<div dir="ltr"><div dir="ltr"><div dir="ltr"><div dir="ltr"><div class="gmail_default" style="font-family:garamond,serif">Another question regarding compiling cp2k with CUDA (I don't know if I should make another post, let me know if I should do).</div><div class="gmail_default" style="font-family:garamond,serif">I compiled cp2k with cuda for V100s. The dbcsr I compiled by myself using the develop branch of Shoshijak, (recursive clone it as a submodule of cp2k like it should be).</div><div class="gmail_default" style="font-family:garamond,serif">Everything was going well with DBCSR and I got parameters for V100.</div><div class="gmail_default" style="font-family:garamond,serif"><br></div><div class="gmail_default" style="font-family:garamond,serif">For the cp2k local_cuda.psmp, I took from the one automatically generated by install_cp2k_toolchain.sh with --gpu-ver=P100.</div><div class="gmail_default" style="font-family:garamond,serif">Then I modify it (attached).</div><div class="gmail_default" style="font-family:garamond,serif"><br></div><div class="gmail_default" style="font-family:garamond,serif">When compiled local_cuda.psmp, it showed this error:</div><div class="gmail_default" style="font-family:garamond,serif">-------------------------------------------------------------</div><div class="gmail_default" style="font-family:garamond,serif"><div class="gmail_default">[phuong@scc5 cp2k]$ make -j 20 ARCH=local_cuda VERSION="psmp"</div><div class="gmail_default">Discovering programs ...</div><div class="gmail_default">make -C /hpchome/phuong/cp2k/exts/dbcsr \</div><div class="gmail_default"> INCLUDEMAKE=/hpchome/phuong/cp2k/arch/local_cuda.psmp \</div><div class="gmail_default"> LIBDIR=/hpchome/phuong/cp2k/lib/local_cuda/psmp/exts/dbcsr \</div><div class="gmail_default"> OBJDIR=/hpchome/phuong/cp2k/obj/local_cuda/psmp/exts/dbcsr \</div><div class="gmail_default"> FYPPEXE=/hpchome/phuong/cp2k/tools/build_utils/fypp</div><div class="gmail_default">Removing stale archives for psmp ... </div><div class="gmail_default">Resolving dependencies for psmp ... </div><div class="gmail_default">Removing stale archives ... </div><div class="gmail_default">Resolving dependencies ... </div><div class="gmail_default">makedep error: Multiple source files with the same basename: dbcsr_list_routinereport</div><div class="gmail_default">make[3]: *** [makedep] Error 1</div><div class="gmail_default">make[2]: *** [dbcsr] Error 2</div><div class="gmail_default">make[2]: *** Waiting for unfinished jobs....</div><div class="gmail_default">make[1]: *** [psmp] Error 2</div><div class="gmail_default">make: *** [all] Error 2</div><div class="gmail_default">------------------------------------------------------------------</div><div class="gmail_default">Can you take a look as well?</div><div class="gmail_default"><br></div><div class="gmail_default">P/s: I forgave to use BLIS + LibFLAME, basically could compile and test with sopt and ssmp version, but I do need the psmp or at least popt.</div><div class="gmail_default">The problem with these libs are they could not work with Scalapack, can compile scalapack links with them without errors but the tests were failed ... Thus no psmp, popt. </div><div><br></div></div></div></div></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Sat, 25 May 2019 at 22:05, Bich Phuong <<a href="mailto:bichphu...@gmail.com">bichphu...@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_default" style="font-family:garamond,serif">Hi,</div><div class="gmail_default" style="font-family:garamond,serif"><br></div><div class="gmail_default" style="font-family:garamond,serif">I'm currently running some benchmarks, tests with different libraries and compilers on 1 node.</div><div class="gmail_default" style="font-family:garamond,serif">I don't know which input file/calculation they will give us during the competition.</div><div class="gmail_default" style="font-family:garamond,serif">also think about tunning with RDMA for job across nodes and try with KNEM to see if the performance changes.</div><div class="gmail_default" style="font-family:garamond,serif"><br></div><div class="gmail_default" style="font-family:garamond,serif">This is output of benchmarks on 1 node (2U).</div><div class="gmail_default" style="font-family:garamond,serif">Thanks for your help.</div><div class="gmail_default" style="font-family:garamond,serif"></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Fri, 24 May 2019 at 10:39, Alfio Lazzaro <<a href="mailto:alfio....@gmail.com" target="_blank">alfio....@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div style="font-family:garamond,serif"><br></div><div style="font-family:garamond,serif">I also appreciate your suggestion for libxsmm and libsmm. Actually we have cp2k as a application task in a cluster competition.</div><div style="font-family:garamond,serif">Do you know any way to boost performance a bit without changing the given input file? </div><div style="font-family:garamond,serif"><br></div></div></blockquote><div><br></div><div>Well, libxsmm can still give you good performance on AMD too, so I suggest to use it.</div><div>Concerning other possible optimizations, it depends on what you are running. Could you share your output?</div><div><br></div><div>Alfio</div><div><br></div><div><br></div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div style="font-family:garamond,serif"></div><div style="font-family:garamond,serif"><br></div><div style="font-family:garamond,serif"></div></div><br><div class="gmail_quote"><div dir="ltr">On Thu, 23 May 2019 at 11:08, Alfio Lazzaro <<a rel="nofollow">al...@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div>Hello Bich Phuong,</div><div>larnv is a LAPACK function. However, I don't see any LAPACK library in your arch file:<br></div><div><br></div><div>
<p><span>LIBS<span> </span>= -lblis -lxsmmf -lxsmm -ldl -lpthread -lsmm_dnn -lxcf03 -lxc -lderiv -lint -lfftw3 -lstdc++</span></p></div><div><br></div><div><br></div><div>BTW, I see that you are lusing libxsmm and libsmm, there are mutual exclusive. You can just use libxsmm.</div><div><br></div><div>Alfio</div><div><br></div><br>Il giorno mercoledì 22 maggio 2019 20:58:20 UTC+2, Bich Phuong ha scritto:<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div>Hello,</div><div><br></div><div>I like to compile CP2K with AMD libs to compare the performances with the one compiled with OpenBLAS, LAPACK and Scalapack</div><div>since I had the system with 2 x EPYC 7601.</div><div><br></div><div>I checked the install_cp2k_toolchain.sh and install_acml.sh but the acml doesn't exist anymore (<a href="https://developer.amd.com/amd-cpu-libraries/" rel="nofollow" target="_blank">https://developer.amd.com/amd-cpu-libraries/</a>), thus can not use this tool.</div><div>I modify the local.sopt (tried with sopt firstly) arch file created by install_cp2k_toolchain.sh by replacing path of openblas and -lopenblas, tried to link either BLIS/BLIS+libflame/LibM but nothing works (the available libraries or compiled them by myself follow this documentation <a href="https://developer.amd.com/wp-content/resources/AMDCPULibrariesUserGuide_1.3.pdf" rel="nofollow" target="_blank">https://developer.amd.com/wp-content/resources/AMDCPULibrariesUserGuide_1.3.pdf</a>).</div><div>It complaint about undefined references to some BLAS functions.</div><div>-----------------------</div><div><div>/home/phuong/cp2k-6.1/lib/amd/sopt/libdbcsrops.a(dbcsr_operations.o): In function `__dbcsr_operations_MOD_dbcsr_init_random':</div><div>/home/phuong/cp2k-6.1/src/dbcsr/ops/dbcsr_operations.F:1104: undefined reference to `dlarnv_'</div><div>/home/phuong/cp2k-6.1/src/dbcsr/ops/dbcsr_operations.F:1104: undefined reference to `dlarnv_'</div><div>/home/phuong/cp2k-6.1/lib/amd/sopt/libdbcsrops.a(dbcsr_blas_operations.o): In function `__dbcsr_blas_operations_MOD_dbcsr_lapack_larnv':</div><div>/home/phuong/cp2k-6.1/src/dbcsr/ops/dbcsr_blas_operations.F:84: undefined reference to `zlarnv_'</div><div>/home/phuong/cp2k-6.1/src/dbcsr/ops/dbcsr_blas_operations.F:82: undefined reference to `clarnv_'</div><div>/home/phuong/cp2k-6.1/src/dbcsr/ops/dbcsr_blas_operations.F:78: undefined reference to `slarnv_'</div><div>/home/phuong/cp2k-6.1/src/dbcsr/ops/dbcsr_blas_operations.F:80: undefined reference to `dlarnv_'</div><div>collect2: error: ld returned 1 exit status</div><div>make[3]: *** [/home/phuong/cp2k-6.1/exe/amd/dbcsr_example_3.sopt] Error 1</div><div>make[3]: *** Waiting for unfinished jobs....</div></div><div>--------------------------</div><div><br></div><div>Does anyone have experiences with this work?</div><div>Which AMD library should I use to alter the ACML one?</div><div><br></div><div>Thank you.</div></div></blockquote></div>
<p></p>
-- <br>
You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.<br>
To unsubscribe from this topic, visit <a href="https://groups.google.com/d/topic/cp2k/4vFftwwDEJg/unsubscribe" rel="nofollow" target="_blank">https://groups.google.com/d/topic/cp2k/4vFftwwDEJg/unsubscribe</a>.<br>
To unsubscribe from this group and all its topics, send an email to <a rel="nofollow">c...@googlegroups.com</a>.<br>
To post to this group, send email to <a rel="nofollow">c...@googlegroups.com</a>.<br>
Visit this group at <a href="https://groups.google.com/group/cp2k" rel="nofollow" target="_blank">https://groups.google.com/group/cp2k</a>.<br>
To view this discussion on the web visit <a href="https://groups.google.com/d/msgid/cp2k/add016e6-b3a7-48f5-8dd9-d38eb9b50330%40googlegroups.com?utm_medium=email&utm_source=footer" rel="nofollow" target="_blank">https://groups.google.com/d/msgid/cp2k/add016e6-b3a7-48f5-8dd9-d38eb9b50330%40googlegroups.com</a>.<br>
For more options, visit <a href="https://groups.google.com/d/optout" rel="nofollow" target="_blank">https://groups.google.com/d/optout</a>.<br>
</blockquote></div>
</blockquote></div>
<p></p>
-- <br>
You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.<br>
To unsubscribe from this topic, visit <a href="https://groups.google.com/d/topic/cp2k/4vFftwwDEJg/unsubscribe" target="_blank">https://groups.google.com/d/topic/cp2k/4vFftwwDEJg/unsubscribe</a>.<br>
To unsubscribe from this group and all its topics, send an email to <a href="mailto:cp...@googlegroups.com" target="_blank">cp...@googlegroups.com</a>.<br>
To post to this group, send email to <a href="mailto:cp...@googlegroups.com" target="_blank">cp...@googlegroups.com</a>.<br>
Visit this group at <a href="https://groups.google.com/group/cp2k" target="_blank">https://groups.google.com/group/cp2k</a>.<br>
To view this discussion on the web visit <a href="https://groups.google.com/d/msgid/cp2k/da5f6707-0a61-481a-9e8c-b697c883f1c0%40googlegroups.com?utm_medium=email&utm_source=footer" target="_blank">https://groups.google.com/d/msgid/cp2k/da5f6707-0a61-481a-9e8c-b697c883f1c0%40googlegroups.com</a>.<br>
For more options, visit <a href="https://groups.google.com/d/optout" target="_blank">https://groups.google.com/d/optout</a>.<br>
</blockquote></div>
</blockquote></div>