OpenMP

Question 1

When was OpenMP first introduced?

Answer 1

1997

Question 2

What is the latest version of OpenMP? When was it released?

Answer 2

Version 6.0, released in November 2024.

Question 3

What are the three main ways OpenMP extends C, C++, and Fortran?

Answer 3

1. Compiler Directives
2. Environment Variables
3. Runtime Library Routines

Question 4

Describe the OpenMP execution model.

Answer 4

Uses a fork-join model:
1. Execution starts with a single thread (master thread)
2. Worker threads start (fork) on entry to a parallel region
3. Worker threads exit (join) at end of parallel region

Question 5

How do you include an OpenMP directive in C/C++?

Answer 5

#pragma omp directive [clause1, ...]

Question 6

How do you create a parallel region in OpenMP?

Answer 6

#pragma omp parallel

Good practice to enclose parallel region in {} braces

Question 7

What compiler flag is required for OpenMP in GCC?

Answer 7

-fopenmp

Question 8

How can you control the number of threads used in OpenMP?

Answer 8

Set the OMP_NUM_THREADS environment variable (e.g., export OMP_NUM_THREADS=4).

Question 9

What happens if the number of threads is not specified?

Answer 9

The implementation chooses a default value.

Question 10

What does the OpenMP runtime library provide?

Answer 10

Functions for:
- Execution Environment Control
- Timing
- Device Memory Management
- Locking

Question 11

Name three OpenMP functions that query execution environment details.

Answer 11

1. omp_get_thread_num(): Returns the thread number
2. omp_get_num_threads(): Returns the total number of threads
3. omp_get_num_procs(): Returns the number of available processors

Question 12

What header file must be included to use OpenMP functions in C?

Answer 12

#include <omp.h>

Question 13

Why would conditional compilation be needed for OpenMP programs?

Answer 13

To allow us to build the serial version of the program.

Question 14

What preprocessor directive is used for conditional compilation in OpenMP?

Answer 14

#ifdef _OPENMP

Question 15

How can you conditionally include OpenMP-specific code?

Answer 15

#ifdef _OPENMP
code_here…
endif

Question 16

What does the _OPENMP macro indicate?

Answer 16

It is defined only when OpenMP is enabled during compilation.

Question 17

Why is loop parallelisation important in OpenMP?

Answer 17

It allows threads to carry out different iterations of a loop simultaneously, improving performance.

Question 18

What directive is used to parallelise a loop inside a parallel region?

Answer 18

#pragma omp for

Question 19

How do you combine parallel region creation and loop parallelisation?

Answer 19

#pragma omp parallel for

Question 20

What are the two types of variable scoping in OpenMP?

Explain each

Answer 20

1. Shared: A variable is accessible by all threads
2. Private: Each thread has its own copy of the variable

Question 21

How can you explicitly define variable scope in OpenMP?

Answer 21

Use the default, private, and shared clauses in OpenMP directives.

Question 22

What does default(none) do in OpenMP?

Answer 22

It forces all variables to be explicitly declared as private or shared.

Question 23

What issue can occur when multiple threads modify a shared variable simultaneously? Why is it bad?

Answer 23

A data race, which can lead to incorrect results.

Question 24

How can you safely sum values in parallel?

Answer 24

Use a reduction variable:
#pragma omp for reduction(+ : variable_name)

Question 25

What is a data dependency?

Answer 25

When the execution of one statement depends on another

Question 26

What is a loop carried dependency?

Answer 26

When there is a dependence between statements executed by different iterations of a loop

Question 27

What are the three types of loop-carried dependencies?

Answer 27

1. **Flow Dependency/True Dependency (Read after Write):** iteration `i+1` depends on iteration `i` 2. **Anti-dependency (Write after Read):** iteration `i` depends on iteration `i+1` 3. **Output Dependency (Write after Write):** Multiple iterations write to the same memory location.

Question 28

What OpenMP feature helps retain the sequentially last computed value in a parallel loop?

Answer 28

The `lastprivate` clause

Question 29

What does `lastprivate(x)` do?

Answer 29

Ensures that the value of `x` from the last loop iteration is preserved after the parallel region ends. ## Footnote Where last does not mean in time, rather the iteration that "should be" last