OpenMP pragma parallel for only using a single thread

Question

I am refactoring some old code which uses OpenMP to parallelise some heavy calculations. The program is written in C.

System specs:

• 2x 48-Core Xeons
• 750GB RAM
• Ubuntu 20

It is worth noting the the system has multiple compilers and libraries installed using the env_modules. The nvhpc-cuda/22.2 and nvhpc/22.2 modules are available.

Background: The code in question was not written be me. I was tasked with refactoring an extremely large file into smaller files. I can't give exact code due to intellectual property.

Here is an example, resembling my problem:

// some other code in a function...


#pragma omp parallel for private(i, j, x, y, z) num_threads(30)
for (i = 0; i < z - 1; i++) 
{
   // do lots of system calls, I/O (each file is only opened by one iteration, should not crash into each other)
   // Call functions which rely on global variables
}

When this runs it appears to only be launching a single thread (as seen by running htop).

I ran this to investigate if we are able to see the cores and do any parallelisation:

#pragma omp parallel num_threads(omp_get_max_threads())
  {
    printf("Thread %d of %d\n", omp_get_thread_num(), omp_get_num_threads());
    printf("Max threads: %d\n", omp_get_max_threads());
  }

// Output:
// >> Thread 0 of 1
// >> Max threads: 96

So it can see the max number of cores, but isn't spawning threads for them

Finally, I am using cmake to build (I think this is where the problem is, as it worked before the refactor, which did not use cmake).

cmake_minimum_required(VERSION 3.16)
project(myexe VERSION 1.0.0 LANGUAGES C)

set(CMAKE_DEBUG_POSTFIX _d)

# Unload nvhpc-cuda module (as this breaks gcc) - we would need to use pgcc instead.
find_package(EnvModules REQUIRED)
env_module(unload nvhpc-cuda/22.2)

# Use gcc (since it is OpenMP compatible)
set(CMAKE_C_STANDARD 11)
set(CMAKE_C_COMPILER gcc)
# disable warnings
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wno-unused-variable -Wno-unused-parameter")

# Add linker flags
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fopenmp")

include_directories(include)
include_directories(/path/to/internal/headers)

link_directories(/path/to/internal/lib)

file(GLOB SOURCES "src/*.c")
add_executable(myexe ${SOURCES})
set_target_properties(myexe PROPERTIES DEBUG_POSTFIX ${CMAKE_DEBUG_POSTFIX})

target_link_libraries(myexe PRIVATE m pthread somelibrary_which_needs_pthread)

I have attempted to build using both pgcc and gcc.

The original build command for the single file was pgcc myfile.c -O3 -fopenmp -lpthread -lsome_library_which_needs_pthread -o myexe