Unify interface of algorithms.

Now all algorithms are used without the parallel pre-/suffix and the for_each method has an specialization for integer ranges.

app/benchmark_fft/main.cpp

@@ -49,7 +49,7 @@ void fft(complex_vector::iterator data, int n) {
     fft(data, n / 2);
     fft(data + n / 2, n / 2);
   } else {
-    pls::invoke_parallel(
+    pls::invoke(
         [&] { fft(data, n / 2); },
         [&] { fft(data + n / 2, n / 2); }
     );

app/benchmark_matrix/main.cpp

@@ -9,12 +9,12 @@ class matrix {
  public:
   T data[SIZE][SIZE];
 
-  matrix(T i = 1) {
+  explicit matrix(T i = 1) {
     std::fill(&data[0][0], &data[0][0] + SIZE * SIZE, i);
   }
 
   void multiply(const matrix<T, SIZE> &a, const matrix<T, SIZE> &b) {
-    pls::algorithm::parallel_for(0, SIZE, [&](int i) {
+    pls::algorithm::for_each_range(0, SIZE, [&](int i) {
       this->multiply_column(i, a, b);
     });
   }

app/benchmark_prefix/main.cpp

@@ -18,7 +18,7 @@ int main() {
   }
 
   pls::internal::helpers::run_mini_benchmark([&] {
-    pls::parallel_scan(vec.begin(), vec.end(), out.begin(), std::plus<double>(), 0.0);
+    pls::scan(vec.begin(), vec.end(), out.begin(), std::plus<double>(), 0.0);
   }, 8, 1000);
 
   PROFILE_SAVE("test_profile.prof")

app/benchmark_unbalanced/main.cpp

@@ -22,9 +22,11 @@ int count_child_nodes(uts::node &node) {
   std::vector<int> results(children.size());
   for (size_t i = 0; i < children.size(); i++) {
     size_t index = i;
-    auto lambda = [&, index] { results[index] = count_child_nodes(children[index]); };
-    pls::lambda_task_by_value<typeof(lambda)> sub_task(lambda);
-    pls::scheduler::spawn_child(sub_task);
+    auto lambda = [&, index] {
+      results[index] = count_child_nodes(children[index]);
+    };
+    using child_type = pls::lambda_task_by_value<typeof(lambda)>;
+    pls::scheduler::spawn_child<child_type>(lambda);
   }
   pls::scheduler::wait_for_all();
   for (auto result : results) {
@@ -41,8 +43,8 @@ int unbalanced_tree_search(int seed, int root_children, double q, int normal_chi
     uts::node root(seed, root_children, q, normal_children);
     result = count_child_nodes(root);
   };
-  pls::lambda_task_by_reference<typeof(lambda)> sub_task(lambda);
-  pls::scheduler::spawn_child(sub_task);
+  using child_type = pls::lambda_task_by_reference<typeof(lambda)>;
+  pls::scheduler::spawn_child<child_type>(lambda);
   pls::scheduler::wait_for_all();
 
   return result;

app/invoke_parallel/main.cpp

@@ -54,7 +54,7 @@ void fft(complex_vector::iterator data, int n) {
     fft(data, n / 2);
     fft(data + n / 2, n / 2);
   } else {
-    pls::invoke_parallel(
+    pls::invoke(
         [&] { fft(data, n / 2); },
         [&] { fft(data + n / 2, n / 2); }
     );

app/playground/main.cpp

@@ -10,5 +10,7 @@
 #include <pls/pls.h>
 
 int main() {
-
+  auto range = boost::irange(0, 10);
+  bool test = std::is_integral<typename std::remove_reference<decltype(range.begin())>::type>::value;
+  std::cout << test << std::endl;
 }

lib/pls/CMakeLists.txt

@@ -2,12 +2,12 @@
 add_library(pls STATIC
         include/pls/pls.h src/pls.cpp
 
-        include/pls/algorithms/invoke_parallel.h
-        include/pls/algorithms/invoke_parallel_impl.h
-        include/pls/algorithms/parallel_for.h
-        include/pls/algorithms/parallel_for_impl.h
-        include/pls/algorithms/parallel_scan.h
-        include/pls/algorithms/parallel_scan_impl.h
+        include/pls/algorithms/invoke.h
+        include/pls/algorithms/invoke_impl.h
+        include/pls/algorithms/for_each.h
+        include/pls/algorithms/for_each_impl.h
+        include/pls/algorithms/scan.h
+        include/pls/algorithms/scan_impl.h
 
         include/pls/internal/base/spin_lock.h
         include/pls/internal/base/tas_spin_lock.h src/internal/base/tas_spin_lock.cpp

lib/pls/include/pls/algorithms/parallel_for.h → lib/pls/include/pls/algorithms/for_each.h

@@ -2,23 +2,17 @@
 #ifndef PLS_PARALLEL_FOR_H
 #define PLS_PARALLEL_FOR_H
 
-// TODO: Replace with own integer iterator to remove dependency
-#include <boost/range/irange.hpp>
-
 namespace pls {
 namespace algorithm {
 
-template<typename RandomIt, typename Function>
-void parallel_for(RandomIt first, RandomIt last, const Function &function);
-
 template<typename Function>
-void parallel_for(size_t first, size_t last, const Function &function) {
-  auto range = boost::irange(first, last);
-  parallel_for(range.begin(), range.end(), function);
-}
+void for_each_range(size_t first, size_t last, const Function &function);
+
+template<typename RandomIt, typename Function>
+void for_each(RandomIt first, RandomIt last, const Function &function);
 
 }
 }
-#include "parallel_for_impl.h"
+#include "for_each_impl.h"
 
 #endif //PLS_PARALLEL_FOR_H

lib/pls/include/pls/algorithms/parallel_for_impl.h → lib/pls/include/pls/algorithms/for_each_impl.h

@@ -7,11 +7,15 @@
 
 #include "pls/internal/helpers/unique_id.h"
 
+// TODO: Replace with own integer iterator to remove dependency
+#include <boost/range/irange.hpp>
+
 namespace pls {
 namespace algorithm {
+namespace internal {
 
 template<typename RandomIt, typename Function>
-void parallel_for(RandomIt first, RandomIt last, const Function &function) {
+void for_each(RandomIt first, RandomIt last, const Function &function) {
   using namespace ::pls::internal::scheduling;
   constexpr long min_elements = 4;
 
@@ -26,18 +30,31 @@ void parallel_for(RandomIt first, RandomIt last, const Function &function) {
     long middle_index = num_elements / 2;
 
     auto second_half_body =
-        [first, middle_index, last, &function] { parallel_for(first + middle_index, last, function); };
+        [first, middle_index, last, &function] { internal::for_each(first + middle_index, last, function); };
     using second_half_t = lambda_task_by_reference<decltype(second_half_body)>;
     scheduler::spawn_child<second_half_t>(std::move(second_half_body));
 
     auto first_half_body =
-        [first, middle_index, last, &function] { parallel_for(first, first + middle_index, function); };
+        [first, middle_index, last, &function] { internal::for_each(first, first + middle_index, function); };
     using first_half_t = lambda_task_by_reference<decltype(first_half_body)>;
     scheduler::spawn_child_and_wait<first_half_t>(std::move(first_half_body));
   }
 }
 
 }
+
+template<typename Function>
+void for_each_range(size_t first, size_t last, const Function &function) {
+  auto range = boost::irange(first, last);
+  internal::for_each(range.begin(), range.end(), function);
+}
+
+template<typename RandomIt, typename Function>
+void for_each(RandomIt first, RandomIt last, const Function &function) {
+  internal::for_each(first, last, function);
+}
+
+}
 }
 
 #endif //PLS_INVOKE_PARALLEL_IMPL_H

lib/pls/include/pls/algorithms/invoke_parallel.h → lib/pls/include/pls/algorithms/invoke.h

@@ -9,15 +9,15 @@ namespace pls {
 namespace algorithm {
 
 template<typename Function1, typename Function2>
-void invoke_parallel(const Function1 &function1, const Function2 &function2);
+void invoke(const Function1 &function1, const Function2 &function2);
 
 template<typename Function1, typename Function2, typename Function3>
-void invoke_parallel(const Function1 &function1, const Function2 &function2, const Function3 &function3);
+void invoke(const Function1 &function1, const Function2 &function2, const Function3 &function3);
 
 // ...and so on, add more if we decide to keep this design
 
 }
 }
-#include "invoke_parallel_impl.h"
+#include "invoke_impl.h"
 
 #endif //PLS_PARALLEL_INVOKE_H

lib/pls/include/pls/algorithms/invoke_parallel_impl.h → lib/pls/include/pls/algorithms/invoke_impl.h

@@ -11,7 +11,7 @@ namespace pls {
 namespace algorithm {
 
 template<typename Function1, typename Function2>
-void invoke_parallel(Function1 &&function1, Function2 &&function2) {
+void invoke(Function1 &&function1, Function2 &&function2) {
   using namespace ::pls::internal::scheduling;
 
   using task_1_t = lambda_task_by_value<Function1>;
@@ -22,7 +22,7 @@ void invoke_parallel(Function1 &&function1, Function2 &&function2) {
 }
 
 template<typename Function1, typename Function2, typename Function3>
-void invoke_parallel(Function1 &&function1, Function2 &&function2, Function3 &&function3) {
+void invoke(Function1 &&function1, Function2 &&function2, Function3 &&function3) {
   using namespace ::pls::internal::scheduling;
 
   using task_1_t = lambda_task_by_value<Function1>;

lib/pls/include/pls/algorithms/parallel_scan.h → lib/pls/include/pls/algorithms/scan.h

@@ -6,10 +6,10 @@ namespace pls {
 namespace algorithm {
 
 template<typename InIter, typename OutIter, typename BinaryOp, typename Type>
-void parallel_scan(InIter in_start, const InIter in_end, OutIter out, BinaryOp op, Type neutral_elem);
+void scan(InIter in_start, const InIter in_end, OutIter out, BinaryOp op, Type neutral_elem);
 
 }
 }
-#include "parallel_scan_impl.h"
+#include "scan_impl.h"
 
 #endif //PLS_PARALLEL_SCAN_H_

lib/pls/include/pls/algorithms/parallel_scan_impl.h → lib/pls/include/pls/algorithms/scan_impl.h

@@ -28,7 +28,7 @@ void serial_scan(InIter input_start, const InIter input_end, OutIter output, Bin
 }
 
 template<typename InIter, typename OutIter, typename BinaryOp, typename Type>
-void parallel_scan(InIter in_start, const InIter in_end, OutIter out, BinaryOp op, Type neutral_elem) {
+void scan(InIter in_start, const InIter in_end, OutIter out, BinaryOp op, Type neutral_elem) {
   constexpr auto chunks_per_thread = 4;
   using namespace pls::internal::scheduling;
 
@@ -43,7 +43,7 @@ void parallel_scan(InIter in_start, const InIter in_end, OutIter out, BinaryOp o
     Type *chunk_sums = reinterpret_cast<Type *>(memory);
 
     // First Pass = calculate each chunks individual prefix sum
-    parallel_for(0, chunks, [&](int i) {
+    for_each_range(0, chunks, [&](int i) {
       auto chunk_start = in_start + items_per_chunk * i;
       auto chunk_end = std::min(in_end, chunk_start + items_per_chunk);
       auto chunk_output = out + items_per_chunk * i;
@@ -59,7 +59,7 @@ void parallel_scan(InIter in_start, const InIter in_end, OutIter out, BinaryOp o
     // Second Pass = Use results from first pass to correct each chunks sum
     auto output_start = out;
     auto output_end = out + size;
-    parallel_for(1, chunks, [&](int i) {
+    for_each_range(1, chunks, [&](int i) {
       auto chunk_start = output_start + items_per_chunk * i;
       auto chunk_end = std::min(output_end, chunk_start + items_per_chunk);
 

lib/pls/include/pls/pls.h

 #ifndef PLS_LIBRARY_H
 #define PLS_LIBRARY_H
 
-#include "pls/algorithms/invoke_parallel.h"
-#include "pls/algorithms/parallel_for.h"
-#include "pls/algorithms/parallel_scan.h"
+#include "pls/algorithms/invoke.h"
+#include "pls/algorithms/for_each.h"
+#include "pls/algorithms/scan.h"
 #include "pls/internal/scheduling/task.h"
 #include "pls/internal/scheduling/scheduler.h"
 #include "pls/internal/helpers/unique_id.h"
@@ -22,9 +22,9 @@ using internal::scheduling::lambda_task_by_reference;
 using internal::scheduling::lambda_task_by_value;
 using internal::scheduling::task;
 
-using algorithm::invoke_parallel;
-using algorithm::parallel_for;
-using algorithm::parallel_scan;
+using algorithm::invoke;
+using algorithm::for_each;
+using algorithm::scan;
 }
 
 #endif