Error compiling with 64 bit ints and MPI使用 64 位整数和 MPI 编译时出错

I'd like to run some semi-large CI expansions, where I need to compile dalton with 64 bit ints (more than 2^32 - 1 configurations). Also, I'd like to use MPI for inter-node parallelism.
我想运行一些半大规模的 CI 扩展，其中需要使用 64 位整数编译 dalton（超过 2^32 - 1 个构型）。此外，我还希望利用 MPI 实现节点间的并行计算。

I can compile with 64 bit ints without problems when not using MPI.
在使用非 MPI 时，我可以毫无问题地编译 64 位整数。

I have tried two setups, both giving too many errors.
我尝试了两种设置，结果都出现了过多的错误。

a) Dalton 2018.2, Intel 2019, mpich and mkl
a) 道尔顿 2018.2, 英特尔 2019, mpich 和 mkl

$ ./setup --int64 --fc=mpifort --cc=mpicc --cxx=mpicxx

b) Dalton 2018.2, GCC 9.2, mpich and lapack-i8
b) Dalton 2018.2, GCC 9.2, mpich 及 lapack-i8

$ ./setup --int64 --fc=mpif90 --cc=mpicc --cxx=mpicxx

Attached are the output from make.
附上 make 的输出结果。

Approach a) gives a bunch of warnings related to the C files in the DFT modules, which are essentially the same as the one I got when compiling without MPI. However, I also get errors related to the lucita parallelization stuff.
方法 a)在编译 DFT 模块的 C 文件时产生了一系列警告，这些警告实质上与我未使用 MPI 编译时遇到的相同。然而，我还遇到了与 lucita 并行化相关的错误。

Approach b) gives errors related to parallel routines from abacus it would seem.
方法 b)似乎会产生与算盘并行程序相关的误差。

log_gcc 日志_gcc

log_intel 日志_英特尔

Any help is appreciated.
任何帮助都感激不尽。

Also, thanks for maintaining dalton 😄
另外，感谢维护 dalton 😄

changed the description

changed the description

changed title from Error compiling with 64 bit ints to Error compiling with 64 bit ints and MPI

@stknecht @hjaaj

So, this was probably just me being silly. Obviously I need a 64-bit MPI library as well. I have installed dalton with intel, mkl and openmpi (built with 64 bit ints) and it's working now.
所以，这可能只是我一时糊涂。显然，我还需要一个 64 位的 MPI 库。我已经用 Intel、MKL 和 OpenMPI（构建时使用 64 位整数）安装了 Dalton，现在它已经正常工作了。

I still get warnings related to the DFT .c files.
我仍然收到与 DFT .c 文件相关的警告。

I suppose one could add a check in the setup stage to see, if the chosen MPI library supports 64 bits.
我认为可以在设置阶段添加一个检查，以确认所选的 MPI 库是否支持 64 位。

I realize this is already being done with Performing Test MPI_F90_I8. When using openmpi, the tests passes and fails with mpich. However, I guess simply giving an error at the setup stage when specifying --int64 and using a 32-bit MPI library would be useful for someone like me.
我意识到这已经在使用 Performing Test MPI_F90_I8 时实现了。在使用 openmpi 时，测试通过，而使用 mpich 时则失败。不过，我想如果在设置阶段指定 --int64 并使用 32 位 MPI 库时直接给出错误提示，对于像我这样的人来说会很有帮助。

Attached is, for reference, the build log when using openmpi.
附上使用 openmpi 时的构建日志，供参考。

log_openmpi-i8 日志_openmpi-i8

There is a mechanism in place to check for 32- and 64-bit MPI, as you noted, but it is possibly broken. Do I understand correctly that the tests pass when you use OpenMPI? All of them?
你提到的检查 32 位和 64 位 MPI 的机制可能已经失效。我理解正确吗？当你使用 OpenMPI 时，所有测试都能通过吗？

I forgot to test the installation. All DFT tests fail, likely due to the warnings mention earlier. Other than than, some tests fail here and there, for a total of 90 failed out of 496. Some coupled cluster tests are failing because the implementation is not MPI parallelizable.
我忘记测试安装了。所有 DFT 测试都失败了，很可能是由于之前提到的警告。除此之外，还有一些测试零星失败，总共 496 个测试中有 90 个失败。一些耦合簇测试失败是因为实现方式无法进行 MPI 并行化。

Attached is LastTest.log of running ctest with 4 MPI processes.
附上使用 4 个 MPI 进程运行 ctest 的 LastTest.log 结果。

LastTest.log

The DFT test are known to fail with 64-bit integers also without MPI (issue #59 (closed)). Can we consider your issue with 64-bit integers and MPI as resolved?
已知 DFT 测试在无 MPI 的情况下对 64 位整数也会失败（问题#59 已关闭）。我们能否认为您关于 64 位整数与 MPI 的问题已得到解决？

Yes 是

closed