OpenMP

Question 1

three primary API components

Answer 1

1 compiler directives
2 runtime library rountines
3 environment variables

Question 2

ICV thread numbers

Answer 2

OMP_NUM_THREADS
num_threads(n) pragma!

Question 3

sections

Answer 3

pragma sections, pragma section
- each sections is run by 1 thread
- implicit barrier at end of sections region

Question 4

for

Answer 4

• no sych at beginning
• implciit barrier at end (nowait to delete it)
• must have no ata dependency

Question 5

schedule

Answer 5

• used for the FOR loop
  1. static: round robin, fixed chunks of n/t
  2. dynamic: given one by one at runtime, chunks 1 (good for load balance)
  3. guided: start with bigger chunks, then smaller exponentially, one by one at runtime
  4. runtime: set at runtime
  5. auto: compiler/runtime chooses

Question 6

sharing attributes

Answer 6

1. private
2. shared
3. default
4. firstprivate
5. lastprivate

Question 7

list synchronization pragmas

Answer 7

barrier, masked region, single region. critical section, atomic statement, ordered contruct

Question 8

barrier

Answer 8

pragma omp barrrier

synchs all threads
can cause load imbalance, use only when needed

Question 9

masked Construct

Answer 9

pragma omp masked [ filter(integer-expression)

• only primary thread executed code, the others skip
• no implied barrier at either end

Question 10

single Construct

Answer 10

pragma omp single

• implicit barrier!
• a thread executes it
• clauses like private(list) firstprivate(list)
• like initializeing data structures

Question 11

ciritical construct

Answer 11

pragma omp critical [(name)

restricts execution of the associated structured block to a single thread at a time
- no implicit barrier?!
- can cause load imbalance, only use when really needed

Question 13

atomic statement

Answer 13

pragma omp atomic

so a memory location is updated atomically
- like critical, but less overhead, but also only 1 operation, avoid locking
-

Question 14

ordered construct

Answer 14

pragma omp ordered

block of code that must be executed in sequential order.
- The ordered construct sequentializes and orders the execution of ordered regions while allowing code outside the region to run in parallel.
- clauses: threads (default), simd
- for!!!

Question 15

omp locks

Answer 15

kinda like mutex

omp_lock_t lockvar; initializes a simple lock
omp_init_lock(&lockvar);
omp_destroy_lock(&lockvar); uninitializes a simple lock
omp_set_lock(&lockvar) waits until a simple lock is available and then sets it
omp_unset_lock(&lockvar) unsets lock
omp_test_lock(&lockvar) tests, if true sets lock

nestable locks possible

Question 16

reduction

Answer 16

reduction (operator: list)

+ - * & ^ | && || max min

Question 17

correctness issues

Answer 17

data races: unsyched conflicting access to shared variables
loop dependencies: prevents parallelization in loops
aliasing: hidden loop dependencies,breaks compiler optimizations

Question 18

types of data races

Answer 18

1. read after write
2. write after read
3. write after write

Question 19

types of dependecies

Answer 19

1. true
2. anti
3. output

Question 20

types of loop dependecies

Answer 20

loop carried
loop independent

Question 21

list loop transformations

Answer 21

4 interchange, distribution, fusion, alignment

Question 22

loop interchange

Answer 22

swap inner and outer loop
not for loop carried dependencies
used to get better cache use (row first!)

Question 23

loop distribution

Answer 23

to eliminate loop carried dep,

Question 24

loop fusion

Answer 24

may create loop carried dep
reduce overhead!!
eliminates need of barrier

Question 25

loop alignment

Answer 25

put 1 operation before and 1 after

Question 26

why is work sharing bad?

Answer 26

1. load imbalance (both static and dynamic scehduling have drawbacks)
2. imbalance because of machine
3. limited program flexibility

this means we need tasks!!

Question 27

tasks

Answer 27

pragma omp task

-inside a single region
- untied: tasks can move to different threads, may lose cache
- if: when to defer task
- sharing: default is firstprivate
- priority influences execution order

taskwait: waits for completion of immediate child tasks

taskyield: current task can be suspended (if takes too long)

dependencies: in, out, inout

PRO: load balance, simple

task granularity:
1. fine: more overhead, better resource use
2. coarse: schedule fragmentation

Question 28

omp runtime routines

Answer 28

omp_set_num_threads(N)
omp_get_max_threads()
omp_get_num_threads() size
omp_get_thread_num() rank

Question 29

omp ICVs

Answer 29

OMP_SCHEDUL
OMP_DYNAMIC
OMP_NESTED