@swmore 你好，我也遇到了用swafort编译mpi+openacc时，提示找不到netcdf.rmod，请问应该怎么解决呢？需要用swafort重新编辑netcdf吗？

编译器和选项是：

错误：

swafort -c -o trc.o -I/home/export/online1/systest/swsduhpc/zhou_sc/nemo/nemo-v4.0.2/cfgs/GYRE_PISCES/BLD/inc -i4 -r8 -O2 -OPT:IEEE_arith=1 -g -I/usr/sw-mpp/apps/bin/WRF/netcdf/include -I/usr/sw-mpp/apps/lib/hdf5-lib/include -c /home/export/online1/systest/swsduhpc/zhou_sc/nemo/nemo-v4.0.2/cfgs/GYRE_PISCES/BLD/ppsrc/nemo/trc.f90
File rmodFileName = netcdf.rmod NOT FOUND (expected to be present)
error: No declaration found for the module: netcdf
Error : cannot find the module (module should have been seen in a previously generated file) : netcdf
swacc_real: ../../../../edg4x-rose/src/frontend/OpenFortranParser_SAGE_Connection/FortranParserActionROSE.C:17796: void c_action_use_stmt(Token_t*, Token_t*, Token_t*, Token_t*, Token_t*, ofp_bool, ofp_bool, ofp_bool): Assertion `false' failed.
File rmodFileName = netcdf.rmod NOT FOUND (expected to be present)
error: No declaration found for the module: netcdf
Error : cannot find the module (module should have been seen in a previously generated file) : netcdf
swacc_real: ../../../../edg4x-rose/src/frontend/OpenFortranParser_SAGE_Connection/FortranParserActionROSE.C:17796: void c_action_use_stmt(Token_t*, Token_t*, Token_t*, Token_t*, Token_t*, ofp_bool, ofp_bool, ofp_bool): Assertion `false' failed.
/usr/sw-mpp/bin//swafort: line 79: 11427 Aborted                 (core dumped) $REAL_SWACC -D__SWACC__ -D_OPENACC=20160806 -I/usr/sw-mpp/mpi2/include -fortran_support -lfortran_slave $NEW_ARG
fcm_internal compile failed (34304)make: *** [fliocom.o] Error 1make: *** Waiting for unfinished jobs....
/usr/sw-mpp/bin//swafort: line 79: 11428 Aborted                 (core dumped) $REAL_SWACC -D__SWACC__ -D_OPENACC=20160806 -I/usr/sw-mpp/mpi2/include -fortran_support -lfortran_slave $NEW_ARG
fcm_internal compile failed (34304)make: *** [flincom.o] Error 1

@段晓辉 我试了下，如果abcd在ab的下面，代码直接读取abcd也是可以一次读出来的，所以感觉不像是必须按顺序来读。感觉是匹配的时候按顺序查找，匹配到了上面的abcd里有ab，就不再往下查找了，造成读取失败。举个不恰当的例子，就像正则匹配最后没有加 $ 只匹配了一部分

最近遇到两个问题，有些不明白

1. 神威上的fortran程序读取namelist文件时，似乎跟顺序有关，即读取的变量名如果和位于它之前的的变量名前半部分重合，就会读取不出来，下面是我抽象出来的一个程序

program main
     implicit none
     integer :: zzzz, ssss
   
     NAMELIST/ab/ ssss
     NAMELIST/abcd/ zzzz

     OPEN(UNIT=100, FILE='namelist', FORM='FORMATTED', ACCESS='SEQUENTIAL', STATUS='OLD', IOSTAT=iost)
     REWIND(100)
     READ(100, ab, IOSTAT = iost)
     print *, 'ab: iost=', iost, ssss
     REWIND(100)
     READ(100, abcd, IOSTAT = iost)
     print *, 'abcd: iost=', iost, zzzz
 end program main

namelist文件：

&abcd
   zzzz        =  6 
/
&ab 
   ssss      =  1
/

这样 ab 是读不出来的，把namelist中的ab和abcd调换位置就可以读出来了。

2. 下面这句 MINLOC(a, mask = b == MINVAL(b), dim = 1) 中的 mask = b == MINVAL(b)写法得到的结果是错误的0，先把b == MINVAL(b)存到中间变量再把中间变量赋值给mask参数结果就正确了

 program main
     implicit none
     INTEGER, DIMENSION(8) :: a, b
     a = (/1, 2, 3, 4, 2, 4, 3, 4/)
     b = (/704, 374, 264, 220, 384, 204, 288, 224/)
     print *, MINLOC(a, mask = b == MINVAL(b), dim = 1)
end program main

上面两个问题我在intel平台ifort试了下都没有问题，不知道是哪里出现了问题？是神威编译器的特性不支持上面两种情况吗？谢谢