[oneDPL][ranges][merge] support size limit for output; fixes for __par_backend::__parallel_for

Author: MikeDvorskiy

include/oneapi/dpl/pstl/algorithm_impl.h

@@ -2948,7 +2948,7 @@ __pattern_remove_if(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec,
 // merge
 //------------------------------------------------------------------------
 
-template<std::random_access_iterator It1, std::random_access_iterator It2, std::random_access_iterator ItOut, typename _Comp>
+template<typename It1, typename It2, typename ItOut, typename _Comp>
 std::pair<It1, It2>
 __brick_merge_2(It1 __it_1, It1 __it_1_e, It2 __it_2, It2 __it_2_e, ItOut __it_out, ItOut __it_out_e, _Comp __comp,
               /* __is_vector = */ std::false_type)
@@ -3082,8 +3082,8 @@ __pattern_merge_2(__parallel_tag<_IsVector>, _ExecutionPolicy&& __exec, _It1 __i
 
                                             //serial merge n elements, starting from input x and y, to [i, j) output range
                                             auto __res = __brick_merge_2(__it_1 + __r, __it_1 + __n_1,
-                                                                       __it_2 + __c, __it_2 + __n_2,
-                                                                       __it_out + __i, __it_out + __j, __comp, _IsVector{});
+                                                                         __it_2 + __c, __it_2 + __n_2,
+                                                                         __it_out + __i, __it_out + __j, __comp, _IsVector{});
 
                                             if(__j == __n_out)
                                             {

include/oneapi/dpl/pstl/omp/parallel_for.h

@@ -29,10 +29,10 @@ namespace __omp_backend
 
 template <class _Index, class _Fp>
 void
-__parallel_for_body(_Index __first, _Index __last, _Fp __f)
+__parallel_for_body(_Index __first, _Index __last, _Fp __f, std::size_t __grainsize)
 {
     // initial partition of the iteration space into chunks
-    auto __policy = oneapi::dpl::__omp_backend::__chunk_partitioner(__first, __last);
+    auto __policy = oneapi::dpl::__omp_backend::__chunk_partitioner(__first, __last, __grainsize);
 
     // To avoid over-subscription we use taskloop for the nested parallelism
     _PSTL_PRAGMA(omp taskloop untied mergeable)
@@ -49,20 +49,21 @@ __parallel_for_body(_Index __first, _Index __last, _Fp __f)
 
 template <class _ExecutionPolicy, class _Index, class _Fp>
 void
-__parallel_for(oneapi::dpl::__internal::__omp_backend_tag, _ExecutionPolicy&&, _Index __first, _Index __last, _Fp __f)
+__parallel_for(oneapi::dpl::__internal::__omp_backend_tag, _ExecutionPolicy&&, _Index __first, _Index __last, _Fp __f,
+               std::size_t __grainsize = __default_chunk_size)
 {
     if (omp_in_parallel())
     {
         // we don't create a nested parallel region in an existing parallel
         // region: just create tasks
-        oneapi::dpl::__omp_backend::__parallel_for_body(__first, __last, __f);
+        oneapi::dpl::__omp_backend::__parallel_for_body(__first, __last, __f, __grainsize);
     }
     else
     {
         // in any case (nested or non-nested) one parallel region is created and
         // only one thread creates a set of tasks
         _PSTL_PRAGMA(omp parallel)
-        _PSTL_PRAGMA(omp single nowait) { oneapi::dpl::__omp_backend::__parallel_for_body(__first, __last, __f); }
+        _PSTL_PRAGMA(omp single nowait) { oneapi::dpl::__omp_backend::__parallel_for_body(__first, __last, __f, __grainsize); }
     }
 }
 

include/oneapi/dpl/pstl/parallel_backend.h

@@ -35,6 +35,9 @@
 #    endif
 #endif
 
+//the parallel backend constants
+#define _ONEDPL_MERGE_CUT_OFF 2000
+
 namespace oneapi
 {
 namespace dpl

include/oneapi/dpl/pstl/parallel_backend_serial.h

@@ -45,7 +45,7 @@ __cancel_execution(oneapi::dpl::__internal::__serial_backend_tag)
 template <class _ExecutionPolicy, class _Index, class _Fp>
 void
 __parallel_for(oneapi::dpl::__internal::__serial_backend_tag, _ExecutionPolicy&&, _Index __first, _Index __last,
-               _Fp __f)
+               _Fp __f, std::size_t __grainsize = 1)
 {
     __f(__first, __last);
 }