current behavior** **Expected/desired behavior** **Reproduction of the...

我看目前mace提供的APU库文件，只有mt67xx的只有6779(P90)和6785(G90)的库文件，想问下6771(P60)支持APU吗？支持的话也是需要使用apu-frontend和apu-platform的方式吗？

引入adreno_opencl_ml_sdk的问题

3

从高通官方博客[blog](https://developer.qualcomm.com/blog/accelerate-your-models-our-opencl-ml-sdk)上得知，从adreno 660开始，高通官方增加了cl_qcom_ml_ops扩展，速度比目前开源实现的opencl 代码都要快一些；我从小米11的手机上adb pull /system/vendor/lib64/libOpenCL.so，本地nm -D看过，确实有clQueryMLInterfaceVersionsQCOM的符号表，但是调用以后会报CL_OUT_OF_HOST_MEMORY的错误，请问是小米还没有更新adreno的驱动导致的吗？

How to use sass_func for computing convolution?

1

Thanks for your great work! I just wanna to compare the latency of running convolution layer between anakin_sass and cudnn, and I have successfully ran convolution layer with cudnn. However,...

How to use api of direct_conv_xxx in sass_func.hpp

2

For convenience, I just take an instance below. If the shape of input tensor of conv op is [1, 32, 110, 94] ([N, C, H, W] order), the shape of...

Dose libanakin_saber_sass_sm_61.a support fp16?

1

@Superjomn Hi, I just want to compare the performance of anakin_sass with the performance of [email protected] for convolution operator, and I have successfully compared on fp32 data-format. However, when I...

opencl的一个bug

8

我在测试bolt的opencl时发现一个bug； 由于bolt采用了NCHW / NCHWC4等数据排布混用、针对opencl在层与层之间的blob混用了buffer、image1d、image2d、image3d，同时内存分配上还采用了内存复用，可能导致了我的一个模型在depth2space_ocl层触发了一个bug，就是depth2space对应的kernel的arg里是写的是buffer的输入类型，但是内存复用以后，对应arg传入了一个image3d的数据类型，导致set_arg报错CL_INVALID_MEM_OBJ，我还在定位是内存复用的代码

@BUG1989 Thanks for your great work ! I have test with mtcnn which the performance just drop a litte, but speed up about 4x, amazing! However, when I test with...

mxnet model

2

@clancylian Thanks for your great work! Could your share your retrained mxnet model?