Commit 2f755525 by Tobias Fuchs

algorithms_cpp: using block size partitioner in MergeSort

parent 40f92327
......@@ -48,57 +48,69 @@ class MergeSortFunctor {
public:
typedef typename std::iterator_traits<RAI>::value_type value_type;
MergeSortFunctor(RAI first, RAI last, RAITemp temporary_first,
ComparisonFunction comparison, const embb::mtapi::ExecutionPolicy& policy,
size_t block_size, const RAI& global_first, int depth)
: first_(first), last_(last), temp_first_(temporary_first),
comparison_(comparison), policy_(policy), block_size_(block_size),
MergeSortFunctor(size_t chunk_first, size_t chunk_last,
RAITemp temporary_first, ComparisonFunction comparison,
const embb::mtapi::ExecutionPolicy& policy,
const BlockSizePartitioner<RAI>& partitioner,
const RAI& global_first, int depth)
: chunk_first_(chunk_first), chunk_last_(chunk_last),
temp_first_(temporary_first),
comparison_(comparison), policy_(policy), partitioner_(partitioner),
global_first_(global_first), depth_(depth) {
}
void Action(mtapi::TaskContext& context) {
typedef typename std::iterator_traits<RAI>::difference_type difference_type;
size_t distance = static_cast<size_t>(std::distance(first_, last_));
if (distance <= 1) {
if(!CloneBackToInput() && distance != 0) {
RAITemp temp_first = temp_first_;
temp_first += std::distance(global_first_, first_);
*temp_first = *first_;
}
void Action(mtapi::TaskContext&) {
typedef typename std::iterator_traits<RAI>::difference_type
difference_type;
size_t chunk_split_index = (chunk_first_ + chunk_last_) / 2;
if (chunk_first_ == chunk_last_) {
// Leaf case: recurse into a single chunk's elements:
ChunkDescriptor<RAI> chunk = partitioner_[chunk_first_];
MergeSortChunkFunctor functor(chunk.GetFirst(),
chunk.GetLast(),
temp_first_,
global_first_,
depth_);
functor.Action();
return;
}
internal::ChunkPartitioner<RAI> partitioner(first_, last_, 2);
MergeSortFunctor<RAI, RAITemp, ComparisonFunction> functorL(
partitioner[0].GetFirst(), partitioner[0].GetLast(), temp_first_,
comparison_, policy_, block_size_, global_first_, depth_ + 1);
MergeSortFunctor<RAI, RAITemp, ComparisonFunction> functorR(
partitioner[1].GetFirst(), partitioner[1].GetLast(), temp_first_,
comparison_, policy_, block_size_, global_first_, depth_ + 1);
if (distance <= block_size_) {
functorL.Action(context);
functorR.Action(context);
} else {
mtapi::Node& node = mtapi::Node::GetInstance();
mtapi::Task taskL = node.Spawn(mtapi::Action(base::MakeFunction(functorL,
&MergeSortFunctor<RAI, RAITemp, ComparisonFunction>::Action),
policy_));
mtapi::Task taskR = node.Spawn(mtapi::Action(base::MakeFunction(functorR,
&MergeSortFunctor<RAI, RAITemp, ComparisonFunction>::Action),
policy_));
taskL.Wait(MTAPI_INFINITE);
taskR.Wait(MTAPI_INFINITE);
}
}
// Recurse further:
// Split chunks into left / right branches:
self_t functor_l(chunk_first_,
chunk_split_index,
temp_first_,
comparison_, policy_, partitioner_,
global_first_, depth_ + 1);
self_t functor_r(chunk_split_index + 1,
chunk_last_,
temp_first_,
comparison_, policy_, partitioner_,
global_first_, depth_ + 1);
mtapi::Node& node = mtapi::Node::GetInstance();
mtapi::Task task_l = node.Spawn(
mtapi::Action(
base::MakeFunction(functor_l, &self_t::Action),
policy_));
mtapi::Task task_r = node.Spawn(
mtapi::Action(
base::MakeFunction(functor_r, &self_t::Action),
policy_));
task_l.Wait(MTAPI_INFINITE);
task_r.Wait(MTAPI_INFINITE);
ChunkDescriptor<RAI> chunk_f = partitioner_[chunk_first_];
ChunkDescriptor<RAI> chunk_m = partitioner_[chunk_split_index + 1];
ChunkDescriptor<RAI> chunk_l = partitioner_[chunk_last_];
if(CloneBackToInput()) {
difference_type first = std::distance(global_first_, functorL.first_);
difference_type mid = std::distance(global_first_, functorR.first_);
difference_type last = std::distance(global_first_, functorR.last_);
SerialMerge(temp_first_ + first, temp_first_ + mid,
temp_first_ + last, functorL.first_, comparison_);
difference_type first = std::distance(global_first_, chunk_f.GetFirst());
difference_type mid = std::distance(global_first_, chunk_m.GetFirst());
difference_type last = std::distance(global_first_, chunk_l.GetLast());
SerialMerge(temp_first_ + first, temp_first_ + mid, temp_first_ + last,
chunk_f.GetFirst(),
comparison_);
} else {
SerialMerge(functorL.first_, functorR.first_, functorR.last_,
temp_first_ + std::distance(global_first_, functorL.first_),
SerialMerge(chunk_f.GetFirst(), chunk_m.GetFirst(), chunk_l.GetLast(),
temp_first_ + std::distance(global_first_, chunk_f.GetFirst()),
comparison_);
}
}
......@@ -114,12 +126,77 @@ class MergeSortFunctor {
}
private:
RAI first_;
RAI last_;
typedef MergeSortFunctor<RAI, RAITemp, ComparisonFunction> self_t;
private:
/**
* Non-parallelized part of merge sort on elements within a single chunk.
*/
class MergeSortChunkFunctor {
public:
MergeSortChunkFunctor(RAI first, RAI last,
RAITemp temp_first,
const RAI & global_first,
int depth)
: first_(first), last_(last),
temp_first_(temp_first), global_first_(global_first),
depth_(depth) {
}
void Action() {
size_t distance = static_cast<size_t>(
std::distance(first_, last_));
if (distance <= 1) {
// Leaf case:
if(!CloneBackToInput() && distance != 0) {
RAITemp temp_first = temp_first_;
std::advance(temp_first, std::distance(global_first_, first_));
*temp_first = *first_;
}
return;
}
// Recurse further. Use binary split, ignoring chunk size as this
// recursion is serial:
ChunkPartitioner<RAI> partitioner(first_, last_, 2);
ChunkDescriptor<RAI> chunk_l = partitioner[0];
ChunkDescriptor<RAI> chunk_r = partitioner[1];
MergeSortChunkFunctor functor_l(
chunk_l.GetFirst(),
chunk_l.GetLast(),
temp_first_, global_first_, depth_ + 1);
MergeSortChunkFunctor functor_r(
chunk_r.GetFirst(),
chunk_r.GetLast(),
temp_first_, global_first_, depth_ + 1);
functor_l.Action();
functor_r.Action();
}
private:
/**
* Determines the input and output arrays for one level in merge sort.
*
* \return \c true if the temporary data range is input and the array to be
* sorted is output. \c false, if the other way around.
*/
bool CloneBackToInput() {
return depth_ % 2 == 0 ? true : false;
}
RAI first_;
RAI last_;
RAITemp temp_first_;
RAI global_first_;
int depth_;
};
private:
size_t chunk_first_;
size_t chunk_last_;
RAITemp temp_first_;
ComparisonFunction comparison_;
const embb::mtapi::ExecutionPolicy& policy_;
size_t block_size_;
const BlockSizePartitioner<RAI>& partitioner_;
const RAI& global_first_;
int depth_;
......@@ -180,8 +257,16 @@ void MergeSort(
"Not enough MTAPI tasks available to perform the merge sort");
}
internal::BlockSizePartitioner<RAI> partitioner(first, last, block_size);
internal::MergeSortFunctor<RAI, RAITemp, ComparisonFunction> functor(
first, last, temporary_first, comparison, policy, block_size, first, 0);
0, partitioner.Size() - 1,
temporary_first,
comparison,
policy,
partitioner,
first,
0);
mtapi::Task task = node.Spawn(mtapi::Action(base::MakeFunction(functor,
&internal::MergeSortFunctor<RAI, RAITemp, ComparisonFunction>::Action),
policy));
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or sign in to comment