How to map private dynamic array to the GPU with OpenMP and nvc?

Accelerated Computing HPC Compilers nvc, nvc++ and nvfortran
13

I don’t know if this is a compiler bug, when I compiled this commit in the openacc branch using make nvc BACKEND=acc, which runs this command

nvc++  -g -Wall -Wextra -pedantic -std=c++11 -O3 -Minfo=mp,accel -Minline -gpu=mem:managed,ptxinfo -static-nvidia -DGPU_OFFLOAD -acc=gpu  -c umcx.cpp -o umcx.o

it produced the following Internal compiler error error, and pointing to this innocent looking line (#639)

NVC++-F-0000-Internal compiler error. Unexpected operation: eok_ref_indirect       0  (umcx.cpp: 639)
NVC++/x86-64-Extractor Linux 24.11-0: compilation aborted
make: *** [Makefile:52: umcx.o] Error 2

PS: never mind, I see your comments on the rationales of defining int ppathlen. I believe the above error is a side effect of the firstprivate handling. After defining int ppathlen, the above error is gone.