[gmx-developers] A HIP version of GROMACS 2020.2 is available

[张驭洲]

Hi Vedran,

Sorry for my late reply. With "the same compilation issues like last time", do you mean the error you mentioned in the email on 6 Jan 2020:

In file included from 
 /home/vedranm/workspace/hipified-gromacs/src/gromacs/mdlib/nbnxn_cuda/nbnxn_cuda.cu:118:
 In file included from 
 /home/vedranm/workspace/hipified-gromacs/src/gromacs/mdlib/nbnxn_cuda/nbnxn_cuda_kernel_pruneonly.cuh:52:
 /home/vedranm/workspace/hipified-gromacs/src/gromacs/mdlib/nbnxn_cuda/nbnxn_cuda_kernel_utils.cuh:440:12: 
 error: no matching function for call to 'lerp'
      return lerp(d01.x, d01.y, fraction);
             ^~~~
 /home/vedranm/workspace/hipified-gromacs/src/gromacs/mdlib/nbnxn_cuda/nbnxn_cuda_kernel_utils.cuh:423:3: 
 note: candidate template ignored: deduced conflicting types for 
 parameter 'T' ('Scalar_accessor<[2 * ...], 0>' vs. 'Scalar_accessor<[2 * 
 ...], 1>')
 T lerp(T d0, T d1, T t)

In my case this can be solved by adding type cast to the line that causes the error, like this:

return lerp((float)d01.x, (float)d01.y, fraction);

I have made this adding in the 2020.2 version that I pushed to github on 4 May, so I think this function should not cause error again. If you encountered any other error, please paste it directly.

Sincerely, 
Zhang



> -----原始邮件-----
&gt; 发件人: "Vedran Miletić" <vedran at miletic.net>
&gt; 发送时间: 2020-05-26 21:33:38 (星期二)
&gt; 收件人: gmx-developers at gromacs.org
&gt; 抄送: 
&gt; 主题: Re: [gmx-developers] A HIP version of GROMACS 2020.2 is available
&gt; 
&gt; Hi again Zheng,
&gt; 
&gt; I actually tried this when I mentioned and I encountered the same 
&gt; compilation issues like last time. Did you do any special configuration 
&gt; on your machine after installing ROCm?
&gt; 
&gt; Cheers,
&gt; Vedran
&gt; 
&gt; 04. 05. 2020. u 09:52, Vedran Miletić je napisao/la:
&gt; &gt; Hi Zhang,
&gt; &gt; 
&gt; &gt; I have been wondering what's going on with this effort just last week. 
&gt; &gt; Will gladly give it a try tomorrow.
&gt; &gt; 
&gt; &gt; Cheers,
&gt; &gt; Vedran
&gt; &gt; 
&gt; &gt; 04. 05. 2020. u 09:38, 张驭洲 je napisao/la:
&gt; &gt;&gt; Dear GROMACS developers,
&gt; &gt;&gt;
&gt; &gt;&gt; I have ported GROMACS 2020.2 to HIP, which can be compiled and run on 
&gt; &gt;&gt; ROCM platform. I tested the code on AMD MI 50 GPU and ROCM 3.3.0. It 
&gt; &gt;&gt; has all the features that the CUDA code has, including computing 
&gt; &gt;&gt; bonded forces, update and constraint on GPU. In my case, the ported 
&gt; &gt;&gt; code is about 20% faster than the official OpenCL code on the same 
&gt; &gt;&gt; platform, but still slower than the CUDA code on nvidia V100 GPU.
&gt; &gt;&gt;
&gt; &gt;&gt; I'm trying optimising the ported code further. If you have interest, 
&gt; &gt;&gt; please clone the code from 
&gt; &gt;&gt; https://github.com/ipe-zhangyz/hipified-gromacs.git 
&gt; &gt;&gt; <https: hipified-gromacs.git="" ipe-zhangyz="" github.com="">and have a try. 
&gt; &gt;&gt; Any feedback is welcome!
&gt; &gt;&gt;
&gt; &gt;&gt;
&gt; &gt;&gt; Sincerely
&gt; &gt;&gt; Zhang
&gt; &gt;&gt;
&gt; &gt; 
&gt; 
&gt; -- 
&gt; Vedran Miletić
&gt; vedran.miletic.net
&gt; -- 
&gt; Gromacs Developers mailing list
&gt; 
&gt; * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before posting!
&gt; 
&gt; * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
&gt; 
&gt; * For (un)subscribe requests visit
&gt; https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers or send a mail to gmx-developers-request at gromacs.org.
</https:></vedran at miletic.net>