java自动卸载原因_英特尔MKL / Xeon Phi卸载运行时问题 - 自动卸载无法正常工作

我已经在Windows 10 Pro中设置了我的Xeon phi 3120A,其中包括MPSS 3.8.4和Parallel XE 2017(初始版本) . 我选择了这个Parallel XE,因为这是x100系列最后支持的XE . 我已经安装了随Parallel XE 2017(初始版本)打包的MKL版本 .

What have I done / setup:

在设置MPSS 3.8.4之后,按照闪烁和ping等步骤,我检查了micctrl -s显示“mic0 ready”(linux图像包含相应的KNC名称),miccheck产生所有“通行证”,micinfo给出我阅读了协处理器提供的所有关键统计数据 .

因此,对我而言,看起来协处理器肯定已安装并被我的计算机识别 . 我还可以看到mic0在micsmc gui中启动并运行 .

然后我设置环境变量以启用自动卸载,即MKL_MIC_ENABLE = 1,OFFLOAD_DEVICES = 0,MKL_MIC_MAX_MEMORY = 2GB,MIC_ENV_PREFIX = MIC,MIC_OMP_NUM_THREADS = 228,MIC_KMP_AFFINITY = balancer .

The Problem - Auto Offload is not working

当我在R-3.4.3中运行一些简单的代码(下面复制,专门为自动卸载而设计)时,它会通过我的主机运行代码而不是通过Xeon phi运行任何代码 .

为了支持这一点,当我看到micsmc gui时,我在Xeon Phis上看不到任何活动 .

因此,自动卸载不起作用 .

The R code:

require(Matrix)

sink("output.txt")

N

cat("Initialization...\n")

a

b

cat("Matrix-matrix multiplication of size ", N, "x", N, ":\n")

for (i in 1:5) {

dt=system.time( c

gflops = 2*N*N*N*1e-9/dt[3]

cat("Trial: ", i, ", time: ", dt[3], " sec, performance: ", gflops, " GFLOP/s\n")

}

Other steps I have tried:

然后我继续设置MKL_MIC_DISABLE_HOST_FALLBACK = 1环境变量,正如预期的那样,当我运行上面的代码时,R终止了 .

在Using Intel® MKL Automatic Offload on Intel Xeon Phi Coprocessors中它表示如果HOST_FALLBACK标志处于活动状态且尝试卸载但失败(由于“卸载运行时无法找到协处理器或无法正确初始化它”),它将终止程序 - 这发生在R完全终止 . 为了完整起见,此问题发生在R-3.5.1,Microsoft R Open 3.5.0和R-3.2.1上 .

So my questions are:

在Xeon phi上运行R代码我错过了什么?你能告诉我我需要做些什么来完成这项工作吗?

(链接到1)有没有办法检查MKL卸载运行时是否可以看到Xeon phi?或者它是否正确设置,或者MKL初始化Xeon phi的问题是什么(如果有的话)?

如果那里有人可以帮助我,我会真诚地感激 - 我相信我错过了一个基本/简单的步骤,并且一直在试图让我的头发工作 .

提前谢谢了,

皮疹

  • 0
    点赞
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
The error message you encountered indicates that there is an issue loading the Intel Math Kernel Library (MKL) dynamic library file `libmkl_rt.so.1` in your Python environment. This error can occur due to various reasons, such as incorrect installation or configuration issues. Here are a few steps you can try to resolve the issue: 1. Check MKL Installation: Ensure that Intel MKL is properly installed on your system. You can visit the official Intel website and follow the instructions to download and install the MKL library for your operating system. 2. Verify Library Path: Confirm that the library path specified in the error message (`/opt/anaconda3/lib/python3.10/site-packages/mkl/../../../libmkl_rt.so.1`) is correct. It should point to the location of the `libmkl_rt.so.1` file on your system. Make sure the file exists at that location. 3. Update LD_LIBRARY_PATH: Set the `LD_LIBRARY_PATH` environment variable to include the directory where the `libmkl_rt.so.1` file is located. You can do this by running the following command before executing your Python script: ``` export LD_LIBRARY_PATH=/path/to/mkl:$LD_LIBRARY_PATH ``` Replace `/path/to/mkl` with the actual directory path containing the `libmkl_rt.so.1` file. 4. Conda Environment Activation: If you are using Anaconda or Miniconda, try activating your conda environment before running the Python script. This ensures that the correct MKL libraries associated with your environment are loaded. 5. Reinstall Libraries: If the above steps do not resolve the issue, you can try reinstalling the relevant libraries, including NumPy, MKL, and any other packages that depend on them. You can use the following commands: ``` conda uninstall numpy mkl conda install numpy mkl ``` This will uninstall and then reinstall the NumPy and MKL libraries using conda. If the issue still persists after trying these steps, it would be helpful to provide more information about your system configuration, the specific steps you followed during the installation, and any additional error messages or traceback information you received.
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值