MT impl WIP
This commit is contained in:
parent
c560bfe4cc
commit
bbf0c0346f
@ -62,14 +62,16 @@ auto main(int argc, char *argv[]) -> int {
|
|||||||
auto delay_ms = std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1);
|
auto delay_ms = std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1);
|
||||||
fmt::print("Sorted {} entries within {} ms in sequential\n", dataset_seq.size(), delay_ms.count());
|
fmt::print("Sorted {} entries within {} ms in sequential\n", dataset_seq.size(), delay_ms.count());
|
||||||
|
|
||||||
|
|
||||||
|
static constexpr int nthreads = 16;
|
||||||
t1 = std::chrono::high_resolution_clock::now();
|
t1 = std::chrono::high_resolution_clock::now();
|
||||||
algo::MergeSort_v2::sort(dataset_par.begin(), dataset_par.end(), [](int32_t a, int32_t b) {
|
algo::MergeSort_v2::sort(dataset_par.begin(), dataset_par.end(), [](int32_t a, int32_t b) {
|
||||||
return (a > b);
|
return (a > b);
|
||||||
}, 16);
|
}, nthreads);
|
||||||
t2 = std::chrono::high_resolution_clock::now();
|
t2 = std::chrono::high_resolution_clock::now();
|
||||||
|
|
||||||
delay_ms = std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1);
|
delay_ms = std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1);
|
||||||
fmt::print("Sorted {} entries within {} ms in parallel using {} threads\n", dataset_seq.size(), delay_ms.count(), 16);
|
fmt::print("Sorted {} entries within {} ms in parallel using {} threads\n", dataset_seq.size(), delay_ms.count(), nthreads);
|
||||||
|
|
||||||
auto eq = (dataset_seq == dataset_par);
|
auto eq = (dataset_seq == dataset_par);
|
||||||
fmt::print("Equality: {}\n", eq);
|
fmt::print("Equality: {}\n", eq);
|
||||||
|
@ -83,10 +83,9 @@ namespace algo {
|
|||||||
|
|
||||||
class MergeSort_v2 {
|
class MergeSort_v2 {
|
||||||
private:
|
private:
|
||||||
//Caution: non thread-safe, must be executed with additional protection (e.g. lock_guard)
|
|
||||||
template<typename Iterator, typename Comparator>
|
template<typename Iterator, typename Comparator>
|
||||||
static auto
|
static auto
|
||||||
merge(Iterator start, Iterator middle, Iterator end, Comparator cmp, Iterator output_start) -> void {
|
merge(Iterator start, Iterator middle, Iterator end, Comparator cmp, Iterator output_start, std::recursive_mutex &dataset_guard) -> void {
|
||||||
Iterator start_m = start;
|
Iterator start_m = start;
|
||||||
Iterator begin = output_start;
|
Iterator begin = output_start;
|
||||||
Iterator start2 = middle + 1;
|
Iterator start2 = middle + 1;
|
||||||
@ -116,20 +115,23 @@ namespace algo {
|
|||||||
output_start++;
|
output_start++;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
dataset_guard.lock();
|
||||||
const auto size = std::distance(start_m, end);
|
const auto size = std::distance(start_m, end);
|
||||||
for (auto i = 0; i <= size; i++, start_m++, begin++) {
|
for (auto i = 0; i <= size; i++, start_m++, begin++) {
|
||||||
*start_m = *begin;
|
*start_m = *begin;
|
||||||
}
|
}
|
||||||
|
dataset_guard.unlock();
|
||||||
|
|
||||||
}
|
}
|
||||||
|
|
||||||
template<typename Container, typename Iterator, typename Comparator>
|
template<typename Container, typename Iterator, typename Comparator>
|
||||||
static auto ms_split(Container &output_vec, Iterator start, Iterator end, Comparator cmp, Iterator output_start,
|
static auto ms_split(Container &output_vec, Iterator start, Iterator end, Comparator cmp, Iterator output_start,
|
||||||
int &nthreads, std::mutex &dataset_guard, std::mutex &depth_guard) -> void {
|
int &nthreads, std::recursive_mutex &dataset_guard, std::mutex &depth_guard) -> void {
|
||||||
Iterator mid = start;
|
Iterator mid = start;
|
||||||
Iterator begin = output_start;
|
Iterator begin = output_start;
|
||||||
|
|
||||||
if (std::distance(start, end) < 1) {
|
if (std::distance(start, end) < 1) {
|
||||||
|
//Quit on smalles list size (one element is always sorted)
|
||||||
return;
|
return;
|
||||||
} else {
|
} else {
|
||||||
if (nthreads > 1) {
|
if (nthreads > 1) {
|
||||||
@ -151,8 +153,7 @@ namespace algo {
|
|||||||
//merge everything together starting from the complete beginning
|
//merge everything together starting from the complete beginning
|
||||||
t1.join();
|
t1.join();
|
||||||
t2.join();
|
t2.join();
|
||||||
const std::lock_guard<std::mutex> lock(dataset_guard);
|
merge(start, mid, end, cmp, begin, dataset_guard);
|
||||||
merge(start, mid, end, cmp, begin);
|
|
||||||
} else {
|
} else {
|
||||||
//move mid iterator litterally to the mid
|
//move mid iterator litterally to the mid
|
||||||
std::advance(mid, std::distance(start, end) / 2);
|
std::advance(mid, std::distance(start, end) / 2);
|
||||||
@ -165,8 +166,7 @@ namespace algo {
|
|||||||
ms_split(output_vec, mid + 1, end, cmp, output_start, nthreads, dataset_guard, depth_guard);
|
ms_split(output_vec, mid + 1, end, cmp, output_start, nthreads, dataset_guard, depth_guard);
|
||||||
|
|
||||||
//merge everything together starting from the complete beginning
|
//merge everything together starting from the complete beginning
|
||||||
const std::lock_guard<std::mutex> lock(dataset_guard);
|
merge(start, mid, end, cmp, begin, dataset_guard);
|
||||||
merge(start, mid, end, cmp, begin);
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
@ -176,7 +176,7 @@ namespace algo {
|
|||||||
static auto sort(Iterator start, Iterator end, Comparator cmp, int nthreads) -> void {
|
static auto sort(Iterator start, Iterator end, Comparator cmp, int nthreads) -> void {
|
||||||
using valtype = typename std::iterator_traits<Iterator>::value_type;
|
using valtype = typename std::iterator_traits<Iterator>::value_type;
|
||||||
std::vector<valtype> temporary_dataset(std::distance(start, end));
|
std::vector<valtype> temporary_dataset(std::distance(start, end));
|
||||||
std::mutex dataset_guard;
|
std::recursive_mutex dataset_guard;
|
||||||
std::mutex depth_guard;
|
std::mutex depth_guard;
|
||||||
|
|
||||||
ms_split(temporary_dataset, start, end - 1, cmp, temporary_dataset.begin(), nthreads, dataset_guard, depth_guard);
|
ms_split(temporary_dataset, start, end - 1, cmp, temporary_dataset.begin(), nthreads, dataset_guard, depth_guard);
|
||||||
|
Loading…
Reference in New Issue
Block a user