ComputeResidual.cpp:60:5: error: variable 'n' must have explicitly specified data sharing attributes about hpcg

[jdayal-ssi]

Fixed this by changing the line
shared(local_residual, v1v, v2v,)
to
shared(local_residual, v1v, v2v, n)

[maherou]

The variable 'n' should be shared. Try adding it to the list of shared variables, just after v2v.

…

________________________________ From: jdayal-ssi ***@***.***> Sent: Friday, September 22, 2023 5:09 PM To: hpcg-benchmark/hpcg ***@***.***> Cc: Subscribed ***@***.***> Subject: [EXTERNAL] [hpcg-benchmark/hpcg] ComputeResidual.cpp:60:5: error: variable 'n' must have explicitly specified data sharing attributes (Issue #81) I am trying to compile hpcg 3.1 on an Ampere neoverse_n1. I am trying to build it with clang, mpich and openmp. Versions: mpich 4.1.1 clang: 16.0.6 OS: rocky linux 8 CXXFLAGS: -O3 -ffast-math -ftree-vectorize -fopenmp When I try to compile, I get the following errors and warnings: ***@***.***:~/hpcg-3.1/neoverse_n1$ make mpicxx -c -I./src -I./src/neoverse_n1 -O3 -ffast-math -ftree-vectorize -fopenmp -I../src ../src/ComputeResidual.cpp -o src/ComputeResidual.o ../src/ComputeResidual.cpp:60:5: error: variable 'n' must have explicitly specified data sharing attributes for (local_int_t i=0; i<n; i++) { ^~~ ../src/ComputeResidual.cpp:56:32: note: explicit data sharing attribute requested here #pragma omp parallel default(none) shared(local_residual, v1v, v2v) ^ 1 error generated. make: *** [Makefile:76: src/ComputeResidual.o] Error 1 Any advice? Thanks in advance. — Reply to this email directly, view it on GitHub<#81>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AAHZIS5VI6FURXHZZQCJ52DX3X5CJANCNFSM6AAAAAA5DUNNUQ>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[maherou]

Sorry. Just saw this. Glad the fix worked.

…

________________________________ From: jdayal-ssi ***@***.***> Sent: Friday, September 22, 2023 6:20 PM To: hpcg-benchmark/hpcg ***@***.***> Cc: Subscribed ***@***.***> Subject: [EXTERNAL] Re: [hpcg-benchmark/hpcg] ComputeResidual.cpp:60:5: error: variable 'n' must have explicitly specified data sharing attributes (Issue #81) Closed #81<#81> as completed. — Reply to this email directly, view it on GitHub<#81 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AAHZIS7OZ4DDDYTMLMCMR4LX3YFMJANCNFSM6AAAAAA5DUNNUQ>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[luszczek]

I cannot reproduce this problem with Clang 14 through 16 nor with godbolt.org where I used this source code:

#include <cmath>

typedef int local_int_t;
struct Vector_STRUCT{local_int_t localLength;double * values;void*optimizationData;};
typedef struct Vector_STRUCT Vector;
int ComputeResidual(const local_int_t n, const Vector & v1, const Vector & v2, double & residual) {

  double * v1v = v1.values;
  double * v2v = v2.values;
  double local_residual = 0.0;

#pragma omp parallel shared(local_residual, v1v, v2v)
  {
    double threadlocal_residual = 0.0;
    #pragma omp for
    for (local_int_t i=0; i<n; i++) {
      double diff = std::fabs(v1v[i] - v2v[i]);
      if (diff > threadlocal_residual) threadlocal_residual = diff;
    }
    #pragma omp critical
    {
      if (threadlocal_residual>local_residual) local_residual = threadlocal_residual;
    }
  }
  residual = local_residual;
  return 0;
}

It would be good to know the origins of the tool chain used for MPICH 4.1.1. Was it system default installation or it came from another packag manager?