L14 & L15

Question 1

What are 2 advantages of std::thread over pthreads?

Answer 1

Type checking is possible

less repetitive code

Question 2

How does the std lock, lock_guard, act?

Answer 2

locks at creating and unlocks at destruction

Question 3

What is parallel STL? What is an advantage of it?

Answer 3

Parallel STL provides high level parallelism through the C++ standard library. It provides parallelized algorithms for performing computations concurrently on multi-core processors.

It can help prevent data races

Question 4

What are the 4 execution policies in Parallel STL?

Answer 4

 Sequenced_policy
 Unsequenced_policy (SIMD)
 Parallel_policy
 Parallel_unsequenced_policy

Question 5

how are Parallel STL algorithms related to functional programming

Answer 5

STL algorithms are compositional, meaning composable transformations can be made on data, rather than sequences of instructions

Question 6

Explain std::future from the C standard library.

Answer 6

Used for accessing results of asynchronous operations through shared memory

Allows offload computation (std::async, std::packaged_task, std::promise)

Question 7

What 3 level abstractions offer an alternative to high level and low level standard parallelism, outlining an advantage of each?

Answer 7

Parallel libraries: low effort from programmer:

Domain specific languages: easier to program, optimise and parallelise (ex: Matlab, Tensorflow

Algorithmic skeletons: focusses on how the computation flows not what it does, utilising composability

Question 8

Is OpenMP an algorithmic skeleton?

Answer 8

No, because it is not particularly composable

Question 9

What are some disadvantages of algorithmic skeletons?

Answer 9

Some parallelism is not structured

Synchronisation automated but still an overhead

Question 10

What is the main goal for OpenMP?

Answer 10

Parallel performance for serial code with minimal effort

Question 11

What is used to start a parallel region in OpenMP?

Answer 11

pragma omp parallel

• # pragma: tells the compiler this is a directive
• omp: pass the directive to the OpenMP system
• parallel: block that follows is a parallel region

Question 12

What is reduction used for in a parallel for loop?

Answer 12

Reduction can be used to combine each thread’s results in a for loop

Question 12

In OpenMP, what is used for data parallelisation? What is used for task parallelisiation?

Answer 12

Loop splitting (parallel for)

Sections & tasks (task farm)

Question 13

Are the following variables shared or private:
a. Variable declared inside parallel region
b. Variable declared outside parallel region

Answer 13

a. shared
b. private

Question 14

What are some ways that OpenMP allows for coordination between threads?

Answer 14

 barrier
 nowait
 omp_(un)set_lock()
 critical
 atomic

Question 15

What are the two types of loop scheduling for loop splitting in OpenMP?

Answer 15

schedule (static, #chunk_size): progress as fast as slowest thread

schedule (dynamic, #chunk_size):each thread processes a chunk of iterations

Question 16

How is the depend() clause used in a OpenMP task?

Answer 16

depend() clause defines a list of items that are input or output dependencies of the task.

The scheduler does not start a task that depends on a value not yet generated.

Question 17

How can the variable visibility be set in OpenMP?

Answer 17

private()
shared()
firstprivate(): initialised to original value