Package: g++
Version: 4:9.2.1-3
When executing code compiled with g++ 9.2.1 that has OpenMP for-loops
with vectors as shared variables, it results in a segmentation fault.
According to Asan (address sanitizer), this might be due to double-
freed variables when running the code.
Example:
#pragma omp parallel for schedule(static) num_threads(nthreads)
shared(some_vector, some_variable)
for (int i = 0; i < some_vector.size(); i++) {
some_func(some_vector[i], some_variable)
}
Setup:
- Processor Ryzen 2700 (8 cores)
- Kernel version 5.2+106
- Compile flags tried: -std=c++11 -fopenmp -march=native -O2
The version currently in buster (4:8.3.0-1) and older do not show this
problem. Tried with different samples of code in different projects and
they all presented the same issue. Potential problems such as out-of-
bounds writes were already discarded. Compiling without OpenMP does not
produce such error.