of2.2.x gpu版在amd平台编译成功后运行一定步数出错

Xiaomin

小白一枚，看不懂错误意思，麻烦老师们看看

编译成功使用smoothed_aggregation预条件后运行17个循环以后报如下错误：

#0  Foam::error::printStack(Foam::Ostream&) at ??:?
#1  Foam::sigFpe::sigHandler(int) at ??:?
#2   in "/lib64/libpthread.so.0"
#3  void cusp::krylov::cg<cusp::ell_matrix<int, float, thrust::hip_rocprim::tag>, cusp::array1d<float, thrust::hip_rocprim::tag>, ofgpu::SolverTelemetry, cusp::precond::aggregation::smoothed_aggregation<int, float, thrust::hip_rocprim::tag, thrust::use_default, thrust::use_default, thrust::use_default> >(cusp::ell_matrix<int, float, thrust::hip_rocprim::tag>&, cusp::array1d<float, thrust::hip_rocprim::tag>&, cusp::array1d<float, thrust::hip_rocprim::tag>&, ofgpu::SolverTelemetry&, cusp::precond::aggregation::smoothed_aggregation<int, float, thrust::hip_rocprim::tag, thrust::use_default, thrust::use_default, thrust::use_default>&) at ??:?
#4  void ofgpu::SparseMatrixSystem::solve<ofgpu::PCGSolver>(ofgpu::SparseMatrixArgs const&) at ??:?
#5  Foam::PCGgpu::solve(Foam::Field<float>&, Foam::Field<float> const&, unsigned char) const at ??:?
#6  Foam::fvMatrix<float>::solveSegregated(Foam::dictionary const&) at ??:?
#7  Foam::fvMatrix<float>::solve(Foam::dictionary const&) at ??:?
#8  Foam::fvMatrix<float>::solve() at ??:?
#9
 in "/public/home/xxx/OpenFOAM/OpenFOAM-2.2.x/platforms/linux64GccSPDebug/bin/simpleFoam"
#10  __libc_start_main in "/lib64/libc.so.6"
#11
 in "/public/home/xxx/OpenFOAM/OpenFOAM-2.2.x/platforms/linux64GccSPDebug/bin/simpleFoam"
/opt/gridview/slurm/spool_slurmd/job5380749/slurm_script: line 22: 83076 Floating point exception~/OpenFOAM/OpenFOAM-2.2.x/platforms/linux64GccSPDebug/bin/simpleFoam

[定位到代码中是blas文件中的函数axpby，如下图]( .)

axpby（）函数的功能是将两个向量相加，y = αx + βy，我尝试用axpy（）函数替代，结果也是在相同位置报错。axpy（）：y = α*x + y

李东岳

22x这个太老了。另外GPU版本我这面也没编译过。没啥经验咯..

smoothed_aggregation预条件

这个能不能换一个别的

Xiaomin

@李东岳好像是因为最后发散了，所以报错，一开始的time step continuity errors就很大，不知道什么原因，时间步长连续性误差会随步长增长，老师有修改的建议嘛

李东岳

一开始的time step continuity errors就很大

稳态是可能得。瞬态要保证很小。

时间步长连续性误差会随步长增长，

这个看起来设置不太对。可以先用CPU跑没问题了之后，再跑GPU

Xiaomin

This post is deleted!