OpenMP Environment Variables

Environment variables provide system-wide configuration for OpenMP programs without code modification. They control thread count, scheduling policies, processor binding, and other runtime behaviors. This enables performance tuning, testing, and deployment flexibility across different systems. Understanding these variables is essential for optimizing parallel applications and adapting to varying hardware configurations without recompilation.

Core Concept

OpenMP environment variables are set in the shell before program execution and affect all OpenMP directives unless overridden by code-level specifications. Key variables include OMP_NUM_THREADS (thread count), OMP_SCHEDULE (loop scheduling policy), OMP_PROC_BIND (thread affinity), and others controlling nested parallelism, stack size, and dynamic adjustment.

Key Points

OMP_NUM_THREADS: Sets default thread count (e.g., export OMP_NUM_THREADS=4)
OMP_SCHEDULE: Default scheduling for loops (e.g., dynamic,10)
OMP_PROC_BIND: Thread-to-processor binding (true, false, master, close, spread)
OMP_NESTED: Enables nested parallelism (deprecated in OpenMP 5.0)
OMP_STACKSIZE: Thread stack size (e.g., 8M, 1G)
OMP_WAIT_POLICY: Thread waiting behavior (active or passive)

Code Example

Program demonstrating environment variable effects on execution

OpenMP Implementation:

#include <stdio.h>
#include <stdlib.h>
#include <omp.h>

int main() {
    // Display current environment variable settings
    printf("Environment Variable Settings:\n");
    printf("  OMP_NUM_THREADS = %s\n", getenv("OMP_NUM_THREADS") ? getenv("OMP_NUM_THREADS") : "not set");
    printf("  OMP_SCHEDULE = %s\n", getenv("OMP_SCHEDULE") ? getenv("OMP_SCHEDULE") : "not set");
    printf("  OMP_PROC_BIND = %s\n\n", getenv("OMP_PROC_BIND") ? getenv("OMP_PROC_BIND") : "not set");

    // Show actual runtime configuration
    printf("Runtime Configuration:\n");
    printf("  Max threads: %d\n", omp_get_max_threads());
    printf("  Nested parallelism: %d\n", omp_get_nested());
    printf("  Dynamic adjustment: %d\n\n", omp_get_dynamic());

    // Parallel region using environment settings
    #pragma omp parallel
    {
        #pragma omp master
        {
            printf("Parallel Execution:\n");
            printf("  Active threads: %d\n", omp_get_num_threads());
        }
    }

    // Loop with schedule affected by OMP_SCHEDULE
    printf("\nParallel loop execution:\n");
    #pragma omp parallel for schedule(runtime)
    for (int i = 0; i < 8; i++) {
        printf("  Iteration %d on thread %d\n", i, omp_get_thread_num());
    }

    return 0;
}

Running with Environment Variables:

export OMP_NUM_THREADS=4
export OMP_SCHEDULE="dynamic,2"
./env_vars_demo

Usage & Best Practices

When to Use

Deploying applications across different systems
Performance tuning without recompilation
Testing scalability with various configurations
Batch job submission with specific requirements

Best Practices

Set OMP_NUM_THREADS to match available cores
Use OMP_SCHEDULE="dynamic" for load-balanced irregular workloads
Enable OMP_PROC_BIND=true for NUMA systems
Document required environment settings in README
Provide sensible defaults in code as fallback

Common Mistakes

Forgetting to export variables (use export)
Setting conflicting values in script vs environment

Key Takeaways

Summary:

Environment variables configure OpenMP runtime without code changes
OMP_NUM_THREADS controls default thread count
OMP_SCHEDULE sets loop scheduling policy for schedule(runtime)
OMP_PROC_BIND manages thread affinity
Code-level specifications override environment variables

Quick Reference

Essential Environment Variables:

export OMP_NUM_THREADS=8           # Thread count
export OMP_SCHEDULE="dynamic,10"   # Loop scheduling
export OMP_PROC_BIND=true          # Thread binding
export OMP_STACKSIZE=8M            # Thread stack size
export OMP_WAIT_POLICY=passive     # Waiting policy
export OMP_DISPLAY_ENV=true        # Show settings at startup

Query in Shell:

echo $OMP_NUM_THREADS
env | grep OMP_

References:

OpenMP Official Site

Go back to OpenMP tutorials.

Learn Parallel Programming