| 查看: 1133 | 回复: 2 | |||
| 【奖励】 本帖被评价2次，作者wuli8增加金币 2 个 | |||
| 当前主题已经存档。 | |||
[资源]
【转帖】Compilation of VASP on Opteron/Rocks cluster Ametisti with Pathscale
|
|||
|
Compilation of VASP on Opteron/Rocks cluster Ametisti with Pathscale 2.2.1 Fortran compiler These observations apply to the serial version of VASP. The parallel section dealing with VASP on Ametisti using the MPI interface will be included later. I used the makefile from Janne Blomqvist at Helsinki University of Technology (Thanks Janne). Few things were noticed after a successful compilation. Serial version problems: Problem 1: Error EDDDAV: Call to ZHEGV failed. Problem 2: lib-4201 : UNRECOVERABLE library error Problem 3: WARNING in EDDRMM: call to ZHEGV failed, returncode = 6 3 9 Parallel version problems: Will be added at some later date Problem 1) The test calculation for copper system (benchmark.tar.gz, download from the VASP ftp-server) stops with an error message: "Error EDDDAV: Call to ZHEGV failed. Returncode = 9 1 8". The actual numbers change at the end of the error, but the message means that a LAPACK library call failed. Solution: The subroutine davidson.F must be handled with lower optimization setting Add the following lines to the end of the VASP Makefile: davidson.o : davidson.F $(CPP) $(FC)$(FFLAGS) -O1 -c $*$(SUFFIX) (You remembered to use the TAB key instead of spaces with the second and the third line, right?) Problem 2) After lowering the davidson subroutine optimization level the calculation ends with another error: "lib-4201 : UNRECOVERABLE library error: Unable to find error message (check NLSPATH, file lib.cat) Encountered during a direct access unformatted READ from unit 21. Fortran unit 21 is connected to a direct unformatted unblocked file: "TMPCAR" /opt/gridengine/default/spool//compute-0-4/job_scripts/10170: line 31: 16997 Aborted (core dumped) ./vasp_path_serial >vasp_path_serial.out" Solution: Change/add the IWAVPR=10 line to your INCAR file. This is from the VASP manual, FAQ section, page 149 "Question: I am running VASP on a SGI Origin, and the simple benchmark (benchmark.tar.gz) fails with lib-4201 : UNRECOVERABLE library error READ operation tried to read past the end-of-record. Encountered during a direct access unformatted READ from unit 21 Fortran unit 21 is connected to a direct unformatted unblocked file: "TMPCAR" IOT Trap Abort (core dumped) Answer: VASP extrapolates the wave functions between molecular dynamics time steps. To store the wave functions of the previous time steps either a temporary scratch file (TMPCAR) is used (IWAVPR=1-9) or large work arrays are allocated (IWAVPR=11-19). On the SGI, the version that uses a temporary scratch file does not compile correctly, and hence the user has to set IWAVPR to 10." Problem 3) When running the Hg benchmark (bench.Hg.tar.gz), the OUTCAR file has numerous lines saying: "WARNING in EDDRMM: call to ZHEGV failed, returncode = 6 3 9" Solution: This issue is addressed in the VASP support forum (http://cms.mpi.univie.ac.at/vasp-forum/forum_viewtopic.php?3.214) Short summary is given here. Possible reasons (this may, once again, be connected to failures calling LAPACK routines): 1) The diagonalization algorithm is not stable for your system -> Change ALGO = Normal or ALGO = Fast in your INCAR file 2) Your geometry is not reasonable. Maybe your initial structure or the algorithm handling ion relaxation is giving a bad structure -> Switch to a different ion relaxation scheme (IBRION line in your INCAR) -> Reduce the step size of the first step by reducing the POTIM value in your INCAR 3) Problem with your LAPACK installation -> Use the LAPACK routines that came with the VASP library tar-file (vasp.4.lib) 4) If the error does not go away even if you already did the thing described in part 1 (ALGO change), then there may be a problem with LAPACK. -> Disable the calls to ZHEGV subroutine. This can be done by commenting the line # define USE_ZHEEVX in subroutines davidson.F, subrot.F and wavpre_noio.F. -> Recompile VASP after you changed the subroutines NOTE TO SELF: These issues have much in common with the problems usually described with Silicon Graphics computers. How curious... |
回复此楼» 收录本帖的淘帖专辑推荐
 计算-vasp |
» 猜你喜欢
93BebMhtakh前后11位开头都是大写
已经有5人回复
-
欢迎发来filecode的Mz6后的代码验证其规律
已经有86人回复
-
感觉是下周放榜了
已经有11人回复
-
filecode=后面第一个是大写字母
已经有10人回复
-
时间戳变了，能看出什么问题？
已经有4人回复
-
哪位老哥知道今年的国自然具体哪一天放榜？
已经有15人回复
-
今天系统多次维护，明天很可能放榜！
已经有9人回复
-
时间戳又变了8-15
已经有25人回复
-
2027广东省杰青
已经有6人回复
-
快农历七夕节了，轻松一下，男人悄悄话，女施主请不要进来。
已经有3人回复
2楼2009-05-23 17:09:17
3楼2009-08-07 15:14:10
