Improving compiler error with OpenACC + OpenMP: "Internal compiler error. confused OMP private processing"

Accelerated Computing HPC Compilers nvc, nvc++ and nvfortran
1

Hello,

This example has an error on user part but the error message could be better and hence I thought I should report it:

If we have simple example with OpenMP as well as OpenACC pragmas together (just for demonstration):

void test(){
  int N = 1e4;
  float *x = new float[N];
  float *y = new float[N];

  #pragma omp target teams distribute parallel for map(to: x[0:N]) map(from: y[0:N])
  #pragma acc parallel loop copyin(x[0:N]) copy(y[0:N])
  for(int i = 0; i < N; i++){
    for(int j = 0; j < 1000; j++){
      y[i] += 3*x[i];
    }
  }

  printf("%g\n", y[5]);

  delete [] x;
  delete [] y;
}

I can compile this with nvc++ -mp=gpu or nvc++ -acc as expected. But if I try to use both -mp=gpu -acc then we get:

NVC++-S-0000-Internal compiler error. confused OMP private processing        0  (test.cpp: 60)
nvc++-Fatal-/gpfs/bbp.cscs.ch/ssd/apps/hpc/jenkins/pulls/1334/deploy/externals/2021-10-15/linux-rhel7-x86_64/gcc-9.3.0/nvhpc-21.9-wsrbqb/Linux_x86_64/21.9/compilers/bin/tools/cpp2 TERMINATED by signal 11
Arguments to /gpfs/.../nvhpc-21.9-wsrbqb/Linux_x86_64/21.9/compilers/bin/tools/cpp2
/gpfs/.../nvhpc-21.9-wsrbqb/Linux_x86_64/21.9/compilers/bin/tools/cpp2 test.cpp -opt 2 -terse 1 -inform warn -x 119 0xa10000 -x 122 0x40 -x 123 0x1000 -x 127 4 -x 127 17 -x 19 0x400000 -x 28 0x40000 -x 120 0x10000000 -x 70 0x8000 -x 122 1 -x 125 0x20000 -quad -x 59 4 -tp skylake -x 120 0x1000 -astype 0 -x 121 1 -fn test.cpp -il /gpfs/.../kumbhar/2790017/nvc++NsRr1NMx-Exe.il -x 117 0x200 -x 123 0x80000000 -x 123 4 -x 119 0x20 -def __pgnu_vsn=90300 -x 70 0x40000000 -x 183 4 -x 121 0x800 -x 6 0x20000 -x 249 110 -x 120 0x200000 -x 70 0x40000000 -x 8 0x40000000 -x 164 0x800000 -x 85 0x2000 -x 85 0x4000 -x 34 0x40000000 -x 53 0x800000 -x 206 0x02 -x 68 0x1 -x 39 4 -x 56 0x10 -x 26 0x10 -x 26 1 -x 56 0x4000 -accel tesla -accel tesla -accel host -x 180 0x4000400 -x 121 0xc00 -x 186 0x80 -x 163 0x1 -x 186 0x80000 -cudaver 11040 -x 194 0x40000 -x 176 0x100 -cudacap 70 -x 180 0x4000400 -x 121 0xc00 -x 186 0x80 -x 180 0x4000400 -x 121 0xc00 -x 194 0x40000 -x 163 0x1 -x 186 0x80000 -cudaver 11040 -x 176 0x100 -cudacap 70 -cudaroot /gpfs/.../nvhpc-21.9-wsrbqb/Linux_x86_64/21.9/cuda/11.4 -x 189 0x8000 -y 163 0xc0000000 -x 201 0xf0000000 -x 189 0x10 -y 189 0x4000000 -cudaroot /gpfs/.../nvhpc-21.9-wsrbqb/Linux_x86_64/21.9/cuda/11.4 -x 187 0x40000 -x 187 0x8000000 -mp -x 69 0x200 -x 69 0x400 -x 69 2 -x 0 0x1000000 -x 2 0x100000 -x 0 0x2000000 -x 161 16512 -x 162 16512 -gnuvsn 90300 -x 69 0x200 -x 123 0x400 -x 180 0x4000000 -x 194 0x20000000 -x 233 0x1 -x 205 2 -x 233 0x100 -x 233 0x400 -cmdline '+nvc++ /gpfs/.../kumbhar/2790017/nvc++NsRr1NMx-Exe.il -D__GCC_ATOMIC_TEST_AND_SET_TRUEVAL=1 -mp=gpu -acc -c -Minfo=accel,mp' -asm /gpfs/.../kumbhar/2790017/nvc++xsRrfA4ol_Tn.ll

This is with NVHPC 21.9. The error could be better here?

2

It’s obviously incorrect to try and use both parallel models’ compute regions on the same loop, but, yes, the compiler shouldn’t give an ICE. I added a problem report (TPR #30770) and sent to engineering for review.

Thanks!
Mat