Threading

Question 1

how to pass data

Answer 1

by pointer
thread(func, &arg)
void* func(int* args) { ++(*args);}

by reference
thread(func, std::ref(arg))
void* func(int &args) { ++args;}

Question 2

user level thread and system level

Answer 2

user level more simple, no os needed
but cant force it to give up hw

pro: lightweight
con: no scheduling guarantees

usecase: light fine grained tasks

Question 3

yield and spin locks

Answer 3

spin
thread will keep probind until lock free
fast, but uses more resources hw

yield
if locked it gives up hw
low resources, but slow and bad for hpc

Question 4

lock granulairyt

Answer 4

coarse: one lock for everythign
easy, but low concurrency

fine: one lock for each data element
good concurrency, but deadlcoking

Question 5

how to avoid lock deadlock

Answer 5

1. order locks
2. hold 1 lock at a time
3. have a central arbiter (overhead)

Question 6

code threads

Answer 6

include mutex
include thread
std::mutex mtx;

void func() {
mtx.lock;
var access
mtx.unlock(); }

main {
std::thread threads[10];
for (loop 10) {
threads[i] = std::thread(func, arg); }
for (auto &th: threads) {
th.join(); }